Estimating monthly rainfall in rural river basins under climate change: an improved bias-correcting statistical downscaling approach

Abstract. This study extended the work of Kim et al. (2008) to generate future rainfall under climate change using a discrete-time/space Markov chain based on historical conditional probabilities. A bias-correction method is proposed by fitting suitable statistical distributions to transform rainfall from the general circulation model (GCM) scale to watershed scale. The demonstration example used the Nam Ngum River Basin (NNRB) in Laos which is a rural river basin with high potential for hydropower generation and significant rain-fed agriculture supporting rural livelihoods. This work generated weekly rainfall for a 100 yr period using historical rainfall data from 1961 to 2000 for ten selected weather stations. The bias-correction method showed the ability to reduce bias of the mean values of GCMs when compared to the observed mean amount at each station. The simulated rainfall series is perturbed using the delta change estimated at each station to project future rainfall for the Special Report on Emission Scenarios (SRES) A2. GCMs consisting of third generation coupled general circulation model (CGCM3.1 T63) and European center Hamburg model (ECHAM5) projected an increasing trend of mean annual rainfall in the NNRB. Seasonal rainfall percent changes showed an increase in the wet and dry seasons with the highest increase in the dry season mean rainfall of about 31% from 2051 to 2090. While the GCM projections showed good results with appropriate bias corrections, the Providing REgional Climates for Impacts Studies (PRECIS) regional climate model significantly underestimated historical behavior and produced higher mean absolute errors compared to the corresponding GCM predictions.