Machine Learning Based Statistical Downscaling Approach for the Assessment of Climate Change Impact on Precipitation in Damaturu, Nigeria

Precipitation forecasts for the future are crucial for the efficient management of water resources. These forecasts are often made using global circulation models (GCMs). Therefore, this paper examined the impact of climate change on precipitation for Damaturu, Yobe State, Nigeria between 2050 and 2080 using GCM variables. To achieve this, an artificial neural network (ANN) was utilized to downscale observed precipitation data using the BNU-ESM GCMs under the emission scenario RCP 4.5. The mutual information (MI) technique was used to rank various climate predictors based on their influence on precipitation. In order to downscale the precipitation data, five distinct predictor combinations were used to create the ANN models. Next, the Root Mean Square Error (RMSE) and Determination Coefficient (DC) performance metrics were used to identify the best downscaling model. It was discovered that M1, which combined the top 8 ranked predictors, performed the best throughout both the projection and downscaling phases. The final M1 results indicated that a decrease in precipitation will likely be experienced in the Damaturu region within the given period. In the months with the highest amounts of precipitation, the decrease will be more pronounced with the greatest amount of up to 20% occurring during the wettest month of August, near the end of the twenty-first century.