Spatiotemporal Exposure to Heavy-Day Rainfall in the Western Himalaya Mapped with Remote Sensing, GIS, and Deep Learning

Heavy rainfall events, characterized by extreme downpours that exceed 100 mm per day,
pose an intensifying hazard to the densely settled valleys of the western Himalaya; however,
their coupling with expanding urban land cover remains under-quantified. This
study mapped the spatiotemporal exposure of built-up areas to heavy-day rainfall (HDR)
across Jammu, Kashmir, and Ladakh and the adjoining areas by integrating daily Climate
Hazards Group InfraRed Precipitation with Stations product (CHIRPS) precipitation (0.05◦)
with Global Human Settlement Layer (GHSL) built-up fractions within the Google Earth
Engine (GEE). Given the limited sub-hourly observations, a daily threshold of ≥100 mm
was adopted as a proxy for HDR, with sensitivity evaluated at alternative thresholds. The
results showed that HDR is strongly clustered along the Kashmir Valley and the Pir Panjal
flank, as demonstrated by the mean annual count of threshold-exceeding pixels increasing
from 12 yr−1 (2000–2010) to 18 yr−1 (2011–2020), with two pixel-scale hotspots recurring
southwest of Srinagar and near Baramulla regions. The cumulative high-intensity areas
covered 31,555.26 km2, whereas 37,897.04 km2 of adjacent terrain registered no HDR events.
Within this hazard belt, the exposed built-up area increased from 45 km2 in 2000 to 72 km2
in 2020, totaling 828 km2. The years with the most expansive rainfall footprints, 344 km2
(2010), 520 km2 (2012), and 650 km2 (2014), coincided with heavy Western Disturbances
(WDs) and locally vigorous convection, producing the largest exposure increments. We also
performed a forecast using a univariate long short-term memory (LSTM), outperforming
Autoregressive Integrated Moving Average (ARIMA) and linear baselines on a 2017–2020
holdout (Root Mean Square Error, RMSE 0.82 km2; measure of errors, MAE 0.65 km2;
R2 0.89), projecting the annual built-up area intersecting HDR to increase from ~320 km2
(2021) to ~420 km2 (2030); 95% prediction intervals widened from ±6 to ±11 km2 and
remained above the historical median (~70 km2). In the absence of a long-term increase in total annual precipitation, the projected rise most likely reflects continued urban encroachment into recurrent high-intensity zones. The resulting spatial masks and exposure
trajectories provide operational evidence to guide zoning, drainage design, and early
warning protocols in the region.