Estimation of latent heat flux over savannah vegetation across the North Australian Tropical Transect from multiple sensors and global meteorological data

Latent heat flux (LE) and corresponding water loss in non-moisture-limited ecosystems are well correlated to radiation and temperature. By contrast, in savannahs and arid and semi-arid lands LE is mostly driven by available water and the vegetation exerts a strong control over the rate of transpiration. Therefore, LE models that use optical vegetation indices (VIs) to represent the vegetation component (transpiration as a function of surface conductance, G_s) generally overestimate water fluxes in water-limited ecosystems. In this study, we evaluated and compared optical and passive microwave index based retrievals of G_s and LE derived using the Penman-Monteith (PM) formulation over the North Australian Tropical Transect (NATT). The methodology was evaluated at six eddy covariance (EC) sites from the OzFlux network. To parameterize the PM equation for retrievals of LE (PM-G_s), a subset of G_s values was derived from meteorological and EC flux observations and regressed against individual and combined satellite indices, from (1) MODIS AQUA: the Normalized Difference Water Index (NDWI) and the Enhanced Vegetation Index (EVI); and from (2) AMSR-E passive microwave: frequency index (FI), polarization index (PI), vegetation optical depth (VOD) and soil moisture (SM) products. Similarly, we combined optical and passive microwave indices (multi-sensor model) to estimate weekly G_s values, and evaluated their spatial and temporal synergies. The multi-sensor approach explained 40–80% of LE variance at some sites, with root mean square errors (RMSE) lower than 20 W/m² and demonstrated better performance to other satellite-based estimates of LE. The optical indices represented potential G_s associated with the phenological status of the vegetation (e.g. leaf area index, chlorophyll content) at finer spatial resolution. The microwave indices provided information about water availability and moisture stress (e.g. water content in leaves and shallow soil depths, atmospheric demand) at a high temporal resolution, thereby providing a scaling factor for potential G_s. We applied the newly proposed G_s model to estimate LE at regional scale using global meteorological data. Our derivation could be extended to continental scales providing equally robust estimates of LE in arid and semi-arid biomes. A more accurate estimation of G_s and LE across different savannah classes will improve the analysis of water use efficiency under drought conditions, which is of importance to climate change studies of water, carbon and energy cycling.