Novel application of a process convolution approach for calibrating output from numerical models

Output from numerical models at high spatial and temporal resolutions is critical for modeling applications in a variety of disciplines. Prior to its use in modeling, output from climate models must be brought to a finer spatial resolution and calibrated with respect to observations. The calibration of model output, referred to as bias-correction, poses many statistical challenges. Here, we develop a bias-correction method in which systematic biases in the mean and standard deviation of model output are corrected. In addition, we employ a novel process convolution approach to correct bias in temporal dependence. We apply this approach to temperature simulations generated by a regional climate model over the Northeastern USA. The goal of this study was to correct systematic bias in model simulations over historical (1976–2005) and future (2006–2099) time periods while simultaneously preserving future trends resulting from carbon emissions scenarios. We compare the proposed method to a quantile mapping method (empirical quantile mapping, EQM). The proposed method resulted in a more effective correction of seasonal biases and temporal dependence compared to EQM.