Simulation of future changes in extreme rainfall and temperature conditions over the Greek area: A comparison of two statistical downscaling approaches

Abstract

An estimation of the future changes in the extreme conditions over the Greek area has been attempted, by applying the output data from a GCM to two different statistical downscaling models. The first was a multiple linear regression model based on a circulation type approach (MLRct) and the second one used artificial neural networks (ANNs). Two extreme temperature indices (txq90, tnq10) and two extreme precipitation indices (pq90, pxcdd) were selected for the study and simulated by the statistical models. From the comparison of the models during a validation period, it was found that the MLRct was relatively more efficient and the circulation types were proven to be a more reliable predictor. However, both the models underestimated the natural variability of the observed time-series. Generally, the extreme temperature indices were simulated better than the precipitation ones and winter was the season that showed the highest downscaling skill. Subsequently, the present day and future extreme temperature and rainfall conditions were examined using four ensemble members (control runs) and five scenarios driven by the A2 and B2 emission SRES scenarios, from the atmospheric general circulation model HadAM3P. Concerning extreme temperatures, an increase of their values is expected during the future period 2070–2100, according to the MLRct model. The climate change signal of the ANNs model was not so strong and in some cases it was of opposite direction. The results for the extreme precipitation indices showed a spatial incoherence and more complex structure of change.