Fidelity assessment of general circulation model simulated precipitation and temperature over Pakistan using a feature selection method

Abstract

General Circulation Models (GCMs) provide vital information on the likely future climate, much needed for the effective planning and management of water resources. The performance assessment of GCMs has received significant attention in recent years for reliable estimation of future climate. Even though many approaches have been trialled in the ranking of GCMs for the selection of an appropriate ensemble, there is still a need to explore the potential of the state-of-the-art ranking approaches for a more dependable selection of GCMs suitable for a given task, over a region of interest. The present study assessed the potential of a state-of-the-art feature selection method known as Symmetrical Uncertainty (SU) in ranking 20 Coupled Model Intercomparison Project Phase 5 (CMIP5) GCMs based on their ability to simulate monthly precipitation and the monthly average of daily maximum and minimum temperature for annual, monsoon and winter seasons. The performance of GCMs was assessed using gridded climate data obtained from Global Precipitation Climatology Centre (GPCC), Climatic Research Unit (CRU) and Princeton Global Meteorological Forcing dataset (Prin) over the period 1961–2005 considering Pakistan as the study area. The ranks obtained with SU were compared with those obtained using two well-established ranking approaches; (1) Compromise Programming (CP) and (2) Wavelet-based Skill Score (WSS). According to the results of this study, for the simulation of all the three climate variables in all seasons; CESM1-CAM5, HadGEM2-AO, NorESM1-M and HadGEM2-ES were identified as the best GCMs by SU, whereas CESM1-CAM5, HadGEM2-AO, NorESM1-M and GFDL-CM3 were identified as the best GCMs by CP, and CCSM4, CESM1-CAM5, GFDL-ESM2G and HadGEM2-ES by WSS. The comparison of ranks of GCMs obtained using the same ranking approach but based on different gridded data products showed more or less similar ranks for a given GCM. However, the differences were noticeable when the ranking was conducted with the same gridded data but employing different ranking approaches. The approach presented in this study can be extended to any number of GCMs and can be applied over any region, for the identification of the best performing ensemble of GCMs for a set of climate variables.