Skip to Main Content
This paper investigates novel supervised machine learning (SML) techniques for bulk soil moisture estimation using cosmic ray sensor. The cosmic ray soil moisture measuring probes are deployed across Australia as a part of CosmOz sensor network. These probes are brand new sensing technology still evaluated. The primary purpose of this paper is to find an alternative well-established SML-based method to estimate bulk soil moisture directly from the cosmic ray sensors that would effectively replace the current calibration equations. The second aspect of this paper is to find an alternative method to replace the cosmic ray sensor's current expensive and time-consuming field soil sample collection protocol. Data collected over 350 consecutive days from Australian Water Availability Project (AWAP) database and Hydroinnova CRS-1000 cosmic ray soil moisture probe deployed in Tullochgorum, Tasmania are used in this paper. Prediction performances of the four supervised estimators, such as sugano type adaptive neuro-fuzzy inference system (S-ANFIS), multilayer perceptron network, probabilistic neural network, and radial basis function network are evaluated using training and testing paradigms. The best result indicates that S-ANFIS is able to match the results achieved using existing calibration equations with a 87% accuracy. Secondly, AWAP data trained S-ANFIS is able to predict bulk soil moisture directly from cosmic ray neutron counts with a 92% accuracy without using any collected field sample-based measurements. Finally, a novel method is also developed to produce an estimated area average bulk soil moisture grid surface map based on multiple ANFIS and cubic grid surface interpolation.