Multi-model optimization integrated vegetation total primary productivity estimation method and system

Abstract

The invention discloses a multi-model optimization integrated vegetation gross primary productivity estimation method and system, and belongs to the technical field of remote sensing, and the method comprises the steps: constructing a plurality of dynamic vegetation sub-models, enabling each model to independently operate according to the initialization and state setting of each model, and carrying out the simulation to generate a GPP result of each model; model weights are optimized, GPP results simulated by all the sub-models serve as independent variables, satellite observation GPP remote sensing data in the same period serve as dependent variables, an XGBoost algorithm is applied to construct a model, hyper-parameter optimization is conducted, and the sub-model weights are obtained; forming an integrated model according to the sub-model weight; and estimating the vegetation total primary productivity of a preset long-time sequence by using the integrated model to obtain an estimation result. According to the method, the multi-model integration weight is optimized through XGBoost, and the problem that the GPP estimation precision is low due to the fact that a traditional method is single in integration mode and unreasonable in weight distribution is solved.

Description

Technical Field

[0001] The present application belongs to the field of remote sensing technology, and specifically relates to a method and system for estimating total primary productivity of vegetation based on multi-model optimization integration. Background Art

[0002] Gross Primary Productivity (GPP) is the total amount of carbon dioxide in the atmosphere converted into organic matter by vegetation in an ecosystem per unit time. It is often used to measure the total amount of carbon fixed by vegetation through photosynthesis. It is a basic indicator for measuring the carbon sequestration capacity of an ecosystem and is widely used in productivity evaluation of ecosystems such as farmland, forests, and water bodies.

[0003] Methods for estimating GPP fall into three main categories: Field observations: These methods rely on ground-based flux tower observations of GPP, such as data from sites provided by the Eddy Covariance Flux Tower Network. These methods offer high accuracy and provide high-temporal-resolution field data, but they suffer from a limited number of sites, poor spatial representativeness, and high operational and maintenance costs, making them suitable only for small-scale studies. Remote sensing estimation: These methods rely on spectral information acquired by remote sensing satellites and combine them with light utilization efficiency models (such as MODIS-GPP, GLASS-GPP, and NIRv-GPP) to invert GPP. These methods offer broad spatial coverage and are suitable for regional and global-scale estimations, but are sensitive to remote sensing data quality and model parameter settings, and lack access to long-term historical series from before satellite launch. Ecosystem process models: These methods dynamically calculate GPP by simulating ecological processes such as photosynthesis, respiration, and the water cycle. These methods, based on plant physiological mechanisms and ecological processes, can overcome the spatial and temporal limitations of observational data and, in theory, reconstruct historical GPP. However, due to differences in model parameterization schemes and data inputs, there are often significant model uncertainties and regional applicability issues.

[0004] Existing model ensemble methods can be divided into two main categories. The first category is based on multi-model ensemble averaging, which assigns equal weight to each model in the ensemble. This approach has been widely used in previous studies and aims to reduce the uncertainty of individual models by averaging the simulation results of multiple models. As a simple and easy-to-implement method, multi-model ensemble averaging has shown some effectiveness in integrating multiple dynamic global vegetation models. However, this method has certain limitations. During the ensemble, it ignores the differences in structure and simulation performance among different dynamic global vegetation models, failing to reflect the relative strengths and weaknesses of the models. Compared with traditional multi-model ensemble averaging, the second category of methods weights multiple dynamic global vegetation models based on model performance or model bias to achieve an optimal ensemble strategy. For example, the Bayesian model averaging method considers model performance information and assigns different weights to each model, providing an optimal strategy for ensembling multi-model simulations. This method has been widely used to provide optimal estimates for climate and dynamic global vegetation models. Compared with multi-model ensemble averaging or single models, ensembles based on the Bayesian model averaging method generally achieve higher simulation accuracy and stability. However, ensemble models based on Bayesian model averaging (BAM) still significantly underestimate or overestimate vegetation dynamics in some vegetation regions. In addition to BAM, numerous optimization methods have emerged in applied mathematics in recent years, such as particle swarm optimization (PSO), cuckoo search, and extreme gradient boosting. These methods have been widely applied in fields such as parameter estimation and model optimization. However, the application of these optimization algorithms to ensemble optimization of dynamic global vegetation models remains relatively rare, leaving considerable room for development and research potential.

[0005] Therefore, it is urgent to provide a solution to improve the above problems. Summary of the Invention

[0006] The purpose of the embodiments of the present application is to provide a method and system for estimating the total primary productivity of vegetation through multi-model optimization integration, which can improve the problem of low accuracy of estimation of the total primary productivity of vegetation in the existing technology due to the single model integration method and unreasonable weight distribution in the process of simulating the total primary productivity of vegetation.

[0007] In the first aspect, the present application provides a method for estimating total primary productivity of vegetation using a multi-model optimization integration method, comprising: Construct multiple dynamic vegetation sub-models, including LM5, CABLE and LPJ-GUESS models; Multiple dynamic vegetation sub-models configure the environment and system according to their respective initialization and state settings, and simulate the total primary productivity of vegetation to obtain the total primary productivity results of vegetation; The total primary productivity results were used as the independent variable, and the satellite-observed vegetation total primary productivity remote sensing data of the same period was used as the dependent variable. A model was constructed based on the XGBoost algorithm, and hyperparameter optimization was performed to obtain the sub-model weights. Integrate multiple dynamic vegetation sub-models according to sub-model weights to obtain an integrated model; The total primary productivity of vegetation in a preset long time series is estimated based on the integrated model to obtain the estimation results.

[0008] As an optional implementation of the first aspect of the present application, integration is a process of using the XGBoost algorithm to solve the minimum value based on nonlinear multivariable constraints. The specific process of integration includes: The minimum mean square error is used as the objective function, and the optimal model weight is obtained with the constraints that the weights of multiple dynamic vegetation sub-models are greater than 0 and the sum of the weights is 1. The minimum mean square error is expressed as the performance of optimizing the GPP simulated by the integrated dynamic vegetation model to approximate the GPP observed by satellite. Obtain the weight coefficients of each dynamic vegetation sub-model according to the optimal model weight; The predicted value of the total primary productivity of vegetation of each dynamic vegetation sub-model is multiplied by its corresponding weight coefficient and weighted to obtain the optimal estimated value of the total primary productivity of vegetation of the integrated model. The mathematical expression of the optimal estimated value of the total primary productivity of vegetation is: ; in, For the The weight of each sub-model, For the Sub-model for samples The predicted value of represents the optimal estimate of the total primary productivity of vegetation from the integrated model, Indicates the number of sub-models.

[0009] As an optional implementation of the first aspect of the present application, the total primary productivity result is used as an independent variable, the remote sensing data of vegetation total primary productivity observed by satellite in the same period is used as a dependent variable, a model is constructed based on the XGBoost algorithm, and hyperparameter optimization is performed to obtain sub-model weights, including: The simulation results of sub-models at different spatial scales and remote sensing data from satellite observations are unified to the same spatial resolution using the nearest neighbor interpolation method. The total primary productivity of vegetation simulated by multiple sub-models within a preset time period is used as the independent variable, and the remote sensing data of total primary productivity of vegetation in the same period is used as the dependent variable. The database for training the integrated model is constructed, and the integrated model is constructed based on the XGBoost algorithm. The total primary productivity of vegetation simulated by multiple sub-models is two-dimensional raster data. The data of any year within the preset time period is used as a validation set to evaluate the generalization ability of the obtained sub-model weights on independent years. The data of the remaining years are synthesized as a training set to fit the mapping relationship between the sub-model simulation results and satellite observation data. Multiple rounds of training and validation were conducted using each year's data as a validation set. In each round of training, the XGBoost algorithm was used to learn the training set, establish the relationship between the sub-model simulation values and satellite observations, and perform hyperparameter tuning through random grid search to obtain multiple model weights to be verified. Based on the mean square error between the simulation results of the total primary productivity of vegetation integrated by the sub-model and the remote sensing data of the total primary productivity of vegetation observed by satellite and the preset constraints, the weights of multiple models to be verified are verified to obtain the sub-model weights.

[0010] As an optional implementation of the first aspect of the present application, the mathematical expression of the mean square error between the vegetation gross primary productivity simulation results integrated by the sub-model and the vegetation gross primary productivity remote sensing data observed by satellite is: ; in, is the minimum mean square error, represents the number of pixels in the total time period, Indicates a point in time, Indicates at a point in time Upper position The true value of Indicates at a point in time Upper position The predicted value of .

[0011] As an optional implementation of the first aspect of the present application, the mathematical expression of the preset constraint condition is: ; ; in, represents the total number of sub-models, Indicates the sequence number of the sub-model, Indicates the The sub-model simulates Line and The weight of the total primary productivity of the column.

[0012] As an optional implementation of the first aspect of the present application, the process of estimating the total primary productivity of vegetation over a preset long time series based on the integrated model and obtaining the estimation result includes: Store the trained integrated model and parameter settings as a portable model file in the joblib format. Download and preprocess the input dataset for the target time period; Based on the portable model file and input data set, the optimal estimate of total primary productivity for the target time period is estimated. This optimal estimate is the estimation result.

[0013] In the second aspect, the present application provides a multi-model optimization integrated vegetation gross primary productivity estimation system, comprising: Dynamic vegetation sub-model module, used to build multiple dynamic vegetation sub-models based on LM5, CABLE and LPJ-GUESS models; The vegetation gross primary productivity result acquisition module is used for multiple dynamic vegetation sub-models to simulate vegetation gross primary productivity according to their respective initialization and state settings, configure the environment and system, and obtain vegetation gross primary productivity results; The sub-model weight generation module is used to use the total primary productivity results as the independent variable and the vegetation total primary productivity remote sensing data observed by satellites in the same period as the dependent variable, build a model based on the XGBoost algorithm, and perform hyperparameter optimization to obtain the sub-model weights; A model integration module is used to integrate multiple dynamic vegetation sub-models according to sub-model weights to obtain an integrated model; The estimation result generation module is used to estimate the total primary productivity of vegetation for a preset long time series based on the integrated model to obtain the estimation result.

[0014] In a third aspect, an embodiment of the present application provides an electronic device comprising a processor, a memory, and a program or instruction stored in the memory and executable on the processor, wherein the program or instruction, when executed by the processor, implements the steps of the method of the first aspect.

[0015] In a fourth aspect, an embodiment of the present application provides a readable storage medium, on which a program or instruction is stored, and when the program or instruction is executed by a processor, the steps of the method of the first aspect are implemented.

[0016] Compared with the prior art, the multi-model optimization integrated vegetation gross primary productivity estimation method provided by the present invention has the following beneficial effects: First, the present invention dynamically calculates the weights of each model through the XGBoost algorithm, and continuously optimizes the weight distribution based on satellite observation data verification, so that the final integrated model approaches the optimal solution, effectively reduces the systematic deviation of a single model, and significantly improves the accuracy of vegetation gross primary productivity inversion; Second, for different regions or different vegetation types, the present invention automatically strengthens the model that performs best in the current scenario by dynamically adjusting weights, avoiding the limitations of "one-size-fits-all" integration for the entire region and achieving optimization for each scenario; based on the integrated model, the historical, present and future gross primary productivity index of vegetation can be seamlessly estimated to support diverse needs such as climate change research, disaster assessment, and precision agriculture management. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 A flowchart showing a method for estimating total primary productivity of vegetation using multi-model optimization integration provided by the first embodiment of the present invention; Figure 2 A flowchart showing a weight optimization process for multiple sub-models provided by the first embodiment of the present invention; Figure 3 A flowchart showing the integration of multiple sub-models provided by the first embodiment of the present invention; Figure 4 A schematic structural diagram of a multi-model optimization integrated vegetation gross primary productivity estimation system provided by the second embodiment of the present invention. DETAILED DESCRIPTION

[0018] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are part of the embodiments of this application, not all of them. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0019] The terms "first," "second," and the like in the specification and claims of this application are used to distinguish similar objects and are not used to describe a particular order or precedence. It should be understood that the terms used in this manner are interchangeable where appropriate so that the embodiments of this application can be implemented in an order other than those illustrated or described herein. In addition, the term "and / or" in the specification and claims refers to at least one of the connected objects, and the character " / " generally indicates that the objects associated with each other are in an "or" relationship.

[0020] In the following, in conjunction with the accompanying drawings, a multi-model optimization integrated vegetation gross primary productivity estimation method and system provided in the embodiment of the present application is described in detail through specific embodiments and their application scenarios.

[0021] Example 1 See also Figure 1 , which represents a flow chart of a multi-model optimization integrated vegetation gross primary productivity estimation method provided by the present invention, including steps S1 to S5.

[0022] S1: Construct multiple dynamic vegetation sub-models, including LM5, CABLE, and LPJ-GUESS models.

[0023] Specifically, in this example, multiple sub-models are selected, driving data and important process selections are prepared, and preliminary preparations for each sub-model are completed. Specifically: S11. Select several typical dynamic vegetation models as sub-models, among which CLM5, CABLE, and LPJ-GUESS have advantages in carbon-nitrogen coupling, biological process simulation, and hydrothermal condition response mechanism.

[0024] S12. Based on the geographic scope of the study area and the set time range (e.g., Yangtze River Economic Belt 2001–2023), obtain historical input data that matches the model operation. This primarily includes climate data such as the CRU TS (Climatic Research Unit gridded Time Series); atmospheric carbon dioxide concentrations, such as long-term carbon dioxide observations released by NOAA; and land use change data, such as the LUH2 (Land Use Harmonization v2) dataset provided by Hurtt.

[0025] S13. Based on the differences in modeling ecological or anthropogenic mechanisms among the sub-models, ecological process modules relevant to the research objectives are enabled. These primarily include land-use change processes, such as deforestation and regeneration after agricultural abandonment, crop rotation, and farmland harvesting; human- and natural-driven ecological processes, such as timber harvesting, forest degradation, peat fires, and fire simulation or suppression; and biogeochemical processes, such as the carbon and nitrogen cycles and their interactions. CLM5 supports the complete process of forest dynamics after agricultural abandonment, crop rotation, fire disturbance, and carbon-nitrogen interactions; CABLE can simulate farmland harvesting, forest disturbance, and carbon-nitrogen processes, making it suitable for semi-natural ecosystems; and LPJ-GUESS can simulate natural fires, regenerated forests, and farmland dynamics, emphasizing long-term ecological succession mechanisms.

[0026] Specifically, the submodel description table is shown in Table 1. Through the process configuration in Table 1, the core parameters required for each submodel are initialized and functional modules are enabled, providing accurate and complete input conditions for subsequent vegetation gross primary productivity simulations. The parameters are: DFRAA (deforestation and forest regeneration after agricultural abandonment); WHFD (wood harvest and forest degradation); SC (forest-farm rotation); CH (crop harvest); PF (peat fire); and FSS (fire simulation or fire suppression).

[0027] Table 1 Sub-model description table Specifically, when executing step S11, relevant land use change processes, human- or naturally-driven ecological processes, and biogeochemical processes are selected based on the sub-model's consideration of ecological or anthropogenic mechanisms. The present invention selects multiple typical dynamic vegetation models as sub-models, including: CLM5 (Community Land Model version 5), CABLE (Community Atmosphere Biosphere Land Exchange), and LPJ_GUESS (Lund–Potsdam–JenaGeneral Ecosystem Simulator).

[0028] Furthermore, the important processes of CLM5 (Community Land Model version 5) include: deforestation and forest regeneration after agricultural abandonment, crop rotation, farmland harvesting, timber harvesting and forest degradation, peat fire, fire simulation or fire suppression, and carbon-nitrogen interaction; the important processes of CABLE (Community Atmosphere Biosphere Land Exchange) include: deforestation and forest regeneration after agricultural abandonment, timber harvesting and forest degradation, farmland harvesting, fire simulation or fire suppression, and carbon-nitrogen interaction; the important processes of LPJ_GUESS (Lund–Potsdam–Jena General Ecosystem Simulator) include: deforestation and forest regeneration after agricultural abandonment, farmland harvesting, fire simulation or fire suppression.

[0029] Furthermore, when executing step S12, the present invention obtains historical climate data (such as precipitation, temperature, and solar radiation) and atmospheric carbon dioxide concentration data based on the geographic scope of the study area and the set time range. The historical climate data (such as temperature, precipitation, and solar radiation) can be obtained from the U.S. Environmental Prediction Center (ftp: / / nacp.ornl.gov / synthesis / 2009 / Frescati / model_driver / cru_ncep) with a spatial resolution of 0.5° × 0.5°. Atmospheric carbon dioxide concentration data and ice core carbon dioxide records can be obtained from the Earth System Research Laboratory (ESRL) of the National Oceanic and Atmospheric Administration (NOAA) (http: / / www.esrl.noaa.gov / gmd / ccgg / trends / ).

[0030] Furthermore, during step S13, the present invention uses the historical climate data (e.g., precipitation, temperature, and solar radiation) and atmospheric carbon dioxide concentration data from step S12, along with the land use data obtained in step S13, to construct the external environmental conditions required for the sub-model. The land use data can be obtained from the Hurtt land use dataset (ftp: / / ftp.pbl.nl / hyde / tmp / 2017), with a spatial resolution of 0.5° × 0.5°.

[0031] In summary, step S1 of the present invention is the preliminary preparation work for the integration of multi-source models, which ensures that the selected model has scientific external driving conditions and ecological process settings, and can improve the authenticity and adaptability of the simulation of vegetation gross primary productivity.

[0032] Step S2: Multiple dynamic vegetation sub-models configure the environment and system according to their respective initialization and state settings, and each simulates the total primary productivity of vegetation to obtain the total primary productivity of vegetation.

[0033] Specifically, the specific process of simulating the total primary productivity of vegetation by configuring the environment and system according to the initialization and state settings of the multiple dynamic vegetation sub-models includes steps S21 to S23, which are as follows: S21. According to the operating structure of the sub-model, configure the system and environment suitable for the operation of the sub-model, including the Linux system.

[0034] S22. Based on the configured environment and system, download the sub-model source code, call the corresponding driving data set and important processes, combine the initialization parameters, use the respective mathematical models and algorithm frameworks, drive the calculation of vegetation physiological and ecological processes within the sub-model, and realize the multi-model optimization integrated vegetation total primary productivity simulation for the set time period in the study area.

[0035] S23. Configure the required dependencies and compilation environment for each model in a Linux system environment, download and compile the source code for the three sub-models: CLM5, CABLE, and LPJ-GUESS. Subsequently, convert and spatially crop the annual driving data for the Yangtze River Economic Belt (including climate data, CO2 concentration data, and land use data) from 2001 to 2023 and load them into the corresponding model frameworks. Simulation tasks are executed based on the initialization state and operating settings of each model, ultimately outputting simulation results for gross primary productivity from 2001 to 2023, providing basic data support for subsequent integrated modeling.

[0036] In summary, step S2 of the present invention emphasizes the standardization of model operation mechanisms and the uniformity of the environment. By establishing a standardized data pipeline, inconsistent interfaces between models can be effectively avoided, improving the comparability of simulation data and the effectiveness of subsequent integration processes. Furthermore, ensuring that the model operates within a unified system environment can avoid operational errors and improve repeatability.

[0037] Step S3: Using the total primary productivity results as the independent variable and the satellite-observed vegetation total primary productivity remote sensing data of the same period as the dependent variable, a model is constructed based on the XGBoost algorithm, and hyperparameter optimization is performed to obtain the sub-model weights.

[0038] Specifically, the present invention uses the total primary productivity results of vegetation simulated by multiple sub-models as independent variables, and the remote sensing data of total primary productivity of vegetation observed by satellites in the same period as the dependent variable, constructs a model based on the XGBoost algorithm, and performs hyperparameter optimization to obtain sub-model weights. The specific process includes steps S31 to S34.

[0039] S31. The nearest neighbor interpolation method is used to unify the spatial resolution of the sub-model simulation results and satellite observation remote sensing data at different spatial scales. Then, the total primary productivity of vegetation simulated by multiple sub-models within a set time period is used as the independent variable, and the total primary productivity of vegetation remote sensing data of the same period is used as the dependent variable to construct an integrated model for training the database, and the integrated model is constructed based on the XGBoost algorithm.

[0040] S32. In a long-term database, the data of a certain year is used as a validation set, and the data of the remaining years are synthesized as a training set. The training set is used to fit the mapping relationship between the sub-model simulation results and the satellite observation data to obtain the sub-model weights to be verified. The validation set is used to evaluate the generalization ability of the obtained sub-model weights on independent years to verify the sub-model weights to be verified. By using each year's data as a validation set in turn, multiple rounds (i.e., the total number of years) of validation are carried out to obtain the sub-model weights. Specifically, the specific process of obtaining the sub-model weight in step S32 includes: S321. In each round of training, the XGBoost algorithm is used to learn the training set and establish the relationship between the sub-model simulation value and the satellite observation value.

[0041] S322. Perform hyperparameter optimization using randomized grid search, using the mean squared error (MSE) between the GPP simulation results of the sub-model ensemble and the GPP remote sensing data observed by satellite as the objective function. The optimized model weights are obtained by constraining the weights of multiple dynamic vegetation sub-models to be greater than 0 and sum to 1. The objective function is the minimum mean squared error, which is expressed as the performance of optimizing the GPP simulated by the ensemble dynamic vegetation model to approximate the GPP observed by satellite. The mathematical expression of the objective function is: ; The constraint condition needs to satisfy that all weights are non-negative and the sum of weights is 1. The mathematical expression of the constraint function is: ; ; in, is the minimum mean square error, represents the number of pixels in the total time period, Indicates a point in time, Indicates at a point in time Upper position The true value of Indicates at a point in time Upper position The predicted value of represents the total number of sub-models, Indicates the sequence number of the sub-model, Indicates the The sub-model simulates Line and The weight of the total primary productivity of the column.

[0042] S33. Leave-One-Year-Out Cross-Validation was used to validate the model weights. The following three statistical indicators were used to evaluate the performance of the model optimization ensemble: coefficient of determination (R 2), root mean square error (RMSE) and mean relative error (MRE).

[0043] S34. Through the optimization and verification process of steps S31 to S33 above, each year's data is used as a validation set in turn, and a total of N groups (i.e., the total number of years) of different model weights can be obtained. According to the performance of statistical indicators between the integrated simulation results corresponding to each group of weights and the satellite observation data, a group of weights with high fit (large R²) and small error (low RMSE and MRE) are selected as sub-model weights to form the final model integration weight distribution scheme.

[0044] See also Figure 2 The following figure shows a flowchart of the weight optimization process for multiple sub-models provided by the present invention. To facilitate data integration and analysis, after obtaining sub-model simulation results, the dynamic vegetation model simulation results and satellite observation data at different spatial scales are unified to the same spatial resolution using the nearest neighbor interpolation method. Subsequently, the GPP simulation results of multiple sub-models for the study period (e.g., 2001–2023) are used as input independent variables, while satellite-observed GPP remote sensing data for the same period are introduced as dependent variables to construct the dataset required for XGBoost algorithm training. Data from a specific year in the dataset is then used as the validation set, while data from the remaining years constitute the training set. For example, data from 2001 is used as the validation set, while data from the remaining years (2002–2023) constitute the training set. In each round of training, the XGBoost algorithm is used to learn from the training set, establishing a relationship between the GPP sub-model simulation values and the satellite-observed GPP values. Hyperparameter optimization is then performed using a random grid search. The specific parameter settings are shown in Table 2. The random grid search objective is to minimize the mean squared error (MSE) between the total primary productivity (TPP) simulated by the sub-model ensemble and the TPP observed by satellite. During the model weight optimization process, the following constraints must be met: all weights are non-negative and their sum is 1.

[0045] Table 2 Random grid search parameter settings Furthermore, after the model weight optimization is completed, the present invention uses the leave-one-year-out cross-validation to verify the optimized model weights and uses the following three statistical indicators to evaluate the performance of the model optimization integration. These three statistical indicators are the coefficient of determination (R2 ), root mean square error (RMSE), and mean relative error (MRE). Using each year as a validation set, they generated 23 different sets of model weights for the Yangtze River Economic Belt. Based on the statistical performance of the ensemble simulation results and satellite observations corresponding to these 23 weight sets, they selected the optimal set of weights with high goodness of fit (large R²) and low error (low RMSE and MRE) for the final model ensemble.

[0046] In summary, the beneficial effects of step S3 of the present invention are: first, the present invention can improve the robustness and adaptability of the model through the training and verification process. Second, the XGBoost algorithm has strong feature selection ability and generalization performance, and can realize model contribution evaluation through automatic learning, thereby achieving more accurate model fusion weight setting. The present invention introduces a machine learning algorithm (XGBoost) to realize the nonlinear mapping relationship between multi-model output and satellite observation data, breaking the limitations of traditional linear weighting methods. Step S4: Integrate multiple dynamic vegetation sub-models according to sub-model weights to obtain an integrated model.

[0047] Specifically, in step S4, the process of obtaining the integrated model includes: performing weighted integration on multiple dynamic vegetation sub-models according to the sub-model weights to obtain the integrated model. The detailed process is as follows: See also Figure 3 , which represents the flow chart of the integration of multiple sub-models provided by the present invention, including: calculating the weights of the three sub-models LM5, CABLE and LPJ-GUESS based on the optimal parameters according to the XGBoost algorithm, the weights of each sub-model are non-negative, and the sum of the weights is equal to 1. Then, the predicted value of the total primary productivity of vegetation of the sub-model is multiplied by its corresponding weight coefficient, and then all the weighted prediction results are superimposed to finally obtain the optimal estimated value of the total primary productivity of vegetation of the integrated model. Next, the predicted value of the total primary productivity of vegetation of each sub-model for the sample is multiplied by the weight coefficient. Its corresponding weight Multiply and superimpose the weighted results of all sub-models to finally obtain the sample The integrated prediction value of vegetation gross primary productivity obtained in this way not only retains the advantages of each model, but also takes into account the complementarity and structural differences between models, providing a more reliable and stable technical support for the long-term, multi-temporal and multi-spatial scale estimation of vegetation gross primary productivity.

[0048] In summary, this step emphasizes integration through the weights of each sub-model. The integration of multiple dynamic vegetation model simulations can be viewed as a nonlinear multivariate constrained minimization problem using the XGBoost algorithm. On a pixel-by-pixel basis, the objective function is the minimum mean squared error (MSE), and the constraints are that the weights of the multiple dynamic vegetation sub-models are greater than 0 and their sum is 1. The mean squared error represents the performance of the optimized ensemble dynamic vegetation model simulated GPP approximating the satellite-observed GPP. Based on the XGBoost algorithm with the optimal parameter configuration, the weights of the three sub-models, CLM5, CABLE, and LPJ-GUESS, are calculated. Introducing the XGBoost algorithm's feature gain metric to evaluate the sub-model weights for estimating gross primary productivity (GPP) significantly improves the prediction accuracy of the model fusion while also ensuring good interpretability and generalizability. Furthermore, the use of a normalized weight distribution mechanism helps maintain the stability of the model fusion process, preventing certain sub-models from being overly amplified or marginalized in the ensemble, thereby achieving a more robust ensemble.

[0049] Step S5: Estimate the total primary productivity of vegetation for a preset long time series based on the integrated model to obtain an estimation result.

[0050] Specifically, the specific process of obtaining the estimation results includes: S51. Storing the trained integrated model and parameter settings as a portable model file, such as a joblib format file; S52, downloading and preprocessing the input data set within the target time period; S53. Based on the portable model file and the input data set, an optimal estimated value of the total primary productivity in the target time period is estimated. The optimal estimated value is the estimation result.

[0051] The process of performing step S53 to estimate the optimal estimated value of the total primary productivity in the target time period includes: ; in, For the The weight of each sub-model, For the Sub-model for samples The predicted value of Indicates the number of sub-models, It represents the optimal estimated value of the total primary productivity of vegetation of the integrated model, which is the estimated result.

[0052] In summary, the focus of step S5 is to apply the estimation capabilities of combining machine learning with ecological models to practical long-term environmental monitoring tasks. Compared to static regression or single-model approaches, this method is transferable, replicable, and produces stable outputs over the long term. It is suitable for dynamic monitoring of a variety of ecosystems and provides a powerful tool for global change research, carbon sink assessment, and agricultural remote sensing applications.

[0053] Example 2 See also Figure 4 , which shows a structural diagram of a multi-model optimization integrated vegetation gross primary productivity estimation system provided by the second embodiment of the present invention, including: The dynamic vegetation sub-model module 100 is used to construct multiple dynamic vegetation sub-models based on the LM5, CABLE and LPJ-GUESS models; The vegetation gross primary productivity result acquisition module 200 is used for multiple dynamic vegetation sub-models to simulate vegetation gross primary productivity according to their respective initialization and state settings, configure the environment and system, and obtain vegetation gross primary productivity results; The sub-model weight generation module 300 is used to use the total primary productivity results as an independent variable and the vegetation total primary productivity remote sensing data observed by satellites during the same period as a dependent variable, build a model based on the XGBoost algorithm, and perform hyperparameter optimization to obtain the sub-model weights; A model integration module 400 is configured to integrate multiple dynamic vegetation sub-models according to sub-model weights to obtain an integrated model; The estimation result generating module 500 is used to estimate the total primary productivity of vegetation of a preset long time series based on the integrated model to obtain an estimation result.

[0054] In the embodiment of the present application, a multi-model optimization integrated vegetation gross primary productivity estimation system can be a device, or a component, integrated circuit, or chip in a terminal. The device can be a mobile electronic device or a non-mobile electronic device. For example, the mobile electronic device can be a mobile phone, a tablet computer, a laptop computer, a PDA, an in-vehicle electronic device, a wearable device, an ultra-mobile personal computer (UMPC), a netbook, or a personal digital assistant (PDA), etc. The non-mobile electronic device can be a server, a network attached storage (NAS), a personal computer (PC), a television (TV), an ATM, or an kiosks, etc., which are not specifically limited in the embodiment of the present application.

[0055] In the embodiments of the present application, a multi-model optimization integrated vegetation gross primary productivity estimation system can be a device having an operating system. The operating system can be an Android operating system, an iOS operating system, or other possible operating systems, which are not specifically limited in the embodiments of the present application.

[0056] The multi-model optimization integrated vegetation total primary productivity estimation system provided in the embodiment of the present application can achieve Figures 1 to 3 In the method embodiment, each process of implementing a multi-model optimization integrated vegetation gross primary productivity estimation method is not described here to avoid repetition.

[0057] Optionally, an embodiment of the present application also provides an electronic device, including a processor, a memory, and a program or instruction stored in the memory and executable on the processor. When the program or instruction is executed by the processor, each process of the above-mentioned multi-model optimization integrated vegetation total primary productivity estimation method embodiment is implemented, and the same technical effect can be achieved. To avoid repetition, it will not be repeated here.

[0058] An embodiment of the present application also provides a readable storage medium, on which a program or instruction is stored. When the program or instruction is executed by a processor, each process of the above-mentioned multi-model optimization integrated vegetation total primary productivity estimation method embodiment is implemented, and the same technical effect can be achieved. To avoid repetition, it will not be repeated here.

[0059] The processor is the processor in the electronic device described in the above embodiment. The readable storage medium includes a computer-readable storage medium, such as a computer read-only memory (ROM), random access memory (RAM), a magnetic disk, or an optical disk.

[0060] It should be noted that, in the present invention, the terms "comprise", "include" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device comprising a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, an element defined by the sentence "comprises a ..." does not exclude the presence of other identical elements in the process, method, article or device comprising the element. In addition, it should be pointed out that the scope of the methods and devices in the embodiments of the present application is not limited to performing functions in the order shown or discussed, and may also include performing functions in a substantially simultaneous manner or in the opposite order according to the functions involved. For example, the described method may be performed in an order different from that described, and various steps may also be added, omitted, or combined. In addition, the features described with reference to certain examples may be combined in other examples.

[0061] Through the description of the above embodiments, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus the necessary general hardware platform. Of course, they can also be implemented by hardware, but in many cases the former is a more preferred embodiment. Based on this understanding, the technical solution of this application, or the part that contributes to the existing technology, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk), and includes a number of instructions for enabling a terminal (which can be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in each embodiment of this application.

[0062] The embodiments of the present application are described above in conjunction with the accompanying drawings, but the present application is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the guidance of this application, ordinary technicians in this field can also make many forms without departing from the purpose of this application and the scope of protection of the claims, all of which are within the protection of this application.

Claims

1. A method for estimating total primary productivity of vegetation based on multi-model optimization integration, characterized in that: include: Constructing multiple dynamic vegetation sub-models, wherein the dynamic vegetation sub-models include LM5, CABLE and LPJ-GUESS models; The multiple dynamic vegetation sub-models configure the environment and system according to their respective initialization and state settings, and each simulates the total primary productivity of vegetation to obtain the total primary productivity of vegetation; The total primary productivity results were used as the independent variable, and the satellite-observed vegetation total primary productivity remote sensing data of the same period was used as the dependent variable. A model was constructed based on the XGBoost algorithm, and hyperparameter optimization was performed to obtain sub-model weights. Integrating the multiple dynamic vegetation sub-models according to the sub-model weights to obtain an integrated model; The total primary productivity of vegetation in a preset long time series is estimated based on the integrated model to obtain an estimation result.

2. The method for estimating total primary productivity of vegetation based on multi-model optimization integration according to claim 1, characterized in that: The integration is a process of using the XGBoost algorithm to solve the minimum value based on nonlinear multivariable constraints. The specific process of the integration includes: The optimal model weight is obtained using the minimum mean square error as the objective function and the constraints that the weights of the multiple dynamic vegetation sub-models are greater than 0 and the sum of the weights is 1, wherein the minimum mean square error is expressed as the performance of the optimized integrated dynamic vegetation model simulated GPP approximating the satellite observed GPP; Obtaining weight coefficients of each dynamic vegetation sub-model according to the optimal model weight; The predicted value of the total primary productivity of vegetation of each dynamic vegetation sub-model is multiplied by its corresponding weight coefficient and weighted to obtain the optimal estimated value of the total primary productivity of vegetation of the integrated model, wherein the mathematical expression of the optimal estimated value of the total primary productivity of vegetation is: ； in, For the The weight of each sub-model, For the Sub-model for samples The predicted value of represents the optimal estimate of the total primary productivity of vegetation from the integrated model, Indicates the number of sub-models.

3. The method for estimating total primary productivity of vegetation based on multi-model optimization integration according to claim 1, characterized in that: The process of using the gross primary productivity results as the independent variable and the satellite-observed vegetation gross primary productivity remote sensing data of the same period as the dependent variable, building a model based on the XGBoost algorithm, and performing hyperparameter optimization to obtain the sub-model weights includes: Unifying the simulation results of the sub-models at different spatial scales and the remote sensing data observed by satellite to the same spatial resolution according to the nearest neighbor interpolation method; The total primary productivity of vegetation simulated by multiple sub-models within a preset time period is used as an independent variable, and the remote sensing data of the total primary productivity of vegetation during the same period is used as a dependent variable to construct an integrated model for training a database, and the integrated model is constructed based on the XGBoost algorithm, wherein the total primary productivity of vegetation simulated by the multiple sub-models is two-dimensional raster data; The data of any year within the preset time period is used as a validation set to evaluate the generalization ability of the sub-model weights on independent years, and the data of the remaining years are synthesized as a training set to fit the mapping relationship between the sub-model simulation results and the satellite observation data; Each year's data is used as a validation set, and multiple rounds of training and validation are carried out. In each round of training, the XGBoost algorithm is used to learn the training set, establish the relationship between the sub-model simulation value and the satellite observation value, and perform hyperparameter tuning through random grid search to obtain multiple model weights to be verified; The plurality of model weights to be verified are verified based on the mean square error between the simulation results of the total primary productivity of vegetation integrated by the sub-models and the remote sensing data of the total primary productivity of vegetation observed by satellite and the preset constraints to obtain the sub-model weights.

4. The method for estimating total primary productivity of vegetation based on multi-model optimization integration according to claim 3, characterized in that: The mathematical expression of the mean square error between the vegetation gross primary productivity simulation results of the sub-model integration and the vegetation gross primary productivity remote sensing data observed by satellite is: ； in, is the minimum mean square error, represents the number of pixels in the total time period, Indicates a point in time, Indicates at a point in time Upper position The true value of Indicates at a point in time Upper position The predicted value of .

5. The method for estimating total primary productivity of vegetation based on multi-model optimization integration according to claim 3, characterized in that: The mathematical expression of the preset constraint condition is: ； ； in, represents the total number of sub-models, Indicates the sequence number of the sub-model, Indicates the The sub-model simulates Line and The weight of the total primary productivity of the column.

6. The method for estimating total primary productivity of vegetation based on multi-model optimization integration according to claim 1, characterized in that: The process of estimating the total primary productivity of vegetation over a preset long time series based on the integrated model and obtaining the estimation result includes: The trained integrated model and parameter settings are stored as a portable model file, wherein the portable model file is a joblib format file; Download and preprocess the input dataset for the target time period; An optimal estimated value of total primary productivity for a target time period is estimated based on the portable model file and the input data set, where the optimal estimated value is the estimation result.

7. A multi-model optimization integrated vegetation gross primary productivity estimation system, characterized by: include: Dynamic vegetation sub-model module, used to build multiple dynamic vegetation sub-models based on LM5, CABLE and LPJ-GUESS models; The module for obtaining the results of vegetation gross primary productivity is used for multiple dynamic vegetation sub-models to simulate vegetation gross primary productivity according to their respective initialization and state settings, configure the environment and system, and obtain vegetation gross primary productivity results; A sub-model weight generation module is used to use the total primary productivity results as an independent variable and the vegetation total primary productivity remote sensing data observed by satellite in the same period as a dependent variable, build a model based on the XGBoost algorithm, and perform hyperparameter optimization to obtain sub-model weights; A model integration module, configured to integrate the plurality of dynamic vegetation sub-models according to the sub-model weights to obtain an integrated model; The estimation result generating module is used to estimate the total primary productivity of vegetation of a preset long time series based on the integrated model to obtain an estimation result.

8. An electronic device, characterized in that: It includes a processor, a memory, and a program or instruction stored in the memory and executable on the processor. When the program or instruction is executed by the processor, the steps of a multi-model optimization integrated vegetation total primary productivity estimation method as described in any one of claims 1 to 6 are implemented.

9. A readable storage medium, characterized in that: The readable storage medium stores a program or instruction, and when the program or instruction is executed by the processor, the steps of the method for estimating the total primary productivity of vegetation by multi-model optimization integration are implemented as described in any one of claims 1-6.

Images