Bias nonstationarity of global climate model outputs: The role of internal climate variability and climate model sensitivity

Abstract

Bias correction methods are developed based on the assumption that the biases of climate model outputs are stationary, that is, the characteristics of the bias are constant over time. However, recent studies have shown the biases are not always stationary. The objectives of this study are to investigate the impacts of bias nonstationarity of climate-model-simulated precipitation and temperature on future climate projections, and the roles of internal climate variability (ICV) and climate model sensitivity (CMS) in bias nonstationarity. A pseudoreality approach is used in this study, in which each of the 24 climate model simulations is alternately selected as a reference to estimate the biases (defined as pseudobias to distinguish it from actual bias estimated by observations) of 23 other simulations. The absolute ratio of the change in pseudobias between two periods to the corresponding climate change signal is calculated to assess the impacts of bias nonstationarity on future climate projections. Furthermore, the roles of ICV and CMS are investigated by comparing the changes in pseudobias between historical and future periods relative to the baseline period. The results show that biases of climate-model-simulated mean annual and seasonal temperature and precipitation vary with time. Bias nonstationarity of temperature is not significant in future temperature projections, while the bias nonstationarity of precipitation plays an important role in future precipitation projections. In addition, the contributions of ICV and CMS to bias nonstationarity are both relatively small for temperature, even though the latter contributes slightly more than the former. However, ICV makes a large contribution to the bias nonstationarity of precipitation for the historical period. In the far future period, the role of CMS is as important as ICV. These results imply that the impacts of ICV and CMS may need to be considered when developing and evaluating a bias correction method, especially for precipitation projections.

A Framework to Quantify the Uncertainty Contribution of GCMs Over Multiple Sources in Hydrological Impacts of Climate Change

“…For example, Ahmadalipour et al, (2017) ranked 20 GCMs based on their skills in simulating multiple aspects of daily observations in Columbia River Basin, which could provide an objective reference for selecting more reliable GCMs for regional climate change impact assessment. However, the ranks of GCMs may be altered due to their nonstationary performances in simulating climates (Hui et al, 2019). Another type of approaches seeks to preserve the uncertainty of climate change signals related to the choice of GCMs (Cannon, 2015; Warszawski et al, 2014).…”

Impacts of Using State‐of‐the‐Art Multivariate Bias Correction Methods on Hydrological Modeling Over North America

“…This may be because the bias of intervariable correlation simulated by climate models is not stationary, and the observed intervariable correlation itself is also not invariable. Previous studies (Chen et al, 2017; Hui et al, 2018; Maraun, 2012) have shown that bias correction methods can deteriorate the original climate simulations when bias directions are different between future and historical periods (or calibration and validation periods) or when future biases reduce to less than half the calibration biases. The nonstationarity of intervariable correlation bias of climate models can be observed in Figure S7.…”

Downscaling climate change of mean climatology and extremes of precipitation and temperature: Application to a Mediterranean climate basin

“…Climate model precipitation biases are not always stationary. Internal climate variability is a substantial source of such bias nonstationarity (Hui et al, ). It was ignored in most climate change impact studies, which resulted in large underestimation of uncertainty.…”