Evaluating Satellite Land Cover Classification Accuracy Using Participatory Remote Sensing

Abstract

Satellite remote sensing has become a foundational tool for monitoring land cover change and urban development. However, classification accuracy remains limited in complex urban environments where vegetation, built infrastructure, and water coexist within small spatial scales. This study evaluated the accuracy of multiple satellite land cover datasets within a 3 km × 3 km area of interest in Hato Rey, Puerto Rico, using the National Aeronautics and Space Administration (NASA) Adopt-a-Pixel participatory remote sensing framework. Ground observations were collected at 23 sampling locations using the Global Learning and Observations to Benefit the Environment (GLOBE) Observer application and compared with land cover classifications derived from European Space Agency (ESA) WorldCover, Dynamic World, ESRI Land Cover, Meta Tree Canopy, and Landsat imagery. Results showed that several satellite datasets frequently overgeneralized land cover as entirely urban, failing to detect localized vegetation and water features observed on the ground. Dynamic World and ESRI Land Cover achieved higher overall accuracy (91.3%) due to consistent classification of urban areas, while Meta Tree Canopy and ESA WorldCover demonstrated improved detection of vegetation in heterogeneous environments, despite slightly lower overall accuracy (87.0%). Landsat time series analysis revealed mixed vegetation trends over time, reflecting both urban expansion and persistent vegetation. These findings demonstrate the importance of integrating participatory ground observations with satellite data to improve land cover classification accuracy and enhance environmental monitoring in rapidly developing urban environments.