Comparing global and local land cover maps for ecosystem management in the Himalayas

Abstract

As a breakthrough in geographical data science, two high-resolution (10m) global land cover products were published by the Environmental Systems Research Institute (ESRI) and European Space Agency (ESA). Yet, the accuracy and applicability of those maps have not been diagnosed in the Himalayas which is characterized by multifunctional and heterogenous landscapes. We compared the ESA and ESRI land cover products in terms of classification performance, overall accuracy, spatial agreement, and applicability in the Hindu-Kush Himalayan region by taking a case of multifunctional landscapes of Nepal. In addition, we generated a local land cover map in the Google Earth Engine (GEE) environment based on Sentinel-2 imagery for 2020 and compared it with the ESRI and ESA land cover products. The overall accuracy of ESRI product was higher (87.3%) than ESA (71.2%) and GEE-generated (73%) map. Shrubland, grassland and forestland were found to be the most common source of confusion in land cover mapping for both products. The spatial agreement (pixel match) of ESA and ESRI products was found to be about 62% of the landscape area, indicating substantial agreement between both products for land cover classification. The ESRI product is more reliable in mapping the Himalayan region. However, we suggest that the global land cover products must be flexible to further improvement by incorporating local topography, vegetation phenotype, spatial heterogeneity, and temporal dynamics of landscape and ecosystem services, particularly in the Himalayas.