Predicting the extent of lakeshore development using GIS datasets

Abstract

Land use along lakeshores impacts littoral habitat. We used raster-based land use and land cover datasets as well as a statewide dock polygon dataset to assess the development status of lake shoreland on 150 Minnesota lakes. A dataset containing dock polygons identified using a semiautomated process performed best statewide. An older raster dataset that classified rural development based on the presence of buildings performed better than a recent dataset based on dominant land cover. We classified points along the shore as developed or undeveloped using proximity to a development indicator: either a raster cell with a developed land use or a dock point. We compared classifications derived from GIS data to actual development, which was defined by the presence of a manually identified dock on aerial photos, and used total operating characteristic (TOC) analysis to evaluate the performance of each dataset. All 3 datasets classified development better than random chance. The dock dataset performed best, and its results were consistent statewide, while the raster datasets' performance varied among ecoregions. Researchers should be aware that the prevalence of the condition being classified has a large impact on some commonly used metrics, such as accuracy and positive predictive value. The costs and tradeoffs of different types of error (false alarms or missed detections) will vary in different situations and should be explicitly considered when deciding how and when to use classification systems like these.