Automated regression-based statistical downscaling tool

Abstract

Many impact studies require climate change information at a finer resolution than that provided by Global Climate Models (GCMs). In the last 10 years, downscaling techniques, both dynamical (i.e. Regional Climate Model) and statistical methods, have been developed to obtain fine resolution climate change scenarios. In this study, an automated statistical downscaling (ASD) regression-based approach inspired by the SDSM method (statistical downscaling model) developed by Wilby, R.L., Dawson, C.W., Barrow, E.M. [2002. SDSM – a decision support tool for the assessment of regional climate change impacts, Environmental Modelling and Software 17, 147–159] is presented and assessed to reconstruct the observed climate in eastern Canada based extremes as well as mean state. In the ASD model, automatic predictor selection methods are based on backward stepwise regression and partial correlation coefficients. The ASD model also gives the possibility to use ridge regression to alleviate the effect of the non-orthogonality of predictor vectors. Outputs from the first generation Canadian Coupled Global Climate Model (CGCM1) and the third version of the coupled global Hadley Centre Climate Model (HadCM3) are used to test this approach over the current period (i.e. 1961–1990), and compare results with observed temperature and precipitation from 10 meteorological stations of Environment Canada located in eastern Canada. All ASD and SDSM models, as these two models are evaluated and inter-compared, are calibrated using NCEP (National Center for Environmental Prediction) reanalysis data before the use of GCMs atmospheric fields as input variables. The results underline certain limitations to downscale the precipitation regime and its strength to downscale the temperature regime. When modeling precipitation, the most commonly combination of predictor variables were relative and specific humidity at 500 hPa, surface airflow strength, 850 hPa zonal velocity and 500 hPa geopotential height. For modeling temperature, mean sea level pressure, surface vorticity and 850 hPa geopotential height were the most dominant variables. To evaluate the performance of the statistical downscaling approach, several climatic and statistical indices were developed. Results indicate that the agreement of simulations with observations depends on the GCMs atmospheric variables used as “predictors” in the regression-based approach, and the performance of the statistical downscaling model varies for different stations and seasons. The comparison of SDSM and ASD models indicated that neither could perform well for all seasons and months. However, using different statistical downscaling models and multi-sources GCMs data can provide a better range of uncertainty for climatic and statistical indices.