A dynamic optimization model for precision irrigation of cherry tomato under mechanized cultivation using CatBoost

• Correlation analysis revealed key indicators for each growth stage, including growth, photosynthesis, yield and quality. • TOPSIS and ML analyzed the characteristics of phenological periods and the water requirements, building dynamic irrigation models. • The CatBoost demonstrated superior performance (R 2 = 0.913, RMSE = 1.730, MAE = 1.238) among 12 machine learning algorithms. • Raspberry Pi automatic irrigation system with Catboost model increased the yield (16.85%) and IWUE (8.77%) over FAO method. Under mechanized cultivation, systemic modifications to the plant growth environment have profoundly impacted crop water requirements. Despite substantial advances in precision irrigation technologies, existing studies remain inadequate in capturing the dynamics of crop water demand across growth stages and in integrating irrigation management under variable cultivation conditions. In this study, a three-factor randomized split-plot experiment was conducted with two temperature regimes (ambient and ambient +2.3°C), two cultivation modes (wide–narrow and equidistant), and three irrigation levels (75% Ep, 100% Ep, and 125% Ep), yielding twelve treatments. Growth traits were quantified at each growth stage, together with final yield and fruit quality. Stage-specific key indicators were identified using Pearson correlation analysis. Optimal irrigation levels for each growth stage under different cultivation conditions were then determined by multi-objective optimization using the TOPSIS comprehensive evaluation model. Twelve machine-learning algorithms were trained to construct the irrigation decision model, among which CatBoost exhibited the highest predictive accuracy (R 2 = 0.913, RMSE = 1.730, MAE = 1.238). The optimized model was implemented and validated using an automatic irrigation system based on Raspberry Pi. Relative to the FAO-recommended irrigation strategy, the automated system significantly increased yield and irrigation water use efficiency (IWUE) by 16.85% and 8.77%, respectively, reduced fruit acidity by 36.36%, and decreased labor costs. Collectively, these findings provide a robust theoretical basis and practical technical framework for precision automatic irrigation in mechanized cultivation systems under future climate-warming scenarios.