Evaluating Satellite Imagery–Based Land Use Data for Describing Forestland Development in Western Washington

and commodity production, agriculture, and residential and other development. Although land cover sometimes can indicate land use (e.g., forest versus development), it is often not a good indicator of whether existing forestland is managed for timber and other forest commodity production. What would be useful is a cost-effective way to obtain data specifically designed to address the information needs of forest policymakers—data that can be used to characterize the extent of forestland development and its effects on private forest management. We examine land use data gathered using multiresolution classification of Landsat imagery for western Washington, west of the crest of the Cascades (Rural Technology Initiative 2006). That data gathering effort, funded by the US Forest Service Pacific Northwest Research Station’s FIA program, sought to test the feasibility of automating the interpretation of satellite imagery to produce data for describing and evaluating land use changes occurring on nonfederal lands, similar to data that FIA gathered over the past decade in collaboration with the Oregon Department of Forestry. Although highly useful, the Oregon data are based on fairly labor-intensive data gathering methods involving interpretation of aerial photos (Lettman 2002, 2004). The purpose of the western Washington effort was to see whether similar data of comparable accuracy could be developed faster and more cost effectively for other areas of the Pacific Northwest. In this study, we evaluate the accuracy of the western Washington land use classifications resulting from the automated process by comparing them to classifications reported by FIA inventories and population densities reported by the US census. We evaluate the usefulness of the data for policy analysis and management by estimating an econometric model describing development as a function of socioeconomic and topographic factors and compare development projections made using the model with those made by other studies using different data and methods. We conclude by discussing implications regarding how best to meet land use informational needs of forestry policymakers and managers. Western Washington Land Use Data The population of western Washington has increased rapidly over the last 25 years, resulting in expanded urban areas and fragmentation of forested and agricultural lands. The increase of forestland in the wildland-urban interface is believed to have implications for the economic viability of forest management focused on timber production, as well as the ecological sustainability of those same lands for habitat and the production of natural resource amenities and other ecosystem services. Increased demand for housing near urban centers, such as Seattle, Olympia, and Tacoma, has increased the value of forestland, often making development more financially appealing to landowners than managing for timber, wildlife, or other ecological values. Decreases in timber production, and in some cases the subsequent conversion of forestland to developed uses, could lead to economic and ecological concerns in the communities where such changes take place. For these reasons, forest policymakers and land use analysts have sought ways to better describe land use changes taking place in the region and their effects on forestry. Despite significant population growth and development in western Washington, much of the land use data and analysis available for the region had been characterized by fairly coarse (e.g., county-level) spatial scales (Alig and White 2007). Although useful for describing general trends, these data tend to constrain evaluation of likely management and ecological effects. Land use data gathered in Oregon served as a model for data desired for western Washington. In a combined effort, the US Forest Service Pacific Northwest Research Station’s FIA program and the Oregon Department of Forestry have developed and periodically updated spatially referenced data describing several discrete categories of land use, as well as building (or structure) counts, on nonfederal lands throughout the state (Lettman 2002, 2004). These data are gathered from high-resolution digital aerial photographs spanning the early 1970s to the present. By tracking land use on individual sample points from one sample period to the next, the data document changes among land use categories and increases in building counts over time. With 37,000 sample points statewide and three or four sample periods depending on region, the data set provides an outstanding resource for observing rates and patterns of land use change and evaluating its potential effect on private forest management in Oregon. However, data based on aerial photography can be time-consuming and costly to develop and update, especially if coverage is desired for relatively large states typical of the western United States. Procuring aerial photos themselves can be costly, difficult, and even sometimes impossible if historical coverage is desired. Aerial photo interpretation can also involve interpreter bias arising from differences in the way individual photo interpreters observe and record land uses. For these reasons, the FIA program sought to develop and test methods for gathering similar land use data from widely available satellite imagery at potentially