Dynamic evaluation method and system for soil carbon storage of southern grassland based on remote sensing monitoring

Abstract

The application relates to a fusion remote sensing monitoring southern grassland soil carbon storage dynamic evaluation method and system, and relates to the technical field of soil carbon storage evaluation, and comprises the following steps: drawing a target region topographic map based on multi-source satellite remote sensing data, inputting an initial soil carbon storage inversion model, and outputting an initial soil carbon storage distribution map and a deterministic thermal map; extracting an uncertain region according to a certainty threshold, configuring a supplementary sampling point according to a division rule and marking; obtaining probe robot parameter information, generating a digital twin model in combination with the topographic map, configuring a supplementary sampling scheme, sampling the supplementary sampling point, and obtaining supplementary soil carbon storage data; returning the supplementary soil carbon storage data to the initial soil carbon storage inversion model, generating a corrected soil carbon storage distribution map and a corrected deterministic thermal map, determining that a task is completed when the certainty is up to standard or reaches a resource consumption upper limit, and otherwise, the task is continuously corrected by supplement sampling. The application solves the problems of inaccurate and low-efficiency traditional soil carbon storage evaluation.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of soil carbon storage evaluation, in particular to a method and system for dynamic evaluation of soil carbon storage of southern grassland by fusing remote sensing monitoring. BACKGROUND

[0002] With the increasing demand for data accuracy of soil carbon storage of southern grassland in ecological protection and carbon cycle research, the reliability of dynamic evaluation becomes a key requirement.

[0003] At present, the traditional method for correcting soil carbon storage data obtained by remote sensing often uses unmanned aerial vehicle soil sampling, but the environmental adaptability of unmanned aerial vehicles is poor, and it is difficult to land and sample in harsh weather or complex ground environment. At the same time, the existing technology lacks targeted sampling setting for the "uncertain area" with the highest model calibration value, and there are only two ways of high-precision low-efficiency ground grid sampling and low-precision high-efficiency pure remote sensing model, resulting in low efficiency of model correction of soil carbon storage data, which is difficult to support accurate evaluation. SUMMARY

[0004] The present application provides a method and system for dynamic evaluation of soil carbon storage of southern grassland by fusing remote sensing monitoring, which improves the current situation of poor environmental adaptability of traditional unmanned aerial vehicle sampling, lack of targeted sampling of "uncertain area" and low correction efficiency.

[0005] The embodiments of the present application disclose the following technical solutions:

[0006] In a first aspect, the embodiments of the present application provide a method for dynamic evaluation of soil carbon storage of southern grassland by fusing remote sensing monitoring, which comprises:

[0007] Based on the multi-source satellite remote sensing data of the target area, a topographic map of the target area is drawn, and the multi-source satellite remote sensing data of the target area is input into an initial soil carbon storage inversion model to output an initial soil carbon storage distribution map and a deterministic thermal map of the target area, wherein the deterministic thermal map is composed of a plurality of pixels labeled with a degree of certainty;

[0008] Based on the initial soil carbon storage distribution map and the deterministic thermal map, an uncertain area is extracted according to a preset degree of certainty threshold, and a plurality of supplementary sampling points are configured for the uncertain area according to a preset division rule, and a plurality of supplementary sampling points are labeled on the topographic map;

[0009] Obtain the parameter information of all detection robots in the target area, generate a digital twin model based on the parameter information of each detection robot and the topographic map of the target area, configure a supplementary sampling scheme based on the digital twin model, and non-destructively sample a plurality of supplementary sampling points to obtain a plurality of supplementary soil carbon storage data;

[0010] The several supplementary soil carbon storage data are fed back to the initial soil carbon storage inversion model to generate a corrected soil carbon storage distribution map and a corrected certainty heat map of the target region, and when the certainty of the entire target region reaches a certainty threshold or reaches a predetermined upper limit of resource consumption, it is determined that the dynamic evaluation task is completed, otherwise, supplementary sampling is continued, and the initial soil carbon storage inversion model and the initial soil carbon storage distribution map are corrected.

[0011] In a second aspect, the embodiments of the present application provide a dynamic evaluation system for soil carbon storage of southern grassland based on remote sensing monitoring, which comprises:

[0012] An initial inversion mapping module is configured to draw a topographic map of a target region based on multi-source satellite remote sensing data of the target region, and input the multi-source satellite remote sensing data of the target region into an initial soil carbon storage inversion model to output an initial soil carbon storage distribution map and a certainty heat map of the target region, wherein the certainty heat map is composed of a plurality of pixels marked with certainty.

[0013] An uncertain region sampling configuration module is configured to extract an uncertain region according to a preset certainty threshold based on the initial soil carbon storage distribution map and the certainty heat map, and configure a plurality of supplementary sampling points for the uncertain region according to a preset division rule, and mark the plurality of supplementary sampling points on the topographic map.

[0014] A digital twin sampling module is configured to obtain parameter information of all detection robots in the target region, generate a digital twin model based on the parameter information of each detection robot and the topographic map of the target region, configure a supplementary sampling scheme based on the digital twin model, and non-destructively sample the plurality of supplementary sampling points to obtain a plurality of supplementary soil carbon storage data.

[0015] A model correction and determination module is configured to feed back the plurality of supplementary soil carbon storage data to the initial soil carbon storage inversion model to generate a corrected soil carbon storage distribution map and a corrected certainty heat map of the target region, and when the certainty of the entire target region reaches a certainty threshold or reaches a predetermined upper limit of resource consumption, it is determined that the dynamic evaluation task is completed, otherwise, supplementary sampling is continued, and the initial soil carbon storage inversion model and the initial soil carbon storage distribution map are corrected.

[0016] One or more technical solutions provided in the present application have at least the following technical effects or advantages:

[0017] The application provides a dynamic evaluation method and system for soil carbon storage of southern grassland based on remote sensing monitoring, which realizes accurate dynamic evaluation of soil carbon storage of southern grassland through the cooperative operation of the following steps: reversing the initial carbon storage distribution and the deterministic thermal map based on remote sensing data, configuring supplementary sampling points for uncertain areas, optimizing the robot sampling scheme by a digital twin model, obtaining measured data by non-destructive sampling, and iterative evaluation of the model by incremental training. First, based on multi-source satellite remote sensing data of the target area, a topographic map containing terrain features is drawn, the remote sensing data is input into the pre-trained initial soil carbon storage inversion model, and the initial soil carbon storage distribution map and the deterministic thermal map are output. Then, the uncertain areas are extracted according to the preset degree threshold, the initial sampling points are generated by using the spatial random sampling algorithm, and the supplementary sampling points are determined by filtering the impassable points of the robot in combination with the topographic map. Then, the parameters of the detection robot are obtained, the digital twin model is constructed in combination with the topographic map, the reachable range and sampling time and energy consumption of the detection robot are simulated, the optimal supplementary sampling scheme is configured, the detection robot is controlled to carry out non-destructive sampling by using the contact near-infrared spectrum probe, and the supplementary soil carbon storage data are obtained. Finally, the supplementary data are returned, the new training samples are formed by using the measured organic carbon data and the corresponding remote sensing data, the initial model is incrementally trained to obtain a modified model, the remote sensing data are reprocessed to generate a modified distribution map and a thermal map, and the global certainty or resource consumption is evaluated. If the result meets the requirements, the result is output; otherwise, the sampling and model modification are repeated.

[0018] The technical scheme solves the problems of poor environmental adaptability of unmanned aerial vehicles, lack of targeted sampling in uncertain areas and low model modification efficiency in traditional soil carbon storage evaluation, realizes the precision, efficiency and ecology of the evaluation of soil carbon storage of southern grassland, and provides technical support for carbon sink research, ecological protection and dynamic monitoring of soil carbon storage of southern grassland. BRIEF DESCRIPTION OF DRAWINGS

[0019] In order to more clearly illustrate the technical solutions in the embodiments of the application, the drawings needed in the embodiment description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the application, and other drawings can be obtained by those skilled in the art without creative labor.

[0020] Figure 1 A flowchart of the dynamic evaluation method for soil carbon storage of southern grassland based on remote sensing monitoring provided by the embodiment of the application is shown in the figure.

[0021] Figure 2 A structure diagram of the dynamic evaluation system for soil carbon storage of southern grassland based on remote sensing monitoring provided by the embodiment of the application is shown in the figure.

[0022] In the drawings, the components represented by the numbers are described as follows:

[0023] An initial inversion drawing module 01, an uncertain area sampling configuration module 02, a digital twin sampling module 03, and a model correction determination module 04. DETAILED DESCRIPTION

[0024] The application provides a dynamic evaluation method and system for soil carbon reserves of southern grasslands by fusing remote sensing monitoring, to solve the technical problems in the prior art that the environment adaptability of unmanned aerial vehicle sampling is poor, targeted sampling of "uncertain areas" is lacking, and only inefficient grid sampling or low-precision remote sensing sampling can be performed, resulting in low efficiency of soil carbon reserve data correction.

[0025] The technical solutions in the embodiments of the application will be clearly and completely described below with reference to the drawings in the embodiments of the application. Obviously, the described embodiments are only part of the embodiments of the application, rather than all the embodiments of the application. Based on the embodiments in the application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the application.

[0026] In the description of the application, the terms "first", "second" are used only for the purpose of description, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined as "first", "second" can explicitly or implicitly include one or more of the features. In the description of the application, the meaning of "multiple" is two or more, unless otherwise specifically limited.

[0027] In the description of the application, the term "for example" is used to indicate "as an example, illustration or explanation". Any embodiment described as "for example" in the application is not necessarily interpreted as more preferred or more advantageous than other embodiments. The following description is given in order to enable any person skilled in the art to implement and use the application. In the following description, details are listed for the purpose of explanation. It should be understood that those skilled in the art can realize the application without using these specific details. In other examples, well-known structures and processes will not be described in detail to avoid unnecessary details making the description of the application obscure. Therefore, the application is not intended to be limited to the embodiments shown, but is consistent with the broadest scope in accordance with the principles and characteristics disclosed in the application.

[0028] Embodiment one, as shown in FIG. 1, the application provides a dynamic evaluation method for soil carbon reserves of southern grasslands by fusing remote sensing monitoring, the method comprising the following steps: Figure 1

[0029] ​S110: Based on the multi-source satellite remote sensing data of the target region, a topographic map of the target region is drawn, and the multi-source satellite remote sensing data of the target region is input into an initial soil carbon storage inversion model to output an initial soil carbon storage distribution map and a deterministic thermal map of the target region, wherein the deterministic thermal map is composed of a plurality of pixels marked with a degree of certainty.

[0030] In the embodiment of the application, in the scenario of dynamically evaluating the soil carbon storage of southern grassland, in order to obtain the topographic basic information and the preliminary carbon storage distribution of the target region, and at the same time to clearly determine the credibility of the carbon storage prediction results of the model, it is necessary to carry out topographic map drawing and model calling based on the multi-source satellite remote sensing data of the target region, to provide initial basis for subsequent targeted supplementary sampling and model correction.

[0031] Specifically, first, a topographic map containing topographic features is generated according to the multi-source satellite remote sensing data of the target region. The generation process can be realized by existing remote sensing data direct mapping technology to ensure that the topographic map can accurately reflect the topographic conditions of the target region.

[0032] Further, the pre-trained initial soil carbon storage inversion model is called, and the multi-source satellite remote sensing data of the target region is input into the model, so that the initial soil carbon storage distribution map and the deterministic thermal map of the target region can be output by the model to quickly obtain the preliminary distribution of the soil carbon storage of the target region and the credibility of the model prediction results.

[0033] This step realizes the preliminary evaluation and prediction of the southern grassland soil carbon storage by relying on the efficient processing of the pre-trained model on the multi-source satellite remote sensing data, and provides data support for subsequent accurate identification of uncertain areas that need to be supplemented sampling and development of a scientific sampling scheme.

[0034] The step S110 in the method provided in the embodiment of the application comprises:

[0035] According to the multi-source satellite remote sensing data of the target region, a topographic map of the target region containing topographic features is generated;

[0036] The pre-trained initial soil carbon storage inversion model is called, and the multi-source satellite remote sensing data of the target region is input into the model, so that the initial soil carbon storage distribution map and the deterministic thermal map of the target region can be output by the model to quickly obtain the preliminary distribution of the soil carbon storage of the target region and the credibility of the model prediction results.

[0037] In the embodiment of the application, in order to establish the dynamic evaluation basis of the soil carbon storage of southern grassland, avoid the deviation of the inversion results caused by single remote sensing data or insufficient generalization ability of the model, and improve the pertinence and overall evaluation efficiency of the subsequent sampling and correction links, a reliable topographic map and a prediction model with self-evaluation ability need to be constructed in the early evaluation stage.

[0038] Specifically, first, a topographic map containing topographic features is generated according to multi-source satellite remote sensing data of the target area. The generation of the topographic map adopts a remote sensing data topographic inversion method in the prior art, and by extracting topographic feature information in the multi-source satellite remote sensing data, the topographic details of the grassland in the target area, such as the topographic undulation, valley distribution, steep slope and gentle slope, are accurately presented, so as to meet the needs of subsequent planning of the sampling path of the exploration robot and screening of the passable sampling points.

[0039] The multi-source satellite remote sensing data includes but is not limited to optical images, radar data, multispectral and hyperspectral data, and by integrating the advantages of different types of data, the surface coverage, topographic structure and soil-related feature information of the target area are comprehensively obtained.

[0040] At the same time of generating the topographic map, a pre-trained initial soil carbon storage inversion model is called, and the multi-source satellite remote sensing data of the target area is inputted, so as to output an initial soil carbon storage distribution map and a deterministic thermal map.

[0041] In the method provided by the embodiment of the application, the method for obtaining the initial soil carbon storage inversion model includes:

[0042] A plurality of southern grasslands similar to the target area in water-heat conditions and vegetation types are selected as a plurality of sample grasslands;

[0043] The historical multi-source satellite remote sensing data of the plurality of sample grasslands and the historical measured soil organic carbon data corresponding to the historical multi-source satellite remote sensing data are obtained, and the historical measured soil organic carbon data at least includes a sampling depth and a soil organic carbon content;

[0044] The initial soil carbon storage inversion model is built based on machine learning;

[0045] The initial soil carbon storage inversion model is trained using the historical multi-source satellite remote sensing data and the historical measured soil organic carbon data until convergence.

[0046] In the embodiment of the application, in order to enable the initial soil carbon storage inversion model to have the ability to adapt to the characteristics of the southern grassland and invert the soil carbon storage, the process of screening similar samples, collecting matching data, building a model framework and completing training is required, so as to ensure that the model can output reliable initial carbon storage distribution and deterministic evaluation results based on the multi-source satellite remote sensing data.

[0047] Specifically, first, a plurality of southern grasslands similar to the target area in water-heat conditions and vegetation types are selected as sample grasslands. The water-heat conditions need to refer to key indicators such as the annual average precipitation and the annual average temperature of the target area, and the vegetation types need to match the dominant herbaceous plant species of the target area.

[0048] Exemplarily, if the target region is a subtropical monsoon climate with dogtooth grass as the main vegetation in the southern grassland, the sample grassland needs to be screened in the region with an average annual rainfall of 800-1500 mm, an average annual temperature of 15-22°C, and also with dogtooth grass as the dominant vegetation, to ensure that the ecological environment of the sample grassland is highly similar to that of the target region.

[0049] Further, the historical multi-source satellite remote sensing data of the sample grassland and the historical measured soil organic carbon data corresponding to the historical period are obtained.

[0050] Among them, the historical multi-source satellite remote sensing data needs to cover the image information of the sample grassland in different time dimensions to fully capture the dynamic characteristics of the surface vegetation coverage and soil moisture of the sample grassland; the historical measured soil organic carbon data needs to be obtained by field sampling, and at least includes the sampling depth and the soil organic carbon content, for example, multiple sampling points are set in each sample grassland, and soil samples of 0-10 cm, 10-20 cm, and 20-30 cm soil layers are collected respectively, and the organic carbon content of each soil layer is detected and recorded, to ensure that the measured data can provide accurate output labels for the model.

[0051] Further, an initial soil carbon storage inversion model is built based on machine learning to establish a precise mapping relationship between "remote sensing data features-soil organic carbon content", providing reliable model support for subsequent carbon storage evaluation of the target region.

[0052] Specifically, a random forest algorithm can be used to build an initial soil carbon storage inversion model. The historical multi-source satellite remote sensing data is used as the input features of the model; at the same time, the historical measured soil organic carbon content is used as the output target of the model, and the multi-decision tree ensemble learning characteristics of the random forest algorithm are used to effectively integrate the advantages of different types of satellite remote sensing data.

[0053] Further, the initial soil carbon storage inversion model is trained using the collected historical multi-source satellite remote sensing data and historical measured soil organic carbon data until convergence.

[0054] Before training, the data needs to be preprocessed, that is, the multi-source satellite remote sensing data is subjected to radiation correction and geometric correction to eliminate atmospheric interference and image distortion; the measured data is subjected to outlier rejection to avoid the influence of individual extreme data on the model training effect.

[0055] During the training process, the preprocessed data set needs to be divided into a training set and a validation set, for example, allocated in a ratio of 7:3, 70% of the data is used for model parameter learning, and 30% of the data is used for real-time evaluation of the model generalization ability.

[0056] During training, the training set is first input into the model, and multiple decision trees are constructed through Bootstrap sampling of the random forest algorithm, each decision tree is split based on a different feature subset to reduce the risk of model overfitting; At the same time, by adjusting parameters such as the number of decision trees, the maximum tree depth, and the minimum number of samples for node splitting, the model fitting effect is optimized.

[0057] After each round of training, the validation set data is input into the model to calculate the prediction error of the model. The root mean square error (RMSE) is commonly used as an evaluation index to intuitively reflect the deviation between the predicted value and the measured value.

[0058] Further, when the root mean square error of the model on the validation set stabilizes within the preset threshold, and there is no significant decrease in error after 5-10 consecutive training rounds, the model training is considered to have converged, and the model has stable soil carbon storage inversion capability, which can be used as an initial soil carbon storage inversion model for subsequent evaluation of the target area.

[0059] Further, the multi-source satellite remote sensing data of the target area is input into the pre-trained initial soil carbon storage inversion model. The model integrates and analyzes information such as the vegetation coverage reflected by optical images, the soil surface structure captured by radar data, and the soil organic matter spectral characteristics contained in multispectral and hyperspectral data, and outputs the initial soil carbon storage distribution map and the certainty heat map of the target area, to respectively present the preliminary spatial distribution rule of carbon storage and the credibility of the model's prediction results in each region, providing core basis for subsequent extraction of uncertain areas and planning of supplementary sampling.

[0060] For example, if the target area is a subtropical monsoon climate with Cynodon dactylon as the main vegetation in southern grassland, the multi-source satellite remote sensing data input into the initial soil carbon storage inversion model includes optical images in summer, radar data in the rainy season, and hyperspectral data in the growing season.

[0061] After the model is analyzed, the initial soil carbon storage distribution map will mark the approximate range of soil organic carbon content in different regions, and the certainty heat map will mark the certainty with different colors to clearly distinguish between credible and suspicious areas predicted by the model.

[0062] S120: Based on the initial soil carbon storage distribution map and the certainty heat map, extract the uncertain area according to the preset certainty threshold, and configure a number of supplementary sampling points for the uncertain area according to the preset division rule, and mark a number of supplementary sampling points on the topographic map;

[0063] In the embodiments of the present application, in order to accurately locate the area with low model prediction reliability, the uncertain area that needs to be verified in priority and the reasonable sampling points are screened out based on the initial carbon reserve distribution and the model certainty evaluation result, so as to obtain reliable measured data through targeted supplementary sampling, and lay a foundation for subsequent model correction and carbon reserve evaluation precision improvement.

[0064] Specifically, first, a determination threshold is preset. The setting of the threshold needs to be determined in combination with the accuracy requirement of the evaluation task, the complexity of the terrain and vegetation of the target area, so as to ensure that the reliable and suspicious areas of the model prediction result can be accurately distinguished.

[0065] Further, the pixel area in the certainty heat map that does not meet the preset determination threshold is divided into an uncertain area. Through this step, the area with low prediction confidence caused by insufficient training samples, complex terrain or blurred remote sensing signals of the model can be circled from the whole area, so as to clearly define the key range for subsequent sampling work.

[0066] Further, a spatial random sampling algorithm is used to generate initial sampling points with a preset density in the uncertain area. The preset density needs to be set according to the area size, terrain complexity and accuracy requirement of the evaluation task of the uncertain area, and the spatial random sampling can ensure that the initial sampling points are uniformly distributed in the uncertain area, so as to comprehensively cover different sub-areas of the area.

[0067] Finally, the generated topographic map is called, and the sampling points in the initial sampling points that are located in the terrain inaccessible to the detection robot are filtered out, and the remaining points are the supplementary sampling points, and these supplementary sampling points are marked on the topographic map to provide guidance for the detection robot to plan the sampling path.

[0068] The step S120 in the method provided by the embodiments of the present application comprises:

[0069] presetting a determination threshold;

[0070] dividing the pixel area in the certainty heat map that does not meet the determination threshold into an uncertain area;

[0071] generating initial sampling points with a preset density in the uncertain area by using a spatial random sampling algorithm;

[0072] calling the topographic map, filtering out the sampling points in the initial sampling points that are located in the terrain inaccessible to the detection robot, and the remaining points are the supplementary sampling points.

[0073] In the embodiments of the present application, in order to accurately lock the area with low model prediction reliability and which needs to be verified, the steps of explicitly determining the threshold, dividing the uncertain area, and generating and screening the initial sampling points are needed to obtain the supplementary sampling points for the model, so as to improve the effectiveness of the subsequent sampling data and the accuracy of the model correction.

[0074] Specifically, first, the threshold of the determination degree is preset. The threshold of the determination degree is determined in combination with the evaluation accuracy requirement of the target area, the complexity of the terrain, and the vegetation coverage characteristics, for example, if the carbon storage data accuracy requirement of the evaluation task is high, the threshold of the determination degree can be set to 80%, and if only the trend needs to be preliminarily evaluated, the threshold of the determination degree can be appropriately reduced to 70%.

[0075] At the same time, it is necessary to avoid that the threshold is too high to cause the uncertainty area to be too large and the sampling cost to increase sharply, or the threshold is too low to cause the low-reliability area to be not identified, so as to ensure that the reliable area and the suspicious area of the model prediction result can be accurately distinguished.

[0076] Further, the pixel area in the obtained certainty heat map that does not meet the threshold of the determination degree is divided into an uncertain area. The area that does not meet the threshold is mostly an area with insufficient model training samples, complex terrain, mixed vegetation, or blurred remote sensing signals. By dividing these areas into the uncertain area, the subsequent sampling work can be focused on the area with the highest model calibration value, and invalid sampling on the area that can be reliably predicted by the model can be avoided, so as to improve the pertinence of the sampling.

[0077] Further, a spatial random sampling algorithm is used to generate initial sampling points with a preset density in the uncertain area.

[0078] The preset density is set according to the area size of the uncertain area, the terrain heterogeneity, and the accuracy requirement of the evaluation task, for example, for an uncertain area with an area of 100 square kilometers and a complex terrain, the density of 5 sampling points per square kilometer can be set to ensure that the initial sampling points can uniformly cover different sub-areas in the area.

[0079] In addition, the spatial random sampling algorithm can avoid excessive concentration or dispersion of the sampling points, ensure that the subsequently obtained sampling data is representative, and can fully reflect the real situation of the carbon storage of the uncertain area.

[0080] Finally, a topographic map containing terrain features generated based on multi-source satellite remote sensing data in the early stage is called, and the sampling points located in the terrain that the detection robot cannot pass through are filtered out from the initial sampling points, and the remaining points are the supplementary sampling points.

[0081] Wherein, the detection robot is limited by the chassis structure (such as light track or six-legged wheel-legged chassis), it is difficult to pass through the terrain such as steep slope (slope more than 30°), deep ditch (depth more than 0.5 meters) or marsh, the terrain map screening can exclude these unreachable points in advance, avoid the robot unable to complete the task due to terrain obstacles during subsequent sampling, and ensure the practical operability of the supplementary sampling points.

[0082] At the same time, the screened supplementary sampling points focus on the uncertain area and adapt to the robot's passing ability, laying a foundation for subsequent efficient non-destructive sampling and obtaining high-quality supplementary carbon storage data.

[0083] S130: Obtain parameter information of all detection robots in the target area, generate a digital twin model based on the parameter information of each detection robot and the topographic map of the target area, configure a supplementary sampling scheme based on the digital twin model, and perform non-destructive sampling on a plurality of supplementary sampling points to obtain a plurality of supplementary soil carbon storage data;

[0084] In the embodiments of the present application, in order to avoid low sampling efficiency or failure of sampling task caused by unclear parameters and unreasonable path planning of the detection robot, while ensuring accurate acquisition of soil carbon storage data of the supplementary sampling points, the robot parameters are obtained, the digital twin model is constructed to configure a scientific sampling scheme and perform non-destructive sampling, so as to efficiently obtain reliable supplementary data to support the correction of the initial soil carbon storage inversion model and ensure the evaluation accuracy.

[0085] Specifically, first, the parameter information of all detection robots in the target area is obtained, which includes at least initial position, battery capacity, motion speed, obstacle crossing performance and core sensor state. By comprehensively mastering the basic parameters of the robot, the working capacity and limitation of each robot can be determined, providing a basis for subsequent scheme configuration.

[0086] Further, based on the parameter information of all detection robots and the topographic map containing terrain features generated in the early stage, a digital twin model is created. The digital twin model can simulate the reachable range of each detection robot, clearly present the terrain area where the robot cannot pass, and also calculate the sampling time and sampling energy consumption of the robot from the initial position to each supplementary sampling point for sampling.

[0087] Further, the labeled positions of a plurality of supplementary sampling points on the topographic map and the initial positions of all detection robots are imported into the digital twin model, and simulation sampling is performed based on the model. Through simulation, the situation of different robots performing sampling tasks can be directly observed, and then the best supplementary sampling scheme with shorter total time consumption, lower total energy consumption and covering all supplementary sampling points is selected.

[0088] Finally, according to the optimal supplementary sampling scheme, the detection robot is controlled to non-destructively sample a plurality of supplementary sampling points. The plurality of supplementary soil carbon storage data obtained in this way can ensure the authenticity and effectiveness of the data, avoid damage to the soil, and meet the needs of southern grassland ecological protection.

[0089] The step S130 in the method provided by the embodiment of the application comprises:

[0090] Parameter information of all detection robots in the target area is acquired, wherein the parameter information at least comprises initial positions, battery capacities, motion speeds, obstacle surmounting performances, and core sensor states;

[0091] Based on the parameter information of all detection robots and the topographic map, a digital twin model is created, which can be used to simulate the reachable range of each detection robot and the sampling time consumption and sampling energy consumption of each detection robot from the initial position to each supplementary sampling point for sampling;

[0092] The plurality of supplementary sampling points are imported into the digital twin model at the marked positions of the topographic map and the initial positions of all detection robots, and the digital twin model is used to simulate sampling to obtain an optimal supplementary sampling scheme, and the plurality of supplementary sampling points are non-destructively sampled to obtain a plurality of supplementary soil carbon storage data.

[0093] In the embodiment of the application, in order to avoid low sampling efficiency caused by unclear operation parameters and unreasonable sampling path planning of the detection robot, a process of comprehensively mastering robot parameters, constructing a digital twin model to optimize a sampling scheme, and using a non-destructive sampling technology is needed to balance sampling efficiency, ecological protection, and data effectiveness, and promote the orderly development of the evaluation work.

[0094] Specifically, first, parameter information of all detection robots in the target area is acquired, and the parameter information at least comprises initial positions, battery capacities, motion speeds, obstacle surmounting performances, and core sensor states.

[0095] The initial position can clearly indicate the starting coordinates of the operation of the detection robot, and provide a starting point basis for subsequent path planning; the battery capacity is directly related to the endurance of the detection robot, and determines the sampling range that can be covered by the detection robot at a time; the motion speed is a key index for calculating sampling time consumption, and affects the overall sampling efficiency; the obstacle surmounting performance needs to be adapted to the rugged terrain of the southern grassland, and judges whether the detection robot can reach a specific sampling point; and the core sensor state (i.e. available or unavailable) directly determines whether the robot can normally acquire soil related data, and avoids invalid sampling data caused by sensor failure.

[0096] Further, based on the parameter information of all detection robots and the topographic map containing the terrain features generated in the early stage, a digital twin model is created.

[0097] Specifically, the digital twin model can fuse the detection robot parameters with the terrain information, which can simulate the reachable range of each detection robot, clearly mark the areas that the robot cannot pass due to terrain restrictions, and avoid the occurrence of unexecutable points in the subsequent sampling scheme.

[0098] On the other hand, the sampling time and sampling energy consumption of each robot from the initial position to reach each supplementary sampling point for sampling can be accurately calculated, for example, the time consumption is calculated according to the terrain distance between the robot movement speed and the sampling point, and the total energy consumption is calculated by combining the energy consumption loss when the obstacle is overcome, which provides a quantitative basis for subsequent optimization of the sampling scheme.

[0099] Further, the labeled positions of the plurality of supplementary sampling points on the terrain map and the initial positions of all detection robots are introduced into the constructed digital twin model, and the best supplementary sampling scheme is obtained by simulating sampling based on the model.

[0100] In the method provided by the embodiments of the present application, "introducing the labeled positions of the plurality of supplementary sampling points on the terrain map and the initial positions of all detection robots into the digital twin model, simulating sampling according to the digital twin model, obtaining the best supplementary sampling scheme, and non-destructively sampling the plurality of supplementary sampling points to obtain a plurality of supplementary soil carbon storage data" includes:

[0101] For each supplementary sampling point, the movement cost from the current position of any detection robot to the supplementary sampling point is calculated in combination with the terrain features in the terrain map and the parameter information of the detection robot, wherein the movement cost is a weighted function of the total sampling time and the total sampling energy consumption.

[0102] The total sampling time and the total sampling energy consumption are taken as the optimization target, and the exclusive supplementary sampling point sequence is allocated to all detection robots in cooperation.

[0103] For the supplementary sampling point sequence allocated to each detection robot, the optimal movement path is planned in combination with the terrain map.

[0104] The detection robots are controlled to sample according to the optimal movement path, and when a passage obstacle is encountered, the supplementary sampling point sequence and the movement path are dynamically adjusted to obtain the best supplementary sampling scheme.

[0105] Specifically, first, for each supplementary sampling point, the movement cost from the current position of any detection robot to the supplementary sampling point is calculated in combination with the terrain features in the terrain map and the parameter information of the detection robot.

[0106] In the topographic map, the topographic features need to be focused on the rugged terrain of the southern grassland, such as the size of the slope, whether there are valleys or protrusions, etc. These features directly affect the difficulty of the robot's passage and energy consumption. The movement cost, as a weighted function of the sampling time and sampling energy consumption, can adjust the weight according to the needs of the evaluation task. If it is necessary to complete the sampling quickly, the weight of the sampling time can be increased. If it is necessary to control energy consumption, the weight of the sampling energy consumption can be increased. Through this calculation, the cost of each exploration robot reaching the sampling point is quantified, providing a basis for subsequent task allocation.

[0107] Further, taking the total sampling time and total sampling energy consumption as the optimization goal, all exploration robots are collaboratively allocated with exclusive supplementary sampling point sequences.

[0108] During the allocation process, the total time and total energy consumption under different allocation schemes need to be simulated based on the digital twin model to avoid unbalanced situations where some exploration robots have too concentrated tasks and some robots are idle, while ensuring that all supplementary sampling points are covered.

[0109] For example, let the exploration robot with an initial position close to a piece of concentrated sampling point and sufficient battery capacity undertake the sampling task in that area, let the robot with fast movement speed but general obstacle crossing performance be responsible for sampling in the flat area, and let the robot with strong obstacle crossing performance but slow speed be responsible for sampling in the complex terrain area. Through collaborative allocation, the total sampling cost is minimized.

[0110] Further, for the supplementary sampling point sequence allocated to each exploration robot, its optimal movement path is planned in combination with the topographic map.

[0111] Specifically, when planning, routes with gentle terrain and no obvious obstacles should be prioritized to avoid steep slopes, deep valleys, and other areas in the southern grassland that exceed the obstacle crossing ability of the exploration robot, reducing energy consumption and failure risk during the passage of the exploration robot. At the same time, the round trip distance between the exploration robots at each sampling point should be minimized to avoid time-consuming detours.

[0112] For example, if the sampling point sequence of a certain exploration robot is concentrated in the eastern part of the target area, and there is a gentle natural passage in that area, the robot can be prioritized to travel along that passage when planning the path to reach each sampling point in sequence, improving the passage efficiency.

[0113] Finally, the exploration robots are controlled to sample according to the optimal movement path, and when encountering passage obstacles, the supplementary sampling point sequence and movement path are dynamically adjusted to obtain the best supplementary sampling scheme.

[0114] Wherein, the field environment of the southern grassland may exist temporary obstacles not marked on the topographic map, such as fallen vegetation, temporary water pits, etc. When the robot encounters such obstacles and cannot pass according to the original path, the obstacle information needs to be fed back in real time through the digital twin model, and the mobile cost is quickly recalculated based on the model to adjust the supplementary sampling point sequence of the detection robot (for example, preferentially collecting sampling points that can be reached around the obstacle, and assisting in collecting sampling points behind the obstacle after other robots complete the task) or to re-plan the mobile path (for example, passing on the other side of the obstacle).

[0115] In the method provided by the embodiments of the present application, the detection robot is equipped with a contact near-infrared spectrum probe, and the supplementary soil carbon storage data of the soil can be obtained without drilling.

[0116] Specifically, all detection robots use non-destructive sampling methods, are equipped with high-precision contact near-infrared spectrum probes, use light track or multi-foot wheel leg type chassis to adapt to rugged terrain, and after driving to the supplementary sampling point, the spectrum probe is closely attached to the ground, so that the soil spectrum data and the soil organic carbon content can be quickly obtained without drilling, which not only avoids the damage to the grassland ecology caused by traditional drilling sampling, but also solves the problem that the unmanned aerial vehicle cannot sample in bad weather or complex terrain.

[0117] After the detection robot drives to the supplementary sampling point, the mechanical arm adjusts the angle and height according to the preset program, stably lowers the high-precision contact near-infrared spectrum probe to the ground, ensures that the probe is closely attached to the soil surface without gap, and avoids air interference to cause distortion of the spectrum data.

[0118] Meanwhile, the probe will quickly emit near-infrared light and receive the spectrum signal reflected by the soil after being attached, and synchronously transmit the spectrum data to the processor built-in the detection robot, the processor calls the pre-calibrated soil organic carbon inversion model, calculates the supplementary soil organic carbon content of the supplementary sampling point by analyzing the characteristic absorption peak in the spectrum data.

[0119] In addition, after completing the data acquisition of each supplementary sampling point, the detection robot will automatically record the coordinates, sampling time and corresponding supplementary soil organic carbon content of the point, forming a complete sampling record. At the same time, the detection robot will real-time return these data to the cloud through the wireless communication module, which is convenient for subsequent integration and model correction.

[0120] S140: return the supplementary soil carbon storage data to the initial soil carbon storage inversion model to generate a corrected soil carbon storage distribution map and a corrected certainty heat map of the target region, when the certainty of the entire target region reaches the certainty threshold or reaches the predetermined upper limit of resource consumption, it is determined that the dynamic evaluation task is completed, otherwise, continue to supplement sampling, and correct the initial soil carbon storage inversion model and the initial soil carbon storage distribution map.

[0121] In the embodiments of the present application, in order to optimize the model inversion capability with the measured sampling data, improve the accuracy of the carbon reserve evaluation result, and at the same time avoid resource waste caused by unlimited sampling, the supplementary soil carbon reserve data needs to be returned to the initial soil carbon reserve inversion model to complete the correction and iterative evaluation until the accuracy requirement is met or the upper limit of resource consumption is reached, so as to output reliable dynamic evaluation results of soil carbon reserve.

[0122] Specifically, first, the obtained supplementary soil carbon reserve data is returned to the initial soil carbon reserve inversion model. That is, the measured soil organic carbon data is extracted from the supplementary soil carbon reserve data, and the corresponding multi-source satellite remote sensing data is used to form a new training sample. The initial soil carbon reserve inversion model is incrementally trained using these training samples to obtain a corrected soil carbon reserve inversion model, so that the model can optimize the parameters in combination with the actual situation of the target area and reduce the prediction bias.

[0123] Further, the multi-source satellite remote sensing data of the target area is reprocessed using the corrected model to generate a corrected soil carbon reserve distribution map and a corrected certainty heat map. The corrected soil carbon reserve distribution map is more consistent with the actual carbon reserve distribution, and the corrected certainty heat map can update the certainty of each region to reflect the prediction credibility of the model after correction.

[0124] Further, the corrected certainty heat map is evaluated. If the global certainty reaches the certainty threshold, it means that the model prediction accuracy meets the standard. If the accuracy does not meet the standard but the task resource consumption (including total sampling time consumption and total sampling energy consumption) has reached the preset upper limit, the cost needs to be weighed to determine the completion of the task.

[0125] On the contrary, if the certainty does not meet the standard and the resources do not exceed the upper limit, the supplementary sampling link needs to be returned to obtain new supplementary data to correct the model and update the carbon reserve distribution map again, and the steps of "supplementary sampling - model correction - heat map evaluation" are executed in a loop.

[0126] When the task is completed, the final corrected soil carbon reserve distribution map is taken as the dynamic evaluation result. This result combines the wide coverage advantage of remote sensing data and the calibration value of the measured data, and can be used as a reliable basis for the study of soil carbon reserve in southern grasslands.

[0127] The step S140 in the method provided in the embodiments of the present application includes:

[0128] The measured soil organic carbon data in the supplementary soil carbon reserve data is combined with the corresponding multi-source satellite remote sensing data to form a new training sample, and the initial soil carbon reserve inversion model is incrementally trained to obtain a corrected soil carbon reserve inversion model.

[0129] The multi-source satellite remote sensing data of the target region is reprocessed by using the corrected soil carbon storage inversion model to generate a corrected soil carbon storage distribution map and a corrected certainty heat map;

[0130] The corrected certainty heat map is evaluated, and if the global certainty reaches a certainty threshold or the task resource consumption has reached a preset upper limit, it is determined that the task is completed, and the final corrected soil carbon storage distribution map is output, wherein the task resource consumption includes total sampling time consumption and total sampling energy consumption;

[0131] Otherwise, the steps of supplementary sampling, correcting the soil carbon storage inversion model, obtaining the corrected certainty heat map, and evaluating the corrected certainty heat map are repeatedly executed until the task is completed.

[0132] The final output corrected soil carbon storage distribution map is taken as the dynamic evaluation result of the soil carbon storage.

[0133] In the embodiments of the present application, in order to continuously optimize the inversion accuracy of the initial soil carbon storage inversion model through measured data, while ensuring that the evaluation task is efficiently completed within a reasonable range of resource consumption, the measured soil organic carbon data in the supplementary soil carbon storage data is combined with the corresponding multi-source satellite remote sensing data to correct the model and update the evaluation result, and a task closed loop is realized through loop iteration and threshold judgment to output a dynamic evaluation result of the soil carbon storage of southern grassland that is accurate and meets actual needs.

[0134] Specifically, the measured soil organic carbon data in the supplementary soil carbon storage data is first combined with the corresponding multi-source satellite remote sensing data to form a new training sample, and the initial soil carbon storage inversion model is incrementally trained.

[0135] The measured soil organic carbon data in the supplementary soil carbon storage data directly reflects the actual carbon storage situation of the target region, and the corresponding multi-source satellite remote sensing data contains the surface feature information of the measured point, and the new training sample formed by combining the two can specifically make up for the limitation of the initial soil carbon storage inversion model which only relies on historical sample grassland data.

[0136] In addition, incremental training does not require reconstruction of the model framework, but only needs to adjust the internal parameters of the model based on the new sample, which can not only retain the learning achievements of the initial soil carbon storage inversion model, but also enable the model to quickly adapt to the carbon storage distribution characteristics of the target region, and finally obtain a corrected soil carbon storage inversion model, further strengthening the inversion ability of the model for the target region.

[0137] Further, after obtaining the corrected soil carbon storage inversion model, the multi-source satellite remote sensing data of the target region is reprocessed by using the model to generate a corrected soil carbon storage distribution map and a corrected certainty heat map.

[0138] Specifically, compared with the initial distribution map, the revised soil carbon storage distribution map can more accurately present the spatial differences of the soil carbon storage in the target region. For example, the region originally predicted by the initial soil carbon storage inversion model as "10-12 g / kg" may be more consistent with the actual "11-13 g / kg" after revision. The revised certainty heat map updates the certainty of each pixel. After the model is calibrated by the measured data, the prediction accuracy of some regions originally with low certainty is improved, and the certainty value is correspondingly increased. The regions still with prediction deviation maintain low certainty, providing accuracy reference for subsequent evaluation.

[0139] Further, the revised certainty heat map is evaluated to determine whether the dynamic evaluation task is completed. Two core conditions need to be focused on during evaluation. First, whether the certainty of the entire target region reaches the preset certainty threshold. If the preset certainty threshold is reached, it means that the model's prediction of the carbon storage of each region in the target region has met the accuracy requirement, and no additional sampling is needed. Second, whether the task resource consumption has reached the preset upper limit.

[0140] The task resource consumption includes total sampling time consumption and total sampling energy consumption. If the resource consumption has reached the upper limit, even if the certainty of some regions does not meet the standard, the task is determined to be completed based on the trade-off between cost and efficiency to avoid wasting resources due to unlimited sampling.

[0141] On the contrary, if the revised certainty heat map shows that the global certainty does not reach the preset certainty threshold, and the task resource consumption does not exceed the upper limit, the steps of "supplementary sampling-revised soil carbon storage inversion model-revised certainty heat map-revised certainty heat map evaluation" need to be executed in a loop. That is, the supplementary sampling link is returned again, and the supplementary sampling points are configured again for the uncertain regions in the revised certainty heat map that do not meet the threshold. New supplementary soil carbon storage data is obtained by the detection robot, and then the model revision and result update process are repeated with the new data until the task completion condition is met.

[0142] Finally, when the task is determined to be completed, the revised soil carbon storage distribution map as the final output is taken as the dynamic evaluation result of the soil carbon storage. This result integrates the wide coverage advantage of multi-source satellite remote sensing data and the precise calibration value of the detection robot measured data, which can clearly reflect the overall distribution trend of the soil carbon storage in the target region and accurately present the carbon storage details in the local region. It can be used as reliable data support for southern grassland carbon sink research, ecological protection planning, and other work.

[0143] Through the specific implementation manner described above, the embodiments of the present application achieve the following technical effects:

[0144] This application proposes a dynamic assessment method for soil carbon storage in southern grasslands based on integrated remote sensing monitoring. First, a topographic map incorporating terrain features is drawn based on multi-source satellite remote sensing data of the target area. Simultaneously, a pre-trained initial soil carbon storage inversion model is retrieved, and the remote sensing data is input into the model to output an initial soil carbon storage distribution map and a deterministic heat map. Next, uncertain regions are extracted from the deterministic heat map according to a preset certainty threshold. A spatial random sampling algorithm is used to generate initial sampling points in these uncertain regions. Points inaccessible to the exploration robots are filtered using the topographic map, and supplementary sampling points are identified and marked. Finally, the parameter information of all exploration robots is acquired. A digital twin model is constructed by combining topographic maps to simulate the reach of the exploration robot and the sampling time and energy consumption. Supplementary sampling points and the robot's initial position are imported to configure the optimal sampling scheme. The robot is controlled to carry a contact near-infrared spectral probe for non-destructive sampling to obtain supplementary soil carbon storage data. Finally, the supplementary data is transmitted back, and the measured organic carbon data and corresponding remote sensing data are used to form new training samples. The initial model is incrementally trained to obtain a corrected model. The remote sensing data is reprocessed to generate a corrected soil carbon storage distribution map and a deterministic heat map. The deterministic nature or resource consumption of the whole area is evaluated. If the criteria are met, the results are output. Otherwise, supplementary sampling and model correction are carried out in a loop.

[0145] The method provided in this application, through the technical solution of "initial inversion and deterministic labeling - precise sampling of uncertain areas - digital twin optimized sampling - non-destructive data acquisition - model iterative correction", solves the problems of poor environmental adaptability of UAVs, lack of targeted sampling, and low model correction efficiency in traditional soil carbon storage assessment. It realizes the accuracy, efficiency and ecological nature of soil carbon storage assessment in southern grasslands, reduces invalid sampling and resource waste, and provides technical support for carbon sink research, ecological protection and dynamic monitoring of carbon storage in southern grasslands.

[0146] Example 2, as shown in the appendix Figure 2 As shown, based on the inventive concept of the method for dynamic assessment of soil carbon storage in southern grasslands using integrated remote sensing monitoring provided in Embodiment 1, this application also provides a system for dynamic assessment of soil carbon storage in southern grasslands using integrated remote sensing monitoring, specifically including:

[0147] The initial inversion mapping module 01 is used to draw a topographic map of the target area based on multi-source satellite remote sensing data of the target area, and input the multi-source satellite remote sensing data of the target area into the initial soil carbon storage inversion model, and output the initial soil carbon storage distribution map and deterministic heat map of the target area, wherein the deterministic heat map is composed of multiple pixels labeled with determinism;

[0148] The uncertain area sampling configuration module 02 is configured to extract an uncertain area according to a preset degree of certainty threshold based on the initial soil carbon storage distribution map and the deterministic thermal map, and configure a plurality of supplementary sampling points for the uncertain area according to a preset division rule, and mark the plurality of supplementary sampling points on the topographic map;

[0149] The digital twin sampling module 03 is configured to obtain parameter information of all detection robots in the target area, generate a digital twin model based on the parameter information of each detection robot and the topographic map of the target area, configure a supplementary sampling scheme based on the digital twin model, and non-destructively sample the plurality of supplementary sampling points to obtain a plurality of supplementary soil carbon storage data.

[0150] The model correction determination module 04 is configured to return the plurality of supplementary soil carbon storage data to the initial soil carbon storage inversion model to generate a corrected soil carbon storage distribution map and a corrected deterministic thermal map of the target area, and determine that the dynamic evaluation task is completed when the determinism of the entire target area reaches the degree of certainty threshold or reaches a predetermined upper limit of resource consumption, otherwise, continue to perform supplementary sampling and correct the initial soil carbon storage inversion model and the initial soil carbon storage distribution map.

[0151] In one embodiment, the initial inversion mapping module 01 is further configured to:

[0152] generate a topographic map containing topographic features of the target area according to multi-source satellite remote sensing data of the target area; call a pre-trained initial soil carbon storage inversion model, input the multi-source satellite remote sensing data of the target area, and output an initial soil carbon storage distribution map and a deterministic thermal map of the target area.

[0153] Further, the initial inversion mapping module 01 further comprises:

[0154] a plurality of sample grasslands similar to the target area in terms of water and heat conditions and vegetation types are selected as a plurality of sample grasslands; historical multi-source satellite remote sensing data of the plurality of sample grasslands and historical measured soil organic carbon data corresponding to the historical multi-source satellite remote sensing data are obtained, the historical measured soil organic carbon data at least including sampling depth and soil organic carbon content; the initial soil carbon storage inversion model is built based on machine learning; and the initial soil carbon storage inversion model is trained using the historical multi-source satellite remote sensing data and the historical measured soil organic carbon data until convergence.

[0155] In one embodiment, the uncertain area sampling configuration module 02 is further configured to:

[0156] a preset certainty threshold; dividing a pixel region in the certainty heat map that does not satisfy the certainty threshold into an uncertain region; using a spatial random sampling algorithm to generate initial sampling points of a preset density in the uncertain region; calling the topographic map and filtering out sampling points in the initial sampling points that are located in impassable terrain for the detection robots, and the remaining points being the supplementary sampling points.

[0157] In one embodiment, the digital twin sampling module 03 is further configured to:

[0158] obtain parameter information of all detection robots in the target region, wherein the parameter information at least includes initial positions, battery capacities, motion speeds, obstacle crossing performances, and core sensor states; create a digital twin model based on the parameter information of all detection robots and the topographic map, wherein the digital twin model can be used to simulate reachable ranges of each detection robot and sampling time consumption and sampling energy consumption of each detection robot from the initial position to each supplementary sampling point; import the supplementary sampling points at the labeled positions in the topographic map and the initial positions of all detection robots into the digital twin model, simulate sampling based on the digital twin model, obtain an optimal supplementary sampling scheme, and non-destructively sample the supplementary sampling points to obtain supplementary soil carbon storage data.

[0159] Further, the digital twin sampling module 03 further includes:

[0160] For each supplementary sampling point, the terrain features in the topographic map and the parameter information of the detection robots are combined to calculate a movement cost from a current position of any detection robot to the supplementary sampling point, wherein the movement cost is a weighted function of the sampling time consumption and the sampling energy consumption; the total sampling time consumption and the total sampling energy consumption are taken as optimization objectives to collaboratively assign exclusive supplementary sampling point sequences to all detection robots; for the supplementary sampling point sequence assigned to each detection robot, the optimal movement path is planned in combination with the topographic map; and the detection robots are controlled to sample according to the optimal movement path, and when encountering impassable obstacles, the supplementary sampling point sequence and the movement path are dynamically adjusted to obtain the optimal supplementary sampling scheme.

[0161] Further, the digital twin sampling module 03 further includes:

[0162] The detection robots are equipped with contact-type near-infrared spectroscopy probes, and the supplementary soil carbon storage data of the soil can be obtained without drilling.

[0163] In one embodiment, the model correction determination module 04 is further configured to:

[0164] The measured soil organic carbon data in the supplementary soil carbon storage data is combined with corresponding multi-source satellite remote sensing data to form a new training sample, the initial soil carbon storage inversion model is incrementally trained to obtain a corrected soil carbon storage inversion model; the multi-source satellite remote sensing data of the target region is reprocessed using the corrected soil carbon storage inversion model to generate a corrected soil carbon storage distribution map and a corrected certainty heat map; the corrected certainty heat map is evaluated, if the global certainty reaches a certainty threshold, or the task resource consumption has reached a preset upper limit, it is judged that the task is completed, and the final corrected soil carbon storage distribution map is output, wherein the task resource consumption includes total sampling time consumption and total sampling energy consumption; otherwise, the steps of supplementing sampling, correcting the soil carbon storage inversion model, obtaining the corrected certainty heat map, and evaluating the corrected certainty heat map are cyclically executed until the task is completed; the final output corrected soil carbon storage distribution map is taken as the dynamic evaluation result of the soil carbon storage.

[0165] It should be noted that the above-mentioned sequence of the embodiments of the present application is only for description, and does not represent the advantages and disadvantages of the embodiments. Moreover, the above-mentioned describes the specific embodiments of the present application. In addition, the processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multi-task processing and parallel processing are also possible or can be advantageous.

[0166] The above only describes the preferred embodiments of the present application, and does not limit the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

[0167] The specification and drawings are merely exemplary of the present application, and any and all modifications, variations, combinations or equivalents that fall within the scope of the present application are considered to be covered by the present application. Obviously, those skilled in the art can make various modifications and variations to the present application without departing from the scope of the present application. Thus, if these modifications and variations of the present application belong to the scope of the present application and its equivalent technology, the present application is intended to include these modifications and variations.

Claims

1. A method for dynamic evaluation of soil carbon storage in southern grassland by fusion remote sensing monitoring, characterized in that, The method comprises the following steps: Based on the multi-source satellite remote sensing data of the target area, a topographic map of the target area is drawn, and the multi-source satellite remote sensing data of the target area is input into an initial soil carbon storage inversion model to output an initial soil carbon storage distribution map and a certainty heat map of the target area, wherein the certainty heat map is composed of a plurality of pixels marked with certainty; Based on the initial soil carbon storage distribution map and the certainty heat map, an uncertain area is extracted according to a preset certainty threshold, and a plurality of supplementary sampling points are configured for the uncertain area according to a preset division rule, and a plurality of supplementary sampling points are marked on the topographic map; Obtain the parameter information of all detection robots in the target area, generate a digital twin model based on the parameter information of each detection robot and the topographic map of the target area, configure a supplementary sampling scheme based on the digital twin model, and non-destructively sample a plurality of supplementary sampling points to obtain a plurality of supplementary soil carbon storage data; The plurality of supplementary soil carbon storage data is returned to the initial soil carbon storage inversion model to generate a corrected soil carbon storage distribution map and a corrected certainty heat map of the target area, and when the certainty of the entire target area reaches the certainty threshold or reaches the predetermined upper limit of resource consumption, the dynamic evaluation task is determined to be completed, otherwise, supplementary sampling is continued, and the initial soil carbon storage inversion model and the initial soil carbon storage distribution map are corrected.

2. The method according to claim 1, wherein, Based on the multi-source satellite remote sensing data of the target area, a topographic map of the target area is drawn, and the multi-source satellite remote sensing data of the target area is input into an initial soil carbon storage inversion model to output an initial soil carbon storage distribution map and a certainty heat map of the target area, comprising: Generating a topographic map containing topographic features of the target area according to the multi-source satellite remote sensing data of the target area; Accessing a pre-trained initial soil carbon storage inversion model, inputting the multi-source satellite remote sensing data of the target area, and outputting an initial soil carbon storage distribution map and a certainty heat map of the target area.

3. The method according to claim 2, wherein, The method for obtaining the initial soil carbon storage inversion model comprises: Selecting a plurality of southern grasslands similar to the water and heat conditions and vegetation types of the target area as a plurality of sample grasslands; Obtaining historical multi-source satellite remote sensing data of the plurality of sample grasslands, and historical measured soil organic carbon data corresponding to the historical multi-source satellite remote sensing data, wherein the historical measured soil organic carbon data at least includes sampling depth and soil organic carbon content; Building the initial soil carbon storage inversion model based on machine learning; Training the initial soil carbon storage inversion model using the historical multi-source satellite remote sensing data and the historical measured soil organic carbon data until convergence.

4. The method according to claim 1, wherein, Based on the initial soil carbon storage distribution map and the certainty heat map, an uncertain area is extracted according to a preset certainty threshold, and a plurality of supplementary sampling points are configured for the uncertain area according to a preset division rule, and a plurality of supplementary sampling points are marked on the topographic map, comprising: A preset certainty threshold is set; Divide the pixel area in the certainty heat map that does not meet the certainty threshold into an uncertain area; An initial sampling point site of a preset density is generated in the uncertain area by using a spatial random sampling algorithm; Access the topographic map, filter out the initial sampling points located in the terrain inaccessible to the detection robot, and the remaining points are the supplementary sampling points.

5. The method of claim 1, wherein Obtain parameter information of all detection robots in the target area, generate a digital twin model based on the parameter information of each detection robot and the topographic map of the target area, configure a supplementary sampling scheme based on the digital twin model, and perform non-destructive sampling on the supplementary sampling points to obtain supplementary soil carbon storage data, including: Obtain parameter information of all detection robots in the target area, wherein the parameter information at least includes initial position, battery capacity, movement speed, obstacle crossing performance, and core sensor state; Based on the parameter information of all detection robots and the topographic map, create a digital twin model, which can be used to simulate the reachable range of each detection robot, the sampling time and sampling energy consumption from the initial position to each supplementary sampling point for sampling; Import the labeled positions of the supplementary sampling points on the topographic map and the initial positions of all detection robots into the digital twin model, simulate sampling based on the digital twin model, obtain the best supplementary sampling scheme, and perform non-destructive sampling on the supplementary sampling points to obtain supplementary soil carbon storage data.

6. The method of claim 5, wherein the method further comprises: Import the labeled positions of the supplementary sampling points on the topographic map and the initial positions of all detection robots into the digital twin model, simulate sampling based on the digital twin model, obtain the best supplementary sampling scheme, and perform non-destructive sampling on the supplementary sampling points to obtain supplementary soil carbon storage data, including: For each supplementary sampling point, calculate the movement cost from the current position of any detection robot to the supplementary sampling point based on the terrain features in the topographic map and the parameter information of the detection robot, wherein the movement cost is a weighted function of the total sampling time and the total sampling energy consumption; Distribute exclusive supplementary sampling points to all detection robots in sequence as the optimization target of the total sampling time and the total sampling energy consumption; For the supplementary sampling point sequence allocated to each detection robot, plan its optimal movement path based on the topographic map; Control each detection robot to sample according to the optimal movement path, and dynamically adjust the supplementary sampling point sequence and the movement path when encountering a passage obstacle to obtain the best supplementary sampling scheme.

7. The method according to claim 5, wherein, The detection robot is equipped with a contact near-infrared spectroscopy probe, which can obtain supplementary soil carbon storage data without drilling.

8. The method of claim 1, wherein The supplementary soil carbon storage data is returned to the initial soil carbon storage inversion model to generate a corrected soil carbon storage distribution map and a corrected certainty heat map of the target area. When the certainty of the entire target area reaches the certainty threshold or reaches the predetermined upper limit of resource consumption, the dynamic evaluation task is completed. Otherwise, continue to perform supplementary sampling to correct the initial soil carbon storage inversion model and the initial soil carbon storage distribution map, including: The measured soil organic carbon data in the supplementary soil carbon storage data and the corresponding multi-source satellite remote sensing data form a new training sample, and the initial soil carbon storage inversion model is incrementally trained to obtain a corrected soil carbon storage inversion model. Reprocess the multi-source satellite remote sensing data of the target region by using the corrected soil carbon storage inversion model to generate a corrected soil carbon storage distribution map and a corrected certainty thermal map; Evaluate the corrected certainty thermal map. If the global certainty reaches the certainty threshold or the task resource consumption has reached the preset upper limit, determine that the task is complete, and output the final corrected soil carbon storage distribution map, wherein the task resource consumption includes total sampling time consumption and total sampling energy consumption; Otherwise, cyclically execute the steps of supplementary sampling, correcting the soil carbon storage inversion model, obtaining the corrected certainty thermal map, and evaluating the corrected certainty thermal map until the task is completed. The final output corrected soil carbon storage distribution map is used as the dynamic evaluation result of the soil carbon storage.

9. A system for dynamic assessment of soil carbon stock in southern grassland by fusion of remote sensing monitoring, characterized in that, The system is used to perform the dynamic evaluation method of the southern grassland soil carbon storage based on fusion remote sensing monitoring according to any one of claims 1-8, and the system comprises: An initial inversion mapping module is configured to draw a topographic map of a target region based on multi-source satellite remote sensing data of the target region, input the multi-source satellite remote sensing data of the target region into an initial soil carbon storage inversion model, and output an initial soil carbon storage distribution map and a certainty thermal map of the target region, wherein the certainty thermal map is composed of a plurality of pixels marked with certainty. An uncertain region sampling configuration module is configured to extract an uncertain region according to a preset certainty threshold based on the initial soil carbon storage distribution map and the certainty thermal map, configure a plurality of supplementary sampling points for the uncertain region according to a preset division rule, and mark the plurality of supplementary sampling points on the topographic map. A digital twin sampling module is configured to obtain parameter information of all detection robots in the target region, generate a digital twin model based on the parameter information of each detection robot and the topographic map of the target region, configure a supplementary sampling scheme based on the digital twin model, and non-destructively sample the plurality of supplementary sampling points to obtain a plurality of supplementary soil carbon storage data. A model correction determination module is configured to return the plurality of supplementary soil carbon storage data to the initial soil carbon storage inversion model to generate a corrected soil carbon storage distribution map and a corrected certainty thermal map of the target region. When the certainty of the entire target region reaches the certainty threshold or reaches the predetermined resource consumption upper limit, it is determined that the dynamic evaluation task is completed. Otherwise, supplementary sampling is continued, and the initial soil carbon storage inversion model and the initial soil carbon storage distribution map are corrected.

Images