Evaluation and improvement of EnMAP’s cloud and cloud-shadow masks – An application in tropical western Kenya

• EnMAP’s cloud and cloud shadow masks show shortcomings in tropical regions. • Masks are improved with thresholding, machine learning, and physical proximity tests. • Global cloud and cloud shadow masks may need local adaptations, especially in the tropics. • The proposed open-source improvement protocol outperforms operational masks. Cloud and cloud shadow (CCS) detection is an integral pre-processing step in optical remote sensing workflows. Residual CCS can distort subsequent analyses, such as land use and land cover classification. The accuracy of operational cloud masks varies across ecoregions, with tropical regions being among the most error-prone. This study evaluates the performance of the EnMAP (Environmental Mapping and Analysis Program) mission’s operational CCS masks over tropical western Kenya and proposes an improvement framework. Our assessment reveals shortcomings in the delineation of CCS objects and a frequent omission of small clouds. Additionally, pixels along the water-land interface and in wetlands are frequently misclassified. To address these shortcomings, we developed EnICCS (EnMAP’s Improved Cloud and Cloud Shadows), an open-source tool that improves upon operational masks using band indices, thresholding, and masking, followed by partial least squares discriminant analysis (PLS-DA) and a post-processing routine that enforces the physical positional relationships between clouds and their shadows. EnICCS outperforms EnMAP’s operational CCS masks for the given region. It achieves F1 scores of 0.93 for clouds and 0.79 for shadows compared to 0.80 and 0.58, respectively. Object delineation is improved by up to 30%. The lightweight framework is designed for execution on standard end-user machines. While optimized for the densely vegetated tropics, its modular structure supports adaptation to different surface characteristics.