Assessing the Influence of a Bias Correction Method on Future Climate Scenarios Using SWAT as an Impact Model Indicator
Precipitation and other outputs generated by global climate models (GCMs) and regional climate models (RCMs) can be biased due to several factors, such as imperfect model conceptualization and discretization. Different methods exist to correct biased climate-model output used as input to hydrological models, including simulations using GCM or RCM output. However, considerable debate has been expressed in the literature as to when bias correction should be used and what methods are the most reliable. In addition, there are implications of large bias corrections relative to smaller projected future climate changes, including for GCM projections that are dynamically downscaled with RCMs. These bias correction issues also have implications for other models interfaced with climate model projections, such as ecohydrological models.
This study was performed with four GCM-RCM combinations, which were used to drive the Soil and Water Assessment Tool (SWAT) ecohydrological model for both historical simulations (1981–2005) and future projections (2030–2050) for the Des Moines River Basin (DMRB) in the north-central U.S. Historical precipitation levels were greatly overestimated by WRF RCM raw precipitation amounts, when interfaced with both the MPI-ESM-LR and GFDL-ESM2M GCMs. The interface of the same two GCMs with the RegCM4 RCM also resulted in overestimated raw precipitation, but to a lesser extent. The historical RCM precipitation outputs were greatly improved by applying the Distribution Mapping (DM) bias correction method. The improved bias-corrected precipitation outputs also resulted in improved historical SWAT streamflow estimates. However, the projected future raw precipitation data resulted in increased SWAT surface runoff and water yield, but the bias-corrected data projected a reduction in these indicators; i.e., the bias correction was often as large or larger than the projected change in precipitation.
Bias correction resulted in improved historical precipitation, including annual volumes, seasonality, spatial distribution, and mean error. The bias-corrected historical precipitation also resulted in improved SWAT-predicted monthly streamflow across 26 DMRB monitoring stations. However, the bias-corrected precipitation data led to changes in future prediction signals, e.g., an increase in SWAT estimates of surface runoff and water yield in response to raw precipitation inputs versus opposite results due to biased-corrected precipitation. In our analysis, the bias correction was often larger than the climate-change signal, indicating that the procedure is not a small correction but a major factor.