As one of the most important spectral invariants, the recollision probability (p) is a crucial physical parameter that describes the interaction process between the photon and the canopy. This study discusses the applicability of the recollision probability in the thermal infrared (TIR) domain by Monte Carlo simulation. By extending the recollision probability from the VNIR to the TIR, an Improved Canopy Brightness Temperature model based on the p theory (ICBT-P) is introduced to simulate the canopy thermal radiation directionality. A cross-validation between the analytical ICBT-P model and 3D DART model was conducted, yielding R2values over 0.99 in two homogeneous scenarios where the leaf and soil components exhibit a significant temperature difference. In other two homogeneous scenarios where the leaf and soil components have the same temperature, the R2values are over 0.98. The average brightness temperature difference across the four typical scenes is 0.08K. This new model (i.e., ICBT-P) has the potential to be used for the canopy temperature estimation and the component temperature separation of leaves and soil.

Ecosystem services (ESs) indicate that species comprising an ecosystem can provide services that are significant to human beings. ES classification of vegetation is significant for accelerating ecological restoration and achieving sustainable development. However, current vegetation classification product cannot effectively reflect their ESs. This study focused on the north Tianshan Mountain, Xinjiang, China, and tried to classify ESs of vegetation based on machine learning and remote sensing techniques. This study utilized multiband Landsat-8 surface reflectance data from Google Earth Engine (GEE) together with meteorological data and Digital Elevation Model data. The classification map accurately portrayed the spatial distribution of seven types of vegetation ES, including agricultural production (AP) cropland, sandstorm prevention (SP) and water retention (WR) forest, grassland, and shrub. The classification results indicated that, the two types of grassland occupy the largest area besides desert, followed by AP cropland, whereas SP forest and WR shrub have the smallest area. The accuracy of ES classification for vegetation is very high, with an impressive overall accuracy of 89.61% and a kappa coefficient of 0.874. This study provides valuable insights into ES classification of vegetation, which can serve as a reference for formulating an ES development plan in north Tianshan Mountain, Xinjiang.

The fraction of absorbed photosynthetically active radiation (FAPAR) is directly linked to the estimation of canopy gross primary production and is a key input parameter in terrestrial ecosystem models. Diffuse FAPAR calculation is necessary for cloudy conditions. Direct and diffuse FAPAR require distinct simulation methods due to different transfer paths, yet most models do not consider diffuse FAPAR independently. In this study, an improved FAPAR-P model （ FAPAR-Pro ） based on the spectral invariant theory, which considers direct and diffuse FAPAR separately, was proposed to simulate all-sky vegetation FAPAR. Based on the proposed FAPAR-Pro model, along with the hourly ratio of diffuse radiance data calculated from Himawari-8 products, the all-sky vegetation FAPAR was calculated. Validation performed with ground measurements showed that the overall RMSE and MAE values are 0.056 and 0.067 for diverse types of vegetation canopies and different diffuse radiation conditions. The results indicate that the model was applicable to diverse types of vegetation canopies and different radiation conditions, and therefore, may provide support for continuous FAPAR time series simulation and various ecological applications in the future.

Leaf area index (LAI) is a critical indicator for modeling global biosphere–climate interactions. Accurately measuring forest LAI significantly affects the objectivity and accuracy of LAI remote sensing products. DHP is the most widely used instrument to measure forest LAI. However, it is challenging to measure forest LAI accurately using DHP at meter-scale as the wide viewing angle range of DHP is not appropriate for validation of high-resolution pixels. To address this issue, a geometric-based method was proposed to obtain the optimal view zenith range for DHP-based LAI field measurements at meter-scale in this study. Three virtual forest scenes were generated by LESS (LargE-Scale remote sensing data and image Simulation framework) and field measurements at Saihanba, in northern China was collected. The validation results indicate a good agreement between LAI estimation using proposed method and reference LAI datasets. The R2between the LESS-derived LAI and DHP-derived LAI are all greater than 0.9, and the RMSE are all less than 0.1 in simulated scenes. Meanwhile, the R2between the UAV lidar-derived LAI and DHP-derived LAI is 0.668 with RMSE of 0.376. In conclusion, this study holds potential in measuring forest LAI at meter-scale.

Leaf area index (LAI) field measurements based on digital hemispherical photography (DHP) and LAI-2200 instruments have been widely used for remote sensing validation in forestry. Both DHP and LAI-2200 utilize fish-eye lens to capture the largest footprint of a canopy with a wide range of view zenith angle (VZA). However, accurately measuring field LAI at high spatial resolution poses a significant challenge since the view scope of fish-eye sensor is much larger than the size of high spatial resolution pixel. Therefore, selecting appropriate VZA ranges is crucial to address this issue. In this letter, we propose an improved geometry-based method that considers the average tree height, crown depth, and high-resolution pixel size. To validate this method, we designed four simulated forest scenes with different crown shapes through the LargE-Scale Remote Sensing Data and Image Simulation Framework (LESS) model and conducted field measurements. The results indicate that our proposed method significantly enhances the accuracy of fisheye-based LAI field measurements at high spatial resolution compared to previous method, with an almost 70% reduction in RMSE. In addition, our method exhibits greater improvement in measuring forest LAI at high resolution with DHP (RMSE < 0.3) compared to LAI-2200 (RMSE < 0.5). In conclusion, our method holds great potential in accurately measuring fisheye-based forest LAI for remote sensing validation at high spatial resolution.