Technical validation of a regional landslide forecasting system in the Eastern Himalayas

Rainfall-induced landslides pose a recurrent threat in the Darjeeling–Kalimpong Himalaya, India. The Geological Survey of India developed a prototype Regional Landslide Forecasting System (RLFS) that issued daily experimental forecast bulletins during the 2020–2023 monsoon seasons. Here we present the first systematic evaluation of an Indian regional landslide early warning system using the EDuMaP framework. Performance was assessed over 2,316 block-days spanning 373 monsoon days across 11 administrative units. Of 521 landslides analysed, 96.50% of block-days recorded no or only minor activity. Aggregated validation yielded 361 correct predictions, 1,862 true negatives, 19 missed alerts, and 74 false alerts. Green and yellow grades together represented 95.60% of forecast–observation pairs, with a mean combined efficiency index of 0.86 across the nine non-overlapping forecast zones. The system’s error structure was dominated by overwarning rather than catastrophic missed events. A stakeholder-driven transition to finer administrative forecast zones improved skill scores. Because the forecasts combined an automated rainfall-threshold model with expert review, the reported skill reflects this hybrid system, not the model alone.