Submitted to: Biological Systems Simulation Group Proceedings
Publication Type: Proceedings
Publication Acceptance Date: 3/22/2010
Publication Date: 4/13/2010
Citation: French, A.N., Hunsaker, D.J., Thorp, K.R. 2010. ET Estimation by assimilating ground and airborne data. Biological Systems Simulation Group Proceedings. 12-13.
Technical Abstract: Within the last three decades researchers have been actively investigating ways to estimate evapotranspiration (ET) with remote sensing. By utilizing visible to near infrared bands for monitoring vegetation canopy densities, and thermal infrared bands for monitoring land surface temperatures it is feasible to estimate spatially distributed surface energy fluxes, which can then be used to estimate ET. This capability, achieved using remote sensors such as Landsat and ASTER, is a major advance over point-based observations because spatial variations in crop water can be observed, rather than ignored or statistically inferred. This in turn means that crop growth differences within a field can be detected and incorporated into decision-support tools. Unfortunately, however, the value of remote sensing data is greatly constrained for field-scale applications because image data with adequate spatial and temporal resolution are usually unavailable. For example, imagery finer than 100 m resolution are typically only available from polar orbiting satellites, but repeat intervals for these commonly 16 days. Where cloudy skies prevail, intervals would naturally be much worse. On the other hand, equatorial/geostationary satellites can provide image data at least every half hour, but at unsuitable spatial resolution, 4-5 km. Although it may be possible to combine image streams from each of these sensors in a data fusion approach, the general situation is clearly unsatisfactory for monitoring and managing actively growing crops. One remedy for this sampling shortcoming is to launch additional satellites and thus fill observational gaps. Unfortunately, however, satellites with all of the needed spectral sampling capabilities are very expensive and are unlikely to be constructed and launched in the near future. Thus alternatives are needed. The one suggested here is to augment occasional satellite observations with ground-based point observations. By combining these data sets within a crop growth modeling framework it is feasible to continuously and realistically estimate crop ET. The proposed approach has two observational components and three modeling components: 1. Acquisition of spaceborne or airborne remote sensing image data in visible, near infrared, and thermal infrared wavelengths to provide spatially distributed but episodic estimates of vegetation cover and temperatures. 2. Acquisition of ground-based thermal radiometric and weather data to provide point-based, but temporally continuous estimates of local surface energy fluxes. 3. Implementation of a semi-empirical land surface temperature data model that provides a way to extend point-based surface temperature data sets to entire fields by considering remotely sensed image data. 4. Implementation of a heat-unit vegetation crop cover assimilation model that provides spatial maps on days without remote sensing observations. 5. A synthesis component that models hourly surface energy fluxes and estimates crop water use at daily time steps. To demonstrate how these components are realized, data collected from a small plot study in cotton are utilized. The study, known as the FAO-56 Irrigation and Scheduling Experiment 2003 (FISE03), was a remote sensing project designed to evaluate the efficacy of modifying standardized crop water use estimates by using vegetation indices. Fortunately, the study included extensive airborne and ground-based thermal infrared data observations, all of which permitted testing the proposed crop ET estimation approach. Based on ten airborne remote sensing sorties and data streams from two ground-based observation sites, daily ET was able to provide 10-day forecasts of daily crop water use to fair accuracy (usually within 1.5 mm/day) for most of the growing season. Tests from the 2003 data also revealed inaccurate model assumptions, such as spatial variabil