Prediction of plant transpiration from environmental parameters and relative leaf area index using the random forest regression algorithm

Abstract

The accurate prediction of plant transpiration is essential for automatic precision irrigation in greenhouses; thus, an accurate transpiration prediction model is needed. The main objective of this study was to establish a plant transpiration prediction model by integrating plant and environmental parameters based on random forest regression (RFR). The relative leaf area index (RLAI) was selected as the plant parameter, which could be obtained in real time by image processing. The environmental parameters included air temperature, relative humidity and illumination intensity, which were obtained in real time by sensors. More than 500 samples were collected in each of the seedling and florescence phases, of which 70% were used for training, and 30% were used for testing the RFR model. The performance of the RFR model was then compared with that of a BP neural network and a GA-BP neural network. The results showed that the RFR model significantly outperformed the other models, achieving the highest prediction accuracy (R2 = 0.9472 and 0.9654) and smallest error (RMSE = 19.75 g and 18.85 g) among the test sets. In addition, the results confirmed the importance of all the parameters used to predict plant transpiration. Among the parameters, illumination intensity had the greatest influence on the mean square of the prediction error rate, and RLAI had the smallest influence. However, RLAI was very important for improving the prediction accuracy of the RFR model and achieving model significance. This research provides a scientific reference for the efficient production and intelligent irrigation of greenhouse tomatoes, providing a way to conserve water and energy.