Predictor Selection Associated With Statistical Downscaling of Precipitation over Zambia

Abstract

A non-generative, analog methodology was used to downscale daily precipitation from CMIP5-CNRM-CM5 developed by Météo-France/CNRS and CMIP5-CANESM2 of the Canadian Centre for Climate Modelling and Analysis. The downscaling reduces the 2° resolution GCM output to point station data. Sensitivity experiments for four different predictor variables (PVs) were carried out to examine the most significant PVs for the case of Zambia. ERA-Interim reanalyses was used for calibration (75%) and validation (25%) for the period 1981 – 2012. The Root Mean Square Error (RMSE) was used to compute the predictive power of CNRM-CM5 and CANESM2 by comparing the difference between their simulation results against ERA-Interim. Pearson correlation coefficient (r) was also used to assess the linear relationship between the datasets. Downscaled and observed data were compared and analysed. Results indicate that both CNRM-CM5 and CANESM2 perform well in perfect prognosis over the period 1970 – 2000 averaged over longitude 19°E - 37°E and latitude 22°S - 4°S. Pearson correlation results show that the combination PV2: T850, Q850, and U850 perform well at 95% confidence level. These results fill the knowledge gap of the behaviour of different variables for climate change projections and impact assessment studies in Zambia. Specifically, this study suggests a starting point in the selection of predictor variables for climate change studies in Zambia.