Comparison of statistically downscaled precipitation in terms of future climate indices and daily variability for southern Ontario and Quebec, Canada

Given the coarse resolution of global climate models, downscaling techniques are often needed to generate finer scale projections of variables affected by local-scale processes such as precipitation. However, classical statistical downscaling experiments for future climate rely on the time-invariance assumption as one cannot know the true change in the variable of interest, nor validate the models with data not yet observed. Our experimental setup involves using the Canadian regional climate model (CRCM) outputs as pseudo-observations to estimate model performance in the context of future climate projections by replacing historical and future observations with model simulations from the CRCM, nested within the domain of the Canadian global climate model (CGCM). In particular, we evaluated statistically downscaled daily precipitation time series in terms of the Peirce skill score, mean absolute errors, and climate indices. Specifically, we used a variety of linear and nonlinear methods such as artificial neural networks (ANN), decision trees and ensembles, multiple linear regression, and k-nearest neighbors to generate present and future daily precipitation occurrences and amounts. We obtained the predictors from the CGCM 3.1 20C3M (1971-2000) and A2 (2041-2070) simulations, and precipitation outputs from the CRCM 4.2 (forced with the CGCM 3.1 boundary conditions) as predictands. Overall, ANN models and tree ensembles outscored the linear models and simple nonlinear models in terms of precipitation occurrences, without performance deteriorating in future climate. In contrast, for the precipitation amounts and related climate indices, the performance of downscaling models deteriorated in future climate.