A line-spread kernel function for angular anisotropy in row-dominated heterogeneous scenarios

Abstract

Land Surface Temperature (LST) is a fundamental variable for determining mass (water, carbon) and energy surface fluxes. LST can be obtained from remote sensing but under varying configuration geometries that create directional effects due to the inherent anisotropy properties of most terrestrial targets. Actually, thermal infrared (TIR) measurements obtained from satellites or unmanned aerial vehicles (UAV) are seriously impacted by varying viewing and solar geometries (Cao et al., 2019). In this regard, a computationally efficient approach to handle them is using kernel-driven models (KDM), as they were shown to be an effective solution. However, in high-resolution scenes, the structural features can be very detailed and, in this case, the assumption of homogeneity in considering traditional KDM no longer holds. This is why we propose to develop a novel KDM that is able to handle typical heterogeneous scenes whose structure is dominated by rows. Rather than improving existing point-spread kernels, we propose a line-spread kernel considering the row orientation and radiative occlusion. This new KDM is validated with both airborne measurements and simulated datasets generated by three-dimensional radiative transfer models. Results indicate that: 1) This proposed Heterogenous KDM captures the directional anisotropies of temperatures in row-planted vineyard canopies, whereas the traditional point-spread KDM show limitations. In most cases, root mean squared errors (RMSE) improved up to 0.5 K. 2) A sensitivity analysis based on simulated datasets also showed a better performance of the new proposed KDM under different cases including LAI and row height/width. 3) Further simple validation using UAV and sandbox measurements has demonstrated the effectiveness of the proposed KDM in urban and mountainous areas, where stripe characteristics in thermal radiation directionality are present. In conclusion, this study proposes a novel KDM with significant practical implications for heterogeneous scenarios.