Development of multi-model ensembles using tree-based machine learning methods to assess the future renewable energy potential: case of the East Thrace, Turkey

Since investigating the long-term trends of the renewable energy potential may help in planning sustainable energy systems, this study intends to forecast the renewable energy potential of the East Thrace, Turkey region, in the future based on CMIP6 Global Circulation Models data using the ensemble mean output of the best-performed tree-based machine learning method. To evaluate the accuracy of global circulation models, Kling-Gupta efficiency, modified index of agreement, and normalized root-mean-square error are applied. The best four global circulation models are detected as a result of the comprehensive rating metric, which combines all accuracy performance results into a single metric. Three different machine learning methods, random forest, gradient boosting regression tree, and extreme gradient boosting, are trained using the historical data of the top-four global circulation models and the ERA5 dataset to calculate the multi-model ensembles of each climate variable, and then, the future trends of those variables are forecasted based on the output of ensemble means of best-performed machine learning methods with the lowest out-of-bag root-mean-square error. It is foreseen that there will not be a significant change in the wind power density. The annual average solar energy output potential is found to be between 237.8 and 240.7 kWh/m²/year depending on the shared socioeconomic pathway scenario. Under the forecasted precipitation scenarios, 356–362 l/m²/year of irrigation water could be harvested from agrivoltaic systems. Thereby, it would be possible to grow crops, generate electricity, and harvest rainwater on the same area. Furthermore, tree-based machine learning methods provide much lower error compared to simple mean methods.