Comparison of transfer functions in statistical downscaling models for daily temperature and precipitation over Canada

Abstract

This study compares three linear models and one non-linear model, specifically multiple linear regression (MLR) with ordinary least squares (OLS) estimates, robust regression, ridge regression, and artificial neural networks (ANNs), to identify an appropriate transfer function in statistical downscaling (SD) models for the daily maximum and minimum temperatures (T max and T min ) and daily precipitation occurrence and amounts (P occ and P amount ). This comparison was made over twenty-five observation sites located in five different Canadian provinces (British Columbia, Saskatchewan, Manitoba, Ontario, and Québec). Reanalysis data were employed as atmospheric predictor variables of SD models. Predictors of linear transfer functions and ANN were selected by linear correlations coefficient and mutual information, respectively. For each downscaled case, annual and monthly models were developed and analysed. The monthly MLR, annual ANN, annual ANN, and annual MLR yielded the best performance for T max , T min , P occ and P amont according to the modified Akaike information criterion (AICu). A monthly MLR is recommended for the transfer functions of the four predictands because it can provide a better performance for the T max and as good performance as the annual MLR for the T min , P occ , and P amount . Furthermore, a monthly MLR can provide a slightly better performance than an annual MLR for extreme events. An annual MLR approach is also equivalently recommended for the transfer functions of the four predictands because it showed as good a performance as monthly MLR in spite of its mathematical simplicity. Robust and ridge regressions are not recommended because the data used in this study are not greatly affected by outlier data and multicollinearity problems. An annual ANN is recommended only for the T min , based on the best performance among the models in terms of both the RMSE and AICu.

Future extreme rainfall change projections in the north of Iran

“…Jeong et al . () compared ANNs with different transfer functions such as regression models for temperature downscaling and showed that multiple regression for monthly series performs better than an ANN except for the minimum temperature for which the ANN performs better.…”

Evaluation of Linear and Non-Linear Downscaling Methods in Terms of Daily Variability and Climate Indices: Surface Temperature in Southern Ontario and Quebec, Canada

“…In general MLP downscaling models give results similar to multiple linear regression downscaling methods for temperature and precipitation (Schoof & Pryor, 2001) and are capable of outscoring linear models when relationships are non-linear and/or interactive (Tang, Hsieh, Monahan, & Tangang, 2000). Nevertheless, there is no consensus on their performance versus linear models (Jeong, St-Hilaire, Ouarda, & Gachon, 2012b). For example, when downscaling temperatures over Europe, Huth et al (2008) concluded that non-linear methods were not an improvement over linear methods.…”

“…Because we are interested in downscaling temperatures in southern Ontario and Quebec, Canada, and we have the same reanalysis data available that Jeong et al (2012b) interpolated to the AOGCM (CGCM3.1) grid, it is natural for us to use the predictors recommended by Jeong et al (2012a). In particular, Jeong et al (2012a) mentioned that their predictor sets " … should be relevant to project anticipated predictand variables because they include sensitive predictors for future climate signal and variability such as temperature at 2 m, specific humidities at 500 hPa, 850 hPa and 1000 hPa, and 500 hPa and 850 hPa geopotential heights."…”

“…Consistent with our project framework and with Jeong et al (2012aJeong et al ( , 2012b, we used a standard set of 25 potential predictors from the NCEP/NCAR reanalysis interpolated to a T47 Gaussian grid (approximately 3.75°latitude by 3.75°longi-tude; i.e., the grid resolution of the Canadian CGCM3.1 (Flato et al, 2000)), over the current climate period . The NCEP/NCAR reanalysis outputs were averaged from 6-hourly to daily data before being linearly interpolated from the 2.5°× 2.5°grid to the CGCM3.1 T47 Gaussian grid (see further details in DAI (2008)).…”

Interpolation of daily rainfall using spatiotemporal models and clustering

“…The inclusion of altitude improved the predictions considerably. Jeong et al (2012) aimed at modelling daily temperatures and precipitation using covariates and historical data. They compared multiple regression models, ordinary least squares estimates, robust regression, ridge regression and artificial neural networks to identify an appropriate transfer function in statistical downscaling models to capture daily precipitation occurrence and amounts.…”