Spatial, Temporal, and Multivariate Bias in Regional Climate Model Simulations

Correction of atmospheric variables to remove systematic biases in global climate model (GCM) simulations before downscaling offers a means of improving climate simulation accuracy in climate change impact assessments. Various mathematical approaches have been used to correct the lateral and lower boundary conditions of regional climate models (RCMs). Most of these techniques correct only the magnitude of each variable individually over time without regard to spatial and multivariate bias. Here, we investigate how well an RCM is able to reproduce the dependence of an observed variable based on three aspects: temporal, spatial, and multivariate. Results show that the RCM simulations with univariate bias‐corrected GCM boundary conditions perform well in capturing both temporal and spatial dependence. However, all RCM simulations do not show improvement in the representation of dependence between variables, indicating the need for alternatives that correct systematic biases in multivariate dependence in both lateral and lower boundary conditions. Plain Language Summary It is well understood that systematic biases within a global climate model simulation can be passed into the input boundary condition of a regional climate model (RCM). To address this, many bias correction approaches have been used to correct the lateral and lower boundary conditions of RCMs, while assuming that each variable that makes up the lateral or lower boundary is independent. This study investigates the consequences of bias correction to assess whether the dependencies in time, space, and between variables are preserved. Using correlation length, it is shown that there is improvement in spatial and temporal dependence but not in inter‐variable dependence which can produce a mismatch in the physical relationships in the simulated outcomes. A more physically consistent approach that considers the relationships between the variables is needed instead of the simplistic univariate correction procedures that have been used to date. Key Points Regional climate model (RCM) simulations with bias‐corrected global climate model (GCM) boundary conditions better capture both temporal and spatial dependence of surface variables RCM simulations show reasonable performance for spatial dependence, implying it is the RCM that imparts the spatial relationships observed All RCM simulations do not show improvement in the dependence across variables