Maximizing multi-source data integration and minimizing the parameters for greenhouse tomato crop water requirement prediction.

Accurate scientific predicting of water requirements for protected agriculture crops is essential for informed irrigation management. The Penman-Monteith model, endorsed by the Food and Agriculture Organization of the United Nations (FAO), is currently the predominant approach for estimating crop water needs. However, the complexity of its numerous parameters and the potential for empirical parameter inaccuracies pose significant challenges to precise water requirement predictions. In this study, we introduce a novel water demand prediction model for greenhouse tomato crops that leverages multi-source data fusion. We employed the ExG (Excess Green) algorithm and the maximum inter-class variance method to develop an algorithm for extracting canopy coverage from image segmentation. Subsequently, Spearman correlation analysis was utilized to select the combination of canopy coverage and environmental data, followed by the random forest feature importance ranking method to identify the most optimal feature variables. We constructed average fusion, weighted fusion, and stacking fusion models based on RandomForest, LightGBM, and CatBoost machine learning algorithms to accurately predict the water requirements of greenhouse tomato crops. The results show that the stacking model has the best prediction effect, and the error is lower than that of RandomForest, LightGBM, CatBoost, Average fusion model and Weighted fusion model. The feature combination of Tmax, Ts, and CC, filtered using Spearman and RandomForest, demonstrated the lowest prediction errors, with reductions in MSE, MAE, and RMSE of over 4%, 14%, and 3%, respectively, compared to other parameter combinations. The R