Development of an Artificial Neural Network Approach for Predicting Plant Water Status in Almonds

Abstract. Stem water potential (SWP) is a commonly used method for determining plant water status (PWS) but requires a significant amount of time and is tedious to measure. To eliminate the necessity for this fieldwork, artificial neural networks (ANNs) were designed to predict PWS using information that is easier to measure, such as leaf temperature and microclimatic variables including ambient air temperature, relative humidity, incident radiation, and soil water content. To collect these variables, leaf and soil water sensors were placed in a 1.6 ha almond orchard. The sensors were interconnected through a wireless mesh network, which allowed remote data access. SWP values were taken in the field at midday three times a week during the growing season. The ANNs were trained using the Levenberg-Marquardt algorithm with the data divided into 70% training, 15% validation, and 15% test data. Each network contained one hidden layer with one to three hidden neurons. For each unique combination of inputs, the network was retrained five times, and the best network was selected based on the lowest mean squared error for the test data. When compared with multiple linear regression models fitting the same data, the networks consistently resulted in better R2 values, and higher values may be achieved with further optimization. These results suggest that there is potential for machine learning techniques that use ANNs to model the relationship between environmental conditions and PWS, which may be used for predicting acceptable temperature differences from target SWP. Keywords: Almonds, Artificial neural network, Leaf monitor, Machine learning, Plant water status, Precision irrigation, Stem water potential.