Multi‐model ensemble approach for statistically downscaling general circulation model outputs to precipitation

Abstract

AbstractTwo statistical downscaling models were developed for downscaling monthly General Circulation Model (GCM) outputs to precipitation at a site in north‐western Victoria, Australia. The first downscaling model was calibrated and validated with the National Centers for Environmental Prediction/National Center for Atmospheric Research (NCEP/NCAR) reanalysis outputs over the periods of 1950–1989 and 1990–2010 respectively. The projections of precipitation into the future were produced by introducing the outputs of HadCM3, ECHAM5, GFDL2.0 and GFDL2.1, pertaining to A2 and B1 greenhouse gas emission scenarios to this downscaling model. In this model, the input data used in the development and future projections are not homogeneous, as they originate from two different sources. As a solution to this issue, the second downscaling model was developed and precipitation projections into the future were produced with a homogeneous set of inputs. To produce a homogeneous set of inputs to this model, regression relationships were formulated between the NCEP/NCAR reanalysis outputs and the twentieth‐century climate experiment outputs corresponding to the variables used in the first downscaling model obtained from the ensemble consisting of HadCM3, ECHAM5 and GFDL2.0. The outputs of these relationships pertaining to the periods of 1950–1989 and 1990–1999 were used for the calibration and validation of this downscaling model respectively. Using the outputs of HadCM3, ECHAM5 and GFDL2.0 pertaining to A2 and B1 emission scenarios on these relationships, inputs for the second downscaling model pertaining to the period of 2000–2099 were generated. The first downscaling model with NCEP/NCAR reanalysis outputs showed a high Nash–Sutcliffe Efficiency (NSE) of 0.75 over the period 1950–1999. When this downscaling model was run with the twentieth‐century climate experiment outputs of HadCM3, ECHAM5, GFDL2.0 and GFDL2.1, it exhibited limited performances over the period 1950–1999, which was indicated by relatively low NSEs of −0.62, −2.54, −0.40 and −0.48 respectively. The second downscaling model displayed an NSE of 0.35 over the period 1950–1999.