Challenges and Successes in Automated Calibration and Operation of Extreme Condition Models such as the METRIC Model in OpenET

OpenET is a collaborative and open source effort on the Google Earth Engine platform to implement a broad suite of spatial ET models and to provide open access to ET data for daily and monthly time periods. The METRIC model, which has been in development and operation since 1999, is one of the models being implemented. The current operational form of METRIC that runs on the Google Earth Engine (GEE) is named EEFlux. By necessity, EEFlux produces spatial maps of ET based on Landsat satellite imagery in a completely hands-free mode with completely automated handling and processing of satellite images and ancillary weather data and land use data.

The automated processing requires the automated calibration of the sensible heat flux component of the surface energy balance and relies on the identification of ‘extreme’ ET conditions on which to base expected near-minimum and near-maximum sensible heat fluxes. EEFlux and its offspring EE-METRIC in OpenET is designed to produce spatial ET for nearly all terrestrial regions of the globe. The automated identification of near-minimum and near-maximum sensible heat flux and near-minimum and near-maximum relative ET rates needs to be robust in climatic regions ranging from arid desert to rain forests, and for latitudes ranging from tropics to near the poles. In addition, calibrations need to be successful for Landsat scenes that include extensive cloudiness, but where ET for cloud-free locations is of value. These requirements and extreme ranges in ET and surface energy balance conditions places a significant burden on the calibration process. Currently, a statistical process is used to identify the near-extreme conditions in an image for calibration, with adjustment for background evaporation stemming from antecedent rain. Generally, agricultural lands are given preference for the identification due to their tendency to be uniform and ‘well-behaved’ relative to relationships between evaporative cooling by ET and thermal impacts. Fortunately, most satellite images have a distinct and consistent relationship between land surface temperature and sensible heat flux so that there is generally a high probability of identifying the near-extreme calibration points.

The presentation will describe mechanics of the automation approach and recent advances that include determination of scene-unique lapse rates that are used to ‘de-lapse’ surface temperature (LST) so that remaining variation in LST across a scene is due to surface energy balance effects, only. A quality assessment score is introduced to a metafile for each scene and date to help users and time integration scripts to select usable and dependable dates for a pixel or scene. The presentation is intended to provide insights into challenges and successes in calibrating fully automated spatial ET processes and provides means to assess expected accuracy and bias.

Beyond EEFlux calibration, another challenge with a reference-ET based spatial ET approach is the identification and correction of air temperature and humidity data to account for aridity-induced biases in gridded weather data. These biases are common to semi-arid and arid regions and can create up to 30% upward bias in reference ET estimates that are used as a basis for potential ET rates. Several approaches for the bias correction are discussed.