Machine learning approaches to multisource geospatial data classification: application to CRP mapping in Texas County, Oklahoma

We develop an Automated Feature Information Retrieval System (AFIRS) for accurate classification of multisource geospatial data, which involves multispectral Landsat imagery, ancillary geographic information system (GIS) data and other derived features. Two machine learning approaches, i.e., decision tree classifier (DTC) and support vector machine (SVM), are implemented as multisource geospatial data classifiers in the AFIRS. Specifically, we apply the AFIRS to the mapping of United States Department of Agriculture (USDA)'s Conservation Reserve Program (CRP) tracts in Texas County, Oklahoma. CRP is a nationwide program, and recently USDA announced payments of nearly $1.6 billion for new CRP enrollments. It is imperative to obtain accurate CRP maps for effective and efficient management and evaluation of the CRP program. However, most existing CRP maps are inaccurate and little work has been done to improve their accuracy. The proposed AFIRS is capable of handling the complex CRP mapping problem with high accuracy when limited training samples are available. Simulation results show that 5-10% improvements can be obtained by incorporating GIS ancillary data and other derived features in addition to multispectral imagery. This work validates the applicability of machine learning approaches to the complex real-world remote sensing applications.