Development of an evapotranspiration estimation method for lettuce via mobile phones using machine vision: Proof of concept

This study aims to elucidate the process of crop evapotranspiration (ETc) and develop models to estimate ETc as a crucial value to efficiently schedule irrigation. Accordingly, we propose a new idea for non-destructive online acquisition of ETc data via mobile phones. The greenhouse environmental data were collected by sensors and uploaded to a mobile phone for the Penman-Monteith (PM) calculations of reference crop evapotranspiration (ET0). The canopy coverage (PGC) was accurately extracted from mobile phone images using the super-green algorithm, and four machine learning algorithms (XGBoost, CatBoost, SVR, and RF) were used to construct an online computing model for crop coefficient (Kc), ETc=Kc*ET0 was used to dynamically monitor the change of water requirement of different types of lettuce. The results showed that the correlation coefficients for the four monitoring networks were in the range of 0.74-0.93, which could be used to estimate the Kc at different stages. The integrated models showed high accuracy and similar performance. Compared with that of the SVR model, the root mean square error (RMSE), mean absolute error (MAE), and mean square error (MSE) of the three al-gorithms were reduced. Regarding the generalisation and computational speed of the proposed approach, the estimation accuracy of the XGBoost model was improved by 17.31%, and the convergence speed was also increased by 12.57% relative to other models on different types of lettuce. The XGBoost model can be recom-mended as the Kc model for estimation. The daily mean values of leaf lettuce and head lettuce as measured in ETc were 2.38 mm d-1 and 6.22 mm d-1, respectively. The average errors were 3.48% and 6.36%, respectively. We further explored models of the mathematical relationship between Kc and the plant height (H), plant width (D), and normalised vegetation index (NDVI), to establish a new approach for reasonable and scientific irrigation management.

Automated summary

This study aims to understand crop evapotranspiration and develop models to estimate it, in order to efficiently schedule irrigation. The study proposes a new approach of non-destructive online acquisition of evapotranspiration data using mobile phones. The models developed in the study show high accuracy and performance in estimating water requirements for different types of lettuce.