Tomato leaf color as predictor of soil moisture value using machine learning techniques

Fresh water supplies for irrigation purposes must be used sparingly and judiciously, as water is an invaluable natural resource that is in short supply in much of the Earth. Soil moisture in fields is not uniform everywhere, and deploying thousands of sensors is unnecessarily expensive. The purpose of this publication is to model and predict the relationship between tomato plants leaf color, soil moisture, and thus manage the irrigation process in an optimal manner. The research was conducted using generally accepted methods, the field method, and the method of statistical evaluation of results. Machine learning algorithms (MLA) and data mining are utilized in this paper to model the relationship between RGB color values from tomato leaves and soil moisture and temperature. The color of the leaves of open field tomato plantations grown without stakes is the focus of this study. Three main tasks are fulfilled: to prove that there is a relationship between leaf color and soil moisture, to study its supposedly nonlinear type and to model this relationship with MLA. First, a classifier is trained, and then a model is created and saved. Finally, the efficiency of the chosen model is tested using a different test data set. The name “12-9-6-3” for the methodology of measurements is fgiven. It is proven that the young leaves are more informative about the need for watering. As a result, there is less than a 1% error in predicting soil moisture using the color of tomato leaves considering also soil temperature, using M5P regression model. This predictive model can be used in creation of automated systems for optimal irrigation management and water saving.