High Spatial-Resolution Land Cover Classification and Wetland Mapping over Large Areas Using Integrated Geospatial Technologies

Abstract

Land Use and Land Cover (LULC) and wetland classification maps are an important prerequisite for many environmental studies. In order to produce accurate LULC and wetland maps at high spatialresolution, a new approach was developed to integrate image classifications, spatial data layers, and analysis methods using Python scripting. Both Maximum Likelihood and Object-based Feature Extraction were adopted into the LULC classification. A spatial analysis approach was applied to wetland mapping based on available wetland inventories and soil data. Python scripts were created and used to automate these processes for each of the 30 reference sites across Minnesota and Wisconsin of the United States, which encompassed the entire study site. Results demonstrated that the proposed method allowed for the integration of geospatial data of varying sources and qualities to produce accurate LULC and wetland maps effectively. The results of accuracy assessment indicated that the classification maps for Minnesota and Wisconsin were of comparable quality. The objectbased classifier extracted LULC effectively from the Wisconsin imagery with acceptable accuracy despite lacking of the NIR spectral band. These maps were used as inputs to create a hydro geomorphicap-proach (HGM) guidebook (Hauer and Smith 1998) for both states (Cook et al. unpublished). The Python-based technique was found to be especially beneficial when dealing with big datasets over large study areas, as it allowed batch processing.