An Analysis of Losses to the Southern Commercial Timberland Base

Demographic and physical factors influencing the conversion of commercial timberland in the south to nonforestry uses between the last two Forest Inventory Analysis (FIA) surveys were investigated. GIS techniques linked Census data and FIA plot level data. Multinomial logit regression identified factors associated with losses to the timberland base. Conversion to agricultural uses represented the largest loss (1.48%) to the commercial timberland base. Slope, forest size, distance to the nearest city, as well as median income and education level were all negatively related to the probability a plot would be converted from forestry to agricultural uses. Conversion to urban uses (1.13%) represented the second largest loss. Forest size, distance to developed areas and distance to the nearest city were all negatively related to the probability a plot would be converted to urban uses. Conversions to a number of miscellaneous uses accounted for an additional 0.38% loss. 1 Approved for publication as Journal Article No. FO-096 of the Forest and Wildlife Research Center, Mississippi State University. Funding for this project was provided by the Southern Resource Assessment Consortium and the Southern Forest Experiment Station, USDA Forest Service. 2 Assistant Professor, Forestry Department, Mississippi State University and Research Forester/Economist, USDA Forest Service Ecosystem Management, respectively. INTRODUCTION Timber supply projections play an important role in both public and private arenas. Forest industries utilize timber supply projections as key components in long-term capital investment decisions and corporate planning. Government policies, regulations and legislation concerning public timberlands, landowner assistance programs, and forest taxation are based in large part on the long-term timber supply outlook. Yet timber supply projections are notoriously inaccurate. Timber supply famines have been predicted repeatedly but have failed to materialize. Consequently, timber supply projections must be refined and updated constantly. A key component of estimating future timber supplies is predicting changes in the land area devoted to timber production. A number of studies focused on this issue. Alig et al. (1983) highlighted the need for new approaches to long range forecasts of forest area change. Parks and Alig (1988) provided a critical review of landbased models for forest resource supply analysis. Alig et al. (1986) examined changes in ownership and cover-type for timberland in the south based on Forest Inventory and Analysis (FIA) data. They found forest farm acreage decreasing, forest industry land holdings steadily increasing and a sharp decrease in natural pine forest types, partially offset by increases in planted pine. Alig (1985, 1986) developed an econometric model of land-use changes for the southeast and examined shifts between timberland and cropland, rangeland and urban uses. Population and personal income were found to be the major determinants of land-use changes. Variations of Alig's basic model have been used in a number of subsequent studies projecting various components of the timberland base. For example, Alig et al. (1987), compared forest acreage changes and the underlying causes for northern and southern regions. Alig et al. (1988) applied the procedure developed for the southeast to the south-central United States. Alig et al. (1990) tracked timberland changes from 1952 through 1987 and projected future changes in the timberland base through 2040 for the entire United States. Total timberland was predicted to decrease by 4% by 2040. Alig and Wear (1992) focused on changes in the private sector of the timberland base through the year 2040. Information recently compiled for the 1992 Resource Planning Act (Powell et al., 1992) provided an opportunity to refine and update the current timberland change projections. Rather than replicate Alig's model utilizing current FIA data, a fundamentally different approach was utilized. Geographic Information Systems (GIS) make it possible to combine raw un-aggregated data from a variety of sources such as Census and FIA data. Plot-level data can easily be linked with the appropriate Census-Tract level data, providing a much more detailed data base with which to investigate changes in the forest land base. This report focuses on losses to the timber land base. DATA The first step in developing the geographic database was to import Census Tract boundaries and associated data for the Southern Region: Alabama, Arkansas, Louisiana, Mississippi, Eastern Oklahoma, Tennessee, and Eastern Texas. The Census Bureau subdivides counties into Census Tracts, areas of similar population characteristics, economic status, and living conditions. Census Tracts average 4,000 people but range from 2,500 to 8,000. Tracts vary widely in area depending on population density. Each Census Tract has an associated set of demographic data. The US Department of Agriculture, Forest Service FIA plots are located on a 3-by-3 mile grid pattern. Each FIA plot that is currently forested, or was previously forested, has an associated set of physical data which is remeasured on roughly an 8-year cycle. Data for 3 The Census Bureau calls these county subdivisions Census Tracts in urban areas and Block Numbering Areas in rural areas. In this paper, the term Census Tract is used for both. plots that have never been forested are limited to the latitude and longitude coordinates. Arc/Info was used to overlay FIA plots on the CensusTract base map based on plot coordinates given in latitude and longitude. The number of FIA plots in each Census Tract depends on the size of the tract. The FIA and Census data were merged, resulting in a combined set of demographic (Census) and physical (FIA) data for each FIA plot. Finally, the straight line distance from each FIA plot to the nearest urban area with a population of 30,000+ was determined using Arc/Info. A total of 32,050 plots are included in the data set. Census variables include population density (people per square mile), population growth rate, median household income (1989 dollars), and the percentage of the population with a bachelor’s degree or higher. The FIA variables include current ground use, past ground use, ownership class, site class, slope, forest size, physiographic type, distance to a truck-operable road, and distance to a “built-up” area of 10 acres or more. “Built-up” land is comprised of areas of intensive human use with much of the land covered by man-made structures. Current and past ground uses are categorized for this study as commercial forest land, non-commercial forest land, urban land, agricultural land, or waste land. Ownership is classified as National Forests, other public, industrial and non-industrial private (NIPF). Site class is the land’s potential timber yield measured in cubic feet/acre/year and is ranked from 1 (lowest) to 7 (highest). Slope is measured in percent. Forest size is the contiguous forest area, measured in acres, surrounding the FIA plot. Forest-area boundaries include public roads, railroads, non-forest uses, and major waterways. Ownership boundaries and power and pipeline right-of-ways are not considered. Physiographic types include pine, upland hardwood and bottomland hardwood types. For this study, distance to a paved or truck-operable road is classified in three categories: less than a mile, between one and three miles, or greater than or equal to three miles. Distance to built-up areas is categorized similarly. Land-use change is classified on the basis of two FIA variables: current ground use and past ground use. All plots with past ground use identified as commercial forest land are included in the analysis. Four outcomes were possible: no change (current ground use is commercial forest land), commercial forest land converted to agricultural use, commercial forest land converted to urban use, and commercial forest land converted to miscellaneous uses. Miscellaneous uses include areas legally removed from commercial forestry such as National Parks or Wilderness Areas, poor quality forest lands unable to support commercial forestry, and waste land. METHODS Changes in land-use were hypothesized to be a function of the physical and demographic characteristics of the plots as follows: Probability (USE CHANGEi = J) = f(SITEi, SLOPEi, SZFORi, PINEi, UPHDWDi, BHDWDi, ROAD1i, ROAD2i, ROAD3i, NFi, PUBi, INDi, NIPFi, DEVELOP1i, DEVELOP2i, DEVELOP3i, INCOMEi, POPDENi, POPGROWi, PCTEDi, DISTANCEi) where, for the ith plot, USE CHANGE is the loss from commercial forest land-use, if any, since the previous survey and J is a categorical variable indicating the specific type of loss, i.e., J = 0 for no land-use change, J=1 for losses to agricultural use, etc. SITE is the site class, ranked from highest (1) to lowest (7); SLOPE is the slope measured in degrees; SZFOR is the size in acres of the contiguous forest area surrounding the plot; PINE, UPHDWD, BHDWD are dummy variables representing different physiographic types: softwoods (PINE), upland hardwoods (UPHDWD), and bottomland hardwoods (BHDWD); ROAD1, ROAD2 and ROAD3 are dummy variables for the distance from the plot to the nearest paved road: one mile or less (ROAD1), between 1 and 3 miles (ROAD2), and three miles or more (ROAD3); NF, PUB, IND, NIPF are dummy variables representing ownership categories: National Forests (NF), other public (PUB), industrial (IND), and non-industrial private (NIPF); DEVELOP1, DEVELOP2, DEVELOP3 are dummy variables representing distance from the plot to the nearest developed area of ten acres or more: less than 1 mile (DEVELOP1), between 1 and 3 miles (DEVELOP2), and three miles or more (DEVELOP3); INCOME is the median household income in 1989 dollars for the surrounding Census Tract; POPDEN is the population density in people per square mile; POPGROW is the population growth rate; PCTED is the percentage of the population with a bachelor’s degree or higher; DISTANCE is the distance in kilometers from the plot to the nearest urban center of 30,000+ people. EMPIRICAL MODEL The probabilities of land-use change were estimated using a multinomial logit model as follows: