A method for evaluating carbon storage of mountain mire wetland based on remote sensing and ground investigation

Abstract

The application provides a mountainous marsh wetland carbon storage evaluation method based on remote sensing and ground investigation, comprising the following steps: obtaining remote sensing data, terrain and hydrological basic data, climate data, human activity and disturbance data and ground investigation data to obtain a multi-source original data set; constructing a global hydrological state grid sequence, a disturbance event library and a terrain partition based on a feature tensor to obtain a space-time reference data package; performing field sampling and measurement according to peat thickness prediction and thickness variance prediction to obtain field measurement data; calculating and performing multi-task learning on a carbon pool set to obtain a multi-carbon pool joint prediction model, and outputting predicted total carbon storage; and constructing an explainability analysis equation to perform explainability analysis on carbon storage changes. The application can realize global, quantifiable uncertainty and explainable carbon storage evaluation in mountainous marsh wetlands.

Description

Technical Field

[0001] This invention relates to the field of carbon storage assessment technology, and in particular to a method for assessing carbon storage in mountainous marsh wetlands based on remote sensing and ground surveys. Background Technology

[0002] Carbon storage assessment of mountainous marsh wetlands refers to the calculation and spatial mapping of carbon pools (carbon storage) in wetland areas characterized by long-term water accumulation, high water content, and marshland soils and vegetation within mountainous environments. It aims to provide, as far as possible, information on trends and uncertainties. Carbon storage includes at least vegetation carbon, dead matter carbon, soil carbon, and peat carbon. Wetlands are high-density and high-risk carbon pools; once drainage, degradation, fire, or warming drought occurs, they may transform from carbon sinks (absorbing carbon dioxide) into carbon sources (emitting carbon dioxide and methane). Carbon storage assessment of mountainous marsh wetlands provides quantifiable indicators for wetland protection, degraded wetland restoration, and ecological red line management, supporting carbon accounting, ecological compensation, and wetland restoration projects.

[0003] Existing methods for assessing carbon storage in mountainous marsh wetlands include traditional laboratory measurements, rapid regional estimation, and machine learning mapping. However, these methods suffer from the following drawbacks: 1) Difficulty in characterizing wetland hydrological processes: Many methods approximate wetland conditions using one-time images or a few indices, failing to reflect the temporal changes in groundwater levels, leading to large spatial errors in carbon storage. 2) Strong influence of mountainous terrain, resulting in systemic bias in cross-topographical modeling: Empirical models from plain wetlands are directly transferred to mountainous areas, severely affected by slope, shadows, and runoff, resulting in poor model generalization. 3) Only point estimates are provided, without credible uncertainty, and the source of error cannot be located: The results are difficult to use for carbon accounting and risk decision-making, and cannot guide subsequent data supplementation or sampling optimization. Summary of the Invention

[0004] To overcome the shortcomings of existing technologies, the purpose of this invention is to provide a method for assessing carbon storage in mountainous marsh wetlands based on remote sensing and ground surveys. In mountainous marsh wetlands, a spatiotemporal benchmark is constructed using multi-source data, and active learning sampling, joint prediction of multiple carbon pools, hierarchical Bayesian scale summarization and uncertainty decomposition are combined to provide causal paths, thereby achieving a comprehensive, quantifiable, and interpretable carbon storage assessment.

[0005] To achieve the above objectives, the present invention provides the following solution: a method for assessing carbon storage in mountainous marsh wetlands based on remote sensing and ground surveys, comprising: Acquire remote sensing data, topographic and hydrological data, climate data, human activity and disturbance data, and ground survey data to obtain multi-source raw datasets; Based on the multi-source raw dataset, a feature tensor is constructed, and then based on the feature tensor, a global hydrological state raster sequence, a disturbance event database, and a terrain partition are constructed to obtain a spatiotemporal reference data package. Based on the spatiotemporal reference data package, peat thickness and thickness variance are predicted. Based on the peat thickness prediction and thickness variance prediction, field sampling and measurement are performed to obtain field measurement data. The field measurement data is then subjected to quality inspection and storage. Based on the field measurement data, a carbon pool set is calculated, and multi-task learning is performed on the carbon pool set to obtain a multi-carbon pool joint prediction model, which outputs the predicted total carbon storage. Based on the predicted total carbon storage and uncertainty analysis results, an interpretability analysis equation is constructed to output the causal path of carbon change in the predicted total carbon storage, thus completing the interpretability analysis of carbon storage change.

[0006] Optionally, remote sensing data, topographic and hydrological baseline data, climate data, human activity and disturbance data, and ground survey data are acquired to obtain a multi-source raw dataset, including: Optical remote sensing data, synthetic aperture radar data, lidar data, and thermal infrared data of the study area were acquired to obtain remote sensing data; Obtain DEM data, elevation, slope, aspect, curvature, catchment area, topographic humidity index, distance from river network, and mountain shadow-related factors of the study area to obtain basic topographic and hydrological data; Acquire precipitation, temperature, radiation, and water vapor pressure deficit data for the study area to obtain climate data, and perform test drive sequence alignment operation on the climate data and the remote sensing data; Data on roads, settlements, tourist facilities, mining sites, nighttime lighting, grazing intensity, drainage ditches, fires, and restoration projects in the study area were obtained to acquire data on human activities and disturbances. Vegetation carbon surveys, soil carbon surveys, and hydrological observations were also conducted to obtain ground survey data. The remote sensing data, the topographic and hydrological basic data, the climate data, the human activity and disturbance data, and the ground survey data are subjected to coordinate system and projection unification, metadata table construction, and ground sample point quality inspection to obtain a multi-source raw dataset.

[0007] Optionally, a feature tensor is constructed based on the multi-source original dataset, and then a global hydrological state raster sequence, a perturbation event database, and a topographic partition are constructed based on the feature tensor to obtain a spatiotemporal reference data package, including: Topographic correction and spatiotemporal alignment are performed on the multi-source raw dataset to obtain a feature tensor. Based on the feature tensor, the hydrological state form of the cell is defined. Based on the hydrological state form, candidate hydrological indicators are constructed to establish a mapping relationship between the candidate hydrological indicators and the observed water level points, and an inversion function is obtained. The inversion function is then used to output a global hydrological state raster sequence. Based on the feature tensor, event types are defined, and abrupt changes and persistent changes are extracted from the time series according to the event types. The abrupt changes and persistent changes are attributed to event categories to obtain a structured perturbation event library. Then, terrain zoning is performed according to landform and wetland types. The feature tensor, the global hydrological state raster sequence, the disturbance event library, and the terrain partition are integrated into a spatiotemporal reference data package.

[0008] Optionally, topographic correction and spatiotemporal alignment are performed on the multi-source original dataset to obtain a feature tensor. Based on the feature tensor, the hydrological state form of the pixel is defined, including: Define a unified spatial grid and time axis, and based on the spatial grid and time axis, perform cloud and fog removal, topographic radiation correction and multi-temporal synthesis on the optical data in the multi-source original dataset to obtain an optical synthesis feature set; The SAR data in the multi-source raw dataset are subjected to orbit correction, radiometric calibration, terrain correction, speckle noise suppression, and time series synthesis to obtain a SAR synthetic feature set, thus completing terrain correction and spatiotemporal alignment. The multi-source original datasets that have undergone terrain correction and spatiotemporal alignment are combined into a learnable data structure to obtain a feature tensor. Based on the feature tensor, the hydrological state form of the pixel is defined.

[0009] Optionally, based on the spatiotemporal reference data package, peat thickness prediction and thickness variance prediction are performed. Based on the peat thickness prediction and thickness variance prediction, on-site sampling and measurement are conducted to obtain on-site measurement data. The on-site measurement data is then subjected to quality inspection and data storage operations, including: Based on the defined core target variable for sampling and the actual constraints, partition regression and uncertainty estimation are performed for each terrain zone to obtain the prior model and output peat thickness prediction and thickness variance prediction. Active learning is performed using the peat thickness prediction and the thickness variance prediction to output the comprehensive pixel sampling revenue. Based on the comprehensive pixel sampling revenue, a sampling execution list for each terrain zone is generated. According to the sampling execution list, probes or core samples are used to probe downwards from the surface until the peat and mineral interface is reached. The total thickness, thickness of each layer, degree of decomposition, water content, color and plant residue characteristics are recorded to obtain peat thickness field sampling information. The thickness, bulk density, organic carbon content and gravel volume fraction of deep SOC are also recorded to obtain deep SOC field sampling information. The field measurement data are obtained by combining the peat thickness field sampling information and the deep SOC field sampling information. The on-site measurement data is associated with the pixel ID and stored in the database. Then, abnormal density checks, carbon content out-of-range checks, layer thickness missing checks, and coordinate checks are performed to complete the data quality inspection operation.

[0010] Optionally, the core target variables include key variables and auxiliary variables. The key variables include peat thickness and deep soil organic carbon, and the auxiliary variables include bulk density, organic carbon content and groundwater level. The actual constraints include safety constraints, cost constraints, ecological constraints and time constraints.

[0011] Optionally, based on the field measurement data, a carbon pool set is calculated, and multi-task learning is performed on the carbon pool set to obtain a multi-carbon pool joint prediction model, outputting a predicted total carbon storage, including: The aboveground biomass carbon pool, the underground biomass carbon pool, the shallow soil organic carbon pool, the deep soil organic carbon pool, and the peat carbon pool are defined. Based on the field measurement data, the pixel area of ​​each carbon pool is converted and the output unit is unified to obtain a set of carbon pools. The spatiotemporal reference data package is set as the input feature, and the carbon pool set is subjected to multi-task learning using the input feature to obtain a multi-carbon pool joint prediction model. Multi-scale hierarchical Bayes is introduced into the multi-carbon pool joint prediction model to output the predicted total carbon storage and uncertainty analysis results with consistent scale.

[0012] Optionally, the spatiotemporal reference data package is set as the input feature, and the input feature is used to perform multi-task learning on the carbon pool set to obtain a multi-carbon pool joint prediction model. A multi-scale hierarchical Bayesian approach is introduced into the multi-carbon pool joint prediction model to output scale-consistent predicted total carbon storage and uncertainty analysis results, including: The spatiotemporal reference data package is set as the input feature, and the carbon library set is used to perform multi-task learning to obtain an initial prediction model and output the pixel prediction result. Multi-scale hierarchical Bayes is introduced into the multi-carbon pool joint prediction model to sum the pixel prediction results within the plot or watershed unit to obtain the pixel summary value. Based on the pixel aggregated value, the peat storage per unit area of ​​the pixel, the predicted value of the peat bulk density of the pixel, the predicted value of the carbon mass fraction of the peat of the pixel, and the predicted value of the peat thickness of the pixel are calculated to output the pixel carbon library. The pixel carbon library is added up by unit area to obtain the total carbon content per unit area of ​​the pixel. Based on the total carbon content of the pixel area, the mean map and standard deviation map of each pixel are output to obtain the pixel uncertainty. The pixel uncertainty is then decomposed according to four sources: missing remote sensing observations, sparse sampling points, peat thickness error, and scale constraint error, to obtain an error decomposition map.

[0013] Optionally, based on the predicted total carbon storage and uncertainty analysis results, an interpretability analysis equation is constructed to output the causal path of carbon change in the predicted total carbon storage, thus completing the interpretability analysis of carbon storage change, including: Based on the predicted total carbon storage and the results of uncertainty analysis, response variables, mediating variables, external driving variables and control variables are defined to obtain the analysis framework; Based on the aforementioned analysis framework, for each pixel, a variable with the same time window is generated, and pixel fitting weights are defined according to the error decomposition map; the calculation expression for the pixel interpretable weights is as follows: ; in, For pixels Weights in interpretability analysis For pixels The variance of carbon variation To prevent extremely small positive numbers with a denominator of 0; Based on the analytical framework and the pixel interpretable weights, hydrological equations and carbon change equations are defined to obtain interpretable analytical equations. Using the interpretable analytical equations, the SEM path map and path coefficients of the predicted total carbon storage are output to obtain the causal path of carbon change.

[0014] Optionally, the response variable consists of pixel carbon change, the mediating variable consists of hydrological state, the external driving variable consists of climate pressure variable and anthropogenic disturbance variable, and the control variable consists of topographic zoning, soil type and initial carbon storage.

[0015] This invention discloses the following technical effects by providing a method for assessing carbon storage in mountainous marsh wetlands based on remote sensing and ground surveys: Comprehensive coverage: By combining multi-source remote sensing, topography and hydrology, climate, disturbance and ground surveys, pixel-level full coverage carbon storage mapping of mountainous marsh wetlands is achieved.

[0016] 1. Stronger process constraints: The hydrological state raster sequence and disturbance event library are explicitly incorporated, making carbon storage estimation closer to the mechanism of wetland carbon formation or loss.

[0017] 2. Optimization of accuracy and cost: Active learning sampling is driven by peat thickness variance to maximize sampling benefits under constraints.

[0018] 3. Consistency across multiple carbon libraries: Multi-task learning jointly predicts five types of carbon libraries, reducing the caliber difference and spatial inconsistency caused by each carbon library being calculated separately.

[0019] 4. Scale consistency and computability: Multi-scale hierarchical Bayesian methods ensure statistical consistency from pixel to plot and watershed aggregation.

[0020] 5. Reliable Uncertainty Closed Loop: Output standard deviation map and perform source decomposition to form a closed loop of error and improvement measures.

[0021] 6. Explainable and decision-making: Based on uncertainty-weighted SEM causal paths, it supports the direction of carbon governance based on which type of disturbance and through which hydrological pathways it affects carbon.

[0022] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0023] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0024] Figure 1 This is a schematic diagram of the method flow provided in an embodiment of the present invention; Figure 2 A flowchart for predicting carbon storage in mountainous marsh wetlands provided in an embodiment of the present invention; Figure 3 A flowchart for interpretability analysis of carbon storage in mountainous marsh wetlands provided in an embodiment of the present invention. Detailed Implementation

[0025] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0026] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0027] like Figure 1 As shown, this invention provides a method for assessing carbon storage in mountainous marsh wetlands based on remote sensing and ground surveys, including: Step 1, as follows Figure 2 As shown, remote sensing data, topographic and hydrological baseline data, climate data, human activity and disturbance data, and ground survey data are acquired to obtain a multi-source raw dataset; specifically including: 1.1 Obtain optical remote sensing data, synthetic aperture radar data, lidar data, and thermal infrared data of the study area to obtain remote sensing data.

[0028] 1.2 Obtain DEM data, elevation, slope, aspect, curvature, catchment area, topographic humidity index, distance from river network, and mountain shadow related factors of the study area to obtain basic topographic and hydrological data; 1.3 Acquire precipitation, temperature, radiation, and water vapor pressure deficit data of the study area to obtain climate data, and perform test drive sequence alignment operation on the climate data and the remote sensing data.

[0029] 1.4 Data on roads, settlements, tourist facilities, mining sites, nighttime lighting, grazing intensity, drainage ditches, fires, and restoration projects in the study area were obtained to acquire data on human activities and disturbances. Vegetation carbon surveys, soil carbon surveys, and hydrological observations were also conducted to obtain ground survey data.

[0030] 1.5 The remote sensing data, the topographic and hydrological basic data, the climate data, the human activity and disturbance data, and the ground survey data are subjected to coordinate system and projection unification, metadata table construction, and ground sample point quality inspection to obtain a multi-source raw dataset.

[0031] Step 2, as follows Figure 2 As shown, a feature tensor is constructed based on the multi-source original dataset, and then a global hydrological state raster sequence, a perturbation event database, and a topographic partition are constructed based on the feature tensor to obtain a spatiotemporal reference data package; specifically including: 2.1 Perform terrain correction and spatiotemporal alignment on the multi-source original dataset to obtain a feature tensor. Based on the feature tensor, define the hydrological state form of the pixel; specifically including: 2.1.1 Define a unified spatial grid and time axis. Based on the spatial grid and time axis, perform cloud and fog removal, terrain radiometric correction and multi-temporal synthesis on the optical data in the multi-source original dataset to obtain an optical synthesis feature set.

[0032] Cloud removal: Use the image's built-in QA (Quality Assessment) band or cloud probability product; Generate cloud mask: Mark cloud, cloud shadow, and snow pixels as invalid values.

[0033] Topographic radiation correction: In mountainous areas, the reflectivity of the same feature differs on shady and sunny slopes, which can misjudge "slope aspect differences" as "wetland differences." Based on the DEM, slope, slope aspect, and solar incidence angle are calculated, topographic correction is performed, and the topographically corrected reflectivity is output to minimize illumination differences and make features more comparable.

[0034] Multi-temporal composite: For each time window (e.g., monthly), all eligible images are composited to produce statistics such as median, reflectance (anti-outlier), P10 / P90 quantiles (to characterize fluctuations), and amplitude (to characterize the magnitude of seasonal variation).

[0035] 2.1.2 Perform orbit correction, radiometric calibration, terrain correction, speckle noise suppression, and time series synthesis on the SAR data in the multi-source raw dataset to obtain a SAR synthetic feature set, and complete terrain correction and spatiotemporal alignment.

[0036] Track calibration: Precision track files are used to correct positioning errors and ensure alignment with the optical and DEM.

[0037] Radiometric calibration: converting raw SAR values ​​into comparable backscattering measures; putting data from different dates and orbits onto the same physical scale.

[0038] Terrain correction: SAR will have overlays and shadows in mountainous areas; terrain correction is performed based on DEM, and overlay and shadow mask layers are generated at the same time: these areas should be reduced in weight or removed in the future.

[0039] Speckle noise suppression: SAR inherently has particle noise. Multi-phase filtering or local filtering is used to improve the stability of time series.

[0040] Time series synthesis: For each time window, calculate: vertical-vertical and vertical-horizontal; statistics: median, P10 / P90, amplitude.

[0041] 2.1.3 The multi-source original datasets that have completed terrain correction and spatiotemporal alignment are combined into a learnable data structure to obtain a feature tensor. Based on the feature tensor, the hydrological state form of the pixel is defined.

[0042] 2.2 Based on the hydrological state form, candidate hydrological indicators are constructed to establish a mapping relationship between the candidate hydrological indicators and the observed water level points, and an inversion function is obtained. The inversion function is then used to output a global hydrological state raster sequence.

[0043] Candidate hydrological indicators: Generated by SAR: flooding probability, water content sensitivity index, scattering change rate, etc.; Generated by topography: confluence potential, valley index, etc.; Climate-related factors: recent cumulative precipitation, number of drought days, etc.

[0044] 2.3 Based on the feature tensor, define event types, extract abrupt changes and persistent changes in the time series according to the event types, attribute the abrupt changes and persistent changes to event categories, obtain a structured perturbation event library, and then perform terrain zoning according to landform and wetland types; 2.4 Integrate the feature tensor, the global hydrological state raster sequence, the disturbance event library, and the terrain partition into a spatiotemporal reference data package.

[0045] Disturbance event database: event type system: drainage, rewetting, fire, grazing trampling, mining, or road impact. Topographic zoning: landform, elevation zone, wetland type, or vegetation type.

[0046] Step 3, as follows Figure 2 As shown, based on the spatiotemporal reference data package, peat thickness and thickness variance are predicted. Based on the predicted peat thickness and thickness variance, on-site sampling and measurement are performed to obtain on-site measurement data. The on-site measurement data is then subjected to quality inspection and data storage operations. Specifically, this includes: 3.1 Based on the defined core target variables and actual constraints, perform zonal regression and uncertainty estimation for each terrain zone to obtain the prior model and output peat thickness prediction and thickness variance prediction.

[0047] The core target variables include key variables and auxiliary variables. The key variables include peat thickness and deep soil organic carbon. The auxiliary variables include bulk density, organic carbon content and groundwater level. The actual constraints include safety constraints, cost constraints, ecological constraints and time constraints.

[0048] Safety restrictions: Areas that are off-limits due to maximum slope, falling rocks, snow line, etc. Cost constraints: arrival time, distance from road, elevation gain; Ecological constraints: Entry permits and disturbance restrictions in the core area of ​​the protected area; Time constraints: seasonal window, i.e. whether drilling is possible during the rainy season or the frozen soil period.

[0049] 3.2 Active learning is performed using the peat thickness prediction and the thickness variance prediction to output the comprehensive pixel sampling revenue. Based on the comprehensive pixel sampling revenue, a sampling execution list for each terrain zone is generated.

[0050] Output a sampling execution list for each selected pixel, for example: Zoning and wetland type prediction; Sampling types: probe, core sampler, soil drill, vegetation quadrat, water level recorder; Target depth: Is it necessary to reach the mineral layer? Layering schemes: 0-10, 10-30, 30-60, 60-100 cm, or dynamically adjusted according to peat thickness; On-site risk warnings: slope, water crossing, communication, etc.; Estimated costs: walking distance, elevation gain, and approval information.

[0051] 3.3 According to the sampling execution list, probes or core samples are used to probe downwards from the surface until the peat and mineral interface is reached. The total thickness, thickness of each layer, degree of decomposition, water content, color, and plant residue characteristics are recorded to obtain peat thickness field sampling information. The thickness, bulk density, organic carbon content, and gravel volume fraction of deep SOC are also recorded to obtain deep SOC field sampling information. The field measurement data are obtained by combining the peat thickness field sampling information and the deep SOC field sampling information.

[0052] 3.4 Associate the field measurement data with the pixel ID and store it in the database. Then perform density anomaly checks, carbon content out-of-range checks, layer thickness missing checks, and coordinate checks to complete the data quality inspection operation.

[0053] Step 4, as follows Figure 2 As shown, based on the field measurement data, a carbon pool set is calculated, and multi-task learning is performed on the carbon pool set to obtain a multi-carbon pool joint prediction model, which outputs the predicted total carbon storage; specifically including: 4.1 Define aboveground biomass carbon pool, underground biomass carbon pool, shallow soil organic carbon pool, deep soil organic carbon pool, and peat carbon pool. Based on the field measurement data, perform pixel area conversion and unify the output unit for each carbon pool to obtain a set of carbon pools.

[0054] 4.2 The spatiotemporal reference data package is set as the input feature. Multi-task learning is performed on the carbon pool set using the input feature to obtain a multi-carbon pool joint prediction model. Multi-scale hierarchical Bayesian methods are introduced into the multi-carbon pool joint prediction model to output scale-consistent predicted total carbon reserves and uncertainty analysis results. Specifically, this includes: 4.2.1 The spatiotemporal reference data package is set as the input feature, and the carbon library set is subjected to multi-task learning using the input feature to obtain an initial prediction model and output the pixel prediction result.

[0055] 4.2.2 In the multi-carbon pool joint prediction model, a multi-scale hierarchical Bayesian method is introduced to sum the pixel prediction results within the plot or watershed unit to obtain the pixel summary value.

[0056] 4.2.3 Based on the pixel aggregated value, calculate the peat storage per unit area of ​​the pixel, the predicted value of the peat bulk density of the pixel, the predicted value of the carbon mass fraction of the peat of the pixel, and the predicted value of the peat thickness of the pixel to output the pixel carbon library. Add the pixel carbon libraries by unit area to obtain the total carbon content per unit area of ​​the pixel.

[0057] 4.2.4 Based on the total carbon content of the pixel area, output the mean map and standard deviation map of each pixel to obtain the pixel uncertainty. According to the four sources of remote sensing observation missing, sample point sparsity, peat thickness error and scale constraint error, the pixel uncertainty is decomposed to obtain the error decomposition map.

[0058] Step 5, as follows Figure 3 As shown, based on the predicted total carbon storage and uncertainty analysis results, an interpretability analysis equation is constructed to output the causal path of carbon change in the predicted total carbon storage, thus completing the interpretability analysis of carbon storage change. Specifically, this includes: 5.1 Based on the predicted total carbon reserves and the uncertainty analysis results, response variables, mediating variables, external driving variables and control variables are defined to obtain the analysis framework.

[0059] The response variable consists of pixel carbon change, the mediating variable consists of hydrological state, the external driving variable consists of climate pressure variable and anthropogenic disturbance variable, and the control variable consists of topographic zoning, soil type and initial carbon storage.

[0060] 5.2 Based on the aforementioned analysis framework, for each pixel, a variable with the same time window is generated, and pixel fitting weights are defined according to the error decomposition diagram; the calculation expression for the pixel interpretable weights is as follows: ; in, For pixels Weights in interpretability analysis For pixels The variance of carbon variation To prevent extremely small positive numbers with a denominator of 0.

[0061] 5.3 Based on the analytical framework and the pixel interpretable weights, hydrological equations and carbon change equations are defined to obtain interpretable analytical equations. Using the interpretable analytical equations, the SEM path map and path coefficients of the predicted total carbon storage are output to obtain the causal path of carbon change.

[0062] Therefore, this invention provides a method for assessing carbon storage in mountainous marsh wetlands based on remote sensing and ground surveys. In mountainous marsh wetlands, a spatiotemporal benchmark is constructed using multi-source data. This method combines active learning sampling, joint prediction of multiple carbon pools, hierarchical Bayesian scaling and uncertainty decomposition, and provides causal paths, thereby achieving a comprehensive, quantifiable, and interpretable assessment of carbon storage.

[0063] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. The same or similar parts between the various embodiments can be referred to each other.

[0064] This document uses specific examples to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. Furthermore, those skilled in the art will recognize that, based on the ideas of the present invention, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A method for assessing carbon storage in mountainous marsh wetlands based on remote sensing and ground surveys, characterized in that, include: Acquire remote sensing data, topographic and hydrological data, climate data, human activity and disturbance data, and ground survey data to obtain multi-source raw datasets; Based on the multi-source raw dataset, a feature tensor is constructed, and then based on the feature tensor, a global hydrological state raster sequence, a disturbance event database, and a terrain partition are constructed to obtain a spatiotemporal reference data package. Based on the spatiotemporal reference data package, peat thickness and thickness variance are predicted. Based on the peat thickness prediction and thickness variance prediction, field sampling and measurement are performed to obtain field measurement data. The field measurement data is then subjected to quality inspection and storage. Based on the field measurement data, a carbon pool set is calculated, and multi-task learning is performed on the carbon pool set to obtain a multi-carbon pool joint prediction model, which outputs the predicted total carbon storage. Based on the predicted total carbon storage and uncertainty analysis results, an interpretability analysis equation is constructed to output the causal path of carbon change in the predicted total carbon storage, thus completing the interpretability analysis of carbon storage change.

2. The method for assessing carbon storage in mountainous marsh wetlands based on remote sensing and ground surveys according to claim 1, characterized in that, Acquire remote sensing data, topographic and hydrological baseline data, climate data, human activity and disturbance data, and ground survey data to obtain multi-source raw datasets, including: Optical remote sensing data, synthetic aperture radar data, lidar data, and thermal infrared data of the study area were acquired to obtain remote sensing data; Obtain DEM data, elevation, slope, aspect, curvature, catchment area, topographic humidity index, distance from river network, and mountain shadow-related factors of the study area to obtain basic topographic and hydrological data; Acquire precipitation, temperature, radiation, and water vapor pressure deficit data for the study area to obtain climate data, and perform test drive sequence alignment operation on the climate data and the remote sensing data; Data on roads, settlements, tourist facilities, mining sites, nighttime lighting, grazing intensity, drainage ditches, fires, and restoration projects in the study area were obtained to acquire data on human activities and disturbances. Vegetation carbon surveys, soil carbon surveys, and hydrological observations were also conducted to obtain ground survey data. The remote sensing data, the topographic and hydrological basic data, the climate data, the human activity and disturbance data, and the ground survey data are subjected to coordinate system and projection unification, metadata table construction, and ground sample point quality inspection to obtain a multi-source raw dataset.

3. The method for assessing carbon storage in mountainous marsh wetlands based on remote sensing and ground surveys according to claim 2, characterized in that, Based on the aforementioned multi-source raw dataset, a feature tensor is constructed. Then, based on the feature tensor, a global hydrological state raster sequence, a perturbation event database, and a topographic partition are constructed to obtain a spatiotemporal reference data package, including: Topographic correction and spatiotemporal alignment are performed on the multi-source raw dataset to obtain a feature tensor. Based on the feature tensor, the hydrological state form of the cell is defined. Based on the hydrological state form, candidate hydrological indicators are constructed to establish a mapping relationship between the candidate hydrological indicators and the observed water level points, and an inversion function is obtained. The inversion function is then used to output a global hydrological state raster sequence. Based on the feature tensor, event types are defined, and abrupt changes and persistent changes are extracted from the time series according to the event types. The abrupt changes and persistent changes are attributed to event categories to obtain a structured perturbation event library. Then, terrain zoning is performed according to landform and wetland types. The feature tensor, the global hydrological state raster sequence, the disturbance event library, and the terrain partition are integrated into a spatiotemporal reference data package.

4. The method for assessing carbon storage in mountainous marsh wetlands based on remote sensing and ground surveys according to claim 3, characterized in that, Topographic correction and spatiotemporal alignment are performed on the multi-source raw dataset to obtain a feature tensor. Based on the feature tensor, the hydrological state form of the pixel is defined, including: Define a unified spatial grid and time axis, and based on the spatial grid and time axis, perform cloud and fog removal, topographic radiation correction and multi-temporal synthesis on the optical data in the multi-source original dataset to obtain an optical synthesis feature set; The SAR data in the multi-source raw dataset are subjected to orbit correction, radiometric calibration, terrain correction, speckle noise suppression, and time series synthesis to obtain a SAR synthetic feature set, thus completing terrain correction and spatiotemporal alignment. The multi-source original datasets that have undergone terrain correction and spatiotemporal alignment are combined into a learnable data structure to obtain a feature tensor. Based on the feature tensor, the hydrological state form of the pixel is defined.

5. The method for assessing carbon storage in mountainous marsh wetlands based on remote sensing and ground surveys according to claim 4, characterized in that, Based on the spatiotemporal reference data package, peat thickness and thickness variance are predicted. Based on the predicted peat thickness and thickness variance, field sampling and measurement are performed to obtain field measurement data. The field measurement data is then subjected to quality inspection and data storage operations, including: Based on the defined core target variable for sampling and the actual constraints, partition regression and uncertainty estimation are performed for each terrain zone to obtain the prior model and output peat thickness prediction and thickness variance prediction. Active learning is performed using the peat thickness prediction and the thickness variance prediction to output the comprehensive pixel sampling revenue. Based on the comprehensive pixel sampling revenue, a sampling execution list for each terrain zone is generated. According to the sampling execution list, probes or core samples are used to probe downwards from the surface until the peat and mineral interface is reached. The total thickness, thickness of each layer, degree of decomposition, water content, color and plant residue characteristics are recorded to obtain peat thickness field sampling information. The thickness, bulk density, organic carbon content and gravel volume fraction of deep SOC are also recorded to obtain deep SOC field sampling information. The field measurement data are obtained by combining the peat thickness field sampling information and the deep SOC field sampling information. The on-site measurement data is associated with the pixel ID and stored in the database. Then, abnormal density checks, carbon content out-of-range checks, layer thickness missing checks, and coordinate checks are performed to complete the data quality inspection operation.

6. The method for assessing carbon storage in mountainous marsh wetlands based on remote sensing and ground surveys according to claim 5, characterized in that, The core target variables include key variables and auxiliary variables. The key variables include peat thickness and deep soil organic carbon. The auxiliary variables include bulk density, organic carbon content and groundwater level. The actual constraints include safety constraints, cost constraints, ecological constraints and time constraints.

7. The method for assessing carbon storage in mountainous marsh wetlands based on remote sensing and ground surveys according to claim 6, characterized in that, Based on the aforementioned field measurement data, a carbon pool set is calculated. Multi-task learning is then performed on the carbon pool set to obtain a multi-carbon pool joint prediction model, which outputs a predicted total carbon storage, including: The aboveground biomass carbon pool, the underground biomass carbon pool, the shallow soil organic carbon pool, the deep soil organic carbon pool, and the peat carbon pool are defined. Based on the field measurement data, the pixel area of ​​each carbon pool is converted and the output unit is unified to obtain a set of carbon pools. The spatiotemporal reference data package is set as the input feature, and the carbon pool set is subjected to multi-task learning using the input feature to obtain a multi-carbon pool joint prediction model. Multi-scale hierarchical Bayes is introduced into the multi-carbon pool joint prediction model to output the predicted total carbon storage and uncertainty analysis results with consistent scale.

8. The method for assessing carbon storage in mountainous marsh wetlands based on remote sensing and ground surveys according to claim 7, characterized in that, The spatiotemporal reference data package is set as the input feature, and the carbon pool set is subjected to multi-task learning using the input feature to obtain a multi-carbon pool joint prediction model. A multi-scale hierarchical Bayesian approach is introduced into the multi-carbon pool joint prediction model, and the outputs scale-consistent predicted total carbon storage and uncertainty analysis results, including: The spatiotemporal reference data package is set as the input feature, and the carbon library set is used to perform multi-task learning to obtain an initial prediction model and output the pixel prediction result. Multi-scale hierarchical Bayes is introduced into the multi-carbon pool joint prediction model to sum the pixel prediction results within the plot or watershed unit to obtain the pixel summary value. Based on the pixel aggregated value, the peat storage per unit area of ​​the pixel, the predicted value of the peat bulk density of the pixel, the predicted value of the carbon mass fraction of the peat of the pixel, and the predicted value of the peat thickness of the pixel are calculated to output the pixel carbon library. The pixel carbon library is added up by unit area to obtain the total carbon content per unit area of ​​the pixel. Based on the total carbon content of the pixel area, the mean map and standard deviation map of each pixel are output to obtain the pixel uncertainty. The pixel uncertainty is then decomposed according to four sources: missing remote sensing observations, sparse sampling points, peat thickness error, and scale constraint error, to obtain an error decomposition map.

9. A method for assessing carbon storage in mountainous marsh wetlands based on remote sensing and ground surveys according to claim 8, characterized in that, Based on the predicted total carbon storage and uncertainty analysis results, an interpretability analysis equation is constructed to output the causal path of carbon change in the predicted total carbon storage, thus completing the interpretability analysis of carbon storage change, including: Based on the predicted total carbon storage and the results of uncertainty analysis, response variables, mediating variables, external driving variables and control variables are defined to obtain the analysis framework; Based on the aforementioned analysis framework, for each pixel, a variable with the same time window is generated, and pixel fitting weights are defined according to the error decomposition map; the calculation expression for the pixel interpretable weights is as follows: ； in, For pixels Weights in interpretability analysis For pixels The variance of carbon variation To prevent extremely small positive numbers with a denominator of 0; Based on the analytical framework and the pixel interpretable weights, hydrological equations and carbon change equations are defined to obtain interpretable analytical equations. Using the interpretable analytical equations, the SEM path map and path coefficients of the predicted total carbon storage are output to obtain the causal path of carbon change.

10. A method for assessing carbon storage in mountainous marsh wetlands based on remote sensing and ground surveys according to claim 9, characterized in that, The response variable consists of pixel carbon change, the mediating variable consists of hydrological state, the external driving variable consists of climate pressure variable and anthropogenic disturbance variable, and the control variable consists of topographic zoning, soil type and initial carbon storage.

Images