Road Construction-Triggered Landslides: daily-resolution attribution and pre-construction suitability mapping in the Lali and Baku Gad watersheds of Himalaya, Far-Western Nepal

Abstract. Monsoon mountain landslide inventories face a trigger-attribution problem: failures caused by road construction can be credited to rainfall when the inventory is observed only quarterly or annually. In two adjacent watersheds in Darchula district, Far-Western Nepal where bulldozer roads began in 2017, we combined daily PlanetScope imagery (2017–2025), gauge and CHIRPS rainfall at each failure, daily road-construction records, and a quarterly inventory. Failures absent on the last clear image before a construction date and present on it were classed construction-coincident failure (CCF, n = 173 along about 69 km of road); failures mapped from 2016 imagery, before any road existed, gave a pre-road natural baseline (PRN, n = 127). Attribution proceeded by exclusion. The failures were not rainfall-driven: 63.6 % fell on dry days by gauge and 78.6 % by CHIRPS, with no rainfall correlation at any timescale. They tracked the construction front rather than the older road, 87.3 % lying within 50 m of the same-day front against 28.9 % of the old road, and they formed a distinct population, smaller and less elongate than the natural failures. One third never reached the inventory, and 60 % of those that did had failed on dry days, events a quarterly record would assign to the monsoon. Persistence was then modelled as a second-stage outcome: south-facing aspect and failure size were its dominant and most stable controls across penalised regression, Random Forest, and SHAP, and a four-variable pre-construction map placed about 90 % of persistent failures in terrain flagged unsuitable beforehand, holding on an independent watershed. Thus, daily imagery isolates a construction trigger that seasonal inventories conflate with rain, and yields a transferable basis for siting spoil before a road is cut.