A method for downscaling VOD derived aboveground biomass based on iterative constrained optimization

Abstract

The application discloses a VOD derived aboveground biomass downscaling method based on iterative constraint optimization, and the method comprises the following steps: inputting coarse and fine scale basic data and completing data preprocessing to obtain a corresponding standardized data set; based on the standardized data set, a candidate matrix and a pixel corresponding relationship index are constructed; based on the pixel corresponding relationship index, an aggregation matrix is constructed; based on the candidate matrix, a mean-variance joint statistical model is constructed; based on the aggregation matrix and the mean-variance joint statistical model, a penalty likelihood objective function is constructed; based on the penalty likelihood objective function, an alternating iterative calculation is performed on a to-be-estimated parameter vector to obtain a downscaling result and generate a pixel-level uncertainty representation. The application realizes AGB product spatial fine scale resolution promotion, and has the advantages of high spatial resolution accuracy, strong physical consistency, controllable error convergence, consistency and interpretability in time series.

Description

Technical Field

[0001] This invention relates to the field of microwave remote sensing inversion and multi-scale data processing technology, and in particular to a VOD-derived aboveground biomass downscaling method based on iterative constraint optimization. Background Technology

[0002] Vegetation is a core regulator of the carbon cycle in terrestrial ecosystems, and its dynamics are significantly affected by climate change and human activities. Remote sensing technology, with its unique spatiotemporal coverage and multi-spectral observation advantages, provides a key means for monitoring vegetation dynamics.

[0003] Among them, VOD (Vegetation Optical Depth) retrieved by microwave remote sensing, because it characterizes the degree of attenuation of microwave signals by vegetation and is closely related to vegetation water content and dry biomass, is an important indicator for AGB (Above-ground Biomass) estimation. Low-frequency passive microwave observations, such as L-band, have strong penetration, are not affected by clouds and rain, are sensitive to vegetation xylem, and have a high saturation threshold. They are more suitable for long-term continuous AGB monitoring than optical remote sensing. Low-frequency VOD can better reflect the moisture status of tree trunks and is suitable for AGB estimation requirements.

[0004] Existing microwave remote sensing AGB estimation methods fall into two categories: one is active methods, represented by synthetic aperture radar (SAR), which construct a statistical mapping between scattering characteristics and AGB. These methods have advantages at local / regional scales, but are limited by observational geometric differences and complex data processing, making it difficult to achieve high spatiotemporal continuous coverage and globally consistent product output, thus restricting long-term monitoring applications. The other is passive methods, which retrieve VOD through multi-band brightness temperature observations, establish a mapping relationship between VOD and AGB using empirical functions, and generate AGB products by fitting parameters with reference data. SMOS L-VOD has a high correlation with global AGB and follows a framework of "physical mechanism + statistical fitting + quality control".

[0005] Existing passive microwave VOD-derived AGBs typically have a spatial resolution of only 25–40 km, which is insufficient to adapt to different types of disturbances and vegetation recovery stages. They also have poor spatiotemporal consistency in characterizing fine-scale heterogeneity and cannot meet the demand for high-resolution, high-reliability AGB products for regional refined ecological monitoring and carbon cycle research. Summary of the Invention

[0006] The purpose of this invention is to address the problems of insufficient spatial resolution accuracy, poor controllability of model error convergence during downscaling, difficulty in ensuring physical consistency, inability to adapt to regional refined ecological monitoring, and poor consistency and interpretability of time series in existing passive microwave VOD-derived aboveground biomass downscaling methods. This invention proposes a VOD-derived aboveground biomass downscaling method based on iterative constraint optimization.

[0007] To solve the above-mentioned technical problems, the present invention provides the following technical solution: A VOD-derived aboveground biomass downscaling method based on iterative constraint optimization includes the following steps: Input the basic data of coarse and fine scales and perform data preprocessing to obtain the corresponding standardized dataset; Based on the standardized dataset, a candidate matrix and pixel correspondence index is constructed. The candidate matrix includes a mean candidate matrix and a variance candidate matrix. The pixel correspondence index is used to indicate the fine resolution pixel corresponding to each coarse resolution pixel. Based on the pixel correspondence index, construct the corresponding aggregation matrix; Based on the standardized dataset and the candidate matrix of the mean term, the corresponding mean vector is calculated. Based on the standardized dataset and the candidate matrix of the variance term, the corresponding fine-scale standardized covariance matrix is ​​calculated. Based on the mean vector and the fine-scale standardized covariance matrix, a joint statistical model of mean and variance is constructed. Based on the aggregation matrix and the mean-variance joint statistical model, the solution is iteratively obtained based on the preset penalized likelihood objective function, and the corresponding pixel-level aboveground biomass is output based on the fine-scale solution obtained when the iteration converges.

[0008] As one possible implementation method, the standardized dataset includes: coarse-scale constraint target quantities, AGB reference data, pixel latitude and longitude, TMF perturbation type data, number of years since the last perturbation, recovery indicator variables, and window climate constraint variable data.

[0009] As one possible implementation method, the specific steps for constructing a joint statistical model of mean and variance based on the candidate matrix are as follows: Based on standardized AGB reference data, pixel latitude and longitude, TMF perturbation type data, annual maximum temperature, maximum cumulative water deficit, number of years since the last perturbation, recovery indicator variable, and window climate constraint variable data, the corresponding mean vector is calculated according to the mean term linear predictor. Based on standardized pixel latitude and longitude and TMF perturbation type data, the corresponding standardized variance value is calculated according to the log-linear predictor of the variance term. Based on the standardized variance values, a fine-scale standardized covariance matrix is ​​constructed; Based on the mean vector and the fine-scale standardized covariance matrix, a joint statistical model of mean and variance is constructed.

[0010] As one possible implementation, the expression for the log-linear predictor is: ; Among them, s k(·) represents the spline smoothing function; t is the target year; n is the year index within the four-year window [t−3,t]; I u For the unperturbed indicator variable, I d For regenerated forest indicator variables, I r x is an indicator variable for degraded forests. i ,y i For pixel latitude and longitude; AGB reference data for a specific year provided to a professional organization; The highest temperature year by year; For maximum cumulative water deficit; M i,t This is a TMF perturbation type; L i , t This represents the number of years since the last disturbance.

[0011] As one possible implementation, the expression for the log-linear predictor of the variance term is: ; Where, x i ,y i For pixel latitude and longitude; M i,t It is a TMF perturbation type.

[0012] As one possible implementation, the expression for the penalized likelihood objective function is: ; Where α is the consistency constraint weight; Ω1 and Ω2 are regularization penalty matrices, corresponding to the smoothing basis function and penalty coefficient; β is the vector of parameters to be estimated for the mean term; λ is the vector of parameters to be estimated for the variance term; y t Y is a fine-scale aboveground biomass vector; t Let A be the coarse-scale constraint objective; A is the aggregation matrix.

[0013] As one possible implementation, the method of iteratively calculating the parameter vector to be estimated based on the penalized likelihood objective function includes the following steps: Based on the penalized likelihood objective function, the variance term parameter vector to be estimated is first fixed, the mean term parameter vector to be estimated is estimated, and the fine-scale solution of the fine-scale aboveground biomass vector is calculated. Calculate the coarse-scale residual based on the initial fine-scale solution of the fine-scale aboveground biomass vector; Based on the coarse-scale residual, iterative optimization is performed alternately, parameters are updated, and fine-scale solutions are calculated. Based on the updated parameters and the fine-scale aboveground biomass vector, it is determined whether convergence has occurred. If convergence is determined, the obtained fine-scale solution is used as the downscaling result; otherwise, the parameters are updated and iterative calculation continues. Based on the downscaling results after iterative convergence, pixel-level aboveground biomass is output, and pixel-level uncertainty characterization is generated.

[0014] As one possible implementation method, the specific steps for determining whether convergence has occurred based on the updated parameters and the fine-scale aboveground biomass vector are as follows: After each iteration, check whether any convergence condition is met: If the relative change in the penalized likelihood objective function between two consecutive iterations is less than a preset threshold, the iteration is considered to have converged; otherwise, it is considered not to have converged. If the coarse-scale aggregation deviation is less than the preset tolerance, the iteration is considered to have converged; otherwise, it is considered to have not converged. If the maximum number of iterations is reached, the iteration is considered to have converged; otherwise, it is considered not to have converged.

[0015] As one possible implementation method, the specific steps for constructing the aggregation matrix based on the pixel correspondence index are as follows: Based on the pixel correspondence index, a coarse-to-fine scale mapping relationship is constructed; Based on the aforementioned coarse-fine scale mapping relationship, aggregate weights are assigned and weight normalization is completed; Based on the aforementioned coarse-to-fine scale mapping relationship and weight normalization aggregated weights, an aggregate matrix is ​​constructed.

[0016] As one possible implementation method, the steps for inputting basic data of coarse and fine scales and performing data preprocessing to obtain the corresponding standardized dataset are as follows: Acquire and input coarse-scale input data and fine-scale input data; Based on the TMF perturbation type data, each pixel is reclassified according to the perturbation type; Based on the reclassified perturbation type, calculate the number of years since the last perturbation for each pixel in each year, and generate a recovery indicator variable according to the corresponding category; Based on the annual maximum temperature and the maximum cumulative water deficit, corresponding window climate constraint variables are constructed for each target year according to the preset window length. Based on the coarse-scale constraint target quantity, AGB reference data, pixel latitude and longitude, TMF perturbation type data, number of years since the last perturbation, recovery indicator variable, and window climate constraint variable data, coarse-scale mapping relationship preprocessing and invalid pixel masking are performed. Based on the AGB reference data, annual maximum temperature, and maximum cumulative water deficit data, the core numerical variables are centered and scaled.

[0017] This invention, by adopting the above technical solutions, has significant technical effects: This invention integrates fine-scale statistical modeling with coarse-scale aggregation consistency constraints into a unified optimization framework. While achieving spatial refinement, it ensures the consistency of statistical caliber between fine-scale results and coarse-scale VOD-derived biomass, significantly improving the physical consistency and numerical stability of downscaling results. Attached Figure Description

[0018] To more clearly illustrate the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying 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.

[0019] Figure 1 This is a flowchart illustrating a VOD-derived aboveground biomass downscaling method based on iterative constraint optimization according to the present invention. Figure 2 This is a schematic diagram comparing the VOD-derived aboveground biomass downscaled to 100 m with forest plot data. Detailed Implementation

[0020] The exemplary embodiments of this disclosure are described below with reference to the accompanying drawings, including various details of the embodiments to aid understanding, and should be considered merely exemplary. Therefore, those skilled in the art will recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of this disclosure. Similarly, for clarity and brevity, descriptions of well-known functions and structures are omitted in the following description.

[0021] High spatial resolution AGB products are difficult to obtain over a long period of time, and their time coverage is limited. The most commonly used high-resolution AGB products currently include Saatchi AGB and ESA CCI AGB. Saatchi products only provide global AGB data from 2015, with a spatial resolution of 1km. The preparation of Saatchi products relies on massive data fusion, including survey data from 4,079 plots of various forest types around the world, more than 3 million GLAS lidar records and LiDAR-based tree height data, combined with multi-source optical and microwave remote sensing observations, to achieve global-scale mapping through the maximum entropy model. Although the ESA CCI product has a resolution of up to 100m, it only covers five years: 2010, 2017, 2018, 2019, and 2020. The ESA CCI product is also built on a large-scale dataset, which not only integrates active microwave observations from Sentinel-1, ENVISAT ASAR, and ALOS-1 / ALOS-2, but also incorporates auxiliary data provided by multiple sensors. In summary, although both methods have high spatial resolution, their complex generation process and high data dependence make it difficult to form continuous time series over many years, thus failing to meet the monitoring needs for long-term changes in vegetation carbon storage.

[0022] Existing long-time AGB inversion methods have low spatial resolution, making them unsuitable for detailed studies. Existing AGB data are typically generated from microwave remote sensing data with low spatial resolution, which are used to generate long-term series data. Passive microwave inversion of vegetation optical thickness (VOD) can characterize the extinction effect of vegetation on microwave radiation and is widely used to monitor the dynamic changes of vegetation water content (VWC). VOD and VWC usually have a linear relationship, and VWC can be parameterized by vegetation volume water content and biomass. After decoupling vegetation biomass from water content in the model, biomass changes can be inverted using time-series VOD. Existing methods for estimating VOD to biomass are mostly based on logistic function inversion methods, which can obtain relatively accurate time series AGB estimates. However, they all rely on VOD as input, and VOD is limited by the observation capabilities of microwave sensors, with its spatial resolution typically only 25–40 km. As a result, the AGB time series generated based on such data can only reach a scale of tens of kilometers, which is difficult to meet the needs of vegetation monitoring at the regional, watershed, or field scale.

[0023] To address the limitations of existing technologies in simultaneously meeting the demands for high resolution and long-term monitoring, this application proposes a VOD-derived aboveground biomass downscaling method based on iterative constraint optimization, such as... Figure 1 As shown, it includes the following steps: S100. Input the basic data for coarse and fine scales and perform data preprocessing; The basic data for coarse and fine scales includes coarse-scale input data and fine-scale input data; The coarse and fine scales here refer to the comparison results of input data with two different resolutions. Those skilled in the art can set the data resolution corresponding to the coarse and fine scales according to actual needs, and this specification does not limit them in detail.

[0024] The specific steps for inputting basic data on coarse and fine scales and completing data preprocessing are as follows: S110. Obtain and input coarse-scale input data and fine-scale input data; Coarse-scale input data includes: Coarse-scale constraint target quantity Y t As a subsequent consistency constraint, it adopts a coarse-resolution VOD-derived aboveground biomass time series product for the target region.

[0025] Fine-scale input data includes: AGB reference data, pixel latitude and longitude, TMF perturbation type data, annual maximum temperature and maximum cumulative moisture deficit; Among them, AGB reference data is fine-resolution reference data corresponding to the target area provided by authoritative institutions. For example, in this embodiment, the 2010 AGB reference data provided by ESA CCI (i.e., AGB) is used. 2010 ); TMF disturbance type data is a classification of disturbance and recovery types for global tropical humid forests.

[0026] S120. Based on the TMF perturbation type data, reclassify each pixel according to the perturbation type; In this embodiment, the disturbance types include: undisturbed, regenerated forest, degraded forest, logged, and others.

[0027] S130. Based on the reclassification results, calculate the number of years since the last perturbation for each pixel in each year, and generate a recovery indicator variable according to the corresponding category; In this embodiment, the number of years since the last disturbance refers to the recovery time from the year the disturbance occurred to the target year; The recovery indicator variable is a variable generated based on the type of disturbance after reclassification. It is divided into three categories, corresponding to three types of pixel disturbances: undisturbed, degraded forest, and regenerated forest. The assignment rule is: if a pixel belongs to a certain category in a certain year, the indicator variable corresponding to that category takes the value of 1; otherwise, it takes the value of 0.

[0028] S140. Based on the annual maximum temperature and the maximum cumulative water deficit, construct window climate constraint variables for each target year according to a preset window length to obtain the corresponding window input sequence. In this embodiment, the window length is four years. That is, the window climate constraint variables refer to the annual maximum temperature and maximum cumulative water deficit for each target year t, taking the annual maximum temperature and maximum cumulative water deficit for a total of 4 years from [t-3,t], and constructing the corresponding window input sequence to characterize the constraint effect of maximum temperature and maximum cumulative water deficit on interannual variation.

[0029] S200. Based on the preprocessed coarse-scale input data and fine-scale input data, construct the corresponding standardized dataset; The specific steps of standardization are as follows: S210. Based on the preprocessed coarse-scale input data and fine-scale input data, perform coarse-fine scale mapping relationship preprocessing and invalid cell masking. This embodiment performs projection, resolution uniformity, and time alignment processing on coarse-scale constraint target quantities, AGB reference data, pixel latitude and longitude, TMF perturbation type data, number of years since the last perturbation, recovery indicator variables, and window climate constraint variables, and removes invalid pixels through masking operations.

[0030] S220. Based on the preprocessed coarse-scale input data and fine-scale input data, the core numerical variables are centered and scaled to improve numerical stability. This embodiment centralizes and scales the AGB reference data, annual maximum temperature, and maximum cumulative water deficit.

[0031] The constructed standardized dataset includes: coarse-scale constraint target quantities, AGB reference data, pixel latitude and longitude, TMF perturbation type data, number of years since the last perturbation, recovery indicator variables, and window climate constraint variable data.

[0032] S300. Based on the standardized dataset, construct an index of the correspondence between candidate matrices and pixels; Based on the standardized dataset, this embodiment constructs a candidate matrix for the mean term and a candidate matrix for the variance term, and establishes an index for the pixel correspondence between coarse and fine resolutions; Save the standardized dataset and the pixel correspondence index between coarse and fine resolutions as necessary inputs for subsequent construction of the aggregation matrix and iterative solution of coarse and fine scale consistency constraints.

[0033] Each coarse-resolution pixel is mapped to a set of fine-resolution pixels, and the pixel correspondence index is used to indicate this mapping relationship. The mean term candidate matrix is ​​used to calculate the relationship between the mean values ​​of coarse-resolution pixels and the corresponding fine-resolution pixels. The variance term candidate matrix is ​​used to calculate the relationship between the variances of coarse-resolution pixels and the corresponding fine-resolution pixels. Given that a standardized dataset has been obtained and that both coarse and fine resolution scales are known, those skilled in the art can construct the aforementioned candidate matrix for the mean term, candidate matrix for the variance term, and pixel correspondence index based on existing technologies. This embodiment will not elaborate on these aspects.

[0034] S400. Construct an aggregation matrix based on the pixel correspondence index; The specific steps for constructing the aggregation matrix are as follows: S410. Based on the pixel correspondence index, construct a coarse-to-fine scale mapping relationship; Based on the coarse and fine resolution pixel correspondence index, this embodiment clarifies the composition range of fine-scale pixels i corresponding to each coarse-scale grid j, and constructs an aggregation mapping from fine scale to coarse scale.

[0035] S420. Based on the coarse-fine scale mapping relationship, allocate aggregation weights and complete weight normalization; In this embodiment, at each time step t, the aboveground biomass of the fine-scale pixels to be estimated is composed into a vector y. t For each coarse-scale grid j, the set of fine-scale pixels it covers is selected, and aggregation weights ω are assigned according to a preset aggregation rule. j,i (When the area of ​​fine-scale pixels is consistent, equal weight average is used; when there are area differences or boundary overlaps, area weight is used.) Invalid fine-scale pixels that are masked are removed and the remaining weights are normalized so that the weights in each coarse-scale grid are equal to 1. Based on this, an aggregation matrix A is constructed, and this aggregation matrix is ​​used as the basic input for implementing coarse-scale aggregation consistency constraints in subsequent iterative solutions. Specifically, based on the coarse and fine resolution pixel correspondence index, all fine-scale pixels i belonging to the spatial range of the current coarse-scale grid j are selected from all fine-scale pixels, and these fine pixels are grouped into a set to obtain the set of fine-scale pixels corresponding to the coarse-scale grid j. Fine-scale aboveground biomass vector for ; Coarse-scale constraint target vector Y t for .

[0036] Then, for each fine pixel i within the coarse grid j, an aggregation weight ω is assigned according to a preset aggregation rule. j,i ; In this embodiment, the preset aggregation rule is: If all fine-scale pixels within the coarse grid j have the same area size, then each fine pixel has the same contribution weight. If there are area differences or boundary overlaps among the fine pixels within the coarse grid j, the weights are assigned according to the actual area proportion of the fine pixels falling within the coarse grid j.

[0037] Finally, invalid fine-scale pixels in the mask are removed, and the remaining weights are normalized so that the weights of all valid fine pixels within the same coarse grid j are equal. The sum equals 1.

[0038] That is, aggregate weight satisfy: ; Where I(j) is the set of effective fine-scale pixels covered by each coarse-scale grid j.

[0039] S430. Based on the aforementioned coarse-fine scale mapping relationship and weight normalization aggregated weights, construct an aggregation matrix; In this embodiment, based on the coarse-to-fine scale mapping relationship and the weight normalized aggregated weights, an aggregation matrix A is constructed: ; in, For aggregate weights; For each coarse-scale grid j, the effective set of fine-scale pixels is covered. Based on the aggregation matrix A, the vector of aboveground biomass estimates for all coarse grids in year t can be calculated. ; ; Among them, y t Y is a fine-scale aboveground biomass vector. t This is the coarse-scale constraint target vector.

[0040] S500. Based on the candidate matrix, construct a joint statistical model of mean and variance; In this embodiment, based on the mean term candidate matrix and variance term candidate matrix, the estimated aboveground biomass for each fine-scale pixel i and each target year t is determined. We construct a statistical model that combines the mean structure and error structure at a fine scale. The specific steps to form a joint statistical model of mean and variance are as follows: S510. Based on the standardized AGB reference data, pixel latitude and longitude, TMF perturbation type data, annual maximum temperature, maximum cumulative water deficit, number of years since the last perturbation, recovery indicator variable, and window climate constraint variable dataset, construct a mean term linear predictor. In this embodiment, the mean term linear predictor η t Its mathematical structure is used to characterize spatial distribution, differences in disturbance categories, and their interannual variations: ; Among them, X t Let be the candidate matrix for the mean term, and β be the vector of parameters to be estimated for the mean term.

[0041] Specifically, the mean term linear predictor It is a linear predictor η of the mean term of all effective pixels at the fine scale in year t. i,t The column vector formed; The linear predictor for the mean term in each year takes pixel latitude and longitude, AGB reference data, reclassified TMF perturbation type, window climate constraint variable, and the number of years since the last perturbation as its core inputs. It also uses a recovery indicator variable model, allowing the same recovery indicator variable to have different recovery curves under different categories. Based on the corresponding static effects, dynamic climate constraints, perturbation types, and grouped recovery terms, the specific construction form is as follows:

[0042] Among them, s k (·) represents the spline smoothing function; t is the target year; n is the year index within the four-year window [t−3,t]; I u For the unperturbed indicator variable, I d For regenerated forest indicator variables, I r x is an indicator variable for degraded forests. i ,y i For pixel latitude and longitude; AGB reference data for a specific year provided to a professional organization; The highest temperature year by year; For maximum cumulative water deficit; M i,t This is a TMF perturbation type; L i , t This represents the number of years since the last disturbance.

[0043] S520. Calculate the mean vector based on the linear predictor of the mean term; To ensure that biomass is non-negative, the mean vector μ t Using an exponential link, the calculation method is as follows: ; Where, η t A linear predictor for the mean term.

[0044] S530. Based on the standardized pixel latitude and longitude and TMF perturbation type dataset, construct a log-linear predictor for the variance term; In this embodiment, the log-linear predictor of the variance term The mathematical structure used to characterize the variation of uncertainty with pixel latitude and longitude and the type of disturbance is as follows: ; Among them, Z t Design a matrix for the variance term, where λ is the vector of parameters to be estimated for the variance term.

[0045] Specifically, the log-linear variance term predictor uses pixel latitude and longitude and TMF perturbation type data as the main inputs to characterize the differences in error amplitude under different land surface types and pixel latitude and longitude. To reflect the variation of error with pixel latitude and longitude and TMF type, a log-linear variance term predictor is constructed. The specific construction form is as follows:

[0046] Among them, s k (·) is the spline smoothing function; x i ,y i For pixel latitude and longitude; M i,t It is a TMF perturbation type.

[0047] S540. Based on the log-linear predictor of the variance term, calculate the standardized variance value; Log-linear predictor using variance term Calculate the standardized variance v i,t To ensure that the variance is positive, exponential operations are used, and the calculation method is as follows: ; in, It is a log-linear predictor of the variance term.

[0048] S550. Based on the standardized variance values, construct a fine-scale standardized covariance matrix;

[0049] Based on the standardized variance of all fine-scale pixels The fine-scale standardized covariance matrix V is constructed using a diagonal matrix construction function. t V t To represent the heteroscedasticity of the model, a diagonal form is used, specifically constructed as follows: .

[0050] S560. Based on the mean vector and the fine-scale standardized covariance matrix, construct a joint statistical model of mean and variance; Based on the constructed mean vector and fine-scale standardized covariance matrix V t We construct a statistical model for the joint mean structure and error structure at a fine scale; assuming the fine-scale aboveground biomass vector to be estimated is y. t It follows a Gaussian distribution, and its mean is the mean vector. The covariance matrix is ​​the global variance scaling factor σ. 2 With fine-scale standardized covariance matrix V t product The specific distribution pattern is as follows: .

[0051] S600. Based on the aggregation matrix and the joint statistical model of mean and variance, construct the penalized likelihood objective function; In this embodiment, in order to simultaneously consider the accuracy of fine-scale statistical fitting and the consistency of coarse-scale data, the aggregation matrix and the mean-variance joint statistical model are incorporated into a unified optimization framework to construct a penalized likelihood objective function that combines fine-scale statistical fitting, coarse-scale consistency constraints, and regularization penalties. First, the function uses the negative log-likelihood of the fine-scale statistical model as the fitting term to ensure the model's fit to fine-scale data. Secondly, a quadratic penalty term centered on the aggregation matrix A is introduced to improve the fine-scale aboveground biomass vector. In the sense of aggregation, and coarse-scale constrained objective quantity To maintain consistency, the balance between the coarse-scale consistency constraint strength and the fine-scale statistical fitting effect is adjusted by constraining the weight parameter α. Finally, for the mean term parameter vector β, the variance term parameter vector λ and the corresponding smoothing function, a regularization penalty term is introduced to suppress model overfitting and ensure parameter identifiability and numerical solution stability. To quantify the above optimization objective, the specific mathematical expression of the penalized likelihood objective function is as follows: ; Where α is the consistency constraint weight; Ω1 and Ω2 are regularization penalty matrices, corresponding to the smoothing basis function and penalty coefficient.

[0052] S700. Based on the penalized likelihood objective function, the parameter vector to be estimated is calculated alternately and iteratively to obtain the downscaling result and generate a pixel-level uncertainty representation. This embodiment, based on the penalized likelihood objective function, includes mean and variance parameters. Direct global optimization is computationally intensive and unstable. Therefore, it employs a method that estimates the mean parameter vector β, the variance parameter vector λ, and the fine-scale solution y. t Perform alternating iterative calculations to obtain the final fine-scale solution. The specific steps are as follows: S710. Based on the penalized likelihood objective function, while fixing the variance term's estimated parameter vector λ, estimate the mean term's estimated parameter vector β, and calculate the fine-scale solution of the fine-scale aboveground biomass vector. .

[0053] S720. Calculate the coarse-scale residual based on the initial fine-scale solution of the fine-scale aboveground biomass vector. The calculation method is as follows: ; in, Let A be the coarse-scale constraint target vector, and A be the aggregation matrix. For the fine-scale solution of the downscaling After iterative convergence, the final y will be t As a downscaling output That is, outputting pixel-level aboveground biomass. .

[0054] This embodiment fits the mean term under the assumption of constant variance and calculates the coarse-scale residual e. t Examine whether there are systematic differences between the residuals and candidate explanatory variables (such as TMF type, spatial location, etc.), and include variables that have explanatory power for the error magnitude in the Z-test. t The estimation is iterated again to improve the stability of the uncertainty characterization; finally, the fine-scale aboveground biomass downscaling results and its uncertainty characterization are output; through residual diagnostic analysis, the systematic correlation between residuals and candidate explanatory variables such as pixel latitude and longitude and perturbation type is examined, and variables that have explanatory power for error magnitude are included in the variance term design matrix. Then iterate and estimate again.

[0055] S730: Based on coarse-scale residuals, perform alternating iterative optimization, update parameters, and calculate fine-scale solutions; Alternating iterative optimization involves updating the mean term's estimated parameter vector β while keeping the variance term's estimated parameter vector λ constant, and updating the variance term's estimated parameter vector λ based on the model residuals while keeping the mean term's estimated parameter vector β constant. This ensures that the error structure reflects the uncertainties caused by pixel latitude and longitude and perturbation type. Furthermore, in each iteration, consistency constraints are used to correct the fine-scale aboveground biomass vector y. t Until it converges.

[0056] To achieve fine-scale field updates under consistency constraints, given the mean vector μ t With fine-scale standardized covariance matrix At that time, let the fine-scale solution y t The update is achieved by minimizing the following objective function: ; in, Let Y be the covariance matrix, α be the consistency constraint weight, and Y be the covariance matrix. t Let A be the coarse-scale constraint target vector, and let A be the aggregation matrix.

[0057] Solving the above equation can be transformed into a system of linear equations: ; Solve and update the fine-scale solution y t After correction, the polymerization result A closer approximation of the coarse-scale constrained target vector Y t .

[0058] In summary, the final estimate of fine-scale aboveground biomass (AGB) in this embodiment is y. t μ t Calculated by the mean term model under given explanatory variables, it represents the statistical prediction baseline of the fine-scale AGB; V t The variance term model characterizes the difference in pixel-level error magnitude, and in the above formula, it adjusts the proximity to the prediction baseline and the target quantity Y that satisfies the coarse-scale constraint in the form of weights. t The trade-off between them. Therefore, y is obtained in each iteration through the above system of linear equations. t This refers to the updated fine-scale AGB field under the combined influence of the mean-variance structure and the coarse-to-fine scale consistency constraint. After iterative convergence, the final y will be... t As a downscaling output .

[0059] S740. Based on the updated parameters and the fine-scale aboveground biomass vector, determine whether convergence has occurred. If convergence is determined, output the downscaling result; otherwise, update the parameters and continue iterative calculation. In this embodiment, when convergence is determined, the fine-scale solution y is... t As a result of downscaling , When convergence is determined, based on the fine-scale solution y t As an update parameter for the next iteration.

[0060] In this embodiment, after each iteration, it is checked whether any convergence condition is met: If the relative change in the penalized likelihood objective function between two consecutive iterations is less than a preset threshold, the iteration is considered to have converged; otherwise, it is considered not to have converged. Or coarse-scale aggregation deviation If the value is less than the preset tolerance, the iteration is considered converged; otherwise, it is considered non-converged. That is, based on... Performing consistency checks is a form of quality control. If the maximum number of iterations is reached, the iteration is considered to have converged; otherwise, it is considered not to have converged.

[0061] S750: Based on the downscaling results after iterative convergence, output pixel-level aboveground biomass and generate pixel-level uncertainty representation; Based on the downscaling results after iterative convergence Output pixel-level aboveground biomass The corresponding relationship is as follows: ; If the AGB reference data variables are standardized or scaled during implementation, then in the output The physical unit is restored by performing the corresponding inverse transformation.

[0062] Based on the variance structure parameters estimated after iterative convergence, a pixel-level uncertainty characterization matching the pixel-level aboveground biomass is generated. The converged variance structure parameters include: the variance term to be estimated parameter vector λ, and the covariance matrix derived from λ. .

[0063] Note that if AGB variables are standardized or scaled during implementation, the output will be... The physical unit is restored by performing the corresponding inverse transformation.

[0064] This embodiment employs alternating iterative estimation and consistency constraint correction. In each iteration, the deviation between the fine-scale aggregation result and the coarse-scale constraint target quantity is gradually reduced, and the error structure is updated synchronously to achieve convergence control of the downscaling error. This effectively avoids the bias diffusion and uncertainty amplification problems commonly found in traditional direct interpolation and one-time regression downscaling methods. The method of this invention adopts a modeling strategy that combines static constraint variables (such as AGB reference data and spatial location), dynamic climate constraint variables (such as annual maximum temperature and window climate constraint variables), perturbation type, and recovery variables (number of years since the last perturbation). This allows the downscaling results to characterize differentiated recovery trajectories under different perturbation categories, while simultaneously enhancing the consistency and interpretability of the product over time series. The high-resolution (100m) VOD-derived aboveground biomass products generated by the method of this invention can accurately characterize the spatial heterogeneity of biomass, providing strong support for forest ecosystem monitoring, disturbance recovery assessment and carbon cycle-related research, and meeting the high-resolution data requirements for refined applications at the regional to watershed scale.

[0065] Case Study: In this case study, the Amazon basin is the scope of the research. Coarse-scale inputs were VOD-derived aboveground biomass time series products with a spatial resolution of 25 km. The fine-scale output resolution is set to 100m; The aboveground biomass (AGB) reference data for a specific year provided by the authoritative organization is the 2010 AGB reference data provided by ESA CCI.

[0066] A comparison diagram of the 100 m downscaling results of VOD-derived aboveground biomass obtained according to the above-mentioned VOD-derived aboveground biomass downscaling method based on iterative constraint optimization and the forest plot data is shown below. Figure 2 As shown, the red line represents the 1:1 reference line; Following the above steps, VOD-derived aboveground biomass (AGB) downscaling data products for the Amazon region from 2010 to 2023 (14 years) with a spatial resolution of 100m can be obtained. To verify the reliability of the product, this embodiment further collects independent plot observation data for comparative evaluation. The comparison results show that the VOD-AGB product obtained by the present invention has high spatial consistency with ground plot observations, with a correlation coefficient of 0.83 and a root mean square error (RMSE) of 53.94 Mg / ha. This embodiment improves the spatial resolution of passive microwave VOD-derived biomass while maintaining the long-term temporal continuity, clearly depicting the spatial heterogeneity and interannual variation characteristics of biomass in the Amazon watershed. This demonstrates that the VOD-derived aboveground biomass downscaling method based on iterative constraint optimization provided by the present invention has the advantages of high spatial resolution accuracy, strong physical consistency, controllable error convergence, consistency and interpretability over time series.

[0067] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. The same or similar parts between the various embodiments can be referred to each other.

[0068] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, apparatus, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0069] This invention is described with reference to flowchart illustrations and / or block diagrams of the method, terminal device (system), and computer program product according to the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing terminal device to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing terminal device, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0070] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing terminal device to operate in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0071] These computer program instructions can also be loaded onto a computer or other programmable data processing terminal equipment, causing a series of operational steps to be performed on the computer or other programmable terminal equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable terminal equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0072] It should be noted that: The phrase "an embodiment" or "an embodiment" used in this specification means that a particular feature, structure, or characteristic described in connection with the embodiment is included in at least one embodiment of the invention. Therefore, the phrase "an embodiment" or "an embodiment" appearing in various places throughout the specification does not necessarily refer to the same embodiment.

[0073] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of the invention.

[0074] Furthermore, it should be noted that the shapes and names of the parts and components described in the specific embodiments described in this specification may differ. All equivalent or simple variations made to the structure, features, and principles described in this patent concept are included within the protection scope of this patent. Those skilled in the art to which this invention pertains may make various modifications or additions to the described specific embodiments or use similar methods to replace them, as long as they do not depart from the structure of this invention or exceed the scope defined in these claims, they should all fall within the protection scope of this invention.

Claims

1. A method for downscaling VOD-derived aboveground biomass based on iterative constraint optimization, characterized in that, Includes the following steps: Input the basic data of coarse and fine scales and perform data preprocessing to obtain the corresponding standardized dataset; Based on the standardized dataset, a candidate matrix and pixel correspondence index is constructed. The candidate matrix includes a mean candidate matrix and a variance candidate matrix. The pixel correspondence index is used to indicate the fine resolution pixel corresponding to each coarse resolution pixel. Based on the pixel correspondence index, construct the corresponding aggregation matrix; Based on the standardized dataset and the candidate matrix of the mean term, the corresponding mean vector is calculated. Based on the standardized dataset and the candidate matrix of the variance term, the corresponding fine-scale standardized covariance matrix is ​​calculated. Based on the mean vector and the fine-scale standardized covariance matrix, a joint statistical model of mean and variance is constructed. Based on the aggregation matrix and the mean-variance joint statistical model, the solution is iteratively obtained based on the preset penalized likelihood objective function, and the corresponding pixel-level aboveground biomass is output based on the fine-scale solution obtained when the iteration converges.

2. The method for downscaling VOD-derived aboveground biomass based on iterative constraint optimization according to claim 1, characterized in that, The constructed standardized dataset includes: coarse-scale constraint target quantities, AGB reference data, pixel latitude and longitude, TMF perturbation type data, number of years since the last perturbation, recovery indicator variables, and window climate constraint variable data.

3. The method for downscaling VOD-derived aboveground biomass based on iterative constraint optimization according to claim 2, characterized in that, The specific steps for constructing the joint statistical model of mean and variance based on the candidate matrix are as follows: Based on standardized AGB reference data, pixel latitude and longitude, TMF perturbation type data, annual maximum temperature, maximum cumulative water deficit, number of years since the last perturbation, recovery indicator variable, and window climate constraint variable data, the corresponding mean vector is calculated according to the mean term linear predictor. Based on standardized pixel latitude and longitude and TMF perturbation type data, the corresponding standardized variance value is calculated according to the log-linear predictor of the variance term. Based on the standardized variance values, a fine-scale standardized covariance matrix is ​​constructed; Based on the mean vector and the fine-scale standardized covariance matrix, a joint statistical model of mean and variance is constructed.

4. The method for downscaling VOD-derived aboveground biomass based on iterative constraint optimization according to claim 3, characterized in that, The expression for the log-linear predictor is: ； Among them, s k (·) represents the spline smoothing function; t is the target year; n is the year index within the four-year window [t−3,t]; I u For the unperturbed indicator variable, I d For regenerated forest indicator variables, I r x is an indicator variable for degraded forests. i ,y i For pixel latitude and longitude; AGB reference data for a specific year provided to a professional organization; The highest temperature year by year; For maximum cumulative water deficit; M i,t This is a TMF perturbation type; L i , t This represents the number of years since the last disturbance.

5. A method for downscaling VOD-derived aboveground biomass based on iterative constraint optimization according to claim 3, characterized in that, The expression for the log-linear predictor of the variance term is: ； Where, x i ,y i For pixel latitude and longitude; M i,t It is a TMF perturbation type.

6. A method for downscaling VOD-derived aboveground biomass based on iterative constraint optimization according to any one of claims 1 to 5, characterized in that, The expression for the penalized likelihood objective function is: ； Where α is the consistency constraint weight; Ω1 and Ω2 are regularization penalty matrices, corresponding to the smoothing basis function and penalty coefficient; β is the vector of parameters to be estimated for the mean term; λ is the vector of parameters to be estimated for the variance term; y t Y is a fine-scale aboveground biomass vector; t Let A be the coarse-scale constraint objective; A is the aggregation matrix.

7. A method for downscaling VOD-derived aboveground biomass based on iterative constraint optimization according to claim 6, characterized in that, Based on the penalized likelihood objective function, the steps for iteratively calculating the parameter vector to be estimated are as follows: Based on the penalized likelihood objective function, when the variance term is fixed, the mean term is estimated, and the fine-scale solution of the fine-scale aboveground biomass vector is calculated. Calculate the coarse-scale residual based on the initial fine-scale solution of the fine-scale aboveground biomass vector; Based on the coarse-scale residual, iterative optimization is performed alternately, parameters are updated, and fine-scale solutions are calculated. Based on the updated parameters and the fine-scale aboveground biomass vector, it is determined whether convergence has occurred. If convergence is determined, the obtained fine-scale solution is used as the downscaling result; otherwise, the parameters are updated and iterative calculation continues. Based on the downscaling results after iterative convergence, pixel-level aboveground biomass is output, and pixel-level uncertainty characterization is generated.

8. A method for downscaling VOD-derived aboveground biomass based on iterative constraint optimization according to claim 7, characterized in that, The specific steps for determining whether convergence has been achieved based on the updated parameters and the fine-scale aboveground biomass vector are as follows: After each iteration, check whether any convergence condition is met: If the relative change in the penalized likelihood objective function between two consecutive iterations is less than a preset threshold, the iteration is considered to have converged; otherwise, it is considered not to have converged. If the coarse-scale aggregation deviation is less than the preset tolerance, the iteration is considered to have converged; otherwise, it is considered to have not converged. If the maximum number of iterations is reached, the iteration is considered to have converged; otherwise, it is considered not to have converged.

9. A method for downscaling VOD-derived aboveground biomass based on iterative constraint optimization according to claim 1, characterized in that, The specific steps for constructing the aggregation matrix based on the pixel correspondence index are as follows: Based on the pixel correspondence index, a coarse-to-fine scale mapping relationship is constructed; Based on the aforementioned coarse-fine scale mapping relationship, aggregate weights are assigned and weight normalization is completed; Based on the aforementioned coarse-to-fine scale mapping relationship and weight normalization aggregated weights, an aggregate matrix is ​​constructed.

10. A method for downscaling VOD-derived aboveground biomass based on iterative constraint optimization according to claim 2, characterized in that, The steps to obtain the corresponding standardized dataset by inputting basic data of coarse and fine scales and performing data preprocessing are as follows: Acquire and input coarse-scale input data and fine-scale input data; Based on the TMF perturbation type data, each pixel is reclassified according to the perturbation type; Based on the obtained reclassification results, the number of years since the last perturbation for each pixel in each year is calculated, and recovery indicator variables are generated according to the corresponding categories. Based on the annual maximum temperature and the maximum cumulative water deficit, corresponding window climate constraint variables are constructed for each target year according to the preset window length. Based on the coarse-scale constraint target quantity, AGB reference data, pixel latitude and longitude, TMF perturbation type data, number of years since the last perturbation, recovery indicator variable, and window climate constraint variable data, coarse-scale mapping relationship preprocessing and invalid pixel masking are performed. Based on the AGB reference data, annual maximum temperature, and maximum cumulative water deficit data, the core numerical variables are centered and scaled.

Images