Modeling the directional anisotropy of fine-scale TIR emissions over tree and crop canopies based on UAV measurements

Land surface temperature (LST) is a vital parameter for the achievement of the surface energy budget and in thorough investigations of water cycle processes. Lightweight thermal infrared (TIR) sensors onboard unmanned aerial vehicles (UAVs) are rapidly becoming key instruments for extracting high-resolution LSTs given the flexibility they offer in capturing different scales. With this expansion, there has been increasing concern regarding the growing demand to obtain a mapping of normalized LST given the directional anisotropy (DA) of surface fine-scale emissions. To date, this topic suffers from a lack of deep analysis and practical solutions for characterizing the DA of fine-scale TIR data from UAV measurements over tree and crop canopies. In this paper, the first objective was to understand the pattern of brightness temperatures (BTs) DAs at a high spatial resolution by using UAV-based multiangle observations and three-dimensional (3D) radiative transfer model simulations. This study highlighted the need for first performing an angular normalization of the BTs of fine-scale pixels prior to any application, as these were easily affected by adjacent pixels and displayed broad spatial variability from 0.5 degrees C to 5.0 degrees C due to 3D occlusion. The second objective of the present study was to appraise the reliability of a modified kernel-driven model, in comparison to the model from which it was derived, with an additional kernel designed to mimic the adjacency effect, plus, a quadratic function used to simplify the estimate of the directional emissivity kernel. The root mean square error of the best fit between the measured UAV dataset and the modified kernel-driven model was approximately 0.65 degrees C, which proves its efficiency since the DA indexes of the BTs were about 1.40 degrees C. This outlined the role of the model to normalize from directional effects the camera image pixels and thereby deliver fine-scale BTs. In addition, results from LESS simulations also demonstrated the good performance of the modified kernel-driven model for simulating the DAs of thermal emissions for both tree and row-planted scenes.