Detailed analysis of Türkiye's agricultural biomass-based energy potential with machine learning algorithms based on environmental and climatic conditions

Abstract

AbstractIn the study, the biomass and energy potential of each province of Türkiye was calculated for the years 2010–2021, using data from 15 different fields and 16 different horticultural crops. The total theoretical energy potential obtained from field and garden products was calculated as 222,620 Terajoule (TJ) and 61,737 TJ for 2010 and 308,888 TJ and 77,002 TJ for 2021, respectively. The agricultural biomass potential for 2021 was estimated using machine learning algorithms, depending on the environmental and climate data covering 2010–2020, which has not been studied in the literature. In this study, agricultural biomass potential for Türkiye was tried to be modeled by using Random Forest, K-Nearest Neighbors (KNN), Gradient Boosting, and eXtreme Gradient Boosting Regressor (XGBR) from machine learning methods. Agricultural biomass potential was tested in a tenfold cross-validation analysis and prediction for 2021 using only climatic and agricultural area data. In addition, by applying feature selection, it has been tried to reduce the features to be used and increase the success rate. Accordingly, when the results of the Random Forest algorithm were generalized, it achieved an R2 value of 0.9328 using all features for the tenfold cross-validation analysis. At the same time, it reached an R2 value of 0.9434 using four features in the prediction of 2021 and was found to be successful. Considering only the 2021 forecast, the KNN algorithm reached the highest result with an R2 value of 0.9560 using only four features. Also, the Wilcoxon rank-sum test result at p = 0.05 shows no significant difference between the predictions and the actual values. Graphical abstract