Agriculture and forestry carbon sink dynamic evaluation method and system

Abstract

This invention discloses a method and system for dynamic assessment of agricultural and forestry carbon sinks, relating to the field of agricultural and forestry carbon sink assessment technology. The method includes: acquiring multi-source data of a target region and performing data preprocessing to form standardized spatiotemporal feature data; extracting regional ecological difference features reflecting climate fluctuations, vegetation growth rhythms, and soil water-heat coupling relationships; constructing a basic carbon sink estimation model based on labeled samples from the source region and retaining relevant parameters of ecological processes; constructing regional latent variables to characterize differences in potential ecological mechanisms based on regional ecological difference features, and identifying ecological parameter drift between the target region and the source region through these regional latent variables; adaptively correcting the carbon allocation coefficient, respiration parameter, and growth response parameter in the basic carbon sink estimation model based on the ecological parameter drift results; inputting time-series data of the target region into the corrected basic carbon sink estimation model, outputting dynamic carbon sink assessment results, and iteratively updating the model as data is updated.

Description

Technical Field

[0001] This invention relates to the field of agricultural and forestry carbon sequestration assessment technology, specifically to a dynamic assessment method and system for agricultural and forestry carbon sequestration. Background Technology

[0002] As global climate change intensifies, farmland and forest ecosystems, as crucial components of the terrestrial carbon cycle, play a vital role in regional carbon balance regulation and the achievement of carbon neutrality goals. Particularly in agricultural production, crop growth cycles, farming practices, and soil management significantly impact carbon sequestration capacity; in forestry management, stand structure, tree species composition, and growth stages directly determine carbon absorption and storage levels. In recent years, research on carbon sequestration assessment in agroforestry systems has deepened, with carbon sequestration estimation techniques based on remote sensing monitoring, ecological process models, and data-driven methods rapidly developing. Remote sensing inversion methods estimate crop or forest biomass using vegetation indices and surface parameters; ecological process models calculate carbon sequestration by simulating photosynthesis, respiration, and carbon allocation processes; and machine learning methods combine meteorological, soil, and management data to construct carbon sequestration prediction models. These technologies have been widely applied to farmland carbon management, forestry resource monitoring, and regional carbon sequestration accounting, and are gradually evolving towards multi-source data fusion, refined management, and time-series dynamic modeling.

[0003] However, existing carbon sink assessment methods for agroforestry systems still have shortcomings. Most methods rely on fixed parameters or model construction based on specific regional conditions, making it difficult to adapt to ecological differences under different climate zones, crop types, or forest stand structures. This results in insufficient generalization ability of the models when applied across regions, failing to meet the accurate assessment needs of complex and variable environments in agricultural and forest areas. Existing methods are mostly based on static analysis of phased observation data, lacking the ability to dynamically respond to changes in crop growth cycles or long-term forest succession processes, and failing to truly reflect the continuous characteristics of carbon sink changes in agroforestry ecosystems over time. In the process of multi-source data fusion, existing technologies usually adopt simple splicing or weighting methods, failing to deeply explore the potential differences in ecological mechanisms in different regions in terms of climate-driven factors, phenological changes, and soil hydrothermal conditions, making it difficult for models to identify the implicit ecological parameter drift between regions. In addition, for dynamic factors such as changes in agricultural management measures (such as adjustments to cropping systems and fertilization strategies) and forestry management interventions (such as thinning and regeneration), existing methods lack effective parameter adaptive adjustment mechanisms. Once the model is established, it is difficult to respond to environmental changes and make corrections in a timely manner, thus limiting its application effect in actual agricultural and forestry production management. Therefore, how to construct a dynamic assessment method for agricultural and forestry carbon sinks that can adapt to different ecological characteristics of farmland and forest land, identify regional ecological mechanism differences, and achieve adaptive correction of model parameters under unlabeled or weakly labeled conditions has become a key technical problem that urgently needs to be solved in this field. Summary of the Invention

[0004] In view of the above-mentioned problems, the present invention is proposed.

[0005] Therefore, the technical problem solved by this invention is that existing agricultural and forestry carbon sink assessment methods have poor cross-regional generalization ability, lack dynamic update mechanism, are difficult to identify implicit drift of ecological parameters, and how to achieve adaptive correction of carbon sink model parameters and conduct dynamic assessment under unlabeled or weakly labeled conditions.

[0006] To address the aforementioned technical problems, this invention provides the following technical solution: a method for dynamic assessment of agricultural and forestry carbon sequestration, comprising: acquiring multi-source data of a target area and performing data preprocessing to form standardized spatiotemporal feature data; extracting regional ecological difference features reflecting climate fluctuations, vegetation growth rhythms, and soil water-heat coupling relationships; constructing a basic carbon sequestration estimation model based on labeled samples from the source area and retaining relevant parameters of ecological processes; constructing regional latent variables to characterize differences in potential ecological mechanisms based on regional ecological difference features, and identifying ecological parameter drift between the target area and the source area through regional latent variables; adaptively correcting the carbon allocation coefficient, respiration parameter, and growth response parameter in the basic carbon sequestration estimation model based on the ecological parameter drift results; inputting time-series data of the target area into the corrected basic carbon sequestration estimation model, outputting dynamic assessment results of carbon sequestration, and iteratively updating the model when the data is updated.

[0007] As a preferred embodiment of the dynamic assessment method for agricultural and forestry carbon sequestration described in this invention, the step of data preprocessing to form standardized spatiotemporal feature data includes: acquiring multi-temporal vegetation index data through a remote sensing data interface, and simultaneously acquiring meteorological observation data and soil attribute data for the corresponding time periods; for data with different time resolutions, using an interpolation method based on the trend changes of adjacent time points for time alignment, so that various types of data are continuously distributed on a unified time axis; performing spatial registration of data from different sources based on a geographic coordinate system, mapping various types of data to a unified spatial grid; standardizing various types of data by mean shifting and scale compression of historical statistical distributions, so that data of different dimensions have a consistent numerical range; and fusing and splicing multi-source data in time series order to form a spatiotemporal feature dataset with a unified structure.

[0008] As a preferred embodiment of the dynamic assessment method for agricultural and forestry carbon sequestration described in this invention, the regional ecological difference characteristics include: calculating climate fluctuation characteristics based on time-series meteorological data, characterizing the degree of climate fluctuation by analyzing the amplitude and stability of temperature and precipitation changes within a fixed time window; extracting dominant growth rhythm information from the vegetation index time series using periodic analysis methods to characterize the seasonal variation characteristics of vegetation; jointly modeling soil moisture and soil temperature to construct a comprehensive index reflecting the water-heat coupling relationship and analyzing the impact of the soil environment on vegetation growth; triggering a compensation mechanism during feature extraction when time-series data is insufficient or missing, supplementing the data through historical data backtracking or interpolation of neighboring regional data; and fusing and encoding climate, vegetation, and soil characteristics to generate a regional ecological difference feature vector.

[0009] As a preferred embodiment of the dynamic assessment method for agricultural and forestry carbon sequestration described in this invention, the construction of a basic carbon sequestration estimation model and retention of ecological process-related parameters includes: selecting carbon storage data observed in source area sample plots as a monitoring basis; inputting multi-source spatiotemporal characteristic data into the basic carbon sequestration estimation model for training; and continuously adjusting the internal parameters of the basic carbon sequestration estimation model to reduce the deviation between the predicted results and observed values; the output of the basic carbon sequestration estimation model is denoted as... , is represented as: , in, Representation model for time index Spatial Unit Index The predicted value of carbon reserves, This represents the set of parameters for the carbon absorption and distribution process. This represents the set of parameters for the respiratory energy expenditure process. This represents the set of parameters for the vegetation growth response process. During training, training stops when the output error of the basic carbon sink estimation model decreases to a preset stable range. After training, a set of key parameters related to ecological processes is extracted from the basic carbon sink estimation model, including parameters of carbon absorption and distribution, parameters of respiration and consumption, and parameters of vegetation growth response.

[0010] As a preferred embodiment of the dynamic assessment method for agricultural and forestry carbon sinks described in this invention, the method for identifying ecological parameter drift between the target region and the source region includes: inputting regional ecological difference characteristics into a latent variable generation module; generating implicit representations characterizing regional ecological mechanisms through feature compression and pattern extraction processes; acquiring latent variable representations for the source region and the target region respectively, and performing distribution difference analysis; determining the degree of difference in ecological mechanisms by comparing the overall distribution pattern and the location of feature centers; determining the existence of ecological parameter drift when the degree of difference exceeds a preset threshold; generating corresponding drift description information based on the direction and magnitude of the difference; and converting the latent variable differences into offsets in the model parameter space through a preset mapping relationship.

[0011] As a preferred embodiment of the dynamic assessment method for agricultural and forestry carbon sequestration described in this invention, the adaptive correction includes: applying the identified parameter shift information to the parameters of the basic carbon sequestration estimation model to make targeted adjustments to various ecological process parameters; introducing an adjustment mechanism based on the degree of drift during the correction process, adopting a weak correction strategy when the parameter correction intensity coefficient is less than 1, and adopting an enhanced correction strategy when the parameter correction intensity coefficient is equal to 1; setting different adjustment weights for different types of parameters, so that carbon allocation, respiration, and growth processes are differentiated according to their corresponding sensitivities; and gradually updating the parameters through continuous iteration until the changes stabilize during the correction process, and writing the updated parameters into the basic carbon sequestration estimation model to replace the original parameters.

[0012] As a preferred embodiment of the dynamic assessment method for agricultural and forestry carbon sequestration described in this invention, the step of outputting the dynamic assessment results of carbon sequestration and iteratively updating them during data updates includes: inputting continuous time series data of the target area into the corrected basic carbon sequestration estimation model, calculating the carbon sequestration estimation values ​​at each time point in chronological order; smoothing the output results using a sliding time window to suppress short-term fluctuations and preserve long-term trends; triggering an update mechanism when new observation data is detected or the time series length changes, and re-executing the latent variable generation and parameter correction process; simultaneously setting trigger conditions based on time intervals or data change amplitudes, executing model updates when preset conditions are met, otherwise maintaining the current state of the basic carbon sequestration estimation model.

[0013] As a preferred embodiment of the dynamic assessment system for agricultural and forestry carbon sequestration described in this invention, the system includes: a data processing module, a drift analysis module, and a dynamic assessment module. The data processing module acquires multi-source data from the target area and performs data preprocessing to form standardized spatiotemporal feature data; it extracts regional ecological difference features reflecting climate fluctuations, vegetation growth rhythms, and soil water-heat coupling relationships. The drift analysis module constructs a basic carbon sequestration estimation model based on labeled samples from the source area and retains relevant parameters of ecological processes; it constructs regional latent variables to characterize differences in potential ecological mechanisms based on regional ecological difference features, and identifies ecological parameter drift between the target area and the source area through these regional latent variables. The dynamic assessment module adaptively corrects the carbon allocation coefficient, respiration parameter, and growth response parameter in the basic carbon sequestration estimation model based on the ecological parameter drift results; it inputs the time-series data of the target area into the corrected basic carbon sequestration estimation model, outputs the dynamic assessment results of the carbon sequestration, and iteratively updates the model when the data is updated.

[0014] A computer device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement a method for dynamic assessment of agricultural and forestry carbon sequestration.

[0015] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of a dynamic assessment method for agricultural and forestry carbon sequestration.

[0016] The beneficial effects of this invention are as follows: The dynamic assessment method for agricultural and forestry carbon sinks provided by this invention achieves stable fusion of multi-source heterogeneous data by introducing a unified spatiotemporal standard data preprocessing mechanism and quality constraint rules, providing a consistent input foundation for the model. Furthermore, by constructing climate fluctuations, vegetation rhythms, and soil hydrothermal coupling characteristics, the original data is elevated to an expression form with ecological mechanism significance, thereby enhancing the ability to represent regional differences. Based on this, a low-dimensional structured expression of complex ecological differences is achieved through latent variable compression mapping, and latent variable differences are innovatively transformed into model parameter offsets, establishing a bridging relationship between the feature space and parameter space. Simultaneously, by introducing a continuous adjustment mechanism based on drift degree and a grouping weight correction strategy, stable adaptive updates of model parameters are achieved, avoiding the instability problems caused by traditional retraining methods. Finally, by constructing a dynamic assessment and triggered update closed loop, the model can continuously evolve with time and environmental changes. Overall, this invention achieves a transformation from static modeling to dynamic adaptive assessment, significantly improving the generalization ability, stability, and engineering application value of carbon sink estimation under cross-regional conditions. Attached Figure Description

[0017] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. 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.

[0018] Figure 1 This is an overall flowchart of a dynamic assessment method for agricultural and forestry carbon sequestration provided in Embodiment 1 of the present invention.

[0019] Figure 2 This is a schematic diagram of a computer device provided in Embodiment 3 of the present invention. Detailed Implementation

[0020] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of the present invention.

[0021] Example 1, referring to Figure 1 As an embodiment of the present invention, a method for dynamic assessment of agricultural and forestry carbon sequestration is provided, comprising: S1: Acquire multi-source data of the target area and perform data preprocessing to form standardized spatiotemporal feature data.

[0022] Furthermore, data preprocessing to form standardized spatiotemporal feature data includes: acquiring multi-temporal vegetation index data through remote sensing data interfaces, and simultaneously acquiring meteorological observation data and soil attribute data for the corresponding time periods; for data with different time resolutions, using interpolation based on the trend changes of adjacent time points for time alignment, so that various types of data are continuously distributed on a unified time axis; spatially registering data from different sources based on a geographic coordinate system, mapping various types of data to a unified spatial grid; standardizing various types of data by mean shifting and scale compression of historical statistical distributions to ensure that data of different dimensions have a consistent numerical range; and fusing and stitching multi-source data in time series order to form a spatiotemporal feature dataset with a unified structure.

[0023] It should be noted that a preferred scheme for data preprocessing to form standardized spatiotemporal feature data specifically includes acquiring multi-source data of the target area within a preset observation period. Multi-source data includes multi-temporal remote sensing vegetation index data, meteorological observation data for the corresponding time period, and soil attribute data; in optional embodiments, it may further include topographic slope, aspect, surface temperature, soil type coding, sample plot observation records, or agricultural management records. In this embodiment, remote sensing data is acquired through a satellite remote sensing interface or a UAV remote sensing interface, meteorological data is acquired through ground meteorological stations, reanalysis products, or gridded meteorological databases, and soil attribute data is acquired through soil survey databases, soil sensors, or historical soil layers. To avoid the inability to directly compare subsequent features due to different time bases and spatial reference systems used by different data sources, a unified time axis and a unified spatial grid are established after data access. Regarding time processing, the preset unified time axis is set as follows: ,in This represents the first time interval obtained by dividing the entire observation period according to a fixed time resolution. At the target moment, This represents the total number of time steps obtained by dividing the entire observation period according to a uniform time resolution; assuming a certain original data source is at adjacent observation times... and The observed values ​​at are respectively and And satisfy For the target time located between the two, Aligned values ​​are obtained by using an interpolation method based on the trend changes of adjacent time points. : , in, This represents the target value after alignment to a unified timeline. and These represent the observations of the original data at adjacent time points. and To reach the target time The two most recent and consecutive original observation times are used. When there are no direct observations at the target time point in the original data, the continuous state at that time is estimated based on the changing trends of adjacent observations, so that remote sensing data, meteorological data, and soil dynamic data can all be mapped onto the same time axis. For cases where daily and monthly scale data are mixed, low-frequency data is preferentially expanded to a unified time scale; when multiple candidate observations exist within the same time period, data with higher quality labels are preferred. It should be noted that not all missing data situations are suitable for the above interpolation method. When the length of consecutive missing data in the original data exceeds a preset missing data threshold... When this occurs, linear interpolation between adjacent points is no longer used; instead, a missing test substitution rule is triggered. The length of consecutive missing values ​​is denoted as... ,when At that time, backfilling is performed using historical data from the same period or the average of neighboring time windows. The optimal approach is to select three consecutive time steps, balancing short-term fluctuation tolerance with long-term missing data error control, thus avoiding interpolation distortion. Specifically, it can be set as follows: if valid observations exist at the current time within the same phenological stage of a historical year, then the historical contemporaneous value is used as the priority compensation value; if the historical contemporaneous value is unavailable, then the mean of valid observations of similar variables within a preset time window before and after the current time is extracted for compensation. This rule avoids introducing unreasonable trends through simple interpolation under long-term missing data conditions. In terms of spatial processing, all raw data are first unified to the same geographic coordinate reference system, and then a unified spatial grid is established according to a preset spatial resolution. The unified spatial grid can be defined as a regular raster divided by fixed grid side lengths, or as a vector surface unit consistent with the target evaluation unit. For remote sensing raster data, reprojection and resampling are used to map it to the unified grid; for station-type meteorological data, spatial interpolation or grid matching is used to assign values ​​to the corresponding spatial units; for soil point or soil areal data, spatial mapping is completed based on point location relationships or area weighting relationships. Preferably, remote sensing data uses bilinear resampling or nearest neighbor resampling, meteorological station data uses inverse distance weighted interpolation or nearest-neighbor mapping, and soil layer data uses the soil category with the largest area proportion for assignment. Specifically, the operation is as follows: First, the target area is divided into uniform spatial cell grids, with each cell preferably having a side length of 1km × 1km; then, the area proportion of each soil category within the cell is calculated. ,in For the first Area of ​​soil type within the unit Select the category with the largest proportion for the total unit area. As a representative category of this unit, the soil properties (such as water content) of this category are vectorized. Assigning values ​​to the feature vector of the spatial unit, when the area ratio of the largest category... Less than the area percentage threshold (Preferred) When calculating the soil properties of a spatial unit, the attribute vectors are weighted and averaged according to their proportions to obtain the final soil properties. , is represented as: , in This is the second largest category in terms of area. This is the second major category attribute vector. Through spatial registration rules, all data can be converted into comparable values ​​on the same spatial unit.

[0024] After completing spatiotemporal alignment, various data types are standardized to eliminate the adverse effects of different units and numerical ranges on subsequent feature fusion. Let the original value of a certain variable within the unified spatiotemporal framework be... ,in Indicates a time index. Indicates spatial cell index, This represents the variable type index; its standardized result is denoted as... Then it can be represented in the following standardized way: , in, Indicates the first The statistical mean of the class variable in the historical sample. Indicates the first The statistical standard deviation of the class variable in the historical sample. A very small positive number is set to prevent the denominator from being zero. Before forming standardized spatiotemporal characteristic data, a quality check is performed on each variable. If the effective observation ratio of a certain variable in a certain spatial unit is lower than a preset efficiency threshold... If the spatial unit does not directly participate in feature fusion in the current period, it will enter the missing measurement compensation process. The effective observation ratio can be defined as: , in, Indicates the first The spatial unit in the first The proportion of effective observations on class variables, Indicates the number of valid observation points. This indicates the total number of observation points that should be present. When If the variable is missing data in a given spatial unit, it indicates a significant risk of missing data. In such cases, it should be filled by historical data backfilling, neighborhood spatial interpolation, or matching with similar land features before being included in subsequent processing. A value of 0.7 is preferred to ensure sufficient data coverage and reduce the impact of missing measurements on feature stability. Increasing this efficiency criterion avoids generating unstable features directly when the original observation coverage is too low. After completing time alignment, spatial registration, standardization, and quality control, the multi-source data is fused and stitched together according to a unified organizational method. In this embodiment, for each target time... Each spatial unit The remote sensing vegetation index, meteorological variables, soil variables, and other auxiliary variables are concatenated into a single feature vector in a preset order. Then, the feature vectors from all times and all spatial units are combined into a unified input tensor. After this processing, any time and any spatial unit corresponds to a set of numerical, same-scale, and same-structured feature data.

[0025] It should also be noted that in the process of agricultural and forestry carbon sink assessment, data from multiple sources such as remote sensing, meteorology, and soil differ in temporal resolution, spatial reference system, and data completeness. This heterogeneity directly makes it difficult for traditional methods to construct a unified input structure, thus affecting the stability and transferability of subsequent models. This invention does not simply adopt conventional interpolation or data stitching methods, but proposes a data alignment mechanism centered on a unified time axis and a unified spatial grid. Combined with an interpolation strategy based on the trend changes of adjacent time points, this ensures that the estimation of missing data points reflects the true trend of change rather than static completion. Furthermore, by setting a missing data threshold to trigger substitution rules, historical contemporaneous and neighborhood information is introduced in cases of long-term missing data, thus avoiding interpolation distortion from a mechanistic perspective. Simultaneously, by introducing effective observation ratio constraints, proactive screening and compensation for low-quality data are achieved, thereby preventing noisy data from directly participating in modeling. This allows data from different sources to be compared and fused at a unified scale, improving the stability, consistency, and engineering feasibility of the input data.

[0026] S2: Extract regional ecological differences that reflect climate fluctuations, vegetation growth rhythms, and soil water-heat coupling.

[0027] Furthermore, the regional ecological differences include: calculating climate fluctuation characteristics based on time-series meteorological data, characterizing the degree of climate fluctuation by analyzing the magnitude and stability of temperature and precipitation changes within a fixed time window; extracting dominant growth rhythm information from vegetation index time series using periodic analysis methods to characterize the seasonal changes in vegetation; jointly modeling soil moisture and soil temperature to construct a comprehensive index reflecting the water-heat coupling relationship and analyzing the impact of the soil environment on vegetation growth; triggering a compensation mechanism during feature extraction when time-series data is insufficient or missing, supplementing the data through historical data backtracking or interpolation of neighboring regional data; and fusing and encoding climate, vegetation, and soil characteristics to generate a regional ecological difference feature vector.

[0028] It should be noted that an optimal scheme for extracting regional ecological differences reflecting climate fluctuations, vegetation growth rhythms, and soil water-heat coupling specifically includes statistical analysis of temperature and precipitation sequences within each spatial unit, using fixed time windows. These fixed time windows can be 7 days, 15 days, 30 days, or windows divided according to phenological stages. For each time window, climate statistics including mean, range, standard deviation, number of consecutive abnormal days, and cumulative deviation are extracted to characterize the strength and stability of climate changes within that window. Specifically, the number of consecutive abnormal days describes the persistence of temperature or precipitation deviations from historical norms, while the cumulative deviation describes the overall deviation from the baseline climate state within the window. The baseline climate state specifically refers to the historical average values ​​of temperature and precipitation within the same time window over the past 10 years, denoted as _____. and If a unified index of climate fluctuation is needed, the magnitude of temperature variation, the magnitude of precipitation variation, and their stability within a window can be weighted and combined to obtain a comprehensive characteristic value reflecting the level of regional climate disturbance. , is represented as: , in, This represents the difference between the maximum and minimum temperatures within a time window. This represents the difference between the maximum and minimum precipitation values ​​within a time window. This represents the cumulative temperature deviation. This represents the cumulative deviation of precipitation. , , , These are the weighting coefficients for each indicator, and can preferably be set to... , , , .

[0029] Secondly, vegetation growth rhythm characteristics are extracted. Since the carbon sequestration capacity of agroforestry ecosystems is highly correlated with the vegetation growth process, simply using the vegetation index at a single moment is insufficient to fully express the ecological differences between regions. Therefore, this embodiment performs periodic analysis on the vegetation index time series to extract dominant growth rhythm information. Specifically, indicators such as the growth start time, the start of the rapid growth phase, the peak time, the decline start time, cycle length, and peak-to-valley difference can be extracted from the vegetation index sequence according to a preset time axis. The vegetation index sequence can be smoothed first, and then the growth start time can be determined by finding the position where the first-order rate of change of the sequence turns from negative to positive. The peak time can be determined by finding the local maximum value, and the decline start time can be determined by finding the point where the rate of decline continues after the peak.

[0030] The characteristics of soil water-heat coupling are extracted again. Agricultural and forestry carbon sequestration processes depend not only on the aboveground vegetation growth status but also on soil moisture supply and soil heat conditions. To avoid treating soil moisture and soil temperature as two isolated variables and losing their synergistic relationship, this embodiment models them jointly to construct a comprehensive index reflecting the impact of the soil environment on vegetation growth. Specifically, within a unified time window, the average soil moisture, average soil temperature, soil moisture fluctuation range, soil temperature fluctuation range, and their synchronous change trends can be statistically analyzed. The synchronous change trend can be characterized by comparing the consistency of their rising and falling directions, the duration of their simultaneous change, or the degree of correlation within the same window. Specifically, within the unified time window, the change directions of soil moisture and soil temperature are compared hourly. When both rise or fall simultaneously, the direction is considered consistent at that moment, and the duration of their simultaneous change is obtained by statistically analyzing the time steps of consecutive consistent directions. Simultaneously, the correlation between the overall change trends of soil moisture and soil temperature within the window can be assessed, and the correlation index reflects the degree of consistency between their changes. If soil moisture continuously decreases and soil temperature continuously increases within a certain time window, it can be determined that the window exhibits significant hydrothermal stress characteristics; if both soil moisture and temperature are within a suitable range and change gradually, it can be determined that the hydrothermal coupling state within the window is relatively stable. The results are recorded as a set of soil hydrothermal coupling characteristic values ​​to reflect the possible differences in underground ecological responses in different regions under the same climatic background.

[0031] To ensure the feasibility of analyzing climate features, vegetation rhythms, and soil hydrothermal coupling features under conditions of data gaps, this embodiment adds missing data triggering and compensation rules to the feature extraction process. Let the effective observation proportion of a spatial unit in the original time-series data required for calculating a certain type of feature be denoted as . ,when If this occurs, it indicates that the reliability of directly calculating this type of feature is insufficient, thus triggering a compensation mechanism. A preset threshold representing the proportion of effective observations is preferred. Specifically, when the effective observation ratio of a spatial unit is below 70%, a missing data compensation mechanism is triggered. This ensures that most spatial units can directly calculate reliable features while effectively avoiding feature anomalies caused by insufficient observations. The compensation mechanism is executed in the following order: first, historical observations of the same type from the same period in the target area are used for retrospective completion; when historical data from the same period in the target spatial unit is missing, or the number of directly obtainable historical observation points within the unit is less than 70% of the total number of observation points, interpolation is used to complete the data from regions with similar land cover types, similar elevation ranges, or adjacent spatial relationships to the current spatial unit; if the effectiveness requirement is still not met, the feature is marked as a low-confidence feature, and its weight is reduced in subsequent feature fusion. Through this process, the feature extraction process will not be interrupted due to local missing data, while avoiding treating low-quality observations and high-quality observations equally.

[0032] After extracting the three types of ecological features, the climate fluctuation features, vegetation growth rhythm features, and soil water-heat coupling features are fused and encoded in a preset order to generate a regional ecological difference feature vector. Let the fused regional ecological difference feature vector be denoted as . Then it can be generated within the same time window. Lower spatial unit The three types of features are concatenated to obtain: , in, This represents the characteristic sub-vector of climate fluctuations corresponding to this time window and spatial unit. This represents a feature vector representing the vegetation growth rhythm. This represents the soil hydrothermal coupling feature subvector. The subsequent input structure, i.e., the regional ecological difference feature vector, must simultaneously contain three types of features from different sources but with complementary ecological meanings, and the arrangement of each type of feature in the vector must be fixed to ensure consistent comparison between different regions by the subsequent latent variable generation module. Furthermore, to enhance the stability of the regional ecological difference features, adjustments can be made after concatenation... Perform another lightweight fusion encoding. Fusion encoding can employ fixed-weighting, principal component compression, shallow autoencoding compression, or dimensionality filtering based on preset rules to compress features with high information redundancy while retaining the main components that best characterize regional ecological differences. Fixed-weighting specifically refers to assigning a fixed weighting coefficient to each type of feature after concatenating climate fluctuation features, vegetation growth rhythm features, and soil hydrothermal coupling features into a unified feature vector in a preset order. The sum of these weighting coefficients is 1. It is preferable to adopt a strategy of concatenation followed by compression, rather than direct compression in the original data stage. The reason is that the ecological meanings of climate, vegetation, and soil features are different. If compressed too early, it is easy to lose structural information useful for subsequent parameter drift identification. First, extracting features according to ecological mechanisms and then unifying and fusing them can more clearly preserve the source of regional differences.

[0033] It should also be noted that by extracting climate fluctuation characteristics, vegetation growth rhythm characteristics, and soil hydrothermal coupling characteristics, a multi-dimensional feature system reflecting regional ecological process differences is constructed, realizing the transformation of regional ecological mechanisms from surface observation data to process-level ecological characteristics. Specifically, by extracting vegetation growth rhythms through periodic analysis, the model can capture phenological differences across regions; hydrothermal coupling modeling avoids isolating soil variables, thereby enhancing the expressive power of underground ecological processes; furthermore, a missing data triggering and compensation mechanism ensures the feasibility of feature extraction even with incomplete data; and by using fusion encoding to form a structurally fixed regional ecological difference feature vector, a consistent input is provided for latent variable modeling. This improves the expressive power of regional ecological differences, enhances the model's sensitivity to changes in ecological mechanisms, and provides high-quality input for parameter drift identification.

[0034] S3: Construct a basic carbon sink estimation model based on source region labeled samples and retain parameters related to ecological processes.

[0035] Furthermore, constructing a basic carbon sink estimation model while retaining relevant ecological process parameters includes selecting carbon storage data from source region sample plots as a monitoring basis, inputting multi-source spatiotemporal characteristic data into the basic carbon sink estimation model for training, and continuously adjusting the internal parameters of the basic carbon sink estimation model to reduce the deviation between the predicted results and the observed values; the output of the basic carbon sink estimation model is denoted as... , is represented as: , in, Representation model for time index Spatial Unit Index The predicted value of carbon reserves, This represents a set of parameters for the carbon absorption and distribution process (such as daily cumulative net photosynthetic carbon uptake and vegetation leaf area index). This represents a set of parameters related to the respiration process (such as soil respiration rate and vegetation respiration rate). This represents the set of parameters for the vegetation growth response process (such as vegetation height and canopy coverage). During training, training stops when the output error of the basic carbon sink estimation model decreases to a preset stable range. After training, the set of key parameters related to ecological processes is extracted from the basic carbon sink estimation model, including parameters of carbon absorption and distribution, parameters of respiration and consumption, and parameters of vegetation growth response.

[0036] It should be noted that a preferred scheme for constructing a basic carbon sink estimation model and retaining relevant parameters of ecological processes specifically includes, in this embodiment, the object of construction of the basic carbon sink estimation model being the source region. The source region refers to a historical regional sample set with relatively complete sample plot observation carbon storage data that can serve as a basis for supervised learning. Its spatial type can be the same as or similar to the target region, and it can also cover multiple climate zones and multiple vegetation types to improve the basic model's ability to represent different ecological conditions. This invention does not directly adapt the model to all regions but first establishes a basic model that can reflect general carbon sink change patterns, and retains key parameter interfaces corresponding to ecological processes in the model for subsequent targeted correction based on regional latent variables. Specifically, for each spatial unit in the source region... and each target moment The corresponding input features are extracted from the standardized spatiotemporal feature dataset and combined with the carbon storage data observed in the source area plots to form supervised training samples. The standardized spatiotemporal feature dataset refers to a structured data set formed within the target area by mapping multiple source raw observation data (such as remote sensing vegetation index, meteorological observation data, soil dynamic data, etc.) through a unified time axis and spatial grid, compensation for missing measurements, and dimensional standardization. Let the carbon storage observed in the plots be denoted as... It indicates that the source region is in the time index. Spatial Unit Index The actual carbon storage observed values; let the corresponding basic model input features be denoted as . , It can be extracted directly from the standardized spatiotemporal feature data, or the extracted regional ecological difference feature vector can be added on top of it. As an auxiliary input. In this embodiment, it is preferable to use Defined as the set of original input features in standardized spatiotemporal feature data, while... It is mainly used for subsequent drift identification to avoid introducing regional correction information too early in the basic model stage. The basic carbon sink estimation model is denoted as... This can be implemented using ecological process models, machine learning models, shallow neural network models, or hybrid models combining both. To ensure the subsequent extraction and correction of parameters related to ecological processes, this embodiment preferably adopts a model form with an explicit parameter grouping structure, ensuring that the model contains at least three independently accessible parameter sets: carbon absorption and distribution process parameters, respiration and consumption process parameters, and vegetation growth response process parameters. Accordingly, the output of the basic carbon sink estimation model is denoted as... ,satisfy: , in, Representation model for time index Spatial Unit Index The predicted value of carbon reserves, This represents the set of parameters for the carbon absorption and distribution process. This represents the set of parameters for the respiratory energy expenditure process. This represents the set of parameters for the vegetation growth response process. The base model must allow the storage of ecological process-related parameters in functional groups, and subsequent application of offset corrections to these three types of parameters is permitted.

[0037] In this embodiment, a preferred implementation of the basic carbon sink estimation model includes three parts, divided according to ecological processes: a growth response sub-model, a carbon absorption and allocation sub-model, and a respiration consumption sub-model. Let the time index be... The spatial unit index is The corresponding standardized spatiotemporal feature vector is The output of the basic carbon sink estimation model It can be represented as: , in, This indicates that the basic carbon sink estimation model is applicable to the first... The moment, the first Predicted carbon sink values ​​for each spatial unit. This indicates the amount of carbon absorbed and distributed at that spatiotemporal location. This represents the amount of respiration consumed at that spatiotemporal location. Furthermore, the vegetation growth response state is calculated by the growth response sub-model and denoted as... , is represented as: , in, Indicates the first The moment, the first The vegetation growth response status of each spatial unit is used to characterize the comprehensive response of vegetation to temperature, precipitation, light, soil hydrothermal conditions and the phenological state of the vegetation itself at that spatiotemporal location. Represents the growth response mapping function; This represents the set of parameters for the vegetation growth response process. This growth response state is not the final carbon sink value, but rather an intermediate quantity in subsequent carbon absorption and allocation calculations. A larger value indicates more active vegetation growth and a stronger potential carbon fixation capacity under current environmental conditions. This growth response state is obtained after... Subsequently, carbon absorption distribution Calculated using the carbon absorption allocator model: , in, Represents the carbon absorption and distribution mapping function. This represents the set of parameters for the carbon absorption and distribution process. The amount of carbon absorbed and distributed is not solely determined by the original input features. The decision is not determined by the vegetation growth response state, but is simultaneously influenced by the vegetation growth response state. The basic model does not jump directly from raw features to carbon absorption results, but first extracts whether the current vegetation can grow effectively through the growth response process, and then determines the carbon absorption capacity based on this growth state, thus better reflecting actual ecological processes. (Respiration consumption) Calculated from the respiratory consumption sub-model: , in, Represents the respiratory consumption mapping function. This represents the set of parameters for the respiratory consumption process. This formula shows that respiratory consumption is not only related to environmental input characteristics... It is related to, and also to, the current state of vegetation growth response. This is because respiratory consumption is usually related to temperature, moisture conditions, and vegetation activity levels, therefore... Introducing a respiration consumption term allows the basic model to respond differently to regions with varying growth states under the same environmental conditions. Connecting the three sub-processes yields the complete expression for the basic carbon sink estimation model, expressed as: , , In one alternative implementation, the growth response state It can be written as: , Carbon absorption distribution It can be written as: , respiratory consumption It can be written as: , in, This means concatenating the original input feature vector with the growth response state. , , These represent the weight matrices of the corresponding sub-models. , , These represent the bias terms, This represents a non-linear activation function. The three sets of parameters can then be further mapped as follows: , During the training phase, carbon storage will be observed in sample plots in the source region. As a monitoring objective, the predicted values ​​are continuously adjusted by adjusting the model's internal parameters. Compared with observed values The deviation between them gradually decreases. In this embodiment, the mean square error is used as the basic training error index, denoted as . , is represented as: , in, This indicates the uniform timeline length for training. This represents the number of spatial units in the source region that participated in the training. and These represent the model's predicted values ​​and the observed values ​​from the sample plots, respectively. During training, gradient descent, stochastic gradient descent, Adam, or other differentiable optimization algorithms can be used to update the parameters. If a non-differentiable model is used, iterative search or ensemble learning can be used to update the parameters. However, regardless of the optimization method used, it should satisfy the condition that as the training iterations progress, The basic rule is gradual reduction. To avoid premature termination of the training process due to local fluctuations, this embodiment does not use the instantaneous change of error in a single round as the stopping condition, but instead adopts a stability determination rule for consecutive multiple rounds. Let the... The error after training rounds is When continuous All rounds of training satisfy: , If the basic carbon sequestration estimation model has reached a stable convergence range, training is stopped. This represents the number of consecutive training rounds required to determine stability. The threshold value for error variation can be 0.001. Only when the model error remains minimally variable across multiple training rounds is the learning of the carbon sink patterns in the source region considered stable, thus avoiding instability in training results due to occasional small fluctuations. For source region samples with a small dataset, an additional maximum number of training rounds can be set as an upper limit to control training costs. After training, the set of key parameters related to ecological processes is extracted from the basic carbon sink estimation model and stored in a structured manner. Specifically, if the basic model has an explicit process structure, the parameters characterizing vegetation carbon absorption and distribution, respiration consumption, and growth response can be directly read out as... , and If the base model is a machine learning or neural network model, it is preferable to pre-define parameter groups or intermediate layer channels corresponding to the three types of processes during model design, and extract the weights, biases, or gating coefficients associated with these channels after training to map them into three types of parameter sets. It should be noted that this step does not simply save all model parameters, but rather groups the model parameters according to their ecological significance, thereby providing a clear target for subsequent regional adaptation corrections. Furthermore, this embodiment preferably establishes a unified index representation for the three types of parameter sets. Let the overall ecological process parameter set of the base model be denoted as... Then we have: , in, This indicates that a uniform parameter vector is formed by concatenating elements in a fixed order.

[0038] It should also be noted that a basic carbon sink estimation model is constructed based on source region labeled samples, and the model parameters are structurally divided into three ecological processes: carbon absorption and distribution, respiration consumption, and growth response. By introducing intermediate state variables of growth response, both carbon absorption and respiration processes are regulated by a unified ecological state, thus enabling the model to reflect the coupling relationship in real ecological processes. At the same time, a stable convergence judgment mechanism is used to prevent the model from stopping training prematurely under local fluctuations, thereby improving the reliability of model training. By saving parameters in groups, clear adjustable objects are provided for cross-regional migration, which enhances the ecological interpretability of the model and improves the controllability of parameters.

[0039] S4: Construct regional latent variables based on regional ecological differences to characterize potential differences in ecological mechanisms, and identify ecological parameter drift between the target region and the source region through regional latent variables.

[0040] Furthermore, identifying ecological parameter drift between the target and source regions includes: inputting regional ecological difference features into a latent variable generation module; generating implicit representations of regional ecological mechanisms through feature compression and pattern extraction processes; acquiring latent variable representations for both the source and target regions and performing distribution difference analysis; determining the degree of ecological mechanism difference by comparing the overall distribution pattern and feature center positions; determining the existence of ecological parameter drift when the degree of difference exceeds a preset threshold; generating corresponding drift description information based on the direction and magnitude of the difference; and converting latent variable differences into offsets in the model parameter space through a preset mapping relationship.

[0041] It should be noted that a preferred scheme for identifying the ecological parameter drift between the target region and the source region through regional latent variables specifically includes indexing the source region and the target region at each time step. Each spatial unit The output of the regional ecological difference feature vector is obtained from the above. Then will Input latent variable generation module We generate latent variable representations that characterize potential ecological mechanisms in a region through feature compression and pattern extraction. Let the corresponding regional latent variables be denoted as... Then we have: , in, Indicates time index Spatial Unit Index The region of latent variable vectors, This is the latent variable generation module. The latent variable generation module can be implemented using shallow coding networks, autoencoders, dimensionality reduction mapping networks, or principal component compression modules. Their common requirement is to generate a regional ecological difference feature vector composed of climate fluctuation characteristics, vegetation growth rhythm characteristics, and soil water-heat coupling characteristics. Compressed into a low-dimensional vector of latent variables that retains information about the main mechanisms of difference. . The input dimension is fixed to the dimension of the S2 output vector, and the output dimension is preferably smaller than the input dimension to highlight its mechanism's role in compressing representation. Since latent variables in a single spatial unit or at a single moment are easily affected by local observation fluctuations, this embodiment further summarizes the latent variables at a regional scale to form regional-level latent variable statistical representations for both the source and target regions. Let the regional-level latent variable center vector of the source region be denoted as... The regional-level latent variable center vector of the target region is denoted as... Then, the values ​​can be obtained by taking the mean of the latent variables of all spatiotemporal samples within the region: , in, Indicates the source region in the time index Spatial Unit Index The latent variable vector on, This represents the latent variable vector of the target region at its corresponding spatiotemporal location. This indicates the number of spatial units in the source region that participated in the statistics. This represents the number of spatial units participating in the statistics within the target region. The regional-level latent variable center vector is used to characterize the central location of the ecological mechanism in the region as a whole, i.e., the average mechanism state of the region in the latent variable space. Comparing only the mean center is insufficient to fully characterize the differences in regional mechanisms; therefore, this embodiment further compares the latent variable distribution patterns of the source region and the target region. This embodiment preferably uses both center difference and dispersion difference to construct a drift index. Let the dispersion of the latent variables in the source region be denoted as... The dispersion of the latent variables in the target region is denoted as These can be calculated from the variance or standard deviation of each latent variable in the regional sample. Furthermore, a regional ecological mechanism drift index is defined. The weighted combination of centrality and dispersion is expressed as: , in, Denotes the Euclidean norm. and Let represent the weight coefficients of the central difference term and the discrete difference term, respectively, and satisfy . , The central difference term mainly reflects the degree of deviation of the mean of regional latent variables, while the discrete difference term mainly reflects the differences in the distribution of regional latent variables. The preferred central difference term is... , This allows the drift index to retain information about the overall central position change while also taking into account the impact of distributional discrepancies. It reflects the degree of shift between the target region and the source region in the central position of the overall ecological mechanism. This reflects the degree of difference between the two in the structure of ecological mechanism fluctuations; the combination of the two can more completely characterize the potential differences in ecological mechanisms between regions. If it is necessary to further consider the differences in distribution tails in practical applications, higher-order statistical terms can be added to this indicator, but at least the two basic difference terms mentioned above should be retained to ensure that subsequent mappings have a stable input basis.

[0042] When calculated Then, compare it with the preset drift threshold. Comparison. When satisfied. When, it is determined that there is a significant shift in ecological parameters in the target area relative to the source area; when If the ecological mechanism differences between the two regions are determined to be within an acceptable range during the current cycle, only weak correction or no correction may be made, and the drift threshold is set. It can be preset to 0.1. To avoid losing drift direction information by relying solely on a single scalar indicator, this embodiment also retains the regional latent variable center difference vector: , in, This represents the directional shift of the target region relative to the source region in the latent variable space. The positive or negative values ​​of each dimension in this vector indicate whether the corresponding underlying ecological mechanism is relatively enhanced or weakened. During parameter correction, the magnitude of the shift is the primary concern, rather than its direction; therefore, absolute values ​​are used. This represents the magnitude of the shift. Therefore, S4 can not only determine whether drift exists, but also generate drift description information including drift direction and magnitude, providing interpretable intermediate results for subsequent parameter correction. Furthermore, to convert the drift results in the latent variable space into shifts in the basic model parameter space, this embodiment predefines a mapping relationship between latent variable differences and ecological process parameter differences. Let the parameter shift be denoted as... Then we have: , in, This represents a mapping function from the latent variable difference vector to the model parameter offsets. Dimensions and the set of overall ecological process parameters in S3 The dimensions are consistent. To maintain consistency with the structure of the parameter set in S3, this embodiment further defines the parameter offset as: , in, This represents the set of parameters acting on the carbon absorption and distribution process. The offset, This represents the set of parameters acting on the respiratory consumption process. The offset, This represents the set of parameters acting on the vegetation growth response process. The offset. Mapping function. Linear mapping, piecewise mapping, or shallow network mapping can be used to achieve this. In this embodiment, a linear or shallow mapping form with a fixed output grouping structure is preferred to ensure that the differences in each hidden variable can be stably transmitted to the corresponding ecological process parameter grouping.

[0043] It should also be noted that by inputting the regional ecological difference feature vector into the latent variable generation module for compression mapping, a low-dimensional regional latent variable representation that retains the main ecological mechanism information is constructed, realizing a structured expression of complex ecological differences. Furthermore, by jointly analyzing the central and distributional differences of latent variables between the source and target regions, an ecological mechanism drift index is constructed, so that regional differences are not only reflected in magnitude but also in changes in distribution structure. At the same time, by introducing a directional difference vector, the drift direction information is avoided from being lost due to relying solely on a single scalar indicator. Finally, by converting the latent variable differences into model parameter offsets through a mapping function, the drift of regional ecological parameters is quantitatively identified, enhancing the cross-regional adaptability.

[0044] S5: Adaptively correct the carbon allocation coefficient, respiration parameters, and growth response parameters in the basic carbon sink estimation model based on the ecological parameter drift results.

[0045] Furthermore, adaptive correction includes applying the identified parameter shift information to the parameters of the basic carbon sink estimation model to make targeted adjustments to various ecological process parameters; introducing a drift-based adjustment mechanism during the correction process, using a weak correction strategy when the parameter correction intensity coefficient is less than 1 and an enhanced correction strategy when the parameter correction intensity coefficient is equal to 1; setting different adjustment weights for different types of parameters, so that carbon allocation, respiration, and growth processes are corrected differently according to their corresponding sensitivities; and gradually updating parameters through continuous iteration until the changes stabilize during the correction process, and writing the updated parameters into the basic carbon sink estimation model to replace the original parameters.

[0046] It should be noted that a preferred scheme for adaptive correction specifically includes, in this embodiment, utilizing the obtained parameter offsets after completing the construction of regional latent variables and the identification of ecological parameter drift. The set of ecological process parameters for the basic carbon sink estimation model Adaptive adjustments are made to enable the model to adapt to the ecological mechanism characteristics of the target region. In the actual adjustment process, to avoid introducing excessive perturbation when regional differences are small, this embodiment introduces an adjustment mechanism based on the degree of drift. Let the ecological mechanism drift index be... Then define the parameter correction strength coefficient. for: , in, This is a preset drift threshold. When... When it is less than the threshold, This indicates the adoption of a weak correction strategy, which specifically includes: adjusting the parameter offset. By proportionality factor The reduced size is then applied to the corresponding parameter set in the basic carbon sink estimation model. This means reducing the correction magnitude for parameters related to carbon absorption and distribution, respiration and consumption, and vegetation growth response to avoid introducing excessive disturbances due to small regional differences and to maintain the stability and continuity of the model output; when When greater than or equal to the threshold, This indicates that the system has entered an enhanced correction state. The enhanced correction strategy specifically includes: adjusting the parameter offset. It directly affects the parameter set of the basic carbon sink estimation model, enabling the model to fully respond to the ecological parameter drift of the target region relative to the source region, and achieve full-range correction of parameters for carbon allocation, respiration, and growth processes. This ensures the model's migration capability and dynamic adaptability across different regions. By directly mapping the degree of regional ecological differences to the parameter adjustment range, the model has the ability to change continuously when migrating between different regions, rather than a simple binary "correction / non-correction" strategy.

[0047] Furthermore, to reflect the varying sensitivities of different ecological processes to regional differences, this embodiment sets independent adjustment weights for each of the three types of parameters. , , ,in: Corresponding carbon absorption and distribution process; Corresponding to the respiratory consumption process; Corresponding vegetation growth response process; satisfying Based on the above correction strength and weights, the basic model parameter update rule is expressed as follows: , , , in, , , This represents the corrected parameter set. To avoid model instability caused by a single correction during parameter updates, this embodiment further introduces an iterative convergence mechanism. Let the... The parameters after the next iteration are Then, calculate the magnitude of parameter change after each update round. , is represented as: , When continuous All iterations satisfy When the parameter correction process reaches a stable state, the iteration stops. A stability threshold of 0.005 can be used. This prevents parameter oscillations caused by latent variable noise or local anomalies, thus ensuring that the corrected model has stable output capability in the target region.

[0048] After completing the iteration, the final set of parameters will be obtained: , It is incorporated into the basic carbon sequestration estimation model to replace the original parameter set. .

[0049] It should also be noted that by applying the parameter offset obtained from the latent variable difference mapping to the basic model parameters and introducing a continuous adjustment mechanism based on the degree of drift, the model parameters are transformed from being fixed to being continuously and adaptively adjusted. Furthermore, by setting adjustment weights for different ecological processes, carbon absorption, respiration, and growth processes can be differentiated according to their respective sensitivities, avoiding the amplification of bias caused by uniform adjustment. At the same time, the parameter update process is controlled by an iterative convergence judgment mechanism, so that the parameter correction gradually tends to stabilize, preventing model oscillations due to noise or anomalies. The updated parameters are written into the model, enabling it to adapt to the ecological mechanisms of the target area, improving the model's cross-regional generalization ability, enhancing the stability of parameter correction, and improving the accuracy of carbon sink estimation.

[0050] S6: Input the time series data of the target area into the corrected basic carbon sink estimation model, output the dynamic assessment results of the carbon sink, and iteratively update the data when it is updated.

[0051] Furthermore, the output of dynamic carbon sink assessment results and iterative updates during data updates include: inputting continuous time series data of the target area into the corrected basic carbon sink estimation model, calculating the carbon sink estimation value at each time point in chronological order; smoothing the output results using a sliding time window to suppress short-term fluctuations and preserve long-term trends; triggering an update mechanism when new observation data is detected or the time series length changes, and re-executing the latent variable generation and parameter correction process; simultaneously setting trigger conditions based on time intervals or data change amplitudes, executing model updates when preset conditions are met, otherwise maintaining the current state of the basic carbon sink estimation model.

[0052] It should be noted that a preferred approach for outputting dynamic carbon sink assessment results and iteratively updating them as the data is updated specifically includes indexing the target region at each time point. and spatial units Standardize the spatiotemporal feature vector Input the corrected model to obtain the carbon sequestration estimate: , in, This represents the carbon sink estimation result for the target region at the corresponding spatiotemporal location. The parameters are determined by... Updated to Therefore, it can reflect the unique ecological mechanisms of the target area. Due to noise and short-term fluctuations in actual observation data, this embodiment uses a sliding window smoothing process to enhance the stability of the output results. Let the smoothed result be... , is represented as: , in, Indicates the length of the sliding window. Indicates the first The weight of each historical moment, satisfying This processing method can suppress short-term abnormal fluctuations while preserving long-term trends, making the output results more consistent with the continuity of actual ecological processes. During dynamic operation, when new observation data is received or the time series is extended, it is necessary to determine whether to re-execute the parameter update process of S4 and S5. To this end, this embodiment sets up a dual triggering mechanism, including time triggering conditions and data change triggering conditions.

[0053] Let the time interval since the last model update be . Let the rate of change of the input data be... The trigger condition is defined as follows: , in, This indicates the time-triggered threshold, which can be selected as 7 days. This represents the threshold for data change, which can be selected as 0.15. When any condition is met, the model update process is triggered, i.e., S4 (latent variable generation and drift identification) and S5 (parameter correction) are re-executed; otherwise, the current model parameters are maintained. The calculations remain unchanged; only standard carbon sequestration estimations are performed. It should be noted that the data change rate... It can be calculated by the difference between the current input features and the historical input features, for example, by using a normalization index of the magnitude of feature vector change.

[0054] It should also be noted that by inputting the time-series data of the target area into the modified model, continuous time-series output of carbon sink results is achieved, enabling the assessment results to reflect the dynamic changes in the ecosystem. Sliding window smoothing suppresses short-term noise fluctuations and enhances the stability of the results. At the same time, by setting a dual-trigger update mechanism based on time interval and data change amplitude, the model can automatically perform parameter updates when environmental changes are significant, avoiding unnecessary calculations in the stable phase, thereby achieving a balance between computational efficiency and accuracy. This improves the timeliness of the results and enhances the model's adaptive update capability.

[0055] Example 2 is an embodiment of the present invention, which provides a dynamic assessment system for agricultural and forestry carbon sequestration, including a data processing module, a drift analysis module, and a dynamic assessment module.

[0056] The data processing module acquires multi-source data from the target area and preprocesses it to form standardized spatiotemporal feature data; it extracts regional ecological difference features reflecting climate fluctuations, vegetation growth rhythms, and soil water-thermal coupling relationships; the drift analysis module constructs a basic carbon sink estimation model based on labeled samples from the source area and retains relevant parameters of ecological processes; it constructs regional latent variables to characterize differences in potential ecological mechanisms based on regional ecological difference features, and identifies ecological parameter drift between the target area and the source area through regional latent variables; the dynamic evaluation module adaptively corrects the carbon allocation coefficient, respiration parameter, and growth response parameter in the basic carbon sink estimation model based on the ecological parameter drift results; it inputs the time-series data of the target area into the corrected basic carbon sink estimation model, outputs the dynamic evaluation results of carbon sink, and iteratively updates the model when the data is updated.

[0057] Example 3, referring to Figure 2 This embodiment also provides a computer device applicable to the dynamic assessment method of agricultural and forestry carbon sinks, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to realize the dynamic assessment method of agricultural and forestry carbon sinks as proposed in the above embodiment.

[0058] The computer device can be a terminal, comprising a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0059] This embodiment also provides a storage medium storing a computer program, which, when executed by a processor, implements the dynamic assessment method for agricultural and forestry carbon sequestration as proposed in the above embodiments. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

Claims

1. A method for dynamic assessment of agricultural and forestry carbon sequestration, characterized in that, include: Acquire multi-source data of the target area and perform data preprocessing to form standardized spatiotemporal feature data; Extract regional ecological differences that reflect climate fluctuations, vegetation growth rhythms, and soil water-heat coupling relationships; A basic carbon sink estimation model is constructed based on source region labeled samples, while retaining parameters related to ecological processes; Based on the characteristics of regional ecological differences, regional latent variables are constructed to characterize the differences in potential ecological mechanisms. Ecological parameter drift between the target region and the source region is identified through regional latent variables. Based on the ecological parameter drift results, the carbon allocation coefficient, respiration parameter and growth response parameter in the basic carbon sink estimation model are adaptively corrected; The time-series data of the target area is input into the modified basic carbon sink estimation model, and the dynamic assessment results of the carbon sink are output and iteratively updated when the data is updated.

2. The method for dynamic assessment of agricultural and forestry carbon sequestration as described in claim 1, characterized in that: The process of preprocessing data to form standardized spatiotemporal feature data includes... Multi-temporal vegetation index data is acquired through remote sensing data interface, and meteorological observation data and soil property data for the corresponding time period are acquired simultaneously. For data with different time resolutions, an interpolation method based on the trend changes of adjacent time points is used for time alignment, so that various types of data are continuously distributed on a unified time axis. Spatial registration of data from different sources is performed based on a geographic coordinate system, mapping various types of data to a unified spatial grid; Standardize various types of data by shifting the mean and compressing the scale of historical statistical distributions to ensure that data of different dimensions have a consistent numerical range. Data from multiple sources are fused and stitched together in chronological order to form a spatiotemporal feature dataset with a unified structure.

3. The method for dynamic assessment of agricultural and forestry carbon sequestration as described in claim 2, characterized in that: The regional ecological differences include, Climate fluctuation characteristics are calculated based on time series meteorological data. The degree of climate fluctuation is characterized by analyzing the magnitude and stability of temperature and precipitation changes within a fixed time window. For the time series of vegetation indices, the dominant growth rhythm information is extracted through periodic analysis to characterize the seasonal variation of vegetation. A joint model of soil moisture and soil temperature was constructed to build a comprehensive index reflecting the hydrothermal coupling relationship, and the impact of the soil environment on vegetation growth was analyzed. During feature extraction, when time series data is insufficient or missing, a compensation mechanism is triggered to complete the data by backtracking historical data or interpolating data from neighboring regions. Climate, vegetation, and soil characteristics are fused and encoded to generate regional ecological difference feature vectors.

4. The method for dynamic assessment of agricultural and forestry carbon sequestration as described in claim 3, characterized in that: The construction of a basic carbon sink estimation model and the retention of ecological process-related parameters include... Carbon storage data from source region sample plots were selected as monitoring data. Multi-source spatiotemporal characteristic data were input into the basic carbon sink estimation model for training. The internal parameters of the basic carbon sink estimation model were continuously adjusted to reduce the deviation between the prediction results and the observed values. The output of the basic carbon sink estimation model is denoted as , is represented as: ， in, Representation model for time index Spatial Unit Index The predicted value of carbon reserves, This represents the set of parameters for the carbon absorption and distribution process. This represents the set of parameters for the respiratory energy expenditure process. This represents the set of parameters for the vegetation growth response process; During training, training is stopped when the output error of the basic carbon sink estimation model decreases to a preset stable range. After training, a set of key parameters related to ecological processes are extracted from the basic carbon sink estimation model, including parameters of carbon absorption and distribution, parameters of respiration and consumption, and parameters of vegetation growth response.

5. The method for dynamic assessment of agricultural and forestry carbon sequestration as described in claim 4, characterized in that: The identification of ecological parameter drift between the target region and the source region includes, The regional ecological differences are input into the latent variable generation module, and implicit representations characterizing the regional ecological mechanisms are generated through feature compression and pattern extraction processes. The latent variable representations of the source and target regions were obtained separately, and the distribution difference analysis was performed. The degree of difference in ecological mechanisms was judged by comparing the overall distribution pattern and the location of the feature center. When the degree of difference exceeds the preset threshold, it is determined that there is ecological parameter drift, and corresponding drift description information is generated according to the direction and magnitude of the difference; The differences in latent variables are converted into offsets in the model parameter space by a pre-defined mapping relationship.

6. The method for dynamic assessment of agricultural and forestry carbon sequestration as described in claim 5, characterized in that: The adaptive correction includes, The identified parameter offset information is applied to the parameters of the basic carbon sink estimation model to make targeted adjustments to various ecological process parameters; A drift-based adjustment mechanism is introduced during the correction process. When the parameter correction intensity coefficient is less than 1, a weak correction strategy is adopted, and when the parameter correction intensity coefficient is equal to 1, an enhanced correction strategy is adopted. Different adjustment weights are set for different types of parameters, so that carbon allocation, respiration and growth processes are modified differently according to their corresponding sensitivities; During the correction process, the parameters are updated gradually through continuous iteration until the changes stabilize, and the updated parameters are written into the basic carbon sink estimation model to replace the original parameters.

7. The method for dynamic assessment of agricultural and forestry carbon sequestration as described in claim 6, characterized in that: The output of dynamic carbon sink assessment results and iterative updates during data updates include, Input the continuous time series data of the target area into the modified basic carbon sink estimation model, and calculate the carbon sink estimation value at each time point in chronological order; The output results are smoothed using a sliding time window to suppress short-term fluctuations while preserving long-term trends. When new observation data is detected or the length of the time series changes, the update mechanism is triggered, and the process of generating latent variables and correcting parameters is re-executed. At the same time, trigger conditions based on time intervals or data change magnitudes are set. When the preset conditions are met, the model is updated; otherwise, the current basic carbon sink estimation model status is maintained.

8. A dynamic assessment system for agricultural and forestry carbon sequestration, employing the dynamic assessment method for agricultural and forestry carbon sequestration as described in any one of claims 1 to 7, characterized in that: It includes a data processing module, a drift analysis module, and a dynamic evaluation module; The data processing module is used to acquire multi-source data of the target area and perform data preprocessing to form standardized spatiotemporal feature data; and to extract regional ecological difference features that reflect climate fluctuations, vegetation growth rhythms and soil water-heat coupling relationships. The drift analysis module is used to construct a basic carbon sink estimation model based on source region labeled samples and retain ecological process-related parameters; it constructs regional latent variables to characterize potential ecological mechanism differences based on regional ecological difference characteristics, and identifies ecological parameter drift between the target region and the source region through regional latent variables; The dynamic assessment module is used to adaptively correct the carbon allocation coefficient, respiration parameter, and growth response parameter in the basic carbon sink estimation model based on the ecological parameter drift results; it inputs the time series data of the target area into the corrected basic carbon sink estimation model, outputs the dynamic assessment results of carbon sink, and iteratively updates the data when it is updated.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the dynamic assessment method for agricultural and forestry carbon sinks as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the dynamic assessment method for agricultural and forestry carbon sinks as described in any one of claims 1 to 7.

Images