Bias correction of WRF output for operational avalanche forecasting in the Indian Himalayan region

Abstract

The forecast of snow avalanches in the Himalayan region has critical importance due to repetitive hazard scenarios and huge loss of property and lives. In recent years, avalanche forecast models have been developed using meteorological data collected from surface observatories (SO) at Defence Geoinformatics Research Establishment (erstwhile Snow and Avalanche Study Establishment (SASE) and now DGRE), India. For operational forecasting, outputs of a Weather Research Forecasting (WRF) model are used in real-time. The objective of this study is to determine a suitable bias correction approach for three key variables: mean temperature (T), relative humidity (RH), and wind speed (WS) which can be implemented in the operational forecast services at four observatory stations in Himachal Pradesh, India. We consider data of seven cold seasons (November–April) from 2011 to 2018 obtained from SO and WRF model output. Three quantile-based bias-correction approaches: Quantile Mapping (QM), Quantile–Quantile Mapping (QQ), and Quantile–Delta Mapping (QD), have been applied. In applying the QD method, two types of delta terms: multiplicative and additive, have been experimented according to the density distribution of the data. For evaluation, a leave-one-season-out-cross-validation approach is used. The results indicate that the QM and QD methods significantly reduced the bias associated with the mean temperature, relative humidity, and wind speed, whereas the performance of QQ method has limitations for all the variables. Furthermore, to meet the objective of generating bias-corrected data for operational forecasting services, uniform parameter set is generated for each variable associated with all the observatory stations. This approach is effective only for temperature, and for other variables, the improvements were not significant, mainly because of the high topographical variability in the Himalayan terrain.