Combining Remotely Sensed Data and Ground-Based Radiometers to Estimate Crop Cover and Surface Temperatures at Daily Time Steps

Estimation of evapotranspiration (ET) is important for monitoring crop water stress and for developing decision support systems for irrigation scheduling. Techniques to estimate ET have been available for many years, while more recently remote sensing data have extended ET into a spatially distributed context. However, remote sensing data cannot be easily used in decision systems if they are not available frequently. For many crops ET estimates are needed at intervals of a week or less, but unfortunately due to cost, weather, and sensor availability constraints, high resolution (< 100 m) remote sensing data are usually available no more frequently than 2 weeks. Since resolution of this problem is unlikely to occur soon, a modeling approach has been developed to extrapolate remotely sensed inputs needed to estimate ET. The approach accomplishes this by combining time-series observations from ground-based radiometers and meteorological instruments with episodic visible, near infrared, and thermal infrared remote sensing image data. The key components of the model are a vegetation density predictor and a diurnal land surface temperature disaggregator, both of which supply needed inputs to a surface energy balance model. To illustrate model implementation, remote sensing and ground-based experimental data were collected for cotton grown in 2003 at Maricopa, Ariz. Spatially distributed cotton canopy densities were forecasted for a 22-day interval using vegetation indices from remote sensing and fractional cover from ground-level photography. Spatially distributed canopy and soil surface temperatures were predicted at 15-min time steps for the same interval by scaling diurnal canopy temperatures according to time of day and vegetative cover. Considering that the predictions span a rapid growth phase of the cotton crop, comparison of spatially projected canopy cover with observed cover were reasonably good, with R-2=0.65 and a root-mean-squared error (RMSE) of 0.13. Comparison of predicted temperatures also showed fair agreement with RMSE=2.1 degrees C. These results show that combining episodic remotely sensed data with continuous ground-based radiometric data are a technically feasible way to forecast spatially distributed input data needed for ET modeling over crops.