Estimating actual evapotranspiration at field-to-continent scales by calibrating the CMRSET algorithm with MODIS, VIIRS, Landsat and Sentinel-2 data

Abstract

Actual evapotranspiration (ETa) can be estimated using optical remote sensing and meteorological data. Herein we calibrated and applied the CMRSET (CSIRO MODIS Reflectance-based Scaling EvapoTranspiration) model at the continental scale in Australia using five remotely sensed data products with spatial resolutions ranging from 500 m (MODIS, VIIRS), 30 m (Landsat) to 20 m (Sentinel-2), and temporal frequencies ranging from daily (MODIS, VIIRS) to multi-days (Landsat and Sentinel-2). The five remotely sensed data products were used to compute a crop factor (kc) based on two vegetation indices which modify Priestley-Taylor potential evapotranspiration estimates (ETp; calculated from nation-wide daily meteorological grids) to obtain ETa. CMRSET was calibrated using daily latent heat observations from 30 eddy covariance flux towers located across Australia, representing a wide variety of land covers and climates. To ensure that CMRSET adequately estimated open water evaporation, daily ETp estimates at six water sites were also added in the calibration network. The calibrated CMRSET model was able to estimate daily ETa observed at the flux towers with a relative Root Mean Squared Error (rRMSE)/coefficient of determination (R2) ranging between 0.15/0.96 (with Sentinel-2) to 0.26/0.93 (VIIRS). Additionally, we independently evaluated the performance of CMRSET by comparing the long-term (5 years or more) differences between mean annual precipitation and ETa with measured streamflow at 638 unimpaired catchments across Australia. The comparison showed a RMSE of 0.50 mm·d-1 (rRMSE of 0.26) and a R2 of 0.70. The CMRSET model performed better than or similarly when compared against two global ETa products based on MODIS optical and thermal data. Extending the suite of remotely sensed data products that CMRSET is calibrated with and using a network of sites located across vast climate and geographic ranges, means that CMRSET can now be used to estimate ETa at continental scales with multiple spatial resolutions and temporal frequencies. Compared to other available ETa products, CMRSET is more straightforward to operationalise, requires less data inputs, can capture ETa dynamics in areas with significant direct evaporation such as floodplains and lakes and does not require manual calibration to detect wet and dry pixels.