Estimation of evapotranspiration using directional thermal infrared data

Overview & Synopsis

Evapotranspiration (ET) is an important component of the hydrological cycle. In many ET estimation methods, surface temperature is an essential input variable. Surface temperature is remotely acquired in the thermal infrared (TIR) domain with varying view-earth-sun configurations. These multi-angular view geometries can lead to varying proportions of the canopy components being sampled in the FOV, and thus different temperatures being observed. This is termed thermal radiation directionality (TRD). It is important to account for TRD to ensure consistency in flux estimation irrespective of viewing configuration. In this study, we focus on single-pixel surface energy balance (SEB) ET models (and initial analyses on contextual SEB), and seek to analyze how thermal directionality influences such methods. For point-based SEB, we extended the SPARSE model, wherein the original SPARSE was modified to distinguish sunlit/shaded soil/vegetation sources, and coupled with a radiative transfer model that links these four element emissions to the directional thermal radiances as observed by remote sensors. Initial evaluations were carried out using historical data that included TIR sensors at nadir (tree cover, wheat and soybean) or in two directions only, on a homogeneous cover (wheat) for which the directional effects were minimal. To compensate for the absence of measurement sites having directional TIR acquisitions in a sufficient range of viewing angles, we set up a flux station on a heterogeneous canopy in Verdu, Spain. Both algorithms showed no observable differences in their estimation of total fluxes when run with nadir-acquired thermal data. Nonetheless, by incorporating the solar direction and discriminating between sunlit/shaded elements, we observed that the partitioning of these overall fluxes between the soil/vegetation could be improved particularly in water-stressed environments. The sensitivity of flux and component temperature estimates to the viewing direction of the sensor was tested by using two - simultaneously observed - sets of thermal data (nadir and oblique) to force the models where it was shown that sensitivity to viewing direction is significantly reduced. Over the heterogeneous Vine site, the sensitivity of component estimates to the viewing direction of the sensor was also tested by using reconstructed TIR data (nadir and oblique) to force the models. Here, we observed degradation in flux retrieval cross-row with better consistency along rows. Overall, it was nevertheless shown that by using the extended method, the sensitivity to viewing direction can be reduced. Separately, a synthetic study was performed in a homogeneous canopy using the SCOPE model, which is able to provide 'references' of surface components. Here, SPARSE4 was observed to improve directional consistency of the flux retrievals over wide-ranging simulation conditions. As with directional anisotropy, different conditions were observed to influence the retrieval consistency, with the main driver observed to be incoming radiation. Wind speed and vegetation fraction cover (as described by LAI) also influence the directional retrievals. Nonetheless, the retrieval inconsistencies (hence weak model performance) were observed to mostly influence retrievals at high zeniths, with minimal effects on near-nadir based retrievals. On how directionality influences the (T-fc) space as applied in contextual ET, initial experiments carried out utilizing the SMEX02 data and SCOPE show insignificant TRD effects when the AOI is close to the satellite's sub-track and in small scenes. This however does not necessarily apply when the AOI is located further off the sub-track (especially in the hotspot region). For a better contextual representation, other aspects (e.g. surface roughness) need to be considered. Nonetheless, simple positioning tools (i.e. Sun path diagrams + specifications of view geometry) can qualitatively help identify this TRD exposure.