Model-parameter fitting using simulations from independent remotely-sensed evapotranspiration algorithms applied in different hydrologic conditions

There are a variety of methodologies applied to compute evapotranspiration (ET) in riparian habitats that yield contrasting results, obviating the usefulness of a descriptive evaluation of model parameters. This work investigates the spatio-temporal relationships of model parameters used in a vegetation-index algorithm and land surface temperature approach. Both methodologies use Landsat imagery with other spatially-gridded data sets and local point-scale measurements. Bayesian hierarchical and non-hierarchical analyses are conducted on the ET algorithms to discover where individual model parameters exist during hydrologically-different growing seasons (March to October). Estimates of ET and the corresponding ranges of model-parameters are calculated for 30 square kilometers of riparian buffer. These analyses experiment on the functional forms of equations used to make recommendations for natural-ecosystem water use. Provisional results indicate our temperature-based approach yields higher ET rates under increased aridity, peaking in June and July. Under decreased aridity, the Enhanced Vegetation Index (EVI) peaks in June and the Normalized Difference Vegetation Index (NDVI) peaks in August. We expect to find that a flexible and naïve Bayesian framework will explain how each model parameter can vary under similar climatic conditions, as well as fluctuate with contrasting degrees of aridity in separate water years. A Bayesian framework can borrow strength from dependent predictor variables to highlight the realm of correctness for each model parameter, and in doing so, it will be used to examine which ET algorithm is most capable of generating the data used to construct the model originally.