Land surface temperature (LST) is a vital parameter for energy budget and land surface process models. With the development of high spatial resolution remote sensing technologies, especially the wide application of unmanned aerial vehicle (UAV) in land surface observations, the acquisition of high-resolution thermal infrared (TIR) data has made the accurate extraction of varied LST products possible. However, the complex heterogeneous structure of land surface may cause adjacency effect, i.e., multiple scattering of radiation and geometric occlusion, which is inevitable in high-resolution UAV-based observations. The existing TIR temperature retrieval methods mainly focus on the one-dimensional (1D) scenarios, lacking consideration for the three-dimensional (3D) structure of actual ground objects, resulting in significant errors in extracting LST from high-resolution data. To address these limitations, we propose a novel retrieval method to eliminate the influence of multiple scattering caused by finer spatial resolution on the accuracy of LST. This method introduces a 3D radiative transfer model to account for the radiative transfer in a heterogeneous scenario, based on remotely sensed imagery, corresponding 3D structure of the ground surface and component attributes. An optimization strategy is adopted to iteratively converge toward physically consistent LST results. The proposed retrieval method is validated using the UAV-based TIR images and in-situ surface temperature measurement data obtained from the Huailai remote sensing test site in Hebei province, China. Simulated TIR image datasets of typical vegetation and urban scenes were additionally employed to validate the applicability at different values of key parameters. The factors influencing the adjacency effect were extensively analyzed. Results show that 1) the effects of adjacent objects on LST can result in an overestimate exceeding 0.9 K in cases of typical vegetation scene for TIR observations when the spatial resolution was finer than 1 m; 2) the proposed LST retrieval method based on a 3D radiative transfer model can significantly reduce the influence of adjacency effect on the accuracy of LST; 3) in addition to the sky view factor (SVF), the irradiance from the adjacent objects can also have a significant impact on the accuracy of the temperature retrieval, especially when the emissivity is relatively low. This retrieval method provides a solution for high-resolution near-surface UAV/airborne TIR data and a promising framework for enhancing the LST accuracy using multi-source geographic information assistance.

Three-dimensional (3D) radiative transfer models (RTMs) have emerged as crucial tools for quantifying uncertainties in remote sensing observations. However, most current 3D radiative transfer modeling is based on static scenes, which differ significantly from the dynamic facts of the real world. The UAV shadow and natural wind were also not considered in previous 3D RTM studies. These simplifications in the modeling caused uncertainties in the 3D radiative transfer modeling. To improve the physical realism of 3D radiative transfer modeling, enhancing the model capacity to dynamically simulate the radiative transfer process of forest scenes under physical disturbances in real-time is essential. This study proposed RAPID-RTX, a novel real-time radiative transfer and force field modeling framework for the RAPID model based on Universal Scene Description (USD) and Ray Tracing Xtreme (RTX) rendering technologies. To quantify the effects of UAV shadows and natural wind on UAV-observed forest BRFs, we used RAPID-RTX to simulate forest scenes with and without UAV shadows under multiple viewing angles, and to assess the impact of wind-induced tree movement. Results showed that UAV shadows significantly impact forest BRF, especially at viewing directions near hotspot direction. In the simple plantation scene, the maximum impact reached 83.9% in the red band and 73.7% in the NIR band, whereas in the dense forest scene (Wytham Woods 2015), the maximum impact reached 51.5% and 40.6% in the red and NIR bands, respectively. To minimize the impact of UAV shadows on UAV observed forest BRF, it is recommended that the sensor FOV is larger than 5° and the flight height is higher than 60 meters. Additionally, at wind speed levels 3 and 5 (4 m/s and 10 m/s), the maximum wind impact on forest BRFs in the red band reaches 23.8% and 28.5%, respectively, while in the NIR band the corresponding values are 26.1% and 33.3%. Therefore, we recommend avoiding remote sensing experiments when the wind speed exceeding 4 m/s.

Photon recollision probability p serves as a critical bond between the spectral characteristics of the canopy at any wavelength and the reflectance, transmittance, or absorptance of the vegetation canopy. However, the estimation of p is predominantly accomplished through canopy structural parameters. In this work, we provide a method to estimate canopy p-value directly from Airborne Laser Scanning (ALS) data. The method was evaluated both in virtual experiments and field measurements. The virtual experiments employed the large-scale remote sensing data and image simulation model (LESS) to simulate virtual ALS scanning data based on three RAdiation transfer Model Intercomparison (RAMI) actual canopies. Our findings showed that p-values can be accurately estimated from ALS point clouds.

3D radiative transfer models (3D RTMs) serve as essential tools for accurately retrieving forest products from multi-source earth observation data and comprehending complex radiation signals from various forest elements, such as branches, leaves, soil and their underlying physical interactions. However, due to the extensive parameter inputs required by the 3D RTM, especially structural parameters, it has become a hindrance to the development of such processes. Although LIDAR provides precise structural information about vegetation, the construction of forest canopy geometric models from LIDAR data for the simulation of reflectance has not been thoroughly investigated. In this study, A method for constructing an explicit forest geometric structure is proposed from ALS point cloud data, and the fitted tree crown's Plant Area Density (PAD) is subsequently estimated by utilizing point cloud intensity information. The various representations of tree crowns (alphashape, voxel, ellipsoid) from ALS are analyzed and compared. Our results demonstrate that the geometric model constructed based on ALS point cloud data can robustly simulate reflectance using 3D RTMs. This approach also holds the potential for relatively high-resolution applications.

Remote sensing observations and landscape modelling are important methods to promote remote sensing accuracy in forestry. However, the current model scenes still have simplifications in the three-dimensional structure and spatial heterogeneity, and they are mostly static scenes, which deviate from the real world. This study proposes a virtual remote sensing observation method based on digital twin technology. A fine 3D scene was reconstructed by digital twin technology, where a computer simulation model was used to achieve the simulation of the radiative transfer process, and a coupled wind model was used to simulate the effect of wind on BRF observations. The results show that the effect of vertical wind is greater than that of parallel wind, and the trend of BRF change is related to the difference between the canopy and the ground surface.

Vegetation Leaf Area Index (LAI) and Leaf Chlorophyll Content (LCC) are crucial indicators for monitoring vegetation growth. Remote sensing combined with radiative transfer models (RTMs) is the primary method for estimating vegetation parameters. However, traditional 1D RTMs make idealized assumptions that are not suitable for fine-scale inversion. Although existing 3D RTMs can intricately consider the radiative propagation mechanisms of complex vegetation structures, they require high computational costs, making it challenging to conduct fine-scale vegetation parameter inversion at a regional scale. In this study, a mixed forest scene is utilized as the foundational application scenario. A semi-empirical 3D acceleration model, named Semi-LESS, is used for fast simulation to generate a large dataset of 3m spatial resolution multispectral imagery. A residual network is used to encode vegetation structure and establish relationships between multi-band reflectance and vegetation parameters for vegetation parameter inversion. Our results indicate favorable outcomes for LAI and LCC inversion in the mixed forest, with RMSE values of 0.24 and 3.68ug/cm2, respectively. The conclusion highlights the capability of quickly achieving fine-scale inversion of vegetation parameters using the coupled Semi-LESS and deep learning. Our approach also holds potential for achieving rapid fine-scale vegetation parameter mapping in orchards and agricultural scenes.

In the field of forestry, there are increasing cases of optical-microwave joint applications. However, most of these cases involve direct statistical regression or machine learning approaches to combine the two data types, with few implementations based on underlying mechanisms, which limits the improvement of inversion accuracy. Therefore, it is necessary to explore the relationship between these two signals and the underlying connection mechanism between them. We first analyzed the correlation between Landsat 8 and Sentinel-1 signals in three regions: Genhe City in Inner Mongolia, Jilin City in Jilin Province, and Sanming City in Fujian Province using the Google Earth Engine (GEE). Subsequently, we employed the three-dimensional radiative transfer model RAPID2 to simulate the forests of the representative tree species in the aforementioned regions based on allometric growth equations. The simulation was used to analyze the correlation between optical and microwave signals and how their relationship varied with changes in branch-to-leaf biomass. We have preliminary conclusions: the correlation between optical and L-band microwave signals may be due to the inherent allometric growth relationship of plants, while the correlation between optical and C-band microwave signals may be due to the combined effects of both signals on the foliage.

Predicting the dynamic process of post-fire regrowth is critical for understanding the specific forest succession trajectory. 3-PGmix model (i.e., Physiological Principles in Predicting Growth for mixed stands) has been reported as a powerful tool for predicting the growth of mixed forest species. However, the prediction of post-fire forest productivity at a regional scale is infrequently documented and not well understood. In this study, we used remote sensing vegetation parameters to retrieve a series of site-specific parameters for driving the 3PGmix model to simulate the spatial-scale dynamics of post-fire vegetation net primary production (NPP) recovery and predict the response of NPP under different future climate conditions. The result indicated that the extended 3PGmix model can accurately simulate the post-fire dynamic of NPP at spatial scales and the predictions are consistent well with the LAI estimated NPP based on the 3PGS model. A higher fertility rating (FR) was predicted to accelerate the process of post-fire forest successional and shorten the duration time when the species proportion achieves balance and Climate change promoted the increase of NPP in the sequence of RCP 8.5 > RCP 4.5 > current climate.

Tree species information is a crucial factor in forest resource inventory. Light detection and ranging (LiDAR), as an emerging active remote sensing technology, has unique advantages in extracting three-dimensional (3-D) vegetation structure information, and its application in forest resource assessment and research is gaining increasing attention. Airborne laser scanning (ALS), unmanned aerial vehicle laser scanning (UAVLS) and terrestrial laser scanning (TLS) are important means to acquire 3-D forest data. The challenge of traditional machine learning based tree classification lies in extracting and selecting numerous key diagnostic features from large amounts of LiDAR data, requiring extensive feature extraction expertise, which limits its scalability. The use of deep learning methods for fast and accurate classification and identification of tree species in individual tree point clouds represents a new development direction of LiDAR technology in forest resource inventory applications. In this study, PointNet++ was used to classify tree species by point cloud data obtained from TLS, ALS and UAVLS, respectively. The research results show that high accuracy in tree species classification can be achieved by using point cloud deep learning methods.

Estimating change in tree growth is important for monitoring forest dynamics, even the whole terrestrial ecosystem. Multi-temporal airborne laser scanning data had been used to accurately assess and predict change in forest attributes such as aboveground biomass (AGB), aboveground carbon density (ACD) and forest growth. In this study, we assessed the ability of dual-temporal airborne laser scanning data and dual-date to analyze forest growth change in Greater Khingan Mountains, Northeastern China.

Monitoring insect and disease disturbances in the lower parts of the forest has always been a hot research topic. Hyperspectral LiDAR (HSL), a new sensor, makes it possible to monitor these pest disturbances. In this study, we used the 3D radiative transfer model LESS to simulate AHSL point cloud data and conducted a sensitivity analysis of the point cloud data for monitoring forest insect and disease. A virtual forest scene was first reconstructed with explicit geometric structures using terrain laser scanning (TLS) data and ground measurement data. Based on measured optical properties of damaged foliage, some different damage scenarios with different stress levels were defined. AHLS and hyperspectral imagery (HI) were simulated for different forest pest disturbance scenarios. For the AHSL point cloud data, we select a commonly used stress index, RENDVI, as the evaluation vegetation index for assessing the sensitivity of the hyperspectral LiDAR signal in monitoring insect and disease disturbances. According to different damaged locations, different damage scenarios were rasterized into images. The corresponding hyperspectral images were also compared with the AHSL data. The result show that AHSL has great potential for monitoring forest insect and disease disturbances compared to HI. This study demonstrates that AHSL has great potential in responding to spectral signal changes in the lower and middle parts of the forest, and which may also be a powerful tool for early detection of forest pest disturbances.

Accurate tree species information is a prerequisite for forest resource management. Combining light detection and ranging (LiDAR) and image data is one main method of tree species classification. Traditional machinelearningmethods rely on expert knowledge to calculatea large number of feature parameters.Deep learning technology can directly use the original image and pointclouddata to classify tree species. However, data with different patterns require the use of different types of deeplearningmethods. In this study, a multimodal deeplearningframework (TSCMDL) that fuses 2D and 3D features was constructed and then used to combine data from multiple sources for tree species classification. This framework uses an improved version of the PointMLP model as its backbone network and uses ResNet50 and PointMLP networks to extract the image features and pointcloudfeatures, respectively. The proposed framework was tested using UAV LiDAR data and RGB orthophotos. The results showed that the accuracy of the tree species classification using the TSCMDL framework was 98.52%, which was 4.02% higher than that based on pointcloudfeatures only. In addition, when the same hyperparameters were used for training the model, the efficiency of the model training was not significantly lower than for models based on pointcloudfeatures only. The proposed multimodal deeplearningframework extracts features directly from the original data and integrates them effectively, thus avoiding manual feature screening and achieving more accurate classification. The feature extraction network used in the TSCMDL framework can be replaced by other suitable frameworks and has strong application potential.

Within-crown heterogeneity exists and needs to be concerned in the field of radiative transfer modeling on vegetation, such as forests containing trees with heterogeneous distribution of damaged foliage. The existing radiative transfer approaches implement the within-crown heterogeneity either using complicated 3D scenes with excessive amounts of parameters, or simplifying the scenes to homogeneous cases by averaging canopy optical properties. We propose a simulation method for canopy bi-directional reflectance factor (BRF) of forests with different foliage damage distributions based on the stochastic radiative transfer (SRT) theory. First, each damaged tree crown was classified into one of three types: top-only damage, bottom-only damage and random damage. Second, canopies with such three types of damage distributions were modeled by extending stochastic moments of SRT. Then, simulations by two 3D models were conducted to evaluate extended SRT model performance. Results suggest that the extended SRT has the capability to model forests with heterogeneous foliage within crowns in an efficient way, and agrees well with other models. It provides a framework to analyze the sensitivity of canopy reflectance to heterogeneous damaged foliage, along with the change of other key scene factors. This model has the prospects for forest damage quantitative inversion.

Fire is a major disturbance in Daxing'anling Mountain in China, affecting ecosystem carbon cycle significantly. However, the return of forest productivity at regional scale in high fire disturbance has not been well documented and understood. The objective of this paper was to evaluate remote sensing driven 3-PG model for simulating the forest recovery in high-intensity burnt areas. Differences in productivity among sites were accounted for by only changing the values of fertility rating and climate condition which calculated from remote sensing data. The adapted model performed well in the experimental plots, and accurately estimated above ground biomass and diameter growth. Then, a case study was conducted on the 2001 burnt area, and NDVI images simulation used for validating the effectiveness of the model in a large scale area.

It is of great interest to inverse the vertical heterogeneity of physiology of a forest canopy using remote sensing. However, there is very radiative transfer models to focus on this effect. The objective of this paper is to properly use a three-dimensional (3D) model, the RAPID (Radiosity Applicable to Porous IndiviDual objects) model, to explore the bidirectional reflectance properties of trees caused by a special pest damage. The innovation is to create a damaged tree with spatially explicit shoot distribution. Three types of spatial distribution of damaged shoots were simulated: totally on the top, fully at bottom (down) and randomly scattered in the crown. The corresponding bidirectional reflectance factor (BRF) of these forest canopies were simulated under different level of damaged shoots ratio (SDR). A case study was conducted on Yunnan Pine canopies attacked by shoot beetles. Forest scene image simulation used for validating the effectiveness of the model.

Multi-angle information extraction is very important for the validation of land surface models, such as bi-directional reflectance distribution function (BRDF) model and directional brightness temperature (DBT) model. The experiment can be done in the near surface, on the plane, and on the satellite. The spatial-temporal average method is widely-used on the scale of plane. In this paper, we try to extend it to extract the multi-angle information from near surface observation sensors. We found this method can obtain good result over homogeneous land surface, but the observation protocol is critical over heterogeneous land surface. For instance, the observation plane should be perpendicular to the row direction. In addition, we found that the hot spot area will be shaded by the observation platform which leads to the BRDF result to be affected seriously.

In order to assess atmospheric effects on the directional reflectance of land surface, we have developed a new approach coupling the 3-D radiosity-based land-surface model [radiosity applicable to porous individual objects (RAPID)] with the atmospheric radiative transfer (RT) model [vector linearized discrete ordinate RT (VLIDORT)]. RAPID is used to generate a lookup table of bidirectional reflectance distribution function (BRDF) elements required by VLIDORT for the surface boundary condition. To test the RAPID-VLIDORT model, we used five natural 3-D scenes along with five aerosol optical depths (AODs). Results for top-of-atmosphere radiances show semiempirical analytical BRDF models are insufficiently accurate to represent bidirectional reflectance factors (BRFs) in hotspot regions and over wide angular variations. The large impact of AOD on BRF hotspot also underlines the importance of precise atmospheric corrections for multiangular remote sensing of the earth's surface.

Land surface temperature (LST) is often needed for using remotely sensed data to study the surface energy budget and hydrological cycle. However, LST is challenging to measure and simulate because of its high sensitivity to atmospheric instability and solar angle, particularly over large-scale heterogeneous scenes. We propose a model that combines radiosity theory and an energy budget method for surface temperatures; we also explore the anisotropic behavior of row-planted crop emissions. The surface thermodynamic equilibrium state is fulfilled via the interaction between the 3-D radiative transfer calculations of the thermal-region radiosity-graphics combined model and the energy balance equation. Despite its shortcomings, such as the time-consuming calculations, the proposed model is feasible according to the results of an intercomparison and validation analysis. The intercomparison shows that the model exhibits similar performance, in terms of surface temperature calculations, to that of the soil-canopy observation, photochemistry and energy balance model (root-mean-square differences) of 0.59 °C and 1.77 °C for the leaf and soil components, respectively. Excellent agreement with the observed directional variation over summer maize canopies is also obtained, with R2values exceeding 0.6 and a mean RMSE of 0.32 °C. Thus, we recommend the new combined model as an option for explaining directional anisotropy due to its potential application to 3-D scenes.

To support ground experiments for multi-sensor fusion in remote sensing, it is necessary to efficiently simulate realistic sensor images at various spatial scales from a few meters to thousands of meters and different sensor supports. The RAPID model, Radiosity Applicable to Porous Individual object, was enhanced to landscape scale (1 km) with natural vegetation and artificial opaque objects as well as functions of atmospheric effect, terrain roughness, perspective projection and fisheye projection.

This study analyzed the relationship between the spatial heterogeneity of land surface temperature (LST) and the spatial scale using the Thermal Airborne Hyperspectral Imager (TASI) and satellite-based Advanced Spaceborne Thermal Emission Reflection (ASTER) data. The spatial heterogeneity of LST was quantified using variogram modeling in univariate and multivariate method. The results show that in both methods, the spatial heterogeneity of LST in a landscape as quantified by the dispersion variance increases with the spatial scale until the scale is larger than the characteristic scale, and the land cover types have significant influences on the spatial heterogeneity of LST. Additionally, the spatial heterogeneity of the land surface decreases obviously as the wavelength increases in the multivariate model.

This study assessed the capability of microclimate model ENVI-met to simulate three-dimensional (3D) land surface temperature (LST) in clear-sky periods. Field measurement data were collected in 3 days to drive and validate the model at point scale. Furthermore, MODIS LST products and Landsat-8 thermal infrared bands were used to validate daily variations and the spatial heterogeneity of simulation, respectively. The results indicate that ENVI-met model is capable of reasonably modeling the diurnal thermal behavior of different ground surfaces in the park, and generate continuous 3D temperature distribution, which provides input for 3D thermal remote sensing models or prior knowledge on monitoring applications.

A multi-level radiation transfer method was proposed to simulate directional reflectance and images for 3D landscapes. This study emphasizes a unified framework to generate multi-level structures from leaf to landscape. Radiosity is adapted to improve calculation efficiency. The program was preliminarily evaluated and further applied to analyze how directional reflectance varies with fragmentation and terrain relief. Results indicate that grass dominated landscapes are more likely to be affected on directional reflectance feature by fragmented forest patches. Also, topographical relief can strongly enhance the hotspot effect and directional variations in the solar principal plane.

Remotely sensed land surface temperature (LST) is of great value to the research in the fields of climatology, hydrology, ecology, and biogeochemistry,as well as a wide range of interdisciplinary research areas, since it isan efficient and practical way of acquiringtemperature variability globally and continuously. In the paper, the generalized split-window algorithm proposed by Wan and Dozier (1996) is used to estimate LST from Visible and Infrared Radiometer (VIRR) onboard the second generation of China's polar-orbiting meteorological satellite (FY3A).MODTRAN 4.0 and the Lhermodynamic Initial Guess Retrieval database 3 (TIGR-3) are used to simulate the data for fitting the algorithm's coefficients. Lhe algorithm fitting accuracy is improved by dividing the LST, the average emissivity (ε) and the water vapor content (WVC) into several sub-ranges. Finally, the validation at five locations is performed and the results show thatthe LSTs estimation from FY3A/VIRR dataagree with the ones extracted from the MODIS 1 km LST products very well.

A two-level radiosity algorithm was developed here for the computation of global radiation transfer at complex land cover called HRAD, which rapidly simulates the surface leaving radiances in three dimensional (3D) landscapes. The algorithm is an integration of a ray casting module and an adapted hierarchical radiosity method with heterogeneous volume clusters. The time cost of our algorithm was nearly linear to the polygon number. The algorithm was first validated by comparing the simulated radiances with that of Radiosity-Graphics combined Model (RGM) at small scenes. The single scattering results were highly correlated with that of RGM (R2> 0.99 and RMSE2> 0.99 and RMSE2> 0.98 and RMSE = 0.015). By using the Gauss scene in RAMI-III website, we also validated our algorithm.

A physical decision method was proposed here to monitor Larch forest insect pests at early stage. Three remote sensing indicators were defined, which are CWC (canopy water content), TVDI (Temperature/Vegetation Dryness Index) and LAI (Leaf Area Index). The Five-scale model and artificial neural network (ANN) were combined to inverse the three factors from Landsat data. Based on training samples of health or attacked pixels, a decision tree was built to classify pest-infected pixels. Field validation showed that the prediction of forest compartments with insect pest were highly consistent with the ground field data.

The Effective Plant Area Index (PAIe) of forest canopy is an important parameter in the canopy reflectance modeling and validation. PAIe can be transformed to leaf area index (LAI) with clumping index. But it is still very difficult to obtain its ground truth value by in situ measurement. In this study, we measured PAIe of 3 typical wood land sites in China by means of three indirect optical techniques: plant canopy analyzer LAI-2000, TRAC, and Digital Fisheye Camera. The 5 measured stands include Qinghai spruce, peach, poplar, willow and silver chain. In this paper, the forest canopy PAIe measured by those three instruments is compared firstly. Then our new approach is how to use the three measured data to get the better PAIe estimation with less overall error. The method of minimizing overall error is adopted to produce PAIe of the sample plots in our study sites, which has less overall error comparing with the PAIe measured by each individual instrument. This approach is also validated by using computer simulated wide-angle viewing pictures when true LAI/PAIe values are given.

Radiosity-graphics combined model (RGM) has been proposed to calculate the radiation regime and bidirectional reflectance distribution function of complex 3D scene, which is limited in visible and near-infrared wavelength (0.3-3 mum) region. In this paper, RGM is extended to thermal region (named as TRGM) based on thermal-radiosity theory and thermal-emission directionality of vegetation canopy. The TRGM has been implemented on Microsoft Windows platform, and a parameterization scheme for crop canopies is introduced in this paper. It is then evaluated by comparing with two row-crop directional thermal emission models and one thermal radiative-transfer model. Field experiment data has been used to validate the TRGM for row structural wheat and maize canopies. The root mean square error of directional brightness temperature (DBT) is smaller than 1.0degC for the wheat canopy and 0.5degC for the maize canopy while the canopy DBTs vary more than 4degC. Model sensitivity analyses have also been conducted to illustrate influences of component temperature distribution, component emissivity, incident atmospheric radiation, and canopy structure on the crop canopy DBT.

Modeling the soil component temperature distribution is useful to study multi-angular thermal remote sensing. SVAT (soil-plant-atmosphere transfer) model could be a good choice because it can predict canopy temperature distribution. However, most of them, including CUPID model 111. were unable to separate shade soil and sunlit soil. They only gave a single temperature for the soil surface. In this paper, based on the difference of net radiance and evaporation rate between the shade and sunlit soil, an extended model from CUPID was proposed to simultaneously retrieve the shaded temperature and sunlit temperature of soil surface. The comparison showed good agreement between simulated soil temperatures and measured ones.