Remotely sensed land-surface energy fluxes at sub-field scale in heterogeneous agricultural landscape and coniferous plantation

Abstract

In this study we evaluate a methodology for disaggregating land surface energy fluxes estimated with the Two-Source Energy Balance (TSEB)-based Dual-Temperature Difference (DTD) model which uses day and night polar orbiting satellite observations of land surface temperature (LST) as a remotely sensed input. The DTD model is run with MODIS input data at a spatial resolution of around 1 km while the disaggregation uses Landsat observations to produce fluxes at a nominal spatial resolution of 30 m. The higher-resolution modelled fluxes can be directly compared against eddy covariance (EC)-based flux tower measurements to ensure more accurate model validation and also provide a better visualization of the fluxes' spatial patterns in heterogeneous areas allowing for development of, for example, more efficient irrigation practices. The disaggregation technique is evaluated in an area covered by the Danish Hydrological Observatory (HOBE), in the west of the Jutland peninsula, and the modelled fluxes are compared against measurements from two flux towers: the first one in a heterogeneous agricultural landscape and the second one in a homogeneous conifer plantation. The results indicate that the coarse-resolution DTD fluxes disaggregated at Landsat scale have greatly improved accuracy as compared to high-resolution fluxes derived directly with Landsat data without the disaggregation. At the agricultural site the disaggregated fluxes display small bias and very high correlation (r ≈ 0.95) with EC-based measurements, while at the plantation site the results are encouraging but still with significant errors. In addition, we introduce a~modification to the DTD model by replacing the "parallel" configuration of the resistances to sensible heat exchange by the "series" configuration. The latter takes into account the in-canopy air temperature and substantially improves the accuracy of the DTD model.