Desert scene albedo simulation method and system based on kernel driving method

Abstract

The invention relates to the technical field of remote sensing, in particular to a desert scene albedo simulation method and system based on a kernel-driven method, and the method comprises the steps: collecting vegetation coverage data, biological soil crust coverage data and topographic feature data, and carrying out the standardization processing of the data, generating dynamic characteristic data by utilizing time dynamic correlation and spatial distribution analysis; the method comprises the following steps: constructing a volume scattering kernel based on a Gutschick-Wiegel scattering phase function, and generating corrected volume scattering kernel data in combination with SCS + C topographic correction and a hotspot correction mechanism; constructing a geometrical optical kernel by jointly using a Li-SparseR kernel and a Rouyean kernel, and generating corrected geometrical optical kernel data by adjusting the weight through dynamic vegetation coverage parameters and combining topographic correction; and finally, combining the volume scattering kernel, the geometrical optical kernel and the isotropic kernel to generate a bi-directional reflection distribution function model, and optimizing parameters to improve the simulation precision of the red band and the near-infrared band. According to the method, the defects of hot spot effect, terrain correction and sparse vegetation optical characteristic representation in the prior art are effectively overcome.

Description

Technical Field

[0001] The present invention relates to the field of remote sensing technology, and in particular to a method and system for simulating the albedo of a desert scene based on a nuclear driving method. Background Art

[0002] Albedo is one of the important parameters of surface energy balance in desert areas, affecting the simulation of surface radiation and climate change. Traditional albedo simulation methods are difficult to effectively characterize the complex surface structure characteristics of desert areas, such as sparse vegetation, undulating terrain, and the scattering characteristics of biological soil crusts.

[0003] Existing technologies mainly use empirical models or radiation transfer models to simulate the albedo of desert scenes. These methods are often based on unified parameterization assumptions and simplified surface scattering models. For example, the standard BRDF model performs well in densely vegetated areas, but has significant defects in characterizing the reflective characteristics of sparse vegetation and complex desert terrain. Specifically, existing models often ignore the hotspot effect of the desert surface and the influence of local terrain, and fail to fully reflect the complex contribution of sparse vegetation characteristics and biological soil crust to light scattering, resulting in large errors and insufficient stability in albedo simulation results. Summary of the invention

[0004] In view of the many problems existing in the above-mentioned prior art, the present invention provides a method and system for simulating the albedo of desert scenes based on a kernel-driven method. The present invention collects and standardizes vegetation coverage data, biological soil crust coverage data and terrain feature data, combines volume scattering kernel and geometric optical kernel construction, and generates a corrected bidirectional reflectance distribution function model (BRDF). After optimizing parameters by the least squares method, the present invention can achieve high-precision albedo simulation of the red band and near-infrared band of desert scenes, thereby effectively overcoming the problems of insufficient hotspot effect simulation and sparse vegetation optical property error in the prior art.

[0005] like Fig.17 As shown, a method for simulating the albedo of a desert scene based on a kernel-driven method comprises the following steps:

[0006] Collect vegetation coverage data, biological soil crust coverage data and terrain feature data of the target desert scene, standardize the collected data, and generate data representing the dynamic characteristics of the desert scene through spatial distribution analysis and temporal dynamic correlation analysis;

[0007] The volume scattering kernel is constructed based on the Gutschick-Wiegel scattering phase function, and the corrected volume scattering kernel data is generated by combining the SCS+C terrain correction method and the hot spot correction mechanism;

[0008] The Li-SparseR kernel and the Roujean kernel are used together to construct a geometric optical kernel. The weights of the Li-SparseR kernel and the Roujean kernel are adjusted by dynamic vegetation coverage parameters, and the corrected geometric optical kernel data are generated in combination with terrain correction.

[0009] The bidirectional reflectance distribution function model of the desert scene is generated by combining the corrected volume scattering kernel data, the corrected geometrical optics kernel data and the isotropic kernel;

[0010] The bidirectional reflectance distribution function model is verified using field measurement data and satellite observation data, and the simulation accuracy in the red band and near-infrared band is improved by optimizing the model parameters.

[0011] Preferably, the vegetation coverage data, biological soil crust coverage data and terrain feature data are obtained by combining on-site measurement and PARASOL satellite multi-angle observation.

[0012] Preferably, after the data is standardized, dynamic characteristic data representing the state of vegetation in the desert scene is generated through time dynamic correlation analysis and spatial distribution analysis.

[0013] Preferably, the volume scattering kernel is constructed based on the Gutschick-Wiegel scattering phase function, and the corrected volume scattering kernel data is generated in combination with the SCS+C terrain correction method and the hot spot correction mechanism.

[0014] Preferably, the corrected volume scattering kernel data further corrects the backscattering characteristics by introducing Chen hot spot correction formula to improve the simulation accuracy of hot spot effect.

[0015] Preferably, the geometric optics kernel consists of a Li-SparseR kernel and a Roujean kernel, wherein the Li-SparseR kernel is used to describe the geometric properties of sparse vegetation canopies in deserts, and the Roujean kernel is used to describe the optical properties of biological soil crusts in desert scenes.

[0016] Preferably, the geometric optics kernel adjusts the weights of the Li-SparseR kernel and the Roujean kernel through dynamic vegetation cover parameters to adapt to the distribution ratios of different surface cover types.

[0017] Preferably, the bidirectional reflectance distribution function model is combined by corrected volume scattering kernel data, corrected geometric optics kernel data and isotropic kernel, and the weight parameters are optimized by the least squares method to generate desert scene reflectance simulation results suitable for red band and near-infrared band.

[0018] Preferably, the desert scene reflectivity simulation results are verified and parameter optimized by on-site multi-angle observation data, bidirectional reflectance distribution function data acquired by the PARASOL satellite, and simulation data generated based on the LESS radiation transfer model.

[0019] like Fig.18 As shown, a system for implementing the desert scene albedo simulation method based on the kernel-driven method comprises:

[0020] A data collection module is used to collect vegetation coverage data, biological soil crust coverage data and terrain feature data of the target desert scene, and to perform standardization on the collected data;

[0021] A dynamic feature generation module is used to generate data representing the dynamic features of the desert scene from the standardized data through spatial distribution analysis and temporal dynamic correlation analysis;

[0022] Volume scattering kernel construction module, which is used to construct volume scattering kernel based on Gutschick-Wiegel scattering phase function, and generate corrected volume scattering kernel data by combining SCS+C terrain correction method and hot spot correction mechanism;

[0023] The geometric optical kernel construction module is used to jointly use the Li-SparseR kernel and the Roujean kernel to construct the geometric optical kernel, adjust the weights of the Li-SparseR kernel and the Roujean kernel through dynamic vegetation coverage parameters, and generate the corrected geometric optical kernel data in combination with terrain correction;

[0024] A reflectivity model generation module is used to generate a bidirectional reflectance distribution function model of a desert scene by combining the corrected volume scattering kernel data, the corrected geometric optics kernel data and the isotropic kernel;

[0025] The model verification and optimization module is used to verify the bidirectional reflectance distribution function model using field measurement data and satellite observation data, and to improve the simulation accuracy of the bidirectional reflectance distribution function model in the red band and the near-infrared band by optimizing model parameters.

[0026] Compared with the prior art, the advantages and beneficial effects of the present invention are:

[0027] The present invention solves the problem of insufficient processing of hot spot effect and terrain influence under sparse vegetation and complex terrain conditions in the prior art by combining the Gutschick-Wiegel scattering phase function, the SCS+C terrain correction method and the hot spot correction mechanism;

[0028] The present invention uses the technical means of adjusting the weights of the Li-SparseR kernel and the Roujean kernel through dynamic vegetation cover parameters to achieve accurate characterization of the geometric characteristics of sparse vegetation canopies and the optical characteristics of biological soil crusts in desert scenes, significantly improving the simulation accuracy.

[0029] The present invention achieves high-precision simulation of albedo in the red band and near-infrared band through optimized bidirectional reflectance distribution function model construction technology, thereby improving the stability and applicability of the model. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 A schematic diagram of a study area in an embodiment of the present invention;

[0031] Figure 2 A schematic diagram of a computer scene with different vegetation coverage in an embodiment of the present invention;

[0032] Figure 3 This is a distribution diagram of POLDER satellite sampling locations in an embodiment of the present invention;

[0033] Figure 4 It is a schematic diagram comparing the simulated BRF values ​​of different models (MHLSRT, RLKB, RTLSR) and the field measured BRF values ​​in the embodiment of the present invention;

[0034] Figure 5 Schematic diagram of Bias distribution between the three models MHLSRT, RLKB, and RTLSR and the site measurement reference value in an embodiment of the present invention;

[0035] Figure 6 Schematic diagram of the red band (670nm) BRF of desert areas simulated by the four models of LESS, MHLSRT, RTLSR and RLKB in the embodiment of the present invention;

[0036] Figure 7 It is a schematic diagram of the BRF of the near infrared band (865nm) in the desert area simulated by the four models of LESS, MHLSRT, RTLSR and RLKB in the embodiment of the present invention;

[0037] Figure 8 It is a comparison diagram of the simulation results of the main plane MHLSRT hot spot area and the reference value of the LESS hot spot area in the embodiment of the present invention, as well as a comparison diagram of desert areas with different vegetation coverage;

[0038] Fig. 9 It is a comparison diagram of the simulation results of the three models MHLSRT, RTLSR and RLKB in the main plane in the embodiment of the present invention with the LESS simulation reference value, as well as a comparison diagram of desert areas with different vegetation coverage;

[0039] Fig.10 It is a schematic diagram of the absolute deviation of the red band and the near infrared band in the main plane MHLSRT, RTLSR, RLKB three model simulation results and the LESS simulation reference value under different vegetation coverage conditions in the embodiment of the present invention;

[0040] Fig.11 It is a comparison diagram of the simulation results of the three models of the cross main plane MHLSRT, RTLSR, and RLKB in the embodiment of the present invention with the LESS simulation reference value, as well as a comparison diagram of desert areas with different vegetation coverage;

[0041] Fig.12 It is a schematic diagram of the absolute deviation of the red band and the near infrared band under different vegetation coverage conditions between the simulation results of the three models MHLSRT, RTLSR and RLKB of the cross principal plane in the embodiment of the present invention and the LESS simulation reference value;

[0042] Fig.13 It is a schematic diagram for comparing the BRF predicted by the POLDER satellite data and the three models of MHLSRT, RLKB and RTLSR in the red band (670nm) and near infrared band (865nm) in the embodiment of the present invention;

[0043] Fig.14 Schematic diagram of the incident and emitted conditions of photons in an embodiment of the present invention;

[0044] Fig.15 Schematic diagram of the relationship between the volume scattering kernel value and VZA of different models in an embodiment of the present invention;

[0045] Fig.16 Schematic diagram of the relationship between the geometric optical kernel value and VZA of different models in the embodiment of the present invention;

[0046] Fig.17 It is a schematic diagram of the process of the present invention;

[0047] Fig.18 It is a structural block diagram of the system of the present invention. DETAILED DESCRIPTION

[0048] Hereinafter, embodiments of the present disclosure will be described with reference to the accompanying drawings. However, it should be understood that these descriptions are exemplary only and are not intended to limit the scope of the present disclosure. In the following detailed description, for ease of explanation, many specific details are set forth to provide a comprehensive understanding of the embodiments of the present disclosure. However, it is obvious that one or more embodiments may also be implemented without these specific details. In addition, in the following description, descriptions of known structures and technologies are omitted to avoid unnecessary confusion of the concepts of the present disclosure.

[0049] The terms used herein are only for describing specific embodiments and are not intended to limit the present disclosure. The terms "comprise", "include", etc. used herein indicate the existence of the features, steps, operations and / or components, but do not exclude the existence or addition of one or more other features, steps, operations or components.

[0050] All terms (including technical and scientific terms) used herein have the meanings commonly understood by those skilled in the art, unless otherwise defined. It should be noted that the terms used herein should be interpreted as having a meaning consistent with the context of this specification, and should not be interpreted in an idealized or overly rigid manner.

[0051] In order to describe the complex radiation transfer process in desert areas, the present invention derives the volume scattering kernel from the analytical solution of the Gutschick-Wiegel (GW) scattering phase function in the radiation transfer theory. The physical mechanism of this function is to describe the physical process of the interaction between photons and the canopy on a macro scale. Although the microscopic scattering details inside the leaves are ignored, it provides a reasonable first approximation. For scenes with relatively simple structures such as desert vegetation, such an assumption can more accurately reflect its scattering characteristics and accurately capture the "anisotropy of light scattering." Taking into account the computational efficiency when calculating desert scenes, the analytical solution of the GW phase function is used, which greatly reduces the numerical iterations or approximate calculations required during the simulation process, reducing the computational cost and time. Please refer to the supplementary materials for the equation derivation process.

[0052] The specific equation is as follows: K myn The variables of the kernel function are explained as follows:

[0053] k1=θ v cosθ L

[0054] k2=θ i cosθ L

[0055]

[0056] A=cotθ L cotθ v

[0057] B=cotθ L cotθ i

[0058] However, in areas with complex terrain, especially in desert areas, terrain effects have a significant impact on reflectivity modeling. This effect originates from the differences in pixels with different terrain slopes and slope directions when receiving and reflecting light, which in turn leads to inconsistency and decreased accuracy of reflectivity in remote sensing data. Desert areas are known for their diversity of landform types and significant changes in terrain undulations, covering a variety of landform types such as deserts, Gobi, and rock deserts. Among them, desert areas are composed of aeolian dunes and sand ridges; Gobi is mainly composed of gravel-covered plains with high surface roughness; rock deserts are mainly composed of exposed rocks and rock debris, and the terrain is rugged. In addition, the terrain undulations in desert areas vary significantly, and the height differences of sand dunes, isolated hills, ridges and gullies in Gobi and rock desert areas, and other terrain features, further aggravate the impact of terrain effects on remote sensing data. Therefore, in order to improve the accuracy and consistency of remote sensing data in desert areas, it is necessary to use the volume scattering kernel developed for flat terrain to enhance its simulation accuracy in desert areas. In the correction of the kernel model, the present invention combines the SCS+C method to correct the kernel. The SCS+C method is an improved terrain correction method based on a physical model (Sun-Canopy-Sensor, SCS) and introduces a semi-empirical C factor to improve the correction effect. The SCS model corrects the reflectance by projecting the sunlit canopy on the slope surface onto the horizontal plane, but it is prone to over-correction problems on steep slopes facing the light source. After introducing the C factor, the SCS+C method can simulate diffuse radiation on shadowed slopes, thereby reducing the over-correction phenomenon, which is specifically expressed as:

[0059]

[0060] In the formula, α represents the terrain slope, and C is an empirical parameter that can be fitted from the least squares method. The SCS model corrects the reflectance by projecting the sunlit canopy on the slope surface onto the horizontal plane, but it is prone to over-correction on steep slopes facing the light source. After introducing the C factor, the SCS+C method can simulate diffuse radiation on shadowed slopes, thereby reducing the over-correction phenomenon.

[0061] The surface vegetation coverage in desert areas is mainly manifested in two categories, desert shrubs represented by Haloxylon ammodendron and Camel Thorn, and soil biological crusts represented by mosses and lichens. Desert vegetation usually grows relatively short, and the shrubs are spherical to reduce wind resistance and help them remain stable in strong wind environments. Such canopy structure characteristics are consistent with the ellipsoid structure assumed when the LiSparseR function was derived. Therefore, the LiSparseR function is selected as the geometric optics kernel in the geometric optics part. The kernel (K Li (θ i ,θ v ,φ)) equation is as follows:

[0062]

[0063] In the formula, O(θ i ',θ v ',φ) is called the overlap function of the view shadow and the sun shadow, which can be written as follows:

[0064]

[0065] Overlap function O(θ i ',θ v ',φ) are interpreted as follows:

[0066]

[0067]

[0068] The two ratios inherent in the equation, namely h / b and b / r, are important structural parameters used to represent the crown shape and relative height, respectively. Specifically, h describes the average height from the crown center, and b and r describe the average vertical and horizontal semi-axes of the modeled ellipsoidal crown, respectively. The RTLSR model used by the MODIS BRDF / Albedo algorithm fixes these two ratios to h / b=2 and r=1 as prior information to simplify the kernel function and maintain the linearity of the kernel-driven model.

[0069] In addition to the growth of biological crusts in desert areas, the coverage of biological crusts in arid and semi-arid areas accounts for about 40%, which is also an important reflectivity contribution component that cannot be ignored. Therefore, it is necessary to consider using two different geometric optics kernels to simulate the surface reflectivity from a geometric optics perspective. At the beginning of the derivation, the RLKB function assumes a reflection model for randomly arranged rectangular protrusions on a flat horizontal surface. The length of these protrusions is l, the height is h, and the width is b. It is assumed that all areas exposed to sunlight, including the ground and protrusions, have the same reflectivity R. This assumption of an opaque rectangle is consistent with the structural characteristics of biological crusts in desert areas. Therefore, it is considered to introduce the RLKB kernel into the geometric optics kernel to simulate the reflectivity contribution of biological crusts in desert areas. This makes up for the limitations of the LiSparseR function in simulating the reflectivity of the geometric optics part in desert areas. The combined use of the two kernels is shown below.

[0070]

[0071] The RLKB function expression is as follows:

[0072]

[0073] Likewise, topographic correction is required for geometric optics developed based on flat terrain.

[0074]

[0075] In order to verify the proposed method, a comprehensive dataset consisting of observational and simulated multi-angle data from various sources in desert areas was used. The research team of the present invention conducted field measurements in April 2023 to obtain a ground multi-angle BRF observation dataset, and used the LESS radiation transfer model based on Monte Carlo ray tracing to obtain a multi-angle BRF observation dataset for computer scenes. The satellite observation data of the desert area came from the POLDER data on the PARASOL satellite.

[0076] The target sample plot is located at the edge of the Gurbantunggut Desert (88.26°N, 44.45°E), in the center of the Junggar Basin in the northern Xinjiang Uyghur Autonomous Region of China. Figure 1 As shown, (a) shows the surface environment of the field measurement area, and (b) shows an aerial view of the field measurement area. The area belongs to the subhumid continental climate with an altitude between 300 and 600 meters. The field measurement plot is the second largest desert and the largest fixed and semi-fixed desert, with exposed gravel as the background. The dataset was collected on April 29, 2023. The weather was clear, windless and cloudless on the day of collection. The portable ground spectrometer ATP9100F produced by Optosky was used for measurement, and its spectral range is 300 to 1100 nanometers. The collected data includes multi-angle information (illumination geometry, observation geometry and BRF) to construct a multi-angle BRF dataset. During the sample measurement, the solar zenith angle and solar azimuth angle changed little, so the present invention assumes that these two angles are fixed values. The measurement interval of the apparent zenith angle (VZA) is set to 30°, ranging from -60° (backscatter) to 60° (forward scatter); the interval of the apparent azimuth angle (VAA) is also 30°, ranging from 0° to 330°.

[0077] The LESS (LargE-Scale remote sensing data and image Simulation framework) model is one of the computer simulation models. The model is based on a ray tracing three-dimensional real structure radiation transfer model, which can simulate the transmission process of incident light in the scene, including absorption, reflection and transmission, and output corresponding simulation data, such as reflectivity, albedo and fPAR. In addition, the LESS model can also simulate the anisotropic reflection characteristics of complex three-dimensional scenes at the pixel scale. Here, it is used as a benchmark for evaluating the accuracy of the BRDF model. Taking into account the influence of different vegetation coverage in desert areas, the present invention constructs three scenes with different vegetation coverage, see Figure 2, where (a) Scene I has a xylem coverage of 20% and a biological crust coverage of 40%; (b) Scene II has a xylem coverage of 40% and a biological crust coverage of 40%; (c) Scene III has a xylem coverage of 60% and a biological crust coverage of 40%; each scene has the same terrain and area except for the different vegetation coverage, and the scene size is 150m*150m; (d) represents a local enlarged detail image of the computer-simulated scene; (e) represents a top view of the computer-simulated scene. The desert vegetation shown in the figure is randomly distributed in the scene in different shapes and sizes. Green represents the biological soil crust in the desert area, which is randomly distributed in the scene. The area of ​​each desert scene is 150m*150m, mainly characterized by Haloxylon ammodendron trees of different sizes and biological soil crusts, and the vegetation is randomly distributed throughout the scene.

[0078] In terms of terrain, the terrain of the scene is set based on the DEM data in the Gurbantunggut Desert, covering the undulating terrain of the desert area. For the three desert scenes with different coverage, the solar zenith angle (SZA) and solar azimuth angle (SAA) are set to fixed values ​​of 30° and 270°, while the positions of the apparent zenith angle and the apparent azimuth angle are evenly distributed in the hemispherical (2π) space with steps of 10° and 15°, respectively, where the range of the apparent zenith angle is [0°, 80°] and the range of the apparent azimuth angle is [0°, 360°].

[0079] The POLDER product data provides a coarse spatial resolution of approximately 6 × 7 km per pixel. 2 , the viewing zenith angle (VZA) ranges from 0° to 70°, covering almost all azimuth directions (i.e., 0° to 360°). A monthly POLDER BRDF database, which includes multi-angle measurements collected from January to December 2008. The present invention selects the 16th category of bare land in the IGBP classification system as the research object, which is mainly distributed in northern Africa, central Asia and other regions. A total of 4836 POLDER BRDF datasets for April 2008 were used, and the observed distribution and solar distribution position are shown in Figure 3 , and mainly focuses on the two bands centered at 670nm and 865nm. The BRDF sampling distribution in the selected POLDER dataset is represented in polar coordinates. The red five-pointed star and blue * symbol indicate the location of the sun and the sensor, respectively.

[0080] In order to comprehensively evaluate the quantitative comparison between the simulated BRF and the reference BRF, the present invention adopts a series of evaluation indicators, including the coefficient of determination (R2), the root mean square error (RMSE) and the absolute bias (bias). These indicators are widely used to evaluate the performance of the semi-empirical kernel-driven model and provide a detailed analysis of the model's predictive ability. The coefficient of determination (R2) is a commonly used evaluation indicator used to measure the degree of fit of the model. It reflects the linear correlation between the model prediction value and the actual observation value, and the value range is between 0 and 1. The closer the value is to 1, the better the model fit effect. The root mean square error (RMSE) is a common indicator for measuring the model prediction error. It calculates the degree of difference between the model prediction value and the actual observation value. The smaller the value, the higher the model prediction accuracy. The absolute bias (bias) is an indicator for evaluating the model prediction deviation. It represents the average deviation between the model prediction value and the actual observation value. The closer the value is to zero, the smaller the model prediction deviation. Finally, the present invention uses field measurement data to analyze and discuss the accuracy optimization and computational efficiency of the MHLSRT model. The calculation method is as follows:

[0081]

[0082] bias = mean(yx)

[0083]

[0084] Where x and y are the simulated and predicted BRFs, respectively. biasA and biasB are the absolute deviations between the simulated BRF and the measured BRF in model A.

[0085] In order to verify the feasibility of the proposed semi-empirical kernel driven bidirectional reflectance distribution function (MHLSRT) model in accurately simulating surface reflectance in arid and semi-arid regions, a comprehensive validation was performed using field measurement data, LESS simulation data, and POLDER satellite data. This section presents detailed results and analysis to demonstrate the performance and accuracy of the model.

[0086] Field measurements were performed at the edge of the Gurbantunggut Desert to provide direct ground truth data for evaluating the MHLSRT model. The dataset includes multi-angle reflectivity measurements, which are critical for evaluating the model's performance in capturing the complex reflective features of arid landforms. The BRFs simulated by different models (MHLSRT, RLKB, RTLSR) are plotted. Figure 4 , Figure 4(a-1) to (a-2) are observed values, (b-1) to (b-2) are MHLSRT model simulation values, (c-1) to (c-2) are RTLSR model simulation values, and (d-1) to (d-2) are RLKB model simulation values, which are compared and analyzed in the red band and near-infrared band, respectively. The deviations between them and the real BRF data obtained by field measurements are calculated, as shown in Figure 2. Figure 5 As shown in (a-1) to (a-3), in the red band, the MHLSRT model shows a high degree of consistency with the field measurement results. The deviation value of the MHLSRT model in the entire field of view is close to zero, showing a high degree of robustness. In contrast, the RLKB model and the RTLSR model have larger deviations in part of the field of view, especially in the position facing away from the sun (the lower half of the polar coordinate diagram), showing obvious deviations. This shows that the RLKB and RTLSR models are insufficient in simulating hot spots. In the near-infrared band, such as Figure 5 As shown in (b-1) to (b-3), the MHLSRT model also shows high consistency, and the overall deviation value is still close to zero. The deviation distribution of the RLKB model and the RTLSR model in the near-infrared band is similar to that in the red band. The overall performance of the MHLSRT model in the near-infrared band is slightly lower than that in the red band, but it still shows significant accuracy. Figure 5 Among them, (a-1), (a-2), and (a-3) are red bands, (b-1), (b-2), and (b-3) are near-infrared bands, and the sun's incidence position is represented by a five-pointed star symbol in the polar coordinate diagram.

[0087] By comparing the reference values ​​of the LESS model with other semi-empirical kernel-driven models (MHLSRT, RLKB, RTLSR), the BRDF shapes of these models and their errors in the principal planes are analyzed. Figure 6 As shown, the first to third columns are high coverage, medium coverage and low coverage respectively. The five-pointed star symbol in the figure represents the solar incidence position. The length of the radius represents the range of the apparent zenith angle (VZA) from -70° to 70°, and the direction of the radius represents the range of the apparent azimuth angle (VAA) from 0° to 360°. Comparing the three models of RLKB, MHLSRT and RTLSR in the red band with the LESS simulated BRF, it can be seen that compared with RTLSR and RLKB, MHLSRT has more accurate BRF values ​​and more significant hotspot effects in the direction of sunlight. This phenomenon is more obvious when SZA is small. In the high coverage area ( Figure 6 (a-1), (b-1), (c-1), (d-1)), MHLSRT model ( Figure 6 (b-1)) performs similarly to the LESS model, and the reflectivity values ​​are also concentrated in the same range, showing good simulation ability. Figure 6(c-1)) The reflectivity value decreases near the viewing angle of 90° and 270°. RLKB model ( Figure 6 The BRF distribution of (d-1)) is close to that of the LESS model and the MHLSRT model, but the reflectivity decreases when the apparent zenith angle is close to ±70°. The situation is similar in low coverage scenarios. Horizontal comparative analysis shows that the MHLSRT model has better consistency with the BRF reference value simulated by LESS in different vegetation coverage scenarios within the zenith angle range of 0°-70°, which proves the anisotropic ability of the MHLSRT model in dealing with different observation zenith angles and relative azimuth changes. The RTLSR model shows an underestimation of the BRF in all observation directions in high BRF areas, and an overestimation of the BRF in low BRF areas. Although the RLKB model has a small simulation error in the small zenith angle area, it has a large error in the hotspot area.

[0088] Similarly, the comparison results of the near-infrared band are as follows Figure 7 As shown in the figure, the first to third columns are high coverage, medium coverage and low coverage respectively, and the five-pointed star symbol in the figure represents the solar incidence position. The length of the radius represents the range of the apparent zenith angle (VZA) from -70° to 70°, and the direction of the radius represents the range of the apparent azimuth angle (VAA) from 0° to 360°. The three models of RLKB, MHLSRT and RTLSR in the near-infrared band (865nm) are compared with the BRF simulated by LESS. It can be seen that compared with RTLSR and RLKB, MHLSRT has more accurate BRF values ​​and more significant hotspot effects in the direction of sunlight. Compared with the BRF reference value simulated by LESS in different vegetation coverage scenarios, the MHLSRT model has better consistency within the range of ±70° zenith angle, which proves the anisotropic ability of the MHLSRT model in dealing with different observation zenith angles and relative azimuth angle changes. The RTLSR model shows an underestimation of BRF in the observation direction of high BRF areas, and an overestimation of BRF in low BRF areas. Although the RLKB model has a small simulation error in the small zenith angle area, it has a large error in the hot spot area. In addition, the characteristics of the MHLSRT anisotropic simulation BRF are very consistent with the reference BRF of LESS. Compared with the semi-empirical kernel-driven model that does not consider the characteristics of the desert scene, the MHLSRT model can significantly improve the simulation performance.

[0089] Figure 8The reflectivity of the MHLSRT model in the hot spot area of ​​the main plane and the reflectivity reference value of the LESS model are plotted. (a-1) and (b-1) represent high coverage areas, (a-2) and (b-2) represent medium coverage areas, (a-3) and (b-3) represent low coverage areas, (a-1), (a-2) and (a-3) represent the simulation results of the red band (670nm), and (b-1), (b-2) and (b-3) represent the simulation results of the near-infrared band (865nm). Figure 8 (a-1) to (a-3) show the reflectance changes of the red band when the vegetation coverage changes from high to low; Figure 8 (b-1) to (b-3) show the reflectivity changes in the near-infrared band when the vegetation coverage changes from high to low. It can be seen that in the red band, the reflectivity of the MHLSRT model and the LESS reference value in the hotspot range almost completely overlap, showing a high degree of consistency. The reflectivity peaks in the hotspot area (when VZA = -30°), and the reflectivity in the near-infrared band is similar to that in the red band. Whether in the red band or the near-infrared band, the MHLSRT model shows extremely high accuracy and stability, and can effectively simulate the reflectivity of the LESS model.

[0090] Fig. 9The simulation results of each model on the main plane are shown. (a-1) and (b-1) represent high coverage areas, (a-2) and (b-2) represent medium coverage areas, (a-3) and (b-3) represent low coverage areas, (a-1), (a-2) and (a-3) represent red band (670nm) simulation results, (b-1), (b-2) and (b-3) represent near infrared band (865nm) simulation results. The proposed MHLSRT models all perfectly capture the roof-shaped BRDF curves, and compared with MLSR and other models without hotspot correction, MHLSRT accurately simulates the location and size of hotspots. The LESS reference value in the high vegetation coverage scene shows a reflectivity peak near the line of sight, that is, the hotspot effect. In the red and near infrared bands, the MHLSRT model is closest to the reference value in shape, while the MLSR and RTLSR models show some deviations at the hotspots. Although the RLKB model captures the overall shape of the reflectance curve, it consistently underestimates the LESS reference values ​​across the entire VZA range. In areas with medium vegetation cover, the reflectance patterns are similar, but there is a clear change in the magnitude of the reflectance values. The MHLSRT model remains the most accurate in both spectral bands. The MHLSRT model performs better in the near-infrared band compared to the red band, and is closer to the reference data. The RTLSR and RLKB models show increased differences at large VZA angles. In both bands in areas with low vegetation cover, the MHLSRT model continues to match the LESS reference values ​​well. The RTLSR model performs poorly in the red band, but improves in the near-infrared band.

[0091] Fig.10The absolute deviations between the main plane LESS reference values ​​and the model simulations were analyzed to evaluate the accuracy of the models. (a-1), (b-1) represent high coverage areas, (a-2), (b-2) represent medium coverage areas, (a-3), (b-3) represent low coverage areas, (a-1), (a-2), (a-3) represent the simulation results of the red band (670nm), (b-1), (b-2), (b-3) represent the simulation results of the near infrared band (865nm). The MHLSRT model maintained the lowest deviations over most of the VZA range in both spectral bands, indicating that its prediction ability is consistent for different vegetation coverage areas, especially in the range from the critical minimum point to 60°, modeling the reflectance with a small deviation, showing excellent performance. In contrast, the RTLSR has larger deviations, especially at positive VZA angles. The RLKB model shows the most significant deviations from the observations, especially at the edges of the VZA range. In contrast, the MLSR describes consistent deviations across the entire VZA spectrum, indicating that the model has systematic overestimations. Absolute deviation analysis highlighted the strengths and limitations of each model at different VZA angles, with MHLSRT providing the best balance between accuracy and consistency, suggesting that the model’s assumptions and derivations are well suited to the desert region in which it was applied.

[0092] Fig.11The reflectivity changes of different kernel-driven models (MHLSRT, RTLSR, RLKB) and LESS reference values ​​on the cross-principal plane were analyzed. (a-1) and (b-1) represent high coverage areas, (a-2) and (b-2) represent medium coverage areas, (a-3) and (b-3) represent low coverage areas, (a-1), (a-2) and (a-3) represent red band (670nm) simulation results, (b-1), (b-2) and (b-3) represent near-infrared band (865nm) simulation results. In Figures (a-1) to (a-3), the changes of the red band at different VZA angles from high to low vegetation coverage are shown; in Figures (b-1) to (b-3), the changes of the near-infrared band at different VZA angles from high to low vegetation coverage are shown. In Figures (a-1) to (a-3), the reflectance trends of different kernel-driven models in the red band with changes in VZA angle can be observed. As the vegetation coverage decreases, the reflectance of each model (MHLSRT, RTLSR, RLKB) and the LESS reference value at different VZA angles shows obvious differences. The reflectance of the RTLSR model at most VZA angles is lower than the LESS reference value, while the MHLSRT model is almost consistent with the LESS reference value in the range of VZA = ±70°. Figures (b-1) to (b-3) show the changes in reflectance of different kernel-driven models in the near-infrared band. Similar to the red band, the MHLSR model is almost consistent with the LESS reference value in the process of vegetation coverage changing from high to low.

[0093] Fig.12The reflectance deviation values ​​of different kernel-driven models (MHLSRT, RTLSR, RLKB) and LESS reference values ​​on the cross-principal plane were calculated. (a-1) and (b-1) represent high coverage areas, (a-2) and (b-2) represent medium coverage areas, (a-3) and (b-3) represent low coverage areas, (a-1), (a-2) and (a-3) represent red band (670nm) simulation results, (b-1), (b-2) and (b-3) represent near-infrared band (865nm) simulation results. As shown in the figure, Figures (a-1) to (a-3) represent the reflectance deviation values ​​of the red band at different VZA angles from high to low vegetation coverage; Figures (b-1) to (b-3) represent the reflectance deviation values ​​of the near-infrared band at different VZA angles from high to low vegetation coverage. In Figures (a-1) to (a-3), the reflectivity deviations between different models and the LESS reference values ​​in the red band are analyzed. It can be observed that the MHLSRT model has the smallest absolute deviation value at all VZA angles, almost close to zero. The RTLSR and RLKB models have the largest deviation values ​​when the VZA is close to ±70°. This shows that the MHLSRT model performs most stably in the red band, while other models have larger errors at larger VZA angles. Figures (b-1) to (b-3) show the reflectivity deviations between different models and the LESS reference values ​​in the near-infrared band. Similar to the red band, the MHLSRT model still has the smallest absolute deviation value at all VZA angles, showing high stability. The deviation values ​​of the RTLSR and RLKB models also increase significantly at larger angles. From the above results, it can be seen that the MHLSRT model has the smallest deviation values ​​at all angles and bands, showing high accuracy and stability.

[0094] In the present invention, the present invention uses the IGBP 16 bare soil category, and based on the comparison of the model-simulated BRF with the observed BRF obtained in the POLDER BRDF database, the present invention performs a one-to-one scatter plot analysis of the simulation results. Fig.13 In the red band (670 nm), the colors correspond to the point density from lowest (purple) to highest (red). For the RLKB model, the correlation is slightly lower, R 2 The RTLSR model showed an R of 0.7739. 2 and RMSE of 0.0407. In particular, for the MHLSRT model, it exhibited an R 2The root mean square error (RMSE) was significantly low at 0.0299, indicating that the MHLSRT application in desert areas has high accuracy relative to the reference data. In the near-infrared band (865 nm), the results also showed a strong linear correlation between the POLDER data and the simulated dataset, with a high R 2 The value is proven. 2 The RLKB model comparison yielded an R of 0.7916. 2 and RMSE of 0.0403, while RTLSR comparison presents an R of 0.8023 2 and RMSE of 0.0401. The scatter plot of the near-infrared band shows a better fit than that of the red band overall. The results show that the MHLSRT model has high accuracy in the red and near-infrared bands, especially in capturing complex reflection features. Compared with the semi-empirical kernel-driven model that does not consider the scene characteristics of desert areas, the MHLSRT model can significantly improve the simulation performance and reproduce the actual surface characteristics more accurately.

[0095] Here the present invention is to illustrate that the Gutschick-Wiegel phase function is a sub-case of the phase function in Ross's radiative transfer theory. In this derivation, first consider an infinitely thin layer (thickness Δz) of a horizontally uniform vegetation canopy, which is densely and randomly filled with leaf elements per unit area. Now, when an outgoing photon hits a leaf element, the angular distribution of the scattered light is given by the scattering phase function P(Ω,Ω'), which satisfies the following relationship:

[0096]

[0097] Wherein, Ω and Ω' are the directional unit vectors before and after scattering, Ω' is the incident unit vector defined by the solar zenith angle, and Ω' is the outgoing unit vector defined by the apparent zenith angle. The present invention averages the scattering phase function P(Ω,Ω') in azimuth to obtain the following:

[0098]

[0099] At the same time, the present invention ensures that the following relationship is satisfied:

[0100]

[0101] The scattering phase function is multiplied by the normalization factor (1 / 4π). The phase function describes the energy distribution of radiation scattered along another direction Ω after the radiation is incident along one direction Ω'. The present invention integrates all possible scattering directions and accumulates the scattered energy in all directions to obtain the total scattered energy:

[0102]

[0103] This is a normalized condition, which means that the radiation energy incident along the Ω' direction is the total energy scattered in all possible directions by the Γ function. The result of the overall integration is the regional scattering coefficient ω a (Ω'), which can be expressed as:

[0104] ω a (Ω')=ω(Ω')G(Ω')≡ωG(Ω')

[0105] The expression of G(Ω') is given in (Ross, 1981). Thus, the present invention can establish the relationship between the surface scattering phase function (1 / 4π)Г(Ω,Ω') and the scattering phase function (1 / 4π)P(Ω,Ω'):

[0106]

[0107] For the case of single-angle blades, the present invention can simplify the above formula to obtain:

[0108]

[0109]

[0110] The kernel function of a single layer of leaves in the canopy can be expressed in the following form:

[0111]

[0112] Assuming that the normal direction of the blades is tilted at a fixed angle and symmetrical in the azimuth direction, which means that their distribution in the horizontal plane is the same in all directions. Then the surface scattering phase function can be expressed as:

[0113] Γ(θ i ,θ v )=K(θ i ,θ v )

[0114] When the above assumptions do not hold, the surface scattering phase function can be expressed as:

[0115]

[0116] So far, the present invention focuses on the derivation of the kernel function K(θi,θv,θ L). Here, the present invention represents the kernel function as two parts. The first part involves the blade projection area in the direction of photon inflow, taking into account the angle of photon incidence and the change in blade orientation. By averaging the leaf orientation, the influence of the leaf orientation change on the result is taken into account. The second part involves the Lambertian reflection factor, which represents the re-radiation of secondary photons in the blade. This part is first calculated in the blade fixed frame and then converted to the space fixed frame to comprehensively consider the directional characteristics of the re-radiated light. In this process, different situations of the outgoing photon direction are taken into account, and a comprehensive representation is provided by averaging the outgoing photon orientation. According to the method of Gutschick and Wiegel, the kernel function K(θi,θv,θ L ) is composed of the dot product of two factors, as shown below:

[0117] Ω L Ω'>0: Photons are incident from the bottom of the canopy

[0118] Ω L Ω'<0: Photons are incident from the top of the canopy

[0119] Ω L Ω>0: Photons are emitted from the top of the canopy

[0120] Ω L Ω<0: Photons are emitted from the bottom of the canopy

[0121] The kernel expression is partitioned according to the integration range, a step based on whether the photon is generated by transmission or reflection, to more accurately describe how the photon interacts with the leaf. This partitioning is critical because it ensures that the model can take into account both possibilities that the photon may pass through the leaf or be reflected by the leaf, thereby more fully capturing the complex optical process. In the formula, the canopy normal vector always points to the upper hemisphere, which provides the model with a physical meaning of the leaf surface direction. When describing the relationship between the photon and the canopy normal vector, it helps to more accurately describe the directional characteristics of the incident and reflected light. The first part of the kernel function accurately quantifies the leaf projected area in the direction of the incident photon flux, integrating the changes in the photon incident angle and the leaf direction. By assuming a mean distribution of leaf orientation, the model can well adapt to changes in leaf arrangement, ensuring that the scattering prediction accurately reflects the combined effects of different leaf orientations. This approach not only clarifies the basic interaction between the incident photon and the leaf, but also provides a detailed understanding of how changes in leaf orientation affect scattering phenomena within the canopy.

[0122] In contrast, the second part focuses on describing the Lambert reflectance factor, which is initially calculated in a leaf-centered coordinate system and then transformed into a spatially fixed frame. This transformation is critical to accurately describe the anisotropy of the canopy. In performing this transformation, the model rigorously evaluates photons that may have emerged from all directions after interacting with the leaf surface. By averaging these emission directions, this does not mean that the algorithm ignores the complexity of the emission; rather, the algorithm finds a way to represent the overall effect of light scattering, thereby capturing the anisotropic nature of light scattering within the vegetation canopy. By integrating these components, the kernel function K(θi,θv,θ L ) is expressed as an equation:

[0123]

[0124] The present invention requires that φ and φ are between 0 and π. L Since the blade normal (and the corresponding outgoing flux) is azimuthally symmetric about π, within this range, the present invention needs to accurately know the two dot products Ω L ,Ω' and Ω L ,Ω changes its sign, the present invention uses φ t’ and φ t To indicate the critical case of the change:

[0125]

[0126] There are three possible cases, depending on whether the integrand is positive, negative, or both. Specifically: Fig.14 (a), (b), and (c) represent the incident conditions, and (d), (e), and (f) represent the exit conditions.

[0127] Case 1: cotθ L cotθv>1, photons are incident from the bottom of the leaf

[0128]

[0129] Case 2: cotθ L cotθv<-1, photons are incident from the top of the blade

[0130]

[0131] Case 3: -1 <cotθ L cotθv<1, photons are incident from the top and bottom of the blade at the same time

[0132]

[0133] Similarly, there are three calculation schemes for the integral in the above formula:

[0134] Case 4: cotθ L cotθ>1, photons are emitted from the top of the blade

[0135]

[0136] Case 5: cotθ L cotθ<-1, photons are emitted from the bottom of the leaf

[0137]

[0138] Case 6: -1 <cotθ L cotθ<1, photons are emitted from the top and bottom of the blade at the same time

[0139]

[0140]

[0141] Here, if r is given to the kernel function L =t L This assumption, then in this case, the present invention summarizes the analytical form of the kernel function as follows:

[0142]

[0143] K myn The variables of the kernel function are explained as follows:

[0144] k1=θ v cosθ L

[0145] k2=θ i cosθ L

[0146]

[0147] A=cotθ L cotθ v

[0148] B=cotθ L cotθ i

[0149] By analyzing the kernel value images of the volume scattering kernel and the geometric optical kernel, it can be clearly seen that the applicability of the MHLSRT model in desert areas is mainly reflected in its stable description of volume scattering and geometric optical scattering. First of all, from the image analysis of the volume scattering kernel, the volume scattering kernel value of the MHLSRT model is low, especially in the low viewing angle (VZA) range. This means that at a smaller viewing angle, the model has a lower response to volume scattering, which is more consistent with the actual situation of the surface in desert areas, because vegetation in desert areas is sparse and the volume scattering effect is weak. In addition, the trend of the kernel value changing with the viewing angle is relatively smooth, without obvious fluctuations, which shows that the model can stably describe the volume scattering characteristics of the desert surface without significant deviations due to changes in viewing angle. The analysis of the volume scattering kernel also supports this observation. The kernel functions of the RTLSR model and the RLKB model fail to accurately express their unique scattering characteristics in desert areas. In contrast, the volume scattering kernel function derived by the present invention shows that the image is overall smooth and approximately symmetrical, and the kernel function value changes smoothly, which is in line with the simple scattering mechanism in desert areas. As Fig.15 As shown, (a-1) represents the volume scattering kernel of the RTLSR model, (a-2) represents the volume scattering kernel of the RLKB model, (b-1) represents the uncorrected volume scattering kernel derived in this paper, and (b-2) represents the volume scattering kernel of the corrected MHLSRT model.

[0150] Secondly, from the image analysis of the geometric optical kernel, the geometric optical kernel values ​​of the MHLSRT model show a consistent distribution trend at different solar zenith angles (SZA), such as Fig.16 As shown in the figure, (a-1) represents the geometric optical kernel of the RTLSR model, (a-2) represents the geometric optical kernel of the RLKB model, (b-1) represents the geometric optical kernel of the MHLSRT model in high vegetation coverage areas, (b-2) represents the geometric optical kernel of the MHLSRT model in medium vegetation coverage areas, and (b-3) represents the geometric optical kernel of the MHLSRT model in low vegetation coverage areas. This shows that the model can stably describe the geometric optical scattering characteristics of the desert surface under different lighting conditions. At the same time, the kernel value range of the MHLSRT model is moderate, and there are no abnormal values ​​that are too high or too low, which is very important for describing the roughness and terrain undulation of the desert surface, because the geometric optical scattering in the desert area mainly comes from these characteristics. In particular, under different vegetation coverage conditions, the geometric optical kernel value gradually decreases, which further demonstrates the applicability of the geometric optical kernel in describing simple surface structures in desert areas. The relationship curve between the geometric optical kernel shape of the MHLSRT model and the viewing angle (VZA) is relatively smooth, which can reasonably reflect the influence of the surface geometric structure on light reflection. The surface structure in desert areas is relatively simple, and this model can better simulate the optical scattering in this environment.

[0151] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems or computer program products. Therefore, the present application may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware.

[0152] The above are only embodiments of the present application and are not intended to limit the present application. For those skilled in the art, the present application may have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included in the scope of the claims of the present application.

Claims

1. A method for simulating the albedo of a desert scene based on a kernel-driven method, characterized in that: The following steps are involved: Collect vegetation coverage data, biological soil crust coverage data and terrain feature data of the target desert scene, standardize the collected data, and generate data representing the dynamic characteristics of the desert scene through spatial distribution analysis and temporal dynamic correlation analysis; The volume scattering kernel is constructed based on the Gutschick-Wiegel scattering phase function, and the corrected volume scattering kernel data is generated by combining the SCS+C terrain correction method and the hot spot correction mechanism; The Li-SparseR kernel and the Roujean kernel are used together to construct a geometric optical kernel. The weights of the Li-SparseR kernel and the Roujean kernel are adjusted by dynamic vegetation coverage parameters, and the corrected geometric optical kernel data are generated in combination with terrain correction. The bidirectional reflectance distribution function model of the desert scene is generated by combining the corrected volume scattering kernel data, the corrected geometrical optics kernel data and the isotropic kernel; The bidirectional reflectance distribution function model is verified using field measurement data and satellite observation data, and the simulation accuracy in the red band and near-infrared band is improved by optimizing the model parameters.

2. The method for simulating albedo of desert scenes based on the kernel-driven method according to claim 1, characterized in that: The vegetation coverage data, biological soil crust coverage data and terrain feature data are obtained by combining on-site measurement and PARASOL satellite multi-angle observation.

3. The method for simulating albedo of desert scenes based on the kernel-driven method according to claim 2, characterized in that: After the data is standardized, dynamic characteristic data representing the state of vegetation in the desert scene is generated through time dynamic correlation analysis and spatial distribution analysis.

4. The method for simulating desert scene albedo based on the kernel-driven method according to claim 1, characterized in that: The volume scattering kernel is constructed based on the Gutschick-Wiegel scattering phase function, and the corrected volume scattering kernel data is generated by combining the SCS+C terrain correction method and the hot spot correction mechanism.

5. The method for simulating desert scene albedo based on the kernel-driven method according to claim 4, characterized in that: The corrected volume scattering kernel data further corrects the back-coherent scattering characteristics by introducing the Chen hot spot correction formula to improve the hot spot effect simulation accuracy.

6. The method for simulating desert scene albedo based on a kernel-driven method according to claim 1, characterized in that: The geometric optics kernel consists of a Li-SparseR kernel and a Roujean kernel, wherein the Li-SparseR kernel is used to describe the geometric properties of sparse vegetation canopies in deserts, and the Roujean kernel is used to describe the optical properties of biological soil crusts in desert scenes.

7. The method for simulating albedo of desert scenes based on the kernel-driven method according to claim 6, characterized in that: The geometric optics kernel adjusts the weights of the Li-SparseR kernel and the Roujean kernel through dynamic vegetation cover parameters to adapt to the distribution ratios of different surface cover types.

8. The method for simulating desert scene albedo based on a kernel-driven method according to claim 1, characterized in that: The bidirectional reflectance distribution function model combines the corrected volume scattering kernel data, the corrected geometric optics kernel data and the isotropic kernel, and optimizes the weight parameters through the least squares method to generate the desert scene reflectance simulation results suitable for the red band and the near-infrared band.

9. The method for simulating desert scene albedo based on the kernel-driven method according to claim 8, characterized in that: The desert scene reflectivity simulation results are verified and parameter optimization is performed using on-site multi-angle observation data, bidirectional reflectance distribution function data acquired by the PARASOL satellite, and simulation data generated based on the LESS radiation transfer model.

10. A system for implementing the desert scene albedo simulation method based on the kernel-driven method according to any one of claims 1 to 9, characterized in that: The system includes: A data collection module is used to collect vegetation coverage data, biological soil crust coverage data and terrain feature data of the target desert scene, and to perform standardization on the collected data; A dynamic feature generation module is used to generate data representing the dynamic features of the desert scene from the standardized data through spatial distribution analysis and temporal dynamic correlation analysis; Volume scattering kernel construction module, which is used to construct volume scattering kernel based on Gutschick-Wiegel scattering phase function, and generate corrected volume scattering kernel data by combining SCS+C terrain correction method and hot spot correction mechanism; The geometric optical kernel construction module is used to jointly use the Li-SparseR kernel and the Roujean kernel to construct the geometric optical kernel, adjust the weights of the Li-SparseR kernel and the Roujean kernel through dynamic vegetation coverage parameters, and generate the corrected geometric optical kernel data in combination with terrain correction; A reflectivity model generation module is used to generate a bidirectional reflectance distribution function model of a desert scene by combining the corrected volume scattering kernel data, the corrected geometric optics kernel data and the isotropic kernel; The model verification and optimization module is used to verify the bidirectional reflectance distribution function model using field measurement data and satellite observation data, and to improve the simulation accuracy of the bidirectional reflectance distribution function model in the red band and the near-infrared band by optimizing model parameters.

Images