NDVI*-BASED ESTIMATION OF EVAPOTRANSPIRATION

NDVI*-Based estimation of Evapotranspiration

Plant water use generally accounts for the largest proportion of evapotranspiration, other components being interception losses and direct soil evaporation. Remote sensing of greenness using satellite data can assess spatial plant vigor using vegetation indices, however, many vegetation indices exist, some with large degrees of associated error.

The first step to build a greenness-based ET estimation method was to test which vegetation index (VI) performed best to track vegetation hydrologic response, while being resistant to accuracy-reducing atmospheric scatter and attenuation. Test data were assembled from a 17-year span (1986–2002) of mid-summer Landsat TM snapshots for San Luis Valley, Colorado that were processed to reflectance and corrected for atmospheric scatter. Data were then extracted and pooled from 2953 pixels contained in 180 131-m-diameter circles of homogeneous native phreatophyte alkali scrub where water tables had remained stable through the study period. All published VIs were calculated from this data base for comparison to antecedent precipitation, a factor known to produce linear promotion of vegetation. Regression statistics (r² values) from the paired antecedent precipitation and average pixel-wise VI values were used to evaluate VI performance to predict the expected linear relationship. A stretched form of normalized difference vegetation index, NDVI*, yielded superior results to all other VIs.

To fit an ET predictive relationship to NDVI*, data measured by micrometeorological methods were obtained from three separate studies conducted in arid/semiarid shallow-groundwater environments in California, New Mexico and Colorado, locations with great diversity in (1) climate expression: monsoonal versus Mediterranean precipitation pattern; (2) comparatively long versus short growing seasons; and (3) vegetation cover: alkali scrub, shallow groundwater meadows, and monocultures of saltcedar and cottonwood. Annual total ET, (ET_a) measured from 24 site- and year-combinations were paired with NDVI* extracted from corresponding locations in nine mid-summer Landsat TM 5 and 7 scenes during 1999-2002, with single mid-summer scenes used for estimation of annual ET_a. In a 1:1 relationship, NDVI* was found to be a competent estimator of a derivation of ET, ET*, calculated by subtracting precipitation and normalizing by corrected reference ET, (ET₀, minus precipitation). This relationship was used to estimate ET_a, derived solely from remote sensing and weather data. Residual error decreased at higher levels of ET_a and data scatter was well balanced indicating lack of systematic error.

NDVI* is proving to be a robust spatial scalar for ET but is only as accurate as estimates of ET₀. Formulations of NDVI* can be used for estimation of groundwater discharge ET, and crop water use in annual, or shorter, time segments.