A Novel Spatial Downscaling Approach for Climate Change Assessment in Regions With Sparse Ground Data Networks

This study proposes a novel approach that expands the existing QDM (quantile delta mapping) to address spatial bias, using Kriging within a Bayesian framework to assess the impact of using a point reference field. Our focus here is to spatially downscale daily rainfall sequences simulated by regional climate models (RCMs), coupled to the proposed QDM‐spatial bias‐correction, in which the distribution parameters are first interpolated onto a fine grid (rather than the observed daily rainfall). The proposed model is validated through a cross‐validatory (CV) evaluation using rainfall data from a set of weather stations in South Korea and climate change scenarios simulated by three alternate RCMs. The results demonstrate the efficacy of the proposed model to simulate the bias‐corrected daily rainfall sequences over large regions at fine resolutions. A discussion of the potential use of the proposed approach in the field of hydrometeorology is also offered. Plain Language Summary Climate models can simulate biased representations of atmospheric processes, necessitating procedures for correction before use in hydrological applications. Such spatial bias can be caused for many reasons, one of which is the use of point data in establishing a spatial reference field to compare model simulations against. The most straightforward way to address this bias is to interpolate the locally observed data at the weather station onto a fine grid and use as a reference. Alternatively, one can define a bias‐correction model that accounts for the systematic impact induced by the use of point data, of special importance when the point data field is sparse and unevenly distributed. Here, we develop a novel approach to better address spatial bias using the Bayesian Kriging model. The results demonstrate the efficacy of the proposed model to simulate the bias‐corrected daily rainfall sequences over large regions at fine resolutions. Key Points All parameters for the spatial downscaling and bias‐correction can be simultaneously estimated and interpolated at the desired points The parameter interpolation is more effective than that of the precipitation in the context of spatial downscaling and bias‐correction The Bayesian Kriging based SD‐QDM can reproduce spatial dependency in the interpolated parameters associated with bias‐correction