Rapid and reliable detection of landslides is essential for disaster response, hazard assessment, and environmental monitoring, particularly in environments where timely information is critical. Recognizing the need, this study proposes a unified framework that integrates pixel-based LULC classification with landslide detection using multispectral satellite imagery. In this framework, landslides are identified as vegetation-to-bare land transitions constrained by temporal, terrain, and morphological conditions, making detection dependent on accurate classification of vegetation and bare land. An optimal classifier configuration was selected through a statistical scheme based on three-way analysis of variance (ANOVA) to ensure reliable classification and applied to pre- and post-event imagery to derive landslide candidate areas. To improve detection reliability, terrain and morphological filters were applied, including a slope constraint selected through sensitivity analysis and a length-to-width (L/W) ratio to reduce false detections. The classification accuracies achieved across the two study sites ranged from 0.85 to 0.89, supporting the detection of approximately 70% of the total landslide area, with precision up to 0.93 and F1-scores of 0.77–0.79. Lower slope thresholds improved detection completeness, while higher thresholds increased omission errors. These findings suggest that the proposed integration of optimized LULC classification and constrained change detection may provide a practical and potentially transferable framework for improving landslide detection performance under varying environmental conditions.
Shukuru et al. (Thu,) studied this question.