A regional gpp assimilation inversion system fusing ground-based and multi-source satellite chlorophyll fluorescence

Abstract

The fusion foundation multi-source satellite chlorophyll fluorescence regional GPP assimilation inversion system belongs to the technical field of terrestrial ecosystem carbon cycle monitoring, and comprises an observation data acquisition module; a signal unmixing and simulation generation module, which is used for unmixing total fluorescence into pure chlorophyll fluorescence SIF, and driving an ecological model to generate SIF and GPP simulation values based on environmental stress data; a state component decoupling module, which is used for decoupling total GPP simulation values into a light response fast component, a slow component and a memory component; a state evolution prediction module, which is used for predicting dynamic evolution of the slow component and the memory component; and an adaptive assimilation inversion module, which is used for taking the slow component and the memory component as state variables, taking SIF observation values as observation data, and combining a prediction result of the state evolution prediction module to perform inversion on regional GPP. The system overcomes systematic deviation caused by model simplification or data singularity in traditional methods.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of carbon cycle monitoring of terrestrial ecosystems, in particular to a regional GPP assimilation inversion system fusing ground-based and multi-source satellite chlorophyll fluorescence. BACKGROUND

[0002] The prior art in estimating gross primary productivity of large areas faces technical bottlenecks such as physiological time lag of vegetation photosynthesis response to the environment, nonlinear mutation of the ecosystem under stress, and nonlinear mixing of multi-vegetation functional type signals within the remote sensing pixel; in addition, there is a fundamental problem in the data source level: the current available SIF remote sensing data sources are numerous, including GOSAT, GOME-2, OCO-2, TROPOMI, etc.; however, due to differences in observation instruments, inversion algorithms, time span, spectral and spatial resolution, and revisit period of each satellite, the quality of various SIF data products is not uniform, which brings great challenges to the effective fusion and collaborative application of multi-source data, and these problems need to be solved.

[0003] To solve the above problems, a new assimilation inversion method is proposed, which first models the dynamic evolution of the state of the ecosystem, then decouples the complex time response of photosynthesis, and then deconstructs the mixed signals of remote sensing observations, and fuses multi-source information in an adaptive framework to achieve accurate inversion.

[0004] The above information disclosed in the background section is only used to strengthen the understanding of the background of the present disclosure, and therefore it can include information that does not constitute the prior art known to those of ordinary skill in the art. SUMMARY

[0005] The purpose of the present application is to provide a regional GPP assimilation inversion system fusing ground-based and multi-source satellite chlorophyll fluorescence, to solve the problems raised in the background art.

[0006] The technical solution of the present application is as follows:

[0007] An observation data acquisition module for acquiring ground-based and remote sensing observations of sunlight-induced chlorophyll fluorescence, photosynthetically active radiation, high-resolution land cover data, and environmental stress data;

[0008] A signal demixing and simulation generation module for demixing the total fluorescence of mixed pixels into pure chlorophyll fluorescence SIF based on the vegetation functional type coverage determined by the high-resolution land cover data, and driving the ecological model to generate SIF and GPP simulation values based on the environmental stress data;

[0009] A state component decoupling module for decoupling the total GPP into a light response fast component, a slow component and a memory component;

[0010] a state evolution prediction module configured to predict the dynamic evolution of the slow component and the memory component based on a preset ecosystem state critical transition model;

[0011] an adaptive assimilation inversion module configured to invert the regional GPP by taking the slow component and the memory component as state variables, taking the SIF observation value as observation data, and combining the prediction result of the state evolution prediction module.

[0012] Preferably, the signal unmixing and simulation generation module is specifically used for:

[0013] applying a preset ecological mechanism constrained mixed pixel unmixing model, combining the vegetation functional type coverage, a preset canopy radiation transmission correction operator, and an ecological interaction term, to calculate the pure chlorophyll fluorescence SIF;

[0014] applying a preset physiological constraint nonlinear conversion operator to establish a quantitative relationship from the pure chlorophyll fluorescence SIF to the maximum carboxylation rate V cmax25 .

[0015] Preferably, the ecological interaction term is modeled as a function with the coverage, leaf area index, and exclusive competition coefficient between different vegetation functional types as inputs, for describing the nonlinear effects between different vegetation functional types due to resource competition or shading.

[0016] Preferably, the physiological constraint nonlinear conversion operator includes a nonlinear adjustment term related to the temperature and vapor pressure difference in the environmental stress data, for representing the regulation of the electron transport chain by the environmental stress.

[0017] Preferably, the state component decoupling module is specifically used for:

[0018] determining the light response fast component based on the photosynthetically active radiation obtained by the observation data acquisition module;

[0019] applying a preset time convolution model to determine the memory component by integrating the historical environmental stress.

[0020] Preferably, the time convolution model adopts a double exponential form of memory kernel function, for capturing the composite memory characteristics of vegetation fast recovery and slow adaptation.

[0021] Preferably, the ecosystem state critical transition model is a nonlinear stochastic differential equation modified based on bifurcation theory, for describing the nonlinear response of the ecosystem under environmental stress.

[0022] Preferably, the adaptive assimilation inversion module is specifically used for:

[0023] constructing an assimilation framework with the slow component and the memory component as state variables;

[0024] The state evolution prediction module is used as a state transition equation of the assimilation framework.

[0025] The pure SIF observation value provided by the signal demixing and simulation generation module is used as an observation input of the assimilation framework, and the state variable and the model parameter are optimized through data assimilation.

[0026] Preferably, the adaptive assimilation inversion module is further used for:

[0027] According to the dominant vegetation functional type in the pixel, corresponding parameters are retrieved from a preset physiological parameter library;

[0028] And based on the retrieved parameters, the constraint term in the assimilation cost function is dynamically adjusted to ensure that the model evolution conforms to the physiological law of the specific vegetation functional type.

[0029] The present application provides a regional GPP assimilation inversion system for fusing ground-based multi-source satellite chlorophyll fluorescence, which has the following improvements and advantages compared with the prior art:

[0030] 1. The present application realizes dynamic tracking and accurate inversion of the carbon cycle process of the ecological system by constructing a complete closed-loop system composed of observation data acquisition, signal processing, state decoupling, dynamic prediction and adaptive assimilation. The overall advantage is that the physical mechanism model and multi-source remote sensing observation data are fused in a unified assimilation framework, so that the model prediction is constrained by real-time observation, and the interpretation of observation data is guided by the physical model, thereby overcoming the systematic bias caused by model simplification or single data in traditional methods;

[0031] 2. The present application realizes more accurate description of this physical process by introducing a mixed pixel demixing model based on ecological mechanism constraint, especially the defined ecological interaction term. The ecological interaction term of the present application can quantify the nonlinear attenuation of herbaceous fluorescence signal caused by shading, so as to calculate more real pure chlorophyll fluorescence SIF value, which cannot be realized by traditional linear model;

[0032] 3. The present application establishes a quantitative relationship from pure chlorophyll fluorescence SIF to maximum carboxylation rate V cmax25 , so that the finally inverted GPP can truly reflect the physiological state of vegetation under different stress conditions, greatly improving the reliability of the inversion result under extreme climate events;

[0033] 4. The adaptive assimilation inversion module of the present scheme solves the problem of the one-size-fits-all of traditional models by introducing a parameterization scheme based on vegetation functional types; the system can retrieve exclusive parameters from a pre-set physiological parameter library according to the dominant vegetation functional type in the pixel, such as C3 and C4 plants, and dynamically adjust the constraint term in the assimilation cost function; this means that the behavior pattern of the model itself will be adaptively adjusted according to the physiological characteristics of the object of simulation, for example, strengthening the adaptability constraint of C4 crops to high light intensity, and enhancing the sensitivity constraint of C3 forests to temperature; this strategy of explicitly embedding macroecological knowledge in the form of mathematics into the assimilation process ensures seamless transition and internal details of the inversion results between different ecosystem types, and its accuracy and reliability are far superior to that of traditional models using general parameters;

[0034] 5. The present scheme establishes a quantitative relationship from pure chlorophyll fluorescence SIF to maximum carboxylation rate V cmax25 , dynamically regulated by environmental stress, through a physiological constraint nonlinear conversion operator; this operator enables the assimilation system to effectively constrain the true physiological state of vegetation using fluorescence observations, thereby significantly improving the reliability of GPP inversion results under extreme climate events such as drought and high temperature; the adaptive assimilation inversion module of the present scheme can retrieve exclusive parameters according to the dominant vegetation functional type in the pixel, and embed macroecological knowledge in the form of mathematics into the assimilation process, ensuring the authenticity of the inversion results; the present invention integrates an automated data preprocessing process, transforming the originally tedious and time-consuming multi-source satellite data standardization work into an automated process. BRIEF DESCRIPTION OF DRAWINGS

[0035] The present invention will be further explained in conjunction with the accompanying drawings and examples:

[0036] Figure 1 is a flowchart of the regional GPP assimilation inversion system of the present invention that fuses ground-based and multi-source satellite chlorophyll fluorescence. DETAILED DESCRIPTION

[0037] In order to make the purpose, technical scheme and advantages of the present invention clearer, the present invention will be further described in detail below in conjunction with specific examples.

[0038] Example 1:

[0039] Please refer to Figure 1 , the present invention provides a regional GPP assimilation inversion system that fuses ground-based and multi-source satellite chlorophyll fluorescence, comprising:

[0040] An observation data collection module is configured to acquire ground-based and remote sensing observation data of sunlight-induced chlorophyll fluorescence, high-resolution land cover data of photosynthetically active radiation, and environmental stress data. The observation data collection module is configured to provide standardized preprocessed multi-source input data for the inversion system. The observation data collection module integrates an automatic preprocessing process to solve the problem of inconsistent SIF data quality from different satellite sources. Specifically, the observation data collection module includes the following steps:

[0041] A data space aggregation unit is configured as an entrance of data processing. The unit performs quality control and regional cropping on original satellite data according to multiple determination criteria, and then uniformly aggregates the screened data in space to a standard grid of 0.05 degrees, and outputs a nc format data file of a day scale.

[0042] A data time aggregation unit is configured to further aggregate the standard data set of a day scale to a month scale and a year scale to facilitate subsequent data mining and research.

[0043] A multi-source data joint unit is configured to match and combine data of different sources and time and space standardized data to form a unified data set, facilitating further research and analysis.

[0044] A signal unmixing and simulation generation module is configured to unmix total fluorescence of a mixed pixel into pure chlorophyll fluorescence SIF based on vegetation functional type coverage determined by high-resolution land cover data, and to drive an ecological model to generate SIF and GPP simulation values based on environmental stress data.

[0045] A state component decoupling module is configured to decouple total GPP into a light response fast component, a slow component and a memory component.

[0046] A state evolution prediction module is configured to predict dynamic evolution of the slow component and the memory component based on a preset ecological system state critical transition model.

[0047] An adaptive assimilation inversion module is configured to take the slow component and the memory component as state variables, take SIF observation values as observation data, and combine a prediction result of the state evolution prediction module to perform inversion on regional GPP.

[0048] The embodiment provides an assimilation inversion system of regional GPP fused with ground-based and multi-source satellite chlorophyll fluorescence. The system integrates multiple functional modules to form a complete technical closed loop from input of original remote sensing data to output of final GPP product.

[0049] The observation data acquisition module is included; the purpose of the module is to provide necessary and multi-source input data for the entire inversion system; in the embodiment, the module periodically acquires four types of data by accessing a data distribution service system and a meteorological reanalysis database: the first type is remote sensing observation of sunlight-induced chlorophyll fluorescence (mixed pixel total fluorescence), which is a core observation quantity and can be obtained from high-spectral satellite sensors such as TanSat, OCO-2 and TROPOMI; the second type is photosynthetically active radiation, which reflects the energy input of photosynthesis and can be obtained from geostationary or polar-orbiting satellite products; the third type is high-resolution land cover data, which is used to identify the type and spatial distribution of surface vegetation and can be obtained from high-resolution imagers such as Landsat and Sentinel; and the fourth type is environmental stress data, mainly including surface temperature and atmospheric water vapor pressure difference, which is used to quantify the degree of environmental limitation on photosynthesis and is usually obtained from ERA5 meteorological reanalysis data set;

[0050] Signal unmixing and simulation generation module; the purpose of the module is to convert the rough mixed signal observed by remote sensing into SIF observation values with clear physiological significance that can be directly used by the assimilation system; high-resolution land cover data provided by the observation data acquisition module is received, and the coverage of different vegetation functional types within each remote sensing pixel is determined; then, the mixed pixel total fluorescence signal is deconstructed to separate the pure chlorophyll fluorescence SIF signal belonging to a specific PFT;

[0051] State component decoupling module; the purpose of the module is to decompose the complex process of total GPP of vegetation into multiple components with different dynamic characteristics according to their different time scales of environmental response; in the embodiment, total GPP is decoupled into three parts: a light response fast component, a slow component, and a memory component; the light response fast component refers to the part of GPP that is directly driven by current light conditions, such as minute-to-hour level PAR changes; the slow component refers to the part of GPP that is related to the physiological state of vegetation, such as photosynthetic capacity and leaf nitrogen content, and changes in the period of several days to several weeks; the memory component refers to the part of GPP that reflects the cumulative effect or residual effect of historical environmental stress, and its change scale can reach several weeks to several months;

[0052] State evolution prediction module; the purpose of the module is to predict the future dynamics of non-instantaneous response components in GPP based on ecological mechanisms; a preset ecosystem state critical transition model is used to perform the prediction; the model specifically describes the dynamic evolution process of the decoupled slow component and memory component, and can simulate the stable, changing and even catastrophic behavior of the ecosystem under internal and external driving forces;

[0053] The adaptive assimilation module is a core calculation unit of the whole system, and aims to fuse the prediction of the model and the actual observation, so as to optimally estimate the regional GPP. A data assimilation framework is constructed, wherein the slow component and the memory component determined by the state component decoupling module are defined as state variables of the system, representing the core physiological state which needs to be tracked and optimized; the SIF observation value output by the signal unmixing and simulation generation module is introduced into the assimilation framework as external observation data; the state variables are continuously adjusted by an optimization algorithm such as Kalman filtering or variational assimilation, by combining the state prediction value at the next moment provided by the state evolution prediction module and the pure SIF observation value at the current moment provided by the signal unmixing and simulation generation module, so as to realize accurate inversion of the total GPP.

[0054] Through the organic combination of the above modules, the system constructs a complete framework capable of dynamically tracking and predicting the photosynthetic physiological state of vegetation; solves the problem that the prior art cannot effectively process physiological time lag, nonlinear mutation and signal mixing, realizes higher precision inversion of the spatiotemporal dynamic change of the regional GPP by decoupling the GPP at multiple scales and fusing multi-source information in the assimilation framework, and provides more reliable data support for carbon cycle research and ecosystem health monitoring.

[0055] Compared with the static or empirical GPP estimation method in the prior art, the present scheme realizes dynamic tracking and accurate inversion of the carbon cycle process of the ecosystem by constructing a complete closed-loop system composed of observation data acquisition, signal processing, state decoupling, dynamic prediction and adaptive assimilation; the overall advantage is that the physical mechanism model and the multi-source remote sensing observation data are fused in a unified assimilation framework, so that the model prediction is constrained by real-time observation, and the interpretation of the observation data is guided by the physical model, thereby overcoming the systematic bias caused by model simplification or data singularity in the traditional method.

[0056] Embodiment 2

[0057] The signal unmixing and simulation generation module is specifically configured to:

[0058] The mixed pixel unmixing model is constrained by the preset ecological mechanism, combined with the vegetation functional type coverage, the preset canopy radiation transmission correction operator and the ecological interaction term, to calculate the pure chlorophyll fluorescence SIF;

[0059] The preset physiological constraint nonlinear conversion operator is applied to establish a quantitative relationship from the pure chlorophyll fluorescence SIF to the maximum carboxylation rate V cmax25 .

[0060] On the basis of the above, the implementation manner of the signal unmixing and simulation generation module is limited in the embodiment.

[0061] The workflow of the module is divided into two core steps; first, a preset ecological mechanism constrained unmixing model is applied to separate the fluorescence signal; the preset ecological mechanism constrained unmixing model refers to a model that is designed to overcome the traditional linear mixing and incorporates ecological and physical school orthodoxy; the weight basis of different vegetation components is received as the coverage of vegetation functional type, and a preset canopy radiation transmission correction operator and an ecological interaction term are introduced; the preset canopy radiation transmission correction operator refers to a mathematical function used to correct the attenuation effect of the signal in the vegetation canopy due to shielding and scattering when propagating, which is based on the public radiation transmission theory, such as Beer-Lambert law; the ecological interaction term is a function specially used to quantify the nonlinear signal superposition effect between different vegetation functional types due to resource competition or physical shielding; by solving the model, the pure chlorophyll fluorescence SIF corresponding to each vegetation functional type can be calculated from the total fluorescence of the mixed pixel;

[0062] Second, a preset physiological constraint nonlinear conversion operator is applied to establish the quantitative relationship between the pure chlorophyll fluorescence SIF and the maximum carboxylation rate V cmax25 ; the preset physiological constraint nonlinear conversion operator refers to a mathematical model that establishes the nonlinear relationship between pure chlorophyll fluorescence SIF and maximum electron transfer rate (J max25 ) based on photosynthetic physiology principles; the internal mechanism is that fluorescence and GPP share the same electron transport chain, but the allocation ratio of the two is dynamically regulated by environmental stress, because their relationship is not a simple linear proportion; the operator is specially used to describe this complex nonlinear conversion;

[0063] By introducing the ecological mechanism constrained unmixing model and the physiological constraint conversion operator, the embodiment significantly improves the physical authenticity of the observation data processing; the fluorescence signals of different vegetation can be more accurately separated, thereby providing higher quality and lower uncertainty observation input for the subsequent assimilation inversion module, which is a key step to improve the final GPP inversion accuracy.

[0064] Embodiment 3

[0065] The ecological interaction term is modeled as a function of the coverage, leaf area index and exclusive competition coefficient between different vegetation functional types, which is used to describe the nonlinear effect between different vegetation functional types due to resource competition or shading;

[0066] The physiological constraint nonlinear conversion operator includes a nonlinear adjustment term related to temperature and vapor pressure difference in the environmental stress data, which is used to represent the regulation of the electron transport chain by the environmental stress;

[0067] In this embodiment, the composition of the ecological interaction term and the physiological constraint nonlinear transformation operator is further described in detail. These two features work together to improve the accuracy of the observation operator.

[0068] To further clarify, the modeling method for ecological interactions is specified; ecological interactions are symbolically represented as follows: Its function is to correct the nonlinear fluorescence effect caused by the coexistence of different vegetation communities; in this embodiment, it is modeled as a function with multiple ecological variables as input, and can be expressed mathematically as follows:

[0069] ;

[0070] in, and They are the first in each pixel species and first The coverage of vegetation functions is a dimensionless parameter, which is derived from high-resolution land cover data acquired by the observation data acquisition module. and These are the leaf area indices of the two vegetation functional types, which are dimensionless parameters used to characterize the density of the canopy, and are derived from remote sensing inversion products or model simulations. This is a specific competition coefficient, a dimensionless parameter, used to quantify a particular competitive relationship between two vegetation types, such as the shading effect between C3 herbaceous plants and C4 crops. It is derived from ground quadrat experimental data or prior ecological knowledge. To ensure dimensional consistency, the function... A function containing empirical parameters in units of fluorescence, to ensure The output dimension is consistent with that of fluorescence;

[0071] To further clarify, the function It can take a non-restrictive specific form, for example, modeling it as a linear summation of the pairwise interactions between different vegetation functional types, for any two different vegetation functional types within a pixel. and The interaction term can be expressed as:

[0072] ;

[0073] in, For any two different vegetation functional types i and j within a pixel, there is an interaction term; It is an empirical conversion factor with the dimension of fluorescence, used to convert dimensionless interaction intensities into fluorescence units. The factor can be obtained through model calibration. Therefore, the total ecological interaction term It can be the sum of all interaction terms between different vegetation functional type pairs:

[0074] ;

[0075] wherein, is the total eco-interaction term; is the interaction term between different vegetation functional types; represents different vegetation functional types; the form describes the competition effect between vegetation, for example, the product of the canopy density and the product of the canopy density of different vegetation functional types, in the form of the product of the coverage and the leaf area index , which is a modeling method that conforms to ecological common sense and can be realized;

[0076] The design of the function aims to capture the nonlinear gain or loss of fluorescence signals caused by competition for resources such as light, water, and nutrients between vegetation or physical mutual shading;

[0077] The core part of the physiological constraint nonlinear conversion operator is specified; the key of the operator is to contain a nonlinear adjustment term; the nonlinear adjustment term, denoted as , acts to represent the dynamic regulation of environmental stress on the electron transport chain in photosynthesis, thereby affecting the energy distribution ratio between fluorescence and GPP; the adjustment term is designed to be closely related to the temperature and water vapor pressure difference in the environmental stress data obtained by the observation data acquisition module;

[0078] ;

[0079] wherein, is the nonlinear adjustment term, which acts to represent the dynamic regulation of environmental stress on the electron transport chain in photosynthesis; the function is set according to the recognized plant physiology model, for example, excessive or low temperature, and excessive water vapor pressure difference, i.e., air dryness, will inhibit photosynthesis, change value, and then adjust the conversion efficiency from fluorescence to GPP; As a power index, it is a dimensionless value, and the function converts the dimensional input and into a dimensionless output;

[0080] As a non-limiting example, the function can be constructed as the product of two independent stress functions:

[0081] ;

[0082] wherein, : nonlinear adjustment term; : the product of two independent stress functions; : water vapor pressure difference stress function; temperature stress function The parabolic form commonly used for photosynthesis can be adopted:

[0083] ;

[0084] wherein, : temperature in the environmental stress data; here is the temperature at which the vegetation functional type grows best, is the minimum temperature at which it can perform photosynthesis, parameters can be obtained from the physiological parameter library, water vapor pressure difference stress function The inhibition of air dryness on stomatal conductance can be characterized by an exponential decay form:

[0085] ;

[0086] wherein, : water vapor pressure difference stress function; : water vapor pressure difference in the environmental stress data; is a vegetation-specific water stress sensitivity coefficient, the dimension of which is the inverse of the dimension of water vapor pressure difference , for example, hPa-1, to ensure that the exponential term is dimensionless, in this way, the value can specifically and nonlinearly respond to environmental changes;

[0087] Specific modeling of the ecological interaction term enables the signal unmixing process to quantify the nonlinear competition effect between different vegetation, which is particularly important in complex surface environments such as agroforestry-pasture complex areas, significantly improving the separation accuracy of pure leaf chlorophyll fluorescence SIF; at the same time, by introducing nonlinear adjustment terms directly related to temperature and humidity, the fluorescence to GPP conversion process can dynamically respond to real environmental stress, overcoming the huge error caused by the use of fixed conversion coefficients in traditional methods; in combination, the physical reality and accuracy of the GPP observation value input into the assimilation system are ensured;

[0088] In the processing link of the observed signal, the present scheme shows significant progress; the prior art usually uses a linear spectral mixing model to process remote sensing mixed pixels, this method assumes that the total signal is a simple weighted sum of component signals, ignoring the complex interactions between vegetation functional types; the present scheme introduces a mixed pixel unmixing model based on ecological mechanism constraints, especially the ecological interaction term defined therein, is modeled as the coverage , leaf area index and exclusive competition coefficient of different vegetation functional types Related functions; for example, in a mixed pixel containing both tall trees and low herbs, the trees have a light-shading effect on the herbs, which is not simply a linear superposition of signals; the ecological interaction term of the present scheme can quantify the non-linear attenuation of herb fluorescence signals caused by shading, so as to calculate more realistic pure chlorophyll fluorescence SIF values, which cannot be achieved by traditional linear models;

[0089] In the step of constraining the model state by pure chlorophyll fluorescence SIF observation, the present scheme discards the fixed and linear relationship adopted by traditional methods; the core lies in the application of a physiological constraint nonlinear conversion operator containing a nonlinear adjustment term , which establishes a quantitative relationship between pure chlorophyll fluorescence SIF and maximum carboxylation rate V cmax25 ; the actual significance of the adjustment term is to capture the dynamic regulation of environmental stress on plant photosynthesis physiology; when the plant is under drought or high temperature stress, the energy distribution of the photosynthetic electron transport chain will tend to heat dissipation rather than photochemical reaction, resulting in a change in the ratio of fluorescence to GPP; the present scheme makes the model accurately depict the nonlinear relationship between SIF and photosynthesis under environmental stress through the nonlinear adjustment term, so that the assimilation system can infer a more realistic vegetation physiological state according to the real SIF observation, thereby improving the reliability of the inversion result under extreme climate events.

[0090] Embodiment 4

[0091] The state component decoupling module is specifically configured to:

[0092] determine the light response fast component based on the photosynthetic active radiation acquired by the observation data acquisition module;

[0093] apply a preset time convolution model to determine the memory component by integrating the historical environmental stress;

[0094] The implementation manner of the state component decoupling module is limited in the embodiment;

[0095] The working process of the module is refined into two parallel steps for determining different components of GPP; in the first step, the light response fast component is determined; the definition of the light response fast component is the part of GPP that responds to light changes immediately; in the present embodiment, it is determined based on the photosynthetic active radiation data acquired by the observation data acquisition module; a standard light response curve model such as the Mitsche equation or the non-right hyperbolic model is used to calculate the instantaneous value directly according to the sub-hour PAR data; the component does not enter the subsequent state evolution prediction because it is considered as a non-memory fast process;

[0096] The module calculates the fast component based on the photosynthetically active radiation The module calculates the memory component by applying a time convolution model The slow component is defined as a core state variable that needs to be predicted by model evolution, and its initial value can be set or obtained from prior knowledge;

[0097] Second, determine the memory component The definition of the memory component is the part of GPP that reflects the cumulative effect of historical environmental stress; In this embodiment, its determination is by applying a preset time convolution model; The preset time convolution model refers to a mathematical tool for describing cumulative effects; The model obtains the current memory component value by integrating the environmental stress sequence over a period of time in the past; Mathematically expressed as:

[0098] ;

[0099] Wherein, is the memory component at time , with the same dimension as GPP; is a memory kernel function, which defines how past stress decays its influence over time, with the dimension of GPP / time; is a dimensionless environmental stress index at past time , which integrates temperature, moisture and other environmental stress factors, and is derived from the environmental stress data obtained by the observation data acquisition module; : time; : time, integral variable;

[0100] To achieve this, the dimensionless environmental stress index can be defined as 1 minus the nonlinear adjustment term defined in the previous embodiment , because it itself quantifies the comprehensive effect of environmental stress on the distribution of photosynthetic electron transport chain, and the calculation formula can be:

[0101] ;

[0102] Wherein, : dimensionless environmental stress index at past time ; : nonlinear adjustment term at past time ; : temperature at past time ; : water vapor pressure difference at past time ; Function The form of which has been illustrated in the above embodiments, such definition ensures that the value of stress index varies between 0, no stress, and 1, physiological activity is completely inhibited, and maintains intrinsic consistency with the physiological regulation mechanism of fluorescence to GPP conversion;

[0103] The physical meaning of integral operation is that all environmental stresses from the past to the present are multiplied by a weight that decays with time, that is, a memory kernel function, and are accumulated to obtain the overall influence on the current photosynthetic capacity;

[0104] Through this decoupling mode, the embodiment clearly separates the response processes with completely different properties in GPP; the fast-changing light response component is directly calculated, and the slow and memory components are modeled and assimilated as core state variables, which greatly simplifies the complexity of the state evolution model and enables the assimilation framework to focus on capturing and predicting slow physiological processes with memory that are essential to the stability of the ecosystem, improving the accuracy and computational efficiency of model prediction;

[0105] The scheme realizes a theoretical breakthrough in describing the response mechanism of the ecosystem to the environment; traditional models often ignore the memory effect of vegetation on environmental stress; the scheme decomposes the total GPP by constructing a state component decoupling module, and assimilates the memory component The time convolution model is used for calculation:

[0106] ;

[0107] The physical connotation of the formula is that the current photosynthetic capacity is the result of the cumulative effect of all historical environmental stresses through a memory kernel function .

[0108] The time convolution model uses a double exponential form of memory kernel function to capture the complex memory characteristics of vegetation fast recovery and slow adaptation;

[0109] The memory kernel function used in the time convolution model is designed in the embodiment;

[0110] In order to more realistically simulate the complex memory behavior of vegetation, the memory kernel function in the embodiment adopts a double exponential form; the double exponential form of memory kernel function refers to a function composed of two exponential decay terms linearly superimposed, and the design motivation is to capture the coexisting memory characteristics of two different time scales in vegetation physiological response; the mathematical form is:

[0111] ;

[0112] In the function: : memory kernel function; and are two weight coefficients, dimension of GPP / time, whose relative magnitude determines the relative importance of the two memory processes, and their initial values can be obtained from the control experiment literature of specific vegetation functional type; and are two characteristic time scales, dimension of time; where, is usually set to a small value, such as a few days, to capture the fast recovery behavior of vegetation after short-term stress relief, such as the rapid rebound of photosynthetic capacity after rainfall after drought; is set to a large value, such as several weeks to months, to capture the physiological structure adjustment of vegetation to adapt to long-term environmental changes, that is, the composite memory characteristics of slow adaptation, such as root growth or stomatal density changes to cope with persistent drought; parameter will be used as an optimization parameter in the assimilation process, and the system will learn and adjust according to the actual observation data; : time evolution; : natural logarithm base, standard mathematical constant;

[0113] Compared with the single exponential kernel function that can only describe a single decay process, the double exponential form used in this embodiment can simultaneously depict the two key memory mechanisms of fast recovery and slow adaptation of vegetation; This design makes the calculation of memory components more consistent with biological reality, and can more accurately simulate the GPP dynamics of vegetation under complex scenarios such as rapid recovery after flash drought, or slow adaptation in seasonal drought, significantly improving the performance of the model under extreme or variable climate conditions;

[0114] Especially crucial is that the memory kernel function is designed in the form of a double exponential:

[0115] ;

[0116] This is not simply a mathematical construct, but has profound physiological significance; It represents that vegetation has at least two time-scale memories: one is a fast recovery process dominated by a characteristic time , such as a few days, and the other is a slow adaptation process dominated by a characteristic time , such as several weeks to months; For example, a forest may quickly recover its GPP within a few days after a short flash drought, which corresponds to ; but if it experiences a drought that lasts throughout the season, its internal physiological structure adjustment, such as root biomass allocation, is a slow adaptation process that will last for months, which corresponds to This accurate characterization of the composite memory property enables the model to capture the realistic dynamics of the ecosystem under complex stress sequences, significantly outperforming existing single-time-scale models.

[0117] The ecosystem state critical transition model is a nonlinear stochastic differential equation based on the modification of bifurcation theory, used to describe the nonlinear response of the ecosystem under environmental stress;

[0118] The form of the ecosystem state critical transition model is limited in this embodiment;

[0119] The purpose of the ecosystem state critical transition model is to describe the dynamic evolution of the slow component and the memory component of GPP, especially the nonlinear behavior of the ecosystem when it approaches its carrying capacity or critical point; in this embodiment, the model is a nonlinear stochastic differential equation based on the modification of bifurcation theory; the bifurcation theory is a theory in mathematics that studies the qualitative change in the behavior of a system as a parameter changes, and its introduction into the ecological model can effectively describe the sudden transition of the system; the core of the equation is modified from the traditional logistic growth model, which is in the form of:

[0120] ;

[0121] To ensure the dimensional consistency of the formula, the original logistic growth term is modified to ; In the equation, the definitions of each term are as follows:

[0122] is the light response fast component, is the slow component, is the memory component, and the sum of the two constitutes the state variable of the system, and the change rate is dimensionless; is the total GPP, equal to the sum of the fast, slow and memory components; is the intrinsic growth rate of the ecosystem, dimensionless, with a dimension of 1 / time; is the maximum GPP under environmental carrying capacity, with the same dimension as GPP; is a comprehensive environmental stress index, dimensionless, calculated from the environmental stress data obtained by the observation data acquisition module; is the critical stress threshold that the ecosystem can withstand, dimensionless, which is a key parameter to be optimized in the assimilation inversion process; is a Heaviside step function, whose value is 1 when , otherwise it is 0; is a bifurcation parameter, dimensionless, which controls the speed of the system collapse after crossing the critical point; is an additive random noise term, representing random disturbances that the model cannot describe, with the same dimension as the left side of the equation;

[0123] To make the equation computable, the environmental stress index and are integrated into the model as where is the integrated environmental stress index, and is the stress index used in the memory component, to ensure that the way of quantifying environmental stress within the model is consistent, and is an additive random noise term, which is usually modeled as a Gaussian white noise process, with contributions to the state variable at discrete time steps given by where is a random variable following a standard normal distribution with mean 0 and variance 1, and is a parameter to be optimized, representing the intensity of the noise, which has the same dimension as , i.e., GPP / time;

[0124] The equation describes the evolution of the state variable ; under normal conditions , the second term in the bracket on the right side of the equation is 1, and the model behaves as a logistic-like growth; when the environmental stress exceeds a critical threshold , the function is activated, and the growth term of the equation is sharply weakened or even becomes negative, thus simulating the sudden decline or collapse of ecosystem productivity;

[0125] By introducing a nonlinear stochastic differential equation based on a modification of the bifurcation theory, the system is no longer just fitting the smooth changes of GPP, but has the ability to predict the nonlinear mutation of the ecosystem, i.e., critical transition; by continuously optimizing the parameters and and tracking the state variable during the assimilation process, the model can provide early warning of the risk of ecosystem collapse due to increasing environmental stress, providing key early warning information for the management and protection of ecosystems;

[0126] The most core progress lies in the prediction ability of the nonlinear behavior of the ecosystem; existing technology models are mostly linear or approximately linear, and cannot provide early warning of the critical transition, i.e., state mutation, that may occur in the ecosystem under continuous stress; the present scheme uses an ecosystem state critical transition model based on a modification of the bifurcation theory as the core of the state evolution prediction module:

[0127] ;

[0128] The derivation of the equation is based on the modification of the standard logistic growth model to ensure dimensional consistency and completeness of physical meaning; among them, the introduction of the term is the key; the Heaviside function constitutes the non-linear switch of the system, and the physical meaning is that when the comprehensive environmental stress index is lower than the critical threshold , the term is 1, and the system evolves normally; once the stress exceeds the critical point, the term becomes , which has a sharp inhibitory effect on the growth of the system, thereby simulating the collapse of ecosystem productivity; by continuously optimizing the critical parameters and in the adaptive assimilation inversion module, the scheme not only can invert the current GPP, but also can provide early warning of the stability of the ecosystem, which is a forward-looking function that the prior art does not have;

[0129] The growth limiting term here uses the total GPP , and the ecological hypothesis is that the resources available to support the growth of slow physiological processes ( ) are subject to instantaneous competition for current total photosynthetic output, i.e. the fast light response component Similarly, the system resources will be consumed, thereby limiting the growth of the slow component, which makes the model more realistically reflect the resource allocation and competition relationship between different time scale processes within the system.

[0130] Embodiment 5

[0131] The regional GPP assimilation inversion system fuses ground-based and multi-source satellite chlorophyll fluorescence, and is characterized in that the adaptive assimilation inversion module is specifically used for:

[0132] Building an assimilation framework with slow component and memory component as state variables;

[0133] Taking the state evolution prediction module as the state transition equation of the assimilation framework;

[0134] Taking the pure SIF observation value provided by the signal unmixing and simulation generation module as the observation input of the assimilation framework, and optimizing the state variables and model parameters through data assimilation;

[0135] This embodiment describes in detail the working mechanism of the adaptive assimilation inversion module;

[0136] The working process of the module is organized as a standard data assimilation framework, which includes three core steps; the first step is to build an assimilation framework with slow component and memory component as state variables; this means that in the assimilation system, the core to be estimated and predicted is the internal physiological state of vegetation, i.e. and It is not directly the total GPP; the assimilation framework can be used in algorithms such as sequence importance resampling particle filtering, ensemble Kalman filtering or variational assimilation.

[0137] The second step uses the state evolution prediction module as the state transition equation of the assimilation framework; this means that during the prediction phase of the assimilation algorithm, the system will invoke the aforementioned nonlinear stochastic differential equation; the equation uses the current state variables ( Using as input, the predicted state for the next time step is calculated by forward integration. ) and its uncertainties; the steps constitute the model-driven physical evolution part of the assimilation process;

[0138] The third step involves using the clean SIF observations as the input to the assimilation framework and optimizing the state variables and model parameters through data assimilation. During the update or analysis phase of the assimilation algorithm, the system introduces clean SIF observations provided by the signal demixing and simulation generation module. The assimilation algorithm compares the differences between the simulated SIF values ​​(after transformation by the observation operator) and the actual SIF observations, considering their uncertainties, to calculate the impact on the state variables. The optimal estimate; at the same time, the difference can also be used to optimize key parameters in the online optimization model, such as... These features enable the model to better learn the behavior of ecosystems in a specific region.

[0139] By constructing the assimilation module in this way, this embodiment dynamically integrates mechanistic model predictions with remote sensing data observations. Instead of simply fitting the data with a model, it allows the data to continuously constrain and correct the model's trajectory in a cyclical iterative process. This makes the GPP inversion results both conform to the basic laws of ecology, guaranteed by the state transition equation, and consistent with real-time satellite observations, thus obtaining GPP estimation results that are far more accurate and robust than single model simulations or simple empirical statistics.

[0140] The adaptive assimilation and inversion module is also used for:

[0141] Based on the dominant vegetation functional type within the pixel, the corresponding parameters are retrieved from the preset physiological parameter library;

[0142] Based on the retrieved parameters, the constraint terms in the assimilation cost function are dynamically adjusted to ensure that the model evolution conforms to the physiological laws of specific vegetation functional types.

[0143] This embodiment enhances the adaptive characteristics of the adaptive assimilation and inversion module;

[0144] The module further comprises two additional adaptive adjustment steps when performing assimilation inversion; the first step is to retrieve corresponding parameters from a preset physiological parameter library according to the dominant vegetation functional type in the pixel; the preset physiological parameter library is a pre-established database, which stores typical physiological parameters for different vegetation functional types, PFTs, such as C3 herb, C4 crop, evergreen coniferous forest, deciduous broad-leaved forest and the like; the parameters include but are not limited to: intrinsic growth rate , environmental carrying capacity , memory kernel function parameters , etc.; when processing each pixel, the system first identifies the dominant PFT according to the high-resolution land cover data, and then selects a set of most suitable initial parameters for the pixel from the library; the adaptive module in the present scheme adopts a set of high-order ecological strategy parameters (such as intrinsic growth rate k, environmental carrying capacity C, etc.) to represent the overall adaptive strategy of different vegetation functional types. These parameters are not traditional biophysical parameters, but through internal preset mapping relationship, are used to dynamically generate and optimize the initial values and ranges of specific parameters (such as V cmax25 , Ω, etc.) required by ecological process models (such as CLM), so as to realize adaptive assimilation inversion based on PFT and real-time stress state.

[0145] As a non-limiting example, the entries in the physiological parameter library can include the following contents: for C3 herb, the parameter set can be:

[0146] ; and for evergreen coniferous forest, the parameter set can be:

[0147] ;

[0148] The initial values will be further optimized in the assimilation process;

[0149] The second step is to dynamically adjust the constraint term in the assimilation cost function based on the retrieved parameters to ensure that the model evolution conforms to the physiological rules of the specific vegetation functional type; the assimilation cost function is the objective function to be minimized by the assimilation algorithm when seeking the optimal solution, and usually includes the difference term between model prediction and observation and the background error term; the innovation of the present embodiment is to introduce an additional PFT-related constraint term into the function; for example, for a pixel dominated by C3 plants, the system will enhance the constraint weight related to the temperature response curve in the cost function, because the photosynthesis of C3 plants is more sensitive to temperature changes; for a pixel dominated by C4 plants, the constraint of light saturation effect is strengthened, because C4 plants usually have a higher light saturation point; for CAM plants, their unique diurnal acidity variation rhythm can be introduced as a strong time constraint; this strategy introduces ecological prior knowledge into the optimization process in the form of mathematical constraints;

[0150] This dynamic adjustment can be achieved by modifying the assimilation cost function , a typical variational assimilation cost function is of the form:

[0151] ;

[0152] where, : assimilation cost function; superscript T: matrix transpose; superscript -1: inverse of a matrix; : observation operator; here, is the state variable, such as ; is the background field, is the observation value, i.e. the pure chlorophyll fluorescence , and are the error covariance matrices of the background and observation, respectively, while is the observation operator, which serves to convert the model state variable (i.e. ) into a quantity comparable to the observation value, i.e. the simulated chlorophyll fluorescence ; the innovation of the present application lies in the introduction and adjustment of the constraint term related to PFT ; for example, the constraint term can be designed as:

[0153] ;

[0154] where, is a function used to express a specific physiological law, while is a weight coefficient retrieved from a parameter library according to the dominant PFT type, for example, when processing C4 crop pixels, a larger value can be selected for from the parameter library, and indicates the degree to which the model state does not reach light saturation under strong light, so as to strongly constrain the model to meet the high light efficiency characteristics of C4 plants in optimization; : weight coefficient retrieved from a parameter library according to the dominant PFT type; : a function used to express a specific physiological law; : state variable;

[0155] Through this double-layer adaptive adjustment mechanism, the assimilation system realizes high intelligence and specificity; it is no longer a one-size-fits-all model and parameter processing all types of land, but for each pixel is configured to meet its main physiological characteristics of the assimilation scheme; this deep integration of macro-ecological knowledge, PFT classification and micro-mathematical model, cost function constraints greatly improves the applicability and accuracy of the model on heterogeneous land, so that the final inversion of regional GPP product in the transition between different ecosystems and internal details are more real and reliable;

[0156] The adaptive assimilation inversion module of the present scheme solves the problem of one-size-fits-all of traditional models by introducing a parameterization scheme based on vegetation functional type; the system can retrieve exclusive parameters from a pre-set physiological parameter library according to the dominant vegetation functional type in the pixel, such as C3 and C4 plants, and dynamically adjust the constraint term in the assimilation cost function; this means that the behavior pattern of the model itself will be adaptively adjusted according to the physiological characteristics of its simulation object, for example, strengthening the adaptability constraint of C4 crops to high light intensity, and enhancing the sensitivity constraint of C3 forests to temperature; this strategy of hard embedding macro-ecological knowledge in the form of mathematics into the assimilation process ensures the seamless transition between different ecosystem types and the authenticity of internal details, and its accuracy and reliability are far superior to that of traditional models using general parameters.

[0157] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present application and not to limit it, although the present application has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present application can be modified or replaced by equivalents without departing from the spirit and scope of the present application, which should be covered in the scope of the claims of the present application.

Claims

1. A regional GPP assimilation inversion system that fuses ground-based and multi-source satellite chlorophyll fluorescence, characterized in that, The method comprises the following steps: an observation data acquisition module is configured to obtain ground-based and remote sensing observations of sunlight-induced chlorophyll fluorescence, photosynthetically active radiation, high-resolution land cover data, and environmental stress data; a signal unmixing and simulation generation module is configured to unmix total fluorescence of mixed pixels into pure chlorophyll fluorescence SIF based on vegetation functional type coverage determined from high-resolution land cover data, and to drive an ecological model to generate SIF and GPP simulation values based on environmental stress data; a state component decoupling module is configured to decouple total GPP into a light response fast component, a slow component, and a memory component; a state evolution prediction module is configured to predict dynamic evolution of the slow component and the memory component based on a preset ecological system state critical transition model; an adaptive assimilation inversion module is configured to invert regional GPP by taking the slow component and the memory component as state variables, taking SIF observations as observation data, and combining prediction results of the state evolution prediction module; the signal unmixing and simulation generation module is specifically configured to: apply a preset ecological mechanism constrained mixed pixel unmixing model, combine vegetation functional type coverage, a preset canopy radiation transfer correction operator, and ecological interaction terms to calculate pure chlorophyll fluorescence SIF; A quantitative relationship from pure chlorophyll fluorescence SIF to the maximum carboxylation rate V cmax25 was established using a pre-set physiological constraint nonlinear conversion operator. the physiological constraint nonlinear conversion operator includes a nonlinear adjustment term related to temperature and water vapor pressure difference in the environmental stress data, and is used to represent the regulation of the electron transport chain by the environmental stress; the state component decoupling module is specifically configured to: determine the light response fast component based on photosynthetically active radiation obtained by the observation data acquisition module; apply a preset time convolution model to determine the memory component by integrating historical environmental stress; the ecological system state critical transition model is a nonlinear stochastic differential equation modified based on bifurcation theory, and is used to describe the nonlinear response of the ecological system under environmental stress; the adaptive assimilation inversion module is specifically configured to: build an assimilation framework taking the slow component and the memory component as state variables; use the state evolution prediction module as a state transition equation of the assimilation framework; use pure SIF observations provided by the signal unmixing and simulation generation module as observation inputs of the assimilation framework, and optimize state variables and model parameters through data assimilation; the adaptive assimilation inversion module is further configured to: retrieve corresponding parameters from a preset physiological parameter library according to a dominant vegetation functional type in the pixel; and based on the retrieved parameters, dynamically adjust constraint terms in the assimilation cost function to ensure that the model evolution conforms to the physiological rules of the specific vegetation functional type.

2. The fused foundation-multiple-source satellite chlorophyll fluorescence based regional GPP assimilative inversion system according to claim 1, wherein, The ecological interaction term is modeled as a function taking coverage, leaf area index, and exclusive competition coefficient between different vegetation functional types as inputs, and is used to describe nonlinear effects generated by resource competition or shading between different vegetation functional types.

3. The fused foundation-multiple-source satellite chlorophyll fluorescence based regional GPP assimilative inversion system according to claim 2, wherein, The time convolution model uses a double exponential form of memory kernel function to capture the composite memory characteristics of rapid recovery and slow adaptation of vegetation.

Images