Integrating Climate and Plant Variables with Machine Learning Models to Forecast Tomato Yield at Different Soil Moisture Levels

Abstract

Accurately predicting crop yield in different environmental conditions and irrigation regimes plays a vital role in optimizing agricultural practices and ensuring food security. This research aims to develop a tomato yield estimation model using machine learning (ML) models such as artificial neural network (ANN), random forest (RF), and decision tree (DT) models, based on climate and plant variables. To enhance the models' performance and prevent overfitting, a hyper-parameter tuning technique was implemented through cross-validation. Field experiments were conducted during the 2022 and 2023 growing seasons, implementing three irrigation regimes: 100%, 75%, and 50% of the full irrigation requirements (FIR). The results demonstrate that the ML models effectively captured the relationship between input variables and tomato production under deficit irrigation, achieving a desirable level of accuracy. Impressively, these models showcased predictive prowess 3-7 weeks before the harvest period. The artificial neural network models yielded an average root mean squared error (RMSE) of 3.9 ton/ha and a coefficient of determination (R2) of 0.95 for tomato yield prediction. The RF model displayed even better accuracy, with an RMSE of 3.50 ton/ha and an R2 of 0.96. The DT model forecasted tomato yield with an RMSE of 3.77 ton/ha and an R2 of 0.95. These findings highlight the practicality and reliability of utilizing climate and plant variables in combination with machine learning models to effectively manage tomato crop production, particularly when facing limited water availability for irrigation.