Prediction of Rainfall Response to the 21st-century Climate Change in Ghana using Machine Learning Empirical Statistical Downscaling

The far-reaching and high spatial variability of climate change dictates the focus on the generation of high-resolution regional and local scale future estimates to help design adaptation measures, vulnerability assessments, and resilience strategies. Developing countries such as Ghana have been tagged with high vulnerability due to their low adaption capacity and the significant impacts of extreme climate events. Moreover, water availability and flood risk are of major concern due to the country's over-dependence on rain-fed agriculture, and hydroelectric power generation. Future trends of such related climatic variables are normally simulated with General Circulation Models (GCMs) or Regional Climate Models (RCMs) with prescribed anthropogenic emission scenarios such as the Representative Concentration Pathways (RCPs) that account for plausible evolution of demographics and technological developments. However, such models have coarse resolution and are affected by systematic biases leading to the inaccurate projection of rainfall patterns on the relevant spatial scale for climate change impacts studies. Therefore, to circumvent their shortcomings, we use Empirical Statistical Downscaling (ESD) to construct a transfer function between the observed large-scale atmospheric patterns (ERA5 reanalysis) and local rainfall measurements at weather stations across Ghana. The models are trained from 1979-2010 and independently validated from 2011-2020. The novelty of our approach is that we experiment with different machine learning algorithms (eg. Artificial Neural Networks (ANN), Support Vector Machines (SVM), and Ensemble methods) to ensure the robust establishment of the empirical transfer function for future predictions. The calibrated models are coupled with GCM simulations to predict station-based future rainfall estimates under the different RCPs (2.6, 4.5, and 8.5). Our modeling approach also contributes to the understanding of the influence of large-scale atmospheric dynamics on the seasonality of future rainfall variability. Policymakers can use the newly established high-resolution climate change products to design efficient climate change adaptation strategies for water resource management, agriculture, and socio-ecological systems.