Forecasting crop yields with climate and economic variables: A machine learning approach for Türkiye

Authors : Burç Arslan Kaleli

Pages : 271-283

Doi:10.24181/tarekoder.1699618

View : 183 | Download : 295

Publication Date : 2025-12-19

Article Type : Research Paper

Abstract :Purpose: This study forecasts the agricultural productivity of five major crops—wheat, barley, maize, sunflower, and cotton—in Türkiye from 1962 to 2022, using climate variables alone and in combination with economic inputs. Design/Methodology/Approach: A panel dataset was constructed by matching annual crop yields with seasonal and annual temperature and precipitation variables, including lagged climate indicators. Two model configurations were tested: (i) climate-only and (ii) climate plus economic controls (fertilizer use, capital stock, labor). Three supervised learning models—Linear Regression, Random Forest, and Gradient Boosting—were evaluated using forward-chaining time-series cross-validation. Findings: Gradient Boosting with economic controls achieved the best out-of-sample performance (R² = 0.44, MAE = 547.8 kg/ha), followed by Random Forest. Climate-only versions of the same models yielded substantially lower accuracy (e.g., Gradient Boosting R² = 0.16), highlighting the added predictive value of structural inputs. Feature importance analysis identified growing season temperature as the most influential climate variable, while fertilizer, capital, and labor emerged as key predictors when included. Originality/Value: This study introduces a robust, time-aware machine learning framework for forecasting crop yields under climate variability. By integrating economic inputs, it enhances predictive accuracy and offers practical insights to support data-driven agricultural planning under climate uncertainty.
Keywords : Tarımsal verimlilik, iklim değişkenliği, ekonomik girdiler, makine öğrenmesi, verim tahmini

ORIGINAL ARTICLE URL