Sensitivity and Uncertainty of Input Sensor Accuracy for Grass-Based Reference Evapotranspiration.

Quantification of evapotranspiration (ET) in agricultural environments is becoming of increasing importance throughout the world, thus understanding input variability of relevant sensors is of paramount importance as well. The Colorado Agricultural and Meteorological Network (CoAgMet) and the Florida Automated Weather Network (FAWN) both utilize an array of sensors to acquire micrometeorological data (temperature, humidity, wind speed, and solar radiation) and can use the ASCE Standardized Reference ET Equation to determine geographically local reference ET values. Multiyear datasets for both networks were evaluated for grass-based reference ET using a local sensitivity analysis which calculated total error range of each individual sensor, as well as Morris and eFAST global sensitivity analysis (GSA) methods which simultaneously evaluate the full accuracy range of each sensor. GSA results were highly correlated with each other, but local sensitivity was poorly correlated for wind input in Colorado. Sensitivity of inputs was generally well-balanced for the Florida network with solar radiation being the most influential in the summer, while the Colorado network’s sensitivity to wind was much higher than the other inputs as shown by all three sensitivity analysis methods due to a large range of quoted sensor accuracy. Uncertainty analysis showed Colorado’s current configuration of sensors to have a higher range of values between 5% and 95% confidence intervals, as compared to Florida. The eFAST GSA method was conducted again using a hypothetical set of “best case” sensors in both stations, showing solar radiation to be the most sensitive input in the high ET months of summer in semi-arid Colorado and humid Florida, and the sensitivity in Colorado to wind to be vastly decreased, suggesting an upgrade of anemometers to the current CoAgMet network. Local sensitivity analysis is suggested as a basic screening method for evaluating input sensor sensitivity.