Statistical downscaling with Bayesian inference: Estimating global solar radiation from reanalysis and limited observed data

Daily global solar radiation (SR) is one of essential weather inputs for crop, hydrological, and other simulation models to calculate biomass production and potential evapotranspiration. The availability of long-term observed SR data is, however, limited, especially in developing countries. This hinders climate applications in various sectors in these countries. To overcome this difficulty, we here propose a method to infer the reasonable daily SR condition for past decades from global reanalysis and limited observed SR data. The method consists of the regression-based statistical downscaling method and two empirical models for estimating the SR condition (i.e. the S-model and the DTR/RH-model). These empirical models were independent in terms of the variables explaining the SR condition. The regression models were trained on the basis of the SR conditions estimated by the S-model and the DTR/RH-model instead of the observed SR data. The Markov Chain Monte Carlo (MCMC) technique was applied to determine the parameter values of these models that guide the models to provide SR conditions that are close in value to each other at both the site and domain-mean scales. After that, we computed the SR condition over the 30 years from 1978 through 2007 at 17 sites in the Vietnam Mekong Delta area using the determined parameter values. The inferred SR condition was close in value to the corresponding observations available from the literature. This suggests that the proposed method yielded a reasonable inference of the SR condition at the sites despite the limited availability of observed SR data. The provided estimates of the daily SR condition over the past 30 years are useful for climate applications in agricultural, hydrological, and other sectors in this area.