Assessing the Need for Downscaling RCM Data for Hydrologic Impact Study

Abstract

Climate change impact studies have generally downscaled large-scale global climate model (GCM) output data; however, few studies have considered downscaling regional climate model (RCM) data. It is unclear whether further downscaling raw RCM data could be beneficial or not in a hydrologic impact study. This study provides some experimental results to address that question. Raw Canadian regional climate model (CRCM4.2) data are downscaled by using a common statistical downscaling method (SDSM) and a data-driven technique called a time-lagged feedforward network (TLFN). Regardless of the downscaling methods and the predictands (e.g., precipitation, temperature), the downscaled CRCM4.2 data are found to be much closer to the observed data than the raw CRCM4.2 data. When the downscaled CRCM4.2 data are used in a hydrologic model (HBV), the model’s ability to accurately simulate streamflow and reservoir inflow is significantly improved as compared to the use of the raw CRCM4.2 data. Simulations of future river flow and reservoir inflow reveal that the general patterns of changes in future flow are quite similar whether downscaled or raw CRCM4.2 data are used. However, the use of downscaled CRCM4.2 data seems to provide more consistent predictions of the magnitude and timing of changes. It appears that the RCM may still suffer from the bias problem inherent to the parent GCM. Further downscaling raw RCM data permits bias correction, improves hydrologic modeling, and provides more consistent changes (magnitude and timing) of future flows.