Estimating time series of land surface energy fluxes using optimized two source energy balance schemes: Model formulation, calibration, and validation

• A canopy conductance model is incorporated into a two source energy balance scheme.
• Both Gc-TSEB and PT-TSEB models are evaluated using EC flux measurements.
• LST is as effective as EC fluxes in calibrating the models unless it is biased.
• The models are robust because their parameters can be adjusted by LST-calibration.
• The predicted daytime and diurnal ET matched the trends of EC flux measurements.

Due to the limited availability of land surface temperature (LST) images, thermal-based evapotranspiration (ET) models can only provide instantaneous ET snapshots. In contrast, models that are based on near surface soil moisture (SM) and leaf area index (LAI) can operate at daily scales. However, their transpiration schemes need to be more physically realistic and their model parameters usually need to be calibrated by flux measurements. In this study, we incorporated a biophysical canopy conductance (Gc) model into a two source energy balance (TSEB) scheme to replace the original Priestly–Taylor (PT) approximation and then optimized both models (Gc-TSEB and PT-TSEB) at pixel scales using qualified MODIS LST data. The results show that using LST is almost as effective in the calibration as using flux measurements. This is promising because unlike flux measurements, LST can be acquired at various spatial resolutions by remote sensing, which makes model calibration feasible for any land pixel. In addition, ET and its partitioning between the canopy and soil layers were found to be reasonable at both validation sites. The day to day and diurnal variations of the predicted ET generally matched the trends and peaks of the flux measurements, although systematic biases were also found due to the decoupling effect of soil moisture at different depths. Furthermore, both models are robust with ±50% changes of SM or LAI because the parameters were automatically adjusted by the LST-calibration. The models are sensitive to LST. However, if the added noise of the LST is less significant than N(±1, 2.52), the medians of the RMSEs in the LE predictions from the LST-calibrated models were quite similar to those from the flux-calibrated models. Both models were found to be accurate, but Gc-TSEB provides slightly more precise and robust predictions than PT-TSEB.