Accurate knowledge of the distribution and interannual variability of precipitation, ET and runoff is critical for studying the biosphere-atmosphere interactions, and is crucial for improving the management of agricultural irrigation, water, and land resources, as well as drought monitoring and assessment. However, ET is difficult to estimate accurately due to the heterogeneity of the landscape and the large number of controlling factors, including climate, plant physiological properties, and soil properties. Developing a robust algorithm to estimate global ET is a significant challenge.

ET is traditionally estimated based on thermodynamic considerations of the surface-atmosphere system, requiring explicit characterization of numerous physical parameters, which are difficult to determine at the global scale. The MODIS sensor provides unprecedented information of vegetation and surface albedo, which can be used to develop a remotely sensed ET model. Using a modification of the Penman-Monteith equation and combining remote sensing and global meteorology data, we have developed an algorithm to estimate both surface resistance and ET.

Preliminary verification of the ET algorithm was performed using measurements at 20 AmeriFlux eddy covariance tower sites in 2001, which cover six typical land cover types and a wide range of climates. Input MODIS data include MOD12Q1 land cover, MOD13A2 EVI, MOD15A2 LAI, and the MOD43C1 albedo. We then calculated the ET for the 3-km x 3-km MODIS pixels surrounding each tower driven by GMAO meteorology data and tower meteorology observations averaging the results across all pixels. At the global scale, the 2000-2005 MODIS data were scaled to a resolution of 0.05„a, and missing/contaminated data were gap-filled.

The ET estimates were compared with the tower ET observations. These initial site-based results indicate that the ET estimates by the ET algorithm agree with the ET observation s well. Daily ET estimates at the tower sites driven by GMAO data are very consistent with those driven by site observed meteorology data. The ET algorithm was then applied globally for 2000-2006. The spatial pattern is reasonable. Tropical forests have the highest ET values, while temperate and boreal forests, the second most productive areas, have the second highest ET. Dry areas and areas with short growing seasons have both low GPP and low ET.

The MODIS ET algorithm also has the ability to capture the response of terrestrial ecosystems to extreme climatic variability at the regional scale. Our MODIS ET also captures seasonal droughts, such as those in Europe in 2003 and in the Amazon in 2005. Figure 1 shows the anomalies of ET from 2000 to 2005 over North America (NA) as estimated from the 0.05¢X MODIS ET, demonstrating the sensitive response of terrestrial ecosystems to (1) widespread drought in the mid-western USA and Canada during 2000, (2) extensive drought over NA in 2002, (3) drought over most of the USA in 2003, and (4) drought in the northern Canada in 2004. This ET dataset is a critical component of continental drought and fire danger monitoring systems.