A Method for Extracting Subforest Ground Photon Elevation Based on ICESat-2

Abstract

The present invention discloses a method for extracting the elevation of understory ground photons based on ICESat-2, including: filtering background noise photons in ATL03 based on confidence; calculating the vertical and horizontal weights of each photon within the research window, and respectively constructing evaluation functions for the ground photon set and the non-ground photon set based on the weights; using the evaluation function as the basis for driving the backpropagation of the Monte Carlo search tree, and adopting the Monte Carlo tree search algorithm to search for ground photons within the window; using ATL08 to fit an approximate surface contour to perform elevation difference threshold filtering on the searched ground photons, obtaining the understory ground photon set, and thus obtaining the understory ground elevation. By constructing evaluation functions for ground and non-ground photons, the present invention guides the best search direction of the Monte Carlo search tree in a way of quantifying scores, overcomes the influence of different vegetation coverage and slope factors on the extraction of ground photons, and ensures that the extracted ground photon elevation points have high accuracy.

Description

Technical Field

[0001] The present invention belongs to the technical field of remote sensing data processing, and particularly relates to a method for extracting understory ground photon elevation based on ICESat-2. Background Art

[0002] Elevation points are one of the most important data products in the surveying and mapping geographic information industry. They provide basic information for terrain analysis, can accurately reflect the undulation characteristics of the earth's surface, and have important research value in aspects such as land resource utilization, forest resource management, hydrological monitoring, and disaster prevention. For areas with less obstacle occlusion, ground elevation data is relatively easy to obtain. However, for areas with high forest cover density and large terrain undulation, obtaining large-scale and high-precision understory ground elevation information poses a higher technical challenge.

[0003] Generally speaking, high-precision ground elevation points are usually obtained by the leveling method or the GNSS static observation method. The data obtained by these two methods has high accuracy, but it requires a large amount of manpower and material resources. At the same time, for large areas of forest areas that are difficult for humans to reach, traditional methods cannot be used, so they are only suitable for obtaining ground elevation points in small and medium-sized areas. Currently, the method for obtaining large-scale ground elevation points mainly relies on spaceborne lidar technology. It calculates the elevation of ground points by emitting laser beams to the ground and based on the principle of light speed ranging and its precise orbital position, and has the advantages of strong timeliness, wide range, and high efficiency.

[0004] Currently, the spaceborne lidar in orbit with global observation capabilities is the Ice, Cloud, and land Elevation Satellite ICESat-2 launched by NASA. It provides global geolocated photon product ATL03 and a series of thematic elevation data products (such as land vegetation height product ATL08, land glacier height product ATL06, sea surface height product ATL12, etc.). However, for areas with relatively dense forest cover, the accuracy of the ATL08 elevation data is significantly affected by vegetation cover and slope. Therefore, currently, the ground elevation acquisition in forest areas mainly focuses on extracting higher-precision and higher-resolution ground elevation points from the geolocated photon product ATL03. Since the ATL03 data contains background noise, interference signals, and reflected photons of all surface features, it is challenging to obtain large-scale, high-precision, and high-resolution ground elevation points from it. Currently, there are two main ideas for extracting understory ground elevation points from ATL03. One is the elevation histogram statistical method based on point cloud density, and the other is the method of carefully fitting the ground contour line based on physical splines. To a certain extent, both methods can obtain ground elevation points with higher accuracy and density. However, the former method is more sensitive to the density of vegetation cover, while the latter method has higher requirements for the continuity quality of signals and is easily affected by terrain undulation changes. Summary of the Invention

[0005] The present invention provides a method for extracting ground photon elevation under forest based on ICESat-2, and the extracted ground photon elevation points have high precision.

[0006] To achieve the above technical objectives, the present invention adopts the following technical solutions:

[0007] A method for extracting ground photon elevation under forest based on ICESat-2, comprising:

[0008] Obtain the ICESat-2 geolocated photon data product ATL03, and filter out the background noise photons therein based on the confidence level of the photons;

[0009] Divide the filtered photons into windows of a preset size, and calculate the vertical weight and horizontal weight of each photon within each window range;

[0010] Based on the vertical weight and horizontal weight of the photons, construct an evaluation function for the ground photon set and an evaluation function for the non-ground photon set;

[0011] Use the evaluation functions of the ground photon set and the non-ground photon set as the basis for driving the backpropagation of the Monte Carlo search tree, and use the Monte Carlo tree search algorithm to search for the ground photons within the window range;

[0012] Use the ICESat-2 land vegetation high product ATL08 to fit an approximate surface profile, and use the approximate surface profile to filter the elevation difference threshold of the ground photons obtained by the search to obtain a ground photon set under the forest, thereby obtaining the ground elevation under the forest.

[0013] Further, based on the confidence level of the photons recorded in the field signal_conf_ph in ATL03, filter the background noise photons in ATL03: regard the photons with signal_conf_ph < 1 as background noise photons, and regard the photons with signal_conf_ph ≥ 2 as signal photons.

[0014] Further, the calculation formulas for the vertical weight and horizontal weight of each photon within a certain window range are:

[0015] ;

[0016] ;

[0017] In the formula, is the vertical weight of the i-th photon within the current window range, is the maximum value of the photon elevation within the current window range, is the elevation value of the i-th photon within the current window range; is the along-track distance of the i-th photon within the current window range, is the along-track distance of the photon with the minimum elevation value within the current window range, is the horizontal weight of the i-th photon within the current window range.

[0018] Furthermore, the evaluation functions for the ground photon set and the non-ground photon set are respectively:

[0019] ;

[0020] ;

[0021] In the formula, is the evaluation function of the ground photon set; is the evaluation function of the non-ground photon set; ε is a preset extremely small constant, N represents the number of photon points in the ground photon set, represents variance, represents the vertical weight vector of the ground photon set, and respectively represent the difference values of the horizontal and vertical adjacent photon weights of all photons in the ground photon set; represents standard deviation, represents the vertical weight vector of the non-ground photon set, and then represent the difference values of the horizontal and vertical adjacent photon weights of all photons in the current non-ground photon set.

[0022] Furthermore, using the evaluation functions of the ground photon set and the non-ground photon set as the basis for driving the backpropagation of the Monte Carlo search tree, the Monte Carlo tree search algorithm is used to search for ground photons within the window range, specifically:

[0023] Define that the nodes in the Monte Carlo tree search algorithm include: reward value Q, visit count N, ground photon set SCT, and non-ground photon set NSCT;

[0024] Initialize the root node, including initializing the first ground photon and the first non-ground photon in the ground photon set SCT and the non-ground photon set NSCT;

[0025] Starting from the root node, select the child node with the largest UCT value from all fully expanded child nodes as the current node, and perform M rounds of total simulation; where each round of total simulation process includes K searches, and any k-th search process includes:

[0026] (1) Selection: If the current node is fully expanded, select the child node with the largest UCT value from all child nodes of the current node as the node to be simulated s;

[0027] (2) Expansion: If the current node is not fully expanded, select ground photons and non-ground photons near the added ground photons and non-ground photons of the current node respectively, incorporate them into the SCT and NSCT of the expanded child node, and use this expanded child node as the node s to be simulated;

[0028] (3) Simulation: Determine whether the elevation values of the SCT of the node s to be simulated are all less than those of the NSCT. If so, randomly incorporate 1 new photon into the SCT and NSCT of the node to be simulated respectively, and then return to the judgment and incorporation of new photons until the judgment condition is not met, and end this search simulation;

[0029] (4) Backpropagation: Calculate the incremental value of the profit of this simulation according to the photon incorporation order in the termination state of the node to be simulated; and update the profit values and access times of each node on the entire path from the node to be simulated back to the root node.

[0030] Further, the calculation formula of the UCT value is:

[0031] ;

[0032] In the formula, Q and N respectively represent the profit value and access times of the current calculation node, represents the access times of the parent node of the current calculation node, and C is a constant;

[0033] The evaluation function for calculating the incremental value Z of the profit during backpropagation is:

[0034] ;

[0035] In the formula, is the evaluation function value when the i-th ground photon is added to the ground photon set SCT, is the evaluation function value when the i-th non-ground photon is added to the non-ground photon set NSCT, n is the number of photons incorporated into the ground photon set and non-ground photon set when the node to be simulated reaches the termination state, and a and b are user-defined parameters.

[0036] Further, determine whether the current node is fully expanded according to the number of child nodes. If the number of child nodes reaches the preset value, it is determined that the current node is fully expanded.

[0037] Further, selecting ground photons and non-ground photons near the ground photons and non-ground photons added at the current node respectively means randomly selecting 1 photon as the ground photon from a preset number of photons with the closest vertical weight to the ground photons added at the current node, and randomly selecting 1 photon as the non-ground photon from a preset number of photons with the closest vertical weight to the non-ground photons added at the current node.

[0038] Further, when using the ICESat-2 land vegetation high-product ATL08 to fit and approximate the surface contour, specifically, a third-order polynomial is used to fit a number of ground elevation points with an adjacent interval of 100 m in the along-track direction of ATL08, so as to obtain an approximate surface contour line:

[0039] ;

[0040] In the formula, is the along-track distance of the elevation point, is the elevation of the elevation point; are the parameters to be fitted of the third-order polynomial, which are obtained by using a series of ground elevation point coordinates ( , ) of ATL08 and solving by the least squares method.

[0041] Further, the approximate surface contour is used to filter the ground photons obtained by search, specifically:

[0042] Substitute the along-track distance of the ground photons obtained by search into the approximate surface contour line to obtain the interpolated elevation value of the fitted surface;

[0043] Calculate the absolute difference between the interpolated elevation value and the elevation value of the ground photons;

[0044] If the absolute difference is less than the set threshold, the ground photons obtained by search are valid ground photons, otherwise they are filtered out.

[0045] Aiming at the interference problem of vegetation coverage and slope factors on the extraction of understory ground points, and considering the diversity of vegetation coverage and terrain undulation in different forest areas, the present invention proposes a method for extracting the elevation of understory ground photons based on ICESat-2. By means of confidence noise filtering and multiple random sampling score quantification, it can identify ground photons in environments with different vegetation coverages and slopes. Finally, homologous prior data information is introduced to eliminate abnormal points, providing further guarantee for the extraction of large-scale and high-precision understory ground elevation points. In addition, the present invention does not require any other data sources, further reflecting the flexibility and efficiency of the method of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 It is a flowchart of the method described in the embodiment of the present invention.

[0047] Figure 2 It is a flowchart of searching for ground photons using the Monte Carlo tree search algorithm in the embodiment of the present invention.

[0048] Figure 3 It is the pseudocode of searching for ground photons using the Monte Carlo tree search algorithm in the embodiment of the present invention.

[0049] Figure 4 It is the extraction result and comparison diagram of the method described in the embodiment of the present invention for Experimental Area 1; where (a) is the extracted elevation points and the original photon point cloud, and (b) is the comparison and verification of the extracted elevation points and the high-precision airborne ground elevation data.

[0050] Figure 5 It is the extraction result and comparison diagram of the method described in the embodiment of the present invention for Experimental Area 2; where (a) is the extracted elevation points and the original photon point cloud, and (b) is the comparison and verification of the extracted elevation points and the high-precision airborne ground elevation data. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0051] The method for extracting the elevation of understory ground photons based on ICESat-2 provided by the embodiment of the present invention has implementation steps referring to Figure 1 as shown, and includes:

[0052] Step 1, background noise filtering: Obtain the ICESat-2 geolocated photon data product ATL03, and filter out the background noise photons based on the confidence level of the photons.

[0053] Since the ICESat-2 geolocated photon data product ATL03 has already performed a preliminary classification on the returned photons and recorded the corresponding photon signal confidence level in the field signal_conf_ph, in this embodiment, threshold filtering is performed according to the signal_conf_ph field. Setting signal_conf_ph≥2 can basically cover about 95% of the signal photons. That is, the effective signal photon set can be expressed as:

[0054] ;

[0055] Therefore, when signal_conf_ph ≤1, background noise photons can be effectively filtered, providing an efficient mode for the preprocessing of large-scale photon data.

[0056] Step 2, Weight calculation: Divide the filtered photons into windows of a preset size, and calculate the vertical weight and horizontal weight of each photon within each window range.

[0057] In this embodiment, a preset window size of 30 m along the orbit is used, and all photons within each 30 m window along the orbit are taken as the research object.

[0058] Vertically, taking the photon with the maximum elevation value as the reference, calculate the vertical weight of the photons at intervals of 2 m in the vertical distance. . Denote the maximum value of the photon elevation as , and the elevation values of each photon are successively , ,…, , then the vertical weight of each photon within the window is calculated as follows:

[0059] .

[0060] Horizontally, taking the photon with the minimum elevation value within the window as the reference point, calculate the horizontal weight of the photons at intervals of 4 m in the horizontal distance. . Denote its along-track distance value as , and the along-track distance values of each photon are successively , ,…, , then the horizontal weight of each photon within the window is calculated as follows:

[0061] .

[0062] Step 3, Establish an evaluation function based on the spatial distribution characteristics: Based on the vertical weight and horizontal weight of the photons, construct an evaluation function for the ground photon set and an evaluation function for the non-ground photon set.

[0063] Based on the spatial distribution characteristics of forest photons, different evaluation functions are established for the ground photon set (SCT) and the non-ground photon set (NSCT). Since ground photons are more likely to show linear aggregation within a short distance (30m), as the number of ground photons increases, the function value is preset to show a decreasing to convergence trend, and the corresponding function is recorded as Loss. Non-ground photons are more likely to show a random discrete distribution. Therefore, as the number of non-ground photon points increases, the function value is preset to show an increasing trend, and the corresponding function is recorded as Gain. The definition of the evaluation function established in this embodiment is as follows:

[0064] ;

[0065] .

[0066] Where ε is a minimum constant defined as 0.001, N represents the number of points in the ground photon set (SCT), represents the longitudinal weight vector of the terrestrial photon set (SCT), represents the longitudinal weight vector of the non-terrestrial photon set (NSCT), and represents the difference between the horizontal and vertical adjacent photon weights of all photons in the ground photon set (SCT); similarly, and It represents the difference value of the horizontal and vertical adjacent photon weights of all photons in the current non-terrestrial photon set (NSCT). and represents the second norm of the weight difference value of the ground photon set (SCT); similarly, and It represents the two-norm of the weight difference value of the non-terrestrial photon set (NSCT). For the terrestrial photon set (SCT) and the non-terrestrial photon set (NSCT), the corresponding and The calculation method is as follows:

[0067] ;

[0068] .

[0069] In the formula and are the vertical and horizontal weight vectors corresponding to the photon set, and D is the difference matrix, whose values are as follows:

[0070] .

[0071] Step 4, Monte Carlo search for understory terrain points: Use the evaluation functions of the ground photon set and the non-ground photon set as the basis for driving the backpropagation of the Monte Carlo search tree, and use the Monte Carlo tree search algorithm to search for ground photons within the window range.

[0072] The classical Monte Carlo tree search algorithm consists of four parts: selection, expansion, simulation, and backpropagation. These processes are all based on the nodes in the search tree. In this embodiment, an improved Monte Carlo tree search algorithm is used to search for ground photons. The flowchart is shown in the appendix Figure 2 , and the pseudocode is as Figure 3 shown, specifically including:

[0073] First, define the nodes in the Monte Carlo tree search algorithm, including: the reward value Q, the visit count N, the ground photon set SCT, and the non-ground photon set NSCT.

[0074] Then initialize the root node, including initializing the first ground photon and the first non-ground photon in the ground photon set SCT and the non-ground photon set NSCT.

[0075] Next, starting from the root node, select the child node with the largest UCT value among the fully expanded child nodes of the current node as the current node, and perform a total of M rounds of simulations in a loop; among them, if it is the root node, then use the root node as the current node, and if it is not the root node, then select the child node with the largest UCT value from its fully expanded child nodes as the current node.

[0076] Among them, each round of the total simulation process includes K searches. Any k-th search process includes:

[0077] (1) Selection: If the current node is fully expanded, then select the child node with the largest UCT value from all the child nodes of the current node as the node to be simulated s;

[0078] (2) Expansion: If the current node is not fully expanded, then select a ground photon and a non-ground photon near the ground photons and non-ground photons added to the current node respectively, add them to the SCT and NSCT of the expanded child node, and use this expanded child node as the node to be simulated s; among them, the SCT and NSCT of the child node are obtained by adding 1 ground photon and 1 non-ground photon respectively to the 6 photons with the closest vertical weights in the SCT and NSCT of the parent node;

[0079] (3) Simulation: Determine whether the elevation values of the SCT of the node to be simulated s are all less than those of the NSCT. If so, randomly incorporate 1 new photon into the SCT and NSCT of the node to be simulated s respectively, and then return to the judgment and incorporation of new photons until the judgment condition is not met, and end this search simulation;

[0080] (4) Backpropagation: Calculate the incremental value of the gain for the current search simulation according to the photon inclusion order in the termination state of the node to be simulated; and update the gain values and access counts of each node on the entire path from the node to be simulated back to the root node: that is, increment the access count of the passing node by 1, and increment the gain of the passing node by Z.

[0081] In the pseudocode of the improved Monte Carlo tree search algorithm for searching ground photons, the meaning of the single-round search number K is the number of simulations when selecting 1 ground photon point and non-ground photon point each time. In the present invention, the value is 8. The meaning of M is the total number of ground photons within every 30 m window, and its default value is 25% of the number of the set P. The UCT formula for node selection is as follows:

[0082] .

[0083] In the formula, Q and N respectively represent the gain value and access count of the current node, represents the access count of the parent node, C is a constant, and its value in this embodiment is 0.001. This UCT selection formula evenly takes into account the average gain value and access count of the node, so it is beneficial to explore various photon distribution states. The full name of UCT is Upper Confidence Bound Applied to Trees, that is, the upper confidence limit applied to the tree search scenario, which is derived from the upper confidence bound (UCB, Upper Confidence Bound) strategy in the multi-armed bandit problem.

[0084] The formula for calculating the incremental gain value Z during backpropagation is as follows:

[0085] ;

[0086] ;

[0087] .

[0088] In the formula, and respectively represent the Loss value and Gain value when the ground photon set (SCT) adds the i-th ground photon point and the non-ground photon set (NSCT) adds the i-th non-ground photon. n is the number of photons included in the ground photon set and non-ground photon set when the node to be simulated reaches the termination state. a and b are user-defined parameters, and their values in this embodiment are 0.1 and 2 respectively. The purpose is to enhance the proportion of the ground photon score and encourage the algorithm to search for more accurate ground photons.

[0089] Step 5, rapid elimination of abnormal ground photon points: Use the ICESat-2 land vegetation high-product ATL08 to fit an approximate surface profile, and use the approximate surface profile to filter the ground photons obtained by searching through an elevation difference threshold to obtain the understory ground photon set, thereby obtaining the understory ground elevation.

[0090] ICESat-2 land vegetation high-product ATL08 can provide relatively reliable land elevation information at a distance of 100 m along the track. Therefore, it can be used as a fuzzy reference object to quickly eliminate outliers in the ground photons extracted by the present invention. Specifically, in this embodiment, a third-order polynomial is used to fit a number of ground elevation points ( , ) at intervals of 100 m along the track of ATL08 to obtain an approximate surface profile line:

[0091] ;

[0092] Among them, the variable is the along-track distance of each elevation point, and the dependent variable is the elevation of each elevation point. The polynomial parameters are solved according to the least squares principle.

[0093] Then, substitute the photon coordinates ( , ) of the ground photon set SCT extracted in step 4 into the following equation. By setting a threshold T to limit the absolute error between the ground photon elevation value and the interpolated elevation value of the fitted surface, photons with an absolute error less than this threshold are the finally qualified ground photons. In this embodiment, the threshold T is taken as 1, that is, photons within a buffer range of 1 m above and below the approximate surface profile line are valid ground photons.

[0094] .

[0095] The extraction results and comparison charts of the method according to the embodiment of the present invention for experimental area 1 and experimental area 2 are respectively as shown in Figure 4 , Figure 5 . Among them, in Figure 4 (a) and Figure 5 (a), the orange points SCT are the extracted understory ground elevation points, and the green points are the original photon point clouds of ATL03; Figure 4 (b) and Figure 5 (b) are the comparison and verification of the extracted SCT with the high-precision airborne ground elevation data.

[0096] The above embodiments are the preferred embodiments of the present invention. Those of ordinary skill in the art can also make various transformations or improvements on this basis. Without departing from the general concept of the present invention, these transformations or improvements should fall within the scope of protection required by the present invention.

Claims

1. A method for extracting understory ground photon elevation based on ICESat-2, characterized in that, Including: Obtain the ICESat-2 geolocated photon data product ATL03, and filter out the background noise photons therein based on the confidence level of the photons. Divide the filtered photons into windows of a preset size, and calculate the vertical weight and horizontal weight of each photon within each window range. Based on the vertical weight and horizontal weight of the photons, construct an evaluation function for the ground photon set and an evaluation function for the non-ground photon set. Use the evaluation functions of the ground photon set and the non-ground photon set as the basis for driving the backpropagation of the Monte Carlo search tree, and use the Monte Carlo tree search algorithm to search for ground photons within the window range. Use the ICESat-2 land vegetation high product ATL08 to fit an approximate surface profile, and use the approximate surface profile to perform elevation difference threshold filtering on the searched ground photons to obtain the understory ground photon set, thereby obtaining the understory ground elevation.

2. The method for extracting the photon elevation of the forest floor according to claim 1, wherein Filter the background noise photons in ATL03 based on the confidence level of the photons recorded in the field signal_conf_ph in ATL03: regard the photons with signal_conf_ph < 1 as background noise photons, and regard the photons with signal_conf_ph ≥ 2 as signal photons.

3. The method for extracting the photon elevation of the forest floor according to claim 1, characterized in that, The calculation formulas for the vertical weight and horizontal weight of each photon within a certain window range are: ； ； In the formula, is the vertical weight of the i-th photon within the current window range, is the maximum value of the photon elevation within the current window range, is the elevation value of the i-th photon within the current window range; is the along-track distance of the i-th photon within the current window range, is the along-track distance of the photon with the minimum elevation value within the current window range, is the horizontal weight of the i-th photon within the current window range.

4. The method for extracting the photon elevation of the forest floor according to claim 1, characterized in that The evaluation functions for the ground photon set and the non-ground photon set are respectively: ； ； Where, is the evaluation function of the ground photon set; is the evaluation function of the non-terrestrial photon set; ε is a preset minimum constant, N represents the number of photon points in the terrestrial photon set, represents the variance, represents the vertical weight vector of the ground photon set, and Respectively represent the difference values of the horizontal and vertical adjacent photon weights of all photons in the ground photon set; represents the standard deviation, represents the vertical weight vector of the non-ground photon set, and It represents the difference between the horizontal and vertical adjacent photon weights of all photons in the current non-ground photon set.

5. The method for extracting the photon elevation of the forest floor according to claim 1, characterized in that, The specific process of using the evaluation functions of the ground photon set and the non-ground photon set as the basis for driving the backpropagation of the Monte Carlo search tree and using the Monte Carlo tree search algorithm to search for ground photons within the window range is as follows: Define that the nodes in the Monte Carlo tree search algorithm include: the reward value Q, the visit count N, the ground photon set SCT, and the non-ground photon set NSCT. Initialize the root node, including initializing the first ground photon and the first non-ground photon in the ground photon set SCT and the non-ground photon set NSCT. Starting from the root node, select the child node with the largest UCT value among the fully expanded child nodes of the current node as the current node, and perform a total of M rounds of simulations in a loop; among them, if it is the root node, then use the root node as the current node, and if it is not the root node, then select the child node with the largest UCT value from its fully expanded child nodes as the current node. Among them, each round of the total simulation process includes K searches, and any k-th search process includes: (1) Selection: If the current node is fully expanded, select the child node with the largest UCT value from all the child nodes of the current node as the node to be simulated s. (2) Expansion: If the current node is not fully expanded, select ground photons and non-ground photons near the ground photons and non-ground photons added to the current node respectively, incorporate them into the SCT and NSCT of the expanded child node, and use the expanded child node as the node to be simulated s. (3) Simulation: Judge whether the elevation values of the SCT of the node to be simulated s are all less than the elevation values of the NSCT. If so, randomly incorporate 1 new photon into the SCT and NSCT of the node to be simulated respectively, and then return to judge and incorporate new photons until the judgment condition is not met, and end this search simulation. (4) Backpropagation: Calculate the incremental gain value of this simulation according to the photon inclusion order in the termination state of the node to be simulated; and update the gain values and access times of each node on the entire path from the node to be simulated back to the root node.

6. The method for extracting the photon elevation of the forest floor according to claim 5, wherein The calculation formula of the UCT value is: ； Where Q and N respectively represent the revenue value and the number of visits of the current computing node, represents the number of visits of the parent node of the current computing node, and C is a constant; The evaluation function for calculating the incremental gain value Z during backpropagation is: ； Wherein, is the evaluation function value when the i-th ground photon is added to the ground photon set SCT, is the evaluation function value when the i-th non-ground photon is added to the non-ground photon set NSCT, n is the number of photons included in the ground photon set and the non-ground photon set when the node to be simulated reaches the termination state, and a and b are user-defined parameters.

7. The method for extracting the photon elevation of the forest floor according to claim 5, wherein Determine whether the current node is fully expanded according to the number of child nodes. If the number of child nodes reaches the preset value, it is determined that the current node is fully expanded.

8. The method for extracting the photon elevation of the forest floor according to claim 5, characterized in that, Selecting a ground photon and a non-ground photon near the ground photon and non-ground photon added at the current node respectively means randomly selecting 1 photon as the ground photon from a preset number of photons closest to the vertical weight of the ground photon added at the current node, and randomly selecting 1 photon as the non-ground photon from a preset number of photons closest to the vertical weight of the non-ground photon added at the current node.

9. The method for extracting the photon elevation of the forest floor according to claim 1, characterized in that, Using the ICESat-2 land vegetation high-product ATL08 to fit an approximate surface profile specifically means using a third-order polynomial to fit several ground elevation points with an adjacent interval of 100m in the along-track direction of ATL08, so as to obtain an approximate surface profile line: ； In the formula, is the along-track distance of the elevation point, is the elevation of the elevation point; are the parameters to be fitted for the third-order polynomial, which are obtained by using a series of ground elevation point coordinates of ATL08 ( , ) and solving by the least squares method.

10. The method for extracting the photon elevation of the forest floor according to claim 9, wherein Filter the ground photons obtained by searching using the approximate surface profile with an elevation difference threshold, specifically: Substitute the along-track distance of the ground photons obtained by the search into the approximate ground contour line to obtain the interpolated elevation value of the fitted ground surface ; Calculate the interpolated elevation value and the elevation value of the ground photon The absolute difference between ; If the absolute difference is less than the set threshold, the ground photons obtained by the search are valid ground photons; otherwise, they are filtered out.

Images