A Dynamic Assessment Method for Wetland Carbon Sequestration Based on Multi-Source Remote Sensing Time-Series Fusion

By using a multi-source remote sensing time-series fusion method, the timing of wetland carbon sink conversion can be accurately located, solving the problem of location offset of carbon balance changes in existing technologies and realizing the accuracy and quantifiability of wetland carbon sink assessment.

CN123117928APending Publication Date: 2026-11-17姚嘉会
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611501431.5
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-09-24
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

Existing wetland carbon sink assessment methods suffer from inaccurate timing of carbon source-sink conversion due to single observation sources and fixed time scales. Furthermore, differences in the acquisition time, coverage, and response speed of different observation data lead to shifts in the location of carbon balance changes, making it difficult to accurately determine the actual timing of carbon source-sink conversion.

Method used

A multi-source remote sensing time-series fusion method was adopted to acquire optical image sequences and SAR image sequences. Combined with carbon flux and hydro-meteorological observation data from flux stations, a fusion grid was generated through asynchronous fusion. Vegetation index and SAR backscatter time series were extracted, inundation status was detected and ecological status intervals were divided. Phase correction was performed using lag time, carbon sink dynamic trajectory was generated, and the sign transition time of net carbon sink rate was detected.

Benefits of technology

Accurately determine the actual moment of carbon source and carbon sink conversion, and the assessment results correspond to both wetland status changes and carbon budget changes, eliminating sampling misalignment at the data level and response delay at the process level, and providing a quantifiable assessment of the degree of carbon sink function recovery.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123117928A_ABST
    Figure CN123117928A_ABST
Patent Text Reader

Abstract

This invention discloses a dynamic assessment method for wetland carbon sinks based on multi-source remote sensing time-series fusion, relating to the fields of remote sensing and carbon sink assessment technology. The method includes: acquiring optical image sequences, SAR image sequences, carbon flux observation data, and hydrological and meteorological observation data of the target wetland; determining the beat frequency period based on the sampling time interval of the two types of images and performing asynchronous fusion to generate a fused grid; extracting vegetation index time series and SAR backscattering time series, performing phenological phase extraction, inundation state detection, and ecological state staging; inverting vegetation net primary productivity, soil heterotrophic respiration, and hydrological carbon transport, and determining the lag time of each component based on cross-correlation analysis to obtain a synchronized carbon sink sequence; constructing a dynamic carbon sink trajectory with relative instantaneous phenological phase as the abscissa and net carbon sink rate as the ordinate, and identifying the carbon source-sink conversion critical point. This invention can reduce the impact of multi-source observation time mismatch and ecological process response delay on the assessment results.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of remote sensing and carbon sequestration assessment technology, specifically to a dynamic assessment method for wetland carbon sequestration based on multi-source remote sensing time-series fusion. Background Technology

[0002] The carbon balance of wetlands is affected by vegetation growth, soil decomposition, precipitation runoff and water level fluctuations, and continues to change with the seasons and flooding conditions. The same area may switch between carbon absorption and carbon release at different times, and the timing and duration of the switch directly affect the carbon sink accounting results.

[0003] Existing wetland carbon sink assessments typically employ single observation sources or fixed timescale statistics. Differences in the acquisition time, coverage, and response speed of different observation data can easily lead to temporal mismatches of the same ecological process. Furthermore, vegetation response, soil respiration, and hydrological transport have different lags in responding to environmental changes. Directly superimposing these data on calendar time may cause a shift in the location of carbon balance changes.

[0004] Therefore, given the continuous changes in wetland hydrological conditions and vegetation growth, and the inconsistent response times of various carbon processes, accurately determining the actual moment of carbon source and carbon sink conversion, and ensuring that the assessment results correspond to both wetland state changes and carbon budget changes, has become a problem that needs to be solved in the dynamic assessment of wetland carbon sinks. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a dynamic assessment method for wetland carbon sinks based on multi-source remote sensing time-series fusion. This method can accurately determine the actual time when carbon sources and carbon sinks are converted, and the assessment results simultaneously correspond to changes in wetland state and carbon budget.

[0006] To achieve the above objectives, the technical solution of the present invention is as follows:

[0007] In a first aspect, this invention discloses a method for dynamic assessment of wetland carbon sequestration based on multi-source remote sensing temporal fusion, comprising the following steps:

[0008] Acquire optical image sequences, SAR image sequences, carbon flux observation data and hydro-meteorological observation data of flux stations within the target wetland;

[0009] The beat frequency period is determined based on the sampling time interval between the optical image sequence and the SAR image sequence. The two types of images are then asynchronously fused using the beat frequency period as a constraint to generate a fused mesh that includes pigment perception dimension and vegetation structure perception dimension.

[0010] Vegetation index time series and SAR backscattering time series are extracted from the fused grid, and phenological phase extraction and flooding state detection are performed accordingly to divide the vegetation growth cycle into multiple ecological state intervals.

[0011] The carbon sink component time series was retrieved based on fused grid and hydro-meteorological observation data. The carbon sink component includes vegetation net primary productivity, soil heterotrophic respiration and hydrological carbon transport.

[0012] At flux stations, cross-correlation analysis is performed between hydrological or temperature sequences processed by seasonal term removal and corresponding carbon flux component sequences according to ecological state intervals to estimate the lag duration of each carbon sink component and then transmit the lag duration to each grid cell.

[0013] Phase correction of the time series of each carbon sink component is performed using the lag duration to obtain a synchronized carbon sink sequence;

[0014] Instantaneous phenological phases are constructed using phenological phase shift parameters extracted from phenological phases. The relative instantaneous phenological phases are determined as the abscissa using the initial boundary of each ecological state interval as the origin, and the net carbon sink rate calculated from the synchronized carbon sink sequence is used as the ordinate to generate a dynamic carbon sink trajectory.

[0015] The sign transition moment of the net carbon sink rate in the dynamic trajectory of the carbon sink is detected and output as the critical point of carbon source-sink conversion.

[0016] Secondly, this invention discloses a dynamic assessment system for wetland carbon sequestration based on multi-source remote sensing time-series fusion, comprising:

[0017] The data acquisition module is used to acquire optical image sequences, SAR image sequences, carbon flux observation data and hydro-meteorological observation data of flux stations within the target wetland.

[0018] The asynchronous fusion module is used to determine the beat frequency period based on the sampling time interval of the optical image sequence and the SAR image sequence, and to perform asynchronous fusion of the two types of images with the beat frequency period as a constraint to generate a fusion mesh containing pigment perception dimension and vegetation structure perception dimension.

[0019] The ecological state classification module is used to extract vegetation index time series and SAR backscatter time series from the fused grid, and perform phenological phase extraction and flooding state detection accordingly, dividing the vegetation growth cycle into multiple ecological state intervals.

[0020] The carbon sink component inversion module is used to invert the time series of carbon sink components based on fused grid and hydro-meteorological observation data. The carbon sink components include vegetation net primary productivity, soil heterotrophic respiration, and hydrological carbon transport.

[0021] The lag analysis module is used at flux stations to perform cross-correlation analysis between hydrological or temperature sequences processed by seasonal term removal and the corresponding carbon flux component sequences according to ecological state intervals, to estimate the lag duration of each carbon sink component, and to transmit the lag duration to each grid cell.

[0022] The phase correction module is used to perform phase correction on the time series of each carbon sink component using the lag time to obtain a synchronized carbon sink sequence.

[0023] The trajectory generation module is used to construct instantaneous phenological phases using phenological phase offset parameters extracted from phenological phases, determine the relative instantaneous phenological phases as the abscissa using the starting boundary of each ecological state interval as the origin, and use the net carbon sink rate calculated from the synchronized carbon sink sequence as the ordinate to generate a dynamic carbon sink trajectory.

[0024] The critical point output module is used to detect the sign transition moment of the net carbon sink rate in the dynamic trajectory of the carbon sink, and output it as the critical point of carbon source-sink conversion.

[0025] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0026] 1. The beat frequency period is determined based on the sampling time interval of the optical image sequence and the SAR image sequence. The two types of images are asynchronously fused with the beat frequency period as a constraint. Each type of observation is registered to the pigment perception dimension and the vegetation structure perception dimension according to its actual sampling time, without time alignment or equivalent quantity conversion, thus preserving the original time structure of the two types of observations from the source. On this basis, the lag time of each carbon sink component is estimated and phase correction is performed at the flux station according to the ecological state interval. The net primary productivity of vegetation, soil heterotrophic respiration and hydrological carbon transport are reduced from their respective observation times to the same ecological driving time. As a result, the sampling misalignment at the data level and the response delay at the process level are eliminated respectively. The three carbon sink components correspond to the same batch of environmental driving conditions when added or subtracted. The zero-crossing position of the net carbon sink rate is no longer shifted due to the misalignment of component peaks and valleys. The actual time of carbon source and carbon sink conversion can be accurately determined.

[0027] 2. This invention enables the assessment results of carbon sink dynamics to correspond simultaneously with wetland state changes and carbon budget changes. On the one hand, it uses SAR dual-branch inundation detection as the main criterion and abrupt changes in phenological phase shift parameters and collapse of vegetation index amplitude as cross-validation to divide the vegetation growth cycle into inundation period, exposure period and transition period, so that the hydrological state transition of wetlands can be time-positioned by two independent types of evidence, microwave and optical. On the other hand, it constructs a relative instantaneous phenological phase as the abscissa with the starting boundary of each ecological state interval as the origin, so that the abscissa advances with the vegetation growth process rather than the calendar process. The resulting carbon sink dynamic trajectory carries at each point the ecological state range of the wetland, the phenological process position of the vegetation, and the net carbon sink rate at that moment. The output carbon source-sink transition critical point can therefore be directly linked to specific hydrological and phenological stages. Moreover, when multi-year data are superimposed in the same phase space, they are not affected by inter-year phenological differences. In addition, since the horizontal axis advances with the vegetation growth process rather than the calendar process, the carbon sink dynamic trajectory of each year before and after restoration can be directly compared at the same phenological position. This makes the change in net carbon sink rate and the movement of the carbon source-carbon sink transition critical point quantifiable indicators, which can be used to evaluate the effectiveness of wetland restoration. By comparing the net carbon sink rate and the carbon source-carbon sink transition time before and after restoration, the degree of restoration of wetland carbon sink function can be evaluated. Attached Figure Description

[0028] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art 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.

[0029] Figure 1 This is an overall block diagram of the method in Embodiment 1 of the present invention;

[0030] Figure 2 This is the timing interaction diagram of the present invention;

[0031] Figure 3 This is an overall block diagram of the system in Embodiment 2 of the present invention. Detailed Implementation

[0032] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments. 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.

[0033] Application Overview: This invention uses the beat frequency period as a constraint to register optical and SAR images into the same fusion grid while retaining their respective actual sampling times. Based on this, vegetation index time series and SAR backscattering time series are extracted, phenological phase extraction and inundation state detection are completed, and ecological state intervals are divided. Then, net primary productivity of vegetation, soil heterotrophic respiration, and hydrological carbon transport are inverted separately. The lag time of each component is estimated using flux station measured sequences, and phase correction is performed to reduce the three to the same ecological driving time reference. Finally, a carbon sink dynamic trajectory is generated with the relative instantaneous phenological phase as the abscissa and the net carbon sink rate as the ordinate. The carbon source-sink conversion critical point is output by detecting the sign conversion time of the net carbon sink rate.

[0034] Example 1:

[0035] like Figures 1-2 As shown, the wetland carbon sequestration dynamic assessment method based on multi-source remote sensing time-series fusion includes the following steps:

[0036] For ease of understanding, the following example uses a temperate reed marsh wetland in northern China. This wetland covers approximately 12 square kilometers, with reeds as the dominant vegetation, and is an emergent herbaceous wetland. The spring flood season occurs annually from late March to early April, causing water levels to rise and resulting in large-scale submersion. The growing season is from May to September, during which most of the wetland surface is exposed. After October, the vegetation withers. A flux tower equipped with an eddy covariance system is located within the wetland. Around the tower are automatic water level gauges, static chamber soil respiration monitoring devices, rain gauges, and joint monitoring sections for water carbon concentration and flow. The assessment output period is from January 1, 2022 to December 31, 2023, with 2022 used for parameter calibration and lag estimation, and 2023 reserved as an independent year for consistency verification.

[0037] This invention is applicable to temperate wetlands with significant annual cycle characteristics, stable hydrological rhythms, and where flux monitoring stations are deployed or can be referenced from nearby stations. It is a retrospective assessment method that allows the use of observation data located after the treatment time point and is not used for real-time inference. The carbon budget assessed is based on carbon dioxide and does not include methane flux and carbon dioxide escape flux from the water surface.

[0038] Step S1: Acquire optical image sequences, SAR image sequences, carbon flux observation data and hydro-meteorological observation data of flux stations within the target wetland;

[0039] Optical image sequence refers to the time series of multispectral images formed by repeated transits of satellite optical sensors over target wetlands; SAR image sequence refers to the time series of images formed by repeated transits of satellite synthetic aperture radar sensors. Its ability to penetrate clouds and fog allows it to still stably acquire data during periods when optical observation is blocked by clouds and rain. This complementarity is the physical basis for incorporating the two types of images into the same grid.

[0040] Furthermore, the SAR image sequence includes a first polarization channel SAR image sequence and a second polarization channel SAR image sequence. In this embodiment, the first polarization channel uses VV polarization, and the second polarization channel uses VH polarization. VV polarization responds strongly to the secondary scattering structure formed by the water surface and stem, while VH polarization is mainly contributed by multiple volume scattering within the canopy. The combination of the two can distinguish between three states: open water surface, submerged under vegetation canopy, and exposed surface.

[0041] Carbon flux observation data include vegetation carbon uptake observation sequences, soil respiration observation sequences, and hydrological carbon transport observation sequences. The vegetation carbon uptake observation sequence was obtained from net ecosystem carbon exchange measured by the eddy covariance system using a nighttime temperature response decomposition method. This method extrapolates daytime ecosystem respiration by utilizing the nighttime carbon exchange-temperature response relationship under no-photosynthetic conditions. The measured time series of total primary productivity is obtained by subtracting the net ecosystem carbon exchange from this extrapolation. Simultaneously, the measured time series of ecosystem respiration was obtained. Soil respiration observation sequences are preferentially derived from static chamber method measurements, which obtains the soil respiration rate by measuring the cumulative rate of carbon dioxide concentration over time within a sealed chamber. Hydrological carbon transport observation sequences are obtained by multiplying measured water body carbon concentration by cross-sectional flow. Hydrometeorological observation data include automatic water level gauge measurements, photosynthetically active radiation sequences, soil temperature sequences, and rain gauge precipitation records. Precipitation records can be supplemented spatially by satellite precipitation products.

[0042] In addition, soil organic carbon pool data for the target wetland were obtained by integrating the organic carbon content and bulk density measured from soil profile sampling with respect to surface depth, or by taking data from a regional soil database. Soil organic carbon pool storage characterizes the total amount of organic carbon in the topsoil that can be decomposed by microorganisms, determining the upper limit of soil carbon release potential.

[0043] Basic preprocessing is performed on the above data: radiometric calibration and atmospheric correction are performed on optical images to obtain surface reflectance images; then, cloud masks are generated based on the spectral discrimination rules of the image quality evaluation band or the visible, near-infrared, and thermal infrared bands. Pixels contaminated by clouds, cloud shadows, and thick aerosols are marked as invalid, and this time point is considered as missing optical data. Radiometric calibration, speckle filtering, radiometric topographic correction, and geometric correction are performed on SAR images to eliminate speckle noise, topographic radiation distortion, and slant range projection distortion in sequence, resulting in SAR backscatter images. Quality control is performed on station data, removing outliers that exceed the physically reasonable range, marking missing segments, and unifying the timestamps to Coordinated Universal Time. The geometric registration accuracy of the two types of images should be controlled within half a pixel.

[0044] Following the example, optical images were taken from satellites with a 10-day revisit period, and SAR images were taken from satellites with a 6-day revisit period. These were resampled to a 30-meter grid, resulting in approximately 13,000 pixels within the wetland area. The station data were aggregated to form a daily-scale sequence. Considering that subsequent sliding window fitting requires data from approximately six months before and after the window center, the actual acquisition period was from July 1, 2021 to June 30, 2024, extending by six months before and after the evaluation output period.

[0045] When the data volume is large, a distributed remote sensing data lake can be built based on the object storage mechanism. The images are indexed by geospatial and temporal coordinates, and the station sequences are indexed by spatial location and temporal coordinates. These are stored in distributed storage nodes, and an elastic computing scheduling framework is established to distribute retrieval and processing tasks to different computing nodes for parallel execution according to spatiotemporal range. This forms a cloud-based distributed spatiotemporal data grid that supports asynchronous queries by spatiotemporal coordinates, providing a parallelization foundation for pixel-by-pixel time-series processing.

[0046] Step S2: Determine the beat frequency period based on the sampling time interval between the optical image sequence and the SAR image sequence, and perform asynchronous fusion of the two types of images with the beat frequency period as a constraint to generate a fusion grid containing pigment perception dimension and vegetation structure perception dimension.

[0047] Existing multi-source fusion methods typically resample images to the same time scale before performing pixel-level fusion. The revisit periods of the two types of satellites are generally not divisible by each other, and their transit times exhibit continuous phase differences over time. Forced alignment inevitably requires temporal interpolation of at least one type of data, which smooths out abrupt changes in state during the rapid transition from flooding to exposure. Converting SAR to optical equivalents introduces conversion model errors and loses structural and water content information. This invention does not perform temporal alignment, but retains the actual sampling times of each type of image and registers them separately according to the sensing dimension, allowing various observations to be accessed downstream as needed in their original form.

[0048] Specifically, asynchronous fusion of the two types of images is performed under the constraint of the beat frequency period. This includes determining the beat frequency period corresponding to one cycle of sampling time difference between the two types of images, based on the sampling time interval between the optical image sequence and the SAR image sequence. Let the optical satellite revisit period be... SAR satellite revisit period is If all are taken as nominal integer days, then: ;

[0049] in, For optical satellite revisit cycles, For SAR satellite revisit cycle, Let be the beat frequency period. The derivation is as follows: the sampling frequencies of the two sources are respectively... and Its sampling phase difference is The rate of accumulation reaches a complete cycle, at which point the relative positions of the two source sampling times return to their initial state. Therefore, the time required for a complete drift cycle is the reciprocal of this rate, which leads to the above formula. The beat frequency period characterizes the time required for the two source sampling times to go from coincidence to maximum offset and back to coincidence. Cross-source statistics must cover at least one beat frequency period to traverse all relative positions and eliminate sampling phase offset.

[0050] In conjunction with this, the least common multiple alignment period can also be calculated: ;

[0051] in, This indicates taking the least common multiple. The least common multiple alignment period represents the shortest interval between the transit times of two types of satellites returning to the same calendar day, and is used as the framework period for organizing statistics in periodic segments. The two are generally not equal, only when... The times are exactly equal, such as when the revisit period is 16 days and 12 days. On the 10th and the 6th day, The days must be used separately. The least common multiple alignment period describes the recurrence rhythm of sampled events on the calendar and is used for segmentation; the beat frequency period describes the drift rhythm of the sampled phase difference and is used to limit the time distance and the length of the statistical window.

[0052] Following the example, day, Day, get day, day.

[0053] After determining the beat frequency period, the optical image data and their actual sampling time in the optical image sequence are registered to the pigment perception dimension, and the SAR image data and their actual sampling time in the SAR image sequence are registered to the vegetation structure perception dimension, using each sampling time in the fusion grid as the time index.

[0054] The division of perception dimensions is based on the different perception mechanisms of the two types of sensors: the optical path receives the reflection of shortwave solar radiation from the Earth's surface, and the signal is mainly determined by the absorption characteristics of pigments such as chlorophyll, which is labeled as the pigment perception dimension, characterizing the spectral response of vegetation chlorophyll content; the SAR path receives the backscattering of active microwaves from the Earth's surface, and the signal is mainly determined by the canopy geometry, which is labeled as the vegetation structure perception dimension, characterizing the scattering response of the canopy geometry; since backscattering also depends on the dielectric constant, and the dielectric constant varies with soil moisture content, the SAR path can also label the soil moisture content perception dimension, characterizing the dielectric response of soil moisture state. The above labeling is performed grid-by-grid cell, forming a perception dimension labeling tensor. Dimensional storage ensures that various observations retain their original physical meaning and are not mutually converted, avoiding the accumulation and propagation of conversion errors in the link.

[0055] Furthermore, for any given sampling time, only another type of image data whose time distance from that sampling time does not exceed the beat frequency cycle is registered, and the actual sampling time of the registered image data is retained.

[0056] This constraint is used to maintain grid index continuity with SAR observations at optically missing nodes, while preventing incorrect association of observations with excessively large temporal distances. Specifically, it involves cross-temporal misalignment observation registration: at time nodes where optical data is missing, the vegetation structure sensing dimension and soil moisture sensing dimension observations from the SAR pathway are written into the observation ledger of that grid cell as misaligned observation entries. The entry fields include sensor type, actual acquisition timestamp, observation type, numerical value, and quality markers. This registration does not perform time alignment or feature transformation to optical equivalents; misaligned entries are only invoked downstream based on their original sensing dimensions. The temporal distance between the entry and the node being supplemented must not exceed the beat frequency cycle. Those exceeding the limit will be marked as missing and will not be registered.

[0057] The basis for using the beat frequency period as the upper limit is that within one beat frequency period, the sampling times of the two sources have not yet completed a full drift, and the registered entry and the current node still belong to the same drift segment in terms of sampling phase, making the surface state they carry comparable. Beyond one beat frequency period, the sampling phase has undergone a complete cycle, the correlation is no longer consistent, and the wetland surface state may have changed substantially, so forced registration would introduce false information. Therefore, this upper limit is a quantitative boundary with physical basis, rather than an empirical value.

[0058] Following the example, the optical data for days 186 and 196 in early July 2022 were marked as invalid by cloud masking, while there was one SAR image each for days 188 and 194. For the node on day 186, the SAR image on day 188 is 2 days away from it, which is less than 15 days away. Therefore, its VV and VH backscatter values, along with the actual acquisition timestamp 188, observation type, and quality label, are registered in the vegetation structure perception dimension and soil moisture perception dimension of this node. If there is a SAR image 20 days away, it is marked as missing because it is more than 15 days away. Although this node has no optical observations, it still holds usable structure and moisture information, and there are no breaks in the grid index.

[0059] In addition, the station observation sequences and meteorological sequences are organized into station observation sub-units according to the spatial location of the stations, and stored and indexed together with the grid.

[0060] The resulting fused mesh has the following characteristics: all types of observations are entered into the database at the actual sampling time, and the sampling interval truly reflects the acquired rhythm without being smoothed by interpolation, so that rapid transitions such as flooding maintain detectable sharpness; each sensing dimension is independent of each other and does not convert to each other, avoiding cross-source conversion errors; the registration rule with the beat frequency period as the upper limit of time distance ensures that the mesh maintains index continuity during optically missing periods, while avoiding erroneous associations across phase cycles.

[0061] Step S3: Extract vegetation index time series and SAR backscattering time series from the fused grid, and perform phenological phase extraction and flooding state detection accordingly to divide the vegetation growth cycle into multiple ecological state intervals.

[0062] The process of phenological phase extraction includes: extracting multispectral reflectance from the optical image data corresponding to the pigment perception dimension in the fused grid, and calculating the vegetation index time series based on the multispectral reflectance.

[0063] In this embodiment, the reflectance of the red band and near-infrared band is used for calculation according to the definition of the normalized vegetation index:

[0064] ;

[0065] in, For a moment Near-infrared surface reflectance, Red band surface reflectance, For a moment The vegetation index value. This formula utilizes the characteristic that green vegetation has low reflectivity in the red band due to strong chlorophyll absorption, and high reflectivity in the near-infrared band due to multiple scattering by mesophyll cells. The value range is [value missing]. It represents the relative abundance and vitality of green vegetation within a pixel.

[0066] The SAR misaligned observation entries registered in step S2 are not included in the calculation of this step. This step only reads the optical pathway entries in the pigment perception dimension. The asynchronous sampling timestamp information is retained, and optical missing nodes are not interpolated. The sampling irregularities are directly handled by the subsequent fitting stage, thereby avoiding the interpolation values ​​from changing the true shape of the phenological curve.

[0067] Next, a sliding window harmonic fitting was performed according to the actual sampling time of each vegetation index value in the vegetation index time series to obtain the phenological phase shift parameters and vegetation index amplitude corresponding to each sliding window.

[0068] Setting the harmonic fitting order , baseband Using a one-year window and the median of the sampling interval as the sliding step, least-squares harmonic fitting is performed on the irregularly sampled vegetation indices within each window:

[0069] ;

[0070] in, For a moment Fitted values ​​of vegetation index; is a constant term, representing the annual average level of vegetation index within the window; , The first Cosine and sine coefficients of the first harmonic; As the fundamental frequency, take Corresponding annual cycle; The harmonic order is represented by the first-order component, which describes the annual cyclical growth rhythm of a single peak, and the second-order component, which describes the asymmetry of the growth curve relative to the sinusoidal shape. A uniform absolute calendar day count is used, and the time origin is fixed throughout to ensure that the phase parameters of different windows are in the same reference system. The order is set to 2 because: an order that is too low is insufficient to characterize the asymmetry of the growth curve, while an order that is too high will incorporate noise such as cloud pollution residuals into the model, reducing the stability of phase estimation. Irregular sampling points directly participate in the fitting, naturally accommodating the sampling unevenness caused by asynchronous sampling and optical measurement deficiencies.

[0071] Each window contains 5 parameters to be estimated, and each window must contain at least 10 valid samples. Parameters with fewer than 10 valid samples will not be included in the extraction process and will be marked as missing. Amplitude and phase shift parameters are extracted window by window:

[0072] ;

[0073] ;

[0074] in, For the first The amplitude parameter of a harmonic wave characterizes the amplitude of that harmonic component. For the first Phase shift parameters of first harmonics; It is the arctangent function in the fourth quadrant, and its principal value range is... The phase quadrant can be determined based on the signs of the two coefficients. This can be achieved through trigonometric identities. It can be seen that the first-order component can be written as The peak occurred at Therefore, the first-order phase shift parameter directly encodes the timing of the vegetation seasonal peak relative to a fixed time origin: its increase corresponds to a delayed peak, and its decrease corresponds to an earlier peak; the first-order amplitude characterizes the strength of the annual cycle fluctuation. Phenology is characterized by harmonic phase rather than the timing of the vegetation index threshold crossing, and its phase parameter is jointly determined by all samples within the window, making it more robust to cloud pollution residuals and sampling sparsity.

[0075] Since the output of the four-quadrant arctangent function is limited to the principal value range, the small fluctuations in the true phase near the endpoints will appear as approximately The transition occurs. Therefore, phase unwinding is performed before the sequence is formed: the phase difference between adjacent windows is compared point by point, and if the difference is greater than a certain threshold, the phase difference is allowed to change. Subtract the last point less than Adding time This ensures sequence continuity; after unwinding, the values ​​can exceed the principal value range without affecting the calculation of the first-order difference and phase difference. Without this processing, entanglement jumps near the endpoints will be misjudged as genuine phenological abrupt changes by subsequent abrupt change criteria.

[0076] Take the first-order phase shift parameter As a phenological phase shift parameter (hereinafter referred to as Take the first order amplitude. As the amplitude of the vegetation index.

[0077] Then, the phenological phase shift parameters and vegetation index amplitudes are arranged according to the time position of each sliding window to obtain the phenological phase shift parameter sequence and the vegetation index amplitude time series, respectively.

[0078] Each window parameter is assigned to its center time and sorted chronologically. The value of the sequence at any given time is obtained by linear interpolation between the centers of adjacent windows, without extrapolation. The window length is one year, and the parameters are assigned to the center, which means that data from approximately six months before and after the center time is used, consistent with the retrospective evaluation positioning. Extending the data acquisition period by six months before and after ensures that every moment within the evaluation output period falls within the complete window coverage.

[0079] Following the example, the sliding step size is taken as the median of the optical sampling interval, 10 days. The first-order amplitude of the window corresponding to the 2022 growing season for the pixel where the flux tower is located is approximately 0.31, and the first-order phenological phase shift parameter is approximately 3.42 radians (value taken after unwinding). The calculated peak occurs around the 199th day, which is consistent with the pattern of the reed canopy being at its peak in mid-to-late July. The same period in 2023 is 3.30 radians, corresponding to the peak around the 192nd day, about 7 days earlier than in 2022, reflecting the earlier phenological events caused by the higher spring temperatures that year.

[0080] Simultaneously with phenological phase extraction, SAR backscattering time series are extracted from SAR image data corresponding to the vegetation structure perception dimension to perform flooding state detection. The SAR backscattering time series includes the SAR backscattering time series of the first polarization channel and the SAR backscattering time series of the second polarization channel. Each polarization channel SAR backscattering time series includes the SAR backscattering values ​​corresponding to each sampling time.

[0081] Wetland submersion manifests in two distinct ways on radar imagery: in open water without vegetation cover, specular reflection causes energy to escape forward with minimal echo, resulting in significantly lower backscattering compared to vegetation and soil; however, submersion beneath emergent vegetation canopy exhibits strong secondary scattering due to the near-right-angle dihedral structure formed by the water surface and stems, causing microwaves to return along their original path after two reflections, leading to an increase in VV polarization backscattering. Using only a low threshold for water body detection would completely miss the latter, and reed marshes are predominantly composed of emergent vegetation. Therefore, a bi-branch criterion designed for each of the two scattering mechanisms is employed. The two branches have opposite directions and each corresponds to a specific mechanism, ensuring a complete and unambiguous representation of the submersion state on radar observations.

[0082] The threshold determination and baseline statistics for backscattering values ​​are performed in the decibel domain, i.e., the logarithm to the base 10 is taken and multiplied by 10. The reason is that the backscattering coefficient spans several orders of magnitude, and the distribution after taking the logarithm is closer to normal, making the criterion based on the mean and standard deviation valid.

[0083] Specifically, the flooding state detection includes: determining the open water surface determination benchmark based on the distribution of each SAR backscatter value in the SAR backscatter time series of the first polarization channel, and determining the grid cells whose SAR backscatter value of the first polarization channel is lower than the open water surface determination benchmark as open water surface grid cells.

[0084] In this embodiment, the Otsu method is used to calculate the segmentation threshold that maximizes the inter-class variance from the VV polarization backscattering histogram for each scene. This threshold is then used as the benchmark. For each scene, a threshold is calculated to automatically absorb differences in incident angle, wind conditions, and calibration at different times of passage. Since the Otsu method relies on a bimodal distribution, the histogram may degenerate into a single peak when the water surface area is extremely small or extremely large, rendering the threshold meaningless. Therefore, a bimodality test is performed before thresholding: the Otsu separability measure is used. Or Sarle bimodal coefficient It is determined to be divisible. The ratio of between-class variance to total variance under optimal segmentation, with a range of values. The closer the value is to 1, the more complete the separation. Calculate using the following formula: ;

[0085] in, This is the Sarle bimodal coefficient; The sample skewness is calculated as the ratio of the third central moment to the cube of the standard deviation. The sample kurtosis is calculated as the ratio of the fourth central moment to the fourth power of the standard deviation. The corresponding value for a uniform distribution is... The closer the structure is to a bimodal distribution, the larger the coefficient becomes. If the separability condition is not met, the system reverts to the global Otsu threshold of the histogram of multiple scenes within the current least common multiple alignment period, and recovers the bimodal structure by utilizing the water surface pixels that appear sufficiently in other scenes.

[0086] Next, the polarization ratio timing is calculated based on the SAR backscattering timing of the first polarization channel and the SAR backscattering timing of the second polarization channel.

[0087] The polarizability ratio is calculated as the ratio of the backscattered values ​​of VH and VV at the same moment, representing the contribution of volume scattering to secondary scattering: multiple scattering within the canopy generates strong cross-polarized echoes, increasing the polarizability ratio; secondary scattering from the water surface to the stem mainly contributes to the same polarized echoes, decreasing the polarizability ratio. As a ratio between the two channels, it can offset systematic factors such as incident angle and calibration deviation that have a consistent effect on both channels.

[0088] Then, water level sequences are extracted from hydrological and meteorological observation data, exposure reference periods are determined based on water level sequences, and exposure state scattering baselines and exposure state polarization ratio baselines are determined based on the SAR backscattering value and polarization ratio value of the first polarization channel within the exposure reference period.

[0089] Water level is the most direct physical quantity determining whether the land surface is submerged. Taking a continuous period where the water level is consistently lower than the beach elevation as the exposure reference period ensures that the baseline is not contaminated by submerged pixels. The beach elevation is provided pixel by pixel by the digital elevation model, or it can be replaced by the upper quantile value of the water level corresponding to the historical exposure time of that pixel; the continuous period must be no less than three SAR revisit cycles and must not contain precipitation events to avoid short-term water accumulation from contaminating the statistics.

[0090] Let the mean value of VV backscattering during this period be denoted as Standard deviation is The exposed state scattering baseline is taken The baseline polarization ratio for exposure status is calculated by subtracting twice the standard deviation from the mean polarization ratio within that time period. Using twice the standard deviation is because, in an approximately normal distribution, this corresponds to about the 97.7th percentile on one side. Observations falling outside the baseline have a probability of less than 3% occurring under exposure conditions, thus striking a balance between missed and false detections. Baseline statistics are performed pixel-by-pixel, automatically incorporating differences in vegetation density, topography, and soil conditions between pixels, allowing the same set of rules to be generalized to all pixels without the need for manually setting thresholds.

[0091] Subsequently, grid cells with SAR backscattering values ​​higher than the exposed state scattering baseline and polarization ratios lower than the exposed state polarization ratio baseline were identified as submerged grid cells under the vegetation canopy.

[0092] This criterion requires both an increase in volume scattering (VV) and a decrease in polarizability to occur simultaneously. This is based on the premise that secondary scattering simultaneously enhances co-polarized echoes and reduces the proportion of cross-polarization. During sub-canopy flooding, VV increases significantly due to secondary scattering, while VH remains essentially unchanged or slightly decreases due to the specular loss of the cross-polarization component by the water surface. Therefore, the decrease in polarizability is caused by the inverse changes in both the numerator and denominator. Conversely, the increased volume scattering caused by biomass growth results in a greater increase in VH than in VV, leading to a corresponding increase in polarizability—in the opposite direction. Requiring both conditions to be met simultaneously eliminates the interference of biomass growth, making this criterion uniquely applicable to sub-canopy flooding.

[0093] Finally, the open water surface grid cells and the submerged grid cells under the vegetation canopy are merged to obtain the submersion state sequence. This sequence is a binary labeled sequence for each sampling time and each pixel, representing the surface water cover status of each pixel.

[0094] Continuing with the example, the cell containing the flux tower is located within an emergent reed community. During the spring flood season at the end of March 2022, its VV backscattering value increased from... about The decibel level rose to approximately decibels, exceeding the baseline of scattering under exposure conditions. decibels (corresponding to) (decibels); the synchronous polarization ratio is approximately dropped to about The polarization ratio is approximately lower than the baseline under exposure conditions. If both conditions are met simultaneously, the pixel is identified as a submerged grid cell beneath the vegetation canopy. Meanwhile, the VV backscattering value of pixels in the permanent waterlogged area at the center of the wetland is consistently below the Otsu threshold for that scene. The decibel level is used to classify the pixels as open water grid cells. If only the open water branch is used, the spring flooding of the cell where the flux tower is located will be completely missed, and the soil anaerobic inhibition and ecological state staging for the corresponding time period will also be incorrect.

[0095] After obtaining the above three sequences, the vegetation growth cycle is divided into multiple ecological state intervals, specifically including:

[0096] The sampling time at which the state transition occurs in the flooded state sequence is determined as the candidate boundary. To suppress false transitions caused by single-scene noise, the new state must be maintained for at least two consecutive SAR sampling times after the transition before it is confirmed. If there are missing measurements in between, the consistency of the valid samples before and after the missing measurement is used for determination. If the two sides are inconsistent, it is not confirmed and is marked for investigation.

[0097] The change between adjacent phenological phase shift parameters in the phenological phase shift parameter sequence is calculated. The statistical distribution of the change is used to determine the phase change judgment benchmark, and the sampling time when the change exceeds the phase change judgment benchmark is determined as the phase change time.

[0098] This change is the first difference of the phenological phase shift parameter sequence: ;

[0099] in, For a moment The change in phenological phase shift parameters at that location and The values ​​are the unwound values ​​of the center times of two adjacent windows. When the phenological rhythm is stable, the peak times of the fitted values ​​for each window are basically consistent, and the first-order difference is close to zero. Disturbances such as flooding cause the waveforms within the window to be suppressed or truncated, and the fitted peak times jump accordingly, which are manifested as significant outliers in the first-order difference. This characteristic makes the first-order difference a flooding conversion corroboration quantity independent of microwave observations.

[0100] Based on this, the criteria for determining phase abrupt changes are as follows: ;

[0101] in, As a criterion for determining phase abrupt changes, The mean of the absolute values ​​of the changes. Its standard deviation. Taking the mean plus twice the standard deviation corresponds to approximately the 97.7th percentile on a one-sided chart, which can screen for significant abrupt changes without misinterpreting normal interannual fluctuations. This benchmark is derived from the statistical distribution of the sequence itself and can automatically adapt to the fluctuation levels of different pixels and years. When The time markers are the moments of phase abrupt change, and their set is denoted as . .

[0102] The baseline of the exposure state amplitude is determined based on the vegetation index amplitude corresponding to the period in the flooded state sequence that is identified as the exposure state, and the period when the vegetation index amplitude is lower than the baseline of the exposure state amplitude is determined as the amplitude collapse period.

[0103] In this embodiment, the median of the first-order amplitude during the exposure period within the current least common multiple alignment period is taken as... The baseline amplitude of the exposed state is taken as The median is used to robustly handle anomalous amplitudes caused by residuals from individual cloud pollution events. If there are fewer than 5 amplitude samples within a given period, the statistical period is extended forward until there are at least 5 samples. If this is still insufficient, the median of the amplitudes from all historical exposure states of that pixel is used. The physical basis for this is that after flooding, the water surface replaces vegetation as the main component of the pixel, causing a sharp drop in near-infrared reflectivity while the red band remains relatively stable. The vegetation index is suppressed overall, and seasonal fluctuations are flattened, resulting in a sharp drop in the first-order amplitude. Using half of the median as a boundary can distinguish the signal collapse caused by flooding from normal interannual intensity variations.

[0104] When the time difference between the candidate boundary and the phase change moment does not exceed the time interval between adjacent SAR sampling moments, or when the candidate boundary is located within the amplitude collapse period, the candidate boundary is determined as the verification consistent boundary.

[0105] The significance of this cross-validation lies in the fact that flooding detection is based on microwave scattering mechanisms, while phase abrupt changes and amplitude collapse are based on optical spectral responses. Since their origins are independent and they corroborate each other temporally, the boundary reliability is significantly improved, thereby reducing the risk of mis-stage predictions due to noise from a single sensor. The tolerance is set at the adjacent SAR sampling interval because the temporal resolution of SAR detection is limited by this interval. When the two criteria conflict, SAR detection takes precedence because SAR is unaffected by clouds and rain, has a direct response to flooding, and exhibits better temporal continuity.

[0106] The flooding period and the exposure period are divided according to the candidate boundary, and the adjacent sampling period containing the candidate boundary is determined as the transition period.

[0107] The process divides the timeframe into three intervals: the interval where the flooding continues is the flooding period, and the interval where the exposure continues is the exposure period. A transition period is defined as a time interval extending a certain width forward and backward from the candidate boundary, drawn from adjacent intervals. These three types of intervals do not overlap on the timeline and collectively cover the entire evaluation output period. The transition period has a half-width interval. The transition period is defined as twice the median of adjacent sampling intervals within the current least common multiple alignment period, and is not less than one SAR revisit period. When the sampling interval contains fewer than 5 samples, the statistical period is also extended forward. Using twice the median ensures that the transition period covers at least one typical sample before and after the boundary, including boundary positioning errors. Using the median ensures that this width is robust to excessively long intervals caused by missing measurements. The lower limit is one SAR revisit period to guarantee that the transition period is not narrower than the temporal resolution of the detection itself. The transition period characterizes the time when the surface state is uncertain or rapidly changing; separating it avoids contaminating the statistical characteristics of each period with mixed state conditions.

[0108] For the Each ecological state interval is recorded, along with its initial boundary time. and the phenological phase shift parameter value at that moment ,in This represents the interval number. Simultaneously, at each sampling time, the proportion of submerged pixels is determined based on the number of grid pixels identified as being in a submerged state and the total number of grid pixels. The submerged pixel proportions are then arranged according to the sampling time to obtain the temporal sequence of the submerged pixel proportions. The denominator is the total number of pixels with valid SAR observations at that time; if SAR observations are missing, that time is marked as missing, and subsequent retrievals are performed using linear interpolation based on adjacent valid times. The submerged pixel proportion characterizes the share of wetland surface water coverage over the entire wetland and is a surface-scale hydrological state indicator.

[0109] Following the example, the median interval between adjacent samplings in this pixel flooding detection sequence is 6 days, therefore The 2022 annual division results are as follows: Days 1 to 72 are the exposure period; a candidate boundary from exposure to flooding was detected on Day 85, with the 12 days before and after (Days 73 to 97) being the transition period; Days 98 to 146 are the flooding period; a candidate boundary from flooding to exposure was detected on Day 159, with Days 147 to 171 being the transition period; and Days 172 to 365 are the exposure period. These five intervals are consecutive and do not overlap. At the boundary on Day 85, the absolute value of the first-order difference of the phenological phase shift parameter is 0.42 radians, exceeding the phase abrupt change judgment benchmark of 0.27 radians. The time difference between the two is 3 days, not exceeding the SAR revisit period of 6 days, thus it is determined to be a consistent boundary. During the same period, the first-order amplitude decreased from the median of 0.30 in the exposure state to 0.11, below the baseline of 0.15, falling into the amplitude collapse period, further confirming this boundary. The proportion of submerged pixels rose from about 0.18 to about 0.71 around day 85, and then fell back to about 0.23 around day 159.

[0110] Step S4: Retrieve the time series of carbon sink components based on fused grid and hydro-meteorological observation data. The carbon sink components include vegetation net primary productivity, soil heterotrophic respiration, and hydrological carbon transport.

[0111] The three components have different driving factors, response times, and spatial patterns. Quantization inversion allows each component to be subject to a hysteresis correction that matches its response mechanism, without having to apply a single average delay to the total.

[0112] The process of retrieving vegetation net primary productivity includes extracting photosynthetically active radiation (PARF) sequences, temperature sequences, and hydrological sequences from hydro-meteorological observation data. The PARF sequence is the temporal series of radiation fluxes reaching the canopy in the wavelength range of 400 to 700 nanometers that can be used for photosynthesis, characterizing the energy supply driving photosynthesis; the temperature sequence is taken from air temperature or canopy temperature; and the hydrological sequence is taken from water level sequences, characterizing water supply and stress conditions.

[0113] The proportion of photosynthetically active radiation absorbed by vegetation is determined based on the vegetation index time series. The temperature response and water response of vegetation light energy utilization are determined based on the temperature series and hydrological series, respectively. The total primary productivity of vegetation is determined by combining the photosynthetically active radiation series.

[0114] This embodiment uses a light energy utilization model: ;

[0115] in, Total primary productivity of vegetation represents the total amount of carbon fixed by vegetation through photosynthesis per unit area per unit time. Photosynthetically active radiation; The proportion of photosynthetically active radiation absorbed by vegetation; The maximum light energy utilization rate characterizes the amount of carbon that can be fixed per unit of absorbed radiation under no-stress conditions; For temperature response, the range of values ​​is... This characterizes the degree to which temperature deviation from the optimum value inhibits light energy utilization. For moisture response, the range of values ​​is... This characterizes the degree to which moisture conditions inhibit light energy utilization. The mechanism is as follows: the amount of fixed carbon is equal to the product of the actual absorbed photosynthetically active radiation and the actual light energy utilization, the latter being obtained by subtracting the maximum light energy utilization from the two stress scalars of temperature and moisture.

[0116] The proportion of photosynthetically active radiation absorbed by vegetation is calculated using the following formula:

[0117] ;

[0118] in, This represents the vegetation index value at the current moment. The lowest quantile value corresponding to bare soil in the study area is taken as the 5th quantile of the full-pixel distribution over the entire time period; The high quantile value corresponding to dense vegetation is taken as the 95th percentile. Indicates truncation to The range is set to prevent exceeding the physical limits; the coefficient 0.95 represents the upper limit of the absorption ratio of a fully covered canopy, reflecting that even when the canopy is closed, there is still a small amount of radiation transmission or reflection. This formula is based on the approximately linear relationship between the vegetation index and the canopy absorption ratio within the medium coverage range, and uses the quantile values ​​at both ends for linear stretching; using quantile values ​​rather than absolute extreme values ​​can avoid the impact of abnormal pixels or residual clouds on the stability of the endpoints.

[0119] Temperature response is determined using a piecewise linear method: 0 is taken when the daily minimum temperature is below the lower limit temperature of the biome, indicating complete inactivation of photosynthetic organs due to low temperature; 1 is taken when it is above the lower limit of the optimum temperature; and linear interpolation is used when it is between the two. The lower limit temperature and the lower limit of the optimum temperature are taken from the herbaceous wetland entry (or the closest entry) in the biome parameter lookup table of the light energy utilization model. Moisture response is determined by the water level sequence: 1 is taken when the water level is within the suitable range; when it is above the upper limit, a reduction of 0.125 is applied for every 0.1 meter exceeding the upper limit, indicating root immersion and hypoxia, and a decrease in stomatal conductance; when it is below the lower limit, a reduction of 0.125 is applied for every 0.1 meter below the lower limit, indicating soil moisture deficit; the reduced value is not lower than 0. The suitable water level range is determined by the physiologically suitable range of the dominant vegetation, or by fitting the response relationship between historical water levels and the measured total primary productivity at the station.

[0120] With site Calibration is performed during the exposure period and growing season because the two stress scalars are close to 1 during this period, making this parameter most sensitive. Since total primary productivity has a linear relationship with it, the least squares solution can be analytically given, i.e., taking... and The regression slopes of the two sequences passing through the origin. Calibration does not use annual data reserved for consistency verification; results are applied according to eco-hydrological units.

[0121] Subtracting vegetation autotrophic respiration from total vegetation primary productivity yields the time series of net primary productivity of vegetation:

[0122] ;

[0123] in, Net primary productivity of vegetation represents the amount of carbon that vegetation net fixes and retains in the ecosystem. Autotrophic respiration refers to the amount of carbon consumed by vegetation to maintain life activities and build new tissues. Its maintenance and growth respiration coefficients are taken from the lookup table above. Explicitly deducting autotrophic respiration clarifies the definition of net primary productivity, and it is unnecessary to deduct it again when calculating the net carbon sink rate.

[0124] The vegetation scattering component is separated from the SAR backscattering time series corresponding to the vegetation structure sensing dimension, and the spatial distribution of the vegetation net primary productivity time series among the grid cells is corrected based on the vegetation scattering component. The time series read here includes the misaligned observation entries registered in step S2. These entries participate in the calculation directly according to the original sensing dimension without being converted.

[0125] The light energy utilization model primarily uses optical vegetation indices as input, but its ability to characterize inter-pixel biomass differences is particularly limited in dense areas where vegetation indices are close to saturation. SAR backscattering, on the other hand, is sensitive to volume scattering and is less prone to saturation, providing independent biomass constraints. This embodiment uses a water cloud model to separate the two scattering components:

[0126] ;

[0127] ;

[0128] ;

[0129] in, This is the total backscattering coefficient of the pixel, i.e., the observed value; The vegetation scattering component originates from volume scattering within the canopy. This represents the soil scattering component, which is scattered from the soil surface below the canopy. The vegetation layer two-way attenuation factor characterizes the proportion of microwaves attenuated after passing through the canopy once and round trip, with a value range of [value missing]. The denser the canopy, the smaller the value; For vegetation description parameters, leaf area index or aboveground biomass is used; The radar incident angle; The volume scattering efficiency coefficient; This represents the attenuation coefficient. The physical picture is as follows: the canopy is considered a uniform scattering body, and the total echo is composed of the superposition of scattering from the canopy and the soil scattering after two-way attenuation. This indicates the energy share involved in volume scattering. Correcting the effect of the incident angle on the effective scattering cross section.

[0130] and Calibration is achieved through field measurements in several sample plots. This involves simultaneously measuring aboveground biomass and corresponding pixel backscattering coefficients in several sample plots, substituting the measured biomass into the above formula, and solving for the minimum mean square error between the simulated and observed values. If no sample plots are available for measurement, literature values ​​from similar emergent herbaceous wetlands are used. After calibration, vegetation description parameters are deduced from the observed total backscattering coefficients, thus obtaining the estimated biomass for that pixel.

[0131] Based on this, space reallocation is performed: ;

[0132] in, For the first Net primary productivity of vegetation after pixel correction; To correct the previous value; The vegetation description parameter obtained by inverse solution of this pixel is the same physical quantity as in the previous formula; For all valid pixels in the region The mean of the values ​​is used as a normalization factor to ensure that the total amount in the region remains unchanged. The spatial heterogeneity moderating index has a range of values. The value of 0 indicates no correction, and 1 indicates a direct proportionality between the two. This formula adjusts high-biomass pixels upwards and low-biomass pixels downwards while keeping the total amount unchanged, thus improving the accuracy of spatial distribution without altering the rate of quantification. In a multi-site scenario, the result is determined by regression analysis of measured biomass and measured total primary productivity at each site; in a single-site scenario, it is determined by plot-scale data, i.e., by simultaneously measuring aboveground biomass and net primary productivity (obtained by the harvest method or biomass increment method) in several plots, taking the logarithm of both, and then performing a linear regression; the slope of the regression is the result. When no sample plots were measured, the default value of 0.5 was used, which corresponds to the increase of net primary productivity with the square root of biomass, reflecting the decrease in carbon sequestration efficiency per unit biomass due to self-shading in high biomass canopies.

[0133] Following the example, the peak photosynthetically active radiation of the pixel where the flux tower is located during the 2022 growing season was approximately 38 moles per square meter per day, and the vegetation index was 0.79. , ,have to The lowest temperature during this period was 18.6 degrees Celsius. The lower limit temperature from the table is [value missing]. The optimal temperature is set to 21 degrees Celsius, and the lower limit of the optimum temperature is taken as 21 degrees Celsius. The water level is 0.24 meters; the suitable range is 0 to 0.20 meters, and exceeding it by 0.04 meters yields... After the rate was determined Take 1.05 grams of carbon per mole, substitute it into the equation, and you will get... Carbon per square meter per day (g / m² / day); after deducting the estimated autotrophic respiration of approximately 12.5 g / m² / day (calculated using a lookup table), the net primary productivity before correction is approximately 18.2 g / m² / day. The vegetation description parameter for this pixel, obtained through inverse solution from the water cloud model, is approximately 12% higher than the regional average. Since there is no actual measured data for this example plot,... Take 0.5, the correction factor is approximately The corrected figure is approximately 19.3 grams of carbon per square meter per day. All net primary productivity figures for vegetation cited below are corrected values.

[0134] Before retrieving the carbon sink component time series, soil organic carbon pool data of the target wetland is obtained. The process of retrieving soil heterotrophic respiration includes: extracting soil temperature series from hydrological and meteorological observation data.

[0135] This analysis uses soil heterotrophic respiration, rather than total soil respiration, as the inversion object. This is because root autotrophic respiration, which is part of soil respiration, has already been deducted as vegetation autotrophic respiration in the calculation of net primary productivity. Including total soil respiration would result in double deduction and a systematic underestimation of the net carbon sequestration rate. Clarifying this approach ensures that the three components do not overlap when added or subtracted.

[0136] Soil scattering components are separated from the SAR backscattering time series corresponding to the vegetation structure perception dimension, and the soil moisture state sequence is determined based on the soil scattering components.

[0137] The soil scattering component is obtained from the water cloud model separation. Since the backscattering intensity increases with increasing dielectric constant, and the dielectric constant monotonically increases with water content, soil water content can be inverted from this component using empirical dielectric models for bare soil (such as the Oh model or the Dubois model). The soil moisture state sequence is obtained by arranging the samples according to their sampling time. This sequence characterizes the degree to which the pores of the surface soil are filled with water, and determines the soil aeration and microbial activity.

[0138] The temperature response of soil organic carbon decomposition was determined based on soil temperature sequences, the water response of soil organic carbon decomposition was determined based on soil moisture state sequences and flooding state detection results, and the soil heterotrophic respiration time series was determined by combining soil organic carbon pool data. ;

[0139] in, For soil heterotrophic respiration, it represents the amount of carbon released per unit area per unit time by microorganisms decomposing soil organic matter; The decomposition rate constant is at the reference temperature; The current soil temperature; For reference temperature, take 10 degrees Celsius; This is the temperature sensitivity coefficient, which characterizes the factor by which the decomposition rate increases for every 10 degrees Celsius increase in temperature; For moisture response, the range of values ​​is... ; This represents the soil organic carbon pool. The formula is a first-order kinetic form, meaning that the decomposition rate is directly proportional to the amount of decomposable substrate. The power term describes the exponential increase in microbial activity with temperature, and the moisture response describes the inhibition at both ends: too low a moisture level restricts the diffusion of substrates and enzymes, while too high a moisture level restricts the diffusion of oxygen.

[0140] in, The solution is obtained by inversely solving the soil respiration observation sequence measured by the static box method. Specifically, the measured values ​​during the exposure reference period are taken together with the soil temperature, inverted water content and organic carbon pool storage during the same period and substituted into the above formula to solve the problem. The median of the solution at each time point is taken. For units without stations, the literature values ​​of similar wetlands are taken and marked as low confidence. The logarithm of soil respiration measured in a static chamber is used to determine the soil temperature, followed by linear regression. The slope is multiplied by 10 and then the exponent is calculated. Due to the altered decomposition pathway under anaerobic conditions, the submerged and exposed states should be calibrated separately. If no submerged state data is available, the default value of 2.0 is used and marked as low confidence. The above calibration also does not use data from the reserved validation year.

[0141] The water response is taken as close to 1 when soil moisture content is moderate, and as the anaerobic inhibition stage is taken when the soil is nearly saturated, i.e., the water decreases to 0.1 to 0.3 times the maximum aerobic value, or the function form of similar studies is used. This is based on the fact that flooding fills pores with water, reduces the oxygen diffusion coefficient by several orders of magnitude, strongly inhibits aerobic microorganisms, and decomposition is switched to the much less efficient anaerobic pathway. Introducing flooding state detection results here as a basis for determining flooding avoids the error of judging saturation state solely based on inverted water content—the soil scattering component itself is strongly altered by the water surface during subsurface flooding, and directly triggering the anaerobic inhibition stage with bi-branch detection results is more reliable. Therefore, flooding state detection is not only used for staged analysis but also directly participates in the mechanism switching of carbon release processes.

[0142] For open water surface grid cells identified by flooding status detection, the corresponding soil heterotrophic respiration is set to zero. Since there is no exposed soil on the surface of these cells, soil heterotrophic respiration is no longer a component of their carbon dioxide balance; the anaerobic processes of the sediments beneath them and water surface escape are not within the scope of this invention's evaluation.

[0143] Following the example, the soil organic carbon pool in the pixel where the flux tower is located contains 4.6 kg of carbon per square meter, obtained from the median of the inverse solution of 38 static chamber measurements during the 2022 exposure reference period. The daily value is 0.00072. During the peak growing season, the soil temperature was 21.5 degrees Celsius, the moisture content was 0.34 cubic meters per cubic meter, and the moisture response was taken as 0.93. Carbon per square meter per day. During the flooding period from day 98 to day 146, this pixel was determined to be sub-canopy flooded, with a moisture response of 0.15 for the anaerobic inhibition segment and a soil temperature of 12.8 degrees Celsius. Carbon per square meter per day reduces the availability of good oxygen conditions by about 90%.

[0144] The inversion of hydrological carbon transport includes: extracting precipitation and water level sequences from hydrological and meteorological observation data, determining the time period of precipitation events based on the precipitation sequences, and determining the time series of hydrological carbon transport based on the water level sequence and the time series of submerged pixels within the precipitation event time period.

[0145] The duration of a precipitation event is determined by a continuous period of precipitation exceeding a set threshold. Precipitation events are used as the trigger because lateral carbon transport is primarily driven by precipitation runoff, while periods without precipitation represent only a very small baseflow term. The duration of the event is calculated using the following formula: ;

[0146] in, Hydrological carbon transport is defined as the amount of carbon that migrates laterally with water flow per unit wetland area per unit time. The cross-sectional flow rate is calculated from the water level sequence using the water level-flow rate curve. The carbon concentration in the water body is determined by sensor measurements or by empirical regression. The wetland area participating in the exchange is the product of the proportion of submerged pixels at that moment and the total wetland area. This formula indicates that the amount of carbon migrating with the water flow is equal to the product of the volumetric flux and the carbon concentration, and then reduced to a unit area based on the area participating in the exchange, consistent with the other two components.

[0147] The water level-flow rate relationship curve is calibrated by regression of the measured flow rate at the cross section and the synchronous water level. That is, the flow rate is obtained by simultaneously measuring the flow velocity and the cross-sectional area of ​​the water flow at several typical water levels, and then a power function regression is performed on the water level and the flow rate. This curve is only applicable within the water level range covered by the calibration. For water levels outside the range, the curve is truncated according to the boundary value and marked with low confidence to avoid non-physical extrapolation. The empirical regression used measured carbon concentration from previous precipitation events as the dependent variable and precipitation amount and number of days without precipitation prior to the event as independent variables in a multiple regression. The rationale was that precipitation amount determines scouring intensity, and the number of days without precipitation prior to the event determines the amount of carbon that can be accumulated through scouring. The transport direction was determined by the flow direction from the cross-section to the sensor or the direction of the water level gradient inside and outside the wetland.

[0148] Specifically, hydrological carbon transport from the target wetland to external water bodies is taken as positive values, while hydrological carbon transport from external water bodies to the target wetland is taken as negative values. This convention aligns the numerical direction with the carbon source direction, facilitating subsequent uniformity through subtraction.

[0149] Using the example, during the flooding period from day 98 to day 146 in 2022, four precipitation events occurred, with cumulative precipitation of 12, 23, 9, and 31 mm, respectively. Among them, the event from day 131 to day 133 had a cumulative precipitation of 31 mm, with the water level rising from 0.41 meters to 0.79 meters, which translates to a peak cross-sectional flow of 2.6 cubic meters per second. During the same period, the measured carbon concentration in the water was 11.4 milligrams of carbon per liter, with a flooded pixel ratio of 0.68 and a corresponding exchange area of ​​8.16 square kilometers. Substituting these values, the peak hydrological carbon transport for this event is approximately 0.31 grams of carbon per square meter per day. Based on the water level gradient, this is determined to be an outward output and is taken as a positive value. During the normal period without precipitation, the peak carbon transport is approximately 0.05 grams of carbon per square meter per day.

[0150] The above three components together constitute the carbon sink component time series, all of which are carbon dioxide caliber.

[0151] Step S5: At the flux station, according to the ecological state interval, perform cross-correlation analysis between the hydrological or temperature series after seasonal term removal and the corresponding carbon flux component series to estimate the lag time of each carbon sink component, and transfer the lag time to each grid cell.

[0152] The responses of various carbon processes to environmental drivers are not instantaneous: after a drop in water level, it takes several to ten days for the reed root system to reflect changes in water state in photosynthetic rate; microbial activation and substrate enzymatic hydrolysis also require time after soil temperature rises; and the runoff accumulation and carbon dissolution and migration after precipitation also involve a propagation time. Ignoring these delays and directly superimposing them according to calendar time will cause the peak and trough positions of each component to be misaligned, and the zero-crossing point of their algebraic sum to deviate from the true transition time. The result of this step directly determines the accuracy of subsequent critical point positioning.

[0153] This step follows a key constraint: each set of reference sequences and carbon flux component sequences used for cross-correlation analysis is formed from different original observations. If they share a structural source, spurious components will be introduced into their correlation, and the estimated lag will no longer reflect the true physical delay. More importantly, the lag duration is estimated solely from ground-based measured sequences; the carbon sink component time series retrieved from remote sensing in step S4 is not included in the estimation—this component time series itself is generated driven by vegetation indices and meteorological sequences, and the results obtained using it as the response sequence only reflect the internal transmission relationships of the inversion model, constituting a circular argument. Thus, the lag duration becomes an external parameter supported by independent measured evidence, supplementing the inversion link with temporal information that the model itself cannot provide.

[0154] Therefore, carbon flux observation data includes vegetation carbon absorption observation sequences, soil respiration observation sequences, and hydrological carbon transport observation sequences; hydrological sequences, temperature sequences, and precipitation event sequences are extracted from hydrological and meteorological observation data, and these two sets of independent observations are analyzed in pairs by interval. The cross-correlation calculation formula is:

[0155] ;

[0156] ;

[0157] in, As a reference sequence, The sequence represents the carbon flux components. , This represents the mean of the two sequences within the window; To test the lag; For lag Cross-correlation coefficient at time, range of values ; The lag amount that maximizes its value, i.e., the estimated lag duration; The upper limit of the search is defined as the product of the L2 norms of the anomalies of the two sequences. This makes the coefficients independent of their magnitude and only reflect the similarity of the waveforms. The shift amount is the optimal alignment of the two sequence waveforms when it exactly cancels out the actual delay, and the coefficients are maximized. This shift amount is the lag time, which physically means the time interval between the occurrence of environmental driving events and the generation of carbon flux response.

[0158] The search range is constrained by non-negative lag. The interval length is set as a limit because the effective sample size decreases sharply and the estimation becomes unreliable when the lag approaches the interval length. A 60-day absolute upper limit is used because the response delay of various carbon processes should not physically exceed this order of magnitude; longer optimal lags are usually spurious peaks caused by seasonal cycle residuals. The basis for limiting non-negative lags is causality. As a robustness check, if the peak value falls within the entire search range including negative lags... At this point, it is judged as an indication of reverse causality (such as vegetation transpiration driving water level), and taken... Mark low confidence and record pending investigation.

[0159] Before performing cross-correlation, seasonal terms need to be removed: water level, temperature, and carbon flux are all driven by the same interannual climate rhythm, and the original sequences are already highly correlated. Their peak values ​​may only reflect the phase difference between the two seasonal curves rather than causal delays. Specifically, the anomaly sequence is obtained by subtracting the multi-year daily climatological mean from each sequence, and then the calculation is performed on the anomaly sequence to remove the hidden variable of the common seasonal cycle.

[0160] Within this framework, estimating the lag time of each carbon sink component specifically includes:

[0161] During the exposure period, seasonal term removal was performed on the hydrological and vegetation carbon uptake observation sequences, and the lag duration of vegetation net primary productivity was determined based on the cross-correlation analysis results. Here, the measured water level sequence is used as the reference sequence (if no actual water level measurements are available, the submerged pixel ratio time series can be used as a substitute). For the response sequence, the offset corresponding to the maximum cross-correlation value is taken as... The exposure period was chosen because vegetation was active during this period, water conditions had the most significant effect on photosynthesis, and the signal-to-noise ratio was the highest. The two sequences were from the water level gauge and the eddy covariance system, respectively, and their sources did not overlap. Characterizes the time interval between changes in hydrological conditions and the generation of vegetation carbon uptake response.

[0162] During the transition period, seasonal term removal was performed on both the temperature and soil respiration observation sequences, and the lag time of soil heterotrophic respiration was determined based on the cross-correlation analysis results. Here, the soil temperature sequence was used as the reference sequence, and the measured soil respiration from the static chamber was used as the response sequence; the offset was taken as... The transition period was chosen because it involves alternating wet and dry conditions, with large and frequent fluctuations in temperature and respiration rate, making it easier to distinguish delays. The static chamber and temperature sensor are two completely independent systems. Switching to the static chamber is only permitted when no static chamber data is available. However, since this sequence was constructed using the nighttime temperature response splitting method and shares temperature information with the reference sequence, its independence is limited and it must be labeled as low confidence. Characterizes the time interval between changes in temperature and substrate and the emergence of a soil respiration response.

[0163] During the inundation period, the lag time of hydrological carbon transport was determined based on the cross-correlation analysis results of precipitation event sequences and hydrological carbon transport observation sequences. Here, the measured precipitation event time series was used as the reference sequence, and the measured lateral carbon transport was used as the response sequence. Impulse cross-correlation analysis was performed on the event baseline; that is, response sequence segments were extracted with the start time of each event as the zero point, and the average was superimposed. The offset corresponding to the maximum value was taken as the threshold. The flooding period was chosen because the water bodies are interconnected and transport paths are unobstructed during this period, and the precipitation-runoff-carbon transport chain is most complete. The event benchmark uses discrete events as alignment anchors, which already avoids interference from seasonal cycles, so no additional seasonal term removal is required. Characterizes the time interval between the occurrence of a precipitation event and the generation of a hydrological carbon transport response.

[0164] Window length not less than The lower limit is the beat frequency period, because cross-source statistics must cover at least one complete sampling phase drift cycle. When the interval is too short, adjacent intervals of the same type are merged; when merging discontinuous intervals, cross-correlation summation is performed only within each interval and does not cross boundaries, i.e., the numerator and denominator terms are accumulated separately for each interval before merging, to avoid meaningless pairings across discontinuities. Coefficients must pass a significance test (…). If the above conditions are met, the median of the historical estimates for that type of interval within the same eco-hydrological unit shall be used.

[0165] In parameter transfer, station estimates are assigned to each grid cell. This is feasible because cells within the same eco-hydrological unit have similar hydrological connectivity, vegetation composition, and soil conditions, resulting in consistent response delays that are approximately constant within the same interval. Flux station observations represent the average state of their upwind source region, which typically spans hundreds of meters and covers tens to hundreds of cells. Unit division should ensure consistency in internal hydrological connectivity, vegetation community type, and soil texture, and that the station's source region falls within it. This can be determined by spatial clustering results of the submerged cell proportions combined with a vegetation type map. For multi-site coverage, a distance-weighted average is used; for units without stations, a regional prior value is used and marked with low confidence. This prior value is the median of the lag times reported in published studies of wetlands with the same climate zone and vegetation type. Furthermore, cross-validation can be performed using a one-year retention method to compare the dispersion of each estimate and generate confidence markers output with the parameter set, providing a quantitative explanation of the reliability of the conclusions.

[0166] Following the example, during the exposure period from day 172 to day 365 in 2022, the water level anomaly and The cross-correlation coefficient of the anomaly in The maximum value is 0.71 at the sun. Therefore As a comparison, if the seasonal term is not removed and the calculation is performed directly on the original sequence, the peak occurs at [date]. The coefficient reached 0.88, indicating that the result actually reflects the phase difference between the two seasonal curves rather than the true delay. The two transition periods, from day 73 to 97 and from day 147 to 171, were each only 25 days long, with insufficient sample size. Following the rules, these were merged into a 50-day window, and summation was performed separately within each interval. The coefficients of soil temperature anomaly and static chamber soil respiration anomaly were... The maximum value is 0.64 at the sun. Therefore Day. During the flooding period from day 98 to day 146, a superimposed epoch pulse cross-correlation analysis was performed using four precipitation events. The maximum value occurred at... day, so Day. Cross-validation over one year shows... The two-year estimates are for the 6th and 7th respectively, marked with high confidence. The values ​​are 2 days and 4 days, respectively, and are marked as medium confidence. The three lag times are transmitted to all pixels within the eco-hydrological unit.

[0167] Step S6: Perform phase correction on the time series of each carbon sink component using the lag duration to obtain a synchronized carbon sink sequence;

[0168] The following are all times without subscript. All refer to the observation time on the calendar timeline, with subscripts. Both refer to the ecological driving moment on the ecological driving timeline, and the two are related by the lag time of each component.

[0169] Specifically, for each sampling moment in the time series of each carbon sink component, the lag time of the corresponding carbon sink component is subtracted from the sampling moment to obtain the corresponding ecological driving moment:

[0170] ;

[0171] ;

[0172] ;

[0173] in, The ecological driving moment represents the actual moment when the environmental driving force that causes the response of this component occurs. , , These are the synchronized net primary productivity of vegetation, synchronized heterotrophic respiration of soil, and synchronized hydrological carbon transport at that time. , , The time sequence of the three components is marked according to the observation time; , , These are the three lag durations estimated in step S5. The three equations indicate that the observation time is... The component value is driven by the actual occurrence of Re-mark it to This means completing the translation of the observation time axis onto the ecologically driven time axis.

[0174] The scope of application for the lag parameter is defined as follows, to cover situations outside the estimation combination in step S5: net primary productivity of vegetation is used uniformly across all intervals. The delayed physiological response of vegetation to hydrological stress is an inherent property of vegetation and is universal across different regions; soil heterotrophic respiration is uniformly used across all activity zones. Open water surface pixels that have been set to zero are not mapped; hydrological carbon transport uses a unified method for precipitation-triggered events within any interval. During periods without precipitation, the baseflow term is either zero or negligible, and the mapping has no substantial impact. These rules ensure that each component has a definite value throughout the entire time period, without any gaps.

[0175] Using the ecological driving time series corresponding to vegetation net primary productivity as a unified time reference, soil heterotrophic respiration and hydrological carbon transport are converted to a unified time reference to obtain a synchronized carbon sink sequence.

[0176] Since the three lag durations are different, the times at which each component falls after translation do not coincide and cannot be directly added point by point. Therefore, the latter two are mapped to this benchmark through linear time-series interpolation. The time series corresponding to net primary productivity of vegetation is chosen as the benchmark because this component has the largest magnitude during the growing season and makes the most dominant contribution to the net carbon sink rate. Using its original sampling time as the benchmark can avoid introducing additional errors from interpolating the dominant component. The sampling time of this component's time series is inherited from the optical path; therefore, the median sampling interval of the unified time benchmark is 10 days in this example. Interpolation is performed only within the envelope of the observation times of each component and is not extrapolated. Items outside the envelope are marked as unavailable.

[0177] The synchronized carbon sink sequence includes synchronized vegetation net primary productivity time series, synchronized soil heterotrophic respiration time series, and synchronized hydrological carbon transport time series.

[0178] Following the example, the net primary productivity of vegetation on the 199th day of observation was 19.3 grams of carbon per square meter per day minus... This will be marked up to day 193 of the ecological driving timeline; the soil heterotrophic respiration value on day 196 of the observation period is 6.8 grams of carbon per square meter per day, minus... The same marking will continue until day 193; the value of hydrological carbon transport on day 195 of the observation time is 0.05 grams of carbon per square meter per day minus... This will also be marked up to day 193. After the three are aligned, a set of directly computable synchronization values ​​will be formed on day 193 of the ecological driving moment. At this time, the environmental driving corresponding to the three values ​​is indeed the same batch of conditions at the same moment.

[0179] Step S7: Construct instantaneous phenological phases using the phenological phase shift parameters extracted from the phenological phases. Use the starting boundary of each ecological state interval as the origin to determine the relative instantaneous phenological phase as the abscissa and the net carbon sink rate calculated from the synchronized carbon sink sequence as the ordinate to generate a dynamic carbon sink trajectory.

[0180] The inherent drawback of using calendar time as the horizontal axis is that the phenological processes corresponding to the same calendar day in different years may differ by more than ten days, which, when accumulated over many years, will cause the carbon sink characteristics to be diffused along the horizontal axis. This step changes the horizontal axis to use phenological processes, aligning the trajectories of different years according to the timing of phenological events.

[0181] First, construct the instantaneous phenological phases: ;

[0182] in, For a moment The instantaneous phenological phase; As the fundamental frequency, take ; For the phenological phase shift parameter sequence at time... The value of is obtained by linear interpolation along the centers of adjacent windows. This is based on the fact that the first harmonic component can be written as... The quantity within parentheses is the angle variable describing the current phase state of this component, corresponding to the seasonal peak when it is zero. The instantaneous phenological phase can be understood as a phase clock that advances over time: when phenology is stable, this parameter is approximately constant, and the phase increases at a rate... Proceed at a constant speed, progressing within one year When phenology advances or delays, this parameter changes accordingly, and its offset is included as a correction term, causing the phase to advance faster in years with advanced phenology and slower in years with delayed phenology. Therefore, the instantaneous phenological phase represents the current position of vegetation in its annual growth cycle, rather than its position in the calendar cycle.

[0183] Then, for any sampling time, the difference between the instantaneous phenological phase at that sampling time and the instantaneous phenological phase corresponding to the initial boundary of the ecological state interval to which that sampling time belongs is determined as the relative instantaneous phenological phase, i.e., the horizontal axis:

[0184] ;

[0185] in, This refers to the relative instantaneous phenological phase; The relationship between the observation time and the ecologically driven time is as follows: ; For a moment The sequence number of the interval to which it belongs; This is the starting boundary time of the interval; and This represents the instantaneous phenological phase and phenological phase offset parameter values ​​at that moment. Taking the initial boundary as the origin for each interval eliminates ambiguity in multi-boundary cases, ensuring that the trajectories of each interval are calculated from zero and are comparable. In phenologically stable years, this coordinate system degenerates into... This refers to the phase clock that advances uniformly from zero within the interval, ensuring that the horizontal axis does not degenerate into a constant; the interannual fluctuation of this parameter causes the trajectories of different years to automatically align according to the phenological time.

[0186] For sampling points that fall within the flooding period or the amplitude collapse period, the phenological phase shift parameter no longer represents the seasonal peak time because the vegetation index is suppressed by the water surface. At this time, the abscissa degenerates into... This means that only the uniform velocity advancement term is retained, excluding the offset correction term, and this segment of the trajectory is marked separately; the complete calculation formula is restored after the surface is exposed again and the first-order amplitude rises above the baseline of the exposed state amplitude. This treatment avoids non-physical lateral distortion of the trajectory caused by the phase parameter losing its phenological meaning. It should be noted that step S3 uses the first-order difference of this parameter to identify phase abrupt changes based on the fact that it jumps under disturbance, regardless of whether the absolute value still represents the peak time point; this step, however, relies on the phenological meaning of its absolute value, so the submergence period must be degraded. The two utilize the differential and absolute characteristics of this parameter respectively, which are not contradictory.

[0187] The horizontal axis is uniformly adopted The conversion between observation time and ecological driving time is completed. Since the unified time reference in step S6 is anchored by the synchronous mapping of vegetation net primary productivity, the coordinate definition is consistent with the reference definition, which can avoid systematic offset.

[0188] No. The range of x-coordinate values ​​for each interval is: ,in This represents the length of the interval. For multi-year overlays, normalized coordinates can also be provided: ;

[0189] in, To normalize the relative instantaneous phenological phase, the range of values ​​is... It represents the relative progress position of the current moment within its own interval, and is used for auxiliary display. The original radian coordinates are still used as the main coordinates.

[0190] Next, subtracting synchronized soil heterotrophic respiration and synchronized hydrological carbon transport from synchronized vegetation net primary productivity at the corresponding ecological driving moment yields the net carbon sink rate, i.e., the ordinate:

[0191] ;

[0192] in, The vertical axis is used as the coordinate. For Ecological Driven Moment The net carbon sink rate represents the amount of carbon absorbed (or released) by wetlands per unit area per unit time. Carbon accounting consistency is maintained because: vegetation autotrophic respiration has already been deducted from net primary productivity; here, only soil heterotrophic respiration and lateral carbon transport are deducted, avoiding double deduction; hydrological carbon transport is treated as positive for outward output and negative for inward input, as previously agreed, and is therefore uniformly treated as subtraction. A positive rate indicates net carbon sink, and a negative rate indicates net carbon source, corresponding to the convention in flux observations where negative values ​​of net ecosystem carbon exchange represent carbon sinks; the measurement is in carbon dioxide.

[0193] Finally, discrete sampling points Connecting points sequentially according to the ecological driving timeline creates a dynamic carbon sink trajectory, with each point representing the net carbon sink status of the wetland under a specific ecological driving phase. Multi-year data can be superimposed in the same phase space to compare carbon budget differences at the same phenological location; this comparison is unaffected by interannual phenological variations.

[0194] Following the example, the exposure period from day 172 to day 365 of 2022 is taken as day 172. Each interval The phenological phase shift parameter at that time was 3.380 radians. Therefore, the range of the horizontal axis is [blank]. Radius. For the sampling point on day 193 of the ecological driving event, the observation time... The phenological phase shift parameter at that time was 3.420 radians, which yielded... radian, ; ordinate A positive value of gram carbon per square meter per day indicates that the wetland exhibits significant net carbon sink characteristics at this phenological phase. Connecting the points in sequence according to ecological drivers yields the dynamic trajectory of carbon sink during this exposure period.

[0195] Step S8: Detect the sign transition moment of the net carbon sink rate in the carbon sink dynamic trajectory and output it as the critical point of carbon source-sink conversion.

[0196] Net carbon sink rates are arranged according to ecological driving moments. When the signs of net carbon sink rates corresponding to two adjacent ecological driving moments are different, the latter ecological driving moment is determined as the sign transition moment. The sign transition represents the wetland's transformation from a carbon source to a carbon sink or from a carbon sink to a carbon source, and is a node with clear physical significance in the evolution of carbon balance. The critical points obtained from this detection include two directions: the carbon source to carbon sink critical point corresponding to the net carbon sink rate changing from negative to positive, and the carbon sink to carbon source critical point corresponding to the net carbon sink rate changing from positive to negative. These correspond to the moment when vegetation carbon sequestration exceeds soil decomposition during the growing season and the moment when carbon sequestration is insufficient to offset decomposition at the end of the growing season, respectively. The output is marked separately for each direction so that the occurrence time and interannual variation can be statistically analyzed separately for each direction.

[0197] Considering that the inversion and interpolation residuals may cause spurious flips in the net carbon sink rate when it is close to zero, a moving average is applied before traversal, and the window length is taken as the median of the sampling interval with a uniform time base to suppress high-frequency noise. At the same time, an amplitude condition is applied to the sign conversion.

[0198] ;

[0199] in, This is the set of critical points for carbon source-sink conversion; For the first on the trajectory The horizontal and vertical coordinates of each sampling point; This indicates that two adjacent points have opposite signs and that there is a zero-crossing point within the interval. For conversion intensity, according to The calculation characterizes the magnitude of the transition between carbon source and sink states; the larger the value, the more significant the transition. As the noise threshold, take , This represents the synthesis uncertainty of the synchronization sequence. The purpose of setting the amplitude condition here is to establish a lower limit for screening above the order of magnitude of the inversion residuals and interpolation residuals, thus preemptively eliminating spurious flips caused by the residuals near zero net carbon sink rate, and avoiding misjudging noise-induced zero crossings as carbon source-sink conversions. The basis for taking three times the synthesis uncertainty is that when the residuals are approximately normal, jumps with amplitudes less than this threshold are more likely to be caused by random noise; excluding them can suppress spurious conversions while preserving the true state changes. It should be noted that this amplitude condition is a necessary screening condition, not a sufficient criterion for true conversions: when the residual distribution deviates from normality, or when there are systematic biases that the inversion model and empirical regression fail to cover, jumps with amplitudes exceeding the threshold may still be caused by residual errors. Therefore, in actual implementation, this condition is only used to determine candidate critical points. Candidate critical points are further confirmed by the following cross-verifiable properties: whether the phase position of the same pixel at the recurring transition time over many years is consistent; whether the transition time matches the independent verification year flux tower measured sequence within one SAR revisit cycle; and whether this time corresponds temporally to the submergence state transition and vegetation index amplitude change. The above confirmation results are output along with each critical point at confidence levels. The combined uncertainty is estimated by the following formula:

[0200] ;

[0201] in, , , These are the root mean square residuals of the three component inversion values ​​relative to the measured values ​​of the same caliber at the flux stations; The residual standard deviation of the linear interpolation in step S6 is obtained by performing leave-one interpolation on known sampling points, and the difference between the statistical interpolation value and the measured value. The four error sources are independent of each other and are synthesized according to the root of square. The synthesized uncertainty level of the synchronized sequence under the current inversion and interpolation link is characterized, and the threshold set accordingly is automatically adjusted according to the data quality: when the error is large, the threshold is raised accordingly and the output is more conservative; when the error is small, the threshold is lowered accordingly and the true conversion with a smaller amplitude can be retained.

[0202] The phase position of the critical point is synchronously mapped. By mapping back to the calendar timeline and combining it with the time markers in the ecological status stages, the dynamic assessment results of wetland carbon sequestration can be obtained, including the transition time and phase position. and The output includes the conversion intensity, conversion direction (carbon source to carbon sink or carbon sink to carbon source), and confidence markers from the cross-validation in step S5. The output also provides the calendar date of the conversion and its position in the phenological process, which can be used for both annual calculations and cross-year phenological comparisons.

[0203] The above detection is performed pixel-by-pixel to obtain the spatial distribution of critical points. When area-scale conclusions are needed, the results are aggregated on the ecological driving timeline. That is, the net carbon sink rate of each pixel is aligned according to its respective ecological driving time, weighted and summed by pixel area, and then the same sign conversion detection is performed on the sequence. Since the starting boundaries of each pixel interval are not exactly the same, the abscissa of the area-scale sequence takes the overall interval starting boundary determined by the time series of the proportion of submerged pixels in the whole wetland as the origin, and constructs the instantaneous phenological phase using the average phenological phase shift parameter of the whole wetland, so that the area-scale trajectory has a unified abscissa definition. Soil heterotrophic respiration of open water surface pixels is set to zero, and its net carbon sink rate is composed only of vegetation net primary productivity and hydrological carbon transport. When summarizing, it participates in the weighting according to the actual area proportion, and the open water surface area proportion is reported separately in the results to explain the scope.

[0204] To verify the reliability of the results, consistency verification can be performed by comparing data with similar calibers. Calculate the synchronized net ecosystem productivity sequence: ;

[0205] in, To synchronize net ecosystem productivity, i.e., to exclude lateral carbon transport from the net carbon sink rate, thus aligning with flux tower observations, the sign transition time and the actual flux tower measurements at the independent validation year (or reserved validation site) are used. The sign transition time of the sequence is compared, where Net ecosystem carbon dioxide exchange; negative values ​​represent carbon sinks, therefore... and The caliber and notation conventions are the same. A time difference within one SAR revisit cycle is considered a pass; if it fails, the bimodality test of the inundation detection is checked back sequentially to see if it triggers a regression, whether the transition period half-width covers the true boundary, and the significance level and interannual dispersion of the lag estimate. In addition, the systematic difference between the net carbon sink rate transition time and the synchronized net ecosystem productivity transition time, which is the contribution of lateral carbon transport, is reported separately, giving this often overlooked pathway an explicit quantitative position in the conclusions.

[0206] The aforementioned consistency verification also serves as a second layer of confirmation beyond the amplitude threshold: in addition to passing the candidate critical point of the amplitude condition, it must also be supported in at least one of the three aspects of multi-year repeatability, comparison with the same caliber, and ecological state grading before it can be output with a higher confidence level. If it is not supported, it is only considered as a candidate result and marked with a lower confidence level, thereby avoiding confirmation of conversion based solely on the amplitude criterion.

[0207] Following the example, the trajectory of the cell containing the flux tower in 2022 showed a sign transition twice, once in spring and once in autumn. (From...) , , , Carbon per square meter per day, synthesized from the formula and root Carbon per square meter per day, noise threshold Carbon per square meter per day.

[0208] The net carbon sequestration rates of two adjacent points at the spring transition point are respectively and Carbon per square meter per day, product is negative, conversion intensity The critical point is defined as when carbon per square meter per day exceeds a certain threshold. The reason this period is a net carbon source is as follows: During the transition period from day 147 to day 171, the reeds are still in the early stages of greening up, and the vegetation's net primary productivity is only about 3.2 grams of carbon per square meter per day. However, after the floodwaters recede, the soil enters an aerobic state, and soil heterotrophic respiration rises to about 4.1 grams of carbon per square meter per day. The two offset each other, resulting in a net carbon source. As the canopy expands, net primary productivity rapidly surpasses soil heterotrophic respiration, thus becoming a net carbon sink. This critical point, driven by ecological factors, occurs on day 168. The Japanese counter-reaction was dated June 23, 2022; among which... The radian is obtained by linear interpolation of the centers of adjacent windows, and substituting it yields... radian, Located at the beginning of the exposure period; confidence markers inherited from High confidence markers.

[0209] The two adjacent points at the autumn transition point are respectively and Carbon per square meter per day, conversion intensity The carbon per square meter per day (g / m² / day) also exceeds the threshold, confirming a critical point. This is because the reeds enter their yellowing stage, net primary productivity drops to approximately 1.8 g / m² / day, while soil temperature remains high and heterotrophic respiration remains at approximately 4.1 g / m² / day. This point corresponds to the 265th day of the ecological driving event, which in turn corresponds to the 271st day of the calendar, September 28, 2022. radian, It is located in the latter part of the middle of this exposure period.

[0210] Consistency validation used 2023 as the independent validation year. Flux tower measurements for that year were not used for maximum light energy utilization calibration, decomposition rate constant and temperature sensitivity coefficient calibration, or lag estimation. The spring sign transition time for the synchronized net ecosystem productivity sequence in that year was day 165. The sequence is dated day 163, with a time difference of 2 days, which does not exceed the SAR revisit period of 6 days, and the verification is successful. The spring transition time of the net carbon sink rate in the same year is day 168, which is 3 days later than the previous day. This systematic difference is the contribution of lateral carbon transport and will be reported separately.

[0211] In contrast, if the phase correction in step S6 is not performed and the three components are directly superimposed using the observation time, the zero-crossing point in spring 2022 will occur on the 181st day, 13 days later than the corrected 168th day. If the 21-day lag estimated from the original sequence is used without further seasonal term removal, the zero-crossing point will shift further. It is evident that explicit correction of component lag differences has a substantial impact on the positioning of the conversion time.

[0212] In summary, this embodiment uses asynchronous fusion constrained by beat frequency period to allow two types of observations to coexist in the same grid according to the perception dimension without performing time alignment and feature transformation. This preserves the temporal sharpness of state abrupt changes while avoiding cross-source conversion errors. By cross-verifying with sliding window harmonic fitting and SAR dual-branch inundation detection, ecological state intervals are robustly divided, enabling phenological rhythms and hydrological rhythms to be uniformly represented within the same periodization framework. Phase correction is performed using the component lag duration estimated from independent ground-based measured sequences, eliminating time mismatches between components and avoiding circular reasoning from a mechanistic perspective. Finally, a dynamic carbon sink trajectory is generated using the relative instantaneous phenological phase as the abscissa, so that the location of the carbon source-sink conversion critical point corresponds to both wetland state changes and carbon budget changes.

[0213] In addition to being used for dynamic monitoring of wetland carbon sinks, this method can also be used in various applications that require understanding the temporal position of wetland carbon balance, including wetland restoration effectiveness assessment, wetland carbon sink project measurement and verification, and evaluation of the effectiveness of ecological water replenishment and water level control measures.

[0214] In assessing the effectiveness of wetland restoration, steps S1 to S8 are performed for both the pre-restoration and post-restoration periods. The results of the two assessments are compared to evaluate the degree of recovery of wetland carbon sink function. The compared results include: the magnitude of the net carbon sink rate and its duration during the growing season, the length of the net carbon source period, the calendar time of the carbon source-carbon sink transition critical point, and the phase position of this critical point within its respective ecological state interval. The determination is based on the fact that the abscissa obtained in step S7 takes the initial boundary of each ecological state interval as the origin, so that the dynamic trajectory of carbon sink in each year before and after restoration falls in the same phenological reference system and can be directly compared. Therefore, when the phase position of the carbon source-carbon sink critical point moves forward towards the start of the interval and the phase position of the carbon sink-carbon source critical point moves backward towards the end of the interval, it indicates that the wetland's transition from carbon source to carbon sink is advanced and the transition from carbon sink to carbon source is delayed, meaning that the net carbon sink function during the growing season has been restored. The degree of restoration is comprehensively characterized by the amount of movement of the above two phase positions and the magnitude of change in the net carbon sink rate.

[0215] Furthermore, the changes in hydrological conditions reflected by the time series of submerged pixel proportions can be compared with the aforementioned carbon sequestration function indicators to determine whether the restoration effect mainly comes from improved hydrological conditions or vegetation restoration: changes in the boundary between the submerged and exposed periods, as well as changes in the magnitude of the submerged pixel proportions and the duration of flooding, along with synchronous changes in the net carbon sequestration rate, indicate that the restoration is mainly driven by improved hydrological conditions; while the hydrological state indicators remain basically unchanged but the net carbon sequestration rate increases, indicate that the restoration is mainly driven by vegetation restoration.

[0216] When used for multi-year restoration monitoring, data from each year before and after restoration can be superimposed and compared in the same phase space to eliminate the interference of interannual climate differences on the judgment of restoration effectiveness.

[0217] Example 2:

[0218] like Figure 3 As shown, the wetland carbon sequestration dynamic assessment system based on multi-source remote sensing time-series fusion includes:

[0219] The data acquisition module is used to acquire optical image sequences, SAR image sequences, carbon flux observation data and hydro-meteorological observation data of flux stations within the target wetland.

[0220] The asynchronous fusion module is used to determine the beat frequency period based on the sampling time interval of the optical image sequence and the SAR image sequence, and to perform asynchronous fusion of the two types of images with the beat frequency period as a constraint to generate a fusion mesh containing pigment perception dimension and vegetation structure perception dimension.

[0221] The ecological state classification module is used to extract vegetation index time series and SAR backscatter time series from the fused grid, and perform phenological phase extraction and flooding state detection accordingly, dividing the vegetation growth cycle into multiple ecological state intervals.

[0222] The carbon sink component inversion module is used to invert the time series of carbon sink components based on fused grid and hydro-meteorological observation data. The carbon sink components include vegetation net primary productivity, soil heterotrophic respiration, and hydrological carbon transport.

[0223] The lag analysis module is used at flux stations to perform cross-correlation analysis between hydrological or temperature sequences processed by seasonal term removal and the corresponding carbon flux component sequences according to ecological state intervals, to estimate the lag duration of each carbon sink component, and to transmit the lag duration to each grid cell.

[0224] The phase correction module is used to perform phase correction on the time series of each carbon sink component using the lag time to obtain a synchronized carbon sink sequence.

[0225] The trajectory generation module is used to construct instantaneous phenological phases using phenological phase offset parameters extracted from phenological phases, determine the relative instantaneous phenological phases as the abscissa using the starting boundary of each ecological state interval as the origin, and use the net carbon sink rate calculated from the synchronized carbon sink sequence as the ordinate to generate a dynamic carbon sink trajectory.

[0226] The critical point output module is used to detect the sign transition moment of the net carbon sink rate in the dynamic trajectory of the carbon sink, and output it as the critical point of carbon source-sink conversion.

[0227] The above description is merely an example and illustration of the structure of the present invention. Those skilled in the art can make various modifications or additions to the specific embodiments described, or use similar methods to replace them, as long as they do not deviate from the structure of the invention or exceed the scope defined in the claims, all of which should fall within the protection scope of the present invention.

[0228] In the description of this specification, references to terms such as "an embodiment," "example," "specific example," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0229] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to any specific implementation. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention. The invention is limited only by the claims and their full scope and equivalents.

Claims

1. A dynamic assessment method for wetland carbon sequestration based on multi-source remote sensing temporal fusion, characterized in that, Includes the following steps: Acquire optical image sequences and SAR image sequences of the target wetland, as well as carbon flux observation data and hydro-meteorological observation data of flux stations within the target wetland; The beat frequency period is determined based on the sampling time interval between the optical image sequence and the SAR image sequence. The two types of images are then asynchronously fused using the beat frequency period as a constraint to generate a fused mesh that includes pigment perception dimension and vegetation structure perception dimension. The vegetation index time series and SAR backscattering time series are extracted from the fused grid, and phenological phase extraction and flooding state detection are performed accordingly to divide the vegetation growth cycle into multiple ecological state intervals. Based on the fused grid and the hydrological and meteorological observation data, the carbon sink component time series was retrieved, which includes vegetation net primary productivity, soil heterotrophic respiration and hydrological carbon transport. At the flux station, according to the ecological state interval, cross-correlation analysis is performed between the hydrological or temperature sequence after seasonal term removal and the corresponding carbon flux component sequence to estimate the lag time of each carbon sink component, and the lag time is transmitted to each grid cell. The phase correction of the time series of each carbon sink component is performed using the lag duration to obtain a synchronized carbon sink sequence. Instantaneous phenological phases are constructed using the phenological phase shift parameters extracted from the phenological phases. The relative instantaneous phenological phases are determined as the abscissa using the starting boundary of each ecological state interval as the origin, and the net carbon sink rate calculated from the synchronized carbon sink sequence is used as the ordinate to generate a dynamic carbon sink trajectory. The sign transition moment of the net carbon sink rate in the dynamic trajectory of the carbon sink is detected and output as the critical point of carbon source-sink conversion.

2. The wetland carbon sequestration dynamic assessment method based on multi-source remote sensing temporal fusion according to claim 1, characterized in that: The process of phenological phase extraction includes: Multispectral reflectance is extracted from the optical image data corresponding to the pigment perception dimension in the fused grid, and the vegetation index time series is calculated based on the multispectral reflectance. Perform sliding window harmonic fitting according to the actual sampling time of each vegetation index value in the vegetation index time series to obtain the phenological phase shift parameter and vegetation index amplitude corresponding to each sliding window. Arrange the phenological phase shift parameters and the vegetation index amplitude according to the time position of each sliding window to obtain the phenological phase shift parameter sequence and the vegetation index amplitude time series, respectively.

3. The wetland carbon sequestration dynamic assessment method based on multi-source remote sensing time-series fusion according to claim 2, characterized in that: The SAR image sequence includes a first polarization channel SAR image sequence and a second polarization channel SAR image sequence. The SAR backscattering time sequence includes a first polarization channel SAR backscattering time sequence and a second polarization channel SAR backscattering time sequence. Each polarization channel SAR backscattering time sequence includes the SAR backscattering value corresponding to each sampling time. The flooding state detection includes: determining the open water surface determination benchmark based on the distribution of each SAR backscatter value in the SAR backscatter time series of the first polarization channel, and determining the grid cells whose SAR backscatter value of the first polarization channel is lower than the open water surface determination benchmark as open water surface grid cells. The polarization ratio timing is calculated based on the SAR backscattering timing of the first polarization channel and the SAR backscattering timing of the second polarization channel; Water level sequences are extracted from the hydrological and meteorological observation data. Exposure reference periods are determined based on the water level sequences. Exposure state scattering baseline and exposure state polarization ratio baseline are determined based on the SAR backscattering value and polarization ratio value of the first polarization channel within the exposure reference periods, respectively. Grid pixels whose SAR backscattering value in the first polarization channel is higher than the exposure state scattering baseline and whose polarization ratio value is lower than the exposure state polarization ratio baseline are identified as submerged grid pixels under the vegetation canopy. By merging the open water surface grid cells and the submerged grid cells under the vegetation canopy, a submerged state sequence is obtained.

4. The wetland carbon sequestration dynamic assessment method based on multi-source remote sensing temporal fusion according to claim 3, characterized in that: The vegetation growth cycle is divided into multiple ecological state intervals, including: The sampling time when the state transition of the flooded state sequence occurs is determined as the candidate boundary; Calculate the change between adjacent phenological phase shift parameters in the phenological phase shift parameter sequence, determine the phase abrupt change judgment benchmark based on the statistical distribution of the change, and determine the sampling time when the change exceeds the phase abrupt change judgment benchmark as the phase abrupt change time; The baseline of the exposure state amplitude is determined based on the vegetation index amplitude corresponding to the period in the flooded state sequence that is determined to be in the exposure state, and the period in which the vegetation index amplitude is lower than the baseline of the exposure state amplitude is determined as the amplitude collapse period. When the time difference between the candidate boundary and the phase change moment does not exceed the time interval between adjacent SAR sampling moments, or when the candidate boundary is located within the amplitude collapse period, the candidate boundary is determined as a verification consistent boundary. The flooding period and the exposure period are divided according to the candidate boundary, and the adjacent sampling period containing the candidate boundary is determined as the transition period.

5. The wetland carbon sequestration dynamic assessment method based on multi-source remote sensing time-series fusion according to claim 4, characterized in that: The process of retrieving net primary productivity of vegetation includes: The photosynthetically active radiation sequence, temperature sequence, and hydrological sequence were extracted from the hydrometeorological observation data. The proportion of photosynthetically active radiation absorbed by vegetation is determined based on the vegetation index time series. The temperature response and water response of vegetation light energy utilization are determined based on the temperature series and the hydrological series, respectively. The total primary productivity of vegetation is determined in combination with the photosynthetically active radiation series. The net primary productivity of vegetation is obtained by subtracting the autotrophic respiration of vegetation from the total primary productivity of vegetation. The vegetation scattering component is separated from the SAR backscattering time series corresponding to the vegetation structure perception dimension, and the spatial distribution of the vegetation net primary productivity time series among each grid cell is corrected based on the vegetation scattering component.

6. The wetland carbon sequestration dynamic assessment method based on multi-source remote sensing time-series fusion according to claim 5, characterized in that: Before retrieving the time series of the carbon sink components, the soil organic carbon pool data of the target wetland is obtained. The process of retrieving soil heterotrophic respiration includes: Soil temperature sequences were extracted from the hydrological and meteorological observation data; Soil scattering components are separated from the SAR backscattering time series corresponding to the vegetation structure perception dimension, and soil moisture state sequence is determined based on the soil scattering components. The temperature response of soil organic carbon decomposition is determined based on the soil temperature sequence, the water response of soil organic carbon decomposition is determined based on the soil moisture state sequence and the results of the flooding state detection, and the soil heterotrophic respiration time series is determined in combination with the soil organic carbon pool data. For open water surface grid cells determined by the flooding state detection, the corresponding soil heterotrophic respiration is set to zero.

7. The wetland carbon sequestration dynamic assessment method based on multi-source remote sensing time-series fusion according to claim 6, characterized in that: The flooding state detection also includes: At each sampling time, the proportion of submerged pixels is determined based on the number of grid pixels identified as being submerged and the total number of grid pixels. The proportion of submerged pixels is then arranged according to the sampling time to obtain the submerged pixel proportion time sequence. The process of inverting the hydrological carbon transport includes: extracting precipitation sequences and water level sequences from the hydrological and meteorological observation data, determining the precipitation event period based on the precipitation sequence, and determining the hydrological carbon transport time series based on the water level sequence and the submerged pixel ratio time series within the precipitation event period; Specifically, the hydrological carbon transport output from the target wetland to external water bodies is taken as a positive value, while the hydrological carbon transport input from external water bodies to the target wetland is taken as a negative value.

8. The wetland carbon sequestration dynamic assessment method based on multi-source remote sensing time-series fusion according to claim 7, characterized in that: The carbon flux observation data includes vegetation carbon uptake observation sequences, soil respiration observation sequences, and hydrological carbon transport observation sequences. Extract hydrological sequences, temperature sequences, and precipitation event sequences from the aforementioned hydrological and meteorological observation data; Estimate the lag time of each carbon sink component, including: During the exposure period, seasonal term removal was performed on the hydrological sequence and the vegetation carbon uptake observation sequence, and the lag time of the vegetation net primary productivity was determined based on the cross-correlation analysis results of the two. During the transition period, seasonal term removal processing is performed on the temperature sequence and the soil respiration observation sequence, and the lag duration of soil heterotrophic respiration is determined based on the cross-correlation analysis results of the two. During the inundation period, the lag duration of hydrological carbon transport is determined based on the cross-correlation analysis results of the precipitation event sequence and the hydrological carbon transport observation sequence. Each set of reference sequences and carbon flux component sequences used for cross-correlation analysis is formed from different original observations.

9. The wetland carbon sequestration dynamic assessment method based on multi-source remote sensing temporal fusion according to claim 8, characterized in that: Phase correction of the time series of each carbon sink component is performed using lag duration, including: For each sampling time in the time series of each carbon sink component, the lag time of the corresponding carbon sink component is subtracted from the sampling time to obtain the corresponding ecological driving time. Using the ecological driving time series corresponding to the net primary productivity of vegetation as a unified time reference, the soil heterotrophic respiration and the hydrological carbon transport are converted to the unified time reference to obtain the synchronized carbon sink sequence. The synchronized carbon sink sequence includes synchronized vegetation net primary productivity time series, synchronized soil heterotrophic respiration time series, and synchronized hydrological carbon transport time series. For any sampling time, the difference between the instantaneous phenological phase of the sampling time and the instantaneous phenological phase corresponding to the starting boundary of the ecological state interval to which the sampling time belongs is determined as the relative instantaneous phenological phase. The net carbon sink rate is obtained by subtracting the synchronized soil heterotrophic respiration and synchronized hydrological carbon transport from the synchronized vegetation net primary productivity at the corresponding ecological driving time. The net carbon sink rates are arranged according to the ecological driving time. When the signs of the net carbon sink rates corresponding to two adjacent ecological driving times are different, the latter ecological driving time is determined as the sign conversion time.

10. A dynamic assessment system for wetland carbon sequestration based on multi-source remote sensing time-series fusion, characterized in that: The wetland carbon sequestration dynamic assessment method based on multi-source remote sensing temporal fusion as described in any one of claims 1-9 includes: The data acquisition module is used to acquire optical image sequences, SAR image sequences, carbon flux observation data and hydro-meteorological observation data of flux stations within the target wetland. The asynchronous fusion module is used to determine the beat frequency period based on the sampling time interval between the optical image sequence and the SAR image sequence, and to perform asynchronous fusion of the two types of images with the beat frequency period as a constraint to generate a fusion mesh containing pigment perception dimension and vegetation structure perception dimension. The ecological state division module is used to extract vegetation index time series and SAR backscatter time series from the fused grid, and perform phenological phase extraction and flooding state detection accordingly, dividing the vegetation growth cycle into multiple ecological state intervals. The carbon sink component inversion module is used to invert the time series of carbon sink components based on the fused grid and the hydrological and meteorological observation data. The carbon sink components include vegetation net primary productivity, soil heterotrophic respiration, and hydrological carbon transport. The lag analysis module is used at the flux station to perform cross-correlation analysis between the hydrological or temperature sequence after seasonal term removal and the corresponding carbon flux component sequence according to the ecological state interval, estimate the lag duration of each carbon sink component, and transmit the lag duration to each grid cell. The phase correction module is used to perform phase correction on the timing of each carbon sink component using the lag time to obtain a synchronized carbon sink sequence. The trajectory generation module is used to construct an instantaneous phenological phase using the phenological phase offset parameters extracted from the phenological phase, determine the relative instantaneous phenological phase as the abscissa using the starting boundary of each ecological state interval as the origin, and use the net carbon sink rate calculated from the synchronized carbon sink sequence as the ordinate to generate a dynamic carbon sink trajectory. The critical point output module is used to detect the sign transition moment of the net carbon sink rate in the dynamic trajectory of the carbon sink, and output it as the critical point of carbon source-sink conversion.