This research compares three different classification algorithms for mapping crops in Pinal County, Arizona, using both present and historical image data. The study area lacked past crop maps, and farmers were dealing with the risk of evolution of resistance to insecticides in the whitefly, a global pest of cotton, fruits, and vegetables. The ability to create historical crop maps without concurrent training data is an invaluable tool for historical integrated pest management research. Comparison of maximum likelihood, object-oriented, and regression tree classifiers was done with Landsat Thematic Mapper imagery and high quality crop maps. Classification outputs for the three years in this research all achieved overall accuracies above the traditional benchmark of 85%. Comparison of the classification results shows that the classification and regression tree technique clearly outperformed the other classifiers. Using training data from one year and applying that data to another year for classification purposes demonstrated that overall accuracies from 71% to 83% are achievable, although accuracies were consistently greater than 85% for some crops.
ASJC Scopus subject areas
- Earth and Planetary Sciences(all)