DESCRIPTION AND EVALUATION OF THE METEOROLOGICAL CASES
USED IB THE RADH NAPAP EVALUATION AND ASSESSMENT
by
Nelson L. Seaaan
Department of Meteorology
The Pennsylvania State University
University Park, Pennsylvania 16802
Reporting under
Inceragency Agreeaent DW49933202
b««"veen U.S. Environaental Protection Agency and
National Center for Atmospheric Research
and
Subavard No. NCAR S8905
between National Center for Atoospheric Research and
The Pennsylvania State University
Project Officer
John F. Clarke
Ataospheric Sciences Modeling Division
Ataospheric Research and Exposure Assessment Laboratory
U.& Environmental Protection Agency
Research Triangle Park, North Carolina 27711
ATMOSPHERIC RESEARCH AND EXPOSURE ASSESSMENT LABORATORY
OFFICE OF RESEARCH AND DEVELOPMENT
U.S. ENVIRONMENTAL PROTECTION AGENCY
• EES1ARCH TRIASGLE PARK, NORTH CAROLINA 27711
December 1989
-------
GOO19893
DESCRIPTION AND EVALUATION OF THE METEOROLOGICAL CASES
USED IN THE BADM NAFAP EVALUATION AND ASSESSMENT
by
Nelson L. Seaman
Department of Meteorology
The Pennsylvania State University
University Park, Pennsylvania 16802
Reporting under
Interagency Agreement DW49933202
between U.S. Environmental Protection Agency and
National Center for Atmospheric Research
and
Subaward No. NCAR S8905
between National Center for Atmospheric Research and
The Pennsylvania State University
Project Officer
John F. Clarke
Atmospheric Sciences Modeling Division
Atmospheric Research and Exposure Assessment Laboratory
U.S. Environmental Protection Agency
Research Triangle Park, North Carolina 27711
ATMOSPHERIC RESEARCH AND EXPOSURE ASSESSMENT LABORATORY
OFFICE OF RESEARCH AND DEVELOPMENT
U.S. ENVIRONMENTAL PROTECTION AGENCY
RESEARCH TRIANGLE PARK, NORTH CAROLINA 27711
December 1989
-------
DISCLAIMER
The information contained in this document has been funded wholly or in
part by the U.S. Environmental Protection Agency (EPA) under Interagency
Agreement DW49933202 between EPA and the National Center for Atmospheric
Research (NCAR) and Subaward No. NCAR S8905 between NCAR and the Pennsylvania
State University. Mention of trade names or commercial products does not
constitute endorsement or recommendation for use. This document is intended
for internal use only.
ii
-------
ABSTRACT
The Penn State/NCAR meteorological model, known as MM4, with four-
dimensional data assimilation (FDDA) has been used to produce meteorological
data sets for input to the Regional Acid Deposition Model (RADM). The
meteorological model, with its extension to apply FDDA, will be referred to in
this document as MM4/FDDA. The meteorological data sets have been produced
for the RADM NAPAP Assessment and Evaluation study. This report provides a
summary description of the numerical model, the meteorological cases, the
methods used to produce the numerical data sets and their statistical
verification.
Additional details about the model and the methodology used to develop
the FDDA scheme are documented elsewhere (refer to References). Also, a set
of volumes, the Case Documentation Reports, have been supplied by Penn State
to EPA. Each volume contains plotted fields of selected meteorological
variables for one case study (generally five days) at 12-hour intervals, in
addition to the case descriptions and statistical summaries also contained in
this report. The Case Documentation Reports can be obtained from EPA,
Atmospheric Research and Exposure Assessment Laboratory, Research Triangle
Park, N.C., 27711.
iii
-------
iv
-------
TABLE OF CONTENTS
Page
Disclaimer ii
Abstract iii
List of Tables vi
List of Figures vii
1. Introduction 1
1.1 Background 1
1.2 Description of Contents 8
2. Description of the Meteorological Modeling System 10
2.1 The PSU/NCAR Model 10
2.2 Model Initialization 12
2.3 Four-Dimensional Data Assimilation 14
2.4 Quality Assurance of Model Simulations 16
3. Verification of Meteorological Data Sets 20
3.1 Description of Statistical Approach 20
3.2 Statistical Results of Model Verification 23
3.2.1 Assimilated Primitive Variables 23
3.2.2 Non-assimilated Primitive Variables 29
3.2.3 Precipitation Scores 35
4. Summary and Conclusions 55
REFERENCES 58
APPENDICES
A. Definition of Statistical Quantities 60
B. Uncertainty Characteristics of Rainfall Analyses 69
C. Statistical Summaries of the Meteorological Cases 88
D. Brief Case Descriptions 151
E. MM4/FDDA Output Volume Names 216
-------
LIST OF TABLES
Page
Table 1. Dates of the cases for RADK Evaluation and
Assessment. The Field Program key appears
below
Table 2. MM4/FDDA seasonally and annually averaged values
of case RMS errors for selected meteorological
variables, based on 61 cases 24
Table 3. MM4/FDDA seasonally and annually averaged values
of case mean-errors for selected meteorological
variables, based on 61 cases 25
Table 4. MM4/FDDA seasonally and annually averaged values
for selected precipitation verification
statistics, based on 60 cases. All precipita-
tion amounts are for 12-h periods 36
Table Al. Typical values of 24-h Threat Score based on
data reported by Anthes (1983) and Black and
Mesinger (1989) for the LFM and NGM operational
mesoscale forecast models 65
Table El. List of volume names for MM4/FDDA output
archived at NCAR 217
vi
-------
LIST OF FIGURES
Figure 1. Mesoscale model domain and the precipitation
verification region (heavy solid line). The
shaded area indicates grid boxes with 10 or
more rain gauges per box 13
Figure 2. MM4/FDDA means (hatched) and standard
deviations (cross-hatched) of seasonal and
annual RMS Error of 500-mb wind (based on
vector wind difference, calculated for 61
cases). Scores are shown at top 26
Figure 3. MM4/FDDA case-mean RMS Errors for 500-mb wind
(based on vector wind difference) versus Julian
Day for 61 cases 27
Figure 4. MM4/FDDA means (hatched) and standard
deviations (cross-hatched) of seasonal and
annual RMS Errors of 850-mb temperature, based
on 61 cases. Scores are shown at top 28
Figure 5. MM4/FDDA means (hatched) and standard
deviations (cross-hatched) of seasonal and
annual RMS Errors of surface mixing ratio,
based on 61 cases. Scores are shown at top 30
Figure 6. MM4/FDDA means (hatched) and standard
deviations (cross-hatched) of seasonal and
annual Sj Score for sea-level pressure
(calculated for a standard distance of 320 km)
based on 61 cases. Scores are shown at top and
dashed line (Sx - 30) indicates an essentially
"perfect" simulation 32
Figure 7. MM4/FDDA case-mean St Scores for sea-level
pressure (calculated for a standard distance of
320 km) versus Julian Day, based on 61 cases 33
Figure 8. MM4/FDDA means (hatched) and standard
deviations (cross-hatched) of seasonal and
annual Sj^ Score for 500-mb height (calculated
for a standard distance of 320 km) based on 61
cases. Scores are shown at top and dashed line
(St - 20) indicates an essentially "perfect"
simulation 34
vii
-------
Figure 9. MM4/FDDA mean seasonal and annual Threat Scores
(0.25 cm/12 h threshold) and standard deviation
from mean threat scores, based on 60 cases 37
Figure 10. MM4/FDDA Threat Scores (0.25-cm threshold)
versus Julian Day for 60 cases. Seasonal mean
TS (solid) and seasonal standard deviations
from the mean (dashed) are shown 38
Figure 11. MM4/FDDA mean seasonal and annual Threat Scores
(0.64 cm/12 h threshold) and standard deviation
from mean threat scores, based on 60 cases 40
Figure 12. MM4/FDDA mean seasonal and annual Bias Scores
(0.25 cm/12 h threshold) and standard deviation
from mean threat scores, based on 60 cases 41
Figure 13. MM4/FDDA Bias Scores (0.25-cm threshold) versus
Julian Day for 60 cases. Seasonal mean BS
(curved solid) and seasonal standard deviations
from the mean (dashed) are shown 42
Figure 14. MM4/FDDA mean seasonal and annual Bias Scores
(0.64 cm/12 h threshold) and standard deviation
from mean threat scores, based on 60 cases 43
Figure 15. Thirty-day mean Bias Scores for MM4/FDDA at the
0.05 cm/12 h threshold. Each period is
approximately 30 days long and is based on a
cluster of 6 cases (usually 5 days each)
arranged seasonally. Months of cases in each
cluster are shown below, and the annual grand
-average BS is shown to the right. The exact
BS values are given above each cluster 45
Figure 16. MM4/FDDA mean annual Bias Score versus
precipitation rate (cm/12h) , based on 24 cases 46
Figure 17. Comparison of mean Bias Scores versus precip-
itation rate based on cases for November 1988,
using MM4/FDDA and the NMC NGM and ETA (80-km
resolution) models. The last two are based on
48-h forecasts (after Black and Mesinger,
1989). (a) Comparison for 24-h precipitation
totals using the three models, (b) Comparison
for N - 12-h versus N - 24-h precipitation
totals using MM4/FDDA 48
viii
-------
Figure 18. Comparison of mean Threat Scores versus precip
-itation rate based on cases for November 1988,
using MM4/FDDA and the NMC NGM and ETA (80-km
resolution) models. The last two are based on
48-h forecasts (after Black and Mesinger, 1989).
(a) Comparison for 24-h precipitation totals
using the three models, (b) Comparison for
N - 12-h versus N - 24-h precipitation totals
using MM4/FDDA 50
Figure 19. MM4/FDDA mean seasonal and annual Mean Absolute
Errors (MAE) of 12-h precipitation and standard
deviation from MAE, based on 60 cases 53
Figure Al. Schematic showing an idealized verification of
model rainfall for which the Threat Score is
0.333 and the Bias Score is 1.00. The solid
(dashed) region represents a forecast (observed)
rain area for an arbitrary threshold. (X - X'
and Y - Y' ) 63
Figure Bl. Map of precipitation (cm) observed during the
storm of 13-15 January 1980. Numbers are in
hundredths of cm, plotted at the climatological
stations (After Forbes, et al., 1987). An
80-km grid is superimposed for reference
(straight thin solids) 70
Figure B2. Representativeness uncertainty of rainfall
observations. The mean absolute deviation from
the mean precipitation is plotted as a function
of the mean precipitation (MP) , valid when a.
single observation is used to represent rain-
fall on an 80-km grid. Based on 14,620 events
in 1985 and 1988 over the eastern U. S. for all
seasons and all terrain. Plots are shown for
rainfall up to (a) 5.0 cm, and (b) 1.0 cm 72
Figure B3. Same as Fig. B2, except for 3,771 events during
winter (Dec., Jan., Feb.) and for all terrain 74
Figure B4. Same as Fig. B2, except for 3,692 events during
spring (Mar., Apr., May) and for all terrain 76
Figure B5. Same as Fig. B2, except for 3,609 events during
summer (Jun., Jul., Aug.) and for all terrain 78
Figure B6. Same as Fig. B2, except for 3,548 events during
autumn (Sep., Oct., Nov.) and for all terrain 80
ix
-------
Figure B7. Same as Fig. B2, except for 7,312 events for
all seasons and for gentle terrain (below 400 m) 82
Figure B8. Same as Fig. B2, except for 7,308 events for
all seasons and for rough terrain (above 400 m) 84
-------
1. INTRODUCTION
This report documents the methodology used to generate meteorological
data sets for the Regional Acid Deposition Model (RADM) Assessment and
Evaluation study and presents case descriptions and statistical summaries of
the data sets. A total of 61 cases covering 3-5 days each has been processed
(see Table 1). These sets are generated using the Pennsylvania State
University (PSU)/National Center for Atmospheric Research (NCAR) mesoscale
meteorological model (MM4) with an extension to use the four-dimensional data
assimilation (FDDA) approach developed at PSU. The extended modeling system
will be referred to as MM4/FDDA. The report also provides information useful
for interpreting the model results. In addition to the material presented
here, a Case Documentation Report has been prepared (one volume per case)
showing plotted fields of selected meteorological variables. The Case
Documentation volumes are available from the U.S. Environmental Protection
Agency (EPA), Atmospheric Sciences Modeling Division, Atmospheric Research and
Exposure Assessment Laboratory, Research Triangle Park, N.C., 27711.
1.1 Background
The effort to develop these meteorological data sets is one part of a
much broader program undertaken by EPA and a number of other Federal agencies
(1) to monitor and understand the chemistry, physics and distribution of acid
precipitation in North America and (2) to study and predict the impact of
alternate pollution emission scenarios. This extended project, known as the
National Acid Precipitation Assessment Program (NAPAP), is intended to aid in
the development of a national pollution abatement strategy by providing the
best scientific understanding currently possible on this highly complex
1
-------
Table 1. Dates of the cases for RADM Evaluation and Assessment.
(Field Program key* appears below.)
Case
No.
1
2
3
4
5
6
7
8
9
10
11
12
13
14
ISA
15B
16
17
18
19
20
21
Dates
(UTC time, day, month, year)
0000,
1200,
0000,
0000,
1200,
0000,
0000,
0000,
0000,
0000,
0000,
1200,
0000,
1200,
0000,
0000,
0000,
1200,
0000,
0000,
0000,
0000,
7
11
20
12
16
27
18
27
6
19
28
1
4
8
12
16
19
23
13
23
13
27
Apr.
Apr.
Apr.
Jul.
Jul.
Jan.
Mar.
Jun.
Aug.
Aug.
Aug.
Sep.
Sep.
Sep.
Sep.
Sep.
Sep.
Sep.
Apr.
Sep.
Dec.
May
1981
1981
1981
1980
1980
1982
1982
1982
1982
1988
1988
1988
1988
1988
1988
1988
1988
1988
1982
1982
1982
1983
- 0000,
- 0000,
- 0000,
- 0000,
- 0000,
- 0000,
- 0000,
- 0000,
- 0000,
- 0000,
- 0000,
- 1200,
- 0000,
- 1200,
- 1200,
- 1200,
- 0000,
- 1200,
- 0000,
- 0000,
- 0000,
- 0000,
12
15
25
17
20
1
23
2
11
24
2
. 4
9
12
16
19
24
28
18
28
18
1
Apr.
Apr.
Apr.
Jul.
Jul.
Feb.
Mar.
Jul.
Aug.
Aug.
Sep.
Sep.
Sep.
Sep.
Sep.
Sep.
Sep.
Sep.
Apr.
Sep.
Dec.
Jun.
Institution Field
Program
1981
1981
1981
1980
1980
1982
1982
1982
1982
1988
1988
1988
1988
1988
1988
1988
1988
1988
1982
1982
1982
1983
PSU
PSU
PSU
PSU
PSU
NCAR
NCAR
NCAR
NCAR
NCAR
NCAR
NCAR
NCAR
NCAR
NCAR
NCAR
NCAR
NCAR
NCAR
PSU
NCAR
NCAR
OSCAR1
OSCAR2
OSCAR3
NEROS
NEROS
TEST
TEST
TEST
TEST
EVAL
EVAL
EVAL
EVAL
EVAL
EVAL
EVAL
EVAL
EVAL
AGG
AGG
AGG
AGG
-------
Table 1. (Continued)
Case No.
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37A
37B
38A
38B
39
40
41
42
0000,
0000,
0000,
0000,
0000,
0000,
0000,
0000,
0000,
0000,
0000,
0000,
0000,
0000,
0000,
0000,
1200,
1200,
0000,
0000,
0000,
0000,
0000,
2
7
30
12
15
19
10
7
14
31
10
30
14
24
9
13
16
23
26
28
4
12
8
(UTC
Aug.
Sep.
Oct.
Sep.
Mar.
Aug.
Jun.
Sep.
Jul.
Oct.
Jul.
Apr.
Nov.
Sep.
Oct.
Aug.
Aug.
Aug.
Aug.
Sep.
Nov.
May
Jun.
Dates
time, day, month, year)
1983
1983
1983
1983
1984
1984
1984
1984
1984
1985
1985
1985
1985
1985
1985
1988
1988
1988
1988
1988
1982
1982
1983
- 0000,
- 0000,
- 0000,
- 0000,
- 0000,
- 0000,
- 0000,
- 0000,
- 0000,
- 0000,
- 0000,
- 0000,
- 0000,
- 0000,
- 0000,
- 0000,
- 1200,
- 1200,
- 1200,
- 0000,
- 0000,
- 0000,
- 0000,
7
12
4
17
20
24
15
12
19
5
15
5
19
29
14
17
19
26
28
1
9
17
13
Aug.
Sep.
Nov.
Sep.
Mar.
Aug.
Jun.
Sep.
Jul.
Nov.
Jul.
May
Nov.
Sep.
Oct.
Aug.
Aug.
Aug.
Aug.
Oct.
Nov.
May
Jun.
1983
1983
1983
1983
1984
1984
1984
1984
1984
1985
1985
1985
1985
1985
1985
1988
1988
1988
1988
1988
1982
1982
1982
Institution
NCAR
PSTJ
PSU
PSU
PSU
PSU
PSU
PSU
NCAR
PSU
PSU
NCAR
PSU
PSU
PSU
NCAR
NCAR
NCAR
NCAR
NCAR
NCAR
NCAR
NCAR
AGG
AGG
AGG
AGG
AGG
AGG
AGG
AGG
AGG
AGG
AGG
AGG
AGG
AGG
AGG
EVAL
EVAL
EVAL
EVAL
EVAL
AGG/D
AGG/D
AGG/D
-------
Table 1. (Continued)
Case No.
Dates
Institution
(UTC time, day, month, year)
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
0000,
0000,
0000,
0000,
0000,
0000,
0000,
0000,
0000,
1200,
0000,
1200,
0000,
1200,
0000,
1200,
15 Jul. 1985 - 0000, 20 Jul.
14 Dec. 1985 - 0000, 19 Dec.
26 Apr. 1985 - 0000, 1 May
30 Jan. 1982 - 0000, 4 Feb.
7 Dec. 1983 - 0000, 12 Dec.
2 Nov. 1985 - 0000, 7 Nov.
7 Feb. 1985 - 0000, 12 Feb.
11 Nov. 1983 - 0000, 16 Nov.
2 Nov. 1988 - 0000, 7 Nov.
6 Nov. 1988 - 1200, 11 Nov.
11 Nov. 1988 - 0000, 16 Nov.
15 Nov. 1988 - 1200, 20 Nov.
20 Nov. 1988 - 0000, 25 Nov.
24 Nov. 1988 - 0000, 29 Nov.
16 Dec. 1988 - 0000, 20 Dec.
19 Dec. 1988 - 0000, 23 Dec.
1985
1985
1985
1982
1983
1985
1985
1983
1988
1988
1988
1988
1988
1988
1988
1988
NCAR
PSU
PSU
PSU
PSU
PSU
PSU
PSU
PSU
PSU
PSU
PSU
NCAR
NCAR
NCAR
NCAR
AGG/D
AGG/D
AGG/D
AGG/W
AGG/W
AGG/W
AGG/W
AGG/W
EVAL
EVAL
EVAL
EVAL
EVAL
EVAL
EVAL
EVAL
Field Program Key:
OSCAR Oxidation and Scavenging of April
NEROS Northeast Regional Oxidant Study
TEST Test Aggregation Cases
EVAL NAPAP Evaluation Cases
AGG
AGG/D
AGG/W
Wet Aggregation Cases
Dry Aggregation Cases
Winter Aggregation Cases
Rains
-------
problem. Detailed understanding of the problem is important, not only from
the scientific standpoint, but also because of the potentially enormous cost
of significantly reducing acid precipitation and the need to equitably share
that cost among appropriate public and private entities (e.g., industry,
states, regions, and nations).
Design and evaluation of possible alternative emission scenarios depends
greatly on the development and validation of a suitable numerical simulation
model capable of reproducing both existing and future acid deposition
patterns. This approach is dictated by the unacceptably high cost and
socioeconomic disruption required to directly alter regional-scale emissions
for the purpose of testing.
Stated simply, the deposition of acidic chemical compounds is a function
of (1) chemical emissions, (2) mixing and transport of the emissions in the
dynamically active atmosphere, (3) transformation of the original constituents
(often highly dependent on concentrations of various chemical species and on
sunlight), and (4) removal, through meteorological processes, leading to
either wet (precipitation related) or dry (direct impaction related)
deposition. Since the time scales involved range from seconds for certain
transformations to several days for transport and removal, and corresponding
length scales range from hundreds of meters to thousands of kilometers,
accurate simulation of these chemical and meteorological processes must be
done by complex regional-scale models. The dependence of the chemistry on the
meteorology for mixing, transport, reaction rates, and deposition means that
detailed treatment of the atmospheric problem is required before an air
chemistry model can succeed. Therefore, EPA has sponsored the development,
testing, and validation of both a state-of-the-art air chemistry model, known
-------
as RADH, and a state-of-the-art dynamic meteorological model capable of
performing FDDA (MM4/FDDA).
The decision to use a numerical model, rather than rely on direct
analyses of atmospheric observations, to supply the necessary meteorological
conditions to RADM is based on several crucial factors:
(1) Observations of the upper atmosphere are taken only
twice daily at spatial separations which average
(over land) about 400 km.
(2) Many of the most important meteorological processes
related to the acid deposition problem (such as
boundary layer dynamics, lower-level and upper
-level jets, fronts, squall lines, rainfall bands)
occur on time and spatial scales which are often
poorly resolved by the observation network.
(3) Acid deposition is highly dependent on the
location, timing, and intensity of precipitation.
(4) Air chemistry and precipitation are strongly
related to convergence/divergence patterns and to
mesoscale vertical motions, which cannot be
resolved or accurately inferred from the
observation network.
(5) No analysis method is known from which three
-dimensional intervariable dynamic consistency, the
result of non-linear interactions, can be derived
between the precipitation observations (highly
dependent on small convective cells) and the 12
-hourly synoptic-scale atmospheric state variables
-------
(pressure, wind, temperature, moisture, etc.).
Thus, a numerical model is the best means currently available for
generating atmospheric information which has sufficient spatial and temporal
detail and which is dynamically consistent, such that it will be suitable for
incorporation into RADM.
However, it is well known that even the most advanced meteorological
forecast models are capable of producing large errors in winds (transport) and
precipitation during the 3 to 5 day episodes to be studied by RADM. In this
form, a numerical model uses only those observations available at the initial
time of the study period. To control error growth in the numerical
simulation, MM4 was adapted to assimilate observations not only at the initial
time, but throughout the simulation period (i.e., FDDA). In this way it is
possible to generate data sets whose accuracy, particularly in terms of
transport, precipitation, and dynamic consistency, is superior to both direct
interpolation of synoptic-scale analyses of observations and a purely
predictive-mode model result.
The advantages of the MM4/FDDA system are based on the ability of the
model to make use of the plentiful observations available after the initial
time of each simulation, while retaining the numerical integration framework
of MM4 to develop mesoscale details, vertical motions, and intervariable
consistency that are not contained in the analyzed observations. The model is
also capable of simulating non-linear interactions that cannot be represented
by direct interpolation between analyses based on synoptic observations
separated by 12 hours. In this way, FDDA prevents accumulation of large-scale
forecast errors (phase and amplitude errors) without serious loss of the
mesoscale details generated by the numerical integration of the model's
-------
primitive equations. Similar methodology has been applied on the global scale
in GARP (Global Atmospheric Research Program) to produce the FGGE-IIIb data
sets using FDDA. Thus, the FDDA technique can be considered as a
sophisticated analysis tool for producing detailed internally consistent data
sets in which the complete primitive equations are used as dynamic constraints
and the numerical model, itself, becomes the analysis tool.
1.2 Description of Contents
Section 2 of this report provides an overview of the PSU/NCAR model MM4,
its initialization scheme and the FDDA scheme. Further details about these
topics as they relate to MM4/FDDA can be found in Seaman and Stauffer (1989)
and Stauffer and Seaman (1990). This section also describes the quality
assurance methods designed for the model data sets. In Section 3, the mean
statistical results of the model verification are summarized for the entire 61
cases and for certain subsets (e.g., seasonal). Also, a brief description of
the verification approach is provided. Section 4 gives the summary and
conclusions.
A series of appendices provides much additional information about the
data assessments. Appendix A gives definitions for each of the statistical
quantities appearing in Section 3 and the other appendices. It also explains
how to interpret the statistics and provides, wherever possible, related
information about typical values of each obtained from the literature. A
discussion of the representativeness of rainfall observations is provided in
Appendix B. This is crucial to understanding the precipitation analyses and
the interpretation of the model verification statistics. Appendix C provides
the case-by-case statistical summaries, while Appendix D gives a brief
description of the synoptic events of each case. A list of the MM4/FDDA
8
-------
output volume names used to archive the numerical model data sets at NCAR
appears in Appendix E.
-------
2. DESCRIPTION OF THE METEOROLOGICAL MODELING SYSTEM
The meteorological data sets are produced with the PSU/NCAR modeling
system, which is a complete system designed to acquire raw data, prepare the
necessary two and three-dimensional initial conditions for the mesoscale
model , supply boundary conditions , integrate the primitive equations , and
store and post-process the numerical results. Thus, the mesoscale model,
itself, is only one component of a much more extensive set of software.
The sequence of programs that make up the modeling software is described
more fully by Anthes, et al. (1987). For the present purpose, it is
sufficient to say that there are three primary components: (1) an
initialization package which produces the objective analyses used to generate
the initial and lateral boundary conditions, (2) the basic PSU/NCAR primitive
equation and physics model (MM4) , and (3) the FDDA package developed at PSU
for use in MM4.
2.1 The PSU/NCAR Model
The numerical model used in this system is an improved version of the
PSU/NCAR mesoscale model described by Anthes and Warner (1978) and Anthes, et
al. (1987). The three-dimensional, limited- area hydrostatic primitive
equation model is written in a terrain- following sigma vertical coordinate,
(1)
P,-Pt P*
where p is pressure, pt is a constant pressure at the top of the model, and ps
is surface pressure. The model equations are written in "flux" form, where
the prognostic variables for horizontal wind, temperature and mixing ratio are
mass weighted by p* (i.e., p*u, p*v, p*T, and p*q) , and where p* - ps - pt.
10
-------
The ground temperature is predicted with a surface energy budget, which
includes longwave and shortwave irradiances which are affected by calculated
cloud cover. The surface parameters (roughness, albedo, moisture
availability, etc.) are variable over the grid and are defined from archived
land use via a look-up table (Anthes, et al., 1987). The planetary boundary
layer (PBL) is represented by a multi-layer Blackadar-type scheme (Zhang and
Anthes, 1982).
The model produces both resolved (grid- scale) precipitation and
parameterized convection (Anthes, 1977). The Anthes-Kuo cumulus
parameterization has been modified (Anthes, et al. , 1987) to allow prescribed
parabolic profiles of heating and moistening (Kuo and Anthes, 1984) resulting
from moist convection to vary in time and space. The upper and lower limits
of these profiles are diagnosed by computing the cloud base and cloud top from
the model sounding. In addition to the standard critical moisture convergence
criterion, moist convection can exist only if the model atmosphere can support
a convective cloud greater than about 300 mb in depth.
The lateral boundary conditions are specified from observations by
interpolating in time the 12 -hourly enhanced analyses described in Section
2.2. The numerical method for incorporating the observed, large-scale
tendency information into the model is based on that of Davies (1976) , and
involves nudging the prognostic model variable, a, toward the observed
analysis, a0. The lateral boundary conditions are then computed by
atn
| - F(n) G!(a0-a) - F(n)G272(af0-a) n-2,3,4 (3)
•'n
where F decreases linearly from the lateral boundary,
11
-------
F(n) - (5-n)/3 n-2,3,4 (4)
F(n) - 0 n>4 (5)
and the nudging factors G^ and G2 are given by
G! - 0.05/Dt (6)
G2 - (Dx)2/(50 Dt) (7)
where Dt is the model time step (s) and Dx is the model grid length.
All cases were simulated with a uniform 46 x 61 mesh (Dx - 80 km)
covering the continental United States (Fig. 1). Fifteen sigma calculation
levels were defined at 0.995, 0.985, 0.970, 0.945, 0.910, 0.865, 0.810, 0.740,
0.650, 0.550, 0.450, 0.350, 0.250, 0.150, and 0.050. The model top was set at
100 mb and the time step (Dt) was 120 s.
2.2 Model Initialization
The mesoscale three-dimensional analyses used to provide initial and
lateral boundary conditions for the MM4 model are obtained by first
interpolating the following operationally-analyzed fields from the synoptic-
scale ECMWF (European Center for Medium-range Weather Forecasting) global data
assimilation system (received as gridded analyses with 2.5° resolution, or
about 275 km) to the mesoscale model grid (80-km resolution): the horizontal
wind components (u and v), temperature (T), and relative humidity at the
standard (mandatory) pressure levels, plus sea-level pressure (SLP) and ground
temperature (TG). TG is surface temperature over land and sea-surface
temperature over water.
Next, these fields are interpolated vertically to the four supplemental
analyses levels used for this study (950, 900, 800, and 600 mb). At this
12
-------
U)
Figure 1. Mesoscale model domain and the precipitation
verification region (heavy solid line). The
shaded area indicates grid boxes with 10 or
more rain gauges per box .
-------
point, the interpolated analyses (80-km resolution) at the standard and
supplemental levels contain only the information found in the original ECMWF
synoptic-scale fields. The interpolated analyses are then used as the
background ("first guess") for a mesoscale objective analysis which enhances
the 80-km fields by blending in data from the standard network of surface and
radiosonde stations. The objective-analysis technique uses a successive-
correction approach that accounts for along-wind correlation of the variables
in curved flow (Benjamin and Seaman, 1985). After the surface pressure is
determined hydrostatically from the SLP and terrain height, all the three-
dimensional fields are interpolated to the model's 15 sigma vertical levels.
2.3 Four-Dimensional Data Assimilation
The method of Newtonian relaxation, or nudging. is used in MM4/FDDA to
perform the data assimilation. This approach to FDDA relaxes the model state
toward the observed state by adding, to one or more of the prognostic
equations, an artificial tendency term that is based on the difference between
the two states. For this study, the model solution is nudged toward gridded
analyses of the data, but the scheme also allows nudging toward individual
observations during a period of time surrounding the observations.
The analysis-nudging term for a given variable is proportional to the
difference between the model simulation and an analysis (derived from
observations) calculated at every grid point. The general form for the
predictive equation of any model variable, a, is written as
|| - F(a,x,t) + Ga.W(x,t).e(x).(a0-o) (8)
All of the model's physical forcing terms (advection, Coriolis effects, etc.)
are represented by F, where a are the model's dependent variables, x are the
14
-------
independent spatial variables, and t is time. The second term on the right-
hand side of (8) is the nudging term, where Ga is a positive nudging factor
which determines the relative magnitude of the term to all the other model
processes in F. The spatial and temporal variation of the term is largely
determined by the four-dimensional weighting function, W. The analysis
quality factor, c, which ranges between 0 and 1, is based on the quality and
density distribution of the data which went into the analysis. The estimate
of the observation for a analyzed to the grid is a0.
The nudging factor Ga is usually selected so that the time scale of the
slowest physical process in the model and the nudging term are similar, and so
that it satisfies the numerical stability criterion, Ga < 1/Dt (Hoke, 1976).
Typical values of Ga are 10~* to 10~3 for meteorological systems. A value of
Ga which is too large will force the model state too strongly toward the
observations. This is undesirable because (1) the ability of the model to
resolve mass-momentum imbalances will be impaired, and (2) the ability of the
model to generate its own mesoscale meteorological structures (e.g., fronts,
squall lines) will be overcome by heavy insertion of the observed analyses.
Such problems arise because the analyses may not resolve these mesoscale
structures or may be contaminated by observational and analysis errors. On
the other hand, if Ga is too small, the observations will have minimal effect
•
on the evolution of the model state, allowing phase and amplitude errors to
grow. Further information on the background of data assimilation can be found
in Stauffer and Seaman (1990).
The assimilation strategy used in the 61 evaluation and assessment cases
was designed and tested by Seaman and Stauffer (1989). Briefly, three-
dimensional analyses of 12-hourly radiosonde data (see Section 2.2) are
assimilated above the model's planetary boundary layer (PEL) for wind,
15
-------
temperature and mixing ratio. The assimilation is performed continuously
(every time step) by interpolation of the enhanced analyses in time between
the two nearest synoptic times. Within the model's PBL, two-dimensional
analyses of surface observations are assimilated for wind (adjusted for the
effects of friction between the surface and the models lowest layer) and
mixing ratio. Above the PBL, mixing ratio is assimilated with Gq - 1 x 10"5
s"1. Elsewhere and for all other variables, Ga - 3 x 10"* s'1.
The four-dimensional weighting function in (8) is given by
W(x,t) - 1 (9)
for this study. The analyses themselves, performed after a quality check of
the observations (see Sec. 2.4), are assumed to be perfect (c - 1). The
analysis confidence factor for the two-dimensional surface analyses is
functionally dependent on the spatial distribution of the surface observations
which contribute to the analyses. Over land, it varies from unity ("perfect")
for grid boxes within 240 km of a surface observation to 0.2 for grid boxes
outside a 480-km radius. Over water, where the data density is much less, the
confidence factor is reduced by a factor of 100 beyond the 480-km radius.
2.4 Quality Assurance of Model Simulations
Because MM4/FDDA is a highly complex mesoscale modeling system, it was
vital to ensure that the model results and the data used for initialization,
boundary conditions and FDDA were thoroughly checked for quality. To this
end, a set of automatic and manual quality assurance procedures were
implemented.
16
-------
First, the objective analysis package performed a quality check on each
observation before it was allowed to affect the analyzed fields. The quality-
checking criteria are situation-adaptive, but are based on the following basic
maximum departures from the first-guess analysis:
sea-level pressure: 5 mb
wind speed: 10 m s"1
temperature: 10 C
There are situations, however, in which the first guess analysis may be
so badly in error that some correct observations may be rejected by the data
quality-checking scheme, although they are crucial for improving the analysis.
For example, pressure observations may be rejected when a rapidly moving
intense storm has a strong pressure gradient either ahead or behind its
center. Although the maximum departure criteria could be increased, this
would not only allow good data to be retained, but might allow many erroneous
observations to be assimilated. Therefore, observations which failed to pass
the automatic checking algorithm were manually checked by an experienced
meteorologist against the analyses. The quality-checking algorithms also
examine radiosonde profiles for vertical consistency and flag any potentially
erroneous soundings for further examination by the meteorologist. At the same
time, the analyses, themselves, were checked for spatial and temporal
consistency by the reviewing scientist. Observations believed to be accurate
were reinserted and used in a subsequent analysis step. Normally, several
hundred observations were checked for each case and many were found to be
correct. This led to significant improvement of the analyses in some areas
for many cases and ensured that the model was using high-quality information.
The 80-km gridded analyses of precipitation used for verification were
generated at EPA using hourly rain gauge data from the National Climate Data
17
-------
Center (NCDC). The methodology was designed by Prof. P. Samson of the
University of Michigan. Precipitation verification is performed only over the
portions of the continental U.S. and extreme southern Canada lying east of the
Rocky Mountains, where high density data area available (Fig. 1). About 1500
hourly observation sites were available in an area encompassing 839 grid
boxes. However, there were typically about 1100-1200 sites reporting
consistently through any given episode. In each grid box, these hourly
precipitation data are summed and averaged over the 12 h verification periods
ending on the synoptic hours 0000 and 1200 UTC. If there are observations
within a box, no data outside the box influences the average value. Any grid
box within the verification region which fails to contain observations for the
period (because of missing data or no rain gauges within that box) may contain
an estimated precipitation based on rainfall analyzed in the immediately
surrounding boxes. At least five of the eight surrounding boxes must contain
observed values, in which case those observations are simply averaged and
applied to the central box. Otherwise, boxes with missing data for a given
12-h period are eliminated from the statistical calculations to avoid
introducing false "zero rain" values and excessive smoothing by interpolating
from boxes further away.
While 1200 observations represents an impressive set of temporally rich
rain data, each grid box in the verification region typically contains only
one or two observations. This raises a question about observation measurement
uncertainty. That is, it is unclear how well one or two observations can
represent the mean rainfall over the area of an 80-km grid box, since rain is
a quantity known to contain much irregularity in its spatial distribution.
This problem is considered in detail in Appendix B and is important to the
interpretation of the rainfall verification statistics.
18
-------
In addition to the steps described above to ensure the quality of the
observations and analyses, plotted fields generated from the model output were
inspected visually by an experienced meteorologist. Each case was checked for
spatial and temporal consistency against the objective analyses.
Representative fields of the analyzed and simulated fields appear in the
volumes of the Case Documentation Report available from EPA. Finally, the
statistical summaries for each case (see Appendix C) were inspected. Only
then were the meteorological data files released for use in RADM.
19
-------
3. VERIFICATION OF METEOROLOGICAL DATA SETS
3.1 Description of Statistical Approach
Statistical summaries for a variety of measures of simulation accuracy
were calculated for each 12-h period of each of the 61 cases in Table 1. The
mean statistics for each case (10 periods for 5 days) are provided in Appendix
C. A definition of each type of statistic is provided in Appendix A and
additional discussion is provided for interpretation of the statistics.
Furthermore, information vital for the interpretation of the rainfall
observations, verifying analyses and statistics appears in Appendix B. The
mean statistics for each case (case means) were then used to calculate grand
averages and seasonal averages to determine the overall performance of the
modeling system (see Section 3.2). The grand average of any quantity is
defined as the annual average of the case-means, based on all the cases.
The statistical verification strategy is based on three categories of
variables. First, we present information about the primitive variables that
are directly assimilated into the model via FDDA. These include winds,
temperatures and moisture (see Section 2.3). Note that these fields are
particularly important to the acid deposition problem because of their direct
influence on transport, atmospheric stability and mixing, and the formation of
clouds and rain. It is anticipated that the errors for these fields should be
"small"; that is, about the size of the errors expected when comparing an
objective analysis to observations.
Second, we will examine the accuracy of dynamic variables that are not
assimilated. These include measures of accuracy for sea-level pressure and
500-mb heights. Although not assimilated directly into the model, these
20
-------
variables are affected by FDDA via the geostrophic and hydrostatic adjustment
processes during the integration of the primitive equations. Since these
fields are not assimilated, it must be expected that it will be difficult to
reduce errors to the same size as those found when comparing objective
analyses to observations. It should be expected, however, that these fields
display improved accuracy compared to typical model solutions without FDDA
(see Seaman and Stauffer, 1989 and Appendix A).
Finally, we will examine the accuracy of the precipitation fields
predicted by the model. This field is crucial to acid deposition modeling
because wet deposition is the most efficient removal mechanism for sulfur and
many other compounds. Unfortunately, there is no single statistical quantity
which adequately represents the accuracy of a precipitation forecast.
Therefore, we use seven separate measures which represent such aspects as the
intensity, areal coverage, and timing of the rain.
To establish the credibility of the MM4/FDDA data sets, two points must
be addressed: model-vs.-model skill and model-vs.-analysis skill. First, it
must be shown that the model performs well compared to other well-documented
high-quality numerical models. This step is important for demonstrating that
MM4/FDDA is a suitable choice among available numerical models for generating
meteorological data sets for use in RADM. We will do this by showing some
comparisons between MM4/FDDA precipitation statistics and similar values from
the National Meteorological Center's (NMC) operational Nested Grid Model (NGM)
and the experimental NMC ETA model. Additional information about the typical
performance of some "standard" models and comparisons with results from
MM4/FDDA appear in Appendix A" and in Seaman and Stauffer (1989).
Second, it is vital to compare the skill of the numerical data sets with
that of objective analyses based on the observed data. Ideally, the magnitude
21
-------
of the model errors should be about the same as found in objectively analyzed
fields based on observations. When this condition is attained, it may be
argued that the mesoscale details and intervariable consistency among the
model's variable fields (neither of which can be well-resolved from the
synoptic-scale observations) make the model-generated data sets superior.to
analyses for use in an air-chemistry model such as RADM.
A comparison of this sort can be made rather easily for the primitive
variables. For precipitation fields, however, there is the further
consideration of observation representativeness. That is, an analysis of
rainfall observations may contain significant errors due to the spatial
variability of the rain which is not resolved by the rain gauge network.
Determination of the rainfall observation representativeness uncertainty is,
then, a vital factor in defining the comparative accuracy of the model-
predicted and analyzed precipitation fields. In addition to the discussion
provided in Section 3.2.3, additional evidence on this subject is given in
Appendix B.
It is important to emphasize that we calculate most statistical measures
of rainfall over periods of 12 hours. This focuses attention on the mesoscale
accuracy of the fields. In contrast, the more familiar precipitation-accuracy
statistics published for operational models generally are calculated over
periods of 24 hours, and the 24-h statistics are then be averaged together for
a month. This operational approach is ideal for examining the synoptic-scale
accuracy of rainfall, but often fails to account for the mesoscale structure.
While important mesoscale features are found in fields such as pressure, wind
and temperature, the rainfall is especially known to have great spatial and
temporal variability. The importance of obtaining mesoscale accuracy in the
acid deposition fields justifies the mesoscale verification approach used
22
-------
here, but it complicates the comparison with standard published statistical
scores for precipitation. This problem is addressed in Section 3.2.3 and also
in Appendix A.
3.2 Statistical Results of Model Verification
3.2.1 Assimilated Primitive Variables
We begin the model verification by examining statistical results for
several representative primitive variables assimilated directly into the model
simulations. Tables 2 and 3 summarize these statistics, along with statistics
for some unassimilated variables. The mean RMS Errors for 500-mb wind are
also shown in Figure 2. The grand-average value of 1.21 m s"1 is comparable
to values expected for obj ective analyses of observations. The figure shows
some seasonal trend with slightly larger RMS errors in winter, when the zonal
mean wind speed in the northern hemisphere is relatively large. The standard
deviations are small compared to the RMS errors, indicating that there is
little case-to-case variability in the magnitude of the errors. A scatter
plot of the individual case-mean RMS Errors confirms this seasonal trend and
shows that there is a relatively small range of RMS error in the wind for any
particular time of year (Figure 3). The grand-average case mean-errors for
500-mb wind (Table 3) is 0.97 m s"1 (positive), indicating that the model has
a slight bias to high speeds at this level. However, the FDDA approach
clearly keeps this bias very small.
Figure 4 shows the mean RMS Errors for 850-mb temperature. As for the
wind statistics, the errors in the temperature simulations are very small and
are about the same as would be expected in objective analyses. The RMS errors
tended to be greatest in winter and smallest in the autumn. Table 3 indicates
23
-------
Table 2. MM4/FDDA seasonally and annually averaged values of
case RMS errors for selected meteorological variables,
based on 61 cases.
Variable Winter Spring Summer Autumn
Type
RMS-SLP 2.02 1.69 1.69 1.68
(mb)
RMS-sfq 0.64 0.85 1.23 Q.99
(gAg)*
RMS-T 0.50 0.42 0.42 0.36
850 mb
(C)
RMS-V 1.48 1.20 1.05 1.24
500 mb
(m s'1)
RMS-Ht. 13.67 11.19 10.60 11.55
500 mb
(m)
Annual
1.72
0.99
0.40
1.21
11.46
sfq. - surface mixing ratio
24
-------
Table 3. MM4/FDDA seasonally and annually averaged values of
case mean-errors for selected meteorological variables,
based on 61 cases.
Variable Winter Spring Summer Autumn Annual
Type
Mean Err.
SLP 0.21 0.02 -0.39 0.01 -0.08
(mb)
Mean Err.
Sfc.Mix.Rat. 0.22 0.42 0.59 0.44 0.45
(gAg)
Mean Err.
T-850 -0.04 -0.12 -0.14 -0.08 -0.10
(C)
Mean Err.
V-500 1.21 0.98 0.84 0.98 0.97
(m s'1)
Mean Err.
Ht.-500 2.33 0.76 -0.77 1.72 0.92
(m)
25
-------
<9
E
Q
i
JO
E
o
a
10
en
S
cc
DEC-FEB MAR-MAY JUN-AUG SEP-NOV
ANNUAL
Figure 2. MM4/FDDA means (hatched) and standard
deviations (cross-hatched) of seasonal and
annual RMS Error of 500-mb wind (based on
vector wind difference, calculated for 61
cases). Scores are shown at top .
26
-------
f"J
1.5
RMS
Error "-0
(ms-')
0.5
nn
i i i i i i i i i i ; i
•
•• • *
* • • 9 •
• * .*« \ • *• **%- *
1 1 1 1 1 1 1 1 1 1 1 1
t> 30 60 90 120
150 I8O 210
Julian Day
240 270 300 33O 36O
Figure 3. MM4/FDDA case-mean RMS Errors for 500-mb wind
(based on vector wind difference) versus Julian
Day for 61 cases
-------
Q.
5
ai
o
in
CO
(0
0.6-
0.5-
0.4-
0.3-
0.2-
0.1 -
0.0-
0.50
0.42
0.42
0.40
0.04
0.35
0.05
0.07
DEC-FEB MAR-MAY JUN-AUG SEP-NOV
ANNUAL
Figure k. MM4/FDDA means (hatched) and standard
deviations (cross-hatched) of seasonal and
annual RMS Errors of 850-mb temperature, based
on 61 cases. Scores are shown at top .
28
-------
that there is only a very slight bias toward cold temperatures at 850-mb
(grand-average case mean-error of -0.1 C).
An examination of the statistics for surface mixing ratio shows a grand-
average RMS Error of 0.99 gAg. with a marked tendency toward higher errors in
the summer (Figure 5). This trend results from the greater capacity of warm
air to hold water vapor before reaching saturation. Considering the many
local effects and high regional variability among surface moisture
observations, this is a very reasonable RMS error. Table 3 shows that there
is a slight bias in the surface mixing ratio (grand average is 0.45 g/kg,
indicating that the lowest level is too moist). It is possible that the soil
moisture availabilities parameterized in the model's surface look-up tables
(see Sec. 2) may average slightly too high.
Clearly, none of the Mean or RMS Errors for the sampled assimilated
variables is large, compared to typical values expected for objective
analyses. While the model does not return fields that are identical to the
analyses (which, of course, is by design), it appears to have only very small
systematic and random error characteristics.
3.2.2 Non-assimilated Primitive Variables
Through the geostrophic and hydrostatic adjustment processes active
during the integration of the primitive equations, the nudging of the model's
wind, temperature and moisture will affect the sea-level pressure and height
solutions. Tables 2 and 3 also show seasonally and annually averaged case RMS
and Mean-Errors, respectively, for these quantities. The tables show, for
example, that the grand-average of mean sea-level pressure errors is only -
0.08 mb, or a very slight bias toward low pressure. The similar grand-average
mean error for 500-mb height is 0.92 m. The RMS Errors for these two
29
-------
1.5
1.23
3
g
cc
o
z
X
o
u.
w
1.0-
0.64
0.5-
0.85
0.0
0.07
0.99
0.17
:
/ t t
, N \
X t t
0.99
DEC-FEB MAR-MAY JUN-AUG SEP-NOV
ANNUAL
Figure 5. MM4/FDDA means (hatched) and standard
deviations (cross-hatched) of seasonal and
annual RMS Errors of surface mixing ratio,
based on 61 cases. Scores are shown at top
30
-------
fields, 1.72 nb and 11.46 ra respectively, indicate that there is some local
variability about the means, but these do not become large. For example, the
Navy NORAPS forecast model produces 500-mb height RMS Errors on order of about
80 m after 5 days (Dr. John B. Hovermale, personal communication).
Another important measure of the skill of model pressure and height
fields is the St score. This statistic measures the skill in simulating the
horizontal gradients of a scalar field (see Appendix A for details). Figure 6
demonstrates that the grand-average mesoscale sea-level pressure Sx Score is
42 (a score of 30 is considered to demonstrate "perfect" skill). This
compares favorably with the typical model value of 55 estimated for mesoscale
Sj (Appendix A). Figure 6 and the accompanying scatter plot (Figure 7) also
show a clear seasonal trend for higher summertime errors in the sea-level
pressure St Score. This is expected because the score tends to give high
values in situations with very weak pressure gradients, which are common in
summer.
Similarly, Figure 8 shows that the grand-average mesoscale S]^ Score for
500-mb height is 24.9. Fawcett (1977) reported that a value of 20 represents
an essentially "perfect" simulation. This value also compares favorably with
the estimated typical model value of about 30 (Appendix A). As for sea-level
pressure, there is a clear tendency for poorer scores in summer when the
height gradients are usually weak. Because the average spacing of radiosonde
data is about 400 km, there is probably little opportunity for FDDA to improve
the height gradient solutions.
The low values of the errors in fields of the non-assimilated variables
indicates that the non-linear adjustment processes are effectively
communicating information from the assimilated variable fields. This is an
important indication that intervariable consistency is being established in
31
-------
60.0
40.2
HI
cc
O
O
V)
a.
_i
(0
48.7
39.6
3.3
42.0
5.2
,
XXX
, \ \ '
XXX
DEC-FEB MAR-MAY JUN-AUG SEP-NOV
ANNUAL
Figure 6. MM4/FDDA means (hatched) and standard
deviations (cross-hatched) of seasonal and
annual S^ Score for sea-level pressure
(calculated for a standard distance of 320 km)
based on 61 cases. Scores are shown at top and
dashed line (S^ - 30) indicates an essentially
"perfect" simulation
32
-------
u>
to
60
50
p 30
'SL
20
10
Q
!•
^
•-—•
1
1
1
1
0 30 60 90 120
150 180 210
Julian Day
240 270 300 330 360
Figure 7. MM4/FDDA case-mean S^ Scores for sea-level
pressure (calculated for a standard distance of
320 km) versus Julian Day, based on 61 cases .
-------
40.0
UJ
cc
o
o
05
T~
w
J-
g
iu
x
o
o
m
30.0-
20.0 •••
10.0-
18.5
0.0
30.9
23.5
1 .8
3.3
23.5
24.9
3.5
5.4
t / t
N X V
t t t
DEC-FEB MAR-MAY JUN-AUG SEP-NOV
ANNUAL
Figure 8. MM4/FDDA means (hatched) and standard
deviations (cross-hatched) of seasonal and
annual S^ Score for 500-mb height (calculated
for a standard distance of 320 km) based on 61
cases. Scores are shown at top and dashed line
(S^ - 20) indicates an essentially "perfect"
simulation
34
-------
the solutions, which was an important goal for the numerically generated
meteorological data sets.
3.2.3 Precipitation Scores
Finally, model verification is performed by examining a set of
precipitation skill scores. As discussed in Section 3.1, it is important to
establish that the results of MM4/FDDA are suitable compared to both
precipitation analyses and simulation results of alternative models. That is,
we seek to show that MM4/FDDA is a good choice among available models for
producing numerical data sets and that those data sets have about the same
synoptic-scale accuracy as analyses of observations, while providing the
mesoscale detail and intervariable consistency characteristic of the model
solutions.
A number of characteristic scores are summarized in Table 4 (definitions
and interpretive discussions are provided in Appendix A). Perhaps the most
exacting of these statistics is the threat score (TS), which measures the
accuracy of precipitation area and position. Figure 9 presents the mean
seasonal and grand-average TS (for the 0.25cm/12h threshold), while Figure 10
shows a scatter plot versus date (1.0 is perfect). The grand-average TS
(0.25cm/12h) is 0.37. The figures show considerable scatter (e.g., the
standard deviation about the grand average is 0.11) and a clear tendency for
lower (worse) TS values during the warmest months (summer average is 0.27).
The seasonal dependency results from the predominance of small-scale
convection in summer rainfall and a few relatively dry autumn and winter cases
for which only a few points received substantial rainfall.
Eight cases have a TS more than one standard deviation below (worse
the seasonal average: cases 4, 5 and 32 in summer, case 1 in the
35
-------
Table 4. MM4/FDDA seasonally and annually averaged values for
selected precipitation verification statistics, based
on 60 cases. All precipitation amounts are for 12-h
periods.
Statistical
Score
Threat Score
at 0.25 cm
Threat Score
at 0.64 cm
Bias Score
at 0.25 cm
Bias Score
at 0 . 64 cm
Mean Error
Precip. (cm)
Mean Abs. Err.
Precip. (cm)
Large Err.
Winter
0.36
0.41
1.71
0.79
-0.05
0.22
10.75
Spring
0.40
0.27
1.39
1.39
0.04
0.28
18.80
Summer
0.27
0.17
1.60
1.81
0.07
0.32
21.88
Autumn
0.42
0.33
1.41
1.16
0.00
0.32
20.46
Annual
Avg. Std.Dev.
0.37 0.11
0.28 0.13
1.50 0.67
1.34 0.74
0.02
0.30 0.08
19.31
Count (1 cm)
36
-------
o
in
CM
DEC-FEB MAR-MAY JUN-AUG SEP-NOV
ANNUAL
Figure 9. MM4/FDDA mean seasonal and annual Threat Scores
(0.25 cm/12 h threshold) and standard deviation
from mean threat scores, based on 60 cases
37
-------
10
oo
1.00
0.75
TS
0.50
0.25
O.OQ
T T
'0 30 60 90 120 150 180 210 240 270 300 330 360
Julian Day
Figure 10. MM4/FDDA Threat Scores (0.25-cra threshold)
versus Julian Day for 60 cases. Seasonal mean
TS (solid) and seasonal standard deviations
from the mean (dashed) are shown
-------
spring, cases 23, 28 and 52 in autumn and case 44 in winter. Of these, the
summer and winter cases are the only ones with a TS below 0.20. These cases
in particular should be used most carefully in RADM.
Figure 11 shows TS (0.64cm/12h) and displays the normal tendency for
lower scores (grand average is 0.28) as the threshold value is increased.
This occurs because fewer grid cells contain higher amounts of precipitation,
and the TS's dependency on position accuracy becomes progressively more
dominant in the verification. Again, the seasonal dependence is found, with
the summer average equal to 0.17 at the 0.64cm/12h threshold.
The model's Bias Scores (BS) are displayed in Figures 12-14 and again in
Table 4. The Bias Score (1.0 is perfect) measures the accuracy of
precipitation area, but does not account for errors in position (see Appendix
A). Figure 12 shows that the grand-average BS (0.25cm/12h) is 1.50. Figure
13 indicates that there is considerable case-to-case variation in the Bias
Scores. Particularly high scores generally occur in light-rainfall
situations, for which over-prediction of the observed area by a few points can
have a great impact on the case-mean statistic. Since there are a number of
relatively "dry" cases among the few winter cases in the set, the standard
deviation in that season appears particularly high (Figure 12).
As with most numerical models, Figure 14 shows that there is a tendency
for the annual-average BS to decline for larger amounts of precipitation
despite the large summertime BS (grand-average BS for 0.64cm/12h is 1.34).
There is less seasonal dependency apparent for the light-rain threshold
(Figure 12). Thus, it appears that the model has a tendency to produce too
great an area of heavier rain (0.64cm/12h) in the summertime. The cause lies
both in the small spatial scale of observed summer convection and in the
39
-------
E
u
CD
DEC-FEB MAR-MAY JUN-AUG SEP-NOV
ANNUAL
Figure 11. MM4/FDDA mean seasonal and annual Threat Scores
(0.64 cm/12 h threshold) and standard deviation
from mean threat scores, based on 60 cases
40
-------
2.0
1.71
E
o
CM
•
O,
CO
03
1.0-
1.12
0.0
1.60
1.39
1.41
0.21
1.50
DEC-FEB MAR-MAY JUN-AUG SEP-NOV
ANNUAL
Figure 12. MM4/FDDA mean seasonal and annual Bias Scores
(0.25 cm/12 h threshold) and standard deviation
from mean threat scores, based on 60 cases
41
-------
ro
3.0
2.0
BS
1.0
O.Q
j _ LI
i i
j _ i
i _ i _ i _ i
0 30760 90 120 150 180 210 240 270 300 330 360
Julian Day
Figure 13. MM4/FDDA Bias Scores (0.25-cra threshold) versus
Julian Day for 60 cases. Seasonal mean BS
(curved solid) and seasonal standard deviations
from the mean (dashed) are shown
-------
u
-------
parameterization assumptions in the Kuo-type convective scheme used in the
model.
Cases with a BS above 2.0 or below 0.50 (an arbitrary choice) can be
considered to have potentially serious errors and results from RADM should be
examined carefully. These categories include cases 5 and 42 in summer
(convective) and cases 6, 24, 57 and 58 in autumn and winter (dry). Cases 5
and 24 also have low values of the Threat Score.
Another picture of the seasonal tendency in the BS can be obtained by
forming the 61 cases into clusters of approximately 30 days each. This was
done by arranging the 61 cases by date and defining clusters of 6 five-day
cases. By averaging the BS over 30 days, the statistics are displayed in a
manner similar to the monthly means reported by NMC (see Appendix A). The
results for the 0.05cm/12h threshold are shown in Figure 15. The grand-
average BS (0.05cm/12h), calculated in this manner, is 1.29, while only
Cluster 4 (June-July) has a BS above 1.4.
Figure 16 presents the MM4/FDDA grand-average (annual average) BS for a
range of rain rates, based on 24 of the cases taken from all seasons. The BS
ranges from about 1.44 at 0.25cm/12h to about 0.70 for 2.50cm/12h (roughly one
inch). While the tendency for lower BS with higher rainfall is clear, the
average BS in the figure remains close to 1.0.
• ~ ~~ " - - ~
Another way of understanding the performance of MM4/FDDA is to compare
its BS and TS to those reported for the NMC models. This step is important to
establish the model as a credible choice among the available high-quality
models which could be used to generate numerical data sets for RADM.
Fortunately, Cases 51 - 56 (see Table 1) cover the month of November 1988, for
which Black and Mesinger (1989) have published statistics for the NGM and the
experimental ETA model (80-km resolution, 16 layers). The ETA model is being
44
-------
2.0
u
in
o
CQ
1.0-
0.0
1.05
1.20
1.34
1.79
1.39
1.26 1-31
1.3S
1.19
1.02
1.29
Cluster
Month
1234567
DEC- FEB- APR- JUN- JUL- AUG-
JAN APR MAY JUL AUG SEP
SEP SEP
1 0
NOV
ANNUAL
Figure 15. Thirty-day mean Bias Scores for MM4/FDDA at the
0.05 cm/12 h threshold. Each period is
approximately 30 days long and is based on a
cluster of 6 cases (usually 5 days each)
arranged seasonally. Months of cases in each
cluster are shown below, and the annual grand
-average B.S. is shown to the right. The exact
BS values are given above each cluster .
45
-------
2.0
1.5
BS
1.0
0.5
0.0
0.0 0.5
I
I
1.0 . 1.5
Rain
(cm/!2h)
2.0 2.5
Figure 16. MM4/FDDA mean annual Bias Score versus
precipitation rate (cm/12h), based on 24 cases.
46
-------
developed as the probable successor to the NGM operational model. Figures 17a
and 18a show the intercomparison of the three models for that month over a
range of precipitation rates using 24-h rainfall totals.
Figure 17a shows that the MM4/FDDA model has a BS similar to or better
than the NGM for virtually the entire range of precipitation rates, especially
those greater than 1.0cm/12h. That is, the BS tends to be closer to 1.0 for
the MM4/FODA results (the Perm State/NCAR model's post-processing software
does not calculate statistics for thresholds above 2.5 cm). The experimental
ETA model was superior to the NGM for rain rates above 1.5 cm/12h, but it
still did not perform as well as MM4/FDDA. Thus, the MM4/FDDA showed greater
skill well at simulating 24-h precipitation areas compared to other prominent
limited area models having similar resolution.
Figure 18a shows that MM4/FDDA clearly produced much better threat scores
than either of the two NMC models. The advantage is particularly great for
heavier rainfall amounts when compared to the NGM. Certainly, MM4/FDDA has an
advantage because FDDA limits the growth of phase errors that lead to low
threat scores in the two forecast models. On the other hand, those models are
run for only 48 h, rather than 120 h. Thus, although these caveats must be
remembered when interpreting the model scores, it appears that the use of FDDA
in the MM4 model has clearly led to a substantial improvement in the
precipitation simulations compared to other advanced models without FDDA.
Figures 17b and 18b compare the BS and TS of MM4/FDDA (November 1988)
over a range of precipitation intensities, when rainfall is summed for 12-h
versus 24-h periods (the summation period is referred to as N hours). These
figures demonstrate that the calculation of the precipitation statistics in
12-h periods, to focus on the mesoscale accuracy of the rainfall, produces
scores which are comparatively "worse" (i.e., TS is lower, BS is further from
47
-------
2.0
T
T
T
1.5
— MM4/FDDA,24h
— NMC NGM.24H
- NMC ETA,24h
BS
1.0
0.5
O.Q
I
1
0.0
1.0
2.0 3.0
Rain
(cm/24h)
4.0
5.0
Figure 17. Comparison of mean Bias Scores versus precip-
itation rate based on cases for November 1988,
using MM4/FDDA and the NMC NGM and ETA (80-km
resolution) models. The last two are based on
48-h forecasts (after Black and Mesinger,
1989). (a) Comparison for 24-h precipitation
totals using the three models, (b) Comparison
for N - 12-h versus N - 24-h precipitation
totals using MM4/FDDA .
48
-------
2.0
T
T
1.5
MM4/FDDA N*24h
MM4/FDDA N* I2h
BS
1.0
0.5
O.Q
0.0 0.5
1.0 1.5
Rain
(cm/Nh)
2.0 2.5
Figure 17(b)
49
-------
1.0
MM4/FDDA,24h
NMC NGM,24h
NMC ETA , 24h
TS
0.5
O.Q
0.0
I
I
1.0
2.0 3.0
Rain
(cm/24h)
4.0
5.0
Figure 18. Comparison of mean Threat Scores versus precip
-itation rate based on cases for November 1988,
using MM4/FDDA and the NMC NGM and ETA (80-km
resolution) models. The last two are based on
48-h forecasts (after Black and Mesinger, 1989)
(a) Comparison for 24-h precipitation totals
using the three models, (b) Comparison for
N - 12-h versus N - 24-h precipitation totals
using MM4/FDDA .
50
-------
1.0
MM4/FDDA N = 24h
MM4/FDDA N= I2h
TS
0.5
0.0
0.0
0.5
I
I
1.0 1.5
Rain
(cm/Nh)
2.0
2.5
Figure 18 (b)
51
-------
1.0) than when a 24-h summation is used. This reflects the greater difficulty
encountered in accurately simulating the position and timing of smaller
(mesoscale) areas of rainfall due to the extreme local variability of this
field. A similar effect would be expected for any mesoscale or synoptic-scale
numerical model.
Nevertheless, Table 4 shows that the grand-average of the mean errors in
12-h precipitation totals is only 0.02 cm. This demonstrates that, for all
cases, MM4/FDDA produces very nearly the correct volume of rainfall.
Examination of the seasonal trend indicates that the greatest departures occur
in winter, when the model tends to under-predict rainfall (seasonal average of
-0.05 cm) and in the summer, when the model over-predicts rain (seasonal
average of 0.07 cm). The large error count (LEG) indicates that about 19 out
of the 839 grid boxes can be expected to have rainfall errors over 1.0 cm.
This last score is used to give an indication of the frequency of large
precipitation errors within the verification domain (see Appendix A).
Although model intercomparison is very helpful for demonstrating the
skill of MM4/FDDA, it remains difficult to establish how accurate a numerical
model's precipitation simulation should be. While both the TS and BS have a
theoretically perfect value at 1.0, this figure is actually unattainable
except by chance. This is discussed more thoroughly in Appendix B by
examining the representativeness uncertainty of observations used to prepare
objective analyses of the rainfall. While these analyses are generally
assumed to be perfect for the purpose of calculating the verification scores,
they cannot, in fact, represent the true rainfall.
Figure 19 shows that the MM4/FDDA grand-average Mean Absolute Error (MAE)
is 0.30 cm for 12-hourly precipitation. Additional calculation found that the
average amount of simulated rainfall per 12 h occurring in grid boxes
52
-------
0.4
0.32
0.32
HI
0.3
0.2-
0.1 -
0.28
0.22
0.0
0.08
0.06
0.04
0.30
DEC-FEB MAR-MAY JUN-AUG SEP-NOV
ANNUAL
Figure 19. MM4/FDDA mean seasonal and annual Mean Absolute
Errors (MAE) of 12-h precipitation and standard
deviation from MAE, based on 60 cases
-------
simulated to have measurable rain (defined here as 0.05 cm) was 0.45 cm. The
MAE gives the most likely difference that occurs between the model
precipitation and the analyzed precipitation.
However, Figure B2.b shows that when a. rainfall observation of 0.45 cm is
applied to represent an 80-km square, the mean absolute deviation is 0.32 cm.
That is, if the true mean rainfall in the 80-km box is 0.45 cm, a single
observation taken within the box will likely misrepresent the rainfall by 0.32
cm. This is the observation representativeness uncertainty. Since there are
typically only 1100-1200 observations available to represent the 839
verification grid boxes, the analyzed value in each box is typically based on
one or two observations. The boxes shaded in Figure 1 have at least 10
climatological rain gauges reporting once in 24 h and do not imply that there
are many hourly reporting gauges needed for verification of MM4/FDDA.
Notice that the uncertainty of a typical rainfall observation used to
represent the average rain amount in an 80-km grid box is actually greater
than the MAE between the model and the analyzed precipitation used for
verification. This suggests that the difference between the model and the
verifying analysis (which is most often based on only one or two observations
per grid box) is about as likely to be due to representativeness uncertainty
as it is to model error, except in the few boxes with more observations.
Therefore, by this measure of rainfall accuracy, the model's precipitation
fields in most areas should be about as accurate, on average and for many
cases, as the analyses based on the available rain gauge network. This
suggests that the model's precipitation fields are reasonably accurate and are
fully suitable for use in the RADM chemistry model, since they also are based
on intervariable consistency between the rainfall and the other model
variables which contribute to the rain (e.g., vertical motion, convergence,
wind, moisture, stability, etc.).
54
-------
4. SUMMARY AND CONCLUSIONS
The PSU/NCAR model with FDDA (MM4/FDDA) has been used to simulate 61
cases (usually for 5 days each). Statistical verification and evaluation has
been performed and documented in this report.
Summarizing the model performance, it is noted that the model produced
very accurate results for the assimilated primitive variables: wind,
temperature and mixing ratio. These variables were evaluated by calculating
Mean Errors and RMS Errors between the simulation and actual observations.
The grand-average RMS Errors for 500-mb wind and 850-mb temperature, for
example, were 1.21 m s"1 and 0.40 C, respectively. For surface mixing ratio,
the grand-average RMS Error was 0.99 g kg'1. These errors are about the size
of those to be expected for objective analyses of the observations.
Generally, the seasonal trends found in the case results are those explained
by the natural variability in annual weather patterns. For example wind
errors grow slightly in winter, and errors in surface mixing ratio are
greatest in summer. The scatter of individual case statistics about the grand
average and seasonal averages is fairly small. These results demonstrate that
the model is performing its designed role as a sophisticated analysis tool.
Evaluation of the primitive variables not assimilated directly into the
model integration demonstrated that the model's geostrophic and hydrostatic
adjustment processes effectively communicate information to these fields from
the assimilated variables (a process is called "induction"). Sx Scores, Mean
Errors and RMS Errors were examined for sea-level pressure and SOO-mb height.
The results were generally superior to similar statistics calculated for
operational forecast models (e.g., the grand-average mesoscale SLP St Score
was 42.0 for MM4/FDDA, while LFM statistics calculated for larger scales
55
-------
generally average about 50.0). Again, the seasonal trend of these statistics
were as expected, with St Scores (a measurement of the accuracy of the
simulated gradients) being higher than average in summer, and RMS and Mean
Errors being higher than average in winter. The scatter of individual case
statistics about the mean seasonal trend was generally found to be rather
small.
Most importantly, the precipitation statistical skill, based on a variety
of scores, was found to be quite impressive. The grand-average Mean Error of
precipitation was only 0.02 cm/12h, indicating that the model produces total
rainfall volumes that are very accurate. Threat Scores and Bias Scores
(calculated using 12-h rainfall totals to focus on the skill of mesoscale
precipitation patterns) were found to be favorable compared to typical
mesoscale model scores. Particularly, the improvement of the TS for heavier
rainfall amounts, compared to the NGM and ETA models, suggests that FDDA is
producing important improvements in rainfall accuracy.
Additionally, the grand-average Mean Absolute Error was found to be 0.30
cm. This quantity measures the average departure of the simulated rain from
the analyzed rainfall. Extensive analysis of actual rainfall data
demonstrated that the verifying analyses, generally based on one or two
observations per 80-km grid square, contain a considerable amount of
uncertainty in most regions due to the representativeness that can be assigned
to the observations. These calculations suggested that the rain observation
representativeness uncertainty was about 0.32 cm for the average rainfall
total of 0.45 cm, or about as large as the MAE. Thus, on average for most of
the verification domain and for many cases, the differences between the
simulated and analyzed rainfall are probably due as much to observation
uncertainty as to model error. Since the model fields contain an internal
56
-------
consistency among the variable fields, including the convergence and vertical
motion fields that are very difficult or impossible to resolve from available
observations, this demonstrates a very important degree of success and skill
in the numerical data sets. On the other hand, it is clearly possible that
the analyzed precipitation is more accurate than the simulated rainfall for
certain well-observed grid boxes or for certain cases. This should not affect
the overall suitability of the simulated rainfall for use in RADM.
Model precipitation skill scores consistently show larger errors for
summertime precipitation. This is understood as the natural result of the
sub-grid scale convection which dominates warm-season rainfall. Additional
cases of poor statistical verification occur in several "dry" autumn and
winter cases for which relatively few points have substantial rainfall. Bias
Scores and Threat Scores typically show greater case-to-case variability than
scores relating to other variables, but this is a consequence of the extreme
spatial variability of rainfall and the complex interaction of the many
physical processes contributing to it. Despite these difficulties, only a few
cases have Threat Scores for the 0.25 cm/12h threshold that are below 0.20 or
Bias Scores above 2.0 (see Section 3.2.3). Numerical results from these cases
should be used cautiously in RADM. Otherwise, the rainfall simulations appear
to be generally suitable for use in the air chemistry model.
In conclusion, careful examination of the mean and case-average
statistics shows that the numerical data sets contain a high degree of skill
for all fields. Most variability among the cases is easily understood in
terms of predictable seasonal trends and normal variability among cases.
Cases having potentially serious problems in the rainfall simulations are
relatively rare and can be readily identified. These occurrences are
attributable to either situations in which isolated local convection is
dominant or those in which rainfall is nearly absent.
57
-------
REFERENCES
Anthes, R.A., 1977: A cumulus parameterization scheme utilizing
a one-dimensional cloud model. Mon. Wea. Rev.. 107. 963
-984.
Anthes, R.A., 1983: Regional models of the atmosphere in middle
latitudes. Mon. Wea. Rev.. 111. 1306- 1335.
Anthes, R.A., E.Y. Hsie, and Y.H. Kuo, 1987: Description of the
Peim State/NCAR mesoscale model version 4 (MM4), NCAR/TN
-282+STR, NCAR, Boulder, CO, 66 pp.
Anthes, R.A. and T.T. Warner, 1978: Development of hydrodynamic
models suitable for air pollution and other
mesometeorological studies, Mon. Wea. Rev.. 106. 1045-1078.
Benjamin, S.G. and N.L. Seaman, 1985: A simple scheme for
objective analysis in curved flow. Mon. Wea. Rev. . 113.
1184-1198.
Black, T.L. and F. Mesinger, 1989: Forecast performance of NMC's
ETA Coordinate Regional Model. 12th Conf. on Wea. Anal, and
For., Oct. 2-6, 1989, Monterey, CA, 551-555.
Davies, H.C., 1976: A lateral boundary formulation for
multi-level prediction models. Quart. J. Roy. Meteor.
Soc.. 103. 225-245.
Fawcett, E.B., 1977: Current capabilities in prediction at the
National Weather Service's National Meteorological Center.
Bull. Amer. Meteor. Soc.. 5.8, 143-149.
58
-------
Forbes, G.S., R.A. Anthes, and D.W. Thomson, 1987: Synoptic and
mesoscale aspects of an Appalachian ice storm associated
with cold-air damming. Mon. Wea. Rev.. 115. 564-591.
Hoke, J.E., 1976: Initialization of models for numerical weather
prediction by a dynamic initialization technique. Ph.D.
dissertation, The Pennsylvania State University, Dept. of
Meteor., University Park, PA, 202 pp.
Kuo, Y.H. and R.A. Anthes, 1984: Mesoscale budgets of heat and
moisture in a convective system over the central United
States. Mon. Wea. Rev.. 112. 1482-1497.
Seaman, H.L. and D.R. Stauffer, 1989: Development of Four
-Dimensional Data Assimilation for Regional Dynamic
Modeling Studies. Final Report to U.S. EPA, Contract CR
-814068-01-0, The Pennsylvania State University, University
Park, PA, 102 pp.
Stauffer, D.R. and N.L. Seaman, 1990: Use of four-dimensional
data assimilation in a limited-area mesoscale model. Part
I: Experiments with synoptic-scale data. Mon. Wea. Rev. .
118. accepted for publication, 37 pp.
Zhang, D.L. and R.A. Anthes, 1982: A high-resolution model of
the planetary boundary layer -- sensitivity tests and
comparisons with SESAME-79 data. J. Appl. Meteor.. 21.
1594-1609.
59
-------
APPENDIX A
Definition of Statistical Quantities
A.I RMS Errors
The RMS errors for wind, temperature, and height were computed at
standard radiosonde locations at the following pressure levels: 850, 700,
500, and 300 mb. The RMS for a scalar, a, is calculated as
N
(Al)
RMS - ^ (a0-a)2/NJ
Verification of the wind is accomplished by computing the RMS error of the
vector wind difference (VWD) between the simulated and observed values, where
F I1/2
VWD - I (xio-u)2 + (v0-v)2 I (A2)
This RMS wind error takes .into account both speed and direction errors.
A. 2 St Scores
The St scores are calculated for SLP and 500 -mb height, as described by
Anthes (1983),
Sx- 100 legl/IGil (A3)
where eG is the error of the forecast pressure or height difference (gradient
times distance between two points) , and GL is the maximum of either the
observed or forecast difference between the points. The distance used in the
calculation is generally chosen to be the scale of the separation of the
observations. For this study, the standard distance was defined as 320 km.
60
-------
The St score measures the skill in simulating the horizontal gradients of
a scalar field. Anthes (1983) reports typical Sx scores for sea-level
pressure in the National Meteorological Center's (NMC) Limited Fine-Mesh (LFM)
model to be about 45, with a score of 30 being "essentially perfect" and a
score of 80 having no skill. These LFM scores are calculated using a 5°-
latitude and 10°-longitude grid, or about at the 550-km scale (Dennis Deaven,
NMC, personal communication). Typical Sx scores for 500-mb heights are given
as 25, and Fawcett (1977) estimated a score of 20 as representing a virtually
"perfect" simulation. However, these scores are sensitive to a number of
factors, including seasonal variations of observed gradients and the length
scale used in the calculations. Scores tend to be higher £or summertime
cases, when the actual pressure and height gradients are weak, and for shorter
length scales used for fine grids capable of resolving finer scales of motion.
For the 80-km resolution used for MM4/FDDA in this study and for verification
at the 320-km scale, typical annually averaged St scores (without FDDA) are
estimated to be about 55 for sea-level pressure and about 30 for 500-mb
height.
A.3 Precipitation Threat Scores
The threat score (TS) measures the model skill in simulating the area and
placement (related to timing) of precipitation exceeding a designated
threshold amount. TS is an exacting measure of skill because it includes a
penalizing factor when precipitation is simulated outside the observed area.
TS is computed from
TS - N1/(N1 + N2 + N3) (A4)
61
-------
where Nl is the number of grid points for which an event (precipitation
exceeding the given threshold amount) is correctly simulated, N2 is the number
of grid points for which the event is simulated but not observed, and N3 is
the number of grid points for which the event is observed but not simulated.
It is important to note, however, that TS, which ranges from zero to one (1.0
representing a perfect score), is not a function of the number of grid points
where the event is neither simulated nor observed (that is, correctly
simulating the non-occurrence of an event does not enter into the calculation
of TS).
Figure Al demonstrates the determination of the threat score for a
simulated area of precipitation having perfectly accurate size, shape and
rainfall totals, but which has a phase error so that it covers one-half of an
identical observed precipitation area. This simulated precipitation field
will have a verifying TS of 0.333, about an average value, despite perfect
skill in several key aspects of accuracy. When the scale of the rain area is
small, as for mesoscale frontal rain bands or thunderstorm clusters, the
possibility of a small phase error leading to a low TS becomes increasingly
large. Similarly, when TS is calculated for rainfall accumulated for short
periods of time, greater mesoscale structure appears in the data and threat
scores will tend to be lower. Most published values of threat score
calculated for operational numerical models are calculated for periods of
24 h. This practice tends to mask much of the mesoscale variability in the
timing and placement of both observed and simulated rainfall.
An important goal for the present study, however, is to ensure
intervariable consistency among the key variables and processes occurring in
the mesoscale model simulation and to ensure that these modeled fields
accurately represent the processes taking place in the atmosphere for specific
62
-------
y'
I
I
Figure Al. Schematic showing an idealized verification of
_model rainfall for which the Threat Score is
0.333 and the Bias Score is 1.00. The solid
(dashed) region represents a forecast (observed)
rain area for an arbitrary threshold. (X - X'
and Y - Y') .
63
-------
cases. Precipitation is a key process relating to intervariable consistency
because of its dependence on so many synoptic, mesoscale and local processes,
and driven by many dynamic, thermodynamic and physical mechanisms. In
addition, precipitation is vital for removal of acid constituents from the
atmosphere. Therefore, verification of mesoscale precipitation features is
performed on time scales of 12 hours, rather than 24 h. For identical
synoptic scale or long term skill, this tends to produce lower TS values,
particularly during summertime situations where rainfall often occurs due to
isolated areas of convection.
For reference, some typical values of 24-h TS calculated for operational
limited-area models are reported in Table Al. The table shows that a TS of
about .33 is typical for annual precipitation accuracy for light rain (0.50
cm/24h), while values decline for heavier rain. It must be emphasized,
however, that these values do not account for the increased mesoscale
variability that occurs in rainfall summed for 12-h periods. For the shorter
period, the values in Table Al certainly overestimate the threat score of
these models.
Another factor affecting the threat scores, as well as other measures of
model skill in predicting rainfall, is the accuracy of the precipitation
verification analyses. The issue of rain gauge representativeness and its
effect on skill scores is examined in Appendix B.
A.4 Precipitation Bias Scores
The bias score (BS) is another measure of a model's skill at predicting
the area of precipitation exceeding a designated threshold amount. However,
it reflects only the tendency for the model to simulate too small or too large
an area with no consideration given to the location of rainfall. Therefore,
64
-------
Table Al. Typical values of 24-h Threat Score (TS) based on
data reported by Anthes (1983) and Black and
Mesinger (1989) for the LFM and NGM operational
mesoscale forecast models. The rainfall intensities
have been expressed as 12-h rates, as for MM4/FDDA, so
that scores for 0.25cm/12h equal those reported in the
literature for 0.50cm/24h.
Rain Threshold Season TS
(cm/12-h)
0.25 annual .33
0.25 summer .20
0.25 winter .40
0.64 annual .22
0.64 summer .12
0.64 winter .28
65
-------
such errors as those occurring due to phase errors in the passage of frontal
rain bands are ignored. The BS is defined as
BS - (Nl + N2)/(N1 + N3) (A5)
where Nl, N2, and N3 are defined above. Figure Al also illustrates a
situation with a perfect bias score (BS - 1.0).
Anthes (1983) reports annual mean 24-h BS for the LFM (following the
introduction of a realistic boundary layer moisture analysis) to be about 1.55
at the 0.025-cm threshold. Black and Mesinger (1989) report that for November
1988, the NGM had a BS of 1.05 at the 0.50-cm threshold and 0.86 at the 1.27-
cm threshold. The reported values indicate that the operational models tend
to over-forecast the area of very light precipitation and under-forecast the
area of heavy precipitation. Black and Mesinger (1989) report that the NGM's
bias score for thresholds above 2.5 cm falls below 0.50. There also tends to
be a. seasonal dependency in bias scores, although the characteristics will
depend on the particular numerical model. The bias scores for the operational
models are normally reported as monthly averages of the 24-h scores.
Ideally, the bias score should be close to 1.0. Values published in the
literature suggest that a range of 0.5 to 1.5 is typical, when the
calculations are based on 30-day averages of scores calculated daily from 24-h
rain totals. It should be noted that this tends to mask certain problems. A
set of bias scores based on 12-h rainfall totals, as calculated for MM4/FDDA
in this report, should be expected to display a wider range of values.
66
-------
A. 5 Precipitation Mean Errors and Mean Absolute Errors
The TS and BS provide no direct information on the accuracy of the
overall magnitude of the precipitation. Therefore, a mean error (ME) and a
mean absolute error (MAE) are computed for the 12-h precipitation fields over
the verification region. These scores are defined by
M
ME - ^ [P - P0]/M (A6)
j-l
and
M
MAE - ^ |P - P0|/M (A7)
j-l
Only those points, j, where either the observed or simulated precipitation
exceeds a trace amount (defined here as 0.05 cm) are used in the calculation.
P is the simulated 12-h precipitation and P0 is the observed 12-h
precipitation total (analyzed to the 80-km mesh). Therefore, similar to TS,
both of these measures have a penalizing effect for simulating precipitation
where it is not observed, and a correct simulation of the non-occurrence of
precipitation does not enter directly into the calculations. Since ME is
based on the simulated-minus-observed rainfall, a positive (negative) value
represents a tendency to over-forecast (under-forecast) the amount of
precipitation. Therefore, the ME provides a direct measure of whether or not
the model has a bias toward producing too much or too little rainfall volume.
a key factor for simulating wet deposition in RADM.
The MAE is always positive and defines the magnitude of the typical error
at any point reporting rainfall. That is, the MAE gives the average error
that would be found by comparing analyzed (observed) and simulated
67
-------
precipitation at randomly selected points within regions experiencing
measurable rainfall.
It is very difficult to give a typical value for the MAE, because
analyses used for verification contain significant uncertainties that depend
on season, size of the grid mesh, terrain roughness and the density of the
rain gauge network used to construct the analyses. Any cursory examination of
rain gauge measurements reveals much local variability in rainfall,
particularly in warm-season cases for which convection is the dominant
precipitation mechanism. Appendix B examines the uncertainty characteristics
of the rainfall analyses used for verification. This data provides a means
for interpreting the MAE by defining a range in which departure of model
precipitation from analyzed precipitation must be considered to be within the
representativeness uncertainty of the rainfall observations themselves, even
assuming no analysis error. Of course, such observational and analysis errors
will inevitably affect the calculation of the other precipitation scores
(e.g., TS and BS) as well.
A.6 Precipitation Large-Error Count
Finally, the number of points for which the 12-hourly precipitation total
is in error by more than 1.0 cm is determined. This provides a rough measure
of the spatial frequency of "very large" rainfall errors which might be
expected to have an adverse effect on acid deposition calculations in RADM.
Of course, this Large-Error Count (LEG) takes no account of the acid
concentration, so that by itself, it cannot be considered a reliable predictor
of deposition error.
68
-------
APPENDIX B
Uncertainty Characteristics of Rainfall Analyses
Any cursory examination of rain gauge measurements reveals much local
variability in rainfall. This variability depends on season, terrain and the
density of the measurements. For example, the predominance of small-scale
connective cells contributing to warm-season rain tends to increase the
variability in that season. Similarly, precipitation distribution will tend
to be less homogeneous in rough terrain. Gridded rainfall analyses based on
such observations will contain uncertainties due to questions about the
representativeness of the gauge measurements representing the mean rainfall in
each grid cell. An example of rainfall spatial variability revealed by a very
dense network of climatological stations (reporting daily totals) is shown in
Fig. Bl.
Ideally, an accurate analysis of rainfall would require "sufficient"
observations to determine a reliable mean value for each grid box. However,
there are rarely enough gauges found in close proximity to ensure that the
analyzed values closely approximate the mean. For example, even the
relatively dense hourly rain gauge network used in this study provides only
about 1100-1200 observations within the 839 verification grid boxes (80-km
resolution), or a little more than one per box. About 55 boxes have 10 or
more 24-h fclimatological) observing stations, while many more have no hourly
rain gauges and are "filled in" by interpolation from adjacent boxes. Thus,
an analysis of the hourly observations will have a significant uncertainty
since only one or two gauges, on average, are used to represent the rainfall
over 6400 square km. Recall that the precipitation analysis scheme described
in Section 2.4 uses only hourly observations within a given 80-km box to
69
-------
0.10
(
%3 025 QIO 0/3 >° ^ 10
0-23^
^ us^o b
OL23
Figure Bl. Map of precipitation (cm) observed during the
storm of 13-15 January 1980. Numbers are in
hundredths of cm, plotted at the climatological
stations (After Forbes, et al., 1987). An
80-km grid is superimposed for reference
(straight thin solids) .
70
-------
represent that box (if no observations exist there, then values in immediately
adjacent boxes may be considered). The problem, then, is to quantify this
observation representativeness uncertainty.
To determine the representativeness uncertainty, data from the dense
climatological (24-h totals) rain gauge network was obtained for fourteen
states from the National Climate Data Center (NCDC) for the years 1985 and
1988. These states (Tennessee, North Carolina, Kentucky, Virginia, Vest
Virginia, Maryland, Delaware, Indiana, Illinois, Ohio, Pennsylvania, New
Jersey, New York, and Michigan) cover most of the precipitation verification
region east of the Mississippi River (see Fig. 1). The data sets were
examined to determine the number of 80-km grid boxes having at least 10 24-h
gauges per box. About 55 such boxes were found, mostly in the states in which
the Appalachian Mountains are located. Only these boxes were considered to
have sufficient observations to provide a reasonably accurate mean rainfall
value for an event. Over the two year period and in all the boxes containing
a minimum of 10 rain gauges, there were many thousands of events from which to
calculate the representativeness uncertainty. For this purpose an event is
defined as a single day for which measurable rain is recorded in a single grid
box.
Next, the mean precipitation (MP) and the mean absolute deviation from MP
were calculated for each event. Figures B2 - B8 show the mean absolute
deviation plotted against MP for all points and then by season and by terrain
roughness. The curves are based on a logarithmic fit to the data, which was
found to give the highest correlation in each case (better that .95).
These figures demonstrate that there is a fairly large uncertainty
introduced when a single observation is used to represent rainfall on the 80-
km scale. For example, Figure B2.b shows that a particular rain observation
71
-------
ALL SEASONS ALL TERRAIN
Correlation = 0.957 Number of Points = 14620
Mean Dev.
from
MP (cm)
2.00
1.75.
1.50.
1.25.
1.00.
0.75.
0.50.
0.25.
0.00
*j£~'.ri*!-.VN*'vrr •.--...•:'-.; •...
/ •^-••••- • •
0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0 4.5 5.0
MP, Mean Precip. (cm)
Figure B2. Representativeness uncertainty of rainfall
observations. The mean absolute deviation from
the mean precipitation is plotted as a function
of the mean precipitation (MP), valid when a
single observation is used to represent rain-
fall on an 80-km grid. Based on 14,620 events
in 1985 and 1988 over the eastern U. S. for all
seasons and all terrain. Plots are shown for
rainfall up to (a) 5.0 cm, and (b) 1.0 cm .
72
-------
1.00
0.80.
0.60.
Mean Dev.
from
MP (cm)
0.40.
0.20.
0.00
0
ALL SEASONS ALL TERRAIN
Correlation » 0.955 Number of Points
12745
0.2 0.4 0.6 0.8
MP, Mean Precip. (cm)
1.0
Figure B2. (b)
73
-------
DEC-FEE ALL TERRAIN
Correlation = 0.953 Number of Points
3771
Mean Dev.
from
MP (can)
2.00
1.75.
1.50.
1.25.
1.00.
0.75-
0.50.
0.25.
0.00
0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0 4.5 5.0
MP, Mean Precip. (cm)
Figure B3. Same as Fig. B2, except for 3,771 events during
winter (Dec., Jan., Feb.) and for all terrain .
74
-------
Mean Dev.
from
MP (cm)
DEC-FEB ALL TERRAIN
Correlation » 0.950 Number of Points
3506
1.00
0.80.
0.60.
0.40.
0.20.
0.00
0
0.2 0.4 0.6 0.8
MP, Mean Precip. (cm)
1.0
Figure B3.(b)
75
-------
MAR-MAY ALL TERRAIN
Correlation = 0.959 Number of Points
3692
Mean Dev.
from
MP (cm)
2.00
1.75.
1.50.
1.25.
1.00.
0.75.
0.50.
0.25.
0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0 4.5 5.0
MP, Mean Precip. (cm)
Figure B4. Same as Fig. B2, except for 3,692 events during
spring (Mar., Apr.. May) and for all terrain .
76
-------
MAR-MAY ALL TERRAIN
Correlation =» 0.956 Number of Points
3237
1.00.
0.80.
0.60.
Mean Dev.
from
MP (cm) 0.40.
0.20.
0.00
0.0
0.2 0.4 0.6 0.8
MP, Mean Precip. (cm)
1.0
Figure B4 (b)
77
-------
JUN-AUG ALL TERRAIN
Correlation - 0.967 Number of Points - 3609
Mean Oev.
from
MP (cm)
2.00
1.75.
1.50.
1.25.
1.00
0.75
0.50
0.25
0.00
0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0 4.5 5.0
MP, Mean Precip. (cm)
Figure B5. Same as Fig. B2, except for 3,609 events during
summer (Jun., Jul., Aug.) and for all terrain .
78
-------
Mean Dev.
from
MP (cm)
JUN-AOG ALL TERRAIN
Correlation =• 0.968 Number of Points
3049
1.00.
0.80.
0.60
0.40
0.20.
0.00
0.2 0.4 0.6 0.8
MP, Mean Precip. (cm)
1.0
Figure B5 (b)
79
-------
SEP-NOV ALL TERRAIN
Correlation = 0.959 Number of Points = 3548
Mean Dev.
from
MP (cm)
2.00
1.75.
1.50.
1.25.
1.00.
0.75.
0.50,
0.25J
0.00
0 0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0 4.5 5.0
MP, Mean Precip. (cm)
Figure B6. Same as Fig. B2, except for 3,548 events during
autumn (Sep., Oct., Nov.) and for all terrain .
80
-------
Mean Dev.
from
MP (cm)
SEP-NOV ALL TERRAIN
Correlation - 0.957 Number of Points =• 2953
1.00.
0.80.
0.60.
0.40.
0.20.
0.00.
/;:./•? • -.'r^xT-' -'-.•••••
0 0
0.2 0.4 0.6 0.8
MP, Mean Precip. (cm)
1.0
Figure B6 (b)
81
-------
Mean Dev.
from
MP (cm)
ALL SEASONS GENTLE TERRAIN
Correlation =» 0.956 Number of Points
7312
2.00.
1.75.
1.50.
1.25.
1.00.
0.75.
0.50.
0.25.
0.00
££&';' ..- , •
•^V^'i. •:" ... '
0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0 4.5 5.0
MP, Mean Precip. (cm)
Figure B7. Same as Fig. B2, except for 7,312 events for
all seasons and for gentle terrain (below 400 m)
82
-------
Mean Dev.
from
MP (cm)
ALL SEASONS GENTLE TERRAIN
Correlation — 0.954 Number of Points
6303
1.00
0.80.
0.60.
0.40
0.20.
0.00
0 0
0.2 0.4 0.6 0.8
MP, Mean Precip. (cm)
1.0
Figure B7(b)
83
-------
ALL SEASONS ROUGH TERRAIN
Correlation = 0.958 Number of Points = 7308
Mean Dev.
from
MP (cm)
2.00
1.75.
1.50.
1.25.
1.00.
0.75
0.50.
0.25.
0.00
0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0 4.5 5 0
MP, Mean Precip. (cm)
Figure B8. Same as Fig. B2, except for 7,308 events for
all seasons and for rough terrain (above 400 m)
84
-------
Mean Dev.
from
MP (cm)
ALL SEASONS ROUGH TERRAIN
Correlation =» 0.956 Number of Points = 6442
1.00
0.80
0.60
0.40
0.20.
0.00
*"''''" ' ' '
0 0
0.2 0.4 0.6 0.8
MP, Mean Precip. (cm)
1 0
Figure B8 (b)
85
-------
of 1.0 cm can be expected to have a characteristic uncertainty of 0.61 cm when
it is used to represent the mean rainfall in an 80-km box. Similarly, a more
typical rainfall amount of 0.45 cm has a characteristic uncertainty of 0.32 cm
at the 80-km scale. Note that the mean absolute deviation from HP (the
observation representativeness uncertainty) is proportionately greater for
smaller amounts of rain. Further examination of Fig. B3 - B8 shows that the
uncertainty is greater during the warm seasons, particularly in summer.
However, no significant dependence was found, at least for these grid boxes,
for gentle versus rugged terrain (defined here as mean elevation below or
above 400 m, respectively). In summer, the characteristic uncertainty for 1.0
cm of rain is 0.80 cm!
These figures have very important implications for the interpretation of
the rainfall verification statistics. They show that, even assuming no
analysis errors whatsoever, the analyzed "observed" precipitation has a very
significant uncertainty which can, indeed, be quantified. Whenever the model-
predicted precipitation for a given grid point falls within the uncertainty
range of the observed precipitation, that prediction must be assumed to be
correct within the ability of the observations to determine what is, in fact,
the true mean precipitation in that box.
The MAE gives the typical error between "observed" and simulated
precipitation for all points having measurable rain. An examination of the
precipitation data indicates that the average amount of simulated rainfall at
points experiencing measurable rain is about 0.45 cm/12h. As shown in Section
3.2.3, the MM4/FDDA model produced, for all cases, a grand-average MAE of 0.30
cm. Fig. B2.b demonstrates that, for the typical rainfall amount of 0.45 cm,
the uncertainty of the observations is 0.32 cm. Thus, on average, the MAE can
be attributed to the uncertainty contained within the analyses about as easily
86
-------
as it can be to model errors. Stated another way, if the model simulated
rainfall were everywhere equal to the true mean rainfall, those boxes
containing 0.45 cm/12h would be expected to have a MAE of 0.32 cm due solely
to the representativeness uncertainty of the observations (assuming an average
of one observation available per box).
Similar calculations could be made to determine the representativeness
uncertainty associated with the use of two, three or more observations per 80-
km box. However, that would involve additional work with the climatological
data sets that is beyond the scope of this study. Only a very few of the grid
boxes contain about 10 hourly reporting rain gauges (these are not the same as
the more numerous 24-h reporting climatological gauges shown in the shaded
region of Figure 1). The few boxes over land and in' the verification domain
with two or more hourly rain gauges are offset by those with no hourly gauges
and for which data is interpolated from adjacent boxes (see Section 2.4).
Thus, on average, an estimate of uncertainty based on the assumption of about
one rain gauge per box will be reasonably accurate for representing the
domain-wide skill.
This calculation certainly does not mean that the model precipitation
fields are "nearly perfect". It does demonstrate, however, that within our
ability to define the actual mesoscale precipitation at the 80-km scale, the
•
numerical model (on average over the region and for many cases) does nearly as
well at representing the rainfall as an analysis based on the relatively
"dense" network of hourly observing stations. At the same time, however, the
model is able to produce mesoscale convergence and vertical motion fields not
obtainable from existing observations and can ensure intervariable consistency
among the model variables.
87
-------
APPENDIX C
Statistical Summaries of the Meteorological Cases
-------
SUMMARY OF MM4-FDDA SIMULATION FOR EPA-RADM
AGGREGATION AND EVALUATION STUDY.
Case No. : 1
Begin Date: 0000 UTC, 7 April 1981
End Date: 0000 UTC, 12 April 1981
Institution: PSU
Recorded by: N. Seaman
SI Score, Sea-Level Fres. - 36.3
SI Score, 500-mb Height - 18.5
Table 1. Mean and RMS statistics for primitive variables.
Variable
( units)
Mean Error
RMS Error
Sea-Lvl.
Pressure
(mb)
0.07
1.44
Sfc.Mix.
Ratio
Ce/k?)
0.47
0.80
850-mb
Temp.
(C)
-0.08
0.40
500-mb
Wind Spd.
(m/s)
0.98
1.20
500-mb
Height
(m)
0.92
9.32
Table 2. 12-hrly Precipitation Statistics.
Samson Precip.
Skill Score
No. of precip. points fcst.
No. of precip. points obs.
Threat score (0.25 cm threshold)
Threat score (0.64 cm threshold)
Bias score (0.25 cm threshold)
Bias score (0.64 cm threshold)
Mean (areal) error (cm)
Mean (areal) absolute error (cm)
Case Mean
220
188
0.28
0.17
1.37
1.49
0.05
0.18
Std. Dev.
103
114
0.16
0.08
0.71
0.60
0.06
0.07
No. of points with errors greater
than 1.0 cm
89
-------
SUMMARY OF MM4-FDDA SIMULATION FOR EPA-RADM
AGGREGATION AND EVALUATION STUDY.
Case No. : 2
Begin Date: 1200 UTC, 11 April 1981
End Date: 0000 UTC, 15 April 1981
Institution: PSU
Recorded by: N. Seaman
Si Score, Sea-Level Pres. - 38.5
SI Score, 500-mb Height .-22.6
Table 1. Mean and RMS statistics for primitive variables.
Variable
( units)
Mean Error
RMS Error
Sea-Lvl.
Pressure
-0.39
1.82
Sfc.Mix.
Ratio
0.41
0.78
850-mb
Temp.
CO
-0.09
0.46
500-mb
Wind Spd.
Cm/s)
0.95
1.21
500-mb
Height
Cm)
-3.10
11.44
Table 2. 12-hrly Precipitation Statistics.
Samson Precip.
Skill Score
No. of precip. points fcst.
No. of precip. points obs.
Threat score (0.25 cm threshold)
Threat score (0.64 cm threshold)
Bias score (0.25 cm threshold)
Bias score (0.64 cm threshold)
Mean (areal) error (cm)
Mean (areal) absolute error (cm)
Case Mean
293
294
0.49
0.35
1.09
1.38
0.04
0.28
Std. Dev.
89
74
0.08
0.10
0.26
0.73
0.09
0.07
No. of points with errors greater
than 1.0 cm
22
14
90
-------
SUMMARY OF MM4-FDDA SIMULATION FOR EPA-RADM
AGGREGATION AND EVALUATION STUDY.
Case No. :3
Begin Date:0000 UTC, 20 April 1981
End Date: 0000 UTC, 25 April 1981
Institution: PSU
Recorded by: N. Seaman
SI Score, Sea-Level Pres. - 37.4
SI Score, 500-mb Height - 20.2
Table 1. Mean and RMS statistics for primitive variables.
Variable
(units)
Mean Error
RMS Error
Sea-Lvl.
Pressure
(mb)
-0.02
1.46
Sfc.Mix.
Ratio
(g/kg)
0.38
0.74
850-mb
Temp.
(C)
-0.19
0.50
500-mb
Wind Spd.
(n/s)
1.01
1.24
500-mb
Height
(m)
1.38
10.27
Table 2. 12-hrly Precipitation Statistics.
Samson Precip.
Skill Score
No. of precip. points fcst.
No. of precip. points obs.
Threat score (0.25 cm threshold)
Threat score (0.64 cm threshold)
Bias score (0.25 cm threshold)
Bias score (0.64 cm threshold)
Mean (areal) error (cm)
Mean (areal) absolute error (cm)
Case Mean
391
334
0.40
0.22
1.58
1.81
0.03
0.24
Std. Dev.
90
105
0.19-
0.16
1.18
1.80
0.11
0.07
No. of points with errors greater
than 1.0 cm - 15 12
91
-------
SUMMARY OF MM4-FDDA SIMULATION FOR EPA-RADM
AGGREGATION AND EVALUATION STUDY.
Case No. :4
Begin Date: 0000 UTC, 12 July 1980
End Date: 0000 UTC, 17 July 1980
Institution: PSU
Recorded by: N. Seaman
SI Score, Sea-Level Pres. - 53.1
SI Score, 500-mb Height - 31.7
Table 1. Mean and RMS statistics for primitive variables.
Variable
(units)
Mean Error
RMS Error
Sea-Lvl.
Pressure
(mb)
-0.74
1.81
Sfc.Mix.
Ratio
(g/kg)
0.83
1.58
850-mb
Temp.
-0.16
0.46
500-mb
Wind Spd.
fm/s")
0.88
1.10
500-mb
Height
(m)
-1.70
11.14
Table 2. 12-hrly Precipitation Statistics.
EPA Precip.
Skill Score
No. of precip. points fcst.
No. of precip. points obs.
Threat score (0.25 cm threshold)
Threat score (0.64 cm threshold)
Bias score (0.25 cm threshold)
Bias score (0.64 cm threshold)
Mean (areal) error (cm)
Mean (areal) absolute error (cm)
Case Mean
200
126
0.15
0.08
1.93
3.71
0.14
0.32
Std. Dev.
71
45
0.10
0.11
1.44
5.51
0.09
0.12
No. of points with errors greater
than 1.0 cm - 16 15
92
-------
SUMMARY OF MM4-FDDA SIMULATION FOR EPA-RADM
AGGREGATION AND EVALUATION STUDY.
Case No. : 5
Begin Date: 1200 UTC, 16 July 1980
End Date: 0000 UTC, 20 July 1980
Institution: PSU
Recorded by: N. Seaman
SI Score, Sea-Level Pres. - 54.8
SI Score, 500-mb Height - 33.0
Table 1. Mean and RMS statistics for primitive variables.
Variable
(units}
Mean Error
RMS Error
Sea-Lvl.
Pressure
(mb)
-0.51
1.65
Sfc.Mix.
Ratio
(g/kg}
0.86
1.65
850-mb
Temp.
(C)
-0.18
0.47
500-mb
Wind Spd.
(m/s)
0.96
1.23
500-mb
Height
(m)
-0.44
11.07
Table 2. 12-hrly Precipitation Statistics.
EPA Precip.
Skill Score
No. of precip. points fcst.
No. of precip. points obs.
Threat score (0.25 cm threshold)
Threat score (0.64 cm threshold)
Bias score (0.25 cm threshold)
Bias score (0.64 cm threshold)
Mean (areal) error (cm)
Mean (areal) absolute error (cm)
Case Mean
285
179
0.17
0.10
2.42
4.82
0.15
0.30
Std. Dev.
86
47
0.08
0.09
1.16
4.69
0.10
0.09
No. of points with errors greater
than 1.0 cm - 17 13
93
-------
SUMMARY OF MM4-FDDA SIMULATION FOR EPA-RADM
AGGREGATION AND EVALUATION STUDY.
Case No. : 6
Begin Date: 0000 UTC, 27 January 1982
End Date: 0000 UTC, 1 February 1982
Institution: NCAR
Recorded by: P. Haagenson
SI Score, Sea-Level Pres. - 34.9
SI Score, 500-mb Height - 17.9
Table 1. Mean and RMS statistics for primitive variables.
Variable
(units}
Mean Error
RMS Error
Sea-Lvl.
Pressure
(mb)
-0.01
1.83
Sfc.Mix.
Ratio
(g/kg)
0.27
0.61
850-mb
Temp.
CO
-0.04
0.53
500-mb
Wind Spd.
fm/s}
1.27
1.55
500-mb
Height
(m)
1.11
12.35
Table 2. 12-hrly Precipitation Statistics.
EPA Precip.
Skill Score
No. of precip. points fcst.
No. of precip. points obs.
Threat score (0.25 cm threshold)
Threat score (0.64 cm threshold)
Bias score (0.25 cm threshold)
Bias score (0.64 cm threshold)
Mean (areal) error (cm)
Mean (areal) absolute error (cm)
Case Mean
220
168
0.40
0.61
2.88
1.23
0.05
0.22
Std. Dev.
159
136
0.30
0.10
3.73
0.21
0.05
0.11
No. of points with errors greater
than 1.0 cm - 14 16
94
-------
SUMMARY OF MM4-FDDA SIMULATION FOR EPA-RADM
AGGREGATION AND EVALUATION STUDY.
Case No. : 7
Begin Date: 0000 UTC, 18 March 1982
End Date: 0000 UTC, 23 March 1982
Institution: NCAR
Recorded by: F. Haagenson
SI Score, Sea-Level Pres. - 39.0
SI Score, 500-mb Height - 21.0
Table 1. Mean and RMS statistics for primitive variables.
Variable
(units}
Mean Error
RMS Error
Sea-Lvl.
Pressure
(mb)
0.20
1.50
Sfc.Mix.
Ratio
fe/kg)
0.42
0.87
850-mb
Temp.
(C)
-0.09
0.40
500-mb
Wind Spd.
(m/s)
1.04
1.26
500-mb
Height
Cm)
2.20
12.07
Table 2. 12-hrly Precipitation Statistics.
EPA Precip.
Skill Score
No. of precip. points fcst.
No. of precip. points obs.
Threat score (0.25 cm threshold)
Threat score (0.64 cm threshold)
Bias score (0.25 cm threshold)
Bias score (0.64 cm threshold)
Mean (areal) error (cm)
Mean (areal) absolute error (cm)
Case Mean
258
186
0.32
0.21
1.37
1.38
0.03
0.24
Std. Dev.
95
71
0.16
0.17
0.69
0.94
0.09
0.09
No. of points with errors greater
than 1.0 cm 15 15
95
-------
SUMMARY OF MM4-FDDA SIMULATION FOR EPA-RADM
AGGREGATION AND EVALUATION STUDY.
Case No. : 8
Begin Date: 0000 UTC, 27 June 1982
End Date: 0000 UTC, 2 July 1982
Institution: NCAR
Recorded by: P. Haagenson
SI Score, Sea-Level Pres. - 46.5
SI Score, 500-mb Height - 28.2
Table 1. Mean and RMS statistics for primitive variables.
Variable
(units)
Mean Error
RMS Error
Sea-Lvl.
Pressure
(rub)
-0.72
1.80
Sfc.Mix.
Ratio
(g/kg)
0.56
1.11
850-mb
Temp.
CO
-0.16
0.45
500-mb
Wind Spd.
(m/s)
0.87
1.16
500-mb
Height
(m)
-3.83
11.77
Table 2. 12-hrly Precipitation Statistics.
EPA Precip.
Skill Score
No. of precip. points fcst.
No. of precip. points obs.
Threat score (0.25 cm threshold)
Threat score (0.64 cm threshold)
Bias score (0.25 cm threshold)
Bias score (0.64 cm threshold)
Mean (areal) error (cm)
Mean (areal) absolute error (cm)
Case Mean
403
289
0.26
0.17
1.41
1.40
0.07
0.32
Std. Dev.
118
84
0.09
0.10
0.42
0.63
0.09
0.06
No. of points with errors greater
than 1.0 cm 30 14
96
-------
SUMMARY OF MM4-FDDA SIMULATION FOR EPA-RADM
AGGREGATION AND EVALUATION STUDY.
Case No. : 9
Begin Date: 0000 UTC, 6 August 1982
End Date: 0000 UTC, 11 August 1982
Institution: NCAR
Recorded by: P. Haagenson
SI Score, Sea-Level Pres. - 48.7
SI Score, 500-mb Height - 31.3
Table 1. Mean and RMS statistics for primitive variables.
Variable
(units}
Mean Error
RMS Error
Sea-Lvl.
Pressure
(mb)
-0.84
1.76
Sfc.Mix.
Ratio
fg/kg>
0.60
1.17
850-mb
Temp.
(C)
-0.12
0.43
500-mb
Wind Spd.
(m/s)
0.77
0.94
500-mb
Height
(m)
-3.97
10.84
Table 2.. 12-hrly Precipitation Statistics.
EPA Precip.
Skill Score
No. of precip. points fcst.
No. of precip. points obs.
Threat score (0.25 cm threshold)
Threat score (0.64 cm threshold)
Bias score (0.25 cm threshold)
Bias score (0.64 cm threshold)
Mean (areal) error (cm)
Mean (areal) absolute error (cm)
Case Mean
405
299
0.28
0.11
1.25
1.32
0.05
0.33
Std. Dev.
123
64
0.11
0.08
0.34
0.68
0.09
0.05
No. of points with errors greater
than 1.0 cm - 29 10
97
-------
SUMMARY OF MM4-FDDA SIMULATION FOR EPA-RADM
AGGREGATION AND EVALUATION STUDY.
Case No. : 10
Begin Date: 0000 UTC, 19 August 1988
End Date: 0000 UTC, 24 August 1988
Institution: NCAR
Recorded by: P. Haagenson
SI Score, Sea-Level Pres. - 49.4
SI Score, 500-mb Height - 30.1
Table 1. Mean and RMS statistics for primitive variables.
Variable
(units)
Mean Error
RMS Error
Sea-Lvl.
Pressure
(mb>
-0.18
1.64
Sfc.Mix.
Ratio
<>/kz)
0.49
1.23
850-mb
Temp.
CO
-0.14
0.38
500-mb
Wind Spd.
(m/s)
0.82
1.00
500-mb
Height
(m)
-0.36
10.79
Table 2. 12-hrly Precipitation Statistics.
EPA Precip.
Skill Score
No. of precip. points fcst.
No. of precip. points obs.
Threat score (0.25 cm threshold)
Threat score (0.64 cm threshold)
Bias score (0.25 cm threshold)
Bias score (0.64 cm threshold)
Mean (areal) error (cm)
Mean (areal) absolute error (cm)
Case Mean
287
230
0.37
0.30
1.33
1.33
0.09
0.42
Std. Dev.
58
44
0.11
0.15
0.29
0.47
0.10
0.08
No. of points with errors greater
than 1.0 cm - 34 10
98
-------
SUMMARY OF MM4-FDDA SIMULATION FOR EPA-RADM
AGGREGATION AND EVALUATION STUDY.
Case No. : 11
Begin Date: 0000 UTC, 28 August 1988
End Date: 0000 UTC, 2 September 1988
Institution: NCAR
Recorded by: F. Haagenson
SI Score, Sea-Level Pres. - 43.6
SI Score, 500-mb Height - 31.4
Table 1. Mean and RMS statistics for primitive variables.
Variable
(units)
Mean Error
RMS Error
Sea-Lvl.
Pressure
fmb)
0.01
1.62
Sfc.Mix.
Ratio
fe/ke)
0.48
1.14
850-mb
Temp.
ro
-0.14
0.36
500-mb
Wind Spd.
Cm/s)
0.77
0.94
500-mb
Height
Cm)
1.42
10.46
Table 2. 12-hrly Precipitation Statistics.
EPA Precio. .
Skill Score
No. of precip. points fcst.
No. of precip. points obs.
Threat score (0.25 cm threshold)
Threat score (0.64 cm threshold)
Bias score (0.25 cm threshold)
Bias score (0.64 cm threshold)
Mean (areal) error (cm)
Mean (areal) absolute error (cm)
Case Mean
251
164
0.35
0.24
1.63
1.62
0.15
0.38
Std. Dev.
52
61
0.16
0.18
0.40
0.50
0.08
0.07
No. of points with errors greater
than 1.0 cm - 27 13
99
-------
SUMMARY OF MM4-FDDA SIMULATION FOR EPA-RADM
AGGREGATION AND EVALUATION STUDY.
Case No. : 12
Begin Date: 1200 UTC, 1 September 1988
End Date: 1200 UTC, 4 September 1988
Institution: NCAR
Recorded by: P. Haagenson
SI Score, Sea-Level Pres. - 41.6
SI Score, 500-mb Height - 31.7
Table 1. Mean and RMS statistics for primitive variables.
Variable
(units)
Mean Error
RMS Error
Sea-Lvl.
Pressure
(mb)
0.35
1.68
Sfc.Mix.
Ratio
fe/ke)
0.51
1.11
850-mb
Temp.
(C)
-0.14
0.32
500-mb
Wind Spd.
Cm/si
0.81
0.96
500-mb
Height
(m)
4.31
11.49
Table 2. 12-hrly Precipitation Statistics.
EPA Precip.
Skill Score
No. of precip. points fcst.
No. of precip. points obs.
Threat score (0.25 cm threshold)
Threat score (0.64 cm threshold)
Bias score (0.25 cm threshold)
Bias score (0.64 cm threshold)
Mean (areal) error (cm)
Mean (areal) absolute error (cm)
Case Mean
336
276
0.39
0.27
1.21
1.07
-0.04
0.43
Std. Dev.
72
78
0.14
0.13
0.25
0.40
0.09
0.10
No. of points with errors greater
than 1.0 cm - 49 26
100
-------
SUMMARY OF MM4-FDDA SIMULATION FOR EPA-RADM
AGGREGATION AND EVALUATION STUDY.
Case No. : 13
Begin Date: 0000 UTC, 4 September 1988
End Date: 0000 UTC, 9 September 1988
Institution: NCAR
Recorded by: P. Haagensdn
SI Score, Sea-Level Pres. - 35.8
SI Score, 500-mb Height - 23.5
Table 1. Mean and RMS statistics for primitive variables.
Variable
(units)
Mean Error
RMS Error
Sea-Lvl.
Pressure
(mb)
0.13
1.57
Sfc.Mix.
Ratio
(g/k?1)
0.58
1.10
850-mb
Temp.
(C)
-0.15
0.41
500-mb
Wind Spd.
rm/s^
0.93
1.12
500-mb
Height
(m)
2.65
11.17
Table 2. 12-hrly Precipitation Statistics.
EPA Precip.
Skill Score
No. of precip. points fcst.
No. of precip. points obs.
Threat score (0.25 cm threshold)
Threat score (0.64 cm threshold)
Bias score (0.25 cm threshold)
Bias score (0.64 cm threshold)
Mean (areal) error (cm)
Mean (areal) absolute error (cm)
Case Mean
130
117
0.47
0.46
1.75
1.03
-0.03
0.38
Std. Dev.
117
121
0.17
0.14
1.79
0.78
0.22
0.22
No. of points with errors greater
than 1.0 cm - 21 27
101
-------
SUMMARY OF MM4-FDDA SIMULATION FOR EPA-RADM
AGGREGATION AND EVALUATION STUDY.
Case No. : 14
Begin Date: 1200 UTC, 8 September 1988
End Date: 1200 UTC, 12 September 1988
Institution: NCAR
Recorded by: P. Haagenson
SI Score, Sea-Level Pres. - 39.1
SI Score, 500-mb Height - 26.2
Table 1. Mean and RMS statistics for primitive variables.
Variable
( units)
Mean Error
RMS Error
Sea-Lvl.
Pressure
(mb)
-0.26
1.63
Sfc.Mix.
Ratio
(e/kz)
0.46
1.15
850-mb
Temp.
(C)
-0.09
0.34
500-mb
Wind Spd.
(m/s)
0.94
1.16
500-mb
Height
Cm)
0.42
11.05
Table 2. 12-hrly Precipitation Statistics.
EPA Precip.
Skill Score
No. of precip. points fcst.
No. of precip. points obs.
Threat score (0.25 cm threshold)
Threat score (0.64 cm threshold)
Bias score (0.25 cm threshold)
Bias score (0.64 cm threshold)
Mean (areal) error (cm)
Mean (areal) absolute error (ca)
Case Mean
173
119
0.48
0.40
1.63
2.33
0.24
0.52
Std. Dev.
90
48
0.22
0.24
0.66
2.48
0.15
0.13
No. of points with errors greater
than 1.0 cm - 2.5
102
-------
SUMMARY OF MM4-FDDA SIMULATION FOR EPA-RADM
AGGREGATION AND EVALUATION STUDY.
Case No. : 15A
Begin Date: 0000 UTC, 12 September 1988
End Date: 1200 UTC, 16 September 1988
Institution: NCAR
Recorded by: P. Haagenson
SI Score, Sea-Level Pres. - 43.3
SI Score, 500-mb Height - 27.7
Table 1. Mean and RMS statistics for primitive variables.
Variable
(units)
Mean Error
RMS Error
Sea-Lvl.
Pressure
(mb)
0.04
1.63
Sfc.Mix.
Ratio
(ff/ke)
0.40
1.07
850-mb
Temp.
(C)
-0.11
0.36
500-mb
Wind Spd.
(m/s)
0.84
1.00
500-mb
Height
(m)
2.21
11.21
Table 2. 12-hrly Precipitation Statistics.
EPA Precip.
Skill Score
No. of precip. points fcst.
No. of precip. points obs.
Threat score (0.25 cm threshold)
Threat score (0.64 cm threshold)
Bias score (0.25 cm threshold)
Bias score (0.64 cm threshold)
Mean (areal) error (cm)
Mean (areal) absolute error (cm)
Case Mean
286
197
0.37
0.25
1.50
1.24
0.09
0.34
Std. Dev.
48
48
0.11
0.18
0.18
0.36
0.05
0.10
No. of points with errors greater
than 1.0 cm - 23 16
103
-------
SUMMARY OF MM4-FDDA SIMULATION FOR EPA-RADM
AGGREGATION AND EVALUATION STUDY.
Case No. : 15B
Begin Date: 0000 UTC, 16 September 1988
End Date: 1200 UTC, 19 September 1988
Institution: NCAR
Recorded by: P. Haagenson
SI Score, Sea-Level Pres. - 39.8
SI Score, 500-mb Height - 30.7
Table 1. Mean and RMS statistics for primitive variables.
Variable
(units)
Mean Error
RMS Error
Sea-Lvl.
Pressure
(mb)
0.03
1.69
Sfc.Mix.
Ratio
(g/kg)
0.47
1.13
850 -mb
Temp.
(C)
-0.06
0.37
500-mb
Wind Spd.
(m/s)
0.92
1.13
500-mb
Height
fm)
2.79
12.48
Table 2. 12-hrly Precipitation Statistics.
EPA Precip.
Skill Score
No. of precip. points fcst.
No. of precip. points obs.
Threat score (0.25 cm threshold)
Threat score (0.64 cm threshold)
Bias score (0.25 cm threshold)
Bias score (0.64 cm threshold)
Mean (areal) error (cm)
Mean (areal) absolute error (cm)
Case Mean
380
327
0.52
0.44
1.15
1.09
0.06
0.51
Std. Dev.
52
40
0.09
0.10
0.11
0.22
0.09
0.06
No. of points with errors greater
than 1.0 cm - 65 15
104
-------
SUMMARY OF MM4-FDDA SIMULATION FOR EPA-RADM
AGGREGATION AND EVALUATION STUDY.
Case No. : 16
Begin Date: 0000 UTC, 19 September 1988
End Date: 0000 UTC, 24 September 1988
Institution: NCAR
Recorded by: P. Haagenson
SI Score, Sea-Level Pres. - 42.7
SI Score, 500-mb Height - 26.3
Table 1. Mean and RMS statistics for primitive variables.
Variable
f units)
Mean Error
RMS Error
Sea-Lvl.
Pressure
(mb)
-0.11
1.70
Sfc.Mix.
Ratio
(g/kg)
0.47
1.17
850 -mb
Temp.
(C)
-0.08
0.35
500-mb
Wind Spd.
fm/s)
0.92
1.12
500-mb
Height
Cm}
1.00
12.03
Table 2. 12-hrly Precipitation Statistics.
EPA Precip.
Skill Score
No. of precip. points fcst.
No. of precip. points obs.
Threat score (0.25 cm threshold)
Threat score (0.64 cm threshold)
Bias score (0.25 cm threshold)
Bias score (0.64 cm threshold)
Mean (areal) error (cm)
Mean (areal) absolute error (cm)
Case Mean
253
183
0.34
0.28
1.61
1.57
0.09
0.36
Std. Dev.
92
90
0.14
0.14
0.66
0.84
0.08
0.12
No. of points with errors greater
than 1.0 cm - 30 28
105
-------
SUMMARY OF MM4-FDDA SIMULATION FOR EPA-RADM
AGGREGATION AND EVALUATION STUDY.
Case No. : 17
Begin Date: 1200 UTC, 23 September 1988
End Date: 1200 UTC, 28 September 1988
Institution: NCAR
Recorded by: P. Haagenson
SI Score, Sea-Level Pres. - 44.9
SI Score, 500-mb Height - 24.7
Table 1. Mean and RMS statistics for primitive variables.
Variable
(units)
Mean Error
RMS Error
Sea-Lvl.
Pressure
(mb)
-0.41
1.61
Sfc.Mix.
Ratio
(g/k^)
0.09
1.01
850 -mb
Temp.
(C)
-0.07
0.31
500-mb
Wind Spd.
(m/s)
0.87
1.05
500-mb
Height
Cm)
-1.12
10.35
Table 2. 12-hrly Precipitation Statistics.
EPA Precip.
Skill Score
No. of precip. points fcst.
No . of precip . points obs .
Threat score (0.25 cm threshold)
Threat score (0.64 cm threshold)
Bias score (0.25 cm threshold)
Bias score (0.64 cm threshold)
Mean (areal) error (cm)
Mean (areal) absolute error (cm)
Case Mean
187
132
0.34
0.26
1.90
1.07
0.08
0.27
Std. Dev.
67
67
0.17
0.20
1.32
0.46
0.07
0.07
No. of points with errors greater
than 1.0 cm - 12 10
106
-------
SUMMARY OF MM4-FDDA SIMULATION FOR EPA-RADM
AGGREGATION AND EVALUATION STUDY.
Case No. : 18
Begin Date: 0000 UTC, 13 April 1982
End Date: 0000 UTC, 18 April 1982
Institution: NCAR
Recorded by: P. Haagenson
SI Score, Sea-Level Pres. - 38.8
SI Score, 500-ob Height - 21.6
Table 1. Mean and RMS statistics for primitive variables.
Variable
(units}
Mean Error
RMS Error
Sea-Lvl.
Pressure
Cmb)
-0.19
1.66
Sfc.Mix.
Ratio
(s/kg}
0.50
0.89
850-mb
Temp.
CO
-0.09
0.37
500-mb
Wind Spd.
(m/s)
1.02
1.25
500-mb
Height
(m)
-1.18
10.66
Table 2. 12-hrly Precipitation Statistics.
EPA Precip.
Skill Score
No. of precip. points fcst.
No. of precip. points obs.
Threat score (0.25 cm threshold)
Threat score (0.64 cm threshold)
Bias score (0.25 cm threshold)
Bias score (0.64 cm threshold)
Mean (areal) error (cm)
Mean (areal) absolute error (cm)
Case Mean
263
187
0.33
0.18
1.60
1.62
0.03
0.27
Std. Dev.
123
95
0.15
0.17
0.47
1.56
0.09
0.11
No. of points with errors greater
than 1.0 cm - 16 20
107
-------
SUMMARY OF MM4-FDDA SIMULATION FOR EPA-RADM
AGGREGATION AND EVALUATION STUDY.
Case No. : 19
Begin Date: 0000 UTC, 23 Sept. 1982
End Date: 0000 UTC, 28 Sept. 1982
Institution: PSU
Recorded by: N. Seaman
SI Score, Sea-Level Fres. - 41.4
SI Score, 500-mb Height - 23.3
Table 1. Mean and RMS statistics for primitive variables.
Variable
(units)
Mean Error
RMS Error
Sea-Lvl.
Pressure
rmb)
-0.41
1.67
Sfc.Mix.
Ratio
(e/kz)
0.53
1.07
850-mb
Temp.
(C)
-0.07
0.32
500-mb
Wind Spd.
(m/s)
0.86
1.04
500-mb
Height
(m)
-0.31
9.76
Table 2. 12-hrly Precipitation Statistics.
EPA Precip.
Skill Score
No. of precip. points fcst.
No. of precip. points obs.
Threat score (0.25 cm threshold)
Threat score (0.64 cm threshold)
Bias score (0.25 cm threshold)
Bias score (0.64 cm threshold)
Mean (areal) error (cm)
Mean (areal) absolute error (cm)
Case Mean
195
132
0.35
0.24
1.32
1.80
0.003
0.22
Std. Dev.
53
34
0.14
0.17
0.45
2.79
0.08
0.08
No. of points with errors greater
than 1.0 cm
108
-------
SUMMARY OF MM4-FDDA SIMULATION FOR EPA-RADM
AGGREGATION AND EVALUATION STUDY.
Case No. : 20
Begin Date: 0000 UTC, 13 December 1982
End Date: 0000 UTC, 18 December 1982
Institution: NCAR
Recorded by: P. Haagenson
SI Score, Sea-Level Pres. - 37.7
SI Score, 500-mb Height - 20.5
Table 1. Mean and RMS statistics for primitive variables.
Variable
( units')
Mean Error
RMS Error
Sea-Lvl.
Pressure
CmM
0.40
2.05
Sfc.Mix.
Ratio
(e/ke)
0.19
0.65
850-mb
Temp.
(C)
-0.04
0.45
500-mb
Wind Spd.
(m/s)
1.16
1.41
500-mb
Height
Cm")
3.59
14.28
Table 2. 12-hrly Precipitation Statistics.
EPA Precip.
Skill Score
No. of precip. points fcst.
No. of precip. points obs.
Threat score (0.25 cm threshold)
Threat score (0.64 cm threshold)
.Bias score (0.25 cm threshold)
Bias score (0.64 cm threshold)
Mean (areal) error (cm)
Mean (areal) absolute error (cm)
Case Mean
106
106
0.46
0.48
0.88
1.03
- -0.10
0.22
Std. Dev.
99
89
0.36
0.27
0.53
0.49
0.11
0.15
No. of points with errors greater
than 1.0 cm - 11 16
109
-------
SUMMARY OF MM4-FDDA SIMULATION FOR EPA-RADM
AGGREGATION AND EVALUATION STUDY.
Case No. : 21
Begin Date: 0000 UTC, 27 May 1983
End Date: 0000 UTC, 1 June 1983
Institution: NCAR
Recorded by: P. Haagenson
SI Score, Sea-Level Pres. - 45.6
SI Score, 500-mb Height - 24.2
Table 1. Mean and RMS statistics for primitive variables.
Variable
( units)
Mean Error
RMS Error
Sea-Lvl.
Pressure
fmb)
0.15
1.75
Sfc.Mix.
Ratio
Cg/kg)
0.40
0.94
850-mb
Temp.
(C)
-0.18
0.46
500-mb
Wind Spd.
(m/s)
0.93
1.16
500-mb
Height
(m)
3.26
11.19
Table 2. 12-hrly Precipitation Statistics.
EPA Precip.
Skill Score
No. of precip. points fcst.
No. of precip. points obs.
Threat score (0.25 cm threshold)
Threat score (0.64 cm threshold)
Bias score (0.25 cm threshold)
Bias score (0.64 cm threshold)
Mean (areal) error (cm)
Mean (areal) absolute error (cm)
Case Mean
330
235
0.31
0.17
1.16
1.22
0.04
0.27
Std. Dev.
124
52
0.06
0.13
0.35
0.52
0.11
0.63
No. of points with errors greater
than 1.0 cm - 16
no
-------
SUMHARY OF MM4-FDDA SIMULATION FOR EPA-RADM
AGGREGATION AND EVALUATION STUDY.
Case No. : 22
Begin Date: 0000 UTC, 2 August 1983
End Date: 0000 UTC, 7 August 1983
Institution: NCAR
Recorded by: P. Haagenson
SI Score, Sea-Level Pres. - 54.6
SI Score, 500-mb Height - 41.4
Table 1. Mean and RMS statistics for primitive variables.
Variable
( units')
Mean Error
RMS Error
Sea-Lvl.
Pressure
Crab1)
-0.51
1.39
Sfc.Mix.
Ratio
fe/ke^
0.61
1.17
850-mb
Temp.
CC")
-0.14
0.39
500-mb
Wind Spd.
fm/s)
0.72
0.85
500-mb
Height
Cm)
-1.52
11.01
Table 2. 12-hrly Precipitation Statistics.
EPA Precip.
Skill Score
No. of precip. points fcst.
No . of precip , points obs .
Threat score (0.25 cm threshold)
Threat score (0.64 cm threshold)
Bias score (0.25 cm threshold)
Bias score (0.64 cm threshold)
Mean (areal) error (cm)
Mean (areal) absolute error (cm)
Case Mean
292
199
0.21
0.13
1.79
2.16
0.08
0.34
Std. Dev.
85
60
0.09
0.11
1.16
1.71
0.14
0.05
No. of points with errors greater
than 1.0 cm - 23
111
-------
SUMMARY OF MM4-FDDA SIMULATION FOR EPA-RADM
AGGREGATION AND EVALUATION STUDY.
Case No. : 23
Begin Dace: 0000 UTC, 7 Sept. 1983
End Date: 0000 UTC,12 Sept. 1983
Institution: PSU
Recorded by: N. Seaman, D. Stauffer
Si Score, Sea-Level Pres. - 42.2
SI Score, 500-nb Height - 27.1
Table 1. Mean and RMS statistics for primitive variables.
Variable
(units)
Mean Error
RMS Error
Sea-Lvl.
Pressure
(mb)
-0.63
1.63
Sfc.Mix.
Ratio
(g/kg)
0.54
1.09
850-mb
Temp.
CO
-0.06
0.34
500-mb
Wind Spd.
fm/s)
0.91
1.12
500-mb
Height
(m)
-2.82
10.41
Table 2. 12-hrly Precipitation Statistics.
EPA Frecip.
Skill Score
No. of precip. points fcst.
No. of precip. points obs.
Threat score (0.25 cm threshold)
Threat score (0.64 cm threshold)
Bias score (0.25 cm threshold)
Bias score (0.64 cm threshold)
Mean (areal) error (cm)
Mean (areal) absolute error (cm)
Case Mean
210
144
0.21
0.13
1.84
1.17
0.04
0.26
Std. Dev.
79
66
0.14
0.12
1.44
0.91
0.10
0.04
No. of points with errors greater
than 1.0 cm - 11
112
-------
SUMMARY OF MM4-FDDA SIMULATION FOR EPA-RADM
AGGREGATION AND EVALUATION STUDY.
Case No. : 24
Begin Date: 0000 UTC, 30 October 1983
End Date: 0000 UTC, 4 November 1983
Institution: PSU
Recorded by: N. Seaman, D. Stauffer
SI Score, Sea-Level Pres. - 38.3
SI Score, 500-mb Height - 29.2
Table 1. Mean and RMS statistics for primitive variables.
Variable
(units)
Mean Error
RMS Error
Sea-Lvl.
Pressure
(mb)
-0.39
1.68
Sfc.Mix.
Ratio
fg/kg)
0.28
0.77
850-mb
Temp.
(C)
-0.06
0.27
500-mb
Wind Spd.
(m/s)
0.73
0.88
500-mb
Height
(m)
-1.16
10.34
Table 2. 12-hrly Precipitation Statistics.
EPA Precip.
Skill Score
No. of precip. points fcst.
No. of precip. points obs.
Threat score (0.25 cm threshold)
Threat score (0.64 cm threshold)
Bias score (0.25 cm threshold)
Bias score (0.64 cm threshold)
Mean (areal) error (cm)
Mean (areal) absolute error (cm)
Case Mean
192
110
0.28
0.19
4.43
0.77
0.04
0.22
Std. Dev.
102
87
0.17
0.15
7.43
0.49
0.07
0.05
No. of points with errors greater
than 1.0 cm - 7 12
113
-------
SUMMARY OF MM4-FDDA SIMULATION FOR EPA-RADM
AGGREGATION AND EVALUATION STUDY.
Case No. : 25
Begin Date: 0000 UTC, 12 September 1983
End Date: 0000 UTC, 17 September 1983
Institution: PSU
Recorded by: N. Seaman, D. Stauffer
SI Score, Sea-Level Pres. - 43.8
SI Score, 500-mb Height - 24.3
Table 1. Mean and RMS statistics for primitive variables.
Variable
( units)
Mean Error
RMS Error
Sea-Lvl.
Pressure
(mb)
-0.22
1.58
Sfc.Mix.
Ratio
Cg/ke)
0.56
1.08
850-mb
Temp.
(C)
-0.11
0.32
500-mb
Wind Spd.
Cm/s)
0.82
0.99
500-mb
Height
(m)
0.35
9.80
Table 2. 12-hrly Precipitation Statistics.
EPA Precip.
Skill Score
No. of precip. points fcst.
No. of precip. points obs.
Threat score (0.25 cm threshold)
Threat score (0.64 cm threshold)
Bias score (0.25 cm threshold)
Bias score (0.64 cm threshold)
Mean (areal) error (cm)
Mean (areal) absolute error (cm)
Case Mean
272
228
0.45
0.33
1.09
0.97
0.01
0.30
Std. Dev.
49
50
0.16
0.15
0.28
0.21
0.05
0.07
No. of points with errors greater
than 1.0 cm - 20 13
114
-------
SUMMARY OF MM4-FDDA SIMULATION FOR EPA-RADM
AGGREGATION AND EVALUATION STUDY.
Case No. : 26
Begin Date: 0000 UTC, 15 March 1984
End Date: 0000 UTC, 20 March 1984
Institution: PSU
Recorded by: N. Seaman, D. Stauffer
SI Score, Sea-Level Fres. - 36.3
SI Score, 500-mb Height - 26.9
Table 1. Mean and RMS statistics for primitive variables.
Variable
(units)
Mean Error
RMS Error
Sea-Lvl.
Pressure
(mb)
0.79
2.04
Sfc.Mix.
Ratio
(g/kg)
0.29
0.81
850-mb
Temp.
(C)
-0.09
0.42
500-mb
Wind Spd.
fm/s")
1.09
1.33
500-mb
Height
Cm')
4.87
13.56
Table 2. 12-hrly Precipitation Statistics.
EPA Precip.
Skill Score
No. of precip. points fcst. -
No. of precip. points obs. -
Threat score (0.25 cm threshold)
Threat score (0.64 cm threshold)
Bias score (0.25 cm threshold) -
Bias score (0.64 cm threshold) -
Mean (areal) error (cm) -
Mean (areal) absolute error (cm) -
No. of points with errors greater
Case Mean Std. Dev.
300
210
0.38
0.27
1.79
1.23
0.05
0.30
19
59
56
0.22
0.18
1.45
0.90
0.13
0.07
10
than 1.0 cm
115
-------
SUMMARY OF MM4-FDDA SIMULATION FOR EPA-RADM
AGGREGATION AND EVALUATION STUDY.
Case No. : 27
Begin Date: 0000 UTC, 19 August 1984
End Date: 0000 UTC, 24 August 1984
Institution: PSU
Recorded by: N. Seaman, D. Stauffer
SI Score, Sea-Level Fres. - 45.5
SI Score, 500-mb Height - 30.9
Table 1. Mean and RMS statistics for primitive variables.
Variable
(units)
Mean Error
RMS Error
Sea-Lvl.
Pressure
Crab)
-0.28
1.46
Sfc.Mix.
Ratio
<-2/ke)
0.54
1.06
850-mb
Temp.
(C)
-0.14
0.39
500-mb
Wind Spd.
(m/s)
0.75
0.89
500-mb
Height
Cm")
-0.69
9.62
Table 2. 12-hrly Precipitation Statistics.
EPA Precip.
Skill Score
No. of precip. points fcst.
No. of precip. points obs.
Threat score (0.25 cm threshold)
Threat score (0.64 cm threshold)
Bias score (0.25 cm threshold)
Bias score (0.64 cm threshold)
Mean (areal) error (cia)
Mean (areal) absolute error (cm)
Case Mean
271
195
0.31
0.15
1.24
1.20
0.06
0.34
Std. Dev.
70
29
0.08
0.09
0.34
0.58
0.10
0.09
No. of points with errors greater
than 1.0 cm . 21 10
116
-------
SUMMARY OF MM4-FDDA SIMULATION FOR EPA-RADM
AGGREGATION AND EVALUATION STUDY.
Case No. : 28
Begin Date: 0000 UTC, 10 June 1984
End Date: 0000 UTC, 15 June 1984
Institution: PSU
Recorded by: N. Seaman, D. Stauffer
Si Score, Sea-Level Pres. - 46.0
SI Score, 500-mb Height - 29.7
Table 1. Mean and RMS statistics for primitive variables.
Variable
(units')
Mean Error
RMS Error
Sea-Lvl.
Pressure
(nib)
-0.37
1.46
Sfc.Mix.
Ratio
(g/ke)
0.56
1.05
850 -mb
Temp.
(C)
-0.14
0.44
500-mb
Wind Spd.
(m/s)
0.88
1.13
500-mb
Height
-1.30
9.24
Table 2. 12-hrly Precipitation Statistics.
EPA Precip.
Skill Score
No. of precip. points fcst.
No. of precip. points obs.
Threat score (0.25 cm threshold)
Threat score (0.64 cm threshold)
Bias score (0.25 cm threshold)
Bias score (0.64 cm threshold)
Mean (areal) error (cm)
Mean (areal) absolute error (cm)
No. of points with errors greater
Case Mean
355
215
0.31
0.19
1.67
1.42
0.06
0.37
32
Std. Dev.
97
58
0.13
0.14
•
0.53
0.65
0.19
0.12
13
than 1.0 cm
117
-------
SUMMARY OF MM4-FDDA SIMULATION FOR EPA-RADM
AGGREGATION AND EVALUATION STUDY.
Case No. : 29
Begin Date: 0000 UTC, September 1984
End Date: 0000 UTC, September 1984
Institution: PSU
Recorded by: N. Seaman, D. Stauffer
SI Score, Sea-Level Pres. - 38.9
SI Score, 500-mb Height - 24.8
Table 1. Mean and RMS statistics for primitive variables.
Variable
f units)
Mean Error
RMS Error
Sea-Lvl.
Pressure
fmb)
-0.16
1.46
Sfc.Mix.
Ratio
fz/kE)
0.52
0.99
850-mb
Temp.
cc->
-0.08
0.37
500-mb
Wind Spd.
(m/s)
0.87
1.05
500-mb
Height
(ml
0.34
10.53
Table 2. 12-hrly Precipitation Statistics.
EPA Precip.
Skill Score
No. of precip. points fcst.
No. of precip. points obs.
Threat score (0.25 cm threshold)
Threat score (0.64 cm threshold)
Bias score (0.25 cm threshold)
Bias score (0.64 cm threshold)
Mean (areal) error (cm)
Mean (areal) absolute error (cm)
Case Mean
250
154
0.31
0.29
1.61
1.93
0.10
0.28
Std. Dev.
59
32
0.12
0.18
0.32
1.04
0.06
0.10
No. of points with errors greater
than 1.0 cm - 15 12
118
-------
SUMMARY OF MM4-FDDA SIMULATION FOR EPA-RADM
AGGREGATION AND EVALUATION STUDY.
Case No. : 30
Begin Date: 0000 UTC, 14 July 1984
End Date: 0000 UTC, 19 July 1984
Institution: NCAR
Recorded by: P. Haagenson
SI Score, Sea-Level Pres. - 46.4
SI Score, 500-mb Height - 28.6
Table 1. Mean and RMS statistics for primitive variables.
Variable
(units)
Mean Error
PUS Error
Sea-Lvl.
Pressure
-0.30
1.62
Sfc.Mix.
Ratio
0.56
1.08
850-mb
Temp.
(C)
-0.14
0.47
500-mb
Wind Spd.
(m/s)
0.86
1.07
500-mb
Height
(m)
-0.66
9.36
Table 2. 12-hrly Precipitation Statistics.
EPA Precip.
Skill Score
No. of precip. points fcst.
No. of precip. points obs.
Threat score (0.25 cm threshold)
Threat score (0.64 en threshold)
Bias score (0.25 cm threshold)
Bias score (0.64 cm threshold)
Mean (areal) error (cm)
Mean (areal) absolute error (cm)
Case Mean
323
90
0.34
0.21
1.27
1.02
0.00
0.34
Std. Dev.
246
51
0.13
0.12
0.53
0.48
0.12
0.07
No. of points with errors greater
than 1.0 cm - 29 15
119
-------
SUMMARY OF MM4-FDDA SIMULATION FOR EPA-RADM
AGGREGATION AND EVALUATION STUDY.
Case No. : 31
Begin Date: 0000 UTC, 31 October 1985
End Date: 0000 UTC, 5 November 1985
Institution: PSU
Recorded by: N. Seaman, D. SGauffer
Si Score, Sea-Level Pres. - 34.6
SI Score, 500-nb Height - 22.7
Table 1. Mean and RMS statistics for primitive variables.
Variable
(units)
Mean Error
RMS Error
Sea-Lvl.
Pressure
fmb")
0.05
1.46
Sfc.Mix.
Ratio
(z/kz)
0.38
0.88
850-mb
Temp.
rci
-0.07
0.30
500 -mb
Wind Spd.
(m/s)
0.96
1.15
500 -mb
Height
(m)
1.60
10.18
Table 2. 12-hrly Precipitation Statistics.
EPA Precip.
Skill Score Case Mean Std. Dev.
No. of precip. points fcst. - 261 56
No. of precip. points obs. - 246 65
Threat score (0.25 cm threshold) - 0.61 0.14
Threat score (0.64 cm threshold) - 0.49 0.18
Bias score (0.25 cm threshold) - 1.07 0.13
Bias score (0.64 cm threshold) - 1.04 0.17
Mean (areal) error (cm) - 0.02 0.07
Mean (areal) absolute error (cm) - 0.36 0.12
No. of points with errors greater
than 1.0 cm 21 9
120
-------
SUMMARY OF MM4-FDDA SIMULATION FOR EPA-RADM
AGGREGATION AND EVALUATION STUDY.
Case No. : 32
Begin Date: 0000 UTC, 10 July 1985
End Date: 0000 UTC, 15 July 1985
Institution: PSU
Recorded by: N. Seaman, D. Stauffer
SI Score, Sea-Level Pres. - 50.6
31 Score, 500-nb Height - 27.6
Table 1. Mean and RMS statistics for primitive variables.
Variable
(units)
Mean Error
RMS Error
Sea-Lvl.
Pressure
(vh)
-0.58
1.73
Sfc.Mlx.
Ratio
(2/ke^
0.75
1.37
850-mb
Temp.
(C~>
-0.17
0.49
500-mb
Wind Spd.
(m/s^
0.86
1.05
500-mb
Height
(ml
-1.78
10.17
Table 2. 12-hrly Precipitation Statistics.
EPA Precip.
Skill Score
No. of precip. points fcst.
No. of precip. points obs.
Threat score (0.25 en threshold)
Threat score (0.64 cm threshold)
Bias score (0.25 cm threshold)
Bias score (0.64 cm threshold)
Mean (areal) error (cm)
Mean (areal) absolute error (cm)
Case Mean
267
175
0.19
0.10
1.66
1.47
0.04
0.29
Std. Dev.
92
58
0.08
0.07
0.38
1.27
0.14
0.04
Nc. of points with errors greater
than 1.0 cm 15 10
121
-------
SUMMARY OF MM4-FDDA SIMULATION FOR EPA-RADM
AGGREGATION AND EVALUATION STUDY.
Case No. : 33
Begin Date: 0000 UTG, 30 April 1985
End Dace: 0000 UTC, 5 May 1985
Institution: NCAR
Recorded by: P. Haagenson
SI Score, Sea-Level Pres. - 41.8
SI Score, 500-mb Height - 25.2
Table 1. Mean and RMS statistics for primitive variables.
Variable
(units ">
Mean Error
RMS Error
Sea-Lvl.
Pressure
i'rnb)
-0.14
1.71
Sfc.Mix.
Ratio
fg/ke)
0.51
0.93
850-mb
Temp.
(C)
-0.15
0.43
500-nb
Wind Spd.
(m/s)
0.85
1.05
500-mb
Height
CE)
0.12
10.54
Table 2. 12-hrly Precipitation Statistics.
EPA Precip.
Skill Score
No. of precip. points fcst.
No. of precip. points obs.
Threat score (0.25 cm threshold)
Threat score (0.64 cm threshold)
Bias score (0.25 ca threshold)
Bias score (0.64 cm threshold)
Mean (areal) error (cm)
Mean (areal) absolute error (cm)
Case Mean
244
175
0.56
0.45
1.39
1.20
0.02
0.30
Std. Dev.
92
71
0.15
0.14
0.36
0.34
0.06
0.09
No. of points with errors greater
than 1.0 cm - 19 11
122
-------
SUMMARY OF MM4-FDDA SIMULATION FOR EPA-RADM
AGGREGATION AND EVALUATION STUDY.
Case No. : 34
Begin Date: OCOO UTC, 14 November 1985
End Date: 0000 UTC, 19 November 1985
Institution: PSU
Recorded by: N. Seaman, D. Stauffer
SI Score, Sea-Level Pres. - 41.7
SI Score, 500-mb Height - 21.7
Table 1. Mean and RMS statistics for primitive variables.
Variable
( units)
Mean Error
EMS Error
Sea-Lvl.
Pressure
(mb)
0.03
2.09
Sfc.Mix.
Ratio
(g/kE)
0.28
0.77
850-mb
Tamp.
(C)
-0.03
0.45
500-mb
Wind Spd.
(m/sN,
1.05
1.33
500-mb
Height
(m)
1.85
14.87
Table 2. 12-hrly Precipitation Statistics.
EPA Precip.
Skill Score
No. of precip. points fcst.
He. of precip. points obs.
Threat score (0.25 cm threshold)
Threat score (0.64 cm threshold)
3ias score (0.25 en threshold)
Bias score (0.64 en threshold)
Mean (areal) error (cm)
Mean (areal) absolute error (cm)
Case Mean
232
263
0.48
0.40
1.03
0.77
-0.08
0.31
Std. Dev.
74
57
0.11
0.18
0.29
0.44
C.09
O.C4
No. of points with errors greater
than 1.0 cm - 22
123
-------
SUMMARY OF MM4-FDDA SIMULATION FOR EPA-RADM
AGGREGATION AND EVALUATION STUDY.
Case No. : 35
Begin Date: 0000 UTC, 24 September 1935
End Date: OOCC UTC, 29 September 1985
Institution: ?SU
Recorded by: N. Seaman, D. Stauffer
SI Score, Sea-Level Pres. - 29.2
SI Score, 500-mb Height - 23.3
Table 1. Mean and RMS statistics for primitive variables.
Variable
( units)
Mean Error
SMS Error
Sea-Lvl.
Pressure
(tub)
0.23
1.73
Sfc.Mix.
Ratio
fz/kz)
0.41
0.37
350-ab
Temp.
CO
-0.10
0 40
500-mb
Wind Spd.
fin/s)
1.05
1.36
500-mb
Height
fm)
4.06
13.43
Table 2. 12-hrly Precipitation Statistics.
EPA Precip.
Skill Score
No. cf precip. points fcst.
No. of precip. points obs.
Threat score (0.25 cm threshold)
Threat score (0.64 cm threshold)
Bias score (0.25 cm threshold)
Bias score (.0.64 cm threshold)
Mean (areal) error (cm)
Mean (areal) absolute error (cm)
Case Mean
219
182
0.46
0.29
1.34
1.99
-0.10
0.40
Std. Dev.
71
60
0.20
0.27
0.81
3.03
0.30
0.28
No. cf points with errors greater
chan 1.0 cm 20 14
124
-------
SUMMARY OF MM4-FDDA SIMULATION FOR EPA-RADM
AGGREGATION AND EVALUATION STUDY.
Case No. : 36
Begin Date: 0000 UTC, 9 October 1985
End Date: 0000 UTC,14 October 1985
Institution: PSU
Recorded by: N. Seaman, D. Stauffer
SI Score, Sea-Level Pres. - 43.1
SI Score, 500-mb Height - 23.0
Table 1. Mean and RMS statistics for primitive variables.
Variable
( units)
Mean Error
RMS Error
Sea-Lvl.
Pressure
(mb)
-0.07
1.72
Sfc.Mix.
Ratio
fg/ke)
0.35
0.85
850-mb
Temp.
(C)
-0.07
0.35
500-mb
Wind Spd.
(m/s)
0.94
1.11
500-mb
Height
Cm)
1.38
12.16
Table 2. 12-hrly Precipitation Statistics.
EPA Precip.
Skill Score
No. of precip. points fcst.
No. of precip. points obs.
Threat score (0.25 cm threshold)
Threat score (0.64 cm threshold)
Bias score (0.25 cm threshold)
Bias score (0.64 cm threshold)
Mean (areal) error (cm)
Mean (areal) absolute error (cm)
Case Mean
285
222
0.43
0.33
1.45
1.71
0.06
0.34
Std. Dev.
28
36
0.14
0.20
0.41
1.09
0.12
0.11
No. of points with errors greater
than 1.0 cm - 21 14
125
-------
SUMMARY OF MM4-FDDA SIMULATION FOR EPA-RADM
AGGREGATION AND EVALUATION STUDY.
Case No. : 37A
Begin Date: 0000 UTC, 13 August 1988
End Date: 0000 UTC, 17 August 1988
Institution: NCAR
Recorded by: P. Haagenson
SI Score, Sea-Level Pres. - 48.1
SI Score, 500-mb Height - 32.5
Table 1. Mean and RMS statistics for primitive variables.
Variable
( units')
Mean Error
RMS Error
Sea-Lvl.
Pressure
(mb)
-0.23
1.74
Sfc.Mix.
Ratio
(g/ke>
0.58
1.30
850-mb
Temp.
(C)
-0.12
0.37
500-mb
Wind Spd.
(m/s)
0.87
1.08
500-mb
Height
(m)
0.48
10.31
Table 2. 12-hrly Precipitation Statistics.
EPA Precip.
Skill Score
No. of precip. points fcst.
No. of precip. points obs.
Threat score (0.25 cm threshold)
Threat score (0.64 cm threshold)
Bias score (0.25 cm threshold)
Bias score (0.64 cm threshold)
Mean (areal) error (cm)
Mean (areal) absolute error (cm)
Case Mean
221
171
0.27
0.17
1.17
0.97
0.00
0.29
Std. Dev.
55
53
0.09
0.11
0.30
0.33
0.09
0.10
No. of points with errors greater
than 1.0 cm 16 15
126
-------
SUMMARY OF MM4-FDDA SIMULATION FOR EPA-RADM
AGGREGATION AND EVALUATION STUDY.
Case No. : 37B
Begin Date: 1200 UTC, 16 August 1988
End Date: 1200 UTC, 19 August 1988
Institution: NCAR
Recorded by: F. Haagenson
Si Score, Sea-Level Pres. - 48.4
SI Score, 500-mb Height - 33.3
Table 1. Mean and RMS statistics for primitive variables.
Variable
( units)
Mean Error
RMS Error
Sea-Lvl.
Pressure
(mb)
-0.17
1.73
Sfc.Mix.
Ratio
(g/ke)
0.50
1.25
850-mb
Temp.
(C)
-0.13
0.37
500-mb
Wind Spd.
(m/s)
0.83
1.02
500-mb
Height
(m)
0.29
11.12
Table 2. 12-hrly Precipitation Statistics.
EPA Precip.
Skill Score
No. of precip. points fcst.
No. of precip. points obs.
Threat score (0.25 cm threshold)
Threat score (0.64 cm threshold)
Bias score (0.25 cm threshold)
Bias score (0.64 cm threshold)
Mean (areal) error (cm)
Mean (areal) absolute error (cm)
Case Mean
224
199
0.25
0.18
1.16
1.34
0.05
0.30
Std. Dev.
68
53
0.06
0.09
0.31
0.57
0.07
0.07
No. of points with errors greater
than 1.0 cm - 18
127
-------
SUMMARY OF MM4-FDDA SIMULATION FOR EPA-RADM
AGGREGATION AND EVALUATION STUDY.
Case No. : 38A
Begin Date: 1200 UTC, 23 August 1988
End Date: 1200 UTC, 26 August 1988
Institution: NCAR
Recorded by: P. Haagenson
SI Score, Sea-Level Pres. - 49.2
SI Score, 500-mb Height - 26.8
Table 1. Mean and RMS statistics for primitive variables.
Variable
(units}
Mean Error
RMS Error
Sea-Lvl.
Pressure
(mb)
0.07
1.71
Sfc.Mix.
Ratio
(g/ke)
0.52
1.24
850-mb
Temp.
(C)
-0.16
0.38
500-mb
Wind Spd.
(m/s)
0.81
0.95
500-mb
Height
(m)
2.94
11.54
Table 2. 12-hrly Precipitation Statistics.
EPA Precip.
Skill Score
No. of precip. points fcst.
No. of precip. points obs.
Threat score (0.25 cm threshold)
Threat score (0.64 cm threshold)
Bias score (0.25 cm threshold)
Bias score (0.64 cm threshold)
Mean (areal) error (cm)
Mean (areal) absolute error (cm)
Case Mean
200
191
0.33
0.15
1.20
1.10
0.00
0.31
Std. Dev.
89
79
0.16
0.16
0.52
0.49
0.06
0.08
No. of points with errors greater
than 1.0 cm - 17 12
128
-------
SUMMARY OF MM4-FDDA SIMULATION FOR EPA-RADM
AGGREGATION AND EVALUATION STUDY.
Case No. : 38B
Begin Date: 0000 UTC, 26 August 1988
End Date: 1200 UTC, 28 August 1988
Institution: NCAR
Recorded by: P. Haagenson
Si Score, Sea-Level Pres. - 45.9
SI Score, 500-mb Height - 25.8
Table 1. Mean and RMS statistics for primitive variables.
Variable
Cunits)
Mean Error
RMS Error
Sea-Lvl.
Pressure
fmb)
-0.50
1.74
Sfc.Mix.
Ratio
(g/k£>
0.37
1.17
850-mb
Temp.
(C)
-0.09
0.34
500-mb
Wind Spd.
(m/s)
0.89
1.05
500-mb
Height
Cm}
-0.45
10.51
Table 2. 12-hrly Precipitation Statistics.
EPA Precip.
Skill Score
No. of precip. points fcst.
No. of precip. points obs.
Threat score (0.25 cm threshold)
Threat score (0.64 cm threshold)
Bias score (0.25 cm threshold)
Bias score (0.64 cm threshold)
Mean (areal) error (cm)
Mean (areal) absolute error (cm)
Case Mean
235
150
0.30
0.11
1.55
1.49
0.09
0.24
Std. Dev.
118
44
0.11
0.11
0.69
1.17
0.15
0.10
No. of points with errors greater
than 1.0 cm - 10 13
129
-------
SUMMARY OF MM4-FDDA SIMULATION FOR EPA-RADM
AGGREGATION AND EVALUATION STUDY.
Case No. : 39
Begin Date: 0000 UTC, 28 September 1988
End Date: 0000 UTC, 1 October 1988
Institution: NCAR
Recorded by: P. Haagenson
SI Score, Sea-Level Pres. - 41.2
SI Score, 500-mb Height - 28.6
Table 1. Mean and RMS statistics for primitive variables.
Variable
( units)
Mean Error
RMS Error
Sea-Lvl.
Pressure
fmb)
0.00
1.65
Sfc.Mix.
Ratio
fe/ke)
0.49
1.10
850-mb
Temp.
(C)
-0.06
0.29
500-mb
Wind Spd.
(m/s)
0.86
1.01
500-mb
Height
(m)
1.89
11.47
Table 2. 12-hrly Precipitation Statistics.
EPA Precip.
Skill Score
No. of precip. points fcst.
No. of precip. points obs.
Threat score (0.25 cm threshold)
Threat score (0.64 cm threshold)
Bias score (0.25 cm threshold)
Bias score (0.64 cm threshold)
Mean (areal) error (cm)
Mean (areal) absolute error (cm)
Case Mean
285
219
0.41
0.30
1.41
1.42
0.07
0.38
Std. Dev.
50
31
0.09
0.07
0.42
0.30
0.11
0.05
No. of points with errors greater
than 1.0 cm - 32 11
130
-------
SUMMARY OF MM4-FDDA SIMULATION FOR EPA-RADM
AGGREGATION AND EVALUATION STUDY.
Case No. : 40
Begin Dace: 0000 UTC, 4 November 1982
End Date: 0000 UTC, 9 November 1982
Institution: NCAR
Recorded by: F. Haagenson
SI Score, Sea-Level Pres. - 37.3
SI Score, 500-mb Height - 18.5
Table 1. Mean and RMS statistics for primitive variables.
Variable
( units')
Mean Error
RMS Error
Sea-Lvl.
Pressure
rmb")
0.00
2.05
Sfc.Mix.
Ratio
(e/kz)
0.35
0.77
850-mb
Temp.
(G)
-0.10
0.40
500-mb
Wind Spd.
(m/s)
0.98
1.22
500-mb
Height
(m)
1.48
12.95
Table 2. 12-hrly Precipitation Statistics.
EPA Precip.
Skill Score
No. of precip. points fcst.
No. of precip. points obs.
Threat score (0.25 cm threshold)
Threat score (0.64 cm threshold)
Bias score (0.25 cm threshold)
Bias score (0.64 cm threshold)
Mean (areal) error (cm)
Mean (areal) "absolute error (cm)
Case Mean
81
67
0.40
0.43
0.52
0.63
-0.06
0.20
Std. Dev.
75
74
0.35
0.18
0.46
0.26
0.10
0.15
No. of points with errors greater
than 1.0 cm - 8 13
131
-------
SUMMARY OF MM4-FDDA SIMULATION FOR EPA-RADM
AGGREGATION AND EVALUATION STUDY.
Case No. : 41
Begin Date: 0000 UTC, 12 May 1982
End Date: 0000 UTC, 17 May 1982
Institution: NCAR
Recorded by: P. Haagenson
SI Score, Sea-Level Pres. - 43.4
SI Score, 500-mb Height - 29.6
Table 1. Mean and RMS statistics for primitive variables.
Variable
(units)
Mean Error
RMS Error
Sea-Lvl.
Pressure
(mb)
-0.42
1.46
Sfc.Mix.
Ratio
(e/ke)
0.40
0.93
850-mb
Temp.
(C)
-0.12
0.38
500-mb
Wind Spd.
(m/s)
0.84
1.09
500-mb
Height
(m)
-1.31
9.68
Table 2. 12-hrly Precipitation Statistics.
EPA Precip.
Skill Score
No. of precip. points fcst.
No. of precip. points obs.
Threat score (0.25 cm threshold)
Threat score (0.64 cm threshold)
Bias score (0.25 cm threshold)
Bias score (0.64 cm threshold)
Mean (areal) error (cm)
Mean (areal) absolute error (cm)
Case Mean
389
277
0.47
- 0.37
1.26
1.21
0.03
0.40
Std. Dev.
68
26
0.10
0.08
0.23
0.29
0.06
0.16
No. of points with errors greater
than 1.0 cm - 32 23
132
-------
SUMMARY OF MM4-FDDA SIMULATION FOR EPA-RADM
AGGREGATION AND EVALUATION STUDY.
Case No. : 42
Begin Date: 0000 UTC, 8 June 1983
End Date: 0000 UTC, 13 June 1983
Institution: NCAR
Recorded by: P. Haagenson
SI Score, Sea-Level Pres. - 45.8
SI Score, 500-mb Height - 31.7
Table 1. Mean and RMS statistics for primitive variables.
Variable
(units)
Mean Error
RMS Error
Sea-Lvl.
Pressure
(mb)
-0.32
1.63
Sfc.Mix.
Ratio
(g/kg)
0.39
1.06
850-mb
Temp.
(C)
-0.16
0.45
500-mb
Wind Spd.
(m/s)
0.91
1.20
500-mb
Height
(m)
-0.36
10.38
Table 2. 12-hrly Precipitation Statistics.
EPA Precip.
Skill Score
No. of precip. points fcst.
No. of precip. points obs.
Threat score (0.25 cm threshold)
Threat score (0.64 cm threshold)
Bias score (0.25 cm threshold)
Bias score (0.64 cm threshold)
Mean (areal) error (cm)
Mean (areal) absolute error (cm)
Case Mean
232
105
0.25
0.31
3.33
3.25
0.14
0.29
Std. Dev.
76
25
0.17
0.30
2.76
4.25
0.10
0.08
No. of points with errors greater
than 1.0 cm - 14
133
-------
SUMMARY OF MM4-FDDA SIMULATION FOR EPA-RADM
AGGREGATION AND EVALUATION STUDY.
Case No. : 43
Begin Date: 0000 UTC, 15 July 1985
End Date: 0000 UTC, 20 July 1985
Institution: NCAR
Recorded by: P. Haagenson
Si Score, Sea-Level Pres. - 51.6
SI Score, 500-mb Height - 31.5
Table 1. Mean and RMS statistics for primitive variables.
Variable
(units}
Mean Error
RMS Error
Sea-Lvl.
Pressure
Crab")
-0.43
1.83
Sfc.Mix.
Ratio
re/ke^
0.70
1.34
850 -mb
Temp.
(C)
-0.23
0.52
500-mb
Wind Spd.
(m/s)
0.91
1.07
500-mb
Height
Cm")
-2.16
9.98
Table 2. 12-hrly Precipitation Statistics.
EPA Precip.
Skill Score
No. of precip. points fcst.
No. of precip. points obs.
Threat score (0.25 cm threshold)
Threat score (0.64 cm threshold)
Bias score (0.25 cm threshold)
Bias score (0.64 cm threshold)
Mean (areal) error (cm)
Mean (areal) absolute error (cm)
Case Mean
266
204
0.29
0.18
1.24
1.11
-0.01
0.34
Std. Dev.
69
44
0.08
0.10
0.41
0.71
0.11
0.10
No. of points with errors greater
than 1.0 cm - 23 13
134
-------
SUMMARY OF MM4-FDDA SIMULATION FOR EPA-RADM
AGGREGATION AND EVALUATION STUDY.
Case No. : 44
Begin Date: 0000 UTC, 14 December 1985
End Date: 0000 UTC, 19 December 1985
Institution: PSU
Recorded by: D. Stauffer
SI Score, Sea-Level Pres. - 39.8
SI Score, 500-mb Height - 15.1
Table 1. Mean and RMS statistics for primitive variables.
Variable
("units )
Mean Error
RMS Error
Sea-Lvl.
Pressure
(mb)
0.36
2.37
Sfc.Mix.
Ratio
(g/ke)
0.23
0.60
850-mb
Temp.
(C)
-0.05
0.58
500-mb
Wind Spd.
(m/s)
1.31
1.60
500-mb
Height
(m)
3.44
15.73
Table 2. 12-hrly Precipitation Statistics.
EPA Precip.
Skill Score
No. of precip. points fcst.
No. of precip. points obs.
Threat score (0.25 cm threshold)
Threat score (0.64 cm threshold)
Bias score (0.25 cm threshold)
Bias score (0.64 cm threshold)
Mean (areal) error (cm)
Mean (areal) absolute error (cm)
Case Mean
52
77
0.10
0.08
1.24
0.37
-0.03
0.10
Std. Dev.
31
28
0.18
0.00
1.89
0.00
0.03
0.05
No. of points with errors greater
than 1.0 cm
135
-------
SUMMARY OF MM4-FDDA SIMULATION FOR EPA-RADM
AGGREGATION AND EVALUATION STUDY.
Case No. : 45
Begin Date: 0000 UTC, 26 April 1985
End Date: 0000 UTC, 1 May 1985
Institution: PSU
Recorded by: D. Stauffer
SI Score, Sea-Level Pres. - 44.9
SI Score, 500-mb Height - 24.9
Table 1. Mean and RMS statistics for primitive variables.
Variable
(units)
Mean Error
RMS Error
Sea-Lvl.
Pressure
(mb)
-0.31
1.81
Sfc.Mix.
Ratio
(g/kg)
0.37
0.83
850-mb
Temp.
(C)
-0.12
0.40
500-mb
Wind Spd.
(m/s)
0.97
1.18
500-mb
Height
(m)
-1.64
11.66
Table 2. 12-hrly Precipitation Statistics.
EPA Precip.
Skill Score
No. of precip. points fcst. -
No. of precip. points obs. -
Threat score (0.25 cm threshold)
Threat score (0.64 cm threshold) -
Bias score (0.25 cm threshold) -
Bias score (0.64 cm threshold) -
Mean (areal) error (cm) -
Mean (areal) absolute error (cm) -
Case Mean
290
219
0.44
0.29
1.31
1.31
0.05
0.38
Std. Dev.
70
44
0.13
0.09
0.22
0.86
0.06
0.08
No. of points with errors greater
than 1.0 cm - 28
136
-------
SUMMARY OF MM4-FDDA SIMULATION FOR EPA-RADM
AGGREGATION AND EVALUATION STUDY.
Case No. : 46
Begin Date: 0000 UTC, 30 January 1982
End Date: 0000 UTC, 4 February 1982
Institution: PSU
Recorded by: D. Stauffer
SI Score, Sea-Level Pres. - 36.3
SI Score, 500-mb Height - 18.0
Table 1. Mean and RMS statistics for primitive variables.
Variable
(units}
Mean Error
RMS Error
Sea-Lvl.
Pressure
(mb)
0.41
1.90
Sfc.Mix.
Ratio
(g/kg)
0.24
0.62
850-mb
Temp.
(O
-0.06
0.49
500-mb
Wind Spd.
(m/s)
1.19
1.47
500-mb
Height
Cm")
3.47
13.75
Table 2. 12-hrly Precipitation Statistics.
EPA Precip.
Skill Score
No. of precip. points fcst;
No. of precip. points obs.
Threat score (0.25 cm threshold)
Threat score (0.64 cm threshold)
Bias score (0.25 cm threshold)
Bias score (0.64 cm threshold)
Mean (areal) error (cm)
Mean (areal) absolute error (cm)
Case Mean
308
279
0.58
0.54
1.14
1.03
- -0.07
0.35
Std. Dev.
63
67
0.19
0.21
0.17
0.24
0.12
0.13
No. of points with errors greater
than 1.0 cm - 32 22
137
-------
SUMMARY OF MM4-FDDA SIMULATION FOR EPA-RADM
AGGREGATION AND EVALUATION STUDY.
Case No. : 47
Begin Date: 0000 UTC, 7 December 1983
End Date: 0000 UTC, 12 December 1983
Institution: PSU
Recorded by: D. Stauffer
SI Score, Sea-Level Fres. - 42.3
SI Score, 500-mb Height - 17.7
Table 1. Mean and RMS statistics for primitive variables.
Variable
(units)
Mean Error
RMS Error
Sea-Lvl.
Pressure
(rob)
-0.09
2.05
Sfc.Mix.
Ratio
(g/ke)
0.22
0.70
850-mb
Temp.
(C)
-0.05
0.46
500-mb
Wind Spd.
(m/s)
1.21
1.47
500-mb
Height
(m)
0.15
12.80
Table 2. 12-hrly Precipitation Statistics.
EPA Precip.
Skill Score
No. of precip. points fcst.
No. of precip. points obs.
Threat score (0.25 cm threshold)
Threat score (0.64 cm threshold)
Bias score (0.25 cm threshold)
Bias score (0.64 cm threshold)
Mean (areal) error (cm)
Mean (areal) absolute error (cm)
Case Mean
132
141
0.33
0.45
0.69
0.72
-0.10
0.23
Std. Dev.
109
88
0.33
0.27
0.47
0.35
0.10
0.15
No. of points with errors greater
than 1.0 cm - 14 21
138
-------
SUMMARY OF MM4-FDDA SIMULATION FOR EPA-RADM
AGGREGATION AND EVALUATION STUDY.
Case No. : 48
Begin Date: 0000 UTC, 2 November 1985
End Date: 0000 UTC, 7 November 1985
Institution: PSU
Recorded by: D. Stauffer
SI Score, Sea-Level Pres. - 37.1
SI Score, 50Q-mb Height - 20.7
Table 1. Mean and RMS statistics for primitive variables.
Variable
(units)
Mean Error
RMS Error
Sea-Lvl.
Pressure
(mb)
0.18
1.62
Sfc.Mix.
Ratio
(g/kg)
0.38
0.81
850-mb
Temp.
(C)
-0.05
0.36
500-mb
Wind Spd.
Cm/s^
1.03
1.25
500-mb
Height
(m)
2.98
11.32
Table 2. 12-hrly Precipitation Statistics.
EPA Precip.
Skill Score
No. of precip. points fcst.
No. of precip. points obs.
Threat score (0.25 cm threshold)
Threat score (0.64 cm threshold)
Bias score (0.25 cm threshold)
Bias score (0.64 cm threshold)
Mean (areal) error (cm)
Mean (areal) absolute error (cm)
Case Mean
198
182
0.60
0.52
1.01
0.95
0.00
0.33
Std. Dev.
53
40
0.17
0.20
0.22
0.29
0.08
0.15
No. of points with errors greater
than 1.0 cm - 15 11
139
-------
SUMMARY OF MM4-FDDA SIMULATION FOR EPA-RADM
AGGREGATION AND EVALUATION STUDY.
Case No. : 49
Begin Date: 0000 UTC, 7 February 1985
End Date: 0000 UTC, 12 February 1985
Institution: PSU
Recorded by: D. Stauffer
SI Score, Sea-Level Fres. - 35.6
SI Score, 500-mb Height - 20.9
Table 1. Mean and RMS statistics for primitive variables.
Variable
( units)
Mean Error
RMS Error
Sea-Lvl.
Pressure
(mb)
0.17
1.93
Sfc.Mix.
Ratio
f^/ke)
0.19
0.51
850-mb
Temp.
(C)
0.02
0.64
500-mb
Wind Spd.
rm/s")
1.09
1.30
500-mb
Height
Cm")
2.23
13.10
Table 2. 12-hrly Precipitation Statistics.
EPA Precip.
Skill Score
No. of precip. points fcst.
No. of precip. points obs.
Threat score (0.25 cm threshold)
Threat score (0.64 cm threshold)
Bias score (0.25 cm threshold)
Bias score (0.64 cm threshold)
Mean (areal) error (cm)
Mean (areal) absolute error (cm)
Case Mean
152
120
0.41
0.47
1.47
0.74
- -0.01
0.16
Std. Dev.
116
96
0.29
0.27
0.98
0.44
0.09
0.10
No. of points with errors greater
than 1.0 cm - 5 11
140
-------
SUMMARY OF MM4-FDDA SIMULATION FOR EPA-RADM
AGGREGATION AND EVALUATION STUDY.
Case No. : 50
Begin Date: 0000 UTC, 11 November 1983
End Date: 0000 UTC, 16 November 1983
Institution: PSU
Recorded by: D. Stauffer
SI Score, Sea-Level Pres. - 40.6
SI Score, 500-mb Height - 19.9
Table 1. Mean and RMS statistics for primitive variables.
Variable
(units}
Mean Error
RMS Error
Sea-Lvl.
Pressure
Cmb}
0.31
1.81
Sfc.Mix.
Ratio
(zSkz)
0.30
0.77
850-mb
Temp.
(0
-0.06
0.38
500-mb
Wind Spd.
fm/s>
1.17
1.42
500-mb
Height
(in)
3.55
12.31
Table 2. 12-hrly Precipitation Statistics.
EPA Precip.
Skill Score
No. of precip. points fcst.
No. of precip. points obs.
Threat score (0.25 cm threshold)
Threat score (0.64 cm threshold)
Bias score (0.25 cm threshold)
Bias score (0.64 cm threshold)
Mean (areal) error (cm)
Mean (areal) absolute error (cm)
Case Mean
164
150 .
0.37
0.25
0.87
0.42
-0.09
0.23
Std. Dev.
63
68
0.26
0.15
0.60
0.28
0.15
0.13
No. of points with errors greater
than 1.0 cm - 10 14
141
-------
SUMMARY OF MM4-FDDA SIMULATION FOR EPA-RADM
AGGREGATION AND EVALUATION STUDY.
Case No. : 51
Begin Date: 0000 UTC, 2 November 1988
End Date: 0000 UTC, 7 November 1988
Institution: PSU
Recorded by: N. Seaman
SI Score, Sea-Level Pres. - 33.6
SI Score, 500-mb Height - 17.3
Table 1. Mean and RMS statistics for primitive variables.
Variable
(units')
Mean Error
RMS Error
Sea-Lvl.
Pressure
(mb)
0.25
1.73
Sfc.Mix.
Ratio
(e/ke)
0.57
1.14
850-mb
Temp.
(C)
-0.08
0.34
500-mb
Wind Spd.
(m/s)
1.21
1.48
500-mb
Height
(m)
2.36
11.51
Table 2. 12-hrly Precipitation Statistics.
EPA Precip.
Skill Score
No. of precip. points fcst.
No. of precip. points obs.
Threat score (0.25 cm threshold)
Threat score (0.64 cm threshold)
Bias score (0.25 cm threshold)
Bias score (0.64 cm threshold)
Mean (areal) error (cm)
Mean (areal) absolute error (cm)
Case Mean
180
174
0.49
0.42
•
1.33
0.79
-0.14
0.38
Std. Dev.
114
105
0.25
0.23
1.08
0.33
0.38
0.31
No. of points with errors greater
than 1.0 cm - 18 20
142
-------
SUMMARY OF MM4-FDDA SIMULATION FOR EPA-RADM
AGGREGATION AND EVALUATION STUDY.
Case No. : 52
Begin Date: 1200 UTC, 6 November 1988
End Date: 1200 UTC, 11 November 1988
Institution: PSU
Recorded by: N. Seaman
SI Score, Sea-Level Pres. - 41.8
SI Score, 500-mb Height - 18.8
Table 1. Mean and RMS statistics for primitive variables.
Variable
(units)
Mean Error
RMS Error
Sea-Lvl.
Pressure
(mb)
0.30
1.62
Sfc.Mix.
Ratio
(g/kg)
0.50
1.04
850-mb
Temp.
(C)
-0.09
0.36
500-ob
Wind Spd.
(m/s)
1.20
1.45
500-mb
Height
(m)
2.61
11.46
Table 2. 12-hrly Precipitation Statistics.
EPA Precip.
Skill Score
No. of precip. points fcst.
No. of precip. points obs.
Threat score (0.25 cm threshold)
Threat score (0.64 cm threshold)
Bias score (0.25 cm threshold)
Bias score (0.64 cm threshold)
Mean (areal) error (cm)
Mean (areal) absolute error (cm)
Case Mean
121
137
0.27
0.18
0.68
0.61
0.00
0.17
Std. Dev.
70
45
0.23
0.22
0.55
0.74
0.11
0.10
No. of points with errors greater
than 1.0 cm
143
-------
SUMMARY OF MM4-FDDA SIMULATION FOR EPA-RADM
AGGREGATION AND EVALUATION STUDY.
Case No. : 53
Begin Date: 0000 UTC, 11 November 1988
End Date: 0000 UTC, 16 November 1988
Institution: PSU
Recorded by: N. Seaman
SI Score, Sea-Level Pres. - 37.1
SI Score, 500-mb Height - 18.6
Table 1. Mean and RMS statistics for primitive variables.
Variable
(units)
Mean Error
RMS Error
Sea-Lvl.
Pressure
fmb)
0.19
1.73
Sfc.Mix.
Ratio
fe/ke)
0.50
1.04
850-mb
Temp.
(C)
-0,06
0.34
500-mb
Wind Spd.
(m/s)
1.44
1.99
500-mb
Height
(m)
2.46
12.19
Table 2. 12-hrly Precipitation Statistics.
EPA Precip.
Skill Score
No. of precip. points fcst.
No. of precip. points obs.
Threat score (0.25 cm threshold)
Threat score (0.64 cm threshold)
Bias score (0.25 cm threshold)
Bias score (0.64 cm threshold)
Mean (areal) error (cm)
Mean (areal) absolute error (cm)
Case Mean
198
168
0.39
0.34
1.99
0.96
-0.01
0.30
Std. Dev.
108
95
0.23
0.23
1.37
0.67
0.18
0.13
No. of points with errors greater
than 1.0 cm - 15 16
144
-------
SUMMARY OF MM4-FDDA SIMULATION FOR EPA-RADM
AGGREGATION AND EVALUATION STUDY.
Case No. : 54
Begin Date: 1200 UTC, 15 November 1988
End Date: 1200 UTC, 20 November 1988
Institution: PSU
Recorded by: N. Seaman
SI Score, Sea-Level Pres. - 36.1
SI Score, 500-mb Height - 19.1
Table 1. Mean and RMS statistics for primitive variables.
Variable
(units}
Mean Error
RMS Error
Sea-Lvl.
Pressure
fmb")
0.73
1.97
Sfc.Mix.
Ratio
(g/ke)
0.48
1.03
850-mb
Temp.
(C)
-0.08
0.38
500-mb
Wind Spd.
Cm/s")
1.27
1.58
500-mb
Height
Cm^
6.12
14.12
Table 2. 12-hrly Precipitation Statistics.
EPA Precip.
Skill Score
No. of precip. points fcst.
No . of precip . points obs .
Threat score (0.25 cm threshold)
Threat score (0.64 cm threshold)
Bias score (0.25 cm threshold)
Bias score (0.64 cm threshold)
Mean (areal) error (cm)
Mean (areal) absolute error (cm)
Case Mean
249
234
0.53
0.39
1.19
0.74
- -0.11
0.41
Std. Dev.
106
99
0.19
0.19
0.63
0.28
0.11
0.16
No. of points with errors greater
than 1.0 cm - 32 32
145
-------
SUMMARY OF MM4-FDDA SIMULATION FOR EPA-RADM
AGGREGATION AND EVALUATION STUDY.
Case No. : 55
Begin Date: 0000 UTC, 20 November 1988
End Date: 0000 UTC, 25 November 1988
Institution: NCAR
Recorded by: P. Haagenson
SI Score, Sea-Level Pres. - 39.3
SI Score, 500-mb Height - 21.5
Table 1. Mean and RMS statistics for primitive variables.
Variable
(units}
Mean Error
RMS Error
Sea-Lvl.
Pressure
(mb)
0.31
2.04
Sfc.Mix.
Ratio
(e/ke)
0.43
0.98
850-mb
Temp.
(C)
-0.07
0.39
500-mb
Wind Spd.
(n/s)
1.28
1.57
500-mb
Height
(m)
2.30
12.41
Table 2. 12-hrly Precipitation Statistics.
EPA Precip.
Skill Score
No. of precip. points fcst.
No. of precip. points obs.
Threat score (0.25 cm threshold)
Threat score (0.64 cm threshold)
Bias score (0.25 cm threshold)
Bias score (0.64 cm threshold)
Mean (areal) error (cm)
Mean (areal) absolute error (cm)
Case Mean
85
101
0.45
0.39
0.80
0.96
- -0.13
0.26
Std. Dev.
124
118
0.29
0.25
0.49
0.95
0.18
0.16
No. of points with errors greater
than 1.0 cm - 14 25
146
-------
SUMMARY OF MM4-FDDA SIMULATION FOR EPA-RADM
AGGREGATION AND EVALUATION STUDY.
Case No. : 56
Begin Date: 1200 UTC, 24 November 1988
End Date: 0000 UTC, 29 November 1988
Institution: NCAR
Recorded by: P. Haagenson
SI Score, Sea-Level Pres. - 36.4
SI Score, 500-mb Height - 19.0
Table 1. Mean and RMS statistics for primitive variables.
Variable
(units)
Mean Error
RMS Error
Sea-Lvl.
Pressure
(mb)
0.69
2.06
Sfc.Mix.
Ratio
re/kg)
0.44
1.01
850-mb
Temp.
(C)
-0.10
0.44
500-mb
Wind Spd.
(m/s)
1.43
1.76
500-mb
Height
Cm1)
4.85
15.74
Table 2. 12-hrly Precipitation Statistics.
EPA Precip.
Skill Score
No. of precip. points fcst.
No. of precip. points obs.
Threat score (0.25 cm threshold)
Threat score (0.64 cm threshold)
Bias score (0.25 cm threshold)
Bias score (0.64 cm threshold)
Mean (areal) error (cm)
Mean (areal) absolute error (cm)
Case Mean
188
195
0.50
0.39
0.89
1.01
- -0.04
0.26
Std. Dev.
108
110
0.21
0.17
0.41
0.49
0.08
0.11
No. of points with errors greater
than 1.0 cm - 15 16
147
-------
SUMMARY OF MM4-FDDA SIMULATION FOR EPA-RADM
AGGREGATION AND EVALUATION STUDY.
Case No. : 57
Begin Date: 0000 UTC, 16 December 1988
End Date: 0000 UTC, 20 December 1988
Institution: NCAR
Recorded by: P. Haagenson
SI Score, Sea-Level Pres. - 41.8
SI Score, 500-mb Height - 18.4
Table 1. Mean and RMS statistics for primitive variables.
Variable
( units)
Mean Error
RMS Error
Sea-Lvl.
Pressure
(wb)
0.41
2.69
Sfc.Mix.
Ratio
(e/ke)
0.25
0.63
850-mb
Temp.
(C)
-0.10
0.51
500-mb
Wind Spd.
Cm/s")
1.18
1.41
500-mb
Height
(m)
3.63
17.69
Table 2. 12-hrly Precipitation Statistics.
EPA Precip.
Skill Score
Case Mean Std. Dev.
No. of precip. points fcst.
No. of precip. points obs.
Threat score (0.25 cm threshold)
Threat score (0.64 cm threshold)
Bias score (0.25 cm threshold)
Bias score (0.64 cm threshold)
Mean (areal) error (cm)
Mean (areal) absolute error (cm)
No. of points with errors greater
than 1.0 cm
33
54
0.20
30
29
0.16
(not applicable)
0.39 0.28
(not applicable)
-0.05 0.02
0.09 0.03
0 0
148
-------
SUMMARY OF MM4-FDDA SIMULATION FOR EPA-RADM
AGGREGATION AND EVALUATION STUDY.
Case No. : 58
Begin Date: 1200 UTC, 19 December 1988
End Date: 0000 UTC, 23 December 1988
Institution: NCAR
Recorded by: F. Haagenson
SI Score, Sea-Level Pres. - 35.3
SI Score, 500-mb Height - 19.4
Table 1. Mean and RMS statistics for primitive variables.
Variable
(units}
Mean Error
RMS Error
Sea-Lvl.
Pressure
fmM
0.67
1.86
Sfc.Mix.
Ratio
(e/ke}
0.29
0.78
850-mb
Temp.
(C)
-0.04
0.36
500-mb
Wind Spd.
(m/s)
1.32
1.60
500-mb
Height
fm")
4.82
13.98
Table 2. 12-hrly Precipitation Statistics.
EPA Precip.
Skill Score
No. of precip. points fcst.
No. of precip. points obs.
Threat score (0.25 cm threshold)
Threat score (0.64 cm threshold)
Bias score (0.25 cm threshold)
Bias score (0.64 cm threshold)
Mean (areal) error (cm)
Mean (areal) absolute error (cm)
Case Mean
170
136
0.23
0.24
3.65
0.44
- -0.05
0.23
Std. Dev.
80
52
0.21
0.15
5.97
0.26
0.10
0.09
No. of points with errors greater
than 1.0 cm - 10 10
149
-------
150
-------
APPENDIX D
Brief Case Descriptions
151
-------
CASE DESCRIPTION
Case No. 1
Begin Date: 0000 UTC, 7 April 1981
End Date: 0000 UTC, 12 April 1981
Institution: PSU
Case no. 1 began with a broad 1028-mb high pressure system over the
eastern United States, and with a 984-mb low over Lake Winnipeg, Canada, and
a 1002-nib low over Denver, CO. A frontal system connected the low centers.
Southwesterly low-level flow occurred from Texas to the Great Lakes ahead of
the front. At 500-mb, troughs lay over the east coast and the western Great
Plains, with a ridge over the Great Lakes. As the coupled western low
pressure systems moved slowly eastward, moisture was advected from the Gulf
of Mexico toward the Midwest. Moderate rain began over Iowa on 8 April and
spread across the midwest and Great Lakes regions on 9 April, and then to the
northeastern states on 10 April, The MM4/FDDA model handled this period
rather well, with no important phase or amplitude errors in the primitive
variables or the precipitation fields.
On 10 April, a high pressure system built over the midwest in back of
the old frontal system then propagating off the east coast. Precipitation
had mostly ended except for light showers along the coast. The 1200 UTC, 10
April, pressures look much like the initial 7 April pressures. Behind this
second high pressure system, a new frontal boundary and moist southwesterly
flow renewed the flux of moist air from the Gulf of Mexico into the midwest.
Precipitation began over Illinois by 1200 UTC, 10 April, but the model did
not initiate this rainfall until after 1200 UTC. The precipitation pattern
was captured well by the model early on 11 April and through the end of the
simulation. However, rainfall totals appear to be too great during the final
24 h and caused some degradation of the average precipitation statistics;
nearly all of the occurrences of errors greater than 1.0 cm occurred at this
last day and caused the standard deviation of that statistic to exceed its
mean. Nevertheless, the accuracy of the primitive variables remained
uniformly good throughout the simulation.
The overall assessment of the simulation for case 1 is that MM4/FDDA
performed very well and the simulation results are of good quality and are
suitable for input to RADM. The overproduction of precipitation during the
final 24 h is not extreme, but unfortunately is concentrated in a narrow band
from Iowa to New Jersey.
Assessment by:
Nelson L. Seaman
152
-------
CASE DESCRIPTION
Case No. : 2
Begin Date: 1200 UTC, 11 April 1981
End Date: 0000 UTC, 15 April 1981
Institution: PSU
As the simulation began on April 11, 1981, a very broad but relatively
weak 500-mb ridge lay over the eastern two-thirds of North America, with a
trough along the Pacific coast. The polar jet was mostly zonal with winds of
40 m s"1 from the Great Lakes to Sable Island. A cold front stretched
southwestward from a 1004-mb low in Quebec to a 1006-mb low in eastern
Colorado, with another cold front southward into western Texas. The Canadian
low moved steadily eastward to the Atlantic Ocean on April 12, while the'
western low drifted very slowly eastward. The frontal boundary between the
two lows became mostly quasi-stationary from Kansas City to Delaware, with
widespread moderate to heavy rains along the front. The model did very well
in reproducing the area and intensity of the precipitation during April 12-
13.
Late on April 12, a new frontal system began moving east from the
northern Rockies into the Great Plains, with a 1004-mb low in Canada and a
weaker 1008-mb low moving east from Wyoming. This system was followed by a
major cold high pushing southeast into the Plains. Initially, there were
only isolated light showers accompanying this new system on 12-13 April. The
model also produced only very light precipitation. Late on April 13, the new
cold front overtook the older low in western Kansas and began to accelerate
and develop that storm along the leading edge of the advancing Canadian high.
A moist southerly flow over the old quasi-stationary front continued to
support the major rain region from the midwest to the east coast. The model
correctly developed the fronts, highs and lows and to reproduced the main
features of rainfall intensity and location.
On the last day, April 14, the cold front advanced eastward from the
Great Plains nearly to the east coast. The Kansas storm moved rapidly
northeastward across Michigan to O^iebec, while deepening to 994 mb. As it
moved east, this system tapped the moist southerly flow from the Gulf of
Mexico and pushed the old stationary front north as a warm front. The
advancing storm produced heavy rains throughout the Midwest and into New
England. MM4/FDDA simulated the rainfall quite well throughout the period.
The overall assessment of the Case 2 simulation is that the model did an
excellent job of reproducing both the synoptic meteorological structures of
highs, lows and fronts, and the precipitation fields throughout the forecast
period. The model results should be quite suitable for use in RADM.
Assessment by:
Nelson L. Seaman
153
-------
CASE DESCRIPTION
Case No. 3
Begin Date: 0000 UTC, 20 April 1981
End Date: 0000 UTC, 25 April 1981
Institution: PSU
Case no. 3 began on April 20 with a 1007-mb low in north central
Oklahoma accompanied by a front lying east-northeastward across the lower
Ohio River Valley toward Virginia. A 1038-mb high was centered in western
Ontario west of James Bay. At 500 mb, a ridge lay over the western Great
Plains and a long-wave trough lay over eastern Canada from Hudson Bay to Nova
Scotia. Moderate precipitation fell along and north of the frontal boundary
from the Great Plains to the east coast. The MM4/FDDA model correctly
carried the front southward to the Gulf of Mexico coast and eastward into the
western Atlantic on April 21 ahead of the advancing Canadian high. Observed
precipitation along the front became less intense, but was somewhat
overpredicted by the model.
By April 22, a new frontal system was beginning to push east from the
Rocky Mtns., with a 997-mb low along the Canadian border of North Dakota and
a 1000-mb low in eastern Colorado. The model reproduced this new development
well, including the outbreak of precipitation in advance of the storms and
front. By April 23, the two lows had consolidated into a 1000-mb low in
Wisconsin, with rainfall organized by a cold front trailing through the
southern Plains and a warm front in the upper Midwest. This system deepened
to 994 mb and occluded near Green Bay later on April 23. The model did an
excellent job of reproducing the movement and intensification of the storm
and its frontal patterns. Rainfall also intensified during this period and
the model simulated the pattern and amounts quite well. On April 24, the
occluded low became quasi-stationary over Michigan and slowly weakened, while
a secondary storm developed over the Northeast coast as the frontal system
continued eastward. The model continued to match the observed phase and
gradually weakening intensity of the precipitation on this final day of the
simulation, until only scattered showers were left in New England and
southern Texas by 0000 UTC, April 25.
The overall assessment of the simulation for Case 3 is that MM4/FDDA
performed very well. The precipitation statistics are of high quality,
especially because the vigorous frontal systems provided excellent
organization of the rainfall patterns. There are no outstandingly poor
periods during this simulation.
Assessment by:
Nelson L. Seaman
154
-------
CASE DESCRIPTION
Case No.: 4
Begin Date: 0000 UTC, 12 July 1980
End Date: 0000 UTC, 17 July 1980
Institution: PSU
On 12 July 1980, a broad 500-mb ridge was located over the central U.S.
with troughs over the Pacific Northwest and New England. A weak polar jet
extended over the ridge along the Canadian border. At the surface, a 998-mb
low lay just north of Maine with a cold front stretching south through New
England to central North Carolina. A second system consisted of a 1003-mb
low in Manitoba with a front southward to a 1006-mb low in southern Minnesota
and then southwest to Trinidad, Colorado. A warm front extended southeast
from the latter low. These systems drifted slowly eastward on July 12,
bringing isolated showers to regions near the weak frontal boundaries. By
July 13, the low in Minnesota had moved southeastward to Kentucky with its
frontal system roughly west to east from Missouri to the western Atlantic.
The MM4/FDDA model correctly simulated the movement of the lows and fronts,
and produced showers of approximately correct intensity and location near
lows and fronts.
As the forecast proceeded during July 14-16, lows, highs and fronts were
simulated very well. A third storm formed just east of northern Rockies and
moved eastward toward the Great Lakes. Convection continued as isolated
showers, but the model gradually began to overpredict the area and amount of
precipitation, especially toward the end of the simulation.
The overall assessment of the Case 4 simulation is that the model
reproduced the movement and intensity of the principal pressure systems and
frontal boundaries rather well, but had more difficulty with the
predominantly isolated convective precipitation. The locations of the small-
scale observed rainfall areas are extremely difficult for the model to
reproduce using its 80-km resolution and radiosonde-scale data assimilation
system (supplemented by higher density surface moisture and winds). This
problem is compounded by the short 12-hr precipitation verification period.
These shortcomings cause the statistical verification for this case to be
among the poorest of the cases simulated. However, given the extreme
difficulty of predicting isolated summertime convection, this may still be an
acceptable simulation for the RADM (Regional Acid Deposition Model).
Assessment by:
Nelson L. Seaman
155
-------
CASE DESCRIPTION
Case No. 5
Begin Date: 1200 UTC, 16 July 1980
End Date: 0000 UTC, 20 July 1980
Institution: PSU
When Case 5 began on 16 July 1980, a 998 mb low was leaving the
northeastern corner of the model domain (northern Newfoundland) , with a cold
front trailing southwestward through northern New England to Lake Erie. A
1005-mb low was moving eastward along the northern border of Minnesota, with
its cold front curving southwestward to the Texas panhandle. At 500-mb, a
weak polar jet lay east-west just south of the Canadian border. Light to
moderate convection was observed in scattered locations along the Gulf coast
and from Minnesota to Pennsylvania and New England. Observed rainfall for
Case 5 was consistently isolated and often triggered by one of the frontal
systems, but failed to organize into large-scale bands.
During July 17-18, the leading system left the domain and the model
correctly simulated the eastward progress of the second storm across the
Great Lakes and toward southern Labrador. A third wave formed and began
moving east from the northern Rockies across the upper Great Plains.
However, the model tended to produce heavier and more widespread rainfall
than observed, particularly near lows and along fronts. These problems are
reflected in the precipitation verification statistics, which indicate that
Case 5 is one of the poorest simulations encountered. On July 19, the second
frontal system was leaving the eastern side of the domain, while the third
system continued eastward across southern Canada and the western Great Lakes.
The model continued to simulate the phase and intensity of these systems
rather well. The observed convection became lighter and more isolated on
this last day, while the model continued to overestimate the area, but did
decrease the rainfall intensity.
The assessment of the Case 5 simulation is that, while reproducing the
phase and intensity of the principal systems, MM4/FDDA tended to overpredict
the amount and coverage of the precipitation. The very small scale of the
observed rainfall regions made this an extremely difficult case to reproduce
accurately. However, a parallel MM4 experiment performed without FDDA showed
considerably less skill in both the large-scale and precipitation statistics.
Much experience with this model strongly suggests that the inaccuracy of the
precipitation fields is not due to any correctable problems with the initial
fields, but is more likely due to insufficient horizontal resolution and an
inability to reproduce the subtle local convergence and stability
perturbations responsible for the isolated showers.
Assessment by:
Nelson L. Seaman
156
-------
CASE DESCRIPTION
Case No. 6
Begin Date: 0000 UTC, 27 January 1982
End Date: 0000 UTC, 1 February 1982
Institution: NCAR
Case No. 6 began with a large 1031-mb high pressure system over the
eastern United States and a 982-mb low centered just north of eastern Montana.
The surface low was coupled with a minor upper-level disturbance, and the
associated frontal structure was rather weak. The low pressure center tracked
eastward across southern Canada and by 1200 UTC, 29 January, the cold front
that extended southward from the surface low had moved off the east coast of
the United States. Very little precipitation was observed or simulated in
conjunction with this frontal system.
On 29 January, a closed upper-level trough moved into the southern Rocky
Mountain states and triggered development of a 999-mb low over western Texas
by 0000 UTC, 30 January. As this storm tracked northeastward, it produced
considerable precipitation on 30 and 31 January (especially along the
associated frontal zone) over much of south-central and eastern United States.
The precipitation amount and pattern were simulated very well by the MM4/FDDA
model except for its early stage of development on 1200 UTC, 29 January. The
phase and amplitude errors in the primitive variables were also small.
The overall assessment of the simulation for case 6 is that MM4/FDDA
performed very well and the results are suitable for input to RADM. The
precipitation threat and bias scores (except for those on 29 January, 1200
UTC) are extremely favorable.
Assessment by:
Phil Haagenson
157
-------
CASE DESCRIPTION
Case No. 7
Begin Date: 0000 UTC, 18 March 1982
End Date: 0000 UTC, 23 March 1982
Institution: NCAR
Case No. 7 began with a broad 1020-mb high pressure system centered over
the northern Great Lakes region. A weak quasi-stationary front, defining the
leading edge of the cooler air associated with the Canadian anticyclone,
extended from Wyoming to North Carolina. On 0000 UTC, 19 March, a 995-mb low
(in association with a developing upper-level disturbance) formed over western
Colorado, and by 0000 UTC on 20 March, it had moved eastward into Nebraska.
Significant amounts of precipitation were observed in the cold air north of a
surface warm front. The MM4/FDDA model simulated the precipitation amounts
fairly well but over predicted the spatial extent in the warm air sector.
During the following 48-h period the low tracked slowly eastward into the
northeastern United States, and by the end of the simulation period it had
moved out of the model domain. The precipitation amount and pattern were
simulated quite well at the 0.25 cm threshold but not so well at the 0.64 cm
threshold. The most significant error occurred in Texas, Louisiana, and
Arkansas at 1200 UTC, 20 March and 0000 UTC, 21 March. The model predicted
precipitation along a cold front extending southward from the low in Indiana--
none was observed. The phase and amplitude errors in the primitive variables
however were small.
The overall assessment of the simulation for case 7 is that the MM4/FDDA
performed very well and the results are suitable for input to RADM. The
precipitation threat and bias scores (except for those on 20 March, 1200 UTC
and 21 March, 000 UTC) are very favorable.
Assessment by:
Phil Haagenson
158
-------
CASE DESCRIPTION
Case No. 8
Begin Date: 0000 UTC, 27 June 1982
End Date: 0000 UTC, 2 July 1982
Institution: NCAR
Case No. 8 began with a minor high pressure cell over the eastern United
States and a 1010-mb low centered over Oklahoma. The surface low was
associated with a weak upper-level trough. By 1200 UTC, 29 June, the frontal
structure associated with the low pressure system was reinforced by a Canadian
anticyclone moving southward into the northern Great Plains states.
Considerable amounts of precipitation were observed in the warm air sector
(southeastern United States) on 27 and 28 June, and also south of the leading
edge of the cold front on 29 June. The MM4/FDDA model simulated the
precipitation amount and pattern fairly well during the most of the period
from 27 to 29 June except for 1200 UTC, 28 June when the model over predicted
both the amount and spatial extent.
During the latter part of the simulation period (30 June - 2 July) the
low tracked northeastward into eastern Canada allowing the Canadian
anticyclone to move into the eastern United States. Since this high pressure
system was characterized by advection of drier air from the northwest, most of
the observed precipitation was in the warm (apparently unstable) air south of
the cold front. The model simulated the movement of the dry cold air very
well as evidenced by the simulated and observed precipitation at 1200 UTC, 30
June. The model had some difficulty with the precipitation simulation during
the final 12 hours of the case study period.
The overall assessment of the simulation for case 8 is that MM4/FDDA
performed quite well and the results are suitable for RADM. Although the SI
scores are larger than for Winter cases, the phase and amplitude errors in the
primitive variables were small. The precipitation threat and bias scores
(except for those on 28 June, 1200 UTC and 2 July, 0000 UTC) are quite
favorable.
Assessment by:
Phil Haagenson
159
-------
CASE DESCRIPTION
Case No. 9
Begin Date: 0000 UTC, 6 August 1982
End Date: 0000 UTC, 11 August 1982
Institution: NCAR
Case No. 9 began with a weak pressure gradient over most of the central
and eastern United States. A quasi-stationary front extended from a 1015-mb
low over Wyoming to Virginia. Some precipitation was observed and simulated
along the front and also in moist unstable air south of the front. On 1200
UTC, 7 August, an upper-level disturbance generated a surface low over Lake
Winnipeg, Canada. By 1200 UTC, 8 August, the cold front associated with the
low pressure system produced a band of precipitation from the Great Lakes to
Oklahoma. The MM4/TDDA model simulated this band fairly well, but it also
predicted warm sector precipitation in Tennessee, Kentucky and Virginia that
was not observed.
During the latter part of the simulation period (9-11 August) the low
tracked eastward across southern Canada, and the cold front moved slowly to
the east coast of the United States. Meanwhile, the southern extent of the
front became quasi-stationary over the southeastern states. The large amounts
of precipitation were observed and simulated in the unstable warm air south
and east of the cold front. The precipitation amount and pattern were
simulated very well by the model on 9 and 10 August. However, the model had
some difficulty with the precipitation simulation during the final 12 hours of
the case study.
The overall assessment of the simulation for case 8 is that the MM4/FDDA
performed quite well and the results are suitable for RADH. Although the SI
scores are larger than for the Winter cases, the phase amplitude errors in the
primitive variables were small. The precipitation threat and bias scores
(except for those on 8 August and perhaps 11 August, 0000 UTC) are reasonably
favorable.
Assessment by:
Phil Haagenson
160
-------
CASE DESCRIPTION
Case No. 10
Begin Date: 0000 UTC, 19 August 1988
End Date: 0000 UTC, 24 August 1988
Institution: NCAR
Case No. 10 began with a large high pressure cell located over the
northern Great Lakes, and a quasi-stationary surface front situated along the
southern boundary of the high pressure system. Considerable amounts of
precipitation were observed and simulated along the frontal zone (on both the
north and south side) during the first 72 hours of the case study period. By
0000 UTC, 22 August the front had moved southward into the southeastern United
States, and a Pacific cold front (associated with an upper-level trough) began
to produce precipitation in the northern Great Plains. At this time, while
precipitation was also occurring in the southeastern United States, most of
the northeastern states were under the influence of dry northeasterly flow and
anticyclonic subsidence. The MM4/FDDA model simulated the precipitation
patterns and amounts quite well for most of the 72-hour period.
During the latter part of the simulation period (22-24 August) the
Pacific front became very active and produced heavy amounts of precipitation
along the frontal zone, particularly on the east side of the front. The
heaviest amounts of precipitation were observed and simulated in Iowa,
Illinois, and Missouri. The threat and bias scores at the 0.25 and 0.64 cm
threshold were very good and the simulated amounts also verified quite well.
The most notable exception was at 0000 UTC, 23 August when the model predicted
9.7 cm in northern Missouri, but only 5.0 cm was observed.
The overall assessment of the simulation for case 10 is that HM4/FDDA
performed quite well and the results are suitable for RADM. The SI scores are
typical of Summer cases, and the phase and amplitude errors in the primitive
variables were small. The precipitation threat and bias scores (0.25 and 0.64
threshold) were better than those for most of the other Summer cases processed
so far.
Assessment by:
Phil Haagenson
161
-------
CASE DESCRIPTION
Case No. 11
Begin Date: 0000 UTC, 28 August 1988
End Date: 0000 UTC, 2 September 1988
Institution: NCAR
Case No. 11 began with a broad upper-level trough located over the
northern Great Plains. A 1002-mb surface low was situated over southern
Ontario from which a surface cold front extended southward into Texas. By
0000 UTC, 31 August, the upper-level trough had moved eastward into the
northeastern United States and the surface front had moved off the east coast.
Modest amounts of precipitation were observed and simulated along the eastward
moving surface front during the first 72-hour period. Heavier amounts of
precipitation were observed and simulated (> 2.5 cm) on 29 August in portions
of the warm air sector of the storm system (North Carolina, South Carolina,
and Virginia). The threat and bias scores at the 0.25 and 0.64 cm threshold
were quite good on 29 and 30 August. On the 31st, however, the MM4/FDDA model
predicted heavy amounts of precipitation in Mississippi that did not verify.
During the latter part of the simulation period (1-2 September) the
weather in most of the central and eastern United States was dominated by a
large anticyclone. Very little precipitation was observed except in
Minnesota, northern Texas, and the extreme southern and southeastern United
States. The precipitation in Minnesota was induced by a surface cold front
that extended southward from southern Ontario to Colorado. The model
over-predicted both the amount and spatial extent of the frontal induced
precipitation. The threat and bias scores, however, were not greatly
affected. The mean threat scores (0.25 and 0.64 cm threshold) were - 0.2 and
the mean bias scores - 1.8.
The overall assessment of the simulation for case 11 is that the MM4/FDDA
performed quite well and the results are suitable for RADM. The SI scores
were better than those for some of the other Summer cases, and the phase and
amplitude errors in the primitive variables were small. The precipitation
threat and bias scores at the 0.25 and 0.64 cm threshold were generally good
except for those on 31 August and 0000 UTC, 1 September. However, on
1 September, very little precipitation was observed or simulated.
Assessment by:
Phil Haagenson
162
-------
CASE DESCRIPTION
Case No. 12
Begin Date:
End Date:
Institution:
Case No. 12 began with a surface cold front and a weak low-pressure
trough located over the northern Great Plains. By 1200 UTC, 3 September, the
surface low was deepening (1008 mb) and was located over the Southern Great
Lakes region. The upper-level trough associated with the storm system was
also intensifying. Initially, the heaviest amounts of precipitation were
observed in Texas and Oklahoma in southerly flow in advance of the cold front.
The MM4/FDDA model simulated the precipitation location and amounts fairly
well, but over predicted the spatial extent. On 3 September, precipitation
was observed and simulated along the frontal zone from the lower Great Lakes
to southern Texas. Most of the precipitation occurred in the warm sector,
south and east of the front. The threat and bias scores at the 0.25 and 0.64
cm threshold were better on 3 September, after the storm began to intensify,
than on 2 September.
During the last 24 hours of the simulation period, the surface low
continued to deepen (from 1008 to 1003 mb) and was located over Lake Erie.
The surface cold front extended southward from the low pressure center to
Louisiana. Considerable amounts of precipitation were observed and simulated
over a large part of eastern and southeastern United States. Again, most of
it occurred in the warm sector of the storm system. The threat and bias
scores at the 0.25 and 0.64 cm threshold on 4 September were very good.
Threat scores exceeded 0.5 and bias scores ranged from 0.85 to 1.15. The
excellent scores probably resulted because much of the precipitation on 4
September wad induced by large-scale dynamics.
The overall assessment of the simulation for Case No. 12 is that MM4/FDDA
performed very well, especially during the time of storm intensification, and
the results are suitable for RADM. The SI scores were good for a Summer case
and the phase and amplitude errors in the primitive variables were quite
small. The threat and bias scores were also good, especially during the last
24 hours of the case study period.
Assessment by:
Phil Haagenson
163
-------
CASE DESCRIPTION
Case No. 13
Begin Date: 0000 UTC, 4 September 1988
End Date: 0000 UTC, 9 September 1988
Institution: NCAR
Case No. 13 began with a closed upper-level low and a surface low located
near Lake Michigan. A surface front (associated with the cyclonic
circulation) extended southward from southern Ontario to Texas. Moderate
amounts of precipitation were observed and simulated along the frontal zone in
the southeastern and eastern United States (particularly in the warm air
sector of the storm system) during the first 48 hours of the simulation
period. By 0000 UTC, 6 September, the low had moved into southeastern Canada
and the surface front extended southward along the east coast from the low
pressure center to northern Florida. The threat and bias scores at the 0.25
and 0.64 cm threshold for the MM4/FDDA simulation were very good on 4 and 5
September, and also at 0000 UTC, 6 September. Most of the threat scores were
> 0.4, and the bias scores ranged between 0.75 and 1.0.
During the latter part of the simulation period (6-9 September) a frontal
wave developed, and stalled, along the southeast coast of the United States.
Consequently, heavy amounts of precipitation were simulated over the water and
in the state of Florida (> 10.0 cm in some areas). Since most of the
precipitation was over water, the only "observable" location was in Florida
where precipitation amounts > 5.0 cm were reported on 6, 7, and 9 September.
Near the end of the case study period a surface cold front moved from
northwestern United States into the Great Plains. However, very little
precipitation was observed or simulated with the system.
The overall assessment of the simulation for case 13 is that the MM4/FDDA
performed very well and the results are suitable for RADM,. The SI scores
were very good and the phase and amplitude errors in the primitive variables
were small. The threat scores at the 0.25 and 0.64 cm threshold were very
good (> 0.45 overall), and the bias scores were fairly good except for those
on 8 September. However, on the 8th almost all of the precipitation (over
land) was confined to extreme southeastern United States.
_==^. Assessment by:
Phil Haagenson
164
-------
CASE DESCRIPTION
Case No. 14
Begin Date: 1200 UTC, 8 September 1988
End Date: 1200 UTC, 12 September 1988
Institution: NCAR
Case No. 14 began with a weak surface front (very little precipitation)
located over the Northern Great Plains and a stationary front over extreme
southeastern United States. A tropical depression was located in the Gulf of
Mexico near 25° N latitude. By 1200 UTC, 9 September, the tropical depression
was upgraded to a minor hurricane which moved onshore near New Orleans,
Louisiana, on 10 September. During the first 36 hours of the simulation
period, large amounts of precipitation were observed along the stationary
front from Florida to North Carolina. The MM4/FDDA model simulated the
amounts and location very well. The threat and bias scores at the 0.25 and
0.64 cm threshold were excellent. Threat scores ranged from 0.60 to 0.75, and
bias scores ranged from 0.90 to 1.20. The hurricane produced moderate amounts
of precipitation in the southern parts of Louisiana, Mississippi, and Alabama.
The observed amounts, however, were less than predicted by the model.
During the latter part of the simulation period (11-12 September) a new
storm system moved into the northern Great Plains. Minor amounts of
precipitation were observed in the cold sector to the rear of the surface cold
front southward from Minnesota to New Mexico. The model over predicted both
the amount and spatial extent. Advection of moist unstable air, on the west
side of a quasi-stationary high located over eastern United States, also
produced some precipitation in the south-central states. The threat and bias
scores at the 0.25 and 0.64 threshold were not very good during the last 36
hours of the case study period. Threat scores ranged from 0.07 to 0.27 and
some of the bias scores exceeded 3.0.
The overall assessment of the simulation for Case No. 14 is that MM4/FDDA
performed quite well (very well during the first 60 h) and the results are
suitable for RADM. The SI scores were better than for most of the Summer
cases and the phase and amplitude errors in the primitive variables were
small. The threat and bias scores were excellent during the first 48 hours of
the case study period, but were not nearly as good during the last 36 hours.
Assessment by:
Phil Haagenson
165
-------
CASE DESCRIPTION
Case No. 15a
Begin Date: 0000 UTC, 12 September 1988
End Date: 1200 UTC, 16 September 1988
Institution: NCAR
Case No. 15a began with a strong anticyclone (1032 mb) centered over
Montana and a low-pressure trough over the Great Plains. The cold front
associated with the surface low moved rapidly eastward, and by 1200 UTC, 14
September, it was over the Atlantic ocean. Most of the storm-related
precipitation occurred in the warm sector with the heaviest amounts reported
from Alabama northeastward into Pennsylvania. The MM4/FDDA model simulated
the precipitation quite well in most areas except for a couple of locations.
For example, the model predicted the observed precipitation very accurately in
the Ohio Valley, but did not predict the heavy precipitation that was observed
in Alabama. A weak upper-level disturbance also produced precipitation over
the southern Great Plains that was simulated fairly well by the model. The
threat and bias scores at the 0.25 and 0.64 cm threshold were reasonably good
except for 0000 UTC, 14 September when precipitation amounts were light, and
the observed and simulated locations did not coincide.
During the last 48 hours of the simulation period, the synoptic pattern
was dominated by a large quasi-stationary high-pressure cell over eastern
United States. The circulation of hurricane Gilbert coupled with the air flow
pattern on the west side of the anticyclone resulted in large-scale advection
of moist unstable air from the Gulf of Mexico into the southern United States
and the central Great Plains. Observed and simulated precipitation amounts
exceeded 2 cm in Texas, Oklahoma, Kansas, Missouri, and Iowa. The threat and
bias scores at the 0.25 and 0.64 cm threshold were quite good. Threat scores
(except for 0000 UTC, 15 September) ranged from 0.25 to 0.55 and bias scores
ranged from 1.05 to 1.75. The direct impact of hurricane Gilbert was
reflected during the last 12 hours of the case study period by simulated
precipitation amounts over water that exceeded 13 cm. The observed
precipitation over land (southeastern Texas), however, suggests that the model
over predicted the amounts.
The overall assessment of the simulation for Case No. 15a is that
MM4/FDDA model performed well and the results are suitable for RADM. The SI
scores were quite good and the phase and amplitude errors in the primitive
variables were small. The threat and bias scores were also quite good except
for those at 0000 UTC, on 14 and 15 September.
Assessment by:
Phil Haagenson
166
-------
CASE DESCRIPTION
Case No. 15b
Begin Date: 0000 UTC, 16 September 1988
End Date: 1200 UTC, 19 September 1988
Institution: NCAR
Case No. 15b began with a large anticyclone over the eastern United
States and hurricane Gilbert approaching the eastern coast of northern Mexico
and southern Texas. By 1200 UTC, 17 September, the high pressure cell had
moved over the Atlantic ocean and hurricane Gilbert was weakening and located
over northern Mexico. A new storm system was also developing over the
northern Rocky Mountain region. The circulation of hurricane Gilbert coupled
with the air flow pattern on the west side of the anticyclone resulted in
large-scale advection of moist unstable air from the Gulf of Mexico into
southern central United States. Observed precipitation exceeded 3 cm in
Texas, Oklahoma, Mississippi, Alabama, Tennessee, and Georgia. The MM4/FDDA
model simulated the precipitation amount and pattern quite well except for
1200 UTC, 16 September when the model predicted moderate precipitation in
Arkansas and Georgia that was instead observed in Oklahoma and Alabama. Also,
precipitation amounts in southern Texas (due to Hurricane Gilbert) were
significantly over-predicted. The threat and bias scores at the 0.25 and 0.64
cm threshold were fairly good on 16 September and very good on 17 September
when the threat scores exceeded 0.5.
During the latter part of the simulation period (18-19 September), an
upper-level disturbance intensified over the northern Great Plains. This
development generated a surface low over the Dakotas and Minnesota that
deepened from 1002 to 994 mb during the last 24 hours of the period.
Considerable precipitation occurred in both warm and cold sectors of the storm
system with the heaviest amounts observed in the warm sector. The cold sector
amounts were > 2 cm and warm sector amounts were > 4 cm. The model simulation
of the precipitation was, for the most part, very good. Threat scores at the
0.25 and 0.64 cm threshold ranged from 0.4 to 0.6 and bias scores ranged from
0.8 to 1.35.
The overall assessment of the simulation for Case No. 15b is that
MM4/FDDA performed well and the results are suitable for RADM. The Si scores
were reasonably good (especially for sea-level pressure) and the phase and
amplitude errors in the primitive variables were quite small. The threat and
bias scores were very good, but the number of grid points with prediction
amount errors exceeding 1 cm was larger than usual.
Assessment by:
Phil Haagenson
167
-------
CASE DESCRIPTION
Case No. 16
Begin Date: 0000 UTC, 19 September 1988
End Date: 0000 UTC, 24 September 1988
Institution: NCAR
Case No. 16 began with a strong surface and upper-level storm system
located over the northern Great Plains. By 1200 UTC, 21 September, the
surface low had moved northeastward into southeastern Canada, and the
associated cold front was located along the east coast of the United States.
Considerable amounts of precipitation occurred in both the warm and cold
sectors of the storm before it moved into Canada. The MM4-FDDA model
simulated the precipitation amount and pattern very well on 19 September and
fairly well on 20 September. The threat scores at the 0.25 and 0.64 threshold
ranged from 0.2 and 0.6, and the bias scores ranged from 1 to 1.5 except for
0000 UTC, 21 September when the bias scores were > 2.
During the last 60 hours of the simulation period, a low pressure trough
developed over the Dakotas, deepened from 1000 to 989 mb, and moved eastward
into southeastern Canada. Light to moderate amounts of precipitation were
observed in the warm sector and along the frontal zone of the storm. The
model simulation of the precipitation was fairly good particularly near the
end of the case study period when it predicted the location and shape of the
frontal zone precipitation and position of the "dry line" north of the surface
cold front. The threat and bias scores at the 0.25 and 0.64 cm threshold were
fairly good except for 0000 UTC, 22 September, when precipitation amounts were
light and the threat scores were < 0.1.
The overall assessment for Case No. 16 is that MM4/FDDA performed
reasonably well and the results are suitable for RADM. The SI scores were
typical of the late Summer cases and the phase and amplitude errors in the
primitive variables were small. The threat and bias scores were fairly good
except for those at 0000 UTC on 21 and 22 September.
Assessment by:
Phil Haagenson
168
-------
CASE DESCRIPTION
Case No. 17
Begin Date: 1200 UTC, 23 September 1988
End Date: 1200 UTC, 28 September 1988
Institution: NCAR
Case No. 17 began with a 991 mb low over southern Hudson Bay and an
associated cold front extending southward from the low pressure center into
Arkansas. By 1200 UTC, 25 September, the surface front has become quasi-
stationary over the southern United States and another weak storm system (very
little precipitation) was located just north of the Great Lakes. Most of the
precipitation during the first 48 hours was light to moderate and occurred in
the cold air sector along the quasi-stationary front. The MM4/FDDA model
simulated the amount and pattern very well. Threat scores at the 0.25 and
0.64 cm threshold ranged from 0.15 to 0.6, and most of the bias scores ranged
from 1.1 to 1.5
During the latter part of the simulation period (26-28 September), the
precipitation was light and scattered. The upper-level winds over the central
and eastern states were generally from the west or northwest resulting in
advection of dry stable air. The pattern did not change until the last day of
the simulation period when an upper-level disturbance intensified over the
Great Plains and induced surface cyclogenesis along the leading edge of a cold
air mass from Canada. Considerable amounts of precipitation were observed and
simulated in the cold air sector (Nebraska, South Dakota, and Iowa) during the
last 12 hours of the case study period. The model, however, predicted that
the precipitation would begin 12 hours earlier than was observed. The threat
and bias scores at the 0.25 and 0.64 cm threshold were not very good on 26 and
27 September because precipitation amounts were light.
The overall assessment of the simulation for Case No. 17 is that MM4/FDDA
performed quite well and the results are suitable for RADM. The SI scores
were similar to the other September cases, and the phase and amplitude errors
in the primitive variable were (as usual) small. The threat and bias scores
were fairly good except for those on 26 and 27 September when the
precipitation was light and scattered.
Assessment by:
Phil Haagenson
169
-------
CASE DESCRIPTION
Case No. 18
Begin Date: 0000 UTC, 13 April 1982
End Date: 0000 UTC, 18 April 1982
Institution: NCAR
Case No. 18 began with a 992 mb surface low located over Wisconsin.
During the first 48 hours of the simulation period the low, which was
associated with a deepening upper-level trough, moved rapidly eastward into
southeastern Canada. Most of the precipitation (amounts were not heavy)
occurred along the frontal zones and in the cold air sectors of the storm
system. The largest amounts were reported in northern Georgia and Alabama.
The MM4/FDDA model simulated the precipitation amounts reasonably well, but
over predicted the spatial extent and also missed the exact location of the
frontal-induced precipitation in Georgia and Alabama.
During the latter part of the simulation period (16-18 April)
considerable amounts of precipitation occurred in conjunction with a surface
cold front that moved slowly eastward from central to eastern United States.
Heaviest amounts were observed along the front and in the warm all-sector
(east and south of the front). The precipitation pattern was simulated quite
well by the model except for over prediction of the spatial extent on 17
April. On the other hand, precipitation amounts were generally under
predicted.
The overall assessment for case 18 is that MM4/FDDA performed
satisfactorily and the results are suitable for RADM. The SI scores were good
and the phase and amplitude errors in the primitive variables were very small.
The precipitation threat and bias scores were favorable, particularly during
most of the last 60 hours of the simulation period when most of the
precipitation occurred.
Assessment by:
Phil Haagenson
170
-------
CASE DESCRIPTION
Case No. 19
Begin Date: 0000 UTC, 23 Sept. 1982
End Date: 0000 UTC, 28 Sept. 1982
Institution: PSU
Case no. 19 was initialized with a broad 1023-nib high over northern
Mississippi which dominated the U.S. from the Great Plains to the east coast.
A front lay just east of the coast from a 1013-mb low southwest of Nova
Scotia. A weakening 1005-mb low was moving southeast toward the Great Lakes
just south of Lake Winnipeg, with a slowly moving cold front extending
southwest to eastern CO. At 500 mb, a closed low was drifting slowly
eastward across Ohio and a major long-wave trough was entrenched over the
eastern U.S. During the first 24 h, a few light showers fell over the Mid-
Atlantic and New England states as a low formed beneath this upper-level
system. Showers also fell across Minnesota and Wisconsin with the Great
Lakes low. MM4/FDDA reproduced this early-period meteorology and rainfall
quite well.
On Sept. 24, the 500-mb low filled and lifted northeastward. A 1028-mb
high entered the northern Great Plains. Light showers continued with the
Great Lakes low and front as they moved east of the Mississippi River on
Sept. 25. The 500-mb trough position remained just east of the Mississippi.
By late on Sept. 26, the surface system had disappeared east of the
Appalachian Mts., but moderate rains fell along the southeast coast as the
weak disturbance tapped moist maritime air. By this time, another low (1000
mb) was moving east from Lake Winnipeg with a. cold front extending through
the northern U.S. Plains states, but virtually no rain accompanied this
system. However, southerly winds behind the preceding ridge and ahead of the
cold front began advecting warmer moist air northward from the Gulf of Mexico
on Sept. 26. MM4/FDDA did rather well at reproducing this sequence of
events.
On Sept. 27, a 1007-mb low reformed in eastern Virginia along the old
frontal system then lying along the east coast. As this low moved north
toward northern New York during the final 24 h, it tapped the moist air over
the western Atlantic and brought fairly heavy rains (2-4 cm) from the
southeast to New England and as far west as eastern Indiana. Meanwhile, a
deep 994-mb storm had formed over the Rocky Mts. and moved across Wyoming to
the Great Plains behind the westernmost frontal system. As this storm
reached the Plains, it tapped the moisture streaming north from the Gulf
states and moderate rains spread across the Dakotas and the Rocky Mt. states.
The model reproduced these events well and precipitation forecasts are quite
suitable for RADM.
Assessment by:
Nelson Seaman
171
-------
CASE DESCRIPTION
Case No. 20
Begin Date: 0000 UTC, 13 December 1982
End Date: 0000 UTC, 18 December 1982
Institution: NCAR
Case No. 20 began with a large high pressure system over central United
States. By 0000 UTC, 14 December, a 998 mb low developed over eastern
Colorado in response to an upper-level trough approaching the west. By 1200
UTC, 15 December, this low was located over the northern Great Lakes, and a
secondary low (frontal wave cyclogenesis) had formed over eastern Arkansas.
Very little precipitation was observed or simulated in conjunction with the
northern-region low, but significant amounts of precipitation occurred in the
vicinity of the frontal wave. The MM4/FDDA model under predicted the amounts
that were reported along the frontal zone and missed the exact location of the
maximum amounts.
During the latter part of the simulation period (16-18 December) the
secondary low deepened from 1014 to 999 mb and moved rapidly northeastward
into eastern Canada. Heavy amounts of precipitation were observed (and
simulated by the model) along the southward-extending cold front from 0000
UTC, 16 December to 0000 UTC, 17 December. The threat and bias scores were
excellent. By the end of the simulation period, most of central and eastern
United States was again dominated by a large high pressure system. Advection
of dry stable air (to the rear of the upper-level trough that triggered the
earlier cyclogenesis) produced only small amounts of precipitation during the
last 24 hours of the simulation period, and very little was simulated in the
verification region.
The overall assessment for case 20 is that MM4/FDDA performed very well
and the results are suitable for RADM. The SI scores were good and the phase
and amplitude errors in the primitive variables were very small. The
precipitation threat and bias scores were very good, particularly during the
latter part of the simulation period when the threat scores at the 0.25 and
0.64 cm threshold were outstanding (> 0.7).
Assessment by:
Phil Haagenson
172
-------
CASE DESCRIPTION
Case No. 21
Begin Date: 0000 UTC, 27 May 1983
End Date: 0000 UTC, 1 June 1983
Institution: NCAR
Case No. 21 began with a high pressure cell centered over the Great
Lakes, and a rather weak surface low over North Dakota. By 1200 UTC, 29 May,
the low was coupled with a well-developed upper-level trough over the Great
Lakes. The heaviest amounts of precipitation were observed in the warm air
sector of the storm system on the 29th (Kansas and Oklahoma), and in the cold
air sector on the 29th (Wisconsin and Michigan). Considerable amounts of
precipitation were also reported along the warm front in Alabama and
Tennessee. The MM4/FDDA model simulated the amounts and spatial extent
reasonably well. The most significant error occurred for the 12-hour period
ending at 0000 UTC, 28 May, when the model produced heavy precipitation in
Iowa and Missouri.
During the latter part of the simulation period (30 May - 1 June), the
upper-level trough remained nearly stationary over the Great Lakes, and
consequently, the surface low and associated surface fronts moved rather
slowly. Most of the precipitation occurred in the warm air sector east of the
cold front, and in the cold air sector north and east of the warm front and
occluded front. The model simulation did quite well except for over
prediction of precipitation (near the end of the case study period) in the
cold air to the rear of the upper-level trough. Large amounts of
precipitation were also observed and simulated north of a surface cold front--
in Texas and Oklahoma on 31 May.
The overall assessment for Case 21 is that MM4/FDDA performed quite well
and the results are suitable for RADM. The SI scores were fairly good
(typical for a Spring-Fall case), and the phase and amplitude errors in the
primitive variables were small. The threat scores were good and the bias
scores very good for most of the time periods. The most notable exceptions
were for those at 0000 UTC on 28 May and 1 June.
Assessment by:
Phil Haagenson
173
-------
CASE DESCRIPTION
Case No. 22
Begin Date: 0000 UTC, 2 August 1983
End Date: 0000 UTC, 7 August 1983
Institution: NCAR
Case No. 22 began with a high pressure cell over central United States.
A weakening cold front, located just east of the high pressure cell, was
moving slowly toward the east coast. Small amounts of precipitation that were
observed along the cold front were under predicted by the MM4-FDDA model.
Some sections of the southeastern coastal states (under the influence of
advection of moist unstable air from the Caribbean and Gulf of Mexico)
reported modest amounts of precipitation during the first 48 hours of the
simulation period. The model simulated the precipitation amounts and pattern
fairly well.
During the latter half of the case study period (4-7 August), a weak
quasi-stationary front that extended from Illinois to New England induced
light amounts of precipitation. The model simulated the precipitation pattern
fairly well,but over predicted the amounts. The model had some difficulty
with the precipitation simulation near the end of the case study, particularly
for the 12-hour period ending at 1200 UTC, 6 August. For that period the
threat and bias scores at the 0.25 threshold were 0.01 and 4.83, respectively.
Examination of the observed precipitation analysis for all time periods
shows a pronounced diurnal oscillation in the amounts. This suggests that a
significant part of the precipitation was not induced by large-scale dynamic
forcing. The surface and upper-level synoptic patterns support this
hypothesis.
The overall assessment of the simulation for case 22 is that MM4/FDDA
performed fairly well and the results are suitable for RADM. Although the SI
scores are larger than for the Winter cases, the phase and amplitude errors in
the primitive variables were small. The precipitation threat and bias scores
were reasonably good for most of the time periods. Notable exceptions,
however, were for those at 1200 UTC on 5 and 6 August.
Assessment by:
Phil Haagenson
174
-------
CASE DESCRIPTION
Case No. 23
Begin Date: 0000 UTC, 7 Sept. 1983
End Date: 0000 UTC,12 Sept. 1983
Institution: PSU
Case no. 23 began with a 500-mb trough centered over the Great Lakes and
a weak upper-level closed low over Lake Superior. A 993-mb low lay east of
James Bay, with a cold front south to New York and then southwest to the
Texas panhandle. Moderate showers fell along the front and the Gulf of
Mexico coast. Relatively weak highs were located over the Southeast and the
central Great Plains. A second cold front was advancing east to the Plains
from the northern Rocky Mts.
These systems progressed slowly eastward on Sept. 7 and 8 with the
leading cold front passing off the New England coast early on the second day.
Showers continued along this front from Virginia to Oklahoma, and along the
Gulf coast. MM4/FDDA produced generally similar precipitation patterns, but
the isolated character of the showers led to fairly low pattern verification
(threat scores). As the second cold front advanced across the Plains on the
Sept. 8 and 9, southerly winds from the Gulf advected moist air northward
ahead of this system. MM4/FDDA correctly spread showers inland from the
Texas coast and initiated showers over Montana and Minnesota. Elsewhere,
high pressures dominated the eastern U.S. By Sept. 10, a wave developing
along the northern front had deepened to 997 mb over Lake Winnipeg, with an
occlusion to Lake Superior and fronts southwest across the Plains and
eastward to northern Maine. MM4/FDDA correctly placed the heaviest rain in
Minnesota and South Dakota, with light showers elsewhere in the verification
domain. This pattern moved slowly eastward through the rest of the period
with the heaviest frontal rains occurring on Sept. 11 across the Midwestern
states and the central Plains.
MM4/FDDA simulated the meteorological patterns very well and also
captured the general precipitation patterns. As during the early part of the
period, some details of the rainfall were in error at any given time because
of the spotty patterns. However, these should average out reasonably well
over the full period, so that the impact on RADM simulations of the episode
should be minor. Overall, the model did well with a difficult precipitation
event.
Assessment by:
Nelson Seaman
175
-------
CASE DESCRIPTION
Case No. 24
Begin Date: 0000 UTC. 30 October 1983
End Date: 0000 UTC, 4 November 1983
Institution: PSU
Case no. 24 began with a 500-mb trough over the Canadian Maritime
provinces and a broad upper-level ridge over the western U.S. The initial
surface pattern was dominated by a 1036-mb high over Michigan which covered
virtually the entire country. No fronts were found over the U.S. This high
moved east to Delaware in the first 24 h, and suppressed precipitation except
for a small patch of showers over Missouri. MM4/FDDA simulated this day very
well, although another small shower,area was created over the Ohio Valley
late on October 30. The high stagnated over the east coast on October 31 and
remained very dry. Although MM4/FDDA correctly simulated the pressure and
wind patterns, it spread some spurious isolated light showers from the Texas
panhandle to Missouri and over parts of the Midwest.
By Nov. 1, a weak 1014-mb low began moving east across the central
Plains, accompanied by a north-south trough. Southerly flow behind the old
high, then centered south of Nova Scotia, advected moisture from the Gulf of
Mexico into the Plains states. Moderate showers began over Iowa on Nov. 1
and spread on the following day. Although the low dissipated, the central
states from Kansas to Ohio experienced rain on Nov. 1 and 2 in the broad
southerly flow. MM4/FDDA did rather well in simulating this rainfall
pattern, although a few spurious areas of showers were simulated in other
areas of the upper Midwest. Totals were quite reasonable.
During the final 24 h, a new 1039-mb high approached the upper Great
Lakes from central Canada, pushing a cold front into the Midwest. This front
became the focus of continued showers as it swept the humid air out of the
Midwest along with the retreating high-pressure ridge. Rain was most intense
from New England through the Ohio Valley to Kansas and northward to the
Dakotas. MM4/FDDA simulated this pattern reasonably well, but over-expanded
another area of showers in the Rio Grande valley and New Mexico.
Generally, the meteorological model reproduced most aspects of this
episode very well, including the precipitation fields. A few periods with
very little observed precipitation had more simulated rain, which contributed
to a high episodic-mean bias score. Despite such problems, the resulting
fields should be quite acceptable for use in RADM.
Assessment by:
Nelson Seaman
176
-------
CASE DESCRIPTION
Case No. 25
Begin Date: 0000 UTC, 12 September 1983
End Date: 0000 UTC, 17 September 1983
Institution: PSU
Case no. 25 began with very weak pressure gradients over the U.S. A
weak cold front was moving slowly eastward from the New England coast. Weak
low pressures were found from the lower Mississippi Valley through the Ohio
Valley. At 500 mb, flow was mostly zonal, with a weak jet stream along the
U.S.-Canadian border. Moderate rains were occurring along the Gulf coast on
Sept. 12, with another line of scattered showers from Kansas to the Ohio
Valley. MM4/FDDA did reasonably well at simulating the rain pattern on Sept.
12, although it failed to predict an isolated area of rain in the western
Great Lakes. On Sept. 13, a broad Canadian high invaded the Great Plains
states, creating cold frontogenesis across the south. The advancing front
gradually pushed the areas of moderate rain to the south and southeast. The
model reproduced this trend rather well, but could not simulate the finer
details in the rain pattern. By Sept. 14, the advancing cold high was over
Minneapolis with its cold front pushing to the Carolinas coast. Heaviest
rains were confined to the southeast and were well simulated by MM4/FDDA.
By Sept. 14, a broad 500 mb trough had formed over the Great Lakes and a.
1011 mb low was dropping southeastward into eastern Wyoming behind the high
over the Plains. Only a few showers were produced with this system. The
model correctly deepened the low to 1008 mb as it moved to the lee of the
Rocky Mts. on Sept. 15, and showers broke out from North Dakota to Nebraska.
MM4/FDDA simulated the expansion of the rain region, but extended it south
through Kansas 12 h too soon. By early on Sept. 16, the 1008 mb low had
moved to southern Minnesota with a cold front southwest to northern Texas and
a warm front to western Ohio. Moderate rains swept in an arc ahead of this
system from eastern North Dakota southward to eastern Oklahoma. The model
produced a very similar system (1009 mb low), including a very reasonable
precipitation forecast. During the final 24 h, on Sept. 16, the low deepened
very slowly as it moved eastward across the upper Great Lakes. Rains covered
most of the Lakes and the Ohio Valley. MM4/FDDA continued to simulate this
pattern well, although it could not reproduce all fine-scale details
accurately.
The meteorological driver did a credible job of simulating this case.
The pressure pattern in the model developed very nearly as observed and most
of the precipitation fields were quite accurate. The model fields should be
very suitable for RADM.
Assessment by
Nelson Seaman
177
-------
CASE DESCRIPTION
Case No. 26
Begin Date: 0000 UTC, 15 March 1984
End Date: 0000 UTC, 20 March 1984
Institution: PSU
When Case no. 26 began, a 500-mb trough lay over New England oriented
southward toward the Bahamas. Another weak trough lay over the western Rocky
Mts. and a broad ridge was centered just east of the Mississippi Valley. At
the surface, a 1011-mb low was located just southwest of Nova Scotia followed
by a 1032-mb high over western New York. A 1000-mb low was over western
Kansas with a frontal system from northern Minnesota to west Texas. This
western storm filled to 1006 mb during the next 12 h as it moved east across
Kansas and moderate precipitation fell over the western Great Lakes, centered
on Wisconsin. MM4/FDDA did not capture the precipitation pattern well at the
outset of this simulation, placing the rain over Iowa and north of Lake
Superior.
The western low moved rapidly northeastward to Illinois at the beginning
of March 16, with a 1040-mb high advancing southward into the Great Plains
from Canada. Moderate to heavy precipitation was centered from Michigan to
Arkansas, moving eastward. The model rapidly corrected its rain pattern
toward that of the observed fields. However, the observed storm was 5 mb
deeper (1007 mb) than was simulated. By early on March 17, the front was
pushing off the east coast ahead of a massive 1050-mb high centered west of
James Bay. A new weak storm along the Rocky Mt. Front Range produced
moderate precipitation over the Great Plains, which was simulated reasonably
by MM4/FDDA.
By March 18, a digging jet streak amplified the western 500-mb trough
and led to strong cyclogenesis over New Mexico. As this system developed, it
caused southerly flow from the Gulf of Mexico to overrun an old stationary
front lying over the Gulf states. Precipitation was widespread over the
Plains states and Mississippi Valley. At 0000 UTC, MM4/FDDA also placed the
precipitation in these regions well ahead of the storm, but failed to
reproduce the details of the distribution. However, the "observed"
precipitation analysis at 1200 UTC on March 18 apparently is unreliable.
because it inexplicably shows virtually no rain with this vigorous storm at
this time, despite heavy precipitation just before and afterward. On March
19, the storm (999 mb) moves across Oklahoma and Missouri, with heavy
precipitation continuing ahead of the storm. While missing a. few details,
MM4/FDDA simulated most aspects of the precipitation patterns correctly
during this final day. Also, another storm developing near Bermuda on March
18 moved northwest on the last day, producing moderate to heavy precipitation
over New England. The model reproduced this feature very well.
In summary, for limited periods, MM4/FDDA had some difficulty simulating
the observed storm intensities and some details of fairly complex
precipitation systems. The overall patterns during the episode, however,
were reasonable. This is confirmed by the model Threat Scores, which were
about average. Furthermore, there were obvious data problems in the
precipitation analyses used for verification, which degraded the statistical
comparison. At 1200 UTC, March 18, the model precipitation simulation is
178
-------
almost certainly more accurate than the analysis. Thus, despite some
problems, the fields produced by MM4/FDDA should be generally suitable for
input to RADM. However, care should be exercised in the interpretation of
results from this case.
Assessment by:
Nelson L. Seaman
179
-------
CASE DESCRIPTION
Case No. 27
Begin Date: 0000 UTC, 19 August 1984
End Date: 0000 UTC, 24 August 1984
Institution: PSU
At the beginning of Case no. 27, a front lay from western New York to
north Texas with two weak lows in northern Ohio and northeast Oklahoma. A
broad 500-mb trough lay over the east coast and a ridge over the western
Great Plains. On August 19, isolated showers were found ahead of the lows and
along the east coast. MM4/FDDA reproduced these showers reasonably well, but
missed some mesoscale details because of the small area covered by the
showers. On August 20, the original frontal system moved eastward and a new
cold front advanced eastward across the northern Great Plains, accompanied by
light to moderate showers. The model reproduced these changing patterns
fairly well, but tended to create too much rain with the new northern front.
By August 21, the original front had drifted off the east coast and a
new low had formed over the Oklahoma panhandle. Moderate to heavy showers
stretched from this low to the northern front nearing Lake Superior. The
model simulated this rain rather well, but by the beginning of August 22, it
produced too much rain in Kansas and Oklahoma. During August 22, a new
Canadian high began pushing slowly into the northern Great Plains, which
gradually swept the fronts and rain south and east. On August 23, the
frontal systems became weakly connected from the Great Lakes to the southern
Plains, but the model continued to simulate too much rain over the southern
end of the system.
These weak lows and fronts continued east and south on the last day of
the episode as the high moved southeast into the western Great Lakes and
Great Plains. The fronts and rain were located along the east coast and the
Gulf states by the end of the period. MM4/FDDA did reasonably well at
simulating this organized precipitation.
In summary, the statistics show that MM4/FDDA did about as well as can
be expected for a summertime episode. However, examination of the 12-hourly
precipitation fields shows that simulation of numerous mesoscale details were
beyond the capability of the model. The meteorological fields for this case
should be acceptable for RADM, but results could be somewhat degraded by
errors in the model precipitation fields. The fields should be used
cautiously and RADM results interpreted carefully.
Assessment by:
Nelson Seaman
180
-------
CASE DESCRIPTION
Case No. 28
Begin Date: 0000 UTC, 10 June 1984
End Date: 0000 UTC, 15 June 1984
Institution: PSU
Case no. 28 began with a 500-mb trough over the Rocky Mts. and a ridge
over the east coast. A 1005-mb low was moving northeast from Omaha toward
the Great Lakes, with a warm front to Toledo, Ohio, and a cold front to west
Texas. Some isolated heavy showers were falling along the cold front and a
broad band of moderate rain fell north of the low from Idaho to Lake Huron.
MM4/FDDA reproduced this rain pattern quite well. By June 11, the low had
moved into southern Canada with most of the rain located over the upper Great
Lakes, while a new storm began to form over the Great Basin and move east to
Colorado. The model began spreading rain east of the mountains a bit too
early, but this was quickly corrected.
Early on June 12, a 1001-mb low was located just east of Denver with a
trough and front system running roughly parallel to the Rocky Mts. Most of
the rain with this system was falling from Idaho to Iowa. MM4/FDDA missed
some details of the rain pattern, but simulated the basic pattern rather
well. The previous low and frontal system had moved to the east coast by
this time and were not producing important rain. By the end of June 12, the
western system had moved northeast with a 1009-mb low in Minnesota. Rain
fell from the northern Plains to the western Great Lakes by June 13. The
model moved the rain with the low and its fronts, but on June 13, it produced
spurious areas of heavy rain over northwestern Kansas and lighter rain
across Michigan, while failing to simulate heavy rain in Iowa. Isolated
showers also occurred over the southeastern states through most of the
period. MM4/FDDA produced isolated showers in the same region, but often
missed the exact locations. This is not unexpected because of the small
scale of these showers.
During the last day, the low moved eastward from the upper Great Lakes
across southern Canada. Rains fell across the Midwest and the Northeast
during its passage. Meanwhile, widespread isolated showers occurred across
the southeast and the entire Great Plains to the Rocky Mts. The model did a
fairly good job of simulating this difficult precipitation pattern, but could
not reproduce many of the mesoscale details.
Generally, the meteorology of this difficult case was simulated rather
well despite the isolated nature of the summertime rainfall. Many details of
the 12-hourly rain pattern were missed, but major failures were rare. The
results should be acceptable for RADM, but should be interpreted cautiously.
Assessment by:
Nelson Seaman
181
-------
CASE DESCRIPTION
Case No. 29
Begin Date: 0000 UTC, 7 September 1984
End Date: 0000 UTC, 12 September 1984
Institution: PSU
Case no. 29 began with a 5448 m closed low at 500 mb located just north
of Montana in a long wave trough over the Rocky Mts. Another 500-mb trough
lay over the east coast and a ridge lay over the eastern Great Plains. A
986-mb low was found north of Montana with an occluded front extending east
and south to a 991-mb low over South Dakota. The second storm had a warm
front to Missouri and a cold front to New Mexico. A 1027-mb high was
located over the eastern states, centered in Pennsylvania.
The storm system filled slowly as it moved east on Sept. 7 and 8, but
produced only isolated rain over the U.S. MM4/FDDA simulated the showers
rather well. On Sept. 8, a low formed over the Bahamas and moved slowly
northwest. This low eventually spread rain from central Florida north along
the coast. By Sept. 9, the storm originally in the northern Rocky Mts. was
north of Minnesota (1002 mb), with light rain over the upper Great Lakes.
Another weak low formed in Oklahoma. About this time, heavy rain broke out
over Missouri and Oklahoma, which the model simulated very well. However,
too much rain was produced in central Florida as the southeastern low
approached the east coast. On Sept. 10, the northern low continued northeast
to James Bay with a trailing cold front that brought more rain to the Great
Lakes. The weak low in the southern Great Plains moved little, with
moderate rains continuing in Arkansas and Missouri. MM4/FDDA simulated the
rain reasonably well during this period, but failed to capture some of the
mesoscale details of the pattern.
By Sept. 11, the southern low had deepened to 1000 mb, but had moved
very little. Moderate rains shifted to Iowa and northern Missouri. The
model simulated the maximum position well, but produced too much rain and
covered too large an area. The coastal storm was located off Charleston,
with moderate rain along the coast of the Carolinas. The model also
simulated this area well, although it produced somewhat heavier rainfall.
Later on the last day, weak frontogenesis occurred from the Oklahoma low
northeastward through the Ohio Valley. Heavy rains were renewed in Missouri
and Illinois, which the model simulated well. By the end of the period, the
rain shifted to New York and New England, and rains rapidly diminished over
the Mississippi Valley. MM4/FDDA captured this change reasonably well, but
was a bit slow to end the rains in the west.
In summary, this case was simulated very reasonably and the
meteorological fields should be suitable for RADM.
Assessment by:
Nelson Seaman
182
-------
CASE DESCRIPTION
Case No. 30
Begin Date: 0000 UTC, 14 July 1984
End Date: 0000 UTC, 19 July 1984
Institution: NCAR
Case No. 30 began with a surface front and weak low-pressure trough over
the northern Great Plains. By 0000 UTC, 16 July, the surface low had deepened
(998 mb) and was located over the southern Hudson Bay. The surface cold front
extended southward from the low pressure center through the Ohio Valley and
then westward into Oklahoma. The heaviest amounts of precipitation (induced
by the storm system) were observed on the 15th in Iowa, Illinois, and Northern
Missouri. The MM4/FDDA model simulated the precipitation amount and location
quite well except for 0000 UTC, 15 July when the model predicted too much
precipitation in Iowa (1.6 cm was observed and 4.1 cm was predicted). The
threat and bias scores at the 0.25 and 0.64 cm threshold were reasonable good
except for 0000 UTC, 16 July when the model simulated widespread light
precipitation in the warm air sector south and east of the surface front--only
scattered amounts were reported.
During the latter part of the simulation period (16-19 July) an upper-
level disturbance intensified over the Great Lakes region. This development
generated a surface cold front that produced modest amounts of precipitation
as it moved southeastward from the Great Plains to the east coast of the
United States. Heaviest amounts of precipitation were reported along the
frontal zone and in the warm air sector east and south of the front. The
model simulation of the precipitation was fairly good, particularly near the
end of the case study period when the model accurately handled the eastward
movement and location of a "dry air line" to the rear of the surface cold
front. The threat and bias scores at the 0.25 and 0.64 cm threshold were also
very good (particularly for a Summer case) during the last 24 hours of the
simulation period.
The overall assessment of the simulation for Case No. 30 is that the
MM4/FDDA performed quite well and the results are suitable for RADM. The SI
scores were better than those for some of the other Summer cases, and the
phase and amplitude errors in the primitive variables were small. The threat
scores (0.25 and 0.64 threshold) were fairly good except for those on 0000
UTC, 16 July. The bias scores, except for those on 0000 UTC, 16 July, were
very good.
Assessment by:
Phil Haagenson
183
-------
CASE DESCRIPTION
Case No. 31
Begin Date: 0000 UTC, 31 October 1985
End Date: 0000 UTC, 5 November 1989
Institution: PSU
When Case no. 31 began, a 500-mb closed low was centered near New
Orleans beneath a broad 500-mb ridge over the U.S. A weak westerly jet
followed close to the Canadian border. A 1025-mb surface high was moving
slowly east through Ontario, while a 991-mb low was found at New Orleans,
with a warm front along the Gulf of Mexico coast and an inverted trough north
into Mississippi. Widespread moderate rain fell throughout the Southeast on
October 31, especially in Arkansas and northern Florida. A 998-mb low was
weakening as it moved east through Alberta. MH4/FDDA simulated the
precipitation pattern very well during this early part of the episode.
On Nov. 1, the deep Gulf low drifted northeast from the Florida
panhandle. The inverted trough extended further north through Illinois,
spreading rain to Wisconsin. The model correctly simulated these changes.
By Nov. 2, the southern low had occluded, weakened to 997 mb and drifted
north into Indiana. The 500-mb low had filled by this time, and an upper-
level trough was digging into the northern Great Plains. A new 1005-mb low
formed in Montana in response to the renewed upper-level forcing. Rain
shifted away from the Gulf coast in an arc from Wisconsin to Cape Hatteras,
where a secondary 1000-mb low formed along the east coast. MM4/FDDA
simulated these changes very well.
By Nov. 3, the precipitation had decreased as the storm systems
weakened. Low pressures were oriented from a 1009-mb low in upper Michigan
southward along a cold front to Georgia and to a new 1006 mb low in the
Florida panhandle. During the last 48 h, the 500-mb trough deepened in the
Mississippi Valley, causing the southern storm to intensify and move
northeast to the Carolinas on Nov. 4. This re-intensified the precipitation,
especially from the eastern Carolinas to Pennsylvania. In the final 12 h,
nearly 9 cm (3.5 in.) of rain fell in West Virginia. MM4/FDDA simulated
these changes in the precipitation and pressures very well, producing 6.4 cm
(2.5 in.) over eastern West Virginia in the last 12 h.
Overall, the timing and intensity of the upper-level dynamics, the low-
level systems and the precipitation were simulated extremely well for this
case. No major errors occurred throughout the period. The statistical
verification indicates this to be one of the very best simulations and
results should be fully suitable for RADM.
Assessment by:
Nelson Seaman
184
-------
CASE DESCRIPTION
Case No. 32
Begin Date: 0000 UTC, 10 July 1985
End Date: 0000 UTC, 15 July 1985
Institution: PSU
Case no. 32 began with a weak 500-mb trough north of the Great Lakes and
an upper ridge over the Rocky Mts. A cold front was moving east through
eastern Canada and New York. A 1007-mb low was drifting east from Illinois
through the Midwest, accompanied by moderate rains. MM4/FDDA began the
period with a very reasonable simulation of pressure and rains associated
with the low and the front, although the totals were less than observed. On
July 11, a new weak low formed over the central Great Plains, but only a few
showers accompanied this system. The model tended to produce larger areas of
rain than the isolated convective showers observed, but smaller rainfall
totals. By July 12, the cold front in the northeast had pushed off the east
coast, leaving nearly flat pressure gradients except for a weak 1005-mb low
over South Dakota. The model simulated these pressure changes rather well.
Rain was extremely disorganized and covered only very small areas. During
the middle of the simulation, the model was unable to match the areas of
isolated showers.
By July 14, the western system had become somewhat better organized with
waves along a front stretching from James Bay southwestward to New Mexico.
However, only a few showers fell along this front until it moved through the
Great Lakes and the Northeast late on July 14. Showers were spread east of
the Mississippi, and became more widespread and organized in the Northeast on
the last day. MM4/FDDA continued to have some difficulty simulating the most
isolated areas of rainfall correctly, but showed noticeable improvement in
predicting the better organized rain associated with the front in the
Northeast.
The model consistently showed the greatest overprediction of the rain
areas during the daytime. While the model was generally unable to match the
specific sites of the isolated showers at the 12-h time scale, the general
regions of rainfall for the episode were reasonable, so that the net acid
deposition occurring in RADM may be fairly accurate. However, the results
of RADM should be carefully examined for this case because of the problems
that may be introduced by the model's rainfall timing and distribution.
Assessment by:
Nelson Seaman
185
-------
CASE DESCRIPTION
Case No. 33
Begin Date: 0000 UTC, 30 April 1985
End Date: 0000 UTC, 5 May 1985
Institution: NCAR
Case No. 33 began with a closed upper-level low over the central Great
Plains, and a surface front extending southwestward from a surface low in
southern Ontario to a surface low located over Colorado. By 1200 UTC, 2 May,
the upper-level low had weakened and was over the Ohio Valley while the
associated surface low was located slightly south of that region.
Considerable amounts of precipitation (up to 4.0 cm) were observed and
simulated in the warm air sector south and east of the frontal zone, and also
in the cold air sector north of the surface warm front. The MM4/FDDA
simulated the precipitation amounts and pattern very well. Some of the threat
scores at the 0.25 and 0.64 cm threshold were > 0.5. Bias scores ranged from
0.7 to 1.5.
During the latter part of the simulation period (3-5 May), the surface
front moved eastward off the east coast of the United States and a Canadian
anticyclone, that produced dry stable conditions, moved into the central and
eastern states. Modest amounts of precipitation were observed and simulated
in the eastern and southeastern United States on 3 May, but only small amounts
were observed or simulated in the verification network during the final 24
hours of the case study period. The threat and bias scores (0.25 and 0.64 cm
threshold) on the 3rd were excellent, and they remained fairly good to the end
of the simulation period even though precipitation amounts were small.
The overall assessment of the simulation for Case 33 is that the MM4/FDDA
performed very well and the results are suitable for RADM. The SI scores were
fairly good (typical for a Spring-Fall case), and the phase and amplitude
errors in the primitive variables were small. The threat scores (0.25 and
0.64 threshold) were excellent for almost all time periods and the bias scores
were above average.
Assessment by:
Phil Haagenson
186
-------
CASE DESCRIPTION
Case No. 34
Begin Date: 0000 UTC, 14 November 1985
End Date: 0000 UTC, 19 November 1985
Institution: PSU
Case no. 34 began with a 500-mb trough in eastern Canada north of a
flattened upper-level ridge along the U.S. east coast. An eastward-tilting
500-mb trough lay over the Rocky Mts. At the surface, a frontal zone
stretched from Nova Scotia to southwest Texas, with a 1015-mb wave over St.
Louis. The western ridge of a Bermuda high lay south of the front, while to
the north a broad cold ridge stretched from a 1031-mb high in Washington east
to New England. Rain fell along the front from Texas to Ohio. On Nov. 14-
15, the wave tracked northeast through New England and moderate rains fell
from New England to Texas. On Nov. 15, a second frontal wave developed in
Oklahoma, which enhanced the rain along the western portion on the front.
MM4/FDDA simulated the wave movement and the rainfall very well, except that
too little rain was predicted in New York and northern New England.
On Nov. 16, the northern ridge built eastward with a 1038-mb high
centered northeast of Lake Huron, which pushed the original front east of the
continent. The second low moved northeast behind the high to Wisconsin,
where it occluded. This system spread moderate rains from the Great Lakes to
Texas. The model simulated the rainfall very well. The occlusion became
quasi-stationary and filled over Lake Superior on Nov. 17, while the
trailing front and rains pushed east across the Appalachian Mts. The
heaviest rains fell from Pennsylvania to southern New England, while most of
the precipitation ended across the South. The model did very well at
reproducing these changes.
By Nov. 18, the eastern frontal system pushed well east of the coast,
and only light rain remained over New England. A new. cold front had formed
over the western Great Plains ahead of another Canadian high then pushing
southward in Alberta. Ahead of the southern end of this front, a wave formed
and southerly flow from the Gulf of Mexico brought moist warm air into the
eastern Plains and Mississippi Valley. Moderate to heavy rains spread from
Louisiana to the Great Lakes. The model correctly simulated the outbreak of
precipitation and the frontal wave development, but was slow to develop the
heavier rainfall that was observed. Areal coverage was reasonable.
In summary, the model simulated the progression of fronts and storms
very well. The accompanying precipitation was forecasted with reasonable
accuracy, except for brief periods. However, the model output should be
quite suitable for RADM.
Assessment by:
Nelson Seaman
187
-------
CASE DESCRIPTION
Case No. 35
Begin Date: 0000 UTC, 24 September 1985
End Date: 0000 UTC, 29 September 1985
Institution: PSU
Case no. 35 began with a deep 500-mb trough over the Great Plains and
ridges over the Pacific and Atlantic coasts. A 988-mb low over Lake Superior
was intensifying and moving north toward Hudson's Bay. A weak 1013-mb low
was found east of Delaware, while high pressure dominated the Rocky Mts. and
southern Plains. Rain was falling over the Great Lakes and east of the
Mississippi River along the north-south cold front from the primary storm.
This front pushed to the east coast late on Sept. 24, with most rain ending
except over New England. MM4/FDDA simulated this pattern well in most
respects.
On Sept. 25, a new front and storm system developed in the lee of the
Rocky Mts. and pushed slowly east across the Great Plains. Rain broke out
with this system from Oklahoma to Iowa. The model simulated this development
well, but also placed some rain in northeast Texas. The low deepened to
1006-mb on Sept. 26 and moved northeast to Detroit, spreading rains from the
Texas coast to the Great Lakes. The model correctly simulated the expansion
of the precipitation area and its movement east with the associated front,
but generally failed to simulate the heaviest observed totals. At the same
time a tropical storm was approaching the Carolinas from the southeast.
Elsewhere, a low and frontal system began moving east from the Canadian
Rockies.
On Sept. 27, the tropical storm hit Cape Hatteras and moved rapidly
north toward western New England. Heavy rains, up to 9.5 cm/12 h, fell from
North Carolina to New England. MM4/FDDA simulated the area of rain very
well, but failed to match the heaviest totals (model maximum of 4.5 cm/12h).
The model's pressure intensity for the storm was also too weak. There was
little rain accompanying the low moving across the northern Great Plains at
this time.
On Sept.- 28, the tropical storm moved from New England northward into
Canada and rains ended over the northeast. In the west, a cold front
stretched from the Canadian low north of Lake Superior to Texas. Regions of
rain broke out on the last day over Texas, along the front, and over the
Rocky Mts. The model simulated the increase of rain, but because the
precipitation tended to be fairly isolated and convective, there were many
errors in simulating the right positions of the mesoscale maxima.
In summary, the progression of storms and fronts in this case was
simulated well, except that the intensity of the tropical storm was too weak.
The model produced heavy rain with the tropical storm, but not as much as
observed. On the final day, the rainfall over the Great Plains was scattered
and the model could not match all the mesoscale areas correctly. Despite
these particular problems, the overall rainfall simulations for this case, in
terms of timing, area and amount, was fairly reasonable. The statistical
measures of rainfall accuracy support the conclusion that the forecast was
188
-------
average to good. The simulated meteorology should be suitable for use in
RADM.
Assessment by:
Nelson Seaman
189
-------
CASE DESCRIPTION
Case No. 36
Begin Date: 0000 UTC, 9 October 1985
End Date: 0000 UTC, 14 October 1985
Institution: PSU
When Case no. 36 began, a 500-mb trough was over the western Rocky Mts.
and a very broad ridge was over the east coast. Upper-level flow was
southwesterly from Arizona to eastern Canada. At the surface, a. 1005-mb low
was moving northeast from Lake Superior toward James Bay, with a. cold front
trailing southwestward into northern Texas. Rain, some of it very heavy,
was mostly limited to the frontal zone. Since the upper-level flow was
nearly parallel to the front, there was little movement on Oct. 9-10.
MM4/FDDA simulated the pressures and rainfall reasonably well during this
initial period, but did not produce totals as high as observed.
On Oct. 11, rains gradually diminished along the original front. A new
storm system developed in the base of the upper-level trough over New Mexico.
Moderate rains rapidly spread over the southwestern states. The model
simulated these changes fairly well, but was somewhat slow to reduce rainfall
along the old front and to produce the heavier rains over New Mexico. The
new low moved northeast to the Dakotas on Oct. 12, spreading rains over most
of the Great Plains. The model produced rainfall much like observed, but
failed to match some of the mesoscale details in terms of areas and totals.
By Oct. 13, the storm was crossing Lake Superior, with its cold front
trailing southwest to Texas. Most precipitation fell over the Great Lakes
region, with lighter amounts along the front. A frontal wave developed in
Texas on the last day, causing precipitation to increase over the central
Plains toward the end of the period. The model reproduced this pattern quite
well. A few isolated areas of convective rain near the Gulf of Mexico were
not simulated accurately.
In summary, the model reproduced the fronts and lows quite well. The
rainfall was simulated with considerable skill in this case, despite some
errors at certain times. Overall, the meteorological fields, including
precipitation, should be suitable for use in RADM.
Assessment by:
Nelson Seaman
190
-------
CASE DESCRIPTION
Case No. 37a
Begin Date: 0000 UTC, 13 August 1988
End Date: 0000 UTC, 17 August 1988
Institution: NCAR
Case No. 37a began with a quasi-stationary front oriented west to east
from northern Montana to Maine. On 14 August, a frontal wave formed over
Minnesota, and by 0000 UTC, 15 August, the wave had deepened into a 1000 mb
low and was located just north of Lake Huron. Considerable amounts of
precipitation were induced by the storm system in the Great Lakes region with
the heaviest amounts in Minnesota. The MM4/FDDA model simulated the
precipitation amounts and pattern fairly well in the vicinity of the surface
low, but did not predict the substantial convective activity that was observed
in Nebraska and Kansas on the 13th. The threat and bias scores ranged from
0.07 to 0.4 and bias scores from 0.6 to 1.6.
During the latter part of the simulation period (15-17 August) the
surface low moved eastward from the northern Great Lakes to the Atlantic and
another weak storm system developed over the northern Great Plains.
Precipitation was mostly light and scattered throughout the period. On 15
August heavier amounts were observed in northern New England and in Florida
and the model simulation was reasonably good. The precipitation in New
England was associated with the storm system in that area and the activity in
Florida was convective. Most of the precipitation during the last 24 hours
was related to small-scale convection, and therefore, the model simulation was
not very good. The model also predicted up to 2.5 cm of precipitation in the
cold air sector of the weak storm system located over the northern Great
Plains, but very little was observed. The threat scores at the 0.25 and 0.64
cm threshold ranged from 0.04 to 0.3 and bias scores from 0.7 to 1.8.
The overall assessment of the simulation for Case No. 37a is that
MM4/FDDA performed reasonably well and the results are suitable for RADAM.
The SI scores were typical of the Summer cases and the phase and amplitude
errors in the primitive variables were quite small. The threat and bias
scores were fairly good except for those during the last 24 hours of the case
study period when the precipitation was associated with small-scale convective
activity.
Assessment by:
Phil Haagenson
191
-------
CASE DESCRIPTION
Case No. 37b
Begin Date: 1200 UTC, 16 August 1988
End Date: 1200 UTC, 19 August 1988
Institution: NCAR
Case No. 37b began with a 1002 nib low over the Dakotas and a rather weak
high-pressure cell over the southeastern United States. By 0000 UTC, 18
August, the surface low had weakened (1010 mb) and was displaced by a large
anticyclone of Canadian origin. A surface cold front along its southern
boundary extended eastward from Montana to New England. Precipitation amounts
exceeded 4 cm in the upper Great Lakes region north of the cold front. The
MM4/FDDA model simulated the cold sector precipitation quite well except for
0000 UTC, 17 August, when precipitation predicted for Michigan was not
observed. The model simulation of scattered convective precipitation that
occurred in some southern states was not very good. Threat scores at the 0.25
and 0.64 threshold did not exceed 0.3 and bias scores ranged from 0.6 to 1.9.
During the last 36 hours of the simulation period, the cold front pushed
southward into the southern states and the Canadian anticyclone moved into the
Great Lakes region. Precipitation continued in the cold air sector along the
frontal zone; amounts, however, did not exceed 2.6 cm. The model simulated
the spatial distribution very well but over-predicted the amounts by a factor
of two. Convective precipitation that occurred in Louisiana was not
predicted. The threat and bias scores at the 0.25 and 0.64 threshold were
better than those during the first part of the case study period. Threat
scores ranged generally from 0.2 and 0.35 and bias scores from 1.0 to 1.7.
The overall assessment of the simulation for Case No. 37b is that
MM4/FDDA performed fairly well and the results are suitable for RADM. The SI
scores were typical for the Summer cases and the phase and amplitude errors in
the primitive variables were small. The threat and bias scores were fairly
good during the latter part of the simulation period but were not too good
during the first 36 hours.
Assessment by:
Phil Haagenson
192
-------
CASE DESCRIPTION
Case No. 38a
Begin Date: 1200 UTC, 23 August 1988
End Date: 1200 UTC, 26 August 1988
Institution: NCAR
Case No. 38a began with a 999 mb low and upper-level trough located over
southern Manitoba. A surface front extended southeastward from the low
pressure center into Illinois and the southwestward into Texas. By 0000 UTC,
25 August, the surface front had moved into the eastern and southeastern
United States and a new storm system was developing over the Great Lakes.
Considerable precipitation was observed, mostly in the warm air sector in
advance of the cold front. The amounts exceeded 2 cm in the Gulf Coast states
in Ohio, Pennsylvania, and New York. The MM4/FDDA model predicted the
precipitation pattern and amounts fairly well, but was slightly out of phase
with the eastward propagation of the precipitation band. The threat and bias
scores at the 0.25 and 0.64 cm threshold were quite good. All but one of the
threat scores exceeded 0.3 and bias scores ranged from 0.7 to 1.6.
During the last 36 hours of the simulation period, a 995 mb low formed
over the northren Great Lakes and moved slowly eastward into southwestern
Quebec. The cold front extending southward from the low did not produce much
precipitation; the heaviest amounts that occurred were not frontal induced.
The model simulation in some areas was not very good. Although precipitation
amounts exceeded 1 cm in New York and 2 cm in Maine and Florida, the model
predicted < 0.5 cm. The threat scores at the 0.25 and 0.64 cm threshold were
mostly < 0.15 and the bias scores ranged from 0.25 to 2.0.
The overall assessment of the simulation for Case No. 38a is that
MM4/FDDA performed fairly well (especially during the first 48 hours) and the
results are suitable for RADM. The SI scores are similar to the other August
cases and the phase and amplitude errors in the primitive variables were quite
small. The threat and bias scores were fairly good except for those during
the last 24 hours of the case study period when some of the precipitation
amounts were significantly under-predicted.
Assessment by:
Phil Haagenson
193
-------
CASE DESCRIPTION
Case No. 38b
Begin Date: 0000 UTC, 26 August 1988
End Date: 1200 UTC, 28 August 1988
Institution: NCAR
Case No. 38b began with a 994 mb low located just north of Lake Huron. A
rather weak (inactive) cold front extended southwestward from the low center
into Kansas. By 1200 UTC, 27 August, the front had reached the Atlantic coast
and a Canadian anticyclone, preceded by a new cold front, was entering the
Great Plains states. Precipitation on the 26th and 27th was relatively light
and scattered with the most significant amounts observed in New York and
Florida. The MM4/FDDA simulated the precipitation pattern quite well but
under-predicted the amounts. The threat and bias scores were fairly good at
the 0.25 cm threshold but threat scores were not good at the 0.64 threshold
(< 0.1) because the heavier amounts of precipitation were under-predicted.
During the last 24 hours of the simulation period, the anticyclone moved
into the central Great Plains and the cold front pushed eastward into the
eastern and southern states. Light to moderate precipitation occurred along
the frontal zone in both the warm and cold air sectors. Substantial
convective precipitation was also observed in Florida. The model simulated
the pattern and amounts fairly well but over-predicted the spatial extent. It
predicted the convective activity in the Florida area, but placed the
precipitation over water instead of land. The threat scores at the 0.25 and
0.64 cm threshold were better than those for the first part of the case study
period, but the bias scores were large. Threat scores ranged from 0.15 to 0.5
and bias scores from 1.7 to 3.0.
The overall assessment of the simulation for Case No. 38b is that
MM4/FDDA performed in a satisfactory manner and the results are suitable for
RADM. The SI scores were a little better than those for the other August
cases and the phase and amplitude errors in the primitive variable were small.
The threat scores were quite good at the 0.25 cm threshold but not as good at
the 0.64 threshold. Most of the bias scores were fairly good except for those
during the last 24 hours of the case study period.
Assessment by:
Phil Haagenson
194
-------
CASE DESCRIPTION
Case No. 39
Begin Date: 0000 UTC, 28 September 1988
End Date: 0000 UTC, 1 October 1988
Institution: NCAR
Case No. 39 began with a large upper-level disturbance over the northern
Great Plains and a Canadian anticyclone located over the Northern Great Lakes.
A low-pressure trough with a complex frontal structure extended from Montana
to Texas. By 1200 UTC, 29 September, the Canadian high-pressure cell was in
eastern Canada and the surface low over the Great Plains had weakened from
1006 to 1013 mb. During the first 24 hours moderate amounts of precipitation
occurred in the cold air sector north of the surface low. On 29 September,
considerable precipitation (amounts > 2 cm) was observed along the frontal
zone south of the surface low. The MM4/FDDA model simulated the precipitation
amount and pattern quite well. The threat and bias scores at the 0.25 and
0.64 cm threshold were fairly good. Threat scores ranged from 0.3 to 0.5, and
bias scores ranged from 0.7 to 1.4.
During the last 36 hours of the case study period the storm system
stalled over the midwestern states and became rather diffuse. Most of the
precipitation occurred along a stationary surface front that extended from
Lake Michigan southward into Louisiana and eastern Texas. Observed and
simulated precipitation amounts exceeded 2 cm in parts of Texas, Louisiana,
Arkansas, and Missouri. The model simulation was fairly good but it over
predicted the spatial extent. The threat scores at the 0.25 and 0.64 cm
threshold were fairly good, they ranged from 0.2 to 0.45; the bias scores,
however, were rather large--they ranged from 1.5 to 2.
The overall assessment of the simulation for Case No. 39 is that MM4/FDDA
performed reasonably well and the results are suitable for RADM. The SI
scores were similar to the other September cases and the phase and amplitude
errors in the primitive variable were quite small. The threat and bias scores
were fairly good, especially during the first half of the simulation period.
Assessment by:
Phil Haagenson
195
-------
CASE DESCRIPTION
Case No. 40
Begin Date: 0000 UTC, 4 November 1982
End Date: 0000 UTC, 9 November 1982
Institution: NCAR
Case No. 40 began with a large upper-level trough over the Great Lakes
region and a surface trough and frontal system extending from Quebec to
Florida. By 1200 UTC, 6 November, the storm system had moved into the western
Atlantic and been replaced by a large anticyclone over the eastern United
States. Considerable precipitation was observed along the frontal zone on 4
and 5 November. Observed amounts exceeded 2 cm in many eastern and
southeastern states. The MM4/FDDA model simulated the precipitation and
movement of the dry line (behind the front) very well and the threat and bias
scores at the 0.25 and 0.64 threshold were very good. During the period when
most of the precipitation occurred threat scores ranged from 0.4 to 0.8 and
bias scores from 0.7 to 1.0.
During the latter part of the simulation period (7-9 November) the large
anticyclone remained stationary over the southeastern United States and
another anticyclone moved into the north central states. A weak surface front
that separated the two high-pressure cells did not induce precipitation. The
combination of anticyclonic subsidence and dry westerly flow aloft produced
very little precipitation (observed or simulated) over the network.
Consequently, the threat and bias scores at the 0.25 and 0.64 cm threshold
were mostly 0.0 and 1.0.
The overall assessment of the simulation for Case No. 40 is that MM4/FDDA
performed very well and the results are suitable for RADM. The SI scores were
very good and the phase and amplitude errors in the primitive variables were
small. The threat and bias scores were also very good during the time when
significant precipitation occurred.
Assessment by:
Phil Haagenson
196
-------
CASE DESCRIPTION
Case No. 41
Begin Date: 0000 UTC, 12 May 1982
End Date: 0000 UTC, 17 May 1982
Institution: NCAR
Case No. 41 began with a surface front that extended from a weak low over
the southern Hudson Bay to a 1000 mb low over New Mexico. The surface low
over New Mexico was deepening in response to an intensifying upper-level low.
By 1200 UTC, 14 May, the surface front and low pressure system over the
southwestern United States had moved into the central Great Plains and the
weak system north of the Great Lakes had dissipated. The storm system in the
southwest and central plains produced heavy precipitation from Texas northward
to the Dakotas. Amounts exceeded 6 cm in the warm air sector in Texas and
Oklahoma and exceeded 3 cm in the cold sector in Nebraska and South Dakota.
The MM4/FDDA simulated the pattern and location of the precipitation very
well, but under-predicted the heavier amounts. The threat and bias scores at
the 0.25 and 0.64 cm threshold were good. Threat scores ranged from 0.3 to
0.6 and bias scores from 0.99 to 1.5.
During the last 60 hours of the case study period, the upper-level and
surface low became stationary over the central Great Plains. A quasi-
stationary front extended southward and eastward from the low pressure center
in South Dakota. Moderate amounts of precipitation continued to occur in both
the warm and cold air sector of the storm system. Heaviest amounts ranged
from 1 cm to 3 cm. The model simulation continued to be quite good,
particularly in the areas where the heaviest precipitation occurred. The
threat and bias scores at the 0.25 and 0.64 cm threshold were nearly as good
as those for the first half of the case study period. Threat scores ranged
from 0.25 to 0.55 and bias scores from 0.8 to 1.8.
The overall assessment of the simulation for Case No. 41 is that MM4/FDDA
performed very well, particularly with respect to the precipitation
simulation, and the results are suitable for RADM. The SI scores were fairly
good and the phase and amplitude errors in the primitive variables were quite
small. The threat and bias scores were better than for most of the other
cases.
Assessment by:
Phil Haagenson
197
-------
CASE DESCRIPTION
Case No. 42
Begin Date: 0000 UTC, 8 June 1983
End Date: 0000 UTC, 13 June 1983
Institution: NCAR
Case No. 42 began with a quasi-stationary front along the southeastern
coast of the United States and a 1005 mb low over the Dakotas. A weak frontal
system was associated with the low. By 1200 UTC, 10 June, the surface low had
moved eastward and weakened, and a large anticyclone covered most of the
eastern United States. Most of the precipitation during the first 60 hours
occurred along the frontal zone in the southeastern United States on the 8th,
with the heaviest amounts reported in Florida. Convective precipitation, not
associated with frontal systems, also occurred in Texas, Oklahoma, and Kansas
on the 10th. The MM4/FDDA model simulated the observed precipitation very
well, but also simulated precipitation that was not observed in the vicinity
of the weak storm system in the northern states. The threat and bias scores
at the 0.25 and 0.64 cm threshold were fairly good except for those at 0000
UTC on 9 and 10 June when the simulation of precipitation that did not occur
in the northern states causes the bias scores to exceed 4.0.
During the latter part of the simulation period (11-13 June) the large
anticyclone remained stationary over the eastern United States and a new storm
system began to organize over the northern Rocky Mountains. By the end of the
period, it had developed into an elongated trough of low pressure that
extended from southern Manitoba to New Mexico. Precipitation was observed
along the frontal zone from the Dakotas to Colorado. Non-frontal related
precipitation also continued to be observed in Oklahoma and Texas. The model
over-predicted the spatial extent of the precipitation, particularly in
Oklahoma and Kansas on the 12th, and in the northern Great Plains on the 13th.
The threat scores at the 0.25 and 0.64 cm threshold were fair, but the bias
scores were large. Threat scores ranged from 0.1 to 0.35 and bias scores from
1.0 to 3.3 except for 0000 UTC, 12 June, when the bias scores at the 0.25
threshold exceeded 8.0.
The overall assessment of the simulation for Case No. 42 is that MM4/FDDA
performed reasonably well and the results are suitable for RADM. The SI
scores were comparable to those for other Summer cases and the phase and
amplitude errors in the primitive variables were quite small. The threat
scores were fairly good but the bias scores were larger than for the other
cases.
Assessment by:
Phil Haagenson
198
-------
CASE DESCRIPTION
Case No. 43
Begin Date: 0000 UTC, 15 July 1985
End Date: 0000 UTC, 20 July 1985
Institution: NCAR
Case No. 43 began with a 1020 mb high over the northern Great Plains and
a slow-moving cold front extending eastward from Colorado to northern Maine.
By 1200 UTC, 17 July, the anticyclone had moved into the Great Lakes region
and the surface cold front had reached the Atlantic coast. Moderate
precipitation occurred along the frontal zone and in the warm air sector in
advance of the cold front. Amounts exceeded 2 cm in several eastern and
southeastern states. The MM4/FDDA model simulated the precipitation fairly
well except for the 16th when it predicted 1.75 cm in Indiana (none was
observed) and did not predict precipitation in Arkansas (3 cm was observed) .
The threat scores at the 0.25 and 0.64 cm threshold were only fair, but the
bias scores were reasonably good. Threat scores ranged from 0.1 to 0.3 and
bias scores from 0.5 to 1.3.
During the last 60 hours of the simulation period, a surface front
associated with a low-pressure trough entered the northern Great Plains and
moved slowly into the central states. The disturbance induced moderate
precipitation, mostly along the frontal zone. The heaviest amounts exceeded 3
cm in Minnesota, Iowa, and Nebraska. Convective precipitation, not associated
with the frontal system, also occurred along the Gulf Coast and in Florida.
The model predicted the frontal precipitation quite well but over-predicted
the spatial extent on the 19th and did not predict some of the precipitation
that was observed in the Gulf States. The threat and bias scores at the 0.25
and 0.64 cm threshold were similar to those for the first 60 hours except for
0000 UTC, 19 July, when the bias scores exceeded 2.0.
The overall assessment of the simulation for Case No. 43 is that MM4/FDDA
performed fairly well and the results are suitable for RADM. The SI scores
were similar to the other Summer cases and the phase and amplitude errors in
the primitive variables were quite small. The threat and bias scores were
fairly good except for the bias scores at 0000 UTC on 19 July.
Assessment by:
Phil Haagenson
199
-------
CASE DESCRIPTION
Case No. 44
Begin Date: 0000 UTC, 14 December 1985
End Date: 0000 UTC, 19 December 1985
Institution: PSU
As Case no. 44 began, a deep 500-mfa long wave trough lay over the
central U.S., with a strong ridge along the west coast and another ridge over
the western Atlantic. A 1001-mb surface storm was moving northeastward south
of New England and a broad 1033-mb high over the western Great Plains was
centered in Texas. A second low in Alberta was slipping southeast along a
lee trough near the Rocky Hts. Rainfall at the beginning of the episode was
confined to New England and central Florida. MM4/FDDA simulated the initial
precipitation regions, but under-forecast the totals during the initial
"spin-up" period.
By Dec. 15, the old storm had moved well east of the coastline, ending
the precipitation in New England and Florida. A 1007-mb low had moved east
from the Rockies to the Great Lakes, but only a few isolated showers occurred
anywhere in the U.S. The model simulated this condition very well. However,
the few isolated showers led to very poor threat scores, degrading the case
average statistics despite a generally excellent precipitation forecast. On
Dec. 16, the original high had drifted east to southern Mississippi,
effectively closing the U.S. to moisture from the Gulf of Mexico. Lows
forming in the lee of the northern Rockies and moving east-northeast across
the Great Lakes on Dec. 15-16 produced virtually no precipitation. The model
correctly simulated this dry pattern.
Little change in this pattern occurred from Dec. 17-19. The 500-mb
trough remained entrenched over the central U.S., and although somewhat
weaker, it continued to advect cold dry air from western Canada into the
entire U.S. east of the Rocky Mts. This flow tended to maintain the high
over the Gulf Coast, thus preventing moist air from penetrating northward.
Only a few light showers accompanied the train of weak lows ("Alberta
clippers") that crossed the northern Plains. MM4/FDDA simulated the pressure
systems very well and even showed considerable correlation of its
precipitation areas to the observed isolated showers accompanying the lows.
In summary, MM4/FDDA did produced an excellent simulation of the
meteorology of this episode. However, the statistical scores for rainfall
are very misleading for this dry case. The model actually performed very
well. The simulated meteorological fields should be very suitable for RADM.
Assessment by:
Nelson Seaman
200
-------
CASE DESCRIPTION
Case No. 45
Begin Date: 0000 UTC, 26 April 1985
End Date: 0000 UTC, 1 May 1985
Institution: PSU
Case no. 45 began with a 500-mb trough over the Rocky Mts. and a broad
ridge over the eastern U.S. A 991-mb surface storm was weakening and
drifting slowly east through New Mexico with a warm front northeastward to
Illinois. Moderate rains fell north of the warm front from Colorado and
Wyoming to Iowa, and also along the Texas-Louisiana coast. MM4/FDDA
simulated the early meteorology and rainfall rather well, but produced too
much rain in Kansas.
By April 27, the original warm front had extended east through the Ohio
Valley to the Atlantic Ocean across Delaware, while the low in the southwest
filled. A frontal wave had formed in Missouri and several mesoscale areas of
rain were observed from New Mexico to the Ohio Valley. The model simulated
most of the rain areas reasonably well, but could not simulate the totals
correctly or some of the smaller-scale spatial details. By April 28, the
rain pattern became better organized along the front and the model's rainfall
became more accurate. Later on April 28-29, the front gradually dissipated
and rainfall became light and isolated in the eastern states. Meanwhile
moderate rains broke out from Arizona to west Texas. MM4/FDDA simulated the
general changes taking place in the precipitation pattern, but failed to
produce many of the details accurately. On April 29 another weak low that
had beneath the 500-mb trough formed over New Mexico moved east to the Texas
panhandle. This system organized the rain from Texas to Kansas and Colorado.
The model made a reasonable forecast of the rain areas, but continued to have
trouble with the location of heaviest rain. On the last day, the low moved
northeastward through Oklahoma, shifting the rain to the central Plains.
Model rain was reasonable in most places, but failed to simulate some details
of distribution and totals.
In summary, the model performance in this case was generally good, but
rainfall across the southern states tended to have considerable mesoscale
structure that the model could not simulate exactly. This is probably due to
the convective nature of rainfall typical for late April at these latitudes.
Averaged over the episode, most of these small-scale errors should become
fairly minor. The case-mean statistical measures of rain skill in MM4/FDDA
are about average and the model fields should be generally suitable for RADM.
However, due to small-scale and temporal inconsistencies in the rain pattern,
RADM output should be reviewed and used with care.
Assessment by:
Nelson Seaman
201
-------
CASE DESCRIPTION
Case No. 46
Begin Date: 0000 UTC, 30 January 1982
End Date: 0000 UTC, 4 February 1982
Institution: PSU
Case no. 46 began with a closed 500-mb low in a trough over the
southwest U.S. and a weak ridge in along the east coast. At the surface, a
1033-mb high was over Chesapeake Bay and a 999-mb low was over west Texas,
with a warm front northeastward to Indiana. A 1014-mb low north of Lake
Superior was moving northeastward and two weak lows were found along a
stationary front paralleling the Rocky Mts. from Montana to Alberta. An
arctic high was moving from the Yukon toward the northern Great Plains. The
Texas low began moving northeastward and slowly filled on Jan. 30, while its
warm front became connected to the cold front extending south from the low
north of the Great Lakes. Moderate to heavy precipitation fell from Texas to
Ohio along the consolidated frontal boundary. MM4/FDDA simulated the pattern
very well, but under-forecast the totals. By Jan. 31, the Arctic high from
Canada had reached Lake Superior, while the southern storm was moving
northeast through western Tennessee and re-intensifying as the baroclinicity
increased along the frontal boundary. Precipitation remained quite heavy,
especially in the Mississippi Valley. The model simulated the area very well
and the amounts of precipitation were very reasonable.
The deepening storm accelerated northeast through New England on Feb. 1
and occluded. The precipitation region was swept rapidly eastward and rain
totals gradually declined. The model simulated this event very well.
Meanwhile, a new 1006-mb low was forming over New Mexico in the 500-mb
trough, but no rain accompanied the early stages of this storm. The model
simulated the storm formation, but with a central pressure of only 1011 mb.
On Feb. 2, the new storm in the southwest moved east into Texas, while a new
polar front raced south from Alberta ahead of another Arctic high. Light
precipitation developed over the central Plains and was simulated reasonably
well by MM4/FDDA. Frontogenesis also occurred south of the Gulf of Mexico
coast, spreading rain onshore and through the southeast. The model
simulated this development, and although it was initially slow to spread the
rain onshore, it recovered well as the rain area expanded across the
southeast.
On the final day, Feb. 3, the low in the south weakened and accelerated
northeast, but baroclinicity increased as the cold air mass over the northern
Plains pushed southeast to the Gulf coast. Meanwhile, warm moist advection
from the south was maintained east of the Mississippi Valley, which supported
heavy precipitation from the Gulf to New England. The model simulated these
events well, including precipitation area and totals, except for some under-
forecasting of the most extreme rain maxima.
202
-------
The statistical scores and subjective evaluation both indicate that this
was one of the most accurately simulated episodes of the set. At most times,
the storm patterns and movements were very close to those observed.
Precipitation was simulated quite well. This case should be very appropriate
as input to RADH.
Assessment by:
Nelson Seaman
203
-------
CASE DESCRIPTION
Case No. : 47
Begin Date: 0000 UTC, 7 December 1983
End Date: 0000 UTC, 12 December 1983
Institution: PSU
Case no. 47 began with a very broad 500-mb trough over the Mississippi
Valley. A ridge was located over the west coast and another over the
Atlantic ocean. A 67 m/s jet streak over Tennessee was just rounding the
base of the trough. At the surface, a 986-mb occluded low over Lake Ontario
was intensifying as it moved down the St. Lawrence Valley. The storm's cold
front lay south along the east coast. During the first 12 h, rain was
confined to the Northeast ahead of the front. MM4/FDDA did a good job of
simulating this initial rainfall.
On Dec. 7, lee cyclogenesis occurred in northeastern Colorado and by
Dec. 8, a 1008-mb low moved east toward St. Louis and the Ohio Valley. A
1027-mb high along the coast of the Gulf of Mexico prevented moist air from
moving north into the U.S. The old occlusion continued northeastward and
rain ended late on Dec. 7 in the New England. The low moving across the
Great Plains produced only a few light showers, the U.S. east of the Rocky
Mts. had virtually no rain. The model simulated this condition very well.
On Dec. 8 another lee cyclone formed in Colorado and tracked east to Illinois
on Dec. 9. This low was also dry, but produced light frontal rain in the
Ohio Valley. MM4/FDDA produced light rain over Ohio, but did not simulate
the banded structure of the rain, nor its maximum in Kentucky.
On Dec. 9, the high along the Gulf coast retreated north and east off
the Atlantic coast and allowed moist warm air to advect northward, first into
Texas and then up the Mississippi Valley. Rain was triggered as this moist
air encountered a cold front trailing southwest from the second lee cyclone
then in the Midwest. On Dec. 10, light rain began on the Texas coast and
spread north to southern Arkansas. The model simulated the initiation of
this rain, including area and totals. At the same time a vigorous jet streak
propagated through the southwestern U.S. and triggering a third lee storm in
Colorado. This last storm, supported by the jet, dropped southeastward into
Oklahoma by Dec. 11 (1006 mb), where it tapped the moist southerly flow from
the Gulf. Heavy rains were triggered in Louisiana, with moderate rain
spreading north to Illinois and Indiana. The model simulated this event
quite well. This storm tracked north and east to Illinois by the end of the
period, shifting the rainfall as it moved. The model predicted these changes
very well and precipitation was reasonably accurate.
In summary, the model fields and rainfall for this case were quite
accurate and should be very suitable for use in RADM.
Assessment by:
Nelson Seaman
204
-------
CASE DESCRIPTION
Case No. 48
Begin Date: 0000 UTC, 2 November 1985
End Date: 0000 UTC, 7 November 1985
Institution: PSU
Case no. 48 began with a 500-mb trough over the Great Plains and a weak
ridge over the east coast. At the surface, an occluded 997-mb low was
weakening over Indiana, while a secondary storm was forming off the South
Carolina coast. A 1001-mb low was moving southwest from Alberta into
Montana, but weakened and dissipated by Nov. 3. Moderate rains accompanied
the occluded low and its coastal secondary. Early on Nov. 2, a frontal zone
formed in the Gulf of Mexico, bringing moderate rain to the Texas-Louisiana
coast. MM4/FDDA simulated this early pattern, including rainfall, very well.
These systems drifted slowly northeast on Nov. 2-3, with rainfall covering
the Appalachian Mt. region on Nov. 3. The model simulated these changes
well.
By Nov. 4, a wave that had formed on the front in the Gulf of Mexico was
located over eastern Georgia with an inverted trough north to Lake Huron.
The northern part of the trough marked the decayed remnant of the occluded
low. Rain continued over the region from Florida to the Great Lakes. The
model simulated this area and correctly placed the heaviest rains east of the
mountains. From Georgia, the southern low moved north-northeast, deepened to
995-mb and occluded on Nov. 5 over Virginia. Heavy rains fell from the Mid-
Atlantic region to New York. The model simulated these changes well and
correctly placed the maximum rain over eastern West Virginia. Meanwhile, a
996-mb low which had formed in the lee of the northern Rocky Mts. moved east
through North Dakota, but there was very little precipitation accompanying
this system.
On Nov. 6, the occluded storm slowly weakened and moved northeast past
Nantucket Island. Rains over New York and New England gradually diminished
as the low moved away from land. The low over the northern Great Plains
traveled northeast to Hudson's Bay, but it trailed a cold front across the
western Great Lakes and south to Texas. A high along the Gulf of Mexico
coast prevented moist air from penetrating the Great Plains, so the western
frontal system remained mostly dry. A few showers fell over the northern
Great Plains and as far east as Lake Michigan. MM4/FDDA simulated these
light showers rather well.
In summary, the model simulation for this case was one of the very best
of the entire case set. All pressure systems, fronts and rain regions were
reproduced quite well and the results should be fully appropriate for use in
RADM.
Assessment by:
Nelson Seaman
205
-------
CASE DESCRIPTION
Case No. 49
Begin Date: 0000 UTC, 7 February 1985
End Date: 0000 UTC, 12 February 1985
Institution: PSU
When Case no. 49 began, flow at 500 mb across the U.S. was mostly zonal,
with a very broad, weak trough over the Mississippi Valley. A deep surface
low lay northeast of the model domain near Labrador, but had no effect on the
U.S. A weak cold front over the Great Lakes was pushing slowly south and
dissipating. A large anticyclone in central Canada supported a strong ridge
south to the Gulf of Mexico. As a result, virtually no rain fell over the
U.S. east of the Rocky Mts. on Feb. 7-8.
On Feb. 8, a Pacific storm crossed the northern Rocky Mts., bringing
snow to the Northwest, western Montana and Wyoming. MM4/FDDA simulated this
event and the precipitation accompanying it. By Feb. 9, rapid lee
cyclogenesis occurred over the Texas panhandle, absorbing the storm crossing
the northern Rockies. This ended precipitation to the north. The ridge over
the Great Plains had drifted east of the Mississippi Valley by Feb. 9, but
warm moist air had not yet penetrated the Plains, so there was virtually no
rain associated with the new cyclogenesis.
By Feb. 10, a 1031-mb Arctic high was sliding south from Alberta into
the northern Plains. Ahead of this high, a cold front ran from Manitoba
south to Kansas City, where it became a warm front near the storm in the
southern Plains, then crossing Oklahoma (1004 mb) . Some light snow fell in
Minnesota along the cold front. The model simulated this area and the totals
very well. To the south, the eastward retreat of the original ridge allowed
a steady southerly flow of moist air from the Gulf of Mexico ahead of the
storm in Oklahoma. Finally, late on Feb. 10, precipitation broke out in
advance of the storm and spread up the Mississippi Valley. The model did
very well at simulating these changes, including the timing, area, and amount
of rain.
On Feb. 11, the storm weakened and turned northeast toward the Great
Lakes, but continued to produce moderate to heavy rain from the Gulf to the
Lakes. As the system drifted east of the Mississippi Valley, a secondary low
formed along its cold front in Mississippi and moved northeast to Kentucky by
the end of the period. By then, rains covered most of the region from Miami
to Lake Superior. The model simulated this complex pressure system very
well, and most aspects of the precipitation were reproduced.
In summary, the model results verified rather well and the
meteorological fields should be appropriate for RADM.
Assessment by:
Nelson Seaman
206
-------
CASE DESCRIPTION
Case No. 50
Begin Date: 0000 UTC, 11 November 1983
End Date: 0000 UTC, 16 November 1983
Institution: PSU
Case No. 50 began with a 500-mb ridge over the Rocky Mts. and another
over Nova Scotia. A 500-mb trough lay over the central U.S. and Canada, with
a closed low over western Kentucky. An occluding 999-mb surface storm over
West Virginia was moving northeast, spreading light to moderate rain from the
Midwest to New England. A surface ridge extended south to Texas from a 1032
mb high in Canada. The eastern storm deepened on 11-12 Nov. and moved
northeast toward the Gaspe Peninsula. Rains slowly tapered off in the
Northeast. MM4/FDDA simulated this change very well. Meanwhile, troughing
and lee cyclogenesis began just east of the Rocky Mts. behind the ridge over
the Great Plains. A weak 1005-mb low formed over the Texas panhandle, but
gradually dissipated as it moved east from the mountains. An area of light
showers occurred over the northern Great Plains ahead of a warm front, but
this disturbance also remained rather weak. The model correctly simulated the
outbreak of showers over the Plains.
On Nov. 13, lee cyclogenesis occurred again east of the northern Rocky
Mts. By Nov. 14, a 997-mb low was moving east across South Dakota and a warm
front was forming over the Gulf of Mexico coast. Later on Nov. 14,
cyclogenesis occurred over the Texas panhandle as the warm front moved
inland. However, no significant precipitation occurred with either system
during Nov. 13-14. MM4/FDDA correctly simulated only a few showers during
this period, but it was unable to match the exact locations, so that the
average threat score statistics for this case were degraded. Unfortunately,
the statistical average fails to reveal the general accuracy of the
precipitation forecast.
By Nov. 15, as a high built over the Rockies, the two lows were moving
east and were connected by a north-south frontal system. The southern low
maintained a warm front across the deep South, which became the sight of
strong overrunning. Moderate rain spread north from the Gulf to the eastern
Great Lakes during the final day. The model simulated the spreading rain
pattern rather well, but failed to simulate some of the heavier rain.
The model simulated the developing pressure patterns quite well in this
case. Precipitation forecasts were actually rather accurate, but the
heaviest rains on the final day were not simulated. Overall, the model
output should be suitable for RADM, although caution should be used when
interpreting the impact of the lighter rainfall simulated on the final day.
Assessment by:
Nelson Seaman
207
-------
CASE DESCRIPTION
Case No. 51
Begin Date: 0000 UTC, 2 November 1988
End Date: 0000 UTC, 7 November 1988
Institution: PSU
Case no. 51 began with a deep 500-mb trough over the eastern U.S. and a
broad ridge over the Rocky Mts. A closed 500-mb low over the lower Great
Lakes was moving northeast toward Maine. At the surface an occluded storm
just south of Long Island was moving northeast toward Nova Scotia. This
brought moderate to heavy rains to New England and New York, which were
simulated well by MM4/FDDA. A weak, but intensifying low (1008 mb) over the
southern Great Plains was moving northeast, but produced no rain. By late on
Nov. 3, precipitation east of the Rocky Mts. had virtually ended, except for
some rain in northern New England.
By Nov. 4, the developing low (then 994 mb) in the Great Plains was over
Iowa. Its cold front trailed southwestward to eastern Texas and its warm
front lay northeast to Michigan. The southerly flow in the storm's warm
sector had advected moist air northward from the Gulf of Mexico since Nov. 3.
During Nov. 4, precipitation broke out in the Ohio and Mississippi Valleys.
An independent area of rain had grown over Florida. MM4/FDDA simulated these
rainfall regions reasonably, but missed some of the mesoscale details of
area, timing and amount. The storm continued northeast on Nov. 5, occluded
and became stationary over northern Michigan (974 mb) by Nov. 6, while its
cold front swept to the east coast. Moderate to heavy rains fell between
the Mississippi Valley and the Appalachian Mts., and between the Gulf and the
Lakes on Nov. 5. By Nov. 6, the rains tapered off in the southeast, but
continued from the northeast to the Great Lakes. As the front continued east
off the coast, the rains ended except for light amounts reported over
northern Maine and near the filling storm center in Michigan. MM4/FDDA
simulated the mature and decaying periods of this major storm very well,
including the precipitation areas and amounts associated with it.
In summary, this case was simulated with considerable accuracy. There
were no important errors in the rainfall simulations and the model's
meteorological fields should be very suitable for application in RADM.
Assessment by:
Nelson Seaman
208
-------
CASE DESCRIPTION
Case No. 52
Begin Date: 1200 UTC, 6 November 1988
End Date: 1200 UTC, 11 November 1988
Institution: PSU
Case no. 52 began with a large 500-mb low over Green Bay that covered
most of the eastern U.S. A 500-mb ridge lay over the eastern Rocky Mts. An
occluded 976-mb surface low was beneath the upper-level center with its cold
front lying east- northeast to Nova Scotia. A second storm (983 mb) was
crossing the northern Rocky Mts. in Alberta. The occluded low was filling
and only light precipitation was falling in the Midwest. MM4/FDDA simulated
this early situation rather well. On Nov. 7, a secondary cold front formed
over the Appalachian Mts. and moved off the east coast. This front brought
only a few showers to New York and Pennsylvania. Meanwhile, lee cyclogenesis
occurred in eastern Colorado and a 1000-mb low drifted east into Oklahoma. A
frontal boundary connected the storm in Alberta to this southern development,
but only a few showers fell over the Dakotas. The model simulated this rain
area very well.
Although the southern lee cyclone weakened as it left the mountains and
approached the central states, southerly winds ahead of the front began
advecting moist air into the Mississippi Valley on Nov. 8. However, showers
were light and confined to fairly small portions of Mississippi and the
Midwest. The model under-forecasted this light rain. The front began
dissipating along the east coast on Nov. 9 and showers remained light and
scattered. The model continued to under-forecast this rainfall.
On Nov. 9, a new cyclone developed over the Texas panhandle.
Initially, precipitation was limited to the Colorado Rockies, but as the low
intensified and moved northeast into the Midwest on Nov. 10, moderate rain
broke out from eastern Texas to the Great Lakes. MM4/FDDA simulated the
timing, area and amount of precipitation quite well, although it was slow to
move the rain maximum north into Michigan. By Nov. 11, the storm (988 mb)
tracked northeast through eastern Canada with its cold front sweeping past
the east coast of the U.S. The light to moderate rains moved through the
eastern U.S. with the front and ended by the final time of the episode. The
model produced roughly the right amount of rain in most areas, but slightly
over-predicted the rains in Pennsylvania and New York.
Despite some problems with simulation of the regions of light rain, the
meteorology and precipitation fields for this case were of about average
accuracy. The model fields should be suitable for use in RADM.
Assessment by:
Nelson Seaman
209
-------
CASE DESCRIPTION
Case No. 53
Begin Date: 0000 UTC, 11 November 1988
End Date: 0000 UTC, 16 November 1988
Institution: PSU
When Case no. 53 began, a 500-mb long wave trough lay over the Great
Lakes, while an upper-level ridge lay over the Rocky Mts. A 988-mb surface
low was tracking northeastward through southern Quebec, with its cold front
approaching the east coast of the U.S. Light rains with the front were
rapidly swept over the Atlantic as a large 1029-mb high centered in Illinois
dominated most of the U.S. east of the Rocky Mts. MM4/FDDA simulated these
early developments very well.
As the high continued eastward through the Midwest on Nov. 12, lee
cyclogenesis occurred in southeastern Colorado and moist air was advected
north into the Great Plains. Moderate to heavy rains broke out in Oklahoma
and spread north to Minnesota as the low approached Kansas City. The model
simulated the rainfall rather well, but also extended it south to New Orleans
too early. The low moved northeast to the Great Lakes on Nov. 13 and was
absorbed into a 999-mb low in central Canada, while its cold front pushed
east to the Appalachian Mts. Rains swept eastward with the front from the
Gulf of Mexico to Canada. The model simulated these changes well, but under-
forecast the rain maximum with the front in the deep South. By Nov. 14, the
front moved off the Atlantic coast and through New England. Moderate to
heavy rains fell from Pennsylvania to Maine. The model simulated the areas
well, but amounts were lighter than observed.
Meanwhile, a strong storm crossed the central Rocky Mts, on Nov. 14,
accompanied by widespread mountain snows. This storm (991 mb) reached
southeastern Colorado on Nov. 15 and then tracked rapidly northeast toward
Iowa. Initially, precipitation over the Great Plains was light and limited
to the Dakotas and a few areas from Missouri to Wisconsin. However, by the
end of the period, moderate rains had overspread most of the eastern Plains
from east Texas to the western Great Lakes. MM4/FDDA simulated the
development of the storm and its precipitation shield rather well, in terms
of area and amounts.
In summary, the model simulated the development of the meteorological
systems quite well. The precipitation patterns and amounts were generally
reasonable despite some periodic mesoscale errors. Overall, the model fields
should be suitable for use in RADM.
Assessment by:
Nelson Seaman
210
-------
CASE DESCRIPTION
Case No. 54
Begin Date: 1200 UTC, 15 November 1988
End Date: 1200 UTC, 20 November 1988
Institution: PSU
Case no. 54 began with a 500-mb trough over the Rocky Mts. and an upper-
level ridge over the Great Lakes. Another upper-level trough was over Nova
Scotia. A 991-mb surface storm was in southeastern Colorado moving rapidly
northeastward. This storm spread rain from east Texas to the Great Lakes
early in the period and MM4/FDDA simulated this pattern rather well. A
strong upper-level short wave contributed to continued intensification of the
surface storm on Nov. 16 as it crossed Lake Superior (976 mb) and then
occluded in Canada. Moderate rain fell in the Great Lakes region and south
through the Mississippi Valley along the trailing cold front. The model
simulated the areas and amounts of precipitation very well.
On Nov. 17 the cold front pushed east across the Appalachian Mts. and to
the Atlantic. The band of moderate rain with the front was simualated well
by MM4/FDDA. Behind the front a 1026-mb high built over the south-central
states and prevented the moist warm air over the Gulf of Mexico from reaching
the mainland. By Nov. 18, rains had ceased east of the Rocky Mts. except for
some rain in New England and along the Mid-Atlantic coast associated with the
old front moving into the Atlantic. A new storm formed over New Mexico on
Nov. 18, but initially produced almost no rain. Late on Nov. 18, the high in
the South moved northeast past the Appalachian Mts. and allowed the
southwestern storm to tap a moist southerly flow from the Gulf.
As the storm built northeast over the Great Plains early on Nov. 19,
heavy rains broke out in Arkansas and spread northeast into the Ohio Valley.
MM4/FDDA simulated the spread of the precipitation reasonably well, but
under-forecast the heaviest rains in Arkansas. On Nov. 20, the system was
oriented north-south through the Mississippi Valley, with frontal waves over
the Great Lakes and the South. Rains over the Great Lakes were light and
covered a small area, but moderate to heavy rain continued from Texas and
Oklahoma northeast to southern New England. The model smiulated this
development well, but rain totals were sometimes too light in areas of
particularly heavy rain.
In summary, the model reproduced the meteorological development well,
including the areas of rainfall. Although the rainfall maxima were often
underforecast, the overall model precipitation was very reasonable. The
model fields should be suitable for use in RADM.
Assessment by:
Nelson Seaman
211
-------
CASE DESCRIPTION
Case No. 55
Begin Date: 0000 UTC, 20 November 1988
End Date: 0000 UTC, 25 November 1988
Institution: NCAR
Case No. 55 began with a 1005 mb low in Arkansas and Louisiana that was
associated with an upper-level trough over the Southern Great Plains. During
the first 36 hours the low deepened to 975 mb and moved rapidly northeastward
into New England. By 1200 UTC, 22 November, the storm system had reached the
North Atlantic and a large anticyclone covered most of the central and eastern
United States. The storm triggered moderate to heavy precipitation in several
of the eastern and southeastern states. Amounts exceeded 4 cm in Alabama,
Kentucky, and southern New England. The MM4/FDDA model simulated the
precipitation very well, particularly the spatial distribution. The threat
and bias scores were very good during the period when most of the
precipitation was observed. The threat scores at the 0.25 and 0.64 cm
threshold exceeded 0.5 and bias scores ranged from 0.7 to 0.95.
During the latter part of the case study period (23-25 November) the
anticyclone remained stationary over the eastern United States and a tropical
storm (Keith) moved eastward from the Gulf of Mexico, across Florida, to the
western Atlantic. By the end of the period, a trough of low pressure had
formed over the northern Great Plains. Very little precipitation was observed
except that induced by Keith in Florida and the Carolinas. Heavy amounts (>
10 cm) were reported in Florida on the 23rd. The model simulation in the
precipitating area was quite good, however, it under-predicted the maximum
amount by a factor of two. The threat and bias scores at the 0.25 and 0.64 cm
threshold were very good (during the period when most of the precipitation
occurred) except for 1200 UTC, 23 November, when the bias score at the 0.64
threshold was 3.0. Threat scores ranged from 0.35 to 0.85.
The overall assessment of the simulation for Case No. 55 is that MM4/FDDA
performed very well and the results are suitable for RADM. The SI scores were
very good and the phase and amplitude errors in the primitive variable were
quite small. The threat and bias scores, particularly the threat scores, were
very good during the periods of significant precipitation.
Assessment by:
Phil Haagenson
212
-------
CASE DESCRIPTION
Case No. 56
Begin Date: 1200 UTC, 24 November 1988
End Date: 0000 UTC, 29 November 1988
Institution: NCAR
Case No. 56 began with a surface low and weak cold front located over the
Great Plains. An upper-level trough was intensifying over the Rocky Mountain
region. By 0000 UTC, 27 November, the surface low had deepened to 991 mb and
the upper-level trough was well developed over the central United States. The
frontal system associated with the storm extended southward from the Great
Lakes to Texas. Very little precipitation was observed during the early
stages of storm development on the 25th, but on the 26th moderate to heavy
precipitation occurred in the warm air sector. Amounts exceeded 2 cm in
several of the south-central states with a maximum of 7 cm in Arkansas. The
MM4/FDDA model simulated the amounts and spatial distribution very well.
Threat scores at the 0.25 and 0.64 threshold (after the onset of significant
precipitation) exceeded 0.5 and bias scores ranged from 0.75 to 1.2.
During the last 48 hours of the simulation period, the upper-level trough
and surface low moved northeastward into southeastern Canada. The frontal
structure of the storm system was quite complex with multiple fronts, but the
most active front was on the east side of the upper-level trough and extended
from Canada to the Gulf states. Moderate amounts of precipitation occurred
along the main frontal zone and the model simulation was very good. Both the
amounts and the spatial distribution were quite accurately predicted. The
threat scores at the 0.25 and 0.64 cm threshold ranged from 0.4 to 0.65 and
bias scores from 0.75 to 1.4.
The overall assessment of the simulation for Case No. 56 is that MM4/FDDA
performed very well and the results are suitable for RADM. The SI scores were
better than for most of the other cases and the phase and amplitude errors in
the primitive variables were quite small. The threat and bias scores were
very good.
Assessment by:
Phil Haagenson
213
-------
CASE DESCRIPTION
Case No. 57
Begin Date: 0000 UTC, 16 December 1988
End Date: 0000 UTC, 20 December 1988
Institution: NCAR
Case No. 57 began with a large anticyclone over the central United States
and a surface cold front along the southeast coast. A weak low-pressure
trough was located over the northern Great Plains. During the first part of
the case study period (16-18 December) the anticyclone remained stationary
over the south-central United States and the low-pressure trough moved
eastward along the Canadian border. The only significant precipitation
occurred along the cold front in the southeastern states during the first 24
hours. The amounts, however, were less than 1 cm. The MH4/FDDA model
simulated the precipitation fairly well. The threat scores at the 0.25 cm
threshold ranged from 0.2 to 0.4 and the bias score from 0.5 to 0.65. Since
the precipitation amounts were light, the threat scores at the 0.64 threshold
were 0.0.
During the last 48 hours of the simulation period, the anticyclone moved
eastward across the southern states while another low-pressure trough
propagated eastward along the Canadian border. By the end of the period a
987 mfa low formed along the Nebraska-Colorado border in response to an upper-
level disturbance. Very little precipitation was observed or simulated over
the network except for light amounts during the last 12 hours in the vicinity
of the Nebraska-Colorado low. Since precipitation was so light the threat
scores at the 0.25 and 0.64 cm threshold were 0.0.
The overall assessment of the simulation for Case No. 57 is that MM4/FDDA
performed quite well and the results are suitable for RADM. The SI scores
were typical of the cold season cases and the phase and amplitude errors in
the primitive variables were quite small. The threat and bias scores were
quite good during the time when there was enough precipitation to make the
scores meaningful.
Assessment by:
Phil Haagenson
214
-------
CASE DESCRIPTION
Case No. 58
Begin Date: 1200 UTC, 19 December 1988
End Date: 0000 UTC, 23 December 1988
Institution: NCAR
Case No. 58 began with an intensifying storm system over the Great Plains
region. A 987 mb low formed on the Nebraska-Colorado border, and by 1200 UTC,
21 December, it had moved northeastward into southeastern Canada. A surface
cold front extended southward from the low-pressure center into the southern
states and light to moderate precipitation occurred along the frontal zone.
Amounts exceeded 1 cm in Iowa and Nebraska on the 20th and exceeded 2 cm in
Kentucky, Tennessee, and Alabama on the 21st. The threat and bias scores were
not very good except for 1200 UTC, 21 December when threat scores at the 0.25
and 0.64 cm threshold were 0.5 and 0.3, respectively, and the bias scores were
1.3 and 0.65, respectively.
During the last 36 hours of the simulation period, the cold front moved
eastward to the Atlantic and a new storm system developed over the Great
Plains. By the end of the period, a 992 mb low was over the eastern border of
South Dakota and a cold front extended southward from the low-pressure center
into eastern Texas. Light to moderate precipitation continued along the
frontal zone of the east coast storm system and moderate precipitation
occurred in the warm air sector of the Great Plains storm system where amounts
exceeded 3 cm in Missouri and Illinois. The model did fairly well in the
precipitating regions, but the bias score at the 0.25 cm threshold for 1200
UTC, 22 December was extremely large (18.0) because it predicted precipitation
in Texas and Oklahoma that did not occur. Except for 1200 UTC, 22 December,
the threat scores at the 0.25 and 0.64 cm threshold ranged from 0.25 to 0.50
and bias scores from 0.5 to 1.25.
The overall assessment of the simulation for Case No. 58 is that MM4/FDDA
performed fairly well and the results are suitable for RADM. The SI scores
were very good and the phase and amplitude errors in the primitive variables
were quite small. The threat and bias scores were reasonably good except for
those during the early part of the simulation period when precipitation was
fairly light and for those at 1200 UTC on 22 December.
Assessment by:
Phil Haagenson
215
-------
APPENDIX E
MM4/FDDA Output Volume Names
The MM4/FDDA program creates output volumes suitable for input to RADM.
Each output volume contains many files. One file is written for each hour of
the simulation period (usually five days, but sometimes less). An End-of-File
mark (EOF) separates each hourly data file. The names of the volumes for the
61 cases appear in Table El.
216
-------
Table El. List of volume names for MM4/FDDA output
archived at NCAR.
Case No. Volume Name Dates
(YR/MO/DY/HR)
1/STAUFFER/EPARADM/OUT/MM4/FDDA/DF/P01A012081040700-
81041200
2 /STAUFFER/EPARADM/OUT/MM4/FDDA/DF/P02A1084 81041112-
81041500
3 /STAUFFER/EPARADM/OUT/MM4/FDDA/DF/P03A0120 81042000-
81042500
4 /STAUFFER/EPARADM/OUT/MM4/FDDA/DF/P04A0120 80071200-
80071700
5 /STAUFFER/EPARADM/OUT/MM4/FDDA/DF/P04B1084 80071612-
80072000
6 /CHENS/EPARADM/OUT/01DAY50Z/MM4/FDDA/MM40UT 82012700-
82020100
7 /CHENS/EPARADM/OUT/02DAY50Z/MM4/FDDA/MM40UTC 82031800-
82032300
8 /CHENS/EPARADM/OUT/03DAY50Z/MM4/FDDA/MM40UT 82062700-
82070200
9 /CHENS/EPARADM/OUT/04DAY50Z/MM4/FDDA/MM40UTC 82080600-
82081100
10 /CHENS/EPARADM/OUT/10DAY50Z/MM4/FDDA/MM40UT 88081900-
88082400
11 /CHENS/EPARADM/OUT/11DAY50Z/MM4/FDDA/MM40UT 88082800-
88090200
12 /CHENS/EPARADM/OUT/12DAY412Z/MM4/FDDA/MM40UT 88090112-
88090412
13 /CHENS/EPARADM/OUT/13DAY50Z/MM4/FDDA/MM40UT 88090400-
88090900
14 /CHENS/EPARADM/OUT/14DAY412Z/MM4/FDDA/MM40UT 88090812-
88091212
15A /CHENS/EPARADM/OUT/15ADAY50Z/MM4/FDDA/MM40UT 88091200-
88091612
217
-------
Table El. (Continued)
Case No. Volume Name Dates
(YR/MO/DY/HR)
15B /CHENS/EPARADM/OUT/15BDAY40Z/MM4/FDDA/MM40UT 88091600-
88091912
16 /CHENS/EPARADM/OUT/16DAY50Z/MM4/FDDA/MM40UT 88091900-
88092400
17 /CHENS/EPARADM/OUT/17DAY512Z/MM4/FDDA/MM40UT 88092312-
88092812
18 /CHENS/EPARADM/OUT/18DAY50Z/MM4/FDDA/MM40UT 82041300-
82041800
19 /STAUFFER/EPARADM/OUT/MM4/FDDA/DF/P19A0120 82092300-
82092800
20 /CHENS/EPARADM/OUT/20DAY50Z/MM4/FDDA/MM40UT 82121300-
82121800
21 /CHENS/EPARADM/OUT/21DAY50Z/MM4/FDDA/MM40UT 83052700-
88060100
22 /CHENS/EPARADM/OUT/22DAY50Z/MM4/FDDA/MM40UT 83080200-
83080700
23 /STAUFFER/EPARADM/OUT/MM4/FDDA/DF/P23A0120 83090700-
83091200
24 /STAUFFER/EPARADM/OUT/MM4/FDDA/DF/P24A0120 83103000-
83110400
25 /STAUFFER/EPARADM/OUT/MM4/FDDA/DF/P25A0120 83091200-
83091700
26 /STAUFFER/EPARADM/OUT/MM4/FDDA/DF/P26A0120 84031500-
84032000
27 /STAUFFER/EPARADM/OUT/MM4/FDDA/DF/P27A0120 84081900-
84082400
28 /STAUFFER/EPARADM/OUT/MM4/FDDA/DF/P28A0120 84061000-
84061500
29 /STAUFFER/EPARADM/OUT/MM4/FDDA/DF/P29A0120 84090700-
84091200
218
-------
Table El. (Continued)
Case No. Volume Name Dates
(YR/MO/DY/HR)
30 /CHENS/EPARADM/OUT/30DAY50Z/MM4/FDDA/MM40UT 84071400-
84071900
31 /STAUFFER/EPARADM/OUT/MM4/FDDA/DF/P31A0120 85103100-
85110500
32 /STAUFFER/EPARADM/OUT/MM4/FDDA/DF/P32A0120 85071000-
85071500
33 /CHENS/EPARADM/OUT/33DAY50Z/MM4/FDDA/MM40UT 85043000-
85050500
34 /STAUFFER/EPARADM/OUT/MM4/FDDA/DF/P34A0120 85111400-
85111900
35 /STAUFFER/EPARADM/OUT/MM4/FDDA/DF/P35A0120 85092400-
85092900
36 /STAUFFER/EPARADM/OUT/MM4/FDDA/DF/P36A0120 85100900-
85101400
37A /CHENS/EPARADM/OUT/37ADAY40Z/MM4/FDDA/MM40UT 88081300-
88081700
37B /CHENS/EPARADM/OUT/37BDAY312Z/MM4/FDDA/MM40UT 88081612-
88081912
38A /CHENS/EPARADM/OUT/38ADAY312Z/MM4/FDDA/MM40UT 88082312-
88082612
38B /CHENS/EPARADM/OUT/38BDAY30Z/MM4/FDDA/HM40UT 88082600-
88082812
39 /CHENS/EPARADM/OUT/39DAY40Z/MM4/FDDA/MM40UT 88092800-
88100100
40 /CHENS/EPARADM/OUT/40DAY50Z/MM4/FDDA/MM40UT 82110400-
82110900
41 /CHENS/EPARADM/OUT/41DAY50Z/MM4/FDDA/MM40UT 82051200-
82051700
42 /CHENS/EPARADM/OUT/42DAY50Z/MM4/FDDA/MM40UT 82060800-
82061300
219
-------
Table El. (Continued)
Case No. Volume Name Dates
(YR/MO/DY/HR)
43 /CHENS/EPARADM/OUT/43DAY50Z/MM4/FDDA/MM40UT 85071500-
85072000
44 /STAUFFER/EPARADM/OUT/MM4/FDDA/DF/P44A0120 85121400-
85121900
45 /STAUFFER/EPARADM/OUT/MM4/FDDA/DF/P45A0120 85042600-
85050100
46 /STAUFFER/EPARADM/OUT/MM4/FDDA/DF/P46A0120 82013000-
82020400
47 /STAUFFER/EPARADM/OUT/MM4/FDDA/DF/P47A0120 83120700-
83121200
48 /STAUFFER/EPARADM/OUT/MM4/FDDA/DF/P48A0120 85110200-
85110700
49 /STAUFFER/EPARADM/OUT/MM4/FDDA/DF/P49A0120 85020700-
85021200
50 /STAUFFER/EPARADM/OUT/MM4/FDDA/DF/P50A0120 83111100-
83111600
51 /STAUFFER/EPARADM/OUT/MM4/FDDA/DF/P51A0120 88110200-
88110700
52 /STAUFFER/EPARADM/OUT/MM4/FDDA/DF/P52A1120 88110612-
88111112
53 /STAUFFER/EPARADM/OUT/MM4/FDDA/DF/P53A0120 88111100-
88111600
54 /STAUFFER/EPARADM/OUT/MM4/FDDA/DF/P54A1120 88111512-
88112012
55 /CHENS/EPARADM/OUT/55DAY50Z/MM4/FDDA/MM40UT 88112000-
88112500
56 /CHENS/EPARADM/OUT/56DAY512Z/MM4/FDDA/MM40UT 88112412-
88112900
57 /CHENS/EPARADM/OUT/57DAY40Z/MM4/FDDA/MM40UT 88121600-
88122000
220
-------
Table El. (Continued)
Case No. Volume Name Dates
(YR/MO/DY/HR)
58 /CHENS/EPARADM/OUT/58DAY412Z/MM4/FDDA/MM40UT 88121912-
88122300
221
------- |