Intercomparison of bias-correction data sources and their influence on watershed-specific downscaling climate projections

Authors: SHESHUKOV, ALEKSEY - Kansas State University; GAO, JUNGANG - Texas A&M University; Mankin, Kyle; YEN, HAW - Texas A&M University

Submitted to: Transactions of the ASABE
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 10/15/2020
Publication Date: 2/10/2021
Citation: Sheshukov, A.Y., Gao, J., Douglas-Mankin, K.R., Yen, H. 2021. Intercomparison of bias-correction data sources and their influence on watershed-specific downscaling climate projections. Transactions of the ASABE. 64(1):203-220. https://doi.org/10.13031/trans.14061.
DOI: https://doi.org/10.13031/trans.14061

Interpretive Summary: Climate models project future rainfall and temperature which are used by hydrologic models to project changes to soil water, streamflow, and other important hydrologic responses. However, climate model projections are coarse and must be downscaled to the fine detail needed by hydrologic models. This downscaling process can impose uncertainties in addition to the uncertainties from the climate models themselves (called general circulation models or GCMs), from the different scenarios that represent human emissions (called representative concentration pathways or RCPs). This study evaluated these effects using three sources of historical data used to downscale the climate projections (ground-based weather station network [NCDC], gridded radar estimates [NEXRAD], and gridded model estimates [PRISM]), six different GCMs, and three different RCPs in the Smoky Hill River Watershed in U.S. Central Great Plains. Downscaling using any of the historical data sources corrected the inherent errors in comparing the raw climate model results to historical rainfall and temperature data. However, downscaling increased variability in future projections, and this occurred regardless of bias-correction data source. For minimum and maximum temperatures, mean and variability of future projections was similar regardless of historical data source. For rainfall, selection of the data source used to downscale the climate projections added less uncertainty than selection of GCM or RCP scenario, but the uncertainty that was added varied greatly among different GCMs. The results showed that future climate projections inherit bias from the selected historical data source used to downscale the model projections.

Technical Abstract: Climate projections developed by General Circulation Models (GCM) are used for projection of future hydrologic changes. In watershed modeling applications, the projections are downscaled to individual map units often represented by subbasins. Uncertainty of downscaled climate projections are a product of uncertainties arising from the model itself, from the representative concentration pathway (RCP), and from the downscaling procedure. Other sources of uncertainty may include the historical observations used for GCM bias correction and data aggregation from GCM grids to subbasin units. This study evaluated effects of three sources of historical data (ground-based weather station network, NCDC, and two gridded datasets, NEXRAD and PRISM) on shifts, variability, and uncertainty in precipitation and temperature projections from six GCMs and three RCPs in 54 subbasins of Smoky Hill River Watershed in U.S. Central Great Plains. Bias correction reduced mean bias of annual precipitation in the historical GCM period to near zero but downscaling increased variability in future projections, and this occurred regardless of bias-correction data source. For minimum and maximum temperatures, mean and variability of future projections was similar regardless of bias-correction data source when GCM ensemble was considered but the uncertainty increased with the increase in RCP forcing. Uncertainty to data source selection was smaller than two other components of uncertainty but varied adversely with the selection of individual GCM. The results demonstrate that statistical downscaling is essential to accounting for locality and spatial variability within a watershed, and future climate projections may inherit the historical bias in a selected data source presented at the subbasin scale. Ensembles of multiple GCMs when bias-corrected to local scale with the historical data source are shown to carry forward the same signal in future change.