AUTHOR=Maia Rodrigo Filev , Lurbe Carlos Ballester , Hornbuckle John TITLE=Machine learning approach to estimate soil matric potential in the plant root zone based on remote sensing data JOURNAL=Frontiers in Plant Science VOLUME=Volume 13 - 2022 YEAR=2022 URL=https://www.frontiersin.org/journals/plant-science/articles/10.3389/fpls.2022.931491 DOI=10.3389/fpls.2022.931491 ISSN=1664-462X ABSTRACT=There is an increasing interest in using the Internet of Things in the agriculture sector to acquire soil and crop-related parameters that provide helpful information to manage farms more efficiently. Soil moisture sensors for monitoring plant water availability are usually deployed in nodes. A more significant number of sensors/nodes is recommended in larger fields, such as those found in broadacre agriculture, to better account for soil heterogeneity. This comes at a higher and often limiting cost for farmers (purchase, labour costs from installation and removal, and maintenance). Methodologies that could enable maintaining the monitoring capability/intensity with a reduced number of sensors in the field would be valuable for the sector and therefore of great interest. In this study, the analysis of sensor data harvested across two irrigation seasons in three cotton fields in two cotton-growing areas of Australia identified a relationship between soil matric potential and satellite-derived cumulative crop evapotranspiration between irrigation events. Such a relationship is represented by a second-degree function, which is affected by the crop development stage, rainfall, irrigation events and the transition between saturated and non-saturated soil. Two machine learning models (a Dense Multilayer Perceptron (DMP) and SVR – Support Vector Regression algorithm) were studied to explore these second-degree function properties and assess whether the models were capable of estimating soil matric potential in the root zone. The performance of the algorithms in predicting soil matric potential was evaluated using the k-fold method in each farm individually and by mixing data from all fields and seasons. Such an approach made it possible to avoid the influence of farm consultants’ decisions regarding when to irrigate the crop in training data. The evaluation indicated that both algorithms accurately estimated soil matric potential for individual (up to 90% of predicted values within ± 10 kPa) and combined (72% of predicted values within ± 10 kPa) datasets. The technique presented here can be used to accurately monitor soil matric potential at the root zone of cotton plants with reduce in-field sensor equipment and offers promising applications for its use in irrigation-decision systems.