In this paper a comparative study of supervised
machine learning methods has been investigated online dynamic
calibrate cosmic ray based bulk soil moisture sensor. Data collected
from the Australian Water Availability Project (AWAP) database has
been used as independent ground truth and the Hydroinnova
CRS-1000 cosmic ray probe deployed in Tullochgorum, Australia was
experimented for this study. Prediction performance of the five
supervised artificial neural network (ANN) estimators, namely
Sugano type Adaptive Neuro-Fuzzy Inference System (S-ANFIS),
Multilayer Perceptron Neural Network (MLPNN), Elman Neural
Network (ENN), Learning Vector Quantization Neural Network
(LVQN) and Radial Basis Function Network (RBFN) were evaluated
using various incremental training and testing paradigms to establish
the best generalisation methodology to calibrate the probes remotely.
AWAP trained five estimators was able to predict bulk soil moisture
directly from cosmic ray neutron counts within the range of 74%-91%
as best accuracies, whereas best sensitivity and specificity was 89%
and 93%. These results proved that supervised artificial neural
network based paradigm could be a valuable alternative calibration
method for cosmic ray sensors against the current expensive and
hydrological assumption based field calibration method.