Climate change impacts on catchment-scale extreme rainfall variability

Abstract

A changing climate alters regional hydrology and thereby virtually affects every aspect of human life. The local impacts of climate change may significantly vary in different locations even within the borders of a country. Turkey, as a part of the Mediterranean region, is located in one of the most vulnerable regions to climate change. In order to figure out the local impacts of climate change on extreme rainfalls across Rize Province, located in the north east of Turkey, this doctoral thesis provides a grid-based performance evaluation of two general circulation models (GCMs) for rainfall reproduction over the province and investigates the variability of extreme rainfall events at three future periods (i.e., 2013-2040; 2041-2070; and 2071-2099) with respect to high emission/concentration greenhouse gas scenarios. In addition, a data-driven projected rainfall-runoff model is provided for projected streamflow prediction concerning the limited data availability in the province. Prior to hydrological assessment of climate change at catchment-scale, an applied methodology is necessary to evaluate the performance of climate models available for a given catchment. Therefor, this thesis, in the first phase, presents a grid-based performance evaluation approach as well as an intercomparison framework to evaluate the uncertainty of climate models for rainfall reproduction over Rize Province. For this purpose, outputs of two general circulation models (GCMs), namely ECHAM5 and CCSM3, downscaled by a regional climate model (RCM), namely RegCM3, are used. To this end, five rainfall-borne climatic statistics are computed from the outputs of ECHAM5-RegCM3 and CCSM3-RegCM3 combinations in order to compare with those of observations in the province for the reference period 1961-1990. Performance of each combination is tested by means of scatter diagram, bias, mean absolute bias, root mean squared error, and model performance index (MPI) measures. The results of this phase of study indicated that ECHAM5-RegCM3 overestimates the total monthly rainfall observations whereas CCSM3-RegCM3 tends to underestimate. In terms of maximum monthly and annual maximum rainfall reproduction, ECHAM5-RegCM3 shows higher performance than CCSM3-RegCM3, particularly in the coastland areas. In contrast, CCSM3-RegCM3 outperforms ECHAM5-RegCM3 in reproducing the number of rainy days, especially in the inland areas. The results also revealed that if a GCM-RCM combination performs well for a portion (statistic) of a catchment, it is not necessarily appropriate for the other portions (statistics). Moreover, the MPI measure demonstrated the superiority of ECHAM5-RegCM3 to CCSM3-RegCM3 up to 33% excelling for annual rainfall reproduction in Rize Province. In the second phase of the thesis, a grid-based methodology is developed and applied to estimate future extreme rainfalls over the province, compatible with adaptive catchment-scale water resources management. Extreme value theory (EVT) is applied to analyse observational and projected extreme rainfall data including dynamically downscaled climate models simulations from two different GCM-RCM combinations including ECHAM5-RegCM3 and HadGEM2-RegCM4 forced by A2 special report on emissions scenarios (SRES-A2) and representative concentration pathway-8.5 (RCP8.5) greenhouse gas scenarios. A new rapid and effective bias correction method is also developed and applied to adjust the climate models simulations. The EVT analysis results demonstrated significant differences between the model runs for both reference and future periods with considerable spatial variability in rainfall extremes. Based upon ensemble mean results, approximately 30% decrease in the median value of extreme rainfall events is projected over the study region for the near future 2013-2039 and middle of the century. This change dramatically decreases down to 15% of its historic value at the end of the century. Results from the implemented scenarios revealed that more intense rainfalls produced by the GCM forced by RCP8.5. In addition, the results from goodness-of-fit tests among different distribution models showed that general extreme value distribution can be used appropriately to characterize the behavior of projected extreme rainfalls over the study region. It would be very helpful to hydrologists if data-driven technology incorporating the existing hydrologic observations and forthcoming climatological projections can be integrated into the catchment-scale hydrological modelling. In the last phase of the study, This thesis delineates the development and implementation of two advancements of state of the art genetic programming (GP) approach, namely linear GP (LGP) and multigene GP (MGGP), and an artificial neural network method, namely feed-forward backpropagation (FFBP), for projected rainfall-runoff (PR-R) modelling at one of the catchments in Rize Province by employing observational streamflow together with projected rainfall and temperature data as model inputs. The proposed models have been evaluated in terms of five efficiency criteria including Nash-Sutcliffe efficiency, persistence index, Kling-Gupta efficiency, volumetric efficiency, and root mean squared error. The results show promising role of GP-based models for both bivariate and multivariate PR-R modelling. An explicit bivariate LGP model exhibited the best overall behaviour for the study region, closely followed by MGGP. The lagged prediction has also been realized as the main drawback of the implemented FFBP algorithm.