Forecasting Spatially Distributed Cotton Evapotranspiration by Assimilating Remotely Sensed and Ground-Based Observations

Estimation of spatially distributed evapotranspiration (ET) with remote sensing could be especially valuable for developing water management tools in arid lands. For decision support over irrigated crops, these spatial ET estimates also depend on good spatial resolution ($<30\mathrm{m}$) at timely intervals, which for practical operations means no less frequent than approximately 5 days. For a variety of reasons, current remote sensing platforms usually cannot meet these needs. Commonly, overpass frequencies are no better than 16 days and sometimes are much worse considering cloudy skies. One way to reduce this problem is to develop an ET estimation approach that utilizes both remotely sensed data and ground-based observations. By combining episodic spatially distributed data with temporally continuous point observations, it could be feasible to provide continuous ET estimates that are better than can be achieved with either technique alone. Using data from a remote sensing irrigation scheduling experiment over cotton, conducted in 2003 at Maricopa, Arizona, an ET modeling approach was developed that used airborne images of vegetation indices (NDVI) and land surface temperatures (LST) along with ground-based thermal infrared radiometry and meteorology. Fractional vegetative cover were forecast from NDVI at daily time steps using a linear Kalman filter consisting of prior data, cumulative heat units, and spatially oriented beta distribution functions. LST were forecast hourly using a diurnal temperature model and a linear cover/LST estimator. ET accuracies derived from using these data as inputs to a surface energy balance model showed good agreement with independent ET estimates determined from 5-day soil depletion observations. Increased ET attributable to increased crop water use and irrigation applications were reflected in model outputs, and sometimes agreement was within 10% of independently observed soil moisture depletion data sets. These results indicated that combining remote sensing and ground-based data sets could be a feasible way to estimate ET at field-scales at daily time steps.