Simulation of urban thermal anisotropy at remote sensing pixel scales: Evaluating three schemes using GUTA-T over Toulouse city

The directional variation in upwelling thermal radiance (known as ‘thermal anisotropy’) affects our understanding of urban land surface temperature (LST) from remote sensing observations. Parametric models have been proposed to quantify and potentially correct the thermal anisotropy from satellite systems. The accurate specification of the coefficients is critical for broadening applications of parametric models. However, the current research is limited to only one study area or one particular approach which often does not sufficiently offer transformative understanding for effective applications over other metropolitan areas. This study focuses on systematically evaluating schemes that determine the model coefficients. We use a geometric model to simulate Urban Thermal Anisotropy Time-series (GUTA-T) and propose different approaches to determine the physically-interpretable parameters. The model and solution schemes reduce the uncertainties caused by observational errors and simplifications of complex urban surfaces. The three schemes include estimating parameter values using component urban surface temperatures obtained from model simulations (Scheme #1mod) or field observations (Scheme #1obs) (‘forward’ approach), inverting parameter values from known anisotropy (Scheme #2) and from multi-angular LST observations (Scheme #3) (‘backward’ approach). The three schemes were separately evaluated and compared to an independent airborne dataset. These schemes have consistent results. The root mean square errors (RMSE) between LST anisotropy from the three schemes and airborne measurements are ranked as: Scheme #3 (1.0 K) < Scheme #2 (1.2 K) < Scheme #1 (based on field data) (1.3 K) < Scheme #1 (based on TUF3D simulation) (1.5 K), whereas the overall amplitude of the variation of directional temperature averaged over the 5 flights is 11.9 K. The inverted parameter values from the three schemes agree well with the results from field measurements. The three schemes have advantages and disadvantages, and are expected to be combined depending on the available input data. These schemes represent multiple options to quantify and/or correct the anisotropic impact from remote sensing LST for urban applications.