Grantee Research Project Results

Final Report: A Dynamic Spatial Socioeconomic and Ecological Landscape Model to Assess Environmental Impacts of Forest Change on the Southern Cumberland Plateau of Tennessee

EPA Grant Number: R829802
Title: A Dynamic Spatial Socioeconomic and Ecological Landscape Model to Assess Environmental Impacts of Forest Change on the Southern Cumberland Plateau of Tennessee
Investigators: Gottfried, Robert , Haskell, David , Williams, Douglass , Evans, Jonathan
Institution: University of the South
EPA Project Officer: Hahn, Intaek
Project Period: September 1, 2002 through August 31, 2005 (Extended to August 31, 2006)
Project Amount: $248,265
RFA: Futures: Research in Socio-Economics (2001) RFA Text
Research Category: Environmental Justice

Objective:

Within the Southeast, where forest issues are politically volatile, the Cumberland Plateau of southeastern Tennessee has attracted regional and national attention as a major hotspot for forestry-related, landscape-level change. The Plateau, and the Southern forested landscape in general, has experienced rapid change as a function of new socioeconomic pressures. However, the removal of the native forest to permit other uses may have significant ecological implications. The Cumberland Plateau in Tennessee contains some of the largest remaining tracts of privately owned, contiguous temperate deciduous forest in North America. From a conservation biology standpoint, this area has received special attention because of the extremely rich animal and plant diversity associated with these remaining tracts of native hardwood forest habitat. All too often the debate concerning the causes, extent, and impact of forest change in the Southeast has been hampered by the lack of reliable data and analysis.

We have attempted to address this information problem by utilizing a relatively rich database of parcel level economic and social data to build an economic model of land use change in a dynamic rural landscape and to use that model to explore future land use change. By linking this model to ecological data, we have been able to explore some environmental implications of individual landowner behavior over a large area under a variety of scenarios. Such models are rare but powerful, both conceptually and socially.

The objectives of the research encompassed: (1) developing a spatial socioeconomic model of change in land use/land cover (LULC) for the Southern Cumberland Plateau for the period of 1980-2000; (2) integrating this model with ecological data to understand the socioeconomic processes bringing about environmental change in the region; (3) using this understanding and the model to assess potential future environmental impacts of likely socioeconomic events or trends; and (4) investigating the impacts of possible policy responses.

Our research represents one of only a few studies that use parcel level data to examine land conversion in a rural context, as opposed to the urban fringe. However, the nature of the available data and of the study area make this study even more interesting. The bluffs that characterize the edges of the Plateau comprise unique amenities that command a high premium. Our GIS data allow us to examine their impact on land use conversion. Using land cover data we also can examine the impact that a parcel’s land cover, and land cover in the surrounding area, has on the probability a home will be built. Accordingly, we can examine the extent to which pine conversion and conversion to grassy cover affect landowners who use their land primarily for recreation and residence; people who may value environmental amenities greatly. To our knowledge, the simulation model we developed is unique in that a home construction model interacts with a forest conversion model to drive land change. It also uses national and regional economic drivers while most models do not and takes into account, at least in part, the impact of timber company land divestitures. Finally, linking these LULC changes to ecological data allows us to begin to explore some of the environmental implications of these socioeconomic changes and their possible implications for the sustainable development of our region.

Summary/Accomplishments (Outputs/Outcomes):

Because spatial data for this very rural area hardly exist, we had to develop a detailed parcel level geographic information system (GIS) database of all nonurban parcels having 10 acres or more, incorporating tax maps we digitized plus county tax data for the year 2000. In addition, we developed GIS coverages for 1990 and 2003 of LULC in addition to our previously developed coverages for 1980, 1997, and 2000. Our LULC categories consisted of native hardwood forest, pine plantation, grass/shrub, pine mix (largely former strip-mined areas with pines and hardwoods), logged, and water/reservoirs (which are few). If land was logged during a given coverage year, we examined its LULC in the following coverage and converted its LULC in the first year to that of the second.

The spatial analyses and model development were performed in the ArcGIS software suite. Due to the size of the study area, the number of features involved, and the fact that some of the variables were the result of several steps, the standard tools provided with the software were found to be inadequate. Accordingly, we ended up developing macros in Visual Basic for Applications (VBA). Development within the ArcObjects environment is very involved and not always documented by the Environmental Systems Research Institute. Often considerable time was spent on errors with the code and processing logic. This approach does, however, offer improved speed and better memory management.

We used logit (in the STATA statistical package) to estimate regression equations for three models based upon our theoretical framework. The first estimated the probability of building a house on a parcel that previously had no residence on it (“raw” parcels). The second estimated the probability of building another residence on parcels that already had at least one house. In either case, we could not locate the residence within the parcel spatially but only record the building of a residence within the space delimited by the parcel boundary. The third model used multinomial logit to estimate the probability of a subparcel polygon of native forest converting to either pine plantation or to grass/shrub.

The three models utilize different time steps. Because of the availability of annual data on home construction, we estimated the housing models annually over the entire study period. Because of the relatively large number of annual observations, we were able to include time series economic variables as drivers of the decision to build a house. However, given that we possessed only five distinct coverages of LULC over time, we only had four observations of land use change for any subparcel polygon. We estimated the LULC model for the periods 1980-1990, 1990-1997, and 1997-2003.

Because landowners often convert only part of their parcel to another use, we based our conversion model on subparcel areas derived from the cumulative changes from 1981-2003. In essence, these comprise management units for the parcels. We based this conversion model on a reduced form multinomial logit model where the probability of a subparcel changing LULC depends upon parcel characteristics, subparcel characteristics, LULC characteristics on and around the parcel, number of homes around the parcel, landowner type, and the presence or absence of a house on the parcel. Similarly, we based our home construction model on a reduced form logit model whereby building a home depends upon parcel characteristics, LULC of the parcel and of the area around the parcel, landowner type, and national and regional economic variables. Having only four observations of land use change prevented us from using economic drivers in the regressions for land cover conversions. Thus, regional and national economic forces enter the model via their effects on house construction, thereby indirectly affecting the decision to change land cover via changes in houses on and around a parcel.

In the course of our research we came to realize that the period 1998-2000 may represent a watershed event for the Plateau. Starting around 1998, large landowning companies, primarily timber and resource-oriented companies, started divesting their landholdings. In Van Buren County, this divestiture represented approximately 40 percent of the land area of the county (and now about 90%). Also, the occurrence of the tremendously destructive pine beetle infestation at around this time probably affected the management decisions of many landowners who never before had encountered such levels of destruction. Consequently, we based the simulation model for the period after 2003 on the results of the 1997-2003 logit analysis.

Logit Results

The logit results for the conversion of native forest confirm that the factors affecting native forest conversion have changed over time. One way to summarize the results is to count how many of the significant coefficients in the 1997-2003 period also are significant with the same sign in the earlier periods of 1980-1990 and 1990-1997. These results are summarized in Table 1.

Table 1. Summary of Logit Results for the Conversion of Native Forest

	Hardwood to Hardwood	Hardwood to Pine	Hardwood to Other
Significant Coefficients 1997-2003	21	16	22
Same Coefficients Significant in at Least One of the Periods	12	10	9
Same Coefficients Significant in Both of the Other Periods	9	5	6

Variables that consistently explain the likelihood of a parcel remaining hardwood (same sign and significant in at least two of the periods) include the parcel size, subparcel size, and the environmental amenity variables capturing average subparcel slope, soil moisture, bluff frontage, presence of a stream, and presence of a waterfall. We can conclude that smaller subparcels with environmental amenities appealing to homeowners that are part of homogeneous parcels are less likely to be converted to something else.

Variables that consistently help explain transitions to pine include whether the owner is a timber company, the environmental amenities variables, the presence of natural gas, parcel size, distance from Fall Creek Falls State Park, and the larger the fraction of surrounding land that has been logged or that is in grass/shrub. We can conclude that subparcels that are part of larger parcels, that are owned by timber companies, that do not have a utility infrastructure, that do not have environmental amenities attractive to homeowners, that are surrounded by active logging and that are further from Fall Creek Falls are more likely to experience pine conversion. Similarly, subparcels part of larger, heterogeneous parcels not owned by not-for-profit owners and not featuring environmental amenities are more likely to convert to grass/shrub.

The logit model for residential construction showed that parcels with bluff frontage, utilities, greater proximity to a nearby house, less proximity to a road (small parcels), greater proximity to a road (large parcels), percentage of parcel in grass/shrub (before 1998), and an owner who was a county or nearby county resident experienced a greater likelihood of having a house built on it. The Wilshire stock index also positively affected this probability in a statistically significant manner. Small parcels for purposes of our analysis consist of parcels between ten and ninety acres in size while large parcels have more than ninety acres. Factors that negatively, and significantly, affected the probability of building a home included having the year be after 1997, bluff frontage (large parcels), percentage of grass/shrub on a parcel (after 1997), percentage of grass/shrub within 1 km of the parcel boundary, percentage of the parcel in pine (small parcels after 1997), location on a paved road, greater proximity to a house (large parcels before 1998), and an owner who was a business or individual acting as a businessperson (prior to 1998). The model generally performed as expected, revealing differences in behavior of small and large parcels, as well as differences before and after 1997.

For parcels that already had a home built on them, the probability of having another home built on it depended significantly on only four variables. The value of housing in the area had a negative effect, number of houses in the area a positive effect, business owner a positive effect, and unemployment rate a negative effect. This would seem to indicate that higher income areas, as proxied by the value of houses in the area, tend to discourage homebuilding. Business owners often may be developers, and so would be more likely to build.

Simulation Results

Land Cover and Home Construction. The model overpredicted by 5.4 percent the amount of forest clearing when it was validated against the historic data for 1997-2003 and subsequently overpredicted the growth in grass/shrub and pine (particularly pine). Overall, the model correctly predicted the land use of a subparcel in 2003 83 percent of the time, which is not too surprising given that most subparcels did not change over the period. The model performed much better in the aggregate and spatially when it came to predicting the total amount and parcel location of home construction. Overall accuracy for home construction was 95 percent.

The difference in performance of the construction and land cover models may reside in the difficulty of directly incorporating economic drivers into the land cover model. Bockstael moved away from using our approach because, while it could perform well at a fine scale, it had difficulty predicting accurately the aggregate amount of land conversion. The hazard approach Bockstael later took can predict the aggregate amount of land use more accurately but cannot incorporate time series economic variables. We had hoped that, given our long time period, we could incorporate economic drivers into our land cover model, thereby enabling us to get the aggregate amount of land cover correct. Our approach indeed worked well in the case of home construction, which was driven by economic time series variables. However, because of the unexpectedly weak link between home construction and land cover (home construction being driven by economic time series variables) and the small number of observations of land cover, our hope that this would be the case for land cover also did not materialize. We hope to explore further, after this grant, the possibility of using a hazard approach for the land cover model and combine this with our current approach for residential construction.

For each scenario we simulated to 2024, starting in 2004. Given that the conversion regression for 1997-2003 is for 7 years, this represents three time steps. The residential models simulate annual home construction, whereas land cover converts every 7 years, based on the 7-year conversion probabilities provided by the regression analysis. Given that each run took 4 hours on average to run, we were able to run the model 30 times for each of the six scenarios.

We ran three base scenarios under differing assumptions of the values of the economic drivers in the future: High Growth (High), Low Growth (Low), and Shock. The former adapt historic data from periods of high and low growth and apply them to our model. Shock uses the historic data for the period of the two OPEC oil shocks and the subsequent adjustment by the economy.

The next three scenarios use the high growth economic drivers to explore three questions. The first two questions address the role that landownership might play in affecting home construction and LULC. The Business and the Individuals scenarios change the 2000/2001 owners of all parcels of 500 acres or more to either business/for profit individuals or to individuals. Using the economic drivers for High, we can compare the results for the High scenario to those where one owner type, Business or Individuals, dominates these large parcels. Given the divesting of timber tracts and the subsequent sale of land to developers, these scenarios help us think about the potential impact of ownership changes. The last scenario, No Housing Change, holds the number of houses constant at the 2003 level, in order to examine the role that home construction may play on forest conversion.

High has the greatest number of homes built over time, followed by Low Growth and Shock. The High scenario exhibits greater swings in construction than the other two. Both graphs also show that changing land ownership in the large parcels has little impact on home construction. The three growth scenarios exhibited very similar results with respect to aggregate native forest, grass/shrub, and pine for the simulated years of 2010, 2017, and 2024. Notably, all the scenarios show a slowing of conversion over time. There is no statistically significant difference between the three scenarios (at 5%) for any of the three land covers. Thus, whereas the number of houses varies dramatically between the three growth scenarios, the LULC does not. This either could be the result of the small number of runs, an artifact of the model, or a reflection of reality. Yet, the standard deviations for the runs are relatively small.

Because the economic drivers enter the simulations explicitly only in the residential model, we may underestimate the differences in LULC between the scenarios. However, should home construction create little clearing of native forest and should changes in LULC result from structural changes in the composition of landowners, then the models may reflect rather accurately the dynamics of change. For instance, it could be that the logit regressions for native forest conversion pick up the impacts of divesting and changes in landowner composition and that these changes drive native forest conversion far more than changes in annual values of economic drivers. If this is so, the sensitivity of home construction to the economic drivers used, and the lack of sensitivity of LULC, may reflect reality more than a shortcoming of the model.

Changing all owners of parcels of 500 acres or more to Business results in statistically significant, larger amounts of native forest than with the mixture of owners used in High (the 2000/2001 owners), while changing owners to Individuals does so even more. The latter difference also is significant. Both Business and Individuals show substantially less pine than High, with Individuals even less than Business. These differences, too, are significant. The differences between Business and Individuals also are significant for native forest and pine, though not for grass/shrub. However, the results for Individuals suggest that, over time individuals may lead to significantly more grass/shrub than business, given the trend in the results. Thus, ownership does make a difference. The pine conversion would be less over time with either Business or Individuals owning the large tracts, and native forest conversion would be smaller, though it still would occur. Given the change in ownership due to divesting, these scenarios suggest what changes may be in store.

Another question still remains: Does home construction explain the large increase in grass/shrub cover? The No Housing Change scenario examines this question by using the same driver values as High but keeping housing constant at the 2003 level. The results are intriguing. Each year, No Housing Change shows less native forest than High and more grass/shrub than High. Although not significant for 2010 and 2017, the growing difference between the two scenarios results in a significant difference in 2024. The two scenarios exhibit no significant difference in pine. Thus, according to the model, home construction actually saves native forest compared to having no home construction. It results in less grass/shrub over time. This suggests that the large increases in grass/shrub result from something other than home construction. These results reflect the regression results.

Environmental Impacts. We examined three general types of environmental impact: water quality, loss of vernal ponds, and changes in bird populations. With respect to water quality, we examined the extent of native forest conversion at the landscape and watershed (hydrologic unit code [HUC] 10) levels. We used as our metric the probability weighted average of the meters of stream in a watershed that would be affected by a subparcel converting to pine or grass/shrub. As of 2003, about 22 percent of the areas’ streams already were impacted by native forest clearing. All the scenarios indicated that approximately 40 percent of the streams would be affected by 2024.

We used an already developed GIS tool to assess the extent of streamside management zone (SMZ) compliance by landowner type. The Tennessee state best management practices (BMPs) are voluntary guidelines. The Sustainable Forestry Initiative^® (SFI) SMZ standard specifies compliance with the state BMPs. The Forest Stewardship Council’s (FSC) SMZ standard for the Appalachia region, of which the study area is a part, requires greater widths than do the state BMPs/SFI. Although state BMPs or SFI standards currently are applied by some landowners in the study area, evaluation of compliance with FSC is purely hypothetical because no private landowners in the region were known to be using that standard as of 2003.

For our research, we declared SMZs as “compliant” or “noncompliant” only in the limited sense of how the width and slope data collected for this study matched the quantitative portions of the SFI and FSC standards. Neither owner type nor owner location had a significant effect on BMP/SFI compliance. FSC compliance, however, was significantly affected by both owner type and owner location (local/nonlocal). Tax parcels owned by timber companies were less FSC-compliant than those owned by individuals or nontimber companies. In addition, tax parcels with owners from outside the state were less FSC-compliant than those with owners classified as “county” or “state.” The 75.8 percent compliance rate with BMPs/SFI is encouraging, but if the SMZ standards of FSC prove to provide requisite watershed protection in this region, then the 15.2 percent FSC-compliance of whole SMZs on the Plateau is cause for concern. The widespread adoption of SFI standards (or mandatory state BMPs) would have little impact on SMZ widths in this landscape. The widespread adoption of FSC standards, however, would necessitate a transformation in the design of riparian buffers on the plateau.

In 2003, 21 percent of blue line streams (National Hydrographic Dataset) and 24 percent of streams from our more detailed layer were affected by native forest clearing. For either measure of streams, all the scenarios indicate that approximately 40 percent of streams will be affected by 2024. Should landowners implement SMZs according to SFI standards at the current rates of compliance, then only about 11 percent of streams will be affected. Whereas affected stream reaches increase about 21 percentage points from 2003 to 2024 without these SMZs, with these SMZs they only rise about 6 percentage points. Implementation of SFI standards at current rates thus could significantly ameliorate water quality problems over time, assuming that these SMZs are wide enough to be effective. Should the wider FSC standards be more appropriate, about 35 percent of streams would be affected by 2024, a rise of 18 percentage points. The lower rate of compliance with this stricter standard implies that water quality problems might well increase over time.

Maps of forest change at the HUC level show that water quality problems may vary spatially. By 2024, HUCs in the southern part of the study area will tend to be more forested, while HUCs in the central region will tend to be in pine and HUCs in the north will tend to be grass/shrub. Maps of change in forest cover show most gains in grass/shrub occurring in the central and northern regions while pine gains most in the central region. When combined with variations in housing density, these spatial differences imply that water quality issues might be highest in the future in the central and northern regions.

Small vernal pools distributed throughout the native forests of the Cumberland Plateau create a complex network. These wetlands support ecologically diverse communities of plants and animals. Because no other natural bodies of water are present throughout this particular ecoregion, these wetlands provide critical breeding and foraging habitat for numerous invertebrate, bird, and amphibian species. As part of another project, the laboratory mapped the locations of all vernal pool wetlands on the southern Cumberland Plateau study area using high-resolution aerial imagery in a GIS. We identified 842 vernal pool wetlands (both natural and roadside) within our study site in 2004. Prior to this project, neither the abundance nor the spatial distribution of these wetlands were known in this region. When a pond was located in a subparcel that experienced native forest conversion in the run of a scenario, we considered that pond lost ecologically. All the scenarios show a substantial loss of ponds that had been intact as of 2003, with losses running from about 16 to 20 percent.

Finally, we calculated the potential impact that forest cover change would have on bird populations. Using the apparent densities previously estimated from point counts, estimates of the total change in the population were generated for two scenarios that represent the possible extremes in terms of land conversion: Individuals and No Housing Change. Eleven of the 18 species are forecast to increase under both scenarios, while seven species (predominantly forest-dwelling species) are forecast to decline. The birds that increase under both scenarios are primarily those that are found in urban habitats and/or early successional habitats.

The predicted decline in native forest and rise in pine and grass/shrub is a cause for concern because the greatest number of rare birds (i.e., Partners in Flight Category I species) are present in this declining habitat. All forest-dwelling bird species are projected to decline in abundance regardless of the scenario. However, the species that will be positively affected most likely will be species that are present in grass/shrub habitats and pine plantations, rather than the rarer species present in the native forests. Some early-successional species may also benefit temporarily from changes in land use. The conversion of native forest may initially benefit these declining species, although there is no evidence to suggest that the habitat will be kept at the early successional stage they prefer.

Overall, species that breed in human-modified habitats are expected to do well under all scenarios. However, species that were only recorded in native forest are projected to decline. This may be a cause for concern because many of these forest-dwelling species are declining elsewhere in North America. While conversion may benefit widespread and common species such as American robins and house finches, it has a detrimental effect on native forest-dwelling species.

Discussion

The spatial pattern of forest cover change and residential construction varies substantially across the landscape in the above simulations. In many ways, the sparsely settled, heavily forest-dominated landscape is “ruralizing,” becoming a patchwork of forest, field, and homes.

This has important implications for the future development of the area. Many citizens, local leaders, and state officials are looking to recreation and tourism as the future economic growth engines for the area. However, particularly in the central and northern regions, but in the south as well, the decline in the amount of native forest raises real questions about this strategy for economic development. Who will come to a region to bird watch, hike, or camp when the little forest that remains is surrounded by pasture and pine? Thus, local government and state officials, as well as the general citizenry, need to look ahead to decide on the role that native forest will play in their development strategy and work proactively to ensure that these resources will be available to fulfill that role.

Similarly, the spatial variation in land cover change and home construction and low rate of FSC SMZ standards, could have serious implications for water quality and, perhaps, quantity, given the dependence of many people on wells as well as water districts. It would appear prudent that citizens, planners, institutions concerned with water quality and quantity, and government officials focus on those watersheds most critical to water quality and quantity (both on the Plateau and downstream) that the model suggests might come under particular stress and work to forestall future problems now. It also suggests that persons wanting to maintain large areas of forest may find it easiest to do so in the southern region where pressures may be relatively less. Given the private landowner makeup of the Plateau and the spatial heterogeneity of forest change, it will be important to devise a spatially sensitive, targeted set of incentives with as many stakeholders as present involved in its crafting.

The model can aid in this process in several ways. First, a number of groups already have utilized the data layers themselves for a variety of purposes. Until recently, spatial ecological and socioeconomic data have been sorely lacking for the Plateau. Second, showing various constituents maps of prospective land cover and residential change can serve as a powerful educational tool and motivator, particularly when presented in a GIS format that allows people to focus on areas of particular interest to them. Third, although the model overestimates forest change, it can identify those areas and particular parcels with the greatest relative probability of conversion. Accordingly, if users overlay a layer of probability of conversion on layers of concern, such as critical habitat, streams, corridors, or watersheds, the model can aid interested parties in identifying parcels and areas most warranting attention. Finally, the model can serve as a vehicle through which people may explore a variety of spatial concerns ranging from habitat conservation to the fiscal implications for county governments of future growth.

Journal Articles on this Report : 2 Displayed | Download in RIS Format

Publications Views

Other project views:	All 10 publications	2 publications in selected types	All 2 journal articles

Publications

Type	Citation	Project	Document Sources
Journal Article	Lemoine D, Evans JP, Smith CK. A landscape-level geographic information system (GIS) analysis of streamside management zones on the Cumberland Plateau. Journal of Forestry 2006;104(3):125-131.	R829802 (Final)	Full-text: ResearchGate-Abstract and Full Text - PDF Exit Abstract: Ingenta-Abstract Exit
Journal Article	McGrath DA, Evans JP, Smith CK, Haskell DG, Pelkey NW, Gottfried RR, Brocket CD, Lane MD, Williams ED. Mapping land-use change and monitoring the impacts of hardwood-to-pine conversion on the southern Cumberland Plateau in Tennessee. Earth Interactions 2004;8(9):1-24.	R829802 (Final)	Full-text: Earth Interactions-Full Text - PDF Exit Abstract: ResearchGate - Abstract & Full Text - PDF Exit

Supplemental Keywords:

forest, public policy, socioeconomic, conservation, Southeast, rural South, Tennessee, TN, land use, land cover, LULC, Economic, social, & behavioral science research program, ecosystem protection/environmental exposure & risk, geographic area, ecology and ecosystems, economics & decision making, environmental monitoring, forestry, monitoring/modeling, social science, state, urban and regional planning, decision-making, GIS, Tennessee, TN, decision making, decision support tool, ecological models, economic research, environmental decision making, environmental impact, environmental impact comparison, forest conservation decisions, forest ecosystem, forest resources, forests, land management, land use, landscape ecology, model-based analysis, modeling, monitoring, remote sensing, remote sensing data, social impact analysis, social sciences, socioeconomics,, RFA, Ecosystem Protection/Environmental Exposure & Risk, Scientific Discipline, Geographic Area, Economic, Social, & Behavioral Science Research Program, Economics & Decision Making, State, Forestry, decision-making, Monitoring/Modeling, Social Science, Ecology and Ecosystems, Environmental Monitoring, Tennesee (TN), economic research, landscape ecology, socioeconomics, water quality, ecological models, social impact analysis, monitoring, spatial landscape model, environmental decision making, forests, remote sensing, forest conservation decisions, environmental impact comparison, forest reources, land use, decision support tool, land management, modeling, forest ecosystem, GIS, model-based analysis, remote sensing data, environmental impact

Relevant Websites:

http://lal.sewanee.edu Exit

Progress and Final Reports:

Original Abstract

2003 Progress Report

2004 Progress Report

2005 Progress Report