Assessment of thermal infrared kernel-driven models over row-planted canopies
Auteurs
Cao B., Du Y., Bian Z., Dong Y., Zhao F., Hu T., Ran X., Qin B., Li H., Xiao Q., Liu Q.
Référence
National Remote Sensing Bulletin, vol. 25, n° 8, pp. 1710-1721, 2021
Description
The ignoring of the land surface thermal radiation directionality hampered the accuracy improvement of current land surface temperature products. It is urgent to develop a practical method to correct the angle effect for the products. More and more attention has been paid to the thermal infrared kernel-drive model because of its simplicity and accuracy. For natural surface, there are widely used eight kernel-driven models. Their fitting abilities over continuous and discrete canopies is well-known, however, no report discussed their performances over the important row-planted stage. The objective of this study is to assess the fitting abilities of all existing eight thermal infrared kernel-driven models over row-planted canopies based on airborne measured datasets.Two multi-angle directional anisotropy datasets over row-planted vineyards were obtained through the airborne observation in several flights. The experiment sites were located at Château Talbot, Médoc region, 30 km north of Bordeaux, France. All the measurements were combined and corrected from nadir temperature to derive the directional anisotropy at 1° steps for view zenith angle (0°-60°) and view azimuth angle (0°-360°). The multi-angle directional anisotropy values were used as input to estimate the kernel coefficients of two 3-parameter models within the traditional visible and near infrared framework (Ross-Li and LSF-Li), two 3-parameter models within the new thermal infrared framework (Vinnikov and RL), and four 4-parameter models within the new thermal infrared framework (Vinnikov-RL, Vinnikov-Chen, LSF-RL, LSF-Chen). Then, the forward simulated directional anisotropy values of all models were compared taking the airborne measured values as reference.Results show that all eight kernel-driven models cannot simulate the row-structure signatures (i.e., axisymmetric feature). They have an overall large RMSE about 2 K and a low R2 less than 0.7. In addition, the RMSE differences between the models are small. For the east-west canopy, RMSE is between 1.930 K and 2.153 K, R2 is between 0.616 and 0.691. For the south-north canopy, RMSE is between 2.005 K and 2.353 K, R2 is between 0.600 and 0.670. Therefore, developing a new specific kernel for row-structure to improve the fitting ability is demanded in the thermal infrared band currently.
