A Nonparametric Statistical Technique for Spatial Downscaling of Precipitation over High Mountain Asia

Precipitation is considered the primary resource supporting the ecosystem services, agriculture, energy and livelihood of over one billion people in the High Mountain Asia (HMA) region. The accurate representation of precipitation sub-daily local-scale variability plays an important role in understanding the hydrological cycle and land-atmosphere interactions of the HMA region. Therefore, the development of hyper-resolution (1-km) precipitation data for HMA is of urgent need. In this study, we propose a framework to downscale the MERRA-2 (Modern-Era Retrospective analysis for Research and Applications, version 2) precipitation dataset based on a nonparametric statistical scheme using a tree-based classification and a regression algorithm. A set of variables representing the atmospheric condition, terrain topography and land cover are considered in the precipitation downscaling scheme. A heuristic type of algorithm that balances the model accuracy and complexity is applied to select important predictors to be used in the downscaling approach. The downscaled precipitation is validated against ground-gauge measurements in terms of rain magnitude and against satellite remote sensing products for spatial pattern consistency. Results indicate that the downscaled precipitation has higher correlation coefficients and Nash-Sutcliffe indices compared to the original coarse resolution MERRA-2 precipitation product. The downscaled precipitation also shows similar patterns with respect to several satellite precipitation products.