Methodology:

We ran version 3.8.1 of the WRF model on a series of three nested domains. The innermost domain had 157 × 100 grid points with a 5 km spatial resolution, and included the Beas and Sutlej river basins. A Lambert conformal conic projection was used, centred on point 23°N, 80°E with one standard parallel at 30°N on a spherical model Earth of radius 6370 km. This was nested within a domain covering the whole of the wider Himalayan region with 124 × 100 grid points and a 15 km spatial resolution, and an outermost domain covering the whole of the Indian subcontinent with 100 × 100 grid points and a 45 km spatial resolution. All domains had 30 vertical levels between the surface and the model top at 50 hPa. The model was forced with atmospheric initial and boundary conditions derived from ERA-Interim reanalysis. The glacier mask used in all three domains was updated using version 5.0 of the Randolph Glacier Inventory. To deal with the complex terrain, the topography was derived from 90 m Shuttle Radar Topography Mission dataset. The model was run from continuously for one calendar year beginning on 15 December, reinitialising every year between 1980 and 2012 (i.e. a total of 33 years) in time zone UTC +5:30. This included a 16-day spin-up period. The model output from the innermost (5 km) domain was archived. We used a power transformation, first proposed by Leander and Buishand (2007), in order to correct the mean and variance statistics of the WRF precipitation output from the model to match the daily in-situ observations. Regression relationships were subsequently derived to allow us to apply a correction to all grid cells in the model innermost domain based on elevation.

Leander, R., and T. A. Buishand (2007), Resampling of regional climate model output for the simulation of extreme river flows, J. Hydrol., 332, 487-496, doi:10.1016/j.jhydrol.2006.08.006.

Data collection:

The model used is version 3.8.1 of the Weather Research and Forecasting (WRF) model.

Data quality:

Both the raw and bias-corrected precipitation output at a spatial resolution of 5 km were compared to precipitation measurements from seven independent sites. For annual as well as seasonal amounts, the bias based on the bias-corrected WRF output is (often considerably) smaller than the raw WRF model output.