Climate Change Indicators in the United States Technical Documentation Freeze Thaw Conditions

Freeze-Thaw Conditions

Identification

1. Indicator Description

This indicator measures the number of days with unfrozen conditions of the land surface in the
contiguous 48 states and Alaska between 1979 and 2019. The balance between frozen and thawed
conditions can be an important factor in determining impacts to surface hydrology—including
evapotranspiration and the timing and extent of seasonal snowmelt (Kim et al., 2017a). It can also be an
important factor in determining the potential growing season for vegetation, which relates to landscape
phenological shifts and important impacts on agriculture and natural resource sectors (Weltzin et al.,
2020). For instance, some pests and pathogens affecting forests and crops are projected to benefit from
warmer temperatures and shorter frozen seasons. A decrease in frozen days may also affect habitat
conditions and wildfire risk (USGCRP, 2018).

This indicator focuses on changes in the number of unfrozen days, which is calculated for each year as a
difference or anomaly compared with the long-term mean (1979-2019). This indicator complements
ground-based measurements by using satellite observations that detect a freeze-thaw (FT) signal from
microwave brightness temperature measurements that are sensitive to changes in the relative
abundance of liquid water (e.g., soil moisture) at the land surface between frozen and non-frozen
conditions. Previous studies using these observational data provide evidence of an increasing annual
thaw cycle and general reduction in temperature constraints on vegetative growth over the Northern
Hemisphere from regional climate warming (Kim et al., 2017a). Components of this indicator include:

• Number of unfrozen days per year in the contiguous 48 states (Figure 1).

• Number of unfrozen days per year in Alaska (Figure 2).

2. Revision History

April 2021: Indicator published.

Data Sources

3. Data Sources

FT data were provided by Drs. Youngwook Kim and John Kimball of the University of Montana. The FT
data were developed from the Nimbus-7 Scanning Multichannel Microwave Radiometer (SMMR)
Pathfinder, Special Sensor Microwave Imager (SSM/I), and SSM/I Sounder (SSMIS) datasets (Armstrong
et al., 2015; Knowles et al., 2000).

4. Data Availability

EPA obtained the data for this indicator from Dr. Youngwook Kim and Dr. John Kimball at the University
of Montana. They published a paper detailing the development of an updated FT Earth system data

Technical Documentation: Freeze-Thaw Conditions

-------
record, or FT-ESDR (Kim et al., 2017a, 2017b), and they provided EPA with a summary file containing
annual unfrozen days for the contiguous 48 states, Alaska, and the contiguous 48 states plus Alaska. The
current indicator represents version 042 of Kim and Kimball's dataset.

All raw source data are available for download in the form of satellite images. SMMR data and data
descriptions are available on the web at: https://nsidc.org/data/nsidc-0071. SSM/I-SSMIS data and data
descriptions are available on the web at: https://nsidc.org/data/nsidc-0032. The global FT-ESDR is
archived and distributed for public access through the National Snow and Ice Data Center (NSIDC) at:
https://nsidc.Org/data/nsidc-0477/versions/4. There are no confidentiality issues that could limit
accessibility.

Methodology

5. Data Collection

This indicator is based on the number of unfrozen days per year in the contiguous 48 states and Alaska.
It was developed by analyzing satellite microwave brightness temperature (Tb) observations, which are
sensitive to changes in the relative abundance of liquid water at the land surface between frozen and
non-frozen conditions. This method specifically focuses on the condition of the land surface (frozen
ground, where soil moisture is frozen). It is not limited to measuring the condition of open bodies of
water; in fact, steps have been taken to avoid relying on map pixels that are dominated by large water
bodies.

The authors developed the FT-ESDR using the SMMR, SSM/I, and SSMIS satellite sensor records:

• SMMR (Knowles et al., 2000): The SMMR onboard the Nimbus-7 Pathfinder satellite collected
brightness temperature data from October 1978 to August 1987. The global data were collected
at a resolution of 25 kilometers at five channels including 37 gigahertz (GHz) for both vertical
and horizontal polarizations. The sensor collected data twice a day on alternate days (with no
data collection on the intervening days) at local noon and local midnight. Documentation and
more information on the data can be found at: https://nsidc.Org/data/nsidc-0071#.

• SSM/I-SSMIS (Armstrong et al., 2015): The SSM/I and SSMIS instruments started collecting
brightness temperature data in July 1987, and data collection continues today. The global data
were collected at a resolution of 25 kilometers at four channels including 37 GHz for both
vertical and horizontal polarizations (only vertical for 22 GHz). The sensors collect twice-daily
data with overpasses at about 6:00 a.m. and 6:00 p.m. The SSM/I and SSMIS sensor series data
consist of measurements from multiple instruments; the intercalibration and data processing
are documented on their website (www.remss.com/support/known-issues/). Documentation
and more information on the data can be found at: https://nsidc.Org/data/nsidc-0032#.

This satellite-based method offers a useful complement to other analyses that are based on air
temperatures measured at weather stations. Air temperature data are widely used, quality controlled,
and highly precise, so they are certainly a viable means to track temperature trends, as EPA has done in
several of its other indicators. Nonetheless, this satellite-based method adds value because it enables
detection of the FT status of the ground, and its coverage of the entire land surface means it can
characterize the condition of locations that might not be well represented by the long-term weather

Technical Documentation: Freeze-Thaw Conditions

-------
station network, due to the limits of station density and unique local topographic and microclimate
conditions.

6. Indicator Derivation

The FT-ESDR was developed for land areas where the average number of days with freezing
temperatures exceeds five per year based on surface air temperature (SAT) daily minima over a 36-year
record (1979-2014). This indicator is also restricted to land areas with at least some vegetation, as
defined from a MODIS land cover map, and limited to areas that are not permanently frozen. Thus, it
excludes large water bodies and permanent ice/snow features. These considerations allow the analysis
to focus on areas (covering most of the country) where FT cycles influence vegetative growth. Figure TD-
1 shows areas that meet the requirement for at least five freezing days per year and were not excluded
for other reasons described above.

This map shows the FT-ESDR domain for the United States and neighboring countries. White shading
shows grid cells where data cannot be computed because there are fewer than five frozen days per year
on average¦, or because the landscape is unvegetated, a large open water body, or permanently frozen.
Black pixels represent the spatial domain covered by this indicator. Data source: Kim and Kimball2020.

The annual grid-cell-wise FT classification thresholds were calculated separately for morning (a.m.) and
evening (p.m.) satellite overpass Tb retrievals using corresponding daily minimum and maximum SAT
measurements. The resulting a.m. and p.m. FT classifications were combined into a daily composite with
four discrete classification levels. Each grid cell day in the FT-ESDR was assigned as frozen, non-frozen,
transitional, or inverse transitional based on how the morning and evening brightness observations
compared with their respective thresholds. Days with a.m. and p.m. frozen were assigned as frozen
days; a.m. and p.m. thawed were non-frozen days; a.m. frozen and p.m. thawed were transitional days;
and a.m. thawed and p.m. frozen were inverse transitional days. The resulting FT-ESDR, with FT values
for each grid cell day in the 1979-2019 period, formed the basis for this indicator. Kim et al. (2017a)
provide a complete description of the analytical procedures used to develop the FT-ESDR.

Technical Documentation: Freeze-Thaw Conditions

-------
The data used in the indicator are derived from the global FT-ESDR, cropped to the shape of the
contiguous 48 states and Alaska. The annual non-frozen season was defined from the daily FT-ESDR as
the total number of non-frozen (a.m. and p.m. thawed) days per year. Non-frozen days averaged over all
grid cells in the year within the contiguous 48 states and Alaska were used to determine annual non-
frozen season anomalies. The anomalies were calculated as annual differences from the long-term mean
based on the period of record (1979-2019).

Figures 1 and 2 show time series for the number of non-frozen days in a calendar year, for the
contiguous 48 states and for Alaska. Each graph shows each year's deviation from the 1979-2019 long-
term average, which is set at zero for a reference baseline. Thus, if year n shows a value of 4, it means
that year had four more unfrozen days than usual. Note that the choice of baseline period will not affect
the shape or the statistical significance of the overall trend; it merely moves the trend up or down on
the graph in relation to the point defined as "zero."

For reference, Figure TD-2 shows the effect of combining the contiguous 48 states and Alaska into a
single graph. The combined results naturally align more closely with the contiguous 48 states (Figure 1
of the indicator) because it has a much larger total land area than Alaska.

Figure TD-2. Number of Unfrozen Days in the Contiguous 48 States and Alaska, 1979-2019

> 20

txO "

E
o

J= -10
c
o

|-20

-30

Linear trend: +0.37 days per year

1979

1984

1989

1994

1999
Year

2004

2009

2014

2019

This figure shows the number of unfrozen days in the contiguous 48 states and Alaska compared with the
1979-2019 average. For each year, the bar represents the number of days shorter or longer than
average. Positive numbers represent years with more unfrozen days than average. The trend line
represents an ordinary least-squares linear regression. Data source: Kim and Kimball2020.

Technical Documentation: Freeze-Thaw Conditions

-------
The map in Figure 3 shows the long-term change in annual unfrozen days across North America
(contiguous 48 states, Alaska, and Canada). This trend analysis uses methods described in Kim et al.
(2012), including Sen's slope regression and Kendall's tau for significance. The annual rate of change was
multiplied by the number of years to derive an estimate of total change. The color ramp accommodates
the vast majority of observations (-40 to 40), a range that contains 99.2 percent of the pixels in the
spatial domain. A few of the remaining pixels range up to nearly +100 unfrozen days.

7. Quality Assurance and Quality Control

Information on quality assurance (QA) and processing protocols for the SMMR dataset can be found at:
https://nsidc.org/data/nsidc-0071. An analysis of processed data and expected residual errors may be
found in Njoku et al. (1998). Information on QA and processing protocols for the SSM/I-SSMIS dataset
can be found at: https://nsidc.org/data/nsidc-0032.

The development of the FT-ESDR involved extensive QA and quality control (QC) measures for assessing
the reliability of the metric. The authors of Kim et al. (2017a) compared the FT-ESDR classifications with
independent in situ daily minimum and maximum SAT measurements from 4,253 +/- 632 (interannual
standard deviation) weather stations and generated a QA map of low and high relative mean annual
classification accuracy. Manual inspection of this map (Figure 6 in Kim et al., 2017a; reproduced in Figure
TD-1) shows the predominance of "good" (85-95 percent agreement) and "best" (>95 percent
agreement) relative quality in the contiguous 48 states and Alaska. The authors of Kim et al. (2017a) also
flagged other potential factors affecting FT classification agreement. They compared grid-cell-wise
metrics of frozen season duration, primary spring thaw date, and non-frozen season duration against
independent cryosphere data records and discussed reasons for discrepancy. These data came from the
Global Lake and River Ice Phenology Database, annual ice breakup dates for the Tanana River in Alaska,
and the NASA MEaSUREs Greenland surface melt record.

Readers can find more details in Kim et al. (2017a), which contains detailed results of the QA assessment
and comparison against independent data. Estimated annual QA maps and grid-cell-level daily QC flag
information is included with the global FT-ESDR database distributed through NSIDC (Kim et al., 2017b).

Analysis

8. Comparability Over Time and Space

All the satellite instruments used for this indicator have collected data in a consistent manner
worldwide. However, there have been some differences in collection and data processing overtime. The
team that developed this indicator has taken several steps to adjust for these differences and ensure
that the resulting data can be compared credibly over the entire period of record.

The SMMR data were collected at a frequency of two days. Time gaps in data also occurred, as described
in more detail in the "Limitations of the Data" section of the data documentation:
https://nsidc.org/data/nsidc-0071. Ground track errors, also described in the preceding link, resulted in
small errors (on the order of 0.1 °C) for January 1981 through May 1983. The missing Tb data attributed
to orbital gaps between satellite overpasses were filled on a grid-cell-wise basis by linear interpolation
of temporally adjacent, successful Tb retrievals to generate spatially and temporally consistent daily (AM
and PM overpass) Tb observations (Kim et al., 2011). The FT-ESDR contains QC flags that are spatially and

Technical Documentation: Freeze-Thaw Conditions

-------
temporally dynamic and assigned on a per-grid-cell basis to denote missing satellite Tb records that are
subsequently gap-filled through temporal interpolation of adjacent Tb retrievals prior to the FT
classification (Kim et al., 2017b).

The SSM-I/SSMIS data may contain geolocation errors in input Remote Sensing Systems swath data of
up to 10 km and additional error from nearest-neighbor interpolation from the over-sampled array of
approximately 6 km. Additionally, the SSM/I and SSMIS sensor series data consist of measurements from
multiple instruments, but steps were taken to harmonize these measurements through intercalibration.
These processing steps are documented at: www.remss.com/support/known-issues.

The SMMR and SSM/I-SSMIS datasets also have differing data collection and processing methodologies,
as described in more detail in their respective documentation: https://nsidc.org/data/nsidc-0071 and:
https://nsidc.org/data/nsidc-0032. For SMMR, the two sensor overpass times are midnight and noon,
and data were collected on alternate days; the SSM/I and SSMIS sensors have overpass times at
approximately 6 AM and 6 PM and collect daily measurements. The authors developed annual FT
thresholds separately for AM and PM overpasses. For most years in the FT-ESDR, the annual thresholds
were developed from data with consistent satellite image collection times, which would account for the
change in measurement collection times across years. However, 1987 FT-ESDR data are based on
measurements from both SMMR and SSM-I/SSMIS, which may affect the accuracy of the FT-ESDR data
for that year because the AM and PM FT thresholds were calibrated on measurements at different times
of the day. Before determining AM and PM FT classification thresholds for each grid cell by empirical
linear regression between model reanalysis daily SAT and satellite Tb retrievals, the SMMR record was
matched to the SSM/I record using pixel-wise adjustment of the SMMR and SSM/I Tb measurements for
1987 to ensure cross-sensor consistency (Kim et al., 2012).

Care has been taken to account for missing data, and thereby prevent missing data from biasing the
resulting annual averages. According to Kim et al. (2017a), on average, annual satellite records were
missing data for 34.3 +/- 24.3 percent of the relevant land area due to orbital gaps between satellite
overpasses. To generate the FT-ESDR, these spatial data were filled on a grid-cell-wise basis by linear
interpolation of temporally adjacent, successful Tb retrievals based on peer-reviewed methods (Kim et
al., 2011). The authors also interpolated large gaps of missing data in January and December 1987 and
January 1988 using empirical relationships developed from ERA-lnterim (Dee et al., 2011) global model
reanalysis SAT and satellite Tb data records. In 2020, a new ERA5-based calibration (instead of ERA-
lnterim) was applied to the entire record. It verified consistent performance relative to the prior (v04)
record and global weather station observations.

The grid-cell-wise Tb threshold and annual calibration used for the FT classification algorithm reduce the
potential influence of spatial and temporal variations in climate and land surface conditions on FT
classification accuracy, and promote greater compatibility in global product performance over the long-
term record (Kim et al., 2017a).

9. Data Limitations

Factors that may impact the confidence, application, or conclusions drawn from this indicator are as
follows:

Technical Documentation: Freeze-Thaw Conditions

-------
1. The agreement of the FT-ESDR with in situ temperature measurements is not perfect, as
described in Section 10. However, the FT-ESDR is well tested and provides a reasonably reliable
complement to ground-based data.

2. The FT classification accuracy was found to be inversely proportional to the spatial fraction of
open water bodies (Fw) within a grid cell. The FT ESDR QC flags distinguish grid cells with large
open water areas (Fw > 0.2) (Kim et al., 2017a).

10. Sources of Uncertainty

Many factors can diminish the accuracy of satellite detection of FT status. These factors include terrain
type; number of SAT validation stations; length of seasonal transitions; and land surface heterogeneity,
including presence of surface water and terrain complexity. Time of day of measurements also
influences agreement with SAT records. Thus, the agreement between the FT-ESDR and in situ SAT
measurements varies, with an average agreement of 90.3 +/-1.4 percent (inter-annual standard
deviation) for the PM overpass and 84.3 +/-1.7 percent (inter-annual standard deviation) for the AM
overpass. The discrepancy may be attributed to spatial mismatch between the in situ weather station
observations and overlying coarser satellite footprint. See Kim et al. (2017a) for further discussion of
factors affecting FT classification agreement.

The authors interpolated to fill in gaps in data, and this may present a source of uncertainty in the
overall trend. Changes in measurement techniques over time and satellite measurement instrument
errors also contribute to uncertainty.

11. Sources of Variability

At any given location, the number of days experiencing frozen conditions naturally varies from year to
year as a result of normal variation in weather patterns, multi-year climate cycles such as the El Nino-
Southern Oscillation and Pacific Decadal Oscillation, and other factors. Overall, this type of variability
should not impact the conclusions that can be inferred from the trends shown in this indicator.

There is also inherent variability over space in any given year, as this indicator is aggregated over a wide
variety of climate zones. To give readers a sense of this natural geographic variation, Figure TD-3 shows
the absolute number of unfrozen days per year (the spatial mean) along with lines that indicate one
standard deviation in each direction to suggest the spread of the distribution. Figure TD-3 shows that
the average annual number of unfrozen days across the entire coverage area is approximately 250, but
there is a substantial geographic spread. If one were to assume an approximately normal distribution,
one would infer that about 68 percent of the grid cells on the map had a number of unfrozen days that
fell between the upper and lower bounds shown in Figure TD-3.

Technical Documentation: Freeze-Thaw Conditions

-------
Figure TD-3. Number of Unfrozen Days in the Contiguous 48 States and Alaska, 1979-2019, with
Standard Deviation Showing Geographic Variation

350

300

><
TO
T3
C

-------
Figure TD-4. Significance of 1979-2019 Trends in the Annual Number of Unfrozen Days

This map shows where long-term trends in unfrozen days are statistically significant. Pixels with p < 0.1
(i.e., significant to a 90 percent level) are shaded black. Data source: Kim and Kimball, 2020.

References	

Armstrong, R., K. Knowles, M. Brodzik, and M.A. Hardman. 2015. DMSP SSM/I-SSMIS Pathfinder daily
EASE-Grid brightness temperatures, Version 2 (1987-2014). NASA DAAC at the National Snow and Ice
Data Center, http://nsidc.org/data/nsidc-0032.html.

Dee, D.P., et al. 2011. The ERA-lnterim reanalysis: Configuration and performance of the data
assimilation system. Q. J. Roy. Meteorol. Soc. 137:553-597.

Kim, Y., and J. Kimball. 2020 update to data originally published in: Kim, Y., J.S. Kimball, J. Glassy, and J.
Du. 2017. An extended global Earth system data record on daily landscape freeze-thaw status
determined from satellite passive microwave remote sensing. Earth Syst. Sci. Data 9:133-147.

Kim, Y., J.S. Kimball, K.C. McDonald, and J. Glassy. 2011. Developing a global data record of daily
landscape freeze/thaw status using satellite passive microwave remote sensing. IEEE Trans. Geosci.
Rem. Sens. 49:949-960.

Kim, Y., J.S. Kimball, K. Zhang, and K.C. McDonald. 2012. Satellite detection of increasing northern
hemisphere non-frozen seasons from 1979 to 2008: Implications for regional vegetation growth.
Remote Sens. Environ. 121:472-487. doi:10.1016/j.rse.2012.02.014

Kim, Y., J.S. Kimball, J. Glassy, and J. Du. 2017a. An extended global Earth system data record on daily
landscape freeze-thaw status determined from satellite passive microwave remote sensing. Earth Syst.
Sci. Data 9:133-147. doi:10.5194/essd-9-133-2017

Technical Documentation: Freeze-Thaw Conditions

9

-------
Kim, Y., J.S. Kimball, J. Glassy, and K.C. McDonald. 2017b. MEaSUREs global record of daily landscape
freeze/thaw status, version 4. NASA DAAC at the National Snow and Ice Data Center.
doi:10.5067/MEASURES/CRYOSPHERE/nsidc-0477.004

Knowles, K., Njoku, E. G., Armstrong, R., and Brodzik, M. 2000. Nimbus-7 SMMR Pathfinder daily EASE-
Grid brightness temperatures (1979-1987). NASA DAAC at the National Snow and Ice Data Center.
http://nsidc.org/data/nsidc-0071.html.

Njoku, E.G., B. Rague, and K. Fleming. 1998. Nimbus-7 SMMR Pathfinder brightness temperature data.
NASA Jet Propulsion Laboratory Publication 98-4.

USGCRP (U.S. Global Change Research Program). 2018. Impacts, risks, and adaptation in the United
States: Fourth National Climate Assessment, volume II. Reidmiller, D.R., C.W. Avery, D.R. Easterling, K.E.
Kunkel, K.L.M. Lewis, T.K. Maycock, and B.C. Stewart (eds.). www.globalchange.gov/nca4.

Weltzin, J., et al. 2020. Seasonality of biological and physical systems as indicators of climatic variation
and change. Climatic Change 163:1755-1771. doi:10.1007/sl0584-020-02894-0

Technical Documentation: Freeze-Thaw Conditions

10

-------