Integration of sentinel optical and radar data in Google Earth Engine for mapping small complex agroforests of shade-grown coffee in northern Peru

Abstract

Coffee is a key global export product; however, there is still limited information on the location and dynamics of its growing areas, especially in light of the implementation of new regulations such as the EUDR. This study used optical and radar data from the Sentinel satellite, processed on the Google Earth Engine (GEE) platform, to map vegetation cover and land use, including coffee farming systems in the province of Rodríguez de Mendoza. The Random Forest algorithm was applied, trained with 531 verification points collected in the field. As a result, six classified maps were generated, showing sustained growth in the area under coffee cultivation, from 7,111.071 ha to 9,443.050 ha over a six-year period. The cartographic products achieved high levels of accuracy, with Overall Accuracy (OA) values above 94% and a Kappa index (KI) above 88.9%. These results demonstrate the potential of GEE for multi-year monitoring of permanent [likely continues, cropped] coffee-growing areas.