Method and system for predicting methane emission flux based on data coupling model

Abstract

This invention discloses a method and system for predicting methane emission flux based on a data coupling model. The method includes: downloading remote sensing, soil, hydrological, observational, and meteorological data from a target wetland area and performing data preprocessing; coupling the TECO model with the Wetland-DNDC model, and transmitting hourly groundwater level data output by the TECO model to the DNDC model; processing the wetland soil data to calculate changes in soil profile moisture content; running the coupled model to simulate methane generation, oxidation, and emission; optimizing key parameters using Bayesian data assimilation based on measured data, quantifying uncertainty through Monte Carlo simulation, and predicting methane emission flux data; and finally, verifying the accuracy of the predicted results against the measured results. Through the coupled model, the present invention achieves high-precision prediction of wetland methane emission flux.

Description

Technical Field

[0001] The present invention relates to the technical field of prediction of methane emission flux in wetland ecosystems, and in particular to a method and system for predicting methane emission flux based on a data coupling model. Background Art

[0002] Wetlands are one of the world's largest natural sources of methane emissions. As a potent greenhouse gas, accurate prediction of methane emission flux is of great significance for understanding the global carbon cycle and climate change.

[0003] Traditional methane emission prediction methods rely primarily on field observations and single-model simulations, but these methods suffer from limitations such as difficulty in data acquisition, insufficient model accuracy, and poor real-time performance. Data acquisition requires extensive equipment and manpower investment in field observations and can only cover limited regions and timescales. Single models, such as the TECO (Terrestrial Ecosystem Carbon Model) or the Wetland-DNDC (Wetland-DeNitrification-DeComposition model, an extension and modification of the well-known agricultural ecosystem model DNDC (DeNitrification-DeComposition). DNDC was originally designed for dryland farmland, and Wetland-DNDC has adapted it by introducing processes unique to wetlands), often fail to fully account for the interactions of multiple factors, such as groundwater levels, soil moisture content, and vegetation cover, when simulating the complex ecological and hydrological processes of wetlands. Furthermore, traditional methods struggle to achieve large-scale, high-spatiotemporal resolution methane emission flux predictions.

[0004] In recent years, with the development of remote sensing technology and data assimilation methods, existing methods still have the following problems: the fusion and processing procedures of multi-source data are relatively cumbersome, and the model parameters are not optimized enough, resulting in low reliability of prediction results.

[0005] To address the above problems, the present invention proposes a wetland methane emission flux prediction method and system based on a data coupling model. By coupling the TECO and Wetland-DNDC models and combining Bayesian data assimilation and Monte Carlo simulation, high-precision, high temporal and spatial resolution methane emission flux prediction is achieved. Summary of the Invention

[0006] In order to achieve large-scale, high-temporal and high-spatial resolution methane emission flux prediction and improve the reliability of predicted methane emission flux, the present invention provides a method and system for predicting methane emission flux based on a data coupling model, which can solve the existing problems raised in the background technology.

[0007] In order to achieve the above object, the present invention provides the following technical solutions:

[0008] A method for predicting methane emission flux based on a data coupling model includes the following specific steps:

[0009] Step 1: Download remote sensing data and soil, hydrological, observational data, and daily and hourly meteorological data of the target wetland area, and preprocess the data;

[0010] Step 2: Process the acquired wetland soil data and calculate the water content change of the first 30 cm soil profile using a water balance model with an hourly time step;

[0011] Step 3: Couple the TECO model and the Wetland-DNDC model, passing the groundwater level data output by the TECO model to the DNDC model; run the coupled TECO-DNDC model to simulate methane generation, oxidation, and emissions (diffusion, plant transport, and boiling);

[0012] Step 4: Based on the measured methane flux emission data from the static chamber-gas chromatograph method, Bayesian data assimilation is used to optimize key parameters. The uncertainty of the parameters and model structure is quantified through Monte Carlo simulation to generate methane flux prediction intervals.

[0013] Step 5: Use the improved wetland coupling model to predict the methane emission flux in the target area and verify its accuracy with the measured results.

[0014] As a method for predicting methane emission flux based on a data coupling model of the present invention, the high spatial resolution remote sensing image of Sentinel-2 with a spatial resolution of 10m downloaded in step 1 uses the latest soil hydrology and observation data, and obtains updated meteorological data in real time.

[0015] In this method for predicting methane emission fluxes based on a data coupling model, the wetland soil data is preprocessed in step 2 using a constructed cubic spline interpolation function to generate hourly data points, thereby obtaining the hourly input data required by the DNDC model. This reduces the amount of data and the data processing process.

[0016] As a method for predicting methane emission flux based on a data coupling model of the present invention, the groundwater depth equation of the water balance model in step 2 is:

[0017]

[0018] As a method for predicting methane emission flux based on a data coupling model of the present invention, the method for evaluating the inversion accuracy in step 4 is: Bayesian data assimilation to optimize key parameters; and Monte Carlo simulation to quantify the uncertainty of parameters and model structure.

[0019] Using Bayesian probabilistic inversion techniques, we estimated the posterior distribution of the model parameters based on prior knowledge of the parameter ranges and field measurements of CH4 emissions. We then randomly selected parameter sets from the posterior distribution of the parameter sets for 100 predictions. We also randomly selected a set of randomly generated environmental variables and used the same set of variables for all predictions.

[0020] Bayes’ theorem provides an equation where the posterior probability density function p(θ|Z) of the model parameters given an observation Z is based on prior knowledge of the parameter distribution p(θ) and the likelihood function p(Z|θ):

[0021] p(θ|Z)∝p(Z|θ)p(θ)

[0022] This paper assumes that the prior knowledge of the parameter distribution p(θ) is uniform. Due to the equivalence and unidentifiable parameters when constraining multiple parameters using only one observation data stream, this paper selects only three highly sensitive parameters for data assimilation, and the a priori range uses the default values ​​for wetlands of the same latitude and type. The error between each observation and the model simulation result follows an independent normal distribution with a mean of zero, so the likelihood function is expressed as:

[0023]

[0024] The posterior probability distribution of parameter sampling was obtained using an adaptive Metropolis-Hastings (MH) algorithm, employing Markov Chain Monte Carlo techniques. The parameter values ​​of the set were randomly accepted with a probability of 0.05. After running the simulation chain, the convergence of the sampling chain was checked using the Gelman-Rubin statistic. Only the second half of the accepted parameter values ​​were used in the posterior analysis.

[0025] Parameter quantization method: select r_me, Q 10_pro , T opt_pro These three parameters have high sensitivity indexes and strong interactions in the formula for calculating methane production. After one of the parameters is input into data assimilation, the distribution of the other two parameters in the parameter histogram is well constrained and the input parameters are within the confidence interval, which can reduce the uncertainty of simulation predictions.

[0026] As a method for predicting methane emission flux based on a data coupling model provided by the present invention, the method for evaluating the inversion accuracy in step 5 is:

[0027] The root mean square error (RMSE) and the coefficient of determination R 2 As an accuracy evaluation indicator, the formula is as follows:

[0028] Root Mean Square Error (RMSE):

[0029]

[0030] Where n is the total number of samples, is the model prediction value, y={y1,y2,K,y n} is the true value; the range of the root mean square error and the mean absolute percentage error are both [0,+∞), and they are equal to 0 when the predicted value is completely consistent with the true value, that is, a perfect prediction.

[0031] Coefficient of determination (R) 2 The calculation formula is as follows:

[0032]

[0033] in:

[0034] is the residual sum of squares, which represents the sum of the differences between the model's predicted values ​​and the actual observed values.

[0035] is the total sum of squares, which represents the sum of the differences between the actual observations and the observed mean.

[0036] R 2 Indicates the proportion of data variance that the model can explain. 2 When it is close to 1, it means that the model can fit the data well; when R 2 When it is close to 0, it means that the model does not fit the data well.

[0037] The present invention also discloses a system for predicting methane emission flux based on a data coupling model, comprising:

[0038] Data acquisition module: used to download remote sensing, meteorological, soil and hydrological data;

[0039] Preprocessing module: performs radiation correction, cubic spline interpolation and spatial and temporal resolution unification;

[0040] Model coupling engine: Dynamically connects the TECO model and the Wetland-DNDC model to achieve hourly transmission of groundwater level data;

[0041] Parameter Optimization Module: Optimize key parameters and quantify uncertainties based on Bayesian data assimilation and Monte Carlo simulation;

[0042] Verification and output module: Calculate RMSE and R 2 , generate methane flux prediction report and visualization results.

[0043] The model coupling engine automatically implements the data interface through Python scripts, specifically including:

[0044] Read the hourly groundwater level output file of the TECO model, convert it into the input format required by the DNDC model, and trigger the running of the coupled model.

[0045] Beneficial effects: Compared with the prior art, the present invention has the following advantages:

[0046] The present invention provides a method and system for predicting methane emission flux based on a data coupling model. By coupling the TECO and Wetland-DNDC models, the model directly inputs the groundwater level depth in the TECO model into the DNDC model, solving the problem of insufficient spatiotemporal continuity of the groundwater level depth. By combining Bayesian data assimilation and Monte Carlo simulation, key parameters are optimized and uncertainties are quantified, achieving high-precision prediction of wetland methane emission flux, providing a scientific basis for wetland management and climate change response. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for describing the embodiments or the prior art.

[0048] Figure 1 This is a flow chart of the method for predicting methane emission flux based on the data coupling model of the present invention. DETAILED DESCRIPTION

[0049] To make the objectives, technical solutions and advantages of the present invention more clear, the embodiments of the present invention will be described in further detail below with reference to the accompanying drawings.

[0050] The present invention provides a method for predicting methane emission flux based on a data coupling model. For detailed process, please refer to Figure 1 , including the following specific steps:

[0051] Step 1: Download remote sensing data and soil, hydrological, observational data, and daily and hourly meteorological data of the target wetland area, and preprocess the data;

[0052] Download high-spatial-resolution Sentinel-2 images to obtain soil, hydrological, and observational data, as well as daily and hourly meteorological data for the site. Preprocess the remote sensing data, including radiometric, atmospheric, and geometric corrections, to obtain accurate multispectral canopy reflectance data.

[0053] Radiation correction, geometric correction and atmospheric correction preprocessing are common remote sensing image processing methods. This embodiment uses ENVI5.6 software to complete the above processing;

[0054] Step 2: Process the acquired wetland soil data and calculate the water content change of the first 30 cm soil profile using a water balance model with an hourly time step;

[0055] The preprocessing is to convert the TECO daily output into DNDC hourly data through cubic spline interpolation. The operation is as follows:

[0056] Time series construction: The daily output data of the TECO model is constructed as a time series, with each data point representing the average value of one day.

[0057] Cubic spline interpolation: This method applies cubic spline interpolation to expand daily data points into hourly data. Cubic spline interpolation approximates the functional relationship between data points by constructing a piecewise cubic polynomial. It ensures continuity and smoothness of the interpolation function at each data point.

[0058] Boundary condition processing: When performing cubic spline interpolation, boundary conditions need to be processed. Usually, natural boundary conditions are used, that is, assuming that the second-order derivative of the function is zero at both ends of the data point.

[0059] Interpolation function construction: For each adjacent daily data point, construct a cubic polynomial so that the value of the polynomial at the data point is consistent with the original data and satisfies the boundary conditions.

[0060] Hourly data generation: The constructed cubic spline interpolation function is used to generate hourly data points, thereby obtaining the hourly input data required by the DNDC model.

[0061] Step 3: Couple the TECO model and the Wetland-DNDC model, passing the groundwater level data output by the TECO model to the DNDC model; run the coupled TECO-DNDC model to simulate methane generation, oxidation, and emissions (diffusion, plant transport, and boiling);

[0062] The groundwater level module calculates the water content change in the first 30 cm of the soil profile using a water balance model characterized by water input and output at hourly time steps:

[0063] In the unsaturated zone, using a quadratic function, the volumetric water content of soil (θ us ) from the vegetation surface volume water content (θ s ) to the groundwater level (z wt ) is incremented as follows:

[0064]

[0065] In the formula The constant value is 0.95, z is the soil depth (mm), θ s Expressed as

[0066]

[0067] where θ smin is the minimum volumetric water content held by peat moss on the soil surface and is set to 0.25, a z is the gradient of a linear decrease;

[0068]

[0069] where z θsmin is the maximum inhalation interval for a given value of 100 mm; therefore, above z b The total amount of water in the soil profile should be:

[0070]

[0071] The first part of the formula is the water content z in the saturated zone mentioned above b , the second part is the water content in the unsaturated zone z wt ; If the entire section is saturated, the still water height is V tot and The difference between

[0072] The groundwater depth equation of the water balance model is:

[0073]

[0074] Table 1 is a graph of the main parameters of methane production, oxidation, diffusion, boiling, and plant-mediated transport used in the present invention;

[0075]

[0076]

[0077] Step 4: Based on the measured methane flux emission data from the static chamber-gas chromatograph method, Bayesian data assimilation is used to optimize key parameters. The uncertainty of the parameters and model structure is quantified through Monte Carlo simulation to generate methane flux prediction intervals.

[0078] Methane emission data are the only observational dataset available for data assimilation. The most sensitive parameters are selected for data assimilation because in data assimilation, an initial sensitivity test is performed when a parameter can be constrained by that variable; the sensitivity of a parameter is determined by the sensitivity index defined as:

[0079]

[0080] Among them, y0 is the model output of the initial value of the independent variable x0, the model output is methane emissions, the independent variable value is ±Δx, the corresponding dependent variable values ​​are y2 and y1, and Δx is set to 0.25 times the initial value;

[0081] The static chamber-gas chromatograph method was used to calculate the methane flux emission data. The target area for predicting methane flux emissions was selected. The ground gas emission flux formula is:

[0082]

[0083] Where F is the gas flux, with positive values ​​indicating emission and negative values ​​indicating absorption; P0, V0, and T0 are the standard atmospheric pressure, gas molar volume, and absolute temperature under standard conditions, respectively; dc / dt is the slope of the linear curve showing the change in sampled gas concentration over time; M is the molar mass of the measured gas; P and T are the actual atmospheric pressure and temperature at the sampling point; and H is the height of the sampling box.

[0084] The formula for gas emission flux on the water surface is:

[0085]

[0086] Where Flux is the greenhouse gas diffusion flux;

[0087] Slop: Slope in the time-concentration graph, ppm·s -1 ;

[0088] V: air volume in static box, m 3 ; S: water surface area covered by the box, m 2 ;

[0089] F1: Standard temperature and pressure of gases in air from ppm to mg·m -3 The conversion coefficient is as follows:

[0090]

[0091] M: gas molar mass, g·mole -1 ;

[0092] P: On-site atmospheric pressure during monitoring, kPa;

[0093] T: Temperature inside the chamber during monitoring, °C;

[0094] F2: is the conversion factor between seconds and days, 86400s·d-1.

[0095] Based on the above formula, the emission of methane flux can be calculated.

[0096] Using Bayesian probabilistic inversion techniques, we estimated the posterior distribution of the model parameters based on prior knowledge of the parameter ranges (Table 1) and field measurements of CH4 emissions. We then randomly selected parameter sets from the posterior distribution of the parameter sets for 100 predictions. We also randomly selected a set of randomly generated environmental variables and used the same set for all predictions.

[0097] Bayes’ theorem provides an equation where the posterior probability density function p(θ|Z) of the model parameters given an observation Z is based on prior knowledge of the parameter distribution p(θ) and the likelihood function p(Z|θ):

[0098] p(θ|Z)∝p(Z|θ)p(θ)

[0099] The prior knowledge of the parameter distribution p(θ) is assumed to be uniformly distributed. Due to the equivalence and unidentifiable parameters when using only one observation data stream to constrain multiple parameters, three highly sensitive parameters are selected for data assimilation. The prior range uses the default values ​​for the same type of wetland at the same latitude. The error between each observation data and the model simulation result is independently normally distributed with a mean of zero, so the likelihood function is expressed as

[0100]

[0101] Where Z i (t) is the only observation flow at time t, X(t) is the corresponding variable of the simulation, The standard deviation of the observation set is used to generate the posterior probability distribution of parameter sampling using the adaptive Metropolis-Hastings (MH) algorithm, employing Markov Chain Monte Carlo techniques. The parameter values ​​of the set are randomly accepted with a probability of 0.05. After running the simulation chain, the Gelman-Rubin statistic is used to check the convergence of the sampling chain. Only the second half of the accepted parameter values ​​are used in the posterior analysis.

[0102] Parameter quantization method: select r_me, Q 10_pro , T opt_pro These three parameters have high sensitivity indexes and strong interactions in the methane production calculation formula. r_me is the potential proportion of anaerobic mineralized carbon released into methane, Q 10_pro is the temperature coefficient of methane production, T opt_pro The optimal temperature for methane production is one of the parameters in the input data assimilation, and the distribution of the other two parameters in the parameter histogram is well constrained and the input parameters are within the confidence interval, which can reduce the uncertainty of the simulation prediction.

[0103] The parameter values ​​used in this process are shown in Table 1.

[0104] Step 5: Use the improved wetland coupling model to predict the methane emission flux in the target area and verify its accuracy with the measured results.

[0105] Among them, the method for evaluating the inversion accuracy in step 5 is:

[0106] The root mean square error (RMSE) and the coefficient of determination R 2 As an accuracy evaluation indicator, the formula is as follows:

[0107] Root Mean Square Error (RMSE):

[0108]

[0109] Where n is the total number of samples, is the model prediction value, y={y1,y2,K,y n} is the true value; the range of the root mean square error and the mean absolute percentage error are both [0,+∞), and they are equal to 0 when the predicted value is completely consistent with the true value, that is, a perfect prediction.

[0110] Coefficient of determination (R) 2 The calculation formula is as follows:

[0111]

[0112] in:

[0113] is the residual sum of squares, which represents the sum of the differences between the model's predicted values ​​and the actual observed values.

[0114] is the total sum of squares, which represents the sum of the differences between the actual observations and the observed mean.

[0115] R 2 Indicates the proportion of data variance that the model can explain. 2 When it is close to 1, it means that the model can fit the data well; when R 2 When it is close to 0, it means that the model does not fit the data well.

[0116] In this regard, the present invention also discloses a system for the above method, comprising:

[0117] Data acquisition module: used to download remote sensing, meteorological, soil and hydrological data;

[0118] Preprocessing module: performs radiation correction, cubic spline interpolation and spatial and temporal resolution unification;

[0119] Model coupling engine: Dynamically connects the TECO model and the Wetland-DNDC model to achieve hourly transmission of groundwater level data;

[0120] Parameter Optimization Module: Optimize key parameters and quantify uncertainties based on Bayesian data assimilation and Monte Carlo simulation;

[0121] Verification and output module: Calculate RMSE and R 2 , generate methane flux prediction report and visualization results.

[0122] The model coupling engine automatically implements the data interface through Python scripts, specifically reading the hourly groundwater level output file of the TECO model, converting it into the input format required by the DNDC model, and triggering the operation of the coupled model.

[0123] In summary, by coupling the TECO and Wetland-DNDC models, the present invention directly inputs the groundwater table depth in the TECO model into the DNDC model, thus solving the problem of insufficient spatiotemporal continuity of the groundwater table depth. Combining Bayesian data assimilation with Monte Carlo simulation, key parameters are optimized and uncertainties are quantified, achieving high-precision prediction of wetland methane emission fluxes, providing a scientific basis for wetland management and climate change response.

Claims

1. A method for predicting methane emission flux based on a data coupling model, characterized in that: The steps include: Step 1: Obtain remote sensing data, soil data, hydrological data, observation data, and daily and hourly meteorological data of the target wetland area, and perform data preprocessing; Step 2: Process the wetland soil data. Use the cubic spline interpolation method to convert the daily output data of the TECO model into hourly data. Then calculate the water content change of the first 30 cm soil profile based on the water balance model. Step 3: Couple the TECO model with the Wetland-DNDC model, inputting hourly groundwater level data output by the TECO model into the DNDC model; run the coupled TECO-DNDC model to simulate methane generation, oxidation, and emission processes; Step 4: Based on the measured methane flux emission data from the static chamber-gas chromatograph method, Bayesian data assimilation is used to optimize key parameters. The uncertainties of the parameters and model structure are quantified through Monte Carlo simulation to generate methane flux prediction intervals. Step 5: Use the improved wetland coupling model to predict the methane emission flux in the target area and verify the accuracy with the measured results; use the root mean square error (RMSE) and determination coefficient (R) 2 As an accuracy evaluation indicator.

2. The method for predicting methane emission flux based on a data coupling model according to claim 1, characterized in that: The data preprocessing in step 1 includes processing the remote sensing data through radiation correction, geometric correction and atmospheric correction to invert vegetation coverage and soil moisture content; the temporal and spatial resolutions of meteorological data, hydrological data and soil data are unified to the hourly scale.

3. The method for predicting methane emission flux based on a data coupling model according to claim 1, characterized in that: The wetland soil data processing in step 2 includes converting the daily output of the TECO model into hourly data through cubic spline interpolation, and the operation is as follows: Time series construction: The daily output data of the TECO model is constructed as a time series, with each data point representing the average value of one day; Cubic spline interpolation: Apply the cubic spline interpolation method to expand daily data points into hourly data; Boundary condition processing: When performing cubic spline interpolation, it is necessary to process boundary conditions and adopt natural boundary conditions, that is, assuming that the second-order derivative of the function is zero at both ends of the data point; Interpolation function construction: For each adjacent daily data point, construct a cubic polynomial so that the value of the polynomial at the data point is consistent with the original data and satisfies the boundary conditions; Hourly data generation: The constructed cubic spline interpolation function is used to generate hourly data points, thereby obtaining the hourly input data required by the DNDC model.

4. The method for predicting methane emission flux based on a data coupling model according to claim 1, characterized in that: Calculate the change in moisture content in the first 30 cm of soil profile: In the unsaturated zone, using a quadratic function, the soil volumetric water content θ us From the vegetation surface volume water content θ s To the groundwater level (z wt ) is incremented as follows: in The constant value is 0.95, z is the soil depth in mm, θ s Expressed as where θ smin is the minimum volumetric water content held by peat moss on the soil surface and is set to 0.25, a z is the gradient of a linear decrease; where z θsmin is the maximum inhalation interval for a given value of 100 mm; therefore, above z b The total amount of water in the soil profile should be: The first part is the water content z in the saturated zone mentioned above. b , the second part is the water content in the unsaturated zone z wt ; If the entire section is saturated, the still water height is V tot and The final groundwater depth equation is 5. The method for predicting methane emission flux based on a data coupling model according to claim 1, characterized in that: In the step 4: Data assimilation method: Using Bayesian probabilistic inversion techniques, based on prior knowledge of the parameter range and field measurements of CH4 emissions, the posterior distribution of the model parameters was estimated. 100 predictions were made by randomly selecting a parameter set from the posterior distribution of the parameter set and a set of randomly generated environmental variables. The same set of variables was used for all predictions. According to the equation in Bayes' theorem, the posterior probability density function p(θ|Z) of the model parameters given the observation value Z is based on the prior knowledge of the parameter distribution p(θ): p(θ|Z)∝p(Z|θ)p(θ) Assuming that the prior knowledge of the parameter distribution p(θ) is uniformly distributed, due to the equivalence and unidentifiable parameters when only one observation data stream is used to constrain multiple parameters, three highly sensitive parameters are selected for data assimilation. The prior range uses the default values ​​of wetlands of the same latitude and type. The error between each observation data and the model simulation result independently follows a normal distribution with a mean of zero. Therefore, the likelihood function is expressed as: where Z i (t) is the only observation flow at time t, X(t) is the corresponding variable of the simulation, is the standard deviation of the observation set. The adaptive Metropolis-Hastings algorithm is used to sample the posterior probability distribution of the parameters. The Markov Chain Monte Carlo technique is used. The parameter values ​​of the set are randomly accepted with a probability of 0.

05. After running the simulation chain, the Gelman-Rubin statistic is used to check the convergence of the sampling chain. Only the second half of the accepted parameter values ​​are used for the posterior analysis. Parameter quantification method: select the potential ratio of anaerobic mineralized carbon released into methane r_me, the temperature coefficient of methane production Q 10-pro , the optimal temperature for methane production T opt-pro These three parameters have high sensitivity indices and strong interactions in the formula for calculating methane production. After one of the parameters is input into data assimilation, the distribution of the other two parameters is well constrained in the parameter histogram and the input parameters are within the confidence interval, which is used to reduce the uncertainty of simulation predictions.

6. The method for predicting methane emission flux based on a data coupling model according to claim 1, characterized in that: The method for evaluating the inversion accuracy in step 5 is: The root mean square error RMSE and the coefficient of determination R 2 As the accuracy evaluation index, among which: Root mean square error RMSE: Where n is the total number of samples, is the model prediction value, y={y1,y2,...,y n } is the true value; the range of the root mean square error and the mean absolute percentage error are both [0, +∞), and when the predicted value is completely consistent with the true value, it is equal to 0, that is, a perfect prediction; Coefficient of determination R 2 The calculation formula is: in: is the residual sum of squares, which represents the sum of the differences between the model predictions and the actual observed values; is the total sum of squares, which represents the sum of the differences between the actual observations and the observed mean; R 2 Indicates the proportion of data variance that the model can explain; when R 2 When it is close to 1, it means that the model can fit the data; when R 2 When it is close to 0, it means that the model cannot fit the data.

7. A system for predicting methane emission flux based on a data coupling model, used to implement the method for predicting methane emission flux based on a data coupling model according to any one of claims 1 to 6, characterized in that: include: Data acquisition module: used to download remote sensing, meteorological, soil and hydrological data; Preprocessing module: performs radiation correction, cubic spline interpolation and spatial and temporal resolution unification; Model coupling engine: Dynamically connects the TECO model and the Wetland-DNDC model to achieve hourly transmission of groundwater level data; Parameter Optimization Module: Optimize key parameters and quantify uncertainties based on Bayesian data assimilation and Monte Carlo simulation; Verification and output module: Calculate RMSE and R 2 , generate methane flux prediction report and visualization results.

8. The system for predicting methane emission flux based on a data coupling model according to claim 7, characterized in that: The model coupling engine automatically implements the data interface through Python scripts, specifically including: Read the hourly groundwater level output file of the TECO model, convert it into the input format required by the DNDC model, and trigger the running of the coupled model.

Images