Comparative Google Earth Engine Workflows for Flood and Landslide Diagnostics in Lower Silesia, Poland, and Batang Kali, Malaysia

Abstract

This study evaluates Google Earth Engine (GEE) as the central computational environment for a unified, repeatable workflow for post-event geospatial diagnostics of two contrasting natural hazards: the September 2024 flood in Lower Silesia, Poland, and the December 2022 Batang Kali landslide in Selangor, Malaysia. Both cases followed the same analytical sequence: pre-event and post-event image selection, generation of a continuous change layer, threshold-based binary mask extraction, z-score standardization of change magnitude, point-based validation, and cartographic export. The flood branch used Sentinel-2 Level-2A imagery and ΔNDWI, whereas the landslide branch used Sentinel-1 GRD SAR imagery and Δσ⁰. For the Lower Silesia AOI, the workflow delineated 70.142 km² of flood-class pixels and produced complete agreement for the 240-point internal reference sample. Because the reference points were selected within the classified domain and interpreted from satellite imagery, this 100% agreement is treated as case-specific and potentially optimistic rather than as independent proof of error-free classification. For Batang Kali, all 11 reference landslide points were detected, giving 100% producer's accuracy, but 109 stable reference points were also classified as landslide-like; overall accuracy and user's accuracy were 9.17%. The landslide output is therefore interpreted as a high-sensitivity disturbance-screening product rather than a precise landslide inventory. The results show that GEE provides a transferable computational architecture, while diagnostic performance remains hazard- and indicator-specific.