Applying multi-site bias correction approach preserving inter-site correlation in the north-western Himalayan region

Due to unavailability of observation data in the north-western Himalayas, reanalysis data is used as an alternative. The reanalysis dataset generally have bias compared to observation data. Hence, different bias correction approaches are used to post-process the data before using it for any hydro-climatic study. Although distribution parameters in the bias correction approaches are adjusted according to observation data, it can modify or misrepresent dependence structure between variables and sites. Ignoring the observed inter-site dependencies in the correction procedure can result in obtaining corrected outputs with mismatched spatial dependence. Hence a multi-site bias correction approach is used with Schaake shuffle approach to reconstruct the inter-site correlation with rank reordering.

 In this study, we apply multivariate bias correction with schaake shuffle approach on 12 observatory stations of north-western Himalayas. The approach is applied on the variables mean temperature, precipitation, downward longwave radiation, downward shortwave radiation, and wind speed obtained from High Asia Reanalysis dataset at 0.25° horizontal resolution. The bias correction is applied using the observation data availed from the Princeton University global meteorological forcings for a time period of 2001 to 2011. The Leave-One-Out-Cross-Validation approach is used to apply the bias correction by leaving one year data for validation on each loop.

The multi-site bias correction applied to all the variables in different seasons shows that it is reducing the inter-station bias considerably for monsoon, winter, and summer seasons. For Post-monsoon season the improvement is not significant. For monsoon season the RMSE, MAE, and Bias percentage is decreasing for all the variables except precipitation. The inter-site correlation is improved after application of multi-site bias correction.