Is Bias Correction in Dynamical Downscaling Defensible?
Fraction of total variability (%) over time in western-U.S.-wide averages of selected variables due to model choice (blue), the effect of a priori bias correction (red), and internal variability (orange). Note that a minimum SWE mask is applied before taking Western U.S. averages for spring snow cover fraction and annual peak SWE.
Mark Risser
The potential for bias correction (BC) to distort climate trends has long been acknowledged. In statistical downscaling, strange artifacts in climate trends may ultimately present that lack physicality. In dynamical downscaling circles, however, BC of the GCM boundary conditions, specifically its potential to distort climate trends, has not been explored. This potential is examined here using a classic variance decomposition. Building on other recent work documenting that a simple delta BC method improves dynamically downscaled hydroclimate states, we show that BC also minimally distorts precipitation and temperature trends through 2100 when compared to other irreducible climate change uncertainties. For snow, however, trends are significantly distorted, calling into question the fitness for the purpose of the original, non-bias-corrected snow projections, rather than the bias-corrected downscaled versions.
Propagated GCM biases into regional climate models lead to an overly wet solution (as much as 60% in the western U.S. historical mean relative to observations). This results in an overly snowy and cold solution, particularly across the intermountain West. As a result, the snow cover fraction, which governs the intensity of the snow-albedo feedback, remains artificially high throughout the future period despite snow water declines. BC implemented before downscaling yields a more skillful historical representation of snow water and fraction, leading to a more realistic expression of the SAF by the mid 21st century due to transient warming.
BC of the GCM boundary conditions is defensible, as it distorts only state-dependent fields such as snow water. However, trend distortion occurs because the historical era is more skillfully represented in the bias-corrected projections, so is it fair to call this a distortion, especially given that trends in precipitation and temperature are preserved at regional scales after BC?