Evaluating Landsat 8 Evapotranspiration for Water Use Mapping in the Colorado River Basin

Abstract:

Evapotranspiration (ET) mapping at the Landsat resolution is essential to fully understand water use and water availability of a basin at the field scale. The Colorado River Basin (CRB) includes diverse ecosystems and complex hydro-climatic regions. The first ever CRB-wide ET mapping was conducted using Landsat 5 and 7 images for 2010. The availability of Landsat 8 images gave the opportunity to map CRB ET for the second time to detect and assess the nature and magnitude of change between the two periods. Using the Operational Simplified Surface Energy Balance (SSEBop) model, 528 Landsat 8 scenes were used to create seamless monthly and annual ET estimates at the inherent 100 meter thermal band resolution. Annual ET values were summarized by land use/land cover classes. Validation with eddy covariance (EC) flux towers and water balance approaches showed good accuracy levels with R² ranging from 0.74 to 0.95 and Nash-Sutcliffe model efficiency coefficient ranging from 0.66 to 0.91 and relative bias ranging from 2 to 35%. The root mean square error ranged from 0.48 – 0.60 mm for daily (days of satellite overpass) ET to 7.75 – 13.04 mm for monthly ET. The magnitude of ET among different land use/land cover between 2010 and 2013 were relatively consistent, showing cropped areas with the highest annual ET next to water bodies; this indicates the reliability of Landsat 8 for providing important information on ET dynamics across the landscape. Annual crop water use was estimated for selected five irrigation districts in the lower CRB where annual ET per district ranged between 681 – 772 mm. A rainfall analysis over these districts suggested that, on average, 69% of the annual ET was met by irrigation. Although the enhanced cloud-masking capability of Landsat 8 improved the removal of contaminated pixels, the ability to reliability estimate ET over these areas remains an important challenge. Although the relative performance of Landsat 8 was consistent for crop lands compared to the 2010 estimation, the large difference in the number of scenes between the two years precluded a meaningful comparison in absolute terms. Overall, the performance of Landsat 8 against available EC datasets and water balance estimates on a complex basin such as the CRB demonstrate the potential of using Landsat 8 for annual water use estimation at national scale.