Using ICAR and En-GARD to understand future climate variability of the Colorado River Basin

Abstract:

Within a modeling chain that takes Global Climate Models (GCM) to evaluate regional scale climate change projections in streamflow, evapotranspiration, and snowpack, downscaling coarse resolution GCMs to produce regional scale temperature and precipitation distributions is a significant step. This study applied two recently developed statistical and dynamical downscaling methods over the Colorado River Basin to investigate effects of downscaling methods on regional climate and streamflow ensembles. One method was a statistical downscaling technique called the Ensemble Generalized Analog Regression Downscaling (En-GARD) tool, which incorporates upper atmospheric variables (e.g., 700 mb meridional and zonal winds) and surface variables (e.g., precipitation and temperature). In the historical period, En-GARD was relatively unbiased and captured the distributions of precipitation and temperature compared to gridded observations. En-GARD future climate projections showed precipitation and temperature changes similar to previously used statistical downscaling methods, such as the Localized Constructed Analog (LOCA) and Bias Corrected Spatial Disaggregation (BCSD) datasets. However, while the BCSD ensemble mean generally gets wetter and warmer over a majority of the Colorado River Basin, LOCA and En-GARD show the lower Colorado River Basin becoming drier and were less sensitive to enhanced warming at higher elevations from the snow albedo feedback within the GCMs. Additionally, we dynamically downscaled GCMs using the Intermediate Complexity Atmospheric Research (ICAR) model. ICAR simulations during the historical period captured inter-annual variability of precipitation and temperature and showed similar distributions to WRF and gridded observations. Due to ICAR’s computational efficiency, we were able to dynamically downscale an ensemble of GCMs from CMIP5. We will show comparisons between dynamically downscaled ICAR simulations and statistically downscaled methods (En-GARD, LOCA, and BCSD). Lastly, we will show streamflow changes using the dynamical and statistically downscaled projections by running projections through a calibrated Variable Infiltration Capacity model.