An Evaluation of Evapotranspiration Model Complexity Against Performance in Comparison with Bowen Ratio Energy Balance Measurements

Abstract

We investigated the relevance of model complexity in 11 evapotranspiration (ET) models, both for two-step and direct predictions for sum-of-hourly, daily, and seasonal time scales and compared their performances against the Bowen ratio energy balance system-measured ET for corn (Zea mays L.) in south central Nebraska. The simplest model we used was the water equivalent of net radiation, the intermediate was the Priestley-Taylor equation, and the complex models were the ASCE-Penman-Monteith (ASCE-PM) based two-step applications (i.e., grass and alfalfa reference surfaces). We also estimated ET with the soil water balance equation from the soil water content measured using a neutron probe soil moisture meter. By constructing the models from the simple towards more complicated, the effects of different components of the energy balance on ET during the growing season were investigated. We reported coefficients of determination (R2) and Nash-Sutcliffe efficiency (R2NS), and root mean square error (RMSE) to assess the models' performance against measured data. The ET from the soil water balance approach agreed well with those measured with the BREBS throughout the growing season. Over a seasonal time scale, the simplistic radiation-based parameterizations gave similar results as those more complex models. In the early season before canopy closure is attained (stage I) and in the late season after crop physiological maturity (stage III), the model predictions significantly deviated from measured data for both daily and seasonal estimates. The two-step ET calculation procedure was found to work very well during the mid-season (stage II) when the crop canopy was fully developed. In stages I and III, all the methods we used revealed negative R2 and R2NS values between the daily predicted and measured ET. We found significant sensible heat losses, indicating low rates of transpiration in both stages I and III. Compared to stage II, the time-dependent crop coefficients were found less capable in scaling the reference ET to represent daily actual crop ET fluctuations in stages I and III. Both the ASCE-PM-based two-step estimates as well as the direct estimates from the Priestley-Taylor and other methods gave more accurate results under full canopy cover. In this period, a sensible heat parameterization implicit in the models improved the daily predictions approximately 10% to 15%, despite model deficiencies. In the mid-season, our results suggested that the daily ASCE-PM reference grass-based approach provided arguably better estimates with respect to both daily and seasonal total fluxes compared to the sum-of-hourly calculations. Compared to the water equivalent of net radiation method, the complex models achieved up to 17% improvement in explaining the variability in daily ET during the mid-season. However, these methods did not improve seasonal ET estimates. Our results suggested that radiation is the dominant driver of evaporative losses over seasonal time scales. Other meteorological variables gained importance at daily and sum-of-hourly calculations.