Skip to main content

Evapotranspiration data from eddy-covariance flux-tower measurements and Landsat imagery in California’s Sierra Nevada from 1985 to 2019

Cite this dataset

Ma, Qin et al. (2020). Evapotranspiration data from eddy-covariance flux-tower measurements and Landsat imagery in California’s Sierra Nevada from 1985 to 2019 [Dataset]. Dryad.


The gridded annual evapotranspiration (ET) from 1985 to 2019 were calculated based on the correlations between eddy-covariance flux-tower measurements of annual evapotranspiration and satellite imagery derived Normalized Difference Vegetation Index (NDVI). Annual ET observations from 12 flux-towers across the Sierra Nevada and Southern California were collected from 2001 to 2016, resulting 97 site-years of ET observations in five main vegetation types (Evergreen Needleleaf Forest, Grasslands, Mixed Forest, Open Shrublands, and Woody Savannas). The NDVI was calculated at 30-m resolution from USGS Landsat Collection Tier 1 surface reflectance over the entire California and aggregated to annual mean values for each water year (Oct. to Sept.) from 1985 to 2019. The gridded annual ET was calculated at 30 m spatial resolution using the exponential relationship between NDVI and ET derived from flux-tower sites.


The ET regression function (1) was estimated based on the  method developed by (Goulden and Bales 2019) using exponential regression in R language. The flux-tower ET observations were summed annually by water year after filling data gaps (Rungee et al. 2018). The total 97 site-year ET observations also included some site-years that were impacted by drought, fire, forest thinning, prescribed fire, reflecting by the significant reduction in annual NDVI. The annual NDVI map was calculated as the mean of all Landsat scenes for a water year (Oct. to Sept.) after masking for shadow, snow or cloud (Zhu and Woodcock 2012). We homogenized Landsat 8 NDVI (L8, 2014-2019) and Landsat 7 (L7, 2012- 2013) to Landsat 5 NDVI (L5, 1985-2011) following equation 2 and 3 (Su et al. 2017; Sulla-Menashe et al. 2016). The NDVI and ET maps were generated using Google Earth Engine. The modeled ET showed a strong correlation to site-level flux-tower observations (coefficient of determination, R2=0.77). Most of the modeled ET fall within ±100 mm ranges of ET measurements, with a root mean square error (RMSE) at 108 mm, and mean absolute error (MAE) at 74 mm. The main estimation error is observed at site-years with high NDVI and ET values, due to NDVI saturation issue. The model’s temporal and spatial sensitivities were assessed using leave-one-out cross validation method by removing an individual water year or flux-tower site for model building and then evaluating on the site-year removed.

ET=112.3×e3.2×NDVI                    (1)

L5=0.9883×L7-0.0367                 (2)

L5=0.8213×L8-0.0403                 (3)


Usage notes

Each annual ET map is named as “ET water year_expNDVI.tif” in a GeoTIFF format under the GCS_WGS_1984 projection system. The ET values is calculated in the unit of mm/year. ET values are mapped over the entire California, but most accurate in Sierra Nevada. The Google Earth Engine code to generate the map can be found in this link:


National Science Foundation Southern Sierra Critical ZoneObservatory, Award: EAR-0725097, 1239521, and 1331939

California Climate Investments program through the Strategic Growth Council , Award: CCR20021

UC Water Security and Sustainability Research Initiative , Award: 13941

USDA National Institute of Food and Agriculture, McIntire Stennis project , Award: 1022944