A framework for constructing machine learning models with feature set optimisation for evapotranspiration partitioning

A deeper understanding of the drivers evapotranspiration and modelling its constituent parts (evaporation transpiration) may be significant importance to monitoring management water resources globally over coming decades. In this work a framework was developed identify best performing machine learning algorithm from candidate set, select optimal predictive features rank in terms their accuracy. The experiments conducted used 3 separate feature sets across 4 wetland sites as input into 8 algorithms, providing 96 experimental configurations. Given high number parameters, our results show strong evidence that there is no singularly or set all studied despite similarities. At each at least one model identified improved on performance baseline. key finding discovered when examining methane flux, whose relationship with not generally examined, contribute further biophysical process understanding. This demonstrates applicability for partitioning independent domain knowledge, producing models identifying new useful features.