|HAIN, CHRIS - University Of Maryland|
|Kustas, William - Bill|
Submitted to: Water Resources Research
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 6/3/2013
Publication Date: 8/5/2013
Citation: Cammalleri, C.N., Anderson, M.C., Gao, F.N., Hain, C., Kustas, W.P. 2013. A data fusion approach for mapping daily evapotranspiration at field scale. Water Resources Research.
Interpretive Summary: Estimation of water used by crops (evapotranspiration)is becoming essential in many agricutural regions due to increasing demands on available water resources. This can be conducted over large agricultural areas using satellites observations. Unfortunately, with the suite of satellites currently available, we can determine crop water use at scales of individual fields (100m pixels) only infrequently (bi-weekly or once per month). Coarser resolution satellites can provide daily estimates, but they do not isolate water use on a field-to-field basis – estimates are averaged over several fields within a pixel 1-km in dimension. In this work, a method to combine these two sources of data (known as data fusion) is tested using evapotranspiration observations collected during the Soil Moisture Experiment of 2002 (SMEX02) in central Iowa. The method facilitates field scale estimates of crop water use at daily timesteps. The proposed method appears to adequately account for the effect of rainfall events that occur between two successive high-resolution satellite acquisitions, which are captured in the daily coarse resolution assessments. The capability for mapping evapotranspiration of agricultural fields on a daily basis is increasingly relevant given forecasted scenarios of reduced water availability worldwide.
Technical Abstract: The capability for mapping water consumption over cropped landscapes on a daily and seasonal basis is increasingly relevant given forecasted scenarios of reduced water availability. Prognostic modeling of water losses to the atmosphere, or evapotranspiration (ET), at field or finer scales in agricultural regions can be challenging due to the strong heterogeneity in surface conditions that typically prevail over managed landscapes. Remote sensing-based methods for mapping ET are generally preferred for such applications, because satellite imagery can capture relevant variability across the landscape. These models generally deal with the strong interconnection between remotely-observed land-surface temperature (LST) and ET fluxes through the surface energy balance. Unfortunately, current satellite-based thermal infrared imaging systems used to map LST are characterized by either low spatial resolution (10^3-10^4 m) and high repeatability (1 day to 15 min) or by moderate/high spatial resolution (10^1-10^2 m) and low temporal frequency (2 weeks). With the goal of obtaining accurate daily ET estimates at field scale, a data fusion modeling framework is evaluated, combining spatiotemporal characteristics of both classes of thermal sensors to provide optimal coverage. The Atmosphere-Land EXchange Inverse (ALEXI) model uses GOES (Geostationary Operational Environmental Satellite) data to create coarse resolution (10-km) hourly ET maps at the U.S. continental scale, whereas the DisALEXI procedure uses ALEXI results as a boundary condition to assess ET periodically at finer spatial scales using daily MODIS (MODerate resolution Imaging Spectroradiometer) 1-km and bi-weekly Landsat 30-m LST imagery. In this study, MODIS and Landsat ET fields are combined by means of the Spatial and Temporal Adaptive Reflectance Fusion Model (STARFM) to obtain continuous daily ET maps at 30-m spatial resolution. The ALEXI/DisALEXI procedure was applied for June-August 2002 over central Iowa, the site of the Soil Moisture Experiment of 2002 (SMEX02). The accuracy of ET maps at 30-m resolution was evaluated using observations from 8 flux towers, which collected continuous flux data in corn and soybean fields during a period of rapid crop development. The comparison between STARFM fused results and a benchmark case (obtained using Landsat images only) highlights that significant improvement in the agreement between modeled and observed ET can be obtained in some cases by fusing MODIS and Landsat data (mean error reduced from 0.75 to 0.58 mm/d on average), especially when a rainfall event occurs between two successive Landsat acquisitions. The improvements are more evident at the seasonal scale, where a 10% systematic underestimation of Landsat-only estimations is reduced to 2% for the MODIS-Landsat fused ET data.