Uncertainty in evapotranspiration from uncertainty in land surface temperature

Land surface temperature (LST) is a critical component in numerous models of evapotranspiration (ET). LST is typically obtained from remote sensing observations in the thermal infrared bands, yet uncertainty in LST retrievals is poorly or rarely quantified, and the impact of LST uncertainty on retrieved ET has rarely been studied. We used a common LST-based ET model, SEBS, to evaluate the uncertainty in ET due to uncertainty in LST from MODIS (MOD11, MOD21). Uncertainty in MODIS LST was obtained using the JPL-LST uncertainty simulator over a wide range of atmospheric conditions and surface types. We evaluated SEBS at 78 AmeriFlux sites where ET was measured in situ and regionally for N. America. We show how measured vs. modeled ET changes as uncertainty in LST increases from ±0 to 5ºC. ET estimation error increases much more rapidly with overestimated LST than with underestimated LST as SEBS reaches its dry edge limit, which is compounded by the increasing data drop as SEBS returns null values due to violation of the physics assumptions surrounding the air-to-surface temperature difference.