Use of artificial neural network for generation of ultra-high resolution meteorological fields over Chowkibal-Tangdhar region of North-West Himalaya
In one sentence. Neural networks post-process 2 km WRF output over the NW Himalaya and interpolate it to 90 m, with NSE above 0.75 at five stations.
Abstract
Abstract High resolution meteorological fields have been found potentially useful for hydrological and agricultural applications and mitigation of hydro-meteorological hazards such as landslide and snow avalanches. These meteorological fields are being generated for Himalaya using Weather Research and Forecast (WRF) model with spatial resolution up to 3 km. However, gridded meteorological data of sub-kilometer resolution is not available for the Himalayan region. In the present study, Numerical Weather Prediction model-WRF has been configured for North-West (N-W) Himalayan region with spatial resolution of 2 km and run in hind cast mode to generate meteorological data of 11 winters (2009-19). Artificial neural networks (ANNs) have been developed for post-processing of maximum temperature, minimum temperature, wind speed, relative humidity, snow depth and snowfall in 24h generated by the WRF model using observed surface weather data of five different locations in Chowkibal-Tangdhar (C-T) region. Post-processed WRF output has been spatially interpolated to a grid resolution of 90 m using quasi-physical relations and inverse distance weighing scheme. The ultra-high resolution meteorological fields generated over the C-T domain have been validated at five locations in the C-T region for two winters (2017-19). For all five stations, the Nash-Sutcliffe Efficiency (NSE) scores of the model for maximum and minimum temperature, relative humidity and snow depth has been found at “very good” level (> 0.75) with considerably high Heidke Skill Score (HSS) (> 0.4). A comparison of observed and simulated cumulative snowfall during major snow storms during 2017-19 has also been discussed.
Labels
- Topic
- Post-processing & calibration, Regional & km-scale models
- Type
- Application study
- Method
- Classical ML (trees, GP, linear, kernels)
- Time scale
- 6 h–3 days
- Domain
- Regional, Local / station
- Data
- NWP analysis or forecasts (IFS, GFS, ...), Station / in-situ
- Status
- Research
- Identifiers
- DOI:10.21203/rs.3.rs-1339694/v1 OpenAlex:W4214883334
- Added
- 2026-10-22
Labels and summary are generated by Claude and reviewed weekly. Wrong? Edit this entry.