How May the Choice of Downscaling Techniques and Meteorological Reference Observations Affect Future Hydroclimate Projections?

We present an intercomparison of a suite of high-resolution downscaled climate projections based on a six-member General Circulation Model (GCM) ensemble from Coupled Models Intercomparison Project (CMIP6). The CMIP6 GCMs have been downscaled using dynamical and statistical downscaling techniques based on two meteorological reference observations over the conterminous United States. We use the regional climate model, RegCM4, for dynamical downscaling, double bias correction constructed analogs method for statistical downscaling, and Daymet and Livneh datasets as the reference observations for statistical training and bias-correction. We evaluate the performances of downscaled data in both historical and future periods under the SSP585 scenario. While dynamical downscaling improves the simulation of some performance evaluation indices, it adds an extra bias in others, highlighting the need for statistical correction before its use in impact assessments. Downscaled datasets after bias-correction compare exceptionally well with observations. However, the choice of downscaling techniques and the underlying reference observations influence the hydroclimate characteristics of downscaled data. For instance, the statistical downscaling generally preserves the GCMs climate change signal but overestimates the frequency of hot extremes. Similarly, simulated future changes are sensitive to the choice of reference observations, particularly for precipitation extremes that exhibit a higher projected increase in the ensembles trained and/or corrected by Daymet than Livneh. Overall, these results demonstrate that multiple factors, including downscaling techniques and reference observations, can substantially influence the outcome of downscaled climate projections and stress the need for a comprehensive understanding of such method-based uncertainties.

Automated summary

We present an intercomparison of high-resolution downscaled climate projections based on a six-member GCM ensemble from CMIP6. The downscaled GCMs were generated using both dynamical and statistical downscaling techniques with two meteorological reference observations. We found that dynamical downscaling improves some performance indices, but introduces bias in others, highlighting the need for statistical correction. Downscaled datasets after bias-correction show good agreement with observations. However, the choice of downscaling techniques and reference observations influences the hydroclimate characteristics of the downscaled data.