|Abstract or Summary
- Evapotranspiration (ET) is an important component of the hydrologic cycle that transfers large quantities of water vapor away from Earth's surface into the atmosphere. In addition to having agricultural water management applications, including monitoring water rights compliance and irrigation scheduling, estimating ET is also important to quantify water used by other landscapes for soil-vegetation-atmosphere-transfer (SVAT) modeling schemes. This can only be done by estimating ET at large scales and this is most efficiently achieved by remote sensing.
Daily ET was retrieved from two remote sensing modeling schemes: a) Reconstructed METRIC: Mapping EvapoTranspiration at high Resolution with Internalized Calibration (R-METRIC) that uses thermal band data from the Landsat 8 satellite; and b) Fusion: ALEXI/DisALEXI (Atmosphere-Land EXchange Inverse/Disaggregated ALEXI) and STARFM (Spatial and Temporal Adaptive Reflectance Fusion Model) that combines GOES (Geostationary Operational Environmental Satellite), MODIS (Moderate-resolution Imaging Spectroradiometer) and Landsat 8. High resolution, daily ET was mapped over two predominately wooded or forested, water stressed study locations at Tonzi Ranch in California and Metolius Forest in Oregon for the summer months of 2013. Both sites have established networks of eddy covariance instruments that acquire high temporal resolution moisture flux data.
Instantaneous surface energy fluxes estimated by R-METRIC at the Tonzi study site showed reasonable agreement with in situ measurements over the flux tower footprint with relative errors (RE) less than 15% for all fluxes except latent heat (λET) which showed RE = 31%. Validation at the Vaira and Metolius towers showed similar results with the exception of significantly overestimating λET (RE = 297%) at Vaira and soil heat flux at Metolius (RE = 169%). DisALEXI showed good agreement for solar and net radiation (RE < 13%) at all sites. Significant overestimations of λET (RE up to 540%) and underestimations of sensible heat (RE up to 65%) were produced at each site. Additionally, soil heat flux at Metolius showed errors up to 167%.
While daily ET at Tonzi modeled with R-METRIC agreed well with observed measurements, modeled values of daily ET were significantly overestimated at Vaira (RE = 395%). DisALEXI showed severe overestimations of daily ET at both Tonzi and Vaira (RE = 428 and 596%, respectively). It should be noted that the observed daily ET at these sites is very low compared to the sensible heat flux which leads to the unexpectedly high error. At Metolius, the two models produced comparable results though they both overestimated the observed daily ET. Because daily ET was overestimated, the seasonal cumulative ET was also overestimated at all sites by both models with the exception of R-METRIC over Tonzi.
Surface and evaporative fluxes retrieved from the two models were also inter-compared over the different land cover types in the scenes. As both schemes were specifically developed for use over agricultural lands, they agree reasonably well with measurements when used over that land cover type. When applied over other land covers, specifically forests, grassland and shrubs, the daily ET showed greater discrepancies.
The results of this study suggest that the current version of the Fusion scheme estimates much higher ET than actually occurs at all three tower locations at both instantaneous and daily scales. This likely results from the ALEXI processing step in which the air temperature for input into DisALEXI is found. Though relatively easy for the user to implement this model, until that step is debugged, it remains unclear how accurate it may be over non-agricultural environments. R-METRIC shows good agreement at the instantaneous scale but more discrepancy at the daily scale. Unlike the Fusion scheme, R-METRIC requires user discretion in order to calibrate it to the study site and is therefore subject to user bias. Though both models have proven their utility over agricultural fields, the water stressed conditions at both sites present a challenging yet important environment that needs improved accuracy in both models.