EPA/600/R-14/467 December
www.epa.gov/researc
MULTI-TEMPORAL LAND USE
GENERATION FOR THE OHIO RIVER
BASIN
Office of Research and Development
Water Supply and Water Resources Division
-------
EPA/600/R-14/467
December2014
FINAL REPORT
MULTI-TEMPORAL LAND USE GENERATION FOR THE OHIO RIVER BASIN
By
Dr. Bryan C. Pijanowski, and Mr. Jarrod Doucette
Human Environment Modeling and Analysis Laboratory,
Department of Forestry and Natural Resources
Purdue University,
West Lafayette, IN 47907-2022
Contract No. EP-12-C-000018 (MOD 1)
Dr. EllyP.H. Best,
Work Assignment Manager
Water Quality Management Branch
Water Supply and Water Resources Division
National Risk Management Research Laboratory
Office of Research and Development
U.S. Environmental Protection Agency
EPA's National Risk Management Research Laboratory,
Andrew W. Breidenbach Environmental Research Center,
25 W M.L. King Drive,
Cincinnati, OH 45268
-------
DISCLAIMER
The U.S. Environmental Protection Agency (EPA), through its Office of Research and
Development, funded and managed, or partially funded and collaborated in, the research described
herein under Contract No. EP-12-C-000018 (MOD 1) to Purdue University.
This document has been reviewed in accordance with U.S. Environmental Protection Agency
(EPA) policy and approved for publication. The views expressed in this report are those of the
author[s] and do not necessarily reflect the views or policies of EPA. Mention of trade names or
commercial products does not constitute endorsement or recommendation for use. The quality of
secondary data referenced in this document was not independently evaluated by EPA and Purdue
University.
11
-------
ABSTRACT
A set of backcast and forecast land use maps of the Ohio River Basin (ORB) was developed that
could be used to assess the spatial-temporal patterns of land use/land cover (LULC) change in this
important basin. This approach was taken to facilitate assessment of integrated sustainable
watershed management (SIWM) planning in the ORB at various spatial scales by providing
information on historical LU patterns, future LU trends, and LU legacy maps illustrating spatial
and temporal changes in LULC in relation to groundwater travel time. The latter information,
combined with water resource-related information on water quality, quantity and ecosystem
service values, is expected to provide a quantitative basis for scenario exploration and optimization
in support of SIWM over short and longer periods of time. Interest into SIWM on a watershed
scale, and supporting research, has increased recently within EPA and other organizations active
in monitoring water quality and quantity, water use, and watershed management planning.
The overarching purpose of this study was to develop a set of backcast and forecast land use maps
for the ORB that could be used to assess the spatial-temporal patterns of LUC in this basin. The
Land Transformation Model (LTM), an artificial neural network and GIS-based tool, was used to
conduct this study. This tool has been designed to forecast LU changes into the future and simulate
LU patterns in the past. The USGS's National Land Cover Database (NLCD) was used to develop
a forecast and backcast set of GIS maps at 30-m resolution. Simulations back in time included the
transformation of land into and out of agriculture, and the loss of urban LU. Backcast LU maps
were generated using a training of two time periods (NLCD 2001 and 1992) with the amount of
agriculture and urban change scaled to data from the USDA Land In Farms database and the US
Census Bureau's decadal Year Built statistic as reported in the 2000 housing census. A recent
version of the LTM (2012) was ported to a super computer and receded to perform the backcast
simulation for the ORB. A GIS was used to create spatial inputs for both models. A separate
urbanization model was merged with the backcast models. Model simulations at 3-km spatial
resolution were considered acceptable.
Backcast results indicated that: (1) approximately 90% of the ORB has remained in the same
LULC class since 1930; (2) agriculture was the dominant LULC class from 1930 to the mid-1960s;
and (3) significant amounts of agriculture have been lost over the last 60 years, largely to forest.
Consequently, LU legacies should be considered in forest management plans for this basin.
Forecast results indicated that: (1) metropolitan areas are likely to have the greatest amount of LU
legacy locations, and (2) the spatial variability of LU legacies across the ORB is significant.
Greatest LU legacies were found in areas nearest to the Ohio River proper and least LU legacies
in the northwestern part of the Basin. The potential impacts of historical LUC on sensitive areas
of watersheds, in particular areas that potentially recharge streams (i.e., riparian zones of
permanent streams and rivers), were examined in the Upper White, the Sugar, the Tippecanoe, and
the Upper Wabash River watersheds. LU persistence was found to be greater within the entirety
of these watersheds than within their riparian zones (83 to 93% versus 74 to 88%, respectively),
suggesting that riparian zones have a greater potential for LU legacies than upland areas. Finally,
an analysis of all HUC-8s in the ORB showed that many have surpassed the regional thresholds
for stream water quality health of > 10% urban or > 38% agricultural LU since 2010, most of
which are located in the northern part of the Basin, and increases in urban LU and associated
negative impacts on water quality are expected by 2050.
in
-------
ACKNOWLEDGEMENTS
This report has been prepared by Dr. Bryan C. Pijanowski and Mr. Jarrod Doucette, of the Human
Analysis and Environment Laboratory of Purdue University. The following individuals are
gratefully acknowledged for their assistance in modeling, analysis and preparing data for use in
this study: Dr. Amin Tayebbi, Dr. Burak Pekin, Jim Plourde, Andrew Bagnara, Dr. David Braun,
and Dr. Kimberly Robinson.
Technical lead, direction and coordination for this project were provided by Dr. Elly P.H. Best,
EPA/ORD/NRMRL/WSWRD/WQMB. Authors are grateful for the guidance on the project and
final report provided by Dr. Best.
IV
-------
EXECUTIVE SUMMARY
A flurry of research in land change science over the last several years has found that historical land
uses significantly shape current ecosystem structure and function. These historical land uses, often
referred to as land use legacies, have been shown to affect plant community structure, animal
abundances and distributions, water quality and biogeochemical fluxes at a variety of spatial and
temporal scales. Historical land use transition pathways at any given location can be complex.
Many areas in the Eastern United States were cleared for agriculture over a century ago, then
abandoned and converted into forest; recent urban sprawl has resulted in a significant amount of
forested landscapes - including the relatively recently developed forests- transforming to urban
use. Knowing the extent and pattern of land use change over time can provide natural resource
managers with valuable information for developing sustainable management plans.
At the same time, considerable work in land change science has focused on simulating current
trends as impacts from certain futures which may require mitigation or adaptation to the effects of
these land use changes. Currently, about 3-4% of the nation's land area is in urban use, and this
amount of land use is predicted to grow, perhaps twice as much, by 2050. If current trends
continue, how might these changes impact ecosystem structure and function?
The overarching purpose of this study was to develop a set of backcast and forecast land use maps
for the Ohio River Basin (ORB) that could be used to assess the spatial-temporal patterns of land
use change (LUC) in this important basin. Specific objectives of this project included: (1)
quantifying land use/land cover (LULC) changes over time for the major LULC classes; (2)
producing historical, future and LU legacy maps for use in GIS; (3) quantifying the spatial
distribution of similarity between historical, current and future LU maps; (4) characterizing the
distribution of LU and legacies in watersheds of the ORB; and (5) assessing the distribution of LU
legacies in high impact surface/ground water areas within four demonstration watersheds. We
employed an artificial neural network and GIS-based tool, called the Land Transformation Model
(LTM), which has been designed to forecast LUC into the future and simulate LU patterns
historically. The USGS's National Land Cover Database (NLCD) was used to develop a forecast
and backcast set of GIS maps at 30-m resolution, the native resolution of the NLCD. Simulations
back in time included the transformation of land into and out of agriculture, and the loss of urban
LU (as described in the reverse direction, as the model simulates in time backwards). As in
previous work with the LTM, backcast LU maps were generated using a training of two time
periods (NLCD 2001 and 1992) with the amount of agriculture and urban change scaled to data
from the USDA Land in Farms database and the US Census Bureau's decadal Year Built statistic
as reported in the 2000 housing census.
Due to the massive size of the ORB (31,644 columns by 31,191 rows representing over 1.0 x 108
cells), a recent (2012) version of the LTM, ported to a high performance computer cluster (i.e.,
super computer), was receded to perform the backcast simulation. A GIS was used to create spatial
inputs for both models, including distance to urban, distance to roads, density of agriculture and
slope. Calibration and validation of the model were conducted using standard land change
modeling statistics reported in the literature and those developed and published by the Purdue
research team. A stable neural network was achieved after about 100,000 training cycles.
Backcast maps for 1930 through 1990 were produced at ten year time steps and a set of forecasts,
2010 through 2050, were also produced. A LU legacy map was generated that contained codes
for LULC for each decade between 1930 and 1990. Analysis of LULC by an 8-digit hydrologic
unit was performed on LULC forecast maps and summary tables of these were created along with
percent area in urban and agriculture.
A separate urbanization model was merged with the backcast models. Previously published as a
national scale simulation, the urbanization model uses a new spatial-temporal statistical routine
that is coupled to state and national population projections and historical per capita urbanization
-------
rates. New calibration techniques used for the urbanization model were applied to the backcast
simulation model.
We found that the forecast and backcast model performed adequately well at 3-km spatial
resolution. Both location and quantity errors were less than 10%, at 3-km, across the ORB. Using
the model, we estimated that (1) approximately 90% of the ORB has remained in the same LULC
class since 1930; (2) that agriculture historically was the dominant LULC class in the ORB until
about the mid-1960s when forest overtook it as the dominant LU class; and (3) significant amounts
of agriculture have been lost over the last 60 years, a majority of it by transit!oning into forest;
and, thus, LU legacies should be considered in forest management plans for the region. With
regards to historical LUC compared to current, we found that (1) metropolitan areas are likely to
have the greatest amount of LU legacy locations, and (2) the spatial variability of LU legacies
across the ORB is significant. We also noted that areas nearest the Ohio River proper have some
of the greatest (measured in area) LU legacies and areas to the northwest have some of the least
amount (measured in area) of LU legacies.
Four demonstration watersheds were selected to examine the potential impact of historical LUC
on sensitive areas of these watersheds - in particular, areas that potentially recharge streams. To
accomplish this, we examined LU legacy patterns in riparian zones of permanent streams and
rivers in these four watersheds. We found that LU persistence was between 83 to 93% within the
entirety of these watersheds, but slightly less within riparian zones (74 to 88%), suggesting that
riparian zones have a greater potential for LU legacies than the upland areas of watersheds.
Finally, an analysis of all 8-digit hydrologic units in the ORB showed that many of these
watersheds have surpassed what we consider as thresholds for stream water quality health (>10%
urban or >38% agriculture). The distribution of watersheds that exceeded either threshold is
similar; much of the northern areas of the ORB have exceeded urban or agriculture amounts that
might lead to decreased stream health. Currently, 32% (38/12) of the 8-digit hydrologic units
surpass 10% urban, and by 2050, more than half (64/120) will surpass this threshold. We also
predict that the ORB will have 11.83% of its area in urban use by 2050, a 32% increase from the
8.98% appearing in the 2001 NLCD map.
VI
-------
CONTENTS
DISCLAIMER II
ABSTRACT Ill
ACKNOWLEDGEMENTS IV
EXECUTIVE SUMMARY V
CONTENTS VII
ACRONYMS AND ABBREVIATIONS X
1 INTRODUCTION 1
1.1 Project Background 1
1.2 Project Objectives 1
1.3 Report Outline 2
2 PLANNED APPROACH 3
3 DATA SOURCES 4
3.1 Study Area 4
3.2 Land Use Data Processing 5
4 SIMULATION APPROACH 6
4.1 Model Inputs 6
4.2 Artificial Neural Network Topology 8
4.3 LTM Using Meso-Scale County Drivers of Urban and Agriculture Quantities 8
4.4 Running the Backcast LTM-MC on an HPC 11
4.5 Transition Rules Applied 12
4.6 Training Goodness of Fit Statistics 12
4.7 Hydrologic Sensitivity and Transition Pathway Analyses 15
5 MODEL CALIBRATION AND VALIDATION 16
6 SIMULATION RESULT 18
6.1 Backcast Results 18
6.2 Forecast Results 24
6.3 Model Output 29
7 DISCUSSION 31
8 REFERENCES 33
9 APPENDIX 36
9.1 Metadata for the Backcast LTM output 36
vn
-------
FIGURES
Figure 3-1. Study area showing ORB boundaries and counties included in the simulation 4
Figure 3-2. NLCD for 1992 5
Figure 4-1. Maps of drivers (6 samples) used for training of the Artificial Neural Network 7
Figure 4-2. Multiple output Artificial Neural Network typology 8
Figure 4-3. Land in Farms statistics summarized by states 10
Figure 5-1. Goodness-of-fit statistics of location-change results of the model when applied to the
entire 48 states 16
Figure 5-2. Goodness of fit statistics of quantity-change results of the model when applied to the
entire lower 48 states, s, on a scale of 0 to 1, of 3 x 3 km simulation 17
Figure 6-1.The most common BLTM land use transition pathways occurring between 1930 and
1990 for the ORB 19
Figure 6-2. Percentage of each case study watershed (top) and within permanent stream riparian
zones (bottom) that persisted in a land use class from 1930 to 1990 21
Figure 6-3. Map of land use persistence from 1930 through 1990 and locations of change (i.e.,
locations where land use legacies may impact ecosystem dynamics) 23
Figure 6-4. Land Transformation Model projections summarized by decade 24
Figure 6-5. Percentage of land use/cover classes simulated over time for the ORB from 1930
through 2050 with 10-year time steps 25
Figure 6-6. Percentage of land use/cover classes that are predicted to be converted into urban
land use during each decade 25
Figure 6-7. Locations where potential urban-water quality thresholds have been met at the scale
of an 8-digit hydrologic unit (i.e., watershed) for years 2010 through 2050 26
Figure 6-8. Number of 8-digit hydrologic units (watersheds) that have exceeded 10% urban land
use by the year indicated 27
Figure 6-9. Locations where potential agriculture-water quality thresholds have been met at the
scale of an 8-digit hydrologic unit (watershed) for the years 2010 through 2050 28
Vlll
-------
TABLES
Table 4-1. Drivers included in the backcast LTM and their rationale 6
Table 4-2.Sample of year built statistics from the 2000 U.S. Census Bureau Housing Data (U.S.
Census Bureau 2000). Units are houses per county 9
Table 6-1. The most common BLTM land use transition pathways occurring 20
Table 6-2. Transition pathways for the selected demonstration watersheds (A) and their riparian
zones (B). Only the top 5 most common transition pathways are listed here and their
percent area for the watershed and riparian zone. Totals are percentage total area 22
Table 6-3. List of model simulation outputs distributed to EPA by Purdue University 29
IX
-------
ACRONYMS AND ABBREVIATIONS
ANN Artificial Neural Network
BLTM Backcast Land Transformation Model
DEM Digital Elevation Model
GIS Geographic Information System
GWTT Ground Water Travel Time Model
HPC High Performance Compute cluster which is the cyber infrastructure used to run
the backcast and forecast LTM
HUC Hydrologic Unit Code
LIF Land In Farms, a USDA NASS statistic that provides the total amount of land, by
county and reported in acres, that is in cropland and pasture. Dates of reporting
vary over time, but the USDA generally has reported on these data approximately
every 5 years since 1900
LTM Land Transformation Model
LTM-MC Multiple-Class Land Transformation Model (more than one class is simulated for
change at each time step)
LTM-HPC High Performance Compute LTM (version that runs on a HPC cluster)
LU Land use
LUC Land Use Change
LULC Land Use/Land Cover
MC Multiple-Classification output node structure used to train artificial neural networks
where two or more changes are being quantified at the same time
MSE Mean Square Error
NASS National Agricultural Statistics Service of the USDA
NHDPlus National Hydrologic Database Plus of the USGS
NLCD National Land Cover Database produced by the USGS, for years 1992, 2001 and
2006
ORB Ohio River Basin
QSPP Quality Assurance Project Plan
PCM Percent Correct Metric which is the proportion of correctly predicted cells divided
by the number of observed changes occurring between two time steps
QAPP Quality Assurance Proj ect Plan
SIWM sustainable integrated watershed management
ti time step number 1 or the first time step
t2 time step number 2 or the second time step
USGS United States Geological Survey
USDA United States Department of Agriculture
YB Year Built, a U.S. Census Bureau housing statistic
-------
1 INTRODUCTION
1.1 Project Background
Historical land use/land cover (LULC) maps are useful for sustainable management and restoration
planning because understanding how landscape structure and ecosystem services are linked
provides valuable information about baseline (or reference condition) as well as legacy signals
from the past. Land use legacy maps provide natural resource managers with information about
the role that slow hydrological processes, such as groundwater travel time, have on current water
quality in surface water bodies such as rivers and streams. This is especially true when
management and restoration need to consider ecosystem services that are directly tied to water
quality and the dynamics of the hydrologic cycle. Future land use maps assist natural resource
managers to determine areas that might be under risk to land transformation and provide early
warnings to them about potential deleterious impacts to ecosystems.
Interest into sustainable integrated watershed management (SIWM) on a watershed scale, and
supporting research, has increased recently within EPA and other organizations active in
monitoring water quality and quantity, water use, and watershed management planning. A recently
(2011) initiated EPA study to evaluate integrated sustainable watershed management planning in
the Ohio River Basin at various spatial scales requires information on historical land use patterns,
future land use trends, and land use legacy maps illustrating spatial and temporal changes in LULC
in relation to groundwater travel time. The latter information, combined with water resource-
related information on water quality, quantity and ecosystem service values, is expected to provide
a quantitative basis for scenario exploration and optimization in support of SIWM over short and
longer periods of time. We intend to produce basin-wide maps of historical land use patterns (at
decadal time steps from pre-settlement to current), future land use trends (also decadal, from
current to 2050) and demonstrate the application of land use legacy maps in a small portion of the
Ohio River Basin (ORB).
This effort is expected to provide information on the potential impacts of dynamic land use patterns
for sustainable watershed management planning and contribute to the ' Safe and Sustainable Water
Research Program' focus areas, 'Sustainable Water Resource Flows' and ' Sustainable Natural and
Engineered Water Infrastructure Systems'.
1.2 Project Objectives
The objectives of this project are to generate land-use legacy maps for watershed management
from historical and recent land-use maps for the ORB and to provide a proof of concept for the use
of the land use legacy concept in a smaller watershed (e.g., within the basin such as portions of the
Wabash or White River watersheds) where these patterns are likely to impact water quality. Areas
with karst topography cannot be reliably modeled for groundwater patterns and are beyond the
scope of this project.
Once developed, these land use legacy maps may serve as a valuable example that greatly
facilitates collaboration in water resources management research and encourages undertaking of
similar activities by EPA colleagues and non-EPA collaborators.
The results of this project are subject to the Quality Assurance Project Plan (QAPP) ID no W-
16753-QP-l-O (Approval date: 03/12/2012).
-------
1.3 Report Outline
The rest of this report is structured as follows. Section 2 describes the planned approach as outlined
in the project QAPP. Section 3 presents an overview of the data sources used in this study. Section
4 summarizes the approach used to develop the land change model simulations. Model calibration
and validation approaches are presented in Section 5. Section 6 contains the results of the
simulations. Our discussion of the results as related to the QAPP is provided in Section 7.
Contents of each section are summarized as follows:
2.0 Planned Approaches
We summarize the objectives of the modeling study with regards to the larger project goals.
A work flow of the modeling steps is provided here.
3.0 Data Sources
Here we summarize briefly the study area and primary data sources used in the modeling
and analysis.
4.0 Simulation Approach
We describe how we prepared the backcast land transformation model (BLTM) for
simulating backwards in time and the land transformation model high performance
compute (LTM-HPC) for the forecasts and the approach used for simulating in both
directions. We describe the topology of the artificial neural network used, how GIS was
used to prepare inputs and the transition rules that were applied in the Backcast version of
the land transformation model (LTM).
5.0 Calibration and Validation Approach
We describe here the calibration and validation approaches used for the backcast and
forecast LTM and the metrics generated for the ORB and nation (for the forward LTM).
6.0 Simulation Results
The results of the simulations for backcast and forecast LTMs are provided. These
backcast summaries examine LULC change as a sequence of LULC classes (at ten year
time steps between 1930 and 1990) across the entire ORB, by 8-digit hydrologic unit (i.e.,
watershed) and then for riparian buffers for four selected demonstration watersheds. The
forecast results are examined for the ORB in its entirety and for 8-digit hydrologic units as
percentage of urban and agriculture as a function of a water quality threshold.
7.0 Discussion
Our discussion presents an overview of the simulation results as related to the five main
objectives outlined in the QAPP and Section 2 of this report.
-------
2 PLANNED APPROACH
The Backcast Land Transformation Model (BLTM) was used to generate legacy land use maps for
a large part of the Ohio River Basin (ORB), i.e., the part included into the only existing basin-wide
conservation plan, the ORB Fish Habitat Partnership Strategic Plan (cf. Stark 2011).
The BLTM is based on a widely used land change model the 'Land Transformation Model' (LTM)
which has been used to forecast land use change in a variety of areas around the world (US, Europe,
and east Africa). The LTM model is an approach by which large-scale land use change is predicted
with a Geographic Information System (GIS) and artificial neural networks. Future land use maps
are also generated through to 2050 from the current year (in ten year time steps). Methods for
projecting large, basin-wide land use maps with the LTM have been recently described by Tayyebi
et al., (2012). The BLTM is often coupled to a Groundwater Travel Time (GWTT) model
(Pijanowski et al., 2007) or a spatial-temporal summary routine that quantifies land use legacies at
a point or within watersheds over time. Previous land uses are well known to influence soil quality
(Foster et al., 2003), water quality (e.g., Allan, 2004), species composition (e.g., Wallin et al.,
1994) and invasive ability (Brudvig et al., 2011).
The BLTM is a spatial-temporal model that uses current land use maps and historical data from
the agricultural census and U.S. population to construct historical land uses. One use of this model
has been its coupling to a groundwater travel time model to develop land use legacy maps. By
quantifying the differences between current land use and legacy land use, a more accurate
representation of linkage between LULC and current water quality is provided than by current land
use alone, in areas dominated by groundwater. Historical signatures of land use impact current
water quality with an extent that depends on landscape geography and should be considered in
land use and watershed management planning.
Basin-wide historical and future land use projections and demonstration site legacy maps provide
quantitative information about the:
1. Changes over time of the major classes of the 7 Anderson level-I land-use/cover categories
(urban, agriculture, forest, shrubland, open water, wetland, and barren);
2. Historical, future and legacy land use/cover maps as digital maps for use in a GIS;
3. Spatial distribution of similarity between historical, current and future land use maps;
4. Distribution of land use legacies as a function of the surface and groundwater watersheds
(riparian zones) and surface watershed subbasins and major rivers in the ORB; and,
5. Distribution of current land-use/cover patches in high-impact groundwater recharge areas
(riparian zones) for demonstration. Demonstration sites are the Upper Wabash Watershed,
Upper White Watershed, Tippecanoe Watershed and the Sugar Watershed.
-------
3 DATA SOURCES
3.1
Study Area
Data used to build and validate the backcast LTM covered the entire Ohio River Basin (ORB),
which included counties that wholly or partially fall within the basin borders (Figure 3-1). The
ORB includes 456 counties in Illinois, Indiana, Ohio, Pennsylvania, West Virginia, Kentucky and
Tennessee, the size of the area including counties is 515,818 km2 (the ORB covers 421,962 km2).
We also selected four demonstration watersheds to examine land use legacy patterns in more detail
and within the context of known human development patterns. These watersheds included: (1) the
Upper White watershed, which contains much of the Indianapolis, Indiana metropolitan area; (2)
the Sugar Watershed, a rural Indiana watershed that contains a lot of forested riparian zones; (3)
the Tippecanoe Watershed, which has been historically a rural, agricultural watershed and the (4)
Upper Wabash watershed, an agricultural watershed that is currently undergoing transitions to
large-scale livestock production, all located in Indiana.
Legend
| | ORB boundary
_ Counties in the ORB
Figure 3-1. Study area showing ORB boundaries and counties included in the
simulation. Modeling was performed for all counties with land area partially or
wholly within the ORB boundary
-------
3.2
Land Use Data Processing
We used the National Land Cover Database (NLCD) to obtain land use data for the ORB for 2001
and 1992 at 30-m resolution. We reclassified NLCD for both years from Anderson Level 1 to four
main land use classes (e.g. urban, forest, agriculture and other classes; Figure 3-2). In order to
accomplish this, the four developed NLCD classes (e.g. 21, 22,23 and 24) were combined to create
a single urban land use class. All forest (e.g. 41,42 and 43), shrubland (e.g. 51, 52), and herbaceous
vegetation classes (e.g. 51, 52) were combined to create a single natural vegetation cover class
referred to as forest from here on. The agricultural land use classes included pasture (e.g. 81) and
cultivated crops (e.g. 82). Urban, agriculture and forest land use classes occupied 8.90, 37.39,
51.55% of the landscape, respectively, in the ORB in 2001; however, these proportions were 8.57,
37.34 and 52.12% in 1992, respectively. Little land use change has occurred overall in the ORB
between 2001-1992; however, this does not indicate to what extent different counties in the ORB
transitioned from one land use class to another.
NICD 1992
I I Caurtict
Water
Uft»r>
J
30 meter
resolution
NLCD 2001
D Afrimlm fi
^H ForvJt
30 meter
resolution
Figure 3-2. Land use maps used for training of the Artificial
Neural Network
-------
4 SIMULATION APPROACH
4.1
Model Inputs
We applied distance and density functions in ArcGISlO to calculate the distance of each cell from
the nearest land use category and density of land use category around (e.g. urban, forest and
agriculture) the central cell, respectively. The Euclidean distance tool in ArcGISlO was used to
create separate raster maps that stored in each cell the distance from the nearest (1) urban, (2) forest
and (3) agriculture cell. Focal statistics in the neighborhood tool were used to calculate the density
of each main land use class around the central cell. Slope was calculated from the DEM using the
ArcGISlO Spatial Analyst tool. Spatial drivers (Figure 4-1; Table 4-1) used as input for the
backcast LTM-MC simulation included: DEM, slope, distance to town, distance to road, distance
to water, distance to urban, distance to forest, distance to agriculture, distance to capital, density
of agriculture in 10, 50 and 250 m windows, and density of urban in 10, 50 and 250 m windows.
Table 4-1. Drivers included in the backcast LTM and their rationale
Driver
Distance to
nearest road
Distance to
nearest town
Slope
Distance to
nearest urban
Description of Rationale
Road construction has been found to be one of the strongest drivers of
urbanization in the U.S.
People live and work near towns and proximity to cities, towns and villages
strongly influences urbanization
Built environment cannot occur on steep slopes; crops are difficult to
manage large scale using mechanized management. Generally, slopes > 8%
are not farmed in the U.S.
Previous urban cells are well known to create new urban cells in future time
steps because infrastructure for urban use likely exists
pixel
Density of
urban within a
fixed window
size
Urban cells tend to fill in once a certain density of this use is reached
Density of
agriculture
within a fixed
window size
Large homogeneous agricultural plots are more sustainable over time
Distance to
nearest surface
water body
People like to place built structures (e.g., houses) next to lakes and rivers and
are, thus, drivers of urbanization
-------
Distance to roads
(major roads only}
30 meter
resolution
fat
Distance Jo towns
30 meter
resolution
Distance to water
IOmeter
resolution
Slope
SO meter
resolution
St«p|
Density of urban
pixels within
10 tan radius
(ArcGIS focal statistics)
30 meter
resolution
Density of agriculture
pixels within
10 km radius
(AraSiS/ecal statis tics)
IOmeter
resolution
High)
I
Figure 4-1. Maps of drivers (6 samples) used for training of the Artificial Neural Network
-------
4.2
Artificial Neural Network Topology
The backcast model artificial neural network (ANN) topology includes two outputs that have been
coded using two digit codes (Figure 4-2): (1) the cells that experience agriculture gain (e.g.
transition from other LULC classes to agriculture) have been coded (e.g. 1, 0) as the first model
outcome, (2) the cells that experience urban loss (e.g. transition from urban to other LULC classes)
have been coded (e.g. 0, 1) as the second model outcome, and (3) other cells that experience other
types of transitions have been coded as 0, 0.
Two Output-Change in Class 1 Two Output - Change in Class 2
Two Output - No-Change
Figure 4-2. Multiple-class output Artificial Neural Network topology illustrating the three
possible outcomes for change or no change
4.3 LTM Using Meso-Scale County Drivers of Urban and Agriculture
Quantities
Most of the land use change models incorporate a separate module to determine the quantity of
land use change and to locate the cells in the map that experience land use change properly. The
drivers of change for quantity and location can differ from each other (Tayyebi et al., 2012). For
example, the meso-scale subcomponents in LTM are responsible for the determination of the
quantity of change in LTM, but the locations of change still have to be determined based on the
suitability of the cells within the spatial units. The ANN component of the LTM is a location-based
driver. The suitability map produced by the model contains the probability of cells in the region
for land use change. Cells with higher probabilities are more likely to convert to land use change
than cells with lower probabilities (Pijanowski et al., 2002 and 2010; Tayyebi et al., 2011).
The amount (i.e., quantity) of each particular land use was determined using county-based
historical data on agriculture for the National Agricultural Statistics Service (NASS) Land In
Farms (LIF) database for agricultural land use (Figure 4-3) and the Year-Built (YB) statistic (Table
4-2) of the 2000 U.S. Census Bureau Housing Data (http://dataferrett.census.gov/). The YB
statistic reports the number of houses built, per county, within each 10-year census period after
-------
1940. The NLCD and the county statistics for urban and agriculture were proportionately scaled
to YB for 1992 (NLCD) and LIF for 1990, respectively, following procedures outlined by Ray and
Pijanowski, 2010, and Pijanowski et al., 2010. These standardized values were then changed over
time according to historical estimates of proportional changes in urban and agricultural land uses,
by county, between 1930 through 1990.
Table 4-2.Sample of year built statistics from the 2000 U.S. Census Bureau Housing Data
(U.S. Census Bureau 2000). Units are houses per county.
County
Summit
Franklin
Marion
Jefferson
Davidson
Allegheny
Rutherford
Cambria
Hamilton
Williamson
Wabash
State
Ohio
Ohio
Indiana
Kentucky
Tennessee
Pennsylvania
Tennessee
Pennsylvania
Ohio
Tennessee
Indiana
Built Built
1939 or 1940 to
earlier 1949
51890
62590
69454
52813
20084
188469
3024
2420
104533
2303
5218
20128
31277
28309
26344
15472
64840
1549
7787
30545
896
722
Built
1950 to
1959
41815
74719
59414
53711
34148
111591
4165
9932
64819
1632
1660
Built
1960 to
1969
31997
73952
61713
51206
42919
69263
6457
4827
5792
3724
1553
Built
1970 to
1979
31537
79490
57714
50796
50935
61424
12070
8830
46385
9008
1585
Built
1980 to
1989
19882
64208
48380
26704
46263
38700
16141
4857
29518
9785
1250
Built
1990 to
1999
27875
78070
52015
3607
35359
29664
26714
3388
27246
19244
1516
-------
Illinois
Indiana
10,800.000
10,600,000
10,400,000
10,200,000
10,000,000
9,800,0 EX)
9,600,000
9,400,000
9,200,000
9,000,000
22.000,000
18,000,000
tfc.OOO.OOO
14,000.000
i i,\ymrmi)
Kentucky
Ohio
West Virginia
y.ow.ow
B. 000.0 00
T.000,000
b,000.000
5,000,000
4.DDD.DOO
i.000,000
?,000,QQG
I.OOO.OOO
*<'^*V*__^u.
^^^^^A
Tfc
^
x
V
"*
' '
' '
1
1900
' '
'
r ,
'
, ,
' ' 1
, ,
19
" '
4.500.000
M0,000
m
10,000.000
5.000.000
3.000,000
2,000,000
\^ .AS*.
V-«r *^^
x^
^^*s^
^^•4—*
00 1920 1940 1960 1980 2000
North Carolina
^^~
^**"^^^,
^v
X
V .
• *
1920 1940 1960 1980 2000
Pennsylvania
•^^^
~"V-
^v'*sw^
V
"K^
^^
Virginia
«,
^ •f^*>.
*~*~^x
>k.
^^N»^
^S*"~*S*
0 1920 1940 1960 19BO 2000
Entire Study Area
**v*k^^^J^*^«*
^^-^».— ^*>v>>».
^^»
v^
x^.
b 19SO I9J5 1940 19&0 19M 1
1969 19/4 1982 198/
Figure 4-3. Land in Farms statistics summarized by state (in acres, as reported by the
USDA NASS (2002))
10
-------
4.4
Running the Backcast LTM-MC on an HPC
Scaling up a land use change simulation often requires re-engineering the model so that it may
handle larger datasets. We recently redesigned the LTM for running at continental scales with fine
(30-m) resolution using a new architecture that employs a windows-based HPC cluster computer
(Pijanowski et al., 2014). We configured the LTM-HPC as a backcast LTM for both MC with the
HPC to run the backcast LTM model at a national scale (Figure 4-4). Simulations in the forward
direction occurred at the level of place polygons. Briefly, place polygons are created in the GIS
using the Delany polygon routine with U.S. Census place locations (i.e., cities, town and villages)
as points for inputs. The Delany polygon represents the largest area of influence for any place.
Urban forecasts for these locations where made using state population forecasts from the U.S.
Census Bureau following the statistical procedure outlined in Tayyebi et al., 2012. Simulations in
the reverse direction occurred at the county level as historical USDA NASS (2002) and U.S.
Census Bureau (2000) are distributed at this scale. In some cases (e.g., Cincinnati), metropolitan
counties were merged to avoid artifacts that arise with back casting of large areas where
urban/population ratios vary considerably from rural areas to inner city locations. The LTM-HPC
was configured so that the ORB was split into these meso-scale regions using the GIS. ANN
routines were then applied and output for each meso-scale region created. The GIS was then used
to integrate these back into one ORB-wide map.
Spatial
Data
Pattern Recognition
©
0Back-
Cast LTM
-
© Multiple
Classifications
Spatial Unit
Meso-Scale
Driver
Quantitative Data
Within Meso-Scale
Back Cast Simulation
Evaluation
Running the Model
Forecastina Backward
in Time
Model
Calibration
ITDi"1
JrirL
Model
Variants
Figure 4-4. Steps in model development and underlying technologies
11
-------
4.5 Transition Rules Applied
Extensive urbanization has occurred over the past decades in the U.S. for a variety of economic,
technological, and population-growth related reasons (Pijanowski and Robinson, 2011). Thus, it
is to be expected that urban areas decrease in backcast land transformation simulation results.
Many agricultural lands have been converted into urban areas (e.g. agricultural land use loss)
because agricultural land is usually conveniently located in the periphery of urban areas and forest
areas have been converted into agricultural lands (e.g. agricultural land use gain) to meet the
demand for agricultural goods. Thus, for a given area where agriculture can be gained or lost, there
are two pathways for land use to transition in a backward manner (Figure 4-4):
(1) Agriculture gain: Agricultural gain quantity is less than urban loss quantity within the
meso-scale boundary, and urban cells at ti (e.g., 2001) are the first candidates to turn into
agriculture. The urban loss suitability map exhibits the locations of urban cells expected to
go to agriculture first, while the rest of the urban loss goes to forest. Thus, urban loss is
equal to agriculture and forest gain in this case.
However, if the total quantity of urban cells in ti cannot satisfy the quantity of agriculture
gain, the rest of the cells (e.g. forest cells first) are ranked based on the urban loss suitability
map and turn into agriculture cells until the total number of agriculture gain cells is met.
Thus, agriculture gain is equal to urban and forest loss in this case.
(2) Agriculture loss: Agriculture cells at ti are the first candidates to turn into forest, and
the agriculture gain suitability map decides the locations of those agriculture cells that
should go to forest first within the meso-scale boundary. The urban cells at ti could also
convert into forest, with the urban loss suitability map deciding the locations of the urban
cells that go to forest first. Thus, forest gain is equal to agriculture and urban loss in this
case.
These model variants are exclusive, and, thus, conflicts resulting from multiple classifications are
prevented (Tayyebi and Pijanowski, 2014). Because barren, open water, wetlands and shrubland
are very minor LULC classes in the ORB and much of it does not change (open water, barren), we
collapsed these into an umbrella "Other Class" in reporting in order to focus on the reporting of
the spatial-temporal dynamics of the major LULC classes (urban, agriculture and forest) located
in the ORB.
4.6 Training Goodness of Fit Statistics
We conducted multiple training cycles with the LTM to identify a training cycle that would
generate model results that deviated to an acceptable extent from observed values. We used the
MSB per cycle and followed these values during training. Briefly, MSB calculates the difference
between observed change (value of "1") and no change (value of "0") and simulated change (value
of "1") and simulated no change (value of "0) for the entire ORB. An MSB value of 0.0 means
that there is a perfect fit between the observed map of change and the simulated map of change
(likewise a value of 1.0 means there is not a fit whatsoever between observed and predicted maps
of change and no change). Early training produced an MSB of 0.182 but the MSB sharply fell after
several hundred cycles (Figure 4-5). We halted the training at 100,000 cycles where the MSB
reached a stable minimum of 0.172. After training the model, the entire dataset in 2001 was used
to generate urban loss and agriculture gain suitability maps (a suitability map contains
"probabilities" or likelihood of change). The urban loss suitability map shows that the cells around
the cities have higher values, and are, thus, the first cells to be converted to other LULC classes
(Figure 4-6). In contrast, the agriculture gain suitability map shows that the cells around the cities
have lower values while the cells around the agriculture or close to the forest cells in 2001 are the
first to be converted to agriculture classes from other LULC class (Figure 4-6). NLCD data
12
-------
between 2001-1992 were used to calculate and fix the amounts of urban, agricultural and forest
transitions within the meso-scale boundary (areas that define place Delany polygons).
0.184
0.182
0.18
0.178 |
0.176
0.174
0.172
0.17
0.168
0.166
Cvcle
Figure 4-5. Mean Square Error saved from training run across training cycles
13
-------
I I Counties in ORW
• High: 100000
• Low:0
0 55 1!Q 220 330 440
I I Counties in ORW
• High: 99259
I Low: 0
0 55 110 220 330 440
Figure 4-6. Suitability maps produced in the training/testing phase of the modeling
application: top - agricultural change; bottom - urban change. Suitability values range
from 0 to 100,000 (0.0 to 1.0 multiplied by 105)
14
-------
4.7 Hydrologic Sensitivity and Transition Pathway Analyses
We extracted the USGS National Hydrologic Database Plus (NHDPlus) hydrography for the entire
Ohio River Basin and used the ArcGIS buffer command (at 150-m) to delineate riparian zones for
the four study site watersheds (Sugar, Tippecanoe, Upper Wabash, Upper White watersheds).
These areas are commonly (cf Ray et al., 2012) the most sensitive areas affecting water quality in
a watershed (Allan, 2004).
Land use maps from 1930 through 1990 were used to create one raster map with each cell coded
with land use class sequences (e.g., a code of 44433322 represents three decades of forest, followed
by three decades of agriculture and then two decades of urban). We calculated the total percentage
of area falling into each possible land use transition pathway; we totaled the percentage of the
watershed and riparian zone that was in the top five most common transition pathways (this
followed previous work by Pijanowski and Robinson, 2011). Finally, we report the percentage of
the watershed or riparian zone that did not change between 1930 through 1990 (this is termed land
use persistence).
Current work by the Purdue team has determined that land use tipping points exist that significantly
negatively impact watersheds or riparian zones, and that these land use category intensities should
not be exceeded if stream macroinvertebrate community structure is to remain healthy. These
tipping points are for watersheds >10% urban or >38% agricultural use (Pijanowski in
preparation).
15
-------
5 MODEL CALIBRATION AND VALIDATION
Model evaluation is needed to ensure that underlying patterns apply to new data (Pontius et al.,
2004; Tayyebi et al., 2012) or that the model can be used for past or future predictions (Ray and
Pijanowski, 2010). The model generated from the training run was applied to the entire dataset in
ti to simulate LUC in t2 using observed (NLCD) data between two times (1992 and 2001) as a
comparison. We then merged simulated LUC maps (e.g. from reference and non-reference data)
in t2 to the observed map in t2to create a map of correctly and incorrectly predicted locations.
Following standard land change modeling practices, we calculated location and quantity errors
from this error map (Pontius et al., 2004). Location errors exist when the model does not predict
the correct cell to transition; two types of related location errors exist, omission (did not predict it
to change) and co-mission (predicted it to change but in reality it did not change). We followed
Pijanowski et al., (2002, 2005, 2006 and 2014) and matched omission/co-mission error pairs at
100 x 100 window sizes (3 km x 3 km) and then reported average values at 4000 x 4000 pixel
window sizes (which we call a simulation tile; dimensions are 30 m/pixel x 4000 pixel length =
120 km x 120 km). The correct location prediction rates for each simulation tile are mapped in
Figure 5-1. In general, the average goodness of fit for urban change ranges from 0.80 to 0.90 (80-
90% accurate at 3 km).
Location Error by Simulation Tile
Simulation Tile Location Error
H OtXttOM . 0.029MH
HOOSMOt-O.iOOSM
0 100001 - 0 MOW9
H 0350001 - 0 KXHWB
^H <9 MOfiOi • 0.80DMO
0 1M J»
m
t.»3
• • :*-
Figure 5-1. Goodness-of-fit statistics of location-change results of the model when applied
to the entire 48 states. As measure for goodness of fit the correctness, on a scale of 0 to 1, of
3x3 km simulation tiles relative to observed values was used
16
-------
Quantity errors exist when the model either under-predicts or over-predicts the amount for LULC
change. To test the quantity error of the backcast model and determine which counties our model
under-estimates and/or over-estimates for three scenarios (versions 1, 2 or 3 in Figure 5-2) with
using non-reference data, we followed the steps listed below. We first compared the classified
NLCD (e.g. with the four land use classes: urban, agriculture, forest and other) between ti and \.i
using a contingency table to generate the NLCD change map. We also compared NLCD at \.i with
the simulated map at ti using a contingency table to generate a NLCD simulated change map. We
then used the tabulate function in ArcGISlO to summarize the NLCD change map and NLCD
simulated change map for each county in a separate table. Comparing the corresponding tables for
the NLCD change map and the NLCD simulated change map enabled us to find where our model
under-estimated and over-estimated each scenario within each county. After model evaluation, the
model could be used for simulating past scenarios.
The quantity errors for urban change for the national simulation (see Pijanowski et al., 2014 for
details) are reported and visualized in Figure 5-2, and illustrate that quantity errors for the ORB
are some of the smallest in the lower 48 states. Past simulations suggest values that are less than
0.5 are satisfactory at this scale of simulation (Pijanowski et al., 2005), this is particularly
important for areas where there is a lot of urban change.
Quantity Error by Simulation Tile
SimulationTileQuantity Error
[ 1 0250001- 1000090
I 1 1 OOGQOt . 2 OOOOW
j UOOMQI-SMOOOO
• 5 000001 - 10 000000
D l!0 300 KD 900 1.200
KI^K^^^^^K^=^^~^^^^^ '' •' ' •
Figure 5-2. Goodness of fit statistics of quantity-change results of the model when applied
to the entire lower 48 states. As measure for goodness of fit the correctness, on a scale of 0
to 1, of 3 x 3 km simulation tiles relative to observed values was used
17
-------
6 SIMULATION RESULT
6.1 Backcast Results
Maps of the historical changes in land use by decade (relative to the calibration period of 1990-
2001) were created by the LTM, from 1930 to 1980 (Figure 6-1). Three general trends emerge
from these simulations: an increase in urban, an increase in forests and a decrease in agriculture.
The increase in forest occurs in the southeastern portion of the study area in large, homogenous
patches. Forests tend to increase in smaller amounts and in a more fragmented pattern in the
northeastern and central regions of the study area. Urban growth is prominent throughout the
region with obvious increases in the major metropolitan areas of the ORB. Urban use in 1930 was
estimated to be 6.9% (a major part of this was in roads) growing to 8.9% by 2010 (36% increase
in the urban use footprint). Nearly one third (33.5%) of the agriculture from 1930 was lost by
2010 (agriculture went from 55.6% in 1930 to 37.0% in 2010). Forest cover gained between these
time periods, representing 35.1% of the land cover in 1930 and 48.8% in 2010. Agriculture was
the dominant land use/cover in 1930 and through gradual loss, forest became the dominant LULC
in the mid-1960s and thereafter.
The GIS was also used to create a time series map of land uses and estimate the proportion of the
map involved in each land use legacy pathway. A land use legacy pathway is a sequence of land
uses in set decadal time steps. For example, one land use legacy pathway is a location staying
forest for 10 years, then converting to agriculture for 20 years and then finally converting to urban
and remaining urban for 30 years. Theoretically, 47 (i.e., 16,384) land use transition pathways are
possible (in all likelihood, fewer than 16,384 of urban is an 'end land use'). We found (Table 6-1)
that 36.7% of the map (ORB) remained agriculture for 60 years (1930 through 1990), 33.5%
remained forest and 6.9% remained urban. As for urban, we classified all roads as urban, and,
since most of the secondary roads in the basin were developed around the early 1900s, a majority
of the urban footprint for 1930 are considered roads. About 2.6% of the map remained other (open
water, barren, shrubland). The most common transition land use legacy pathways was the
conversion of agriculture to forest (16.4% of the transition pathways in the basin), followed by the
conversion of agriculture to urban (1.11% of the transition pathways basin; Table 6-1).
18
-------
1980
N
W
Legend
State Boundaries
ORB Boundary
Other
Urban
Agriculture
Forest
0 100200 400 600 800
Figure 6-1.The most common BLTM land use transition pathways occurring between 1930
and 1990 for the ORB
19
-------
Table 6-1. The most common BLTM land use transition pathways occurring
Percentage of Pathways
in Specific Transition
Pathway
36.7703
33.5545
6.9823
5.2397
4.3761
2.4097
2.2673
2.1635
1.4411
0.6978
0.6953
0.3773
0.3397
0.2711
0.2405
0.2387
0.2236
Status 1
Stay Ag for 60
Years
Stay Forest for 60
Years
Stay Urban for 60
Years
Ag for 30 Years
Ag for 40 Years
Ag for 60 Years
Ag for 50 Years
Stay Other
Classes for 60
Years
Ag for 20 Years
Forest for 10
Years
Ag for 10 Years
Ag for 10 Years
Forest for 10
Years
Ag for 50 Years
Ag for 30 Years
Forest for 20
Years
Ag for 40 Years
Change To
— -
— -
Forest
Forest
Forest
Forest
Forest
Ag
Forest
Urban
Ag
Urban
Urban
Ag
Urban
Status 2 Change To Status 3
— -
— -
Forest for
30 Years
Forest for
20 Years
Forest in
1990
Forest for
10 Years
Forest for
40 Years
Ag for 10 wt Forest for
Years 40 Years
Forest for
50 Years
Urban for
50 Years
Agfor20 Forest for
Years 30 Years
Urban for
10 Years
Urban for
30 Years
Ag for 10 _ Forest for
° Forest ,„ ,,
Years 30 Years
Urban for
20 Years
20
-------
Approximately 90% of the area in our four demonstration watersheds (Figure 6-2) remained in a
single land use over the 1930 to 1990 simulation period. The Upper White watershed had less area
(83.4%) that persisted in a single land use during the 60-year period; this also means that over 16%
of the watershed has experienced at least one land use change. Of the four watersheds, it is the
only one to contain a large city (Indianapolis). We also examined land use persistence in riparian
zones and found that these areas have undergone more change than the watersheds as a whole,
suggesting that these are more dynamic locations. Of the four demonstration watersheds, the
Upper White had a quarter of its riparian zone transformed during the 1930 to 1990 period. The
Sugar Watershed had the greatest amount of forest cover (-12%) that persisted over the backcast
simulation period.
100
90
80
CD
O
CD
0_
Q) 50
30
CD O
0)N 60
!•! 5°
O TO
CD .9- 40
o_ or
30
92.1%
90.7%
89.9%
83.4%
I forest
agriculture
I urban
I other
Sugar
Tippecanoe
Upper Wabash
Upper White
I forest
agriculture
I urban
I other
Sugar
Tippecanoe Upper Wabash Upper White
Figure 6-2. Percentage of each case study watershed (top) and within permanent
stream riparian zones (bottom) that persisted in a land use class from 1930 to 1990.
Percentage value over each aggregated bar indicates the total area that persisted
in a single land
21
-------
The land use transition pathways for all four of our demonstration watersheds are relatively
similar. The top 5 most frequent long-term pathways are shown in Table 6-2 for entire
watersheds (A) and for riparian zones (B). Between 2 to 6 percent of these watersheds were in
agricultural land use for 3 to 6 time steps. In three of the four watersheds, agriculture for one
time period (1930) followed by urban (1940 and thereafter) was among the top 5 transition
pathways. When examined within riparian zones, compared to values for entire watersheds,
there were larger proportions of land use in agriculture followed by forest for each of the four
demonstration watersheds. For example, the Sugar Watershed had 7.58% of the riparian zone in
agriculture for 5 time steps and forest for 2; only 2.71% of the entire watershed exhibited this
exact transition pathway. Only the Upper White Watershed had a transition pathway for its
riparian zone that included urban.
Table 6-2. Transition pathways for the selected demonstration watersheds (A) and their
riparian zones (B). Only the top 5 most common transition pathways are listed here and
their percent area for the watershed and riparian zone. Totals are percentage total area.
A. Top Five Transition Pathways for All Areas within Demonstration Watersheds
Sugar
AAAAAFF
AAAAAAF
FAFFFFFF
FAAFAFA
AUUUUUU
Percent
2.71
1.13
0.82
0.46
0.25
Tippecanoe
AAAAAAF
AAAAAFF
FAAAAFF
AAAAFFF
FFAFFFF
Percent
2.07
1.57
0.78
0.67
0.47
Upper
Wabash
AAAAAAF
AAAAAFF
AAAAFFF
AUUUUUU
FFFFFAFFF
Percent
4.24
1.94
0.99
0.59
0.34
Upper
White
AAAAAFF
AAAAAAF
AAAFFFF
AAAAFFF
AUUUUUU
Percent
2.83
2.48
1.88
1.78
1.35
total for top 5 5.36
5.56
8.11
10.31
B. Top Five Transition Pathways within Riparian Zones of the Demonstration Watersheds
Sugar
AAAAAFF
AAAAAAF
FAFFFFF
FAAFAFF
AAAFAFF
Percent
7.58
2.88
1.83
1.05
0.56
Tippecanoe
AAAAAAF
AAAAAFF
FAAAAFF
AAAAFFF
FFAFFFF
Percent
2.60
2.01
1.26
0.84
0.59
Upper
Wabash
AAAAAAF
AAAAAFF
AAAAFFF
FAAAFFF
AAAFFFF
Percent
7.19
2.83
1.45
0.48
0.27
Upper
White
AAAAAFF
AAAAAAF
AAAAFFF
AAAFFFF
AUUUUU
Percent
6.68
6.45
2.99
2.32
1.24
total for top 5 6.32
7.31
12.23
19.69
22
-------
The spatial distribution of persistence and land use change in the ORB between 1930 and 1990
varies considerably spatially. Many areas, such as those in the northeastern portion of the ORB
(Figure 6-3, area labeled A), have a very scattered distribution of persistence. The northwestern
portion of the watershed (area labeled B) has few locations of change; much of the land use that
was in place in 1930 persists today. Areas along the Ohio River proper, especially north of the
river, have clumped areas of change and persistence (labeled C). Finally, one area in southern
West Virginia (labeled D) experienced large homogeneous amounts of change. Inspection of the
time series maps from Figure 6-1 indicates that the latter area used to be largely agricultural in the
1930s and 1940s but it transit!oned to forest in later years, possibly as a result of farm failures
during the Great Depression.
Legend
| 1 agriculture
| land use legacy
Counties in the ORB
ORB boundary
States
0 40 80 160 240 320
I Kilometers
Figure 6-3. Map of land use persistence from 1930 through 1990 and locations of
change (i.e., locations where land use legacies may impact ecosystem dynamics). See
Section 6.1 for explanations of areas indicated by letters A-D. ArcGIS file name is
legacy_all
23
-------
6.2
Forecast Results
The same training approach was used to forecast land use change into the future at 10-year time
steps (2010-2050). The decadal maps of land use change are presented in Figure 6-4, and show
that metropolitan areas will continue to expand at historical rates (Pijanowski and Robinson, 2011,
Pijanowski and Plourde, unpublished) and these are reflected in our estimates. We predict that the
entire study area will reach 11.83% urban in 2050 with agriculture decreasing to 35.8% and forest
decreasing to 47.6%. Figure 6-5 shows a complete trend for all major land use classes from 1930
through 2050. Note that in 1930, a majority of the ORB was in agriculture but over time there was
a steady decline in agriculture with an increase in forest cover. Urban land use increases gradually
over time and by 2050 we estimate that urban should be almost half the footprint of agriculture in
2050.
2001
2010
Legend
State Boundaries
ORB Boundary
Other
Urban
Agriculture
Forest
0 95190 380 570
Kilometers
Figure 6-4. Land Transformation Model projections summarized by decade
24
-------
In 2010, a majority of the areas converted to urban are from forest (51.7%), followed closely by
agriculture (41.2%). Only 7% of the "other" LULC class was converted to urban in 2010. These
rankings stay the same but the trends differ slightly, as less forest and more agriculture is converted
to urban with each successive decade (Figure 6-6). By 2050, 48.4% of the new urban in that
decade is from forest and 45.5% is from agriculture.
100%
I Forest
Agriculture
I Urban
I Other
1930 1940 1950 1960 1970 1980 1990 2001 2010 2020 2030 2040 2050
Figure 6-5. Percentage of land use/cover classes simulated over time for the ORB from 1930
through 2050 with 10-year time steps
i Other
Agriculture
i Forest
2010
2020
2030
2040
2050
Figure 6-6. Percentage of land use/cover classes that are predicted to be converted into
urban land use during each decade
25
-------
2010
2020
North Carolina
SftaCT Uaro*Kaa—-v
North Carolina
Scurni Car3Kna -*y
2030
2040
North Carolina
"irniTtl 1'irBhm ",
2050
New York
North Carolina
SjuiWCarollna—s.
''''•'
Legend
^^ Canada
C^J5 state Boundaries
8-Digit Hydrologic Units
Percent of Watershed
in Urban
0.00-10.00
10.01-100.00
0 85 170 340 510 680
I Kilometers
Figure 6-7. Locations where potential urban-water quality thresholds have been met at the
scale of an 8-digit hydrologic unit (i.e., watershed) for years 2010 through 2050. Watersheds
with a red color have exceeded 10% urban land use by the year indicated
26
-------
Almost one third of the ORB watersheds exceeded 10% urban in 2010 (Figure 6-7). Much of
these watersheds (we call these threshold watersheds) in 2010 are located in the northern portion
of the ORB and along the Pennsylvania-West Virginia border. By 2050, the distribution of these
"threshold" watersheds have spread further east, south and north. By 2050, over half (64/120) of
the 8-digit hydrologic units have more than 10% urban (Figure 6-8).
2010
2020
2030
2040
2050
Figure 6-8. Number of 8-digit hydrologic units (watersheds) that have exceeded 10% urban
land use by the year indicated. There are 120 8-digit hydrologic units within the ORB
Most of the watersheds in Indiana, Illinois and Ohio have also exceeded 38% agriculture (Figure
6-9) by 2010. Between 2010 and 2040, fifty 8-digit hydrologic units have more than 38%
agriculture; by 2050 one of these is predicted to have enough agriculture transitioned to
urbanization that it is no longer a member of this threshold condition.
27
-------
2010
2020
North Carolina
Spafll Carolina y
2030
2040
Wisconsin .
| Michigan
"A
North Carolina
Sotrm t,aroriflii —N.
2050
North Carolina
Sjitrm larATini "^
N
\V
Legend
^^^ Canada
State Boundaries
8-Digit Hydrologic Units
Percent of Watershed
in Agriculture
000-38.00
38.01 - 100.00
0 85 170 340 510 680
i Kilometers
Figure 6-9. Locations where potential agriculture-water quality thresholds have been met
at the scale of an 8-digit hydrologic unit (watershed) for the years 2010 through 2050.
Watersheds with a red color have exceeded 38% agricultural land use by the year indicated
28
-------
6.3
Model Output
We are distributing five sets of files that are output from the simulations. The first set contains
seven backcast results at 10-year time steps from 1930 through 1990. Maps are in ArcGIS 10
raster file format in 30-m resolution. Land use codes are urban, agriculture, forest and an "other"
class that collapses four minor LULC classes. The second set contains LULC maps for the
forecasts from 2010 through 2050, also at 10-year time steps. One version contains urban
forecasts coded as 1XX where XX is the original Anderson level 2 LULC code from the 2001
NLCD. Locations from 2001 that are not predicted to change contain the original LULC codes.
We have also collapsed the LULC classes to the major four classes and will distribute these as
well as they corresponded to the same LULC class codes as the backcast maps. The fourth
database distributed is a raster legacy map that contains a sequence of codes for LULC for years
in the sequence 1930, 1940, 1950, 1950, 1970, 1980 and 1990. We coded "other" = '0',
urban=T, agriculture='2' and forest ='3'; avalueinthis database that is 2223331 is alocation
that has had a sequence of agriculture for three decades (1930-1950), forest for three more
decades (1960-1980) and urban for the final decade (1990). The fifth database we are
distributing is a shape file with the summary of the percentage of urban and agricultural land use
by 8-digit hydrologic unit. We used the tabulate area command in ArcGIS 10 Spatial Analyst to
summarize the area for each land use and Excel to calculate the percentage areas for urban and
agriculture. All raster files contain grids of 31,644 columns by 31,191 rows (slightly more than
1.0 x 108 cells) and are being distributed in the simulated projection of Albers Equal Area (the
North American standard datum).
Table 6-3. List of model simulation outputs distributed to EPA by Purdue University
Output
Backcast
maps
1930-1990
Forecast
maps
2010-2050
Forecast
maps
2010-
2050,
collapsed
classes
Legacy
map
Naming
Orb xxxx vl 1
where xxxx is
year
Orb xxxx urb
where xxxx is
year
Orb xxxx
where xxxx is
year
Legacy orb
Description
Maps of LULC for urban, agriculture,
forest and other category
Maps of land use/cover for native
NLCD 2001 LULC classes. Urban
forecasts are included in maps as 100
+ original code (e.g., 182 is future
urban from agriculture class of 82)
Maps of LULC for the four major
LULC classes of urban, agriculture,
forest and other.
Map of LULC change sequences for
each location. Code contains the land
use for year sequences as seven
sequential digits, the first is the LULC
code for 1930 and the last is the
LULC code for 1990
Data Format
Seven ArcGIS
10.0 raster maps
at 30 x 30m
resolution (31,644
columns by 31,191
rows)
Five ArcGIS 10.0
raster maps at 30 x
30m resolution
(31,644 columns
by 3 1,191 rows)
Five ArcGIS 10.0
raster maps at 30 x
30m resolution
(31,644 columns
by 3 1,191 rows)
One ArcGIS 10.0
raster data at 30 x
30m resolution
(31,644 columns
by 3 1,191 rows)
29
-------
Land
Use/Cover
Percent by
8-Digit
Hydrologic
Unit
HUC_ORB_future
urb_ag_percents_
Final, shp
Percentage of each land use class
summarized by 8-digit hydrologic unit
for years 20 10-2050
One shape file
with 8-digit
hydrologic units,
codes and percent
land use/cover
classes as attribute
table
There are 120 8-digit hydrologic units in the ORB and the attribute table for the shape files
contains summary information for each hydrologic unit. Table 6-3 summarizes these simulation
output files.
30
-------
7 DISCUSSION
The Ohio River Basin (ORB) has undergone tremendous changes in land use from 1930 to current,
and, if contemporary trends continue, it is likely that the entire ORB will reach 10% urban by
2050. Historical patterns are not unlike other areas of the Midwest (see Pijanowski et al., 2007;
and Pijanowski and Robinson, 2011). In 1930, the dominant land use was agriculture, but
afforestation patterns between 1950 to current have "greened" portions of the ORB yielding forest
as the dominant land use/land cover (LULC) class today. The return of some of the landscapes to
forests has been well characterized by Brown et al. (2005), who showed that many landscapes east
of the Mississippi River from 1970 to present have increased forest cover. However, these new
forested landscapes exist today with land use legacies that include past agriculture. Areas that
have had agriculture in the past are known to retain the biogeochemical signature of the inputs
from cropping, most notably phosphorus and herbicides (Dupouev et al., 2002; Flinn and Vellend,
2005; Standish et al. 2006; Baeten et al., 2010; Christiansen et al., 2010; Brudvig et al., 2013),
which can be retained in the soil for long periods of time. Past agricultural use influences plant
community structure through processes such as germination (Flinn and Vellend, 2005; Hermy and
Verheven, 2007; Feurdean et al., 2009) or facilitates the colonization and spread of invasive
species (Vila and Ibanez 2011). Water quality of streams has been shown to harbor the "ghost" of
previous land uses (Harding et al. 1998) through biogeochemical signatures that remain from
historical land use management practices, such as fertilizer application.
Our model shows that the spatial patterns of land use persistence vary considerably across the
ORB. A highly fragmented pattern of land use persistence exists in the northeastern portion of the
ORB, a large homogenous distribution of agriculture to the north and west and one "patch" of
almost exclusive agriculture conversion to forest exists in southern West Virginia. Areas just to
the north of the Ohio River proper contain large, but dispersed, areas which have had a history of
change in land use. We also observed that many metropolitan areas have undergone more land
use change and, thus, have more frequent occurrences of land use legacies than rural areas. The
most common land use transition pathway that occurred throughout the ORB was for land to
remain agriculture for 30 years, then to transition to forest, and stay forest at the current time step.
The next most common land use transition pathways were (1) agriculture for 40 years and then
forest, (2) agriculture for 60 years and then forest, and (3) agriculture for 50 years and then forest.
Thus, agricultural abandonment leading to forest represents the most common land use pathway
even when urbanization is accounted for. Three-phase land use transition pathways, such as
agriculture to forest and then to urban, accounted for some pathways but these were only 1/1 Oth
as frequent as the two-phase agriculture to forest pathway. Thus, land use legacies have
tremendous implications for forest management in the ORB and, based on recent literature on the
topic (Vila and Ibanez 2011), for invasive plant species management.
Analysis of four demonstration watersheds was conducted so that we might examine trends that
reflect different human histories. The Upper White watershed contains much of the Indianapolis,
Indiana metropolitan area representing a highly urbanized watershed. The Sugar Watershed is a
rural watershed and contains a lot of forested riparian zones. The Tippecanoe Watershed is
historically a rural, agricultural watershed. Finally, the Upper Wabash watershed is an agricultural
watershed that is currently undergoing transitions to large-scale livestock production. We found
three of the four watersheds (the Upper White with some differences) to have very similar land
use persistence patterns historically. Nearly 90% of these watersheds stayed in the same land use
from 1930 to 1990. The most common transitions were agriculture to forest. The Upper White
watershed had less land use persistence and the most common land use transition pathway involved
urban. Interestingly, land use persistence was less in riparian zones for all four watersheds as
compared to the watersheds as a whole. As the most common land use transition pathways
involved the conversion of agriculture to forest, conservation efforts by groups and land owners
may have created situations where stream health was a concern and planting of trees or agriculture
abandonment may have resulted in more forested landscapes along the riparian zones. An analysis
31
-------
of historical photographs along different portions of these rivers could verify the model outcome.
Restoration efforts along streams and rivers should examine historical patterns of land use
transition pathways as the model suggests that they are more common there than in the upland
portions of the watershed.
Calibration of the forecast and backcast models suggest the model performed satisfactorily in the
forward and backward directions at about 3-km spatial resolution, making the output suitable for
larger scale coupled simulations to hydrologic and/or climate at this resolution or larger.
Describing general trends at coarser resolutions is possible as we have done.
The future of the size of the agricultural footprint for the ORB remains uncertain. Two factors
may lead the agricultural footprint to remain the same despite the need for a large area of cropland
in the U.S. to be used for biofuels. First, as Plourde et al. (2013) recently found, much of the
ethanol production occurs west of the ORB. Cropland there is also being changed to monoculture
inter-annual planting of corn (i.e., corn-soybean rotation is not occurring) with a majority of the
corn-soybean rotation in the U.S. occurring in the ORB. It is uncertain as to whether monocultural
corn cropping would move east and into the ORB as a result of increased ethanol demand. The
second factor contributing to an unknown future of the agricultural footprint for ORB is the
economics of energy; the relatively low price of fossil fuel relative to the cost of ethanol production
makes ethanol less profitable. However, if fossil fuel prices rise dramatically then pressure to
grow more corn could result and corn-soybean rotation in the ORB may be dropped in favor of
monocultural corn. Such corn-corn crop rotation practices are known to increase the need for
fertilizer, thus, potentially reducing water quality of streams and rivers (Plourde et al. 2013).
Current work in the Purdue lab has focused on how to incorporate crop rotation patterns into the
LTM so that crop-type patterns can be accounted for in land-hydrologic simulations.
The ORB watersheds currently contain levels of land use, particularly the percentages of urban
and agriculture, which are known to threaten water quality and stream macroinvertebate
community structure in the Great Lakes watersheds (Pijanowski in prep.). Our analysis suggests
that a majority of the 8-digit hydrologic units to the north have exceeded 10% urban already.
Likely, nearly all of the 8-digit hydrologic units in the north and some in the central (e.g.,
Kentucky) watersheds have exceeded 38% agriculture.
The modeling approach here differs from similar efforts to simulate future and historical land
use/cover at large scales. A notably similar effort has recently been undertaken by Sohl and
colleagues (Sohl et al. 2012a, 2012b) with a model called FORE-SEC. Both models use a
"demand" and "allocate" structure where the demands for each use are constrained by scenario
and allocation is driven by spatial pattern characterization. Both approaches also use historical
NLCD data to parameterize and calibrate the models.
Very recently, the 2011 NLCD maps have become available and having a fourth time step to
determine how well these models perform will help support their use and further development.
Using two maps from different time steps to parameterize a model and then another two time steps
is likely to lead to more insight on how we could improve models that simulate these very complex
phenomena.
In summary, the ORB has been a dynamic basin historically. An agriculturally dominant land
cover has given rise to a forested landscape which in turn may become mostly urban in the future.
Land use legacy patterns in the ORB are complex and are likely to be key factors in consideration
of forest and water quality management.
Disclaimer
32
-------
The views expressed in this article are those of the authors and do not necessarily reflect the views
or policies of the U.S. Environmental Protection Agency.
8 REFERENCES
Allan, J.D. 2004. "Landscapes and riverscapes: the influence of land use on stream
ecosystems," Annual Review of Ecology, Evolution, and Systematics, 257-284. Not cited
in text.
Baeten, L.. M. Vanhellemont. P.D. Frenne. M. Hermv. and K. Verheven. 2010. "The phosphorus
leaacv of former agricultural land use can affect the production of germinable seeds in
forest herbs," Ecoscience, 17(4), 365-371.
Brown, D.G.. K.M. Johnson. T.R. Loveland. and D.M. Theobald. 2005. "Rural land-use trends in
the conterminous United States, 1950-2000," Ecological Applications, 75(6), 1851-1863.
Brudvig. L.A.. and E.I. Damschen. 2011. "Land-use history, historical connectivity, and land
manaaement interact to determine lonaleaf pine woodland understory richness and
composition," Ecography, 34(2), 257-266.
Brudvig. L.A.. E. Grman. C.W. Habeck. J.L. Orrock. and J.A. Ledvina. 2013. "Strong leaacv of
agricultural land use on soils and understorv plant communities in longleaf pine
woodlands,' Forest Ecology and Management, 310, 944-955.
Christiansen. J.R.. L. Vesterdal. I. Callesen. B. Elberlina. I.K. Schmidt, and P. Gundersen. 2010.
"Role of six European tree species and land-use legacy for nitrogen and water budgets in
forests," Global Change Biology, 16(8), 2224-2240.
Dupouev. J.L.. E. Dambrine. J.D. Laffite. and C. Moares. 2002. "Irreversible impact of past land
use on forest soils and biodiversity," Ecology, 83(11), 2978-2984.
Feurdean. A.N.. K.J. Willis, and C. Astalos. 2009. "Leaacv of the past land-use changes and
manaaement on the 'natural' upland forest composition in the Apuseni Natural Park,
Romania," The Holocene, 19(6), 967-981.
Flinn, K.M.. and M. Vellend. 2005. "Recovery of forest plant communities in post-aaricultural
landscapes," Frontiers in Ecology and the Environment, 3(5), 243-250. Not cited in text
Foster, D.. F. Swanson. J. Aber. I. Burke. N. Brokaw. D. Tilman. and A. Knapp. 2003. "The
importance of land-use legacies to ecology and conservation," BioScience, 53(1), 77-88.
Harding. J.S.. E.F. Benfield. P.V. Bolstad. G.S. Helfman. and E.B.D. Jones. 1998. "Stream
biodiversity: The ahost of land use past," Proceedings of the National Academy of
Sciences, 95(25), 14843-14847.
Hermy, M.. and K. Verheven. 2007. "Leaacies of the past in the present-day forest biodiversity:
A review of past land-use effects on forest plant species composition and diversity."
In: Sustainability and Diversity of Forest Ecosystems (pp. 361-371). Springer Japan. Not
cited in text.
33
-------
Pijanowski. B.C.. and K.D. Robinson. 2011. "Rates and patterns of land use change in the Upper
Great Lakes States. USA: A framework for spatial temporal analysis," Landscape and
Urban Planning, 102(2), 102-116.
Pijanowski. B.C.. K.T. Alexandridis. and D. Mueller. 2006. "Modelling urbanization patterns in
two diverse regions of the world," Journal of Land Use Science., 1(2-4), 83-108.
Pijanowski, B.C., O.K. Ray, A.D. Kendall, J.M. Duckies, and D.W. Hyndman. 2007. "Using
backcast land-use change and groundwater travel-time models to generate land-use legacy
maps for watershed management," Ecology and Society, 12, 25;
http: //www. ecol ogy andsoci ety. org/vol 12/i s s2/art2 5
Pijanowski. B.C.. S. Pithadia. B.A. Shellito. and K. Alexandridis. 2005. "Calibrating a neural
network-based urban change model for two metropolitan areas of the Upper Midwest of
the United States," InternationalJournal of Geographical Information Science, 19(2), 197-
215.
Pijanowski. B.C.. A. Tavvebi. M.R. Delavar. and MJ. Yazdanpanah. 2010. "Urban expansion
simulation using geospatial information system and artificial neural
networks," InternationalJournal of Environmental Research, 3(4), 493-502.
Pijanowski. B.C.. A. Tavvebi. J. Doucette. B.K. Pekin. D. Braun. and J. Plourde. 2014. "A big data
urban growth simulation at a national scale: Configuring the GIS and neural network based
Land Transformation Model to run in a High Performance Computing (HPC)
environment," Environmental Modelling and Software, 51, 250-268.
Pijanowski, B.C., D.G. Brown, G. Manik, and B. Shellito. 2002. "Using neural nets and GIS to
forecast land use changes: a land transformation model," Computers, Environment and
Urban Systems, 26, 553-575.
Plourde. J.D.. B.C. Piianowski. and B.K. Pekin. 2013. "Evidence for increased monoculture
cropping in the Central United States," Agriculture, Ecosystems and Environment, 165, 50-
59.
Pontius Jr. R.G.. D. Huffaker. and K. Denman. 2004. "Useful techniaues of validation for spatially
explicit land-change models," Ecological Modelling, 179(4), 445-461.
Ray, D.K.. and B.C. Piianowski. 2010. "A backcast land use change model to generate oast land
use maps: Application and validation at the Muskegon River watershed of Michigan,
USA," Journal of Land Use Science, 5(1), 1-29.
Ray, D.K.. J.M. Duckies, and B.C. Piianowski. 2010. "The impact of future land use scenarios on
runoff volumes in the Muskegon River Watershed," Environmental management, 46(3),
351-366.
Ray, D. K.. B.C. Piianowski. A.D. Kendall, and D.W. Hvndman. 2012. "Coupling land use and
groundwater models to map land use legacies: Assessment of model uncertainties relevant
to land use planning," Applied Geography, 34, 356-370.
Sohl, T.L.. B.M. Sleeter. K.L. Savler. M.A. Bouchard. R.R. Reker. S.L. Bennett. R.R. Sleeter. R.L.
Kanegister. and Z. Zhu. 2012a. "Spatiallv explicit land-use and land-cover scenarios for
the Great Plains of the United States," Agriculture, Ecosystems and the Environment, 153,
1-15.
34
-------
Sohl, T.L.. B.M. Sleeter. Z. Zhu. K.L. Savler. S. Bennett. M.A. Bouchard. R.R. Reker. T.
Hawbaker. A. Wein, S. Liu, R. Kanengieter, and W. Acevedo. 2012b. Applied Geography,
34, 111-124.
Standish. R.J.. V.A. Cramer. RJ. Hobbs. and H.T. Kobrvn. 2006. "Leaacv of land-use evident in
soils of Western Australia's wheatbelt," Plant and Soil, 280 (1-2), 189-207.
Stark, J. 2011. "Ohio River Basin Fish Habitat Partnership Strategic Plan. Version 6," April 14
2011. Produced by The Nature Conservancy in Ohio. 64 pp.
Tayyebi. A., and B.C. Piianowski. 2014. "Modeling multiple land use changes using ANN. CART
and MARS: Comparing tradeoffs in goodness of fit and explanatory power of data mining
tools," InternationalJournal of Applied Earth Observation and Geoinformation, 28, 102-
116.
Tayyebi. A.. B.K. Pekin. B.C. Piianowski. J.D. Plourde. J.S. Doucette. and D. Braun. 2012.
"Hierarchical modeling of urban growth across the conterminous USA: developing meso-
scale quantity drivers for the Land Transformation Model," Journal of Land Use Science,
1-21.
Tayyebi. A.. B.C. Piianowski. and A.H. Tavvebi. 2011. "An urban growth boundary model using
neural networks. GIS and radial parameterization: An application to Tehran,
Iran," Landscape and Urban Planning, 100(1), 35-44.
U.S. Census Bureau. 2000. "Housing data, (http://www.census.gov/main/www/cen2000.html),
last accessed May 30, 2013.
U.S.D.A. NASS. 2002. "Land-in-Farms Data from the USDA NASS,"
(http://www.nass.usda.gov/Charts_and_Maps/Farms_and_Land_in_F arms/index.asp),
last accessed May 30, 2013.
Vila, M.. and I. Ibafiez. 2011. "Plant invasions in the landscape," Landscape Ecology, 26(4), 461-
472.
Wallin, D.O.. FJ. Swanson. and B. Marks. 1994. "Landscape pattern response to changes in
pattern generation rules: land-use legacies in forestry," Ecological Applications, 569-580.
35
-------
9 APPENDIX
9.1 Metadata for the Backcast LTM output
Metadata:
• Identification_Information
• Data Quality Information
• Spatial_Data_Organization_Information
• Spatial Reference Information
• Entity_and_Attribute_Information
• Di stributi on Informati on
• Metadata Reference Information
Identification Information:
Citation:
Citation Information:
Originator:
Human-Environment Modeling and Analysis (HEMA) Laboratory, Department of
Forestry and Natural Resource, Purdue University
Publication Date: December 31, 2013
Title: Historic Land Use for the Ohio River Basin 1930 - 1990
GeospatialData Presentation Form: ESRI GRID
Description:
Abstract:
Past land cover predictions were created for 1990 to 1930 in 10-year increments for the
Ohio River Basin. The 2001 National Land Cover Dataset version 2 served as the basis
for all predictions. Change in land use between 2001 and 1992 along with topography,
infrastructure accessibility, proximity to water, and land use density were used to determine
the probability of change for a given area. Rates of agriculture change were based on "land
in farm" from the U.S. Census of Agriculture, and "year built" from the U.S. Census for
urban areas.
Purpose:
Past land cover is meant to serve as an example of one possible scenario of past conditions.
Supplemental Information:
This metadata applies to all data from 1930 to 1990. The backcast year can be determined
based on the file name (e.g. ORB_1930 is the Ohio River Basin data for 1930).
Time Period of Content:
Time Period Information:
Range of Dates/Times:
Beginning Date: 1930
Ending Date: 1990
Currentness Reference: ground condition
Status:
Progress: Complete
36
-------
Maintenance and Update Frequency: As needed
Spatial Domain:
Bounding Coordinates:
Top: 2310075
Left: 566325
Right: 1515645
Bottom: 1374345
Keywords:
Theme:
Theme Keyword Thesaurus: None
Theme Keyword: Land Cover
Theme Keyword: Land Use
Theme Keyword: Historic
Place:
Place Keyword Thesaurus: None
Place Keyword: Ohio River Basin
Access Constraints: None
Use Constraints: None
Point of Contact:
Contact Information:
Contact Person Primary:
Contact Person: Dr. Bryan Pijanowski
Contact Organization: Department of Forestry and Natural Resources, Purdue University
ContactAddress:
Address Type: mailing and physical address
Address: 195 Marsteller St.
City: West Lafayette
State or Province: IN
Postal Code: 47906
Country: USA
Contact Voice Telephone: 765-496-2215
Contact Facsimile Telephone: 765-496-2422
Contact ElectronicMa/7Address: bpij anow@purdue.edu
Data Set Credit:
Human-Environment Modeling and Analysis Laboratory, Department of Forestry and
Natural Resources, Purdue University
Native Data Set Environment:
ArcGIS Desktop 10.1
Cross Reference:
Citation Information:
Originator:
Data Quality Information:
A ttribute A ccuracy:
A ttribute A ccuracy Report:
37
-------
Base land cover classes are as accurate as the NLCD 2001 on which they are based
(http://www.epa.gov/mrlc/accuracy-2001.html). No formal accuracy assessment for
projections was completed.
Lineage:
Source Information:
Source Citation:
Citation Information:
Originator: Human-Environment Modeling and Analysis Laboratory, Department of
Forestry and Natural Resources, Purdue University
Title:
Geospatial Data Presentation Form: ESRI GRID
Type of"Source Media: Digital
Process Step:
Process Description:
The 2001 National Land Cover Dataset version 2 served as the basis for all predictions. Change in
land use between 2001 and 1992 along with topography, infrastructure accessibility, proximity to
water, and land use density were used to determine the probability of change for a given area.
Rates of agriculture change were based on "land in farm" from the U.S. Census of Agriculture,
and "year built" from the U.S. Census for urban areas.
Spatial Data Organization Information:
Direct Spatial Reference Method: Raster
Raster Object Information:
Raster Object Type: Pixel
Row Count: 31191
Column Count: 31644
Vertical Count: 1
Spatial Reference Information:
Horizontal Coordinate System Definition:
Planar:
Map Projection:
Projection: NAD 83 Albers
false easting: 0.000000
false northing: 0.000000
central meridian: -96.000000
standard_parallel 1: 29.500000
standard parallel'_2: 45.500000
latitude of origin: 23.000000
Linear Unit: Meter (1.000000)
Geographic Coordinate System: GCS North American 1983
Angular Unit: Degree (0.017453292519943295)
Prime Meridian: Greenwich (0.000000000000000000)
Datum: D North American 1983
Spheroid: GRS1980
Semimajor Axis: 6378137.000000000000000000
38
-------
Semiminor Axis: 6356752.314140356100000000
Inverse Flattening: 298.257222101000020000
Entity and Attribute Information:
Detailed Description:
Entity Type:Table
Attribute Label: Rowid
Attribute:Table Row
Attribute Label: Value
Attribute:Pixel value denoting land use class
Overview Description:
All class codes are based on NLCD 2001 V2 Level II schema
There are no class 12 - Perennial Ice/Snow pixels
0: Other Land Use (11 - Open Water, 31 - Barren Land, 90 - Woody Wetlands, and 95 -
Emergent Herbaceous Wetland)
1: Urban (21 - Developed Open Space, 22 - Developed Low Intensity, 23 - Developed
Medium Intensity, and 24 - Developed High Intensity)
2: Agriculture (81 - Pasture/Hay and 82 - Cultivated Crops)
3: Forest and Rangeland (41 - Deciduous Forest, 42 - Evergreen Forest, 43 - Mixed Forest,
52 - Scrub/Shrub, and 71 - Grassland/Herbaceous)
Attribute Label: Count
Attribute.'Number of Pixels
Distribution Information:
Distributor:
Contact Information:
Contact Organization Primary:
Contact Organization Human-Environment Modeling and Analysis Laboratory,
Department of Forestry and Natural Resources, Purdue University
Contact Person: Jarrod Doucette
Contact Address:
Address Type: mailing and physical address
Address: 195 Marsteller St.
City: West Lafayette
State or Province: IN
Postal Code: 47907
Country: USA
Contact Voice Telephone:
Contact Facsimile Telephone:
Contact Electronic^Mail Address: j doucett@purdue.edu
Distribution Liability:
The Human-Environment Modeling and Analysis Laboratory, Department of Forestry and
Natural Resources, Purdue University assumes no liability for results or conclusions drawn
from use of this data.
Standard Order Process:
Digital Form:
39
-------
Digital Transfer Information:
Format Name: ESRI GRID
Format Specification:
Format Information Content: Approximately 3GB per decadal file
Transfer Size: 100 MB per file compressed with standard zip Digital Transfer Option:
Online Option:
Computer Contact Information:
Network Address:
Metadata Reference Information:
Metadata Date: 20010501
Metadata Contact:
Contact Information:
Contact Organization Primary:
Human-Environment Modeling and Analysis Laboratory, Department of Forestry and
Natural Resources, Purdue University
Contact Person: Jarrod Doucette
Contact Address:
Address Type: mailing and physical address
Address: 195 Marsteller St.
City: West Lafayette
State or Province: IN
Postal Code: 47907
Country: USA
Contact Voice Telephone:
Contact Facsimile Telephone:
Contact ElectronicjVlail Address: j doucett@purdue.edu
Metadata Standard Name: FGDC Content Standards for Digital Geospatial Metadata
Metadata Standard Version: FGDC-STD-001-1998
Metadata Time Convention: local time
Metadata Extensions:
Online Linkage:
Profile Name: ESRI Metadata Profile
40
------- |