Method for optimizing water carbon flux based on carbonyl sulfur and light-induced chlorophyll fluorescence

Abstract

The application discloses a water-carbon flux optimization method based on carbonyl sulfide (COS) and sun / solar-induced chlorophyll fluorescence (SIF), and belongs to the technical field of land ecosystem model optimization and application. The application combines COS and SIF observation information, adopts a four-dimensional variation algorithm to perform data assimilation on observation data and simulation data, simultaneously restrains plant photosynthesis processes and gas transmission processes, and then realizes joint optimization of carbon flux and water flux. Meanwhile, the application compares and verifies simulation data before and after optimization and observation data, and proves effectiveness and reliability of the method.

Description

Technical Field

[0001] This invention belongs to the field of terrestrial ecosystem model optimization and application technology, and more specifically, relates to a method for optimizing water and carbon flux based on carbonyl sulfur and sunlight-induced chlorophyll fluorescence. Background Technology

[0002] Against the backdrop of global change and my country's announcement of its "Dual Carbon Plan," it is imperative to strengthen research on the water and carbon cycles and enhance our understanding of their responses and feedbacks under the dual impacts of climate change and human activities. Models are an effective method for expressing formalized knowledge about the Earth system and can be used to simulate and predict key variables in the Earth system, including water and carbon fluxes. However, models themselves are only approximations of truth and inevitably contain inherent uncertainties. In the current era of big data, observational data is becoming increasingly abundant and multi-sourced. Using data assimilation techniques, direct and indirect observations from different sources and at different resolutions can be integrated within the dynamic framework of models, thereby enhancing the predictability and observability of the system.

[0003] Carbonyl sulfide (COS) is a trace gas with properties similar to CO2. Both are absorbed by enzymes involved in photosynthesis via the same pathway, but plants do not produce COS. Therefore, carbonyl sulfide observation can be used to estimate plant photosynthetic capacity, stomatal conductance, and other parameters. Sun / Solar-induced chlorophyll fluorescence (SIF) is the spectral signal emitted by the photosynthetic center of a plant under sunlight conditions, directly reflecting the dynamic changes in actual photosynthesis. Summary of the Invention

[0004] 1. The problem to be solved

[0005] To address the problem that existing methods using only COS or SIF data are insufficient for effectively constraining plant photosynthesis and gas transport processes, this invention provides a method for optimizing water and carbon fluxes based on carbonyl sulfide and light-induced chlorophyll fluorescence. By combining COS and SIF observation information, both plant photosynthesis and gas transport processes are constrained simultaneously, thereby achieving joint optimization of carbon and water fluxes.

[0006] 2. Technical Solution

[0007] To address the aforementioned issues, we have, for the first time, proposed and successfully implemented a joint assimilation method of COS and SIF based on data assimilation technology and the BEPS (Boreal Ecosystem Productivity Simulator) ecosystem model, effectively optimizing water and carbon fluxes. The technical solution adopted in this invention is as follows:

[0008] A method for optimizing water and carbon flux based on carbonyl sulfide and sunlight-induced chlorophyll fluorescence includes the following steps:

[0009] S1 runs the BEPS model based on prior model parameters and initial conditions to obtain prior simulation variables; the initial conditions include meteorological data, boundary conditions, and leaf area index (LAI) data.

[0010] S2 uses a four-dimensional variation (4D-Var) algorithm to assimilate the observed data with the prior simulation data (simulation data is the set of simulation variables) to obtain the optimized model parameters; the observed data includes observed COS data and observed SIF data, and the prior simulation data includes the prior COS data and prior SIF data in the prior simulation variables obtained in step S1.

[0011] S3 uses the optimized model parameters to re-drive the BEPS model under the initial conditions, obtaining optimized simulation data including variables related to water and carbon fluxes.

[0012] Preferably, in step S1, there are a total of 76 prior model parameters. Some of these parameters change with the vegetation function type (PFT) or soil texture (Txt), while others do not change with PFT and Txt. The specific details are shown in Table 1.

[0013] Preferably, the simulation variables of the BEPS model in step S1 include COS flux (the sum of vegetation and soil fluxes, unit: μmol / L). 2 / ), evaporation rate (m / s), photosynthetically active radiation absorption ratio, and total primary productivity (GPP, unit: kg / m²). 2 / ), Latent heat flux (LH, unit: W / m³) 2 ), net primary productivity of ecosystems (kg / 2 / ), Sensible heat flux (W / m 2 SIF(mW / ) 2 / / sr), soil moisture (m 3 / 3 ), transpiration rate (m / s) and vegetation optical thickness.

[0014] Preferably, in step S2, based on the prior data and observed data, the cost function is calculated using a four-dimensional variational algorithm:

[0015]

[0016] Where M and O represent simulated data and observed data, respectively; x and x0 represent the control parameter vector and prior control parameter vector, respectively; C represents the error covariance matrix between the observed data and the model parameters; and J(x) represents the cost function corresponding to the control parameter vector x.

[0017] Preferably, a gradient-based optimization algorithm is used to solve for the control parameter vector that minimizes the cost function. The gradient optimization algorithm is implemented through backward computation using an adjoint model, employing the classical difference algorithm to calculate the gradient, thereby determining the adjustment direction of the control parameters, and iteratively solving for the optimized parameter vector; the formula is as follows:

[0018]

[0019] in, x represents the partial derivative of the cost function with respect to the i-th model parameter in the control parameter vector; J represents the cost function; x i ε represents the i-th model parameter in the control parameter vector; ε represents the difference step size.

[0020] Preferably, the BEPS model in steps S1 and S3 considers a total of 10 Plant Function Types (PFTs) and 11 Soil Textures (represented by Txt). The Plant Function Types are evergreen coniferous forest, deciduous coniferous forest, deciduous broad-leaved forest, evergreen broad-leaved forest, mixed forest, shrubland, grassland, farmland, C4 grassland, and C4 farmland. The Soil Textures are sandy soil, loamy sandy soil, sandy loam, loam, silty loam, sandy clay loam, clay loam, silty clay loam, sandy clay, silty clay, and clay.

[0021] Preferably, the meteorological data in step S1 includes temperature, precipitation, shortwave radiation, relative humidity, and wind speed.

[0022] Preferably, the meteorological data in step S1 is sourced from FLUXNET (https: / / fluxnet.org).

[0023] Preferably, the boundary conditions in step S1 include surface data (including aggregation index, proportion of each PFT, etc.), historical LAI and NPP (Net primary productivity) data, and soil carbon pool data, all provided by the BEPS model.

[0024] Preferably, the COS data observed in step S2 is obtained by measuring the eddy covariance method, which is derived from other research (https: / / doi.org / 10.5281 / zenodo.6940750).

[0025] Preferably, the SIF data observed in step S2 is GOME-2 downscaled SIF data (Joint Research Centre Data Catalogue-Downscaled GOME2 SIF-European Commission, europa.eu) retrieved based on the JJ algorithm.

[0026] Preferably, in step S2, the RMSE (root mean square error) within 24 hours surrounding each COS data point is used as an estimate of the error of that COS data point, and the RMSE within a 3×3 window surrounding each SIF data point is used as an estimate of the error of that SIF data point.

[0027] 3. Beneficial effects

[0028] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0029] (1) In terms of innovation, the method of this invention is the first to propose and successfully realize the joint optimization of water and carbon flux based on COS, SIF observation and data assimilation technology, which provides a reference for the simulation and evaluation of water and carbon flux.

[0030] (2) Structurally, the method of this invention is based on the BEPS ecosystem model, which contains the mechanisms of plant photosynthesis and respiration. Therefore, when performing joint optimization of water and carbon fluxes, this method does not directly optimize the simulated variables, but indirectly optimizes them by adjusting the model parameters. This allows the method of this invention to analyze the optimization results from a deeper and more mechanistic perspective.

[0031] (3) From an algorithmic perspective, the method of this invention adopts a four-dimensional variational algorithm that better reflects complex nonlinear constraint relationships. It can effectively utilize historical observation data at different observation frequencies in a scientific and reasonable manner, and can also consider the evolution information provided by the background mode of state variables.

[0032] (4) From the results, for example Figure 6As shown, the GPP of the FI-Hyy station in July 2015 was optimized using three observation combinations: COS, COS+SIF, and SIF. The results show that the root mean square error between the optimized GPP and the GPP observations is significantly smaller than that between the unoptimized GPP and the GPP observations, which proves the effectiveness and reliability of the method of the present invention. Attached Figure Description

[0033] Figure 1 This is a schematic diagram illustrating the principle of water and carbon flux optimization based on COS and SIF data.

[0034] Figure 2 This refers to the observation data file information in the embodiment;

[0035] Figure 3 To show the changes in model parameters when conducting twin experiments using COS and SIF virtual observations;

[0036] Figure 4 The changes in model parameters during joint assimilation of COS and SIF;

[0037] Figure 5 The graph shows a comparison between the model parameter values ​​obtained by COS, COS+SIF, and SIF assimilation and the prior model parameter values.

[0038] Figure 6 A comparison chart showing the prior simulated GPP, observed GPP, and GPP values ​​obtained by optimization using COS, COS+SIF, and SIF, respectively.

[0039] Figure 7 The graph shows a comparison of the LH values ​​obtained from prior simulation, observation, and optimization using COS, COS+SIF, and SIF, respectively. Detailed Implementation

[0040] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains; the term “and / or” as used herein includes any and all combinations of one or more of the associated listed items.

[0041] Unless otherwise specified in the examples, the procedures should be performed under standard conditions or conditions recommended by the manufacturer. Reagents or instruments whose manufacturers are not specified are all commercially available products.

[0042] As used herein, the term “about” is used to provide for the flexibility and imprecision associated with a given term, measure, or value. Those skilled in the art can readily determine the degree of flexibility for a particular variable.

[0043] The present invention will be further described below with reference to specific embodiments.

[0044] Example

[0045] The technical solution adopted in this invention is as follows:

[0046] A method for optimizing water and carbon flux based on carbonyl sulfide and sunlight-induced chlorophyll fluorescence includes the following steps:

[0047] S1 runs the BEPS model based on prior model parameters and initial conditions to obtain prior simulation variables; the initial conditions include meteorological data, boundary conditions, and leaf area index (LAI) data.

[0048] In step S1, there are a total of 76 prior model parameters. Some of these parameters change with the vegetation function type (PFT) or soil texture (Txt), while others do not change with PFT and Txt. The specific details are shown in Table 1.

[0049] In step S1, the simulation variables of the BEPS model include COS flux (the sum of vegetation and soil fluxes, unit: ρmol / L). 2 / ), evaporation rate (m / s), photosynthetically active radiation absorption ratio, and total primary productivity (GPP, unit: kg / m²). 2 / ), Latent heat flux (LH, unit: W / m³) 2 ), net primary productivity of ecosystems (kg / 2 / ), Sensible heat flux (W / m 2 SIF(mW / ) 2 / / sr), soil moisture (m 3 / 3 ), transpiration rate (m / s) and vegetation optical thickness;

[0050] The BEPS model in steps S1 and S3 considers a total of 10 plant function types (PFTs) and 11 soil textures (represented by Txt). The plant function types are evergreen coniferous forest, deciduous coniferous forest, deciduous broad-leaved forest, evergreen broad-leaved forest, mixed forest, shrubland, grassland, farmland, C4 grassland, and C4 farmland. The soil textures are sandy soil, loamy sandy soil, sandy loam, loam, silty loam, sandy clay loam, clay loam, silty clay loam, sandy clay, silty clay, and clay.

[0051] The meteorological data in step S1 includes temperature, precipitation, shortwave radiation, relative humidity, and wind speed;

[0052] The meteorological data in step S1 comes from FLUXNET (https: / / fluxnet.org);

[0053] The boundary conditions in step S1 include surface data (including aggregation index, proportion of each PFT, etc.), historical LAI and NPP (Net primary productivity) data, and soil carbon pool data, all provided by the BEPS model.

[0054] S2 uses a four-dimensional variation (4D-Var) algorithm to assimilate the observed data with the prior simulation data (simulation data is the set of simulation variables) to obtain the optimized model parameters; the observed data includes observed COS data and observed SIF data, and the prior simulation data includes the prior COS data and prior SIF data in the prior simulation variables obtained in step S1.

[0055] In step S2, based on the prior data and observed data, the cost function is calculated using a four-dimensional variational algorithm:

[0056]

[0057] Where M and O represent simulated data and observed data, respectively; x and x0 represent the control parameter vector and prior control parameter vector, respectively; C represents the error covariance matrix between the observed data and the model parameters; and J(x) represents the cost function corresponding to the control parameter vector x.

[0058] Gradient-based optimization algorithms are used to find the control parameter vector that minimizes the cost function. The gradient optimization algorithm is implemented through backward computation using an adjoint model, employing the classical difference algorithm to calculate the gradient, thereby determining the adjustment direction of the control parameters, and iteratively solving for the optimized parameter vector; the formula is as follows:

[0059]

[0060] in, x represents the partial derivative of the cost function with respect to the i-th model parameter in the control parameter vector; J represents the cost function; x i This represents the i-th model parameter in the control parameter vector; ε represents the difference step size, which is set to 0.01 here.

[0061] The COS data observed in step S2 were obtained by measuring the eddy covariance method, which is derived from other research (https: / / doi.org / 10.5281 / zenodo.6940750).

[0062] In step S2, the observed SIF data uses GOME-2 downscaled SIF data retrieved based on the JJ algorithm (Joint Research Centre Data Catalogue-Downscaled GOME2 SIF-European Commission, europa.eu).

[0063] In step S2, the RMSE (root mean square error) within 24 hours surrounding each COS data point is used as the estimated error of that COS data point. Similarly, the RMSE within a 3×3 window surrounding each SIF data point is used as the estimated error of that SIF data point.

[0064] S3 uses the optimized model parameters to re-drive the BEPS model under the initial conditions, obtaining optimized simulation data including variables related to water and carbon fluxes.

[0065] The specific steps of the method of the present invention are as follows:

[0066] (1) Preparation of materials required for the method, including preparation of observation files and preparation and setup of the model (such as setting the start date of simulation data in the model). Preparation of observation files mainly includes COS and SIF observations and determination of corresponding observation errors; preparation and setup of the model mainly includes preparation of model driving data (such as leaf area index (LAI), boundary conditions (such as soil carbon pool) and meteorological data (including temperature, precipitation, shortwave radiation, relative humidity, wind speed)), setting the start time of meteorological data in the model code, and both observation files and driving data need to be generated in a certain format.

[0067] (2) Based on the prior model parameters (Table 1) and initial conditions, run the model to obtain prior simulation data. The prior simulation data will affect the calculation of the cost function, and the final optimization effect also needs to be evaluated by comparing with the prior simulation data.

[0068] Table 1. Parameters of the BEPS model optimized by this method and the uncertainty of the model parameters.

[0069]

[0070]

[0071] (3) Based on the prior model parameters and initial conditions, run the model to obtain virtual observation data. The virtual observation data is used to conduct a twin experiment to verify the data assimilation capability of this method. This step prepares the data for verifying the assimilation capability and parameter optimization results of the method.

[0072] (4) Conduct twin experiments to verify the assimilation capability and parameter optimization results of this method. Virtual observation data is generated under the conditions of prior model parameters. Therefore, if a perturbation is applied to the prior model parameters and the virtual observation data is assimilated, the model parameters should be adjusted from the perturbated state to the original state, and the cost function and its gradient at the control parameter vector should ultimately reach a value close to zero. Therefore, this method can be verified by observing whether the prior model parameters can recover from the perturbated state and the changes in the cost function and its gradient at the control parameter vector before and after assimilation using virtual observation data with a certain perturbation.

[0073] (5) Calculate the Jacobian matrix of the model, that is, calculate the derivative information of the model at the given control parameter vector. This step is related to the best linear approximation of the optimized control parameter vector.

[0074] (6) Conduct a real assimilation experiment. This step truly combines the actual observation data with the simulated data, performs data assimilation, adjusts the control parameter vector, and finally obtains the optimized model parameters.

[0075] (7) Using the optimized model parameters obtained in the previous step, generate a parameter file, and then combine it with the driving data to re-drive the BEPS model to obtain optimized posterior simulation data. This is the final step, ultimately yielding optimized information such as water and carbon flux. Afterward, this data can be compared with prior simulation data, observational data, etc., for analysis and discussion.

[0076] The following uses the FI-Hyy station (61.845°N, 24.288°E) as an example to illustrate each step of the method. Unless otherwise specified, the time period for each experiment is July 2015.

[0077] For the relevant code and introduction to the BEPS model, including data preparation, please refer to https: / / github.com / JChen-UToronto / BEPS_hourly_site. Our meteorological data comes from FLUXNET (https: / / fluxnet.org), LAI data uses GLOBMAP LAI product (Version 3) (GLOBMAP global Leaf Area Index since 1981 | Zenodo), SIF data uses GOME-2 downscaled SIF data retrieved based on the JJ algorithm (Joint Research Centre Data Catalogue-Downscaled GOME2 SIF-EuropeanCommission (europa.eu)), and COS data comes from other research (https: / / doi.org / 10.5281 / zenodo.6940750). Here, the RMSE (root mean square error) within 24 hours surrounding each COS data point is used as the estimate of the error for that COS data point, and the RMSE within a 3*3 window surrounding each SIF data point is used as the estimate of the error for that SIF data point. The final output is a NetCDF (Network Common DataForm) file containing observed COS and SIF data and their errors. The file includes variable and dimensional information as follows: Figure 2 As shown.

[0078] The observation data comprises three dimensions: nlp, time, and ntc, representing the number of sites, the length of the time series (in hours), and the time series storage dimension (the example file stores time in four dimensions: year, month, day, and second, hence ntc is 4). In addition, the observation file contains nine variables: yyyymmddss, lon, lat, sif, sif_unc, sm, sm_unc, COS_flux, and COS_flux_unc. The dimensional information and meanings of these variables differ, and will be detailed below.

[0079] The yyyymmddss variable uses four dimensions—year, month, day, and second—to store the corresponding time of the station's observation data. Here, we take the data from the FI-Hyy station from 0:00 on January 1, 2015 to 23:00 on December 31, 2015 as an example: the above time range contains a total of 8760 hours, so the size of the yyyyddmmss variable is (8760, 4). The data in the first row is (2015, 1, 1, 0), the data in the second row is (2015, 1, 1, 3600), and the data in the last row is (2015, 12, 31, 82800). Similarly, the storage method for other times can be understood.

[0080] The `lon` variable is used to store the latitude information of the station. A positive value indicates that the station is located in the Northern Hemisphere.

[0081] The lat variable is used to store the longitude information of the station. It is defined as the longitude of the Prime Meridian as 0 degrees. A positive value indicates that the longitude of the station is east of the Prime Meridian.

[0082] SIF represents the daylight-induced chlorophyll fluorescence data for this site. This example site uses resampled GOME-2 SIF data with a nadir spatial resolution of approximately 0.05° and a temporal resolution of 8 days. The unit is mW / 2 / / sr. Because the minimum time interval for the time dimension in this file is 1 hour, the GOMESIF data, which originally had an eight-day time resolution for each year, has been processed here. Starting from the first day of each year, the data is placed sequentially at the time of 12:00 noon on the fourth day of each eight-day period. For example, in the 2015 SIF data from the FI-Hyy site, the first SIF data point is placed at 12:00 noon on January 4, 2015 (i.e., 43200 seconds).

[0083] sm represents the soil moisture data for this site. The sm data for this example site has a time resolution of 1 day and is processed in a similar way to sif data, placing it at 12 noon each day.

[0084] COS_flux represents the carbonyl sulfide flux data at this site, with a spatial resolution of 1 hour and units of ρmol / L. 2 Since the model considers both plant uptake of COS and the influence of soil on COS flux, the carbonyl sulfur flux data used should be the net ecosystem flux.

[0085] sif_unc, sm_unc, and COS_flux_unc represent the observation errors of the sunlight-induced chlorophyll fluorescence data, soil moisture data, and carbonyl sulfur flux data, respectively. The number of data points, their locations, spatial resolutions, and units are consistent with the corresponding observation data.

[0086] In this experiment, soil moisture data was not assimilated; therefore, both the `sm` and `sm_unc` variables contain null values. Besides the driving data from the BEPS model and the assimilated observational data, it is also necessary to validate the results of this method in practical applications. The validation data, like the meteorological data, comes from FLUXNET (https: / / fluxnet.org). Here, validation data, including GPP and LH, were used to validate the water and carbon flux optimization results.

[0087] Figure 3 To validate the model's assimilation ability, solid lines represent parameter changes on the left axis, and dashed lines represent changes on the right axis. The suffixes "PFT" and "Txt (Soil Texture)" indicate whether the parameter is related to PFT or soil texture. "PFT01" represents the vegetation type of this site as evergreen coniferous forest, and "Txt02" represents the soil texture as loam. Vcmax represents the maximum carboxylation rate of rubisco enzyme at 25°C; VJ_slope represents the slope between Vcmax and Jmax (maximum photosynthetic electron transport rate); and SIF alpha and SIF beta parameters are determined by the relationship between the additional heat dissipation corresponding to maximum dark-adapted fluorescence and the relative reduction in photochemical reaction. Ksat_scalar represents soil saturated hydraulic conductivity, and b_scalar is the Campbell parameter used to simulate soil moisture characteristics; both are related to soil texture. f_leaf represents the leaf respiration ratio, which is independent of vegetation type and soil texture. The prior values ​​of the model parameters were standardized to 4. A 20% perturbation was applied to the prior model parameter values ​​as initial values, and the errors of the COS and SIF data in the virtual observation data were both set to 1. It can be seen that after assimilation, the parameters eventually reach a stable state at 4, indicating that the model can effectively assimilate the data and correctly adjust the model parameters.

[0088] Figure 4 This is used to represent the changes in model parameters when using COS and SIF observation data for real data assimilation. It can be seen that after hundreds of iterations, the model parameters eventually reach a stable state, and the final stable values ​​differ from the prior model parameter values, indicating that the model can effectively adjust its parameters by incorporating COS and SIF data.

[0089] Figure 5 This indicates the change between the prior model parameter values ​​and the model parameter values ​​obtained by parameter optimization using COS, COS+SIF, and SIF observation combinations, respectively.

[0090] Figure 6 The paper plots the changes in prior GPP simulation values, GPP simulation values ​​obtained after optimization using COS, COS+SIF, and SIF, and GPP observation values, and calculates the RMSE between the four types of GPP simulation values ​​and GPP observation values.

[0091] It can be seen that all four simulation scenarios can capture the diurnal and intraday variations of GPP and simulate them effectively. However, the prior simulated GPP values ​​are significantly overestimated compared to the observed values. After optimization using COS, COS+SIF, and SIF respectively, the overestimation of the simulated values ​​is significantly corrected, and the three optimized GPP simulation results all have smaller RMSEs and achieve better simulation effects. There are also differences among the three optimized results. Overall, the GPP simulated value obtained by SIF optimization is greater than that obtained by COS optimization, which is greater than that obtained by SIF+COS optimization. It should be noted that optimization using COS+SIF may lead to parameter overtuning, resulting in an underestimate of the simulated GPP and a larger RMSE compared to other optimization results.

[0092] Latent heat flux refers to the heat exchange between the atmosphere and the underlying surface moisture, mainly in the form of phase change of water. Figure 7 The paper plots the changes in prior LH simulated values, LH simulated values ​​obtained after optimization using COS, COS+SIF, and SIF, and LH observed values, and calculates the RMSE between the four types of LH simulated values ​​and LH observed values.

[0093] It can be seen that all four simulation scenarios can effectively capture the diurnal and intraday variations of LH. Similar to the GPP simulation, the LH simulation generally shows a pattern where the prior simulation is better than the SIF-optimized simulation, which is better than the COS-optimized simulation, which is better than the COS+SIF simulation. Unlike the GPP simulation, the results of the four LH simulations are not significantly different, with no obvious overestimation or underestimation. However, compared with the prior simulation results, the three optimized simulations still have smaller RMSE and achieve better simulation results.

[0094] The above description provides an illustrative overview of the present invention and its embodiments. This description is not restrictive, and the embodiments shown are merely one example of the invention's implementation. Actual implementations are not limited to these examples. Therefore, if those skilled in the art are inspired by this description and design similar implementations and examples without departing from the spirit of the invention, such designs should fall within the scope of protection of the present invention.

Claims

1. A method for optimizing water and carbon flux based on carbonyl sulfide and sunlight-induced chlorophyll fluorescence, characterized in that, Includes the following steps: S1 runs the BEPS model based on prior model parameters and initial conditions to obtain prior simulation variables; the initial conditions include meteorological data, boundary conditions, and leaf area index data. S2 uses a four-dimensional variational algorithm to assimilate the observed data with the prior simulation data to obtain the optimized model parameters; the observed data includes observed COS data and observed SIF data, and the prior simulation data includes the prior COS data and prior SIF data in the prior simulation variables obtained in step S1. S3 uses the optimized model parameters to re-drive the BEPS model under the initial conditions, obtaining optimized simulation data including variables related to water and carbon fluxes.

2. The method for optimizing water and carbon flux based on carbonyl sulfide and sunlight-induced chlorophyll fluorescence as described in claim 1, characterized in that, In step S2, based on the prior data and observed data, the cost function is calculated using a four-dimensional variational algorithm: Where M and O represent simulated data and observed data, respectively. and These represent the control parameter vector and the prior control parameter vector, respectively, and C represents the error covariance matrix between the observed data and the model parameters; The control parameter vector is The corresponding cost function.

3. The method for optimizing water and carbon flux based on carbonyl sulfide and sunlight-induced chlorophyll fluorescence as described in claim 2, characterized in that, Gradient-based optimization algorithms are used to find the control parameter vector that minimizes the cost function. The gradient optimization algorithm is implemented through backward computation using an adjoint model, employing the classical difference algorithm to calculate the gradient, thereby determining the adjustment direction of the control parameters, and iteratively solving for the optimized parameter vector; the formula is as follows: in, It represents the partial derivative of the cost function with respect to the i-th model parameter in the control parameter vector; Represents the cost function; This represents the i-th model parameter in the control parameter vector; This represents the difference step size.

4. The method for optimizing water and carbon flux based on carbonyl sulfide and sunlight-induced chlorophyll fluorescence as described in any one of claims 1 to 3, characterized in that, The simulated variables of the BEPS model in step S1 include: COS flux, where COS flux refers to the sum of vegetation and soil fluxes, unit: ; Evaporation rate, unit: m / s ; Photosynthetically active radiation absorption ratio; Total primary productivity of ecosystems, unit: ; Latent heat flux, unit: ; Ecosystem net primary productivity ; Sensible heat flux ; SIF, unit: ; Soil moisture, unit: ; transpiration rate, unit: m / s ;and Optical thickness of vegetation.

5. The method for optimizing water and carbon flux based on carbonyl sulfide and sunlight-induced chlorophyll fluorescence as described in any one of claims 1 to 3, characterized in that, The BEPS model in steps S1 and S3 considers a total of 10 vegetation functional types and 11 soil textures; wherein the vegetation functional types are evergreen coniferous forest, deciduous coniferous forest, deciduous broad-leaved forest, evergreen broad-leaved forest, mixed forest, shrub, grassland, farmland, C4 grassland and C4 farmland; and the soil textures are sandy soil, loamy sandy soil, sandy loam, loam, silty loam, sandy clay loam, clay loam, silty clay loam, sandy clay, silty clay and clay.

6. The method for optimizing water and carbon flux based on carbonyl sulfide and sunlight-induced chlorophyll fluorescence as described in any one of claims 1 to 3, characterized in that, The meteorological data in step S1 includes temperature, precipitation, shortwave radiation, relative humidity, and wind speed.

7. The method for optimizing water and carbon flux based on carbonyl sulfide and sunlight-induced chlorophyll fluorescence as described in any one of claims 1 to 3, characterized in that, The meteorological data in step S1 is sourced from FLUXNET.

8. The method for optimizing water and carbon flux based on carbonyl sulfide and sunlight-induced chlorophyll fluorescence as described in any one of claims 1 to 3, characterized in that, The boundary conditions in step S1 include surface data, historical LAI and NPP data, and soil carbon pool data.

9. The method for optimizing water and carbon flux based on carbonyl sulfide and sunlight-induced chlorophyll fluorescence as described in any one of claims 1 to 3, characterized in that, The COS data observed in step S2 is obtained by measuring the eddy covariance method; the SIF data observed in step S2 is the GOME-2 downscaled SIF data inverted based on the JJ algorithm.

Images