Bias‐corrections on aridity index simulations of climate models by observational constraints

Abstract

AbstractAridity Index (AI) indicates the balance between water supply and water demand on the atmosphere–land interface. Despite continuous improvements, coupled climate models still have significant systematic errors in simulating AI in terms of temporal and spatial variabilities. One of the approaches to bias‐correct simulations is extracting the linear relationship between historical observations and model outputs by utilizing the empirical orthogonal function (EOF). In this study, the methodology of ensemble EOF‐based bias‐correction by observational constraints is developed based on previous bias‐correction approach, with the improvement on seeking the optimal combinations of the leading modes with sensitivity test and replacing the certain correction with the ensemble means of optimal members. In verification, the ensemble mean of Coupled Model Intercomparison Project phase 5 (CMIP5‐EM) is bias‐corrected towards the CPC/GLDAS observations, and the extracted leading modes present high correlations with internal climate variability. By cross‐validation and posteriori independent validation of hindcasts over the historical period (1948–2005), the ensemble EOF‐based bias‐correction could better present spatial patterns compared to the CMIP5‐EM after systematic bias‐correction, as indicated by the anomaly correlation and the root mean square error. The verifications also indicate that the temporal variability in aridity over different dryland regions is much closer to that in the observations and that the dryland subtype changes are improved significantly by bias‐corrections. Besides, another observational dataset of UDel/CRU is applied to assess the uncertainty on different datasets and the improvement on skill scores is robust. The above results verify that the ensemble EOF‐based bias‐corrections provide better reference for assessing and projecting global aridity changes by climate models.