A low-cost stem water potential monitoring method using proximate sensor and scikit-learn classification algorithms
Haoyu Niu, Tiebiao Zhao, Cameron Zuber, Jacqueline Vasquez-Mendoza, David Doll, K. A. Arnold, YangQuan Chen · 2020 ASABE Annual International Virtual Meeting, July 13-15, 2020 · 2020
Abstract. Stem water potential is currently one of the most discriminating indicators for water deficit. It has been commonly used in agricultural applications, such as grapevine management in both non-irrigated and irrigated vineyards, almond water stress monitoring, and fruit size estimation. Traditionally, to determine the stem water potential of a tree, midday stem water potential is measured by using a pressure chamber. By measuring the pressure that it takes to make the water appear at the petiole, the pressure chamber can tell how much tension the leaf is experiencing. A higher value of pressure means a higher degree of water stress. However, this method involves a hazardous level of pressure. The operator must pay attention to safety precautions. Before sampling, the operator usually needs to cover a leaf for about ten minutes to stop the process of the water loss of leaves, which is time consuming if there are many samples in a field. The labor cost can also be a consideration. In this study, the authors proposed a new low-cost proximate radio frequency tridimensional sensor Walabot and machine learning classification algorithms. Walnut leaves from trees of different stem water potential were placed on this sensor to test if the Walabot can detect small changes in the water stress levels. Hypothetically, waveforms generated by different signals may be useful to classify stem water potential levels. Scikit-learn classification algorithms, such as Neural Networks, Random forest, Adam optimizer, and Decision tree were applied for data processing. Results showed that the Walabot predicted stem water potentials with an accuracy of 78% using the Decision Tree classifier so far. These findings highlight the powerful potential and the outstanding challenges in applying Walabot as a proximate sensor of water status.