Upscaling method and system for carbon flux in flooded wetlands integrating hydrodynamic model and integrated machine learning

Abstract

The present invention discloses a method and system for upscaling carbon flux in floodplain wetlands by integrating a hydrodynamic model and integrated machine learning. The present invention relates to the technical field of water conservancy and ecological environment monitoring and management. By collecting data on eddy covariance measurement, digital elevation model, hydrometeorology, and remote sensing images of floodplain wetlands, multi-scale segmentation of the underlying surface of floodplain wetlands is carried out to obtain calculation units for upscaling carbon flux; a hydrodynamic model of floodplain wetlands is established to calculate relevant hydrological situation index factors affecting NEE in floodplain wetlands; multiple machine learning models are constructed, and the prediction weights of each model are optimized to obtain an optimal integrated prediction scheme for model integration; NEE spatial upscaling is carried out, and the prediction uncertainty is quantitatively characterized by using the prediction differences of multiple models to obtain the spatio-temporal distribution map and uncertainty map of NEE; the present invention realizes high-precision prediction of wetland NEE upscaling, providing a new technical method and reliable data support for the research on carbon flux in floodplain wetlands and ecological management.

Description

Technical Field

[0001] The present invention relates to the technical field of water conservancy and ecological environment monitoring and management, and particularly relates to a method and system for upscaling carbon fluxes in floodplain wetlands by integrating a hydrodynamic model and integrated machine learning. Background Art

[0002] Floodplain wetlands are one of the most important wetland types, with multiple functions such as flood regulation, water purification, habitat for animals and plants, and carbon sequestration. The frequent water level fluctuations in floodplain wetlands, combined with rich carbon and nitrogen reserves, make them hotspots for global carbon flux changes. Climate change and human activities are accelerating the transformation of floodplain wetland ecosystems, leading to deteriorated water environments, fragmented underlying surfaces, and disrupted natural carbon cycles. Understanding the spatio-temporal trends of carbon fluxes in floodplain wetlands under changing environments is of great significance for accurate carbon budgeting and effective wetland management. However, due to the high heterogeneity of wetland landscapes and hydrology and the limitations of regional carbon flux monitoring, it is challenging to map the net ecosystem exchange (NEE) of floodplain wetlands at the regional or global scale.

[0003] Traditional site-scale eddy covariance observations are difficult to meet the needs of regional or global scale NEE assessments. Currently, pixel-scale remote sensing data and single machine learning models are mostly used for NEE upscaling, which have problems such as difficulty in segmenting wetland patch heterogeneity, neglecting hydrological processes, and limited prediction accuracy of single models. Therefore, there is an urgent need to propose a method and system that comprehensively consider the impacts of wetland heterogeneity and hydrological processes and accurately upscale to solve the above problems. Summary of the Invention

[0004] The purpose of the present invention is to provide a method and system for upscaling carbon fluxes in floodplain wetlands by integrating a hydrodynamic model and integrated machine learning to solve the problems in the prior art, such as difficulty in segmenting wetland patch heterogeneity, neglecting hydrological processes, and limited prediction accuracy of single models.

[0005] In the first aspect, the present invention provides a method for upscaling carbon fluxes in floodplain wetlands by integrating a hydrodynamic model and integrated machine learning, including:

[0006] Step 1: Collect eddy covariance measurement data, digital elevation model data, hydrometeorological data, and remote sensing image data of floodplain wetlands, and conduct multi-scale segmentation of the underlying surface of floodplain wetlands using an object-oriented calculation method to obtain calculation units for carbon flux upscaling;

[0007] Step 2: Based on the hydrometeorological data of floodplain wetlands, establish a hydrodynamic model of floodplain wetlands to obtain water level parameters of floodplain wetlands, and calculate relevant hydrological regime index factors affecting the NEE of floodplain wetlands;

[0008] Step 3: Integrate the eddy covariance measurement data, hydrological regime index factors, and environmental variables based on remote sensing image data, construct multiple machine learning models, and use the Powell algorithm to optimize the prediction weights of each model to obtain the optimal model ensemble prediction scheme;

[0009] Step 4: Based on the optimal model ensemble scheme, carry out the spatial upscaling of NEE, realize the characterization of the spatio-temporal variation of the carbon flux in the floodplain wetland ecosystem, use the multi-model prediction differences to quantitatively characterize the prediction uncertainty, and obtain the spatio-temporal distribution map and uncertainty map of NEE.

[0010] Furthermore, Step 1 includes:

[0011] Obtain NEE data through the eddy covariance tower installed on-site in the floodplain wetland, and use EddyPro software for coordinate rotation correction, trend correction, data synchronization, density correction, ultrasonic virtual temperature correction, spectral correction, angle-of-attack correction, and data quality control processing;

[0012] Obtain the band reflectance data of remote sensing data, digital elevation model data of the floodplain wetland, as well as daily flow, water level, precipitation, temperature, wind speed, vegetation index, and vegetation photosynthetically active radiation data;

[0013] Use the multi-resolution segmentation algorithm in eCognition Developer software to perform object-oriented multi-scale segmentation on the remote sensing image. By testing different combinations of ratio parameters, shape weights, and compactness weights, determine the optimal segmentation parameters, so as to divide the spatial segmentation object units of the floodplain wetland.

[0014] Furthermore, Step 2 includes:

[0015] Based on the collected hydrometeorological data, set the upper and lower boundaries of the region, define the roughness parameters of the model, and use the two-dimensional incompressible Navier-Stokes equation to construct a hydrodynamic model. The continuous equation and momentum equation obtained by integrating the two-dimensional incompressible Navier-Stokes averaged equation along the water depth are shown as follows:

[0016] Continuous equation:

[0017]

[0018] Momentum equation:

[0019]

[0020] In the formula, h is the water depth; t is the time step; and are the flow velocities in the transverse x and longitudinal y directions respectively;u , v are the components of the vertical average velocity in the transverse x and longitudinal y directions respectively; z is the water level; g is the acceleration due to gravity; C is the Chezy coefficient; v t is the turbulent viscosity coefficient;

[0021] Considering the hydrological regime indicators closely related to NEE of floodplain wetlands, based on the water depth data of each upscaled NEE calculation unit of floodplain wetlands obtained from the hydrodynamic model, the hydrological regime indicators of each unit are obtained.

[0022] Furthermore, step three includes:

[0023] Integrate the upscaled calculation units generated by object-oriented image analysis with NEE data, as well as the remote sensing bands, environmental variables, and hydrological regime indicators of the corresponding calculation units in space and time to construct a unified dataset;

[0024] Based on the integrated dataset, establish multiple machine learning models, and determine the optimal model parameters through grid search. The machine learning models include random forest, support vector regression, extreme gradient boosting algorithm, gradient boosting decision tree, K-nearest neighbor, and multi-layer perceptron;

[0025] Use the Powell algorithm to integrate and optimize the prediction results of multiple machine learning models, determine the optimal weight combination of each model, and aim to minimize the prediction error to achieve the optimal model integration scheme:

[0026]

[0027] In the formula, is the final predicted NEE; is the number of integrated models; is the weight of the i-th machine learning model; is the i th machine learning model's predicted NEE;

[0028] For each machine learning model participating in the integration , the Powell algorithm finds the optimal weights by minimizing the difference between the observed NEE and the predicted NEE, and is restricted by the following constraints:

[0029]

[0030] The Powell algorithm iteratively explores the search space along a set of linear directions to determine the weight combination that can minimize the prediction error, and finally obtains the predicted NEE. The objective function is as follows:

[0031]

[0032] In the formula, is the observed NEE of the k th observation value, and

[0033] is the number of observations.

[0034] Further, step four includes:

[0035] Based on the optimal integrated model solution determined in the early stage, and taking the upscaling calculation unit as the basis, conduct refined NEE spatial upscaling analysis for the entire floodplain wetland research area;

[0036] Second invention, the present invention provides a floodplain wetland carbon flux upscaling system integrating a hydrodynamic model and integrated machine learning, including a memory, a processor, and a computer program stored on the memory and operable on the processor. When the computer program is loaded into the processor, it implements the above-mentioned method for upscaling the carbon flux of floodplain wetlands integrating a hydrodynamic model and integrated machine learning.

[0037] The present invention has the following beneficial effects: The method and system for upscaling the carbon flux of floodplain wetlands integrating a hydrodynamic model and integrated machine learning of the present invention collect floodplain wetland eddy covariance measurement data, digital elevation model data, hydrometeorological data, and remote sensing image data, and use an object-oriented calculation method to conduct multi-scale segmentation of the floodplain wetland underlying surface to obtain the calculation unit for carbon flux upscaling; based on the floodplain wetland hydrometeorological data, establish a floodplain wetland hydrodynamic model to obtain the floodplain wetland water level parameters and calculate the relevant hydrological situation index factors affecting the floodplain wetland NEE; integrate the eddy covariance measurement data, hydrological situation index factors, and environmental variables based on remote sensing image data, construct multiple machine learning models, and use the Powell algorithm to optimize the prediction weights of each model to obtain the optimal model integrated prediction solution; based on the optimal model integrated solution, conduct NEE spatial upscaling to realize the characterization of the spatio-temporal changes of the carbon flux of the floodplain wetland ecosystem, use the differences in multi-model predictions to quantitatively characterize the prediction uncertainty, and obtain the spatio-temporal distribution map and uncertainty map of NEE; achieve high-precision prediction of wetland NEE upscaling, and provide a new technical method and reliable data support for floodplain wetland carbon flux research and ecological management. Description of the Drawings

[0038] To more clearly illustrate the technical solution of the present invention, the following will briefly introduce the attached drawings required in the embodiments. Obviously, for those of ordinary skill in the art, without creative efforts, other attached drawings can also be obtained based on these drawings.

[0039] Figure 1 It is a flowchart of the upscaling method for floodplain wetland carbon flux that integrates a hydrodynamic model and integrated machine learning provided by the present invention;

[0040] Figure 2 It is a comparison schematic diagram of predicted NEE based on the integrated model and NEE monitored at the eddy covariance site. Detailed implementation manners

[0041] To make the objectives, technical solutions, and advantages of the present invention clearer, the following will clearly and completely describe the technical solutions of the present invention in combination with specific embodiments and corresponding attached drawings of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present invention. The following will detail the technical solutions provided by each embodiment of the present invention in combination with the attached drawings.

[0042] Please refer to Figure 1 , an upscaling method for floodplain wetland carbon flux that integrates a hydrodynamic model and integrated machine learning provided by an embodiment of the present invention, includes:

[0043] Step 1: Collect eddy covariance measurement data, digital elevation model data, hydrometeorological data, and remote sensing image data of floodplain wetlands, and use an object-oriented calculation method to carry out multi-scale segmentation of the floodplain wetland underlying surface to obtain calculation units for carbon flux upscaling.

[0044] This step specifically includes: obtaining continuous and high-frequency NEE data through an eddy covariance tower installed on-site in the floodplain wetland, and using EddyPro software for coordinate rotation correction, trend correction, data synchronization, density correction, ultrasonic virtual temperature correction, spectral correction, angle-of-attack correction, and data quality control processing.

[0045] Obtain multi-temporal remote sensing data band reflectance data, digital elevation model (DEM) data, and daily data such as flow, water level, precipitation, temperature, wind speed, enhanced vegetation index (EVI), and fraction of absorbed photosynthetically active radiation (Fpar) of the floodplain wetland.

[0046] The multi - resolution segmentation algorithm in eCognition Developer software is used to perform object - oriented multi - scale segmentation on remote sensing images. By testing different combinations of scale parameters, shape weights, and compactness weights, the optimal segmentation parameters are determined, so as to accurately divide the spatial segmentation object units of floodplain wetlands. In this example, the optimal segmentation parameters are determined as a scale parameter of 250 meters, a shape weight of 0.1, and a compactness weight of 0.5.

[0047] Step 2: Based on the hydrometeorological data of floodplain wetlands, establish a hydrodynamic model of floodplain wetlands, obtain the water level parameters of floodplain wetlands, and calculate the relevant hydrological regime index factors affecting the NEE of floodplain wetlands.

[0048] Specifically, based on the collected hydrometeorological basic data, set the upper and lower boundaries of the region, define the roughness parameters of the model, and use the two - dimensional incompressible Navier - Stokes equation to construct a hydrodynamic model. The continuous equation and momentum equation obtained by integrating the two - dimensional incompressible Navier - Stokes averaged equation along the water depth are as follows:

[0049] Continuous equation:

[0050]

[0051] Momentum equation:

[0052]

[0053] In the formula, h is the water depth; t is the time step; 、 are the flow velocities in the transverse x transverse and longitudinal y directions respectively; u 、 v are the components of the vertical average flow velocity in the transverse x 、longitudinal y directions respectively; z is the water level; g is the acceleration of gravity; C is the Chezy coefficient; v t is the eddy viscosity coefficient.

[0054] Considering the hydrological regime indicators closely related to NEE in floodplain wetlands, in this embodiment, the average water level (WL) within 8 days, the maximum flood level (MFL), the inundation frequency (IF), and the water level fluctuation (WLF) are used as representative hydrological regime indicators, as shown in Table 1. The water depth data of each upscaled NEE calculation unit in the floodplain wetland obtained from the hydrodynamic model is used to obtain the hydrological regime indicators of each unit.

[0055] Table 1. Definitions and Descriptions of Hydrological Regime Indicators

[0056]

[0057] Among them, T is the 8-day time interval; is the water level on the t th day; is the number of days when the water level exceeds the flood critical value; , respectively represent the highest water level and the lowest water level within 8 days.

[0058] Step 3: Integrate the eddy covariance measurement data, hydrological regime indicator factors, and environmental variables based on remote sensing image data, construct multiple machine learning models, and use the Powell algorithm to optimize the prediction weights of each model to obtain the optimal model ensemble prediction scheme.

[0059] This step specifically includes: spatiotemporally integrating the upscaled calculation units generated by the object-based image analysis (OBIA) method with the NEE data, as well as the remote sensing bands, environmental variables, and hydrological regime indicators of the corresponding calculation units to construct a unified dataset. Among them, the environmental variables include precipitation, temperature, EVI, and Fpar; the hydrological regime indicators include WL, MFL, IF, and WLF.

[0060] Considering the time resolution characteristics of remote sensing images, in this study, an 8-day time window is used to address the differences in time resolution between variables. The hydrological indicators are calculated using established formulas, and the environmental and hydrological variables are taken as the average values every 8 days. The NEE, as the response variable, is taken as the average value every 8 days to ensure the spatiotemporal consistency of the model input data.

[0061] Based on the integrated dataset, multiple machine learning models are established, and the best model parameters are determined through grid search. In this example, six popular machine learning models are selected, including random forest, support vector regression, extreme gradient boosting algorithm (XGBoost), gradient boosting decision tree, K-nearest neighbor, and multi-layer perceptron.

[0062] The Powell optimization algorithm is used to integrate and optimize the prediction results of multiple machine learning models, determine the optimal weight combination for each model, and aim to minimize the prediction error to achieve the optimal model integration scheme:

[0063]

[0064] where is the final predicted NEE; is the number of integrated models; is the weight of the i-th machine learning model; is the NEE predicted by the i-th machine learning model.

[0065] For each machine learning model participating in the integration , the Powell method finds the best weights by minimizing the difference between the observed NEE and the predicted NEE, and is restricted by the following constraints:

[0066]

[0067] The Powell method iteratively explores the search space along a set of linear directions to determine the weight combination that can minimize the prediction error, and finally obtains the predicted NEE, as Figure 2 . The objective function is as follows:

[0068]

[0069] where is the observed NEE of the k-th observation, is the number of observations.

[0070] Step 4: Based on the optimal model integration scheme, conduct NEE spatial upscaling to characterize the spatio-temporal variation of the carbon flux in the floodplain wetland ecosystem, use the multi-model prediction differences to quantitatively characterize the prediction uncertainty, and obtain the spatio-temporal distribution map and uncertainty map of NEE.

[0071] This step specifically includes: Based on the optimal integrated model scheme determined in the early stage, taking the upscaling calculation unit as the basis, conduct refined NEE spatial upscaling analysis for the entire floodplain wetland research area. Specifically, use an 8-day time window as the unit, and conduct continuous dynamic NEE predictions for each spatial unit in the region one by one, fully capturing the changing rules of the carbon exchange process in the wetland ecosystem. The refined NEE spatial distribution and its continuous time series data of all spatial units in the region are obtained, realizing the accurate characterization of the spatial heterogeneity and temporal dynamics of the carbon flux in the wetland ecosystem.

[0072] Based on spatial upscaling, the standard error (STDE) of the predicted values of the multi-model integration scheme is calculated by using the differences in the predictions of each single model during the model integration process, so as to quantitatively characterize the prediction accuracy of the model. Based on the uncertainty assessment results, a high-precision uncertainty map of the spatio-temporal distribution of NEE is made to provide more scientific and robust data support for wetland carbon flux research and ecological environment management.

[0073] An embodiment of the present invention further provides a floodplain wetland carbon flux upscaling system integrating a hydrodynamic model and integrated machine learning, which is characterized by comprising a memory, a processor, and a computer program stored on the memory and executable on the processor. When the computer program is loaded into the processor, the above-mentioned method for upscaling floodplain wetland carbon flux integrating a hydrodynamic model and integrated machine learning is implemented.

[0074] The above-described embodiments of the present invention do not constitute a limitation on the protection scope of the present invention.

Claims

1. A method for upscaling carbon fluxes in flooded wetlands by integrating hydrodynamic models and integrated machine learning, characterized in that, Including: Step 1: Collect eddy covariance measurement data, digital elevation model data, hydrometeorological data, and remote sensing image data of floodplain wetlands, and conduct multi-scale segmentation of the underlying surface of floodplain wetlands using object-oriented calculation methods to obtain calculation units for upscaling carbon flux. Among them, NEE data is obtained through an eddy covariance tower installed on-site in floodplain wetlands, and EddyPro software is used for coordinate rotation correction, trend correction, data synchronization, density correction, ultrasonic virtual temperature correction, spectral correction, angle-of-attack correction, and data quality control processing. Obtain remote sensing data band reflectance data, digital elevation model data of floodplain wetlands for multiple time phases, as well as daily flow, water level, precipitation, temperature, wind speed, vegetation index, and vegetation photosynthetically active radiation data. Use the multi-resolution segmentation algorithm in eCognition Developer software to perform object-oriented multi-scale segmentation on remote sensing images. By testing different combinations of scale parameters, shape weights, and compactness weights, determine the optimal segmentation parameters, thereby dividing the spatial segmentation object units of floodplain wetlands. Step 2: Based on the hydrometeorological data of floodplain wetlands, establish a hydrodynamic model of floodplain wetlands to obtain water level parameters of floodplain wetlands, and calculate relevant hydrological regime index factors affecting NEE of floodplain wetlands. Among them, based on the collected hydrometeorological data, set the upper and lower boundaries of the region, define the roughness parameters of the model, and use the two-dimensional incompressible Navier-Stokes equation to construct a hydrodynamic model. The continuous equation and momentum equation obtained by integrating the two-dimensional incompressible Navier-Stokes averaged equation along the water depth are shown as follows: Continuous equation: Momentum equation: In the formula, h is the water depth; t is the time step; , are the flow velocities in the transverse x and longitudinal y directions respectively; u , v are the components of the vertical average flow velocity in the transverse x and longitudinal y directions respectively; z is the water level; g is the acceleration due to gravity; C is the Chezy coefficient; v t is the turbulent viscosity coefficient; Considering the hydrological regime indicators closely related to NEE of floodplain wetlands, based on the water depth data of each upscaling NEE calculation unit of floodplain wetlands obtained from the hydrodynamic model, obtain the hydrological regime indicators of each unit. Step 3: Integrate eddy covariance measurement data, hydrological regime index factors, and environmental variables based on remote sensing image data, construct multiple machine learning models, and use the Powell algorithm to optimize the prediction weights of each model to obtain an optimal model integration prediction scheme. Step 4: Based on the optimal model integration scheme, conduct NEE spatial upscaling, realize the characterization of the spatio-temporal variation of the carbon flux of the floodplain wetland ecosystem, use the multi-model prediction difference to quantitatively characterize the prediction uncertainty, and obtain the spatio-temporal distribution map and uncertainty map of NEE.

2. The upscaling method for floodplain wetland carbon flux integrating a hydrodynamic model and integrated machine learning according to claim 1, characterized in that, Step 3 includes: Spatially and temporally integrate the upscaling calculation units generated by object-oriented image analysis with NEE data, remote sensing bands, environmental variables, and hydrological regime indicators of the corresponding calculation units to construct a unified dataset. Based on the integrated dataset, establish multiple machine learning models, and determine the best model parameters through grid search. The machine learning models include random forest, support vector regression, extreme gradient boosting algorithm, gradient boosting decision tree, K-nearest neighbor, and multi-layer perceptron. Use the Powell algorithm to integrate and optimize the prediction results of multiple machine learning models, determine the optimal weight combination of each model, and aim to minimize the prediction error to achieve the optimal model integration scheme: wherein, is the final predicted NEE; is the number of integrated models; is the weight of the i-th machine learning model; is the i -th machine learning model's prediction of NEE; For each machine learning model participating in the integration , the Powell algorithm finds the optimal weights by minimizing the difference between the observed NEE and the predicted NEE, and is restricted by the following constraints: The Powell algorithm iteratively explores the search space along a set of linear directions to determine the weight combination that minimizes the prediction error, and finally obtains the predicted NEE. The objective function is as follows: In the formula, is the observed NEE of the k th observation value, is the number of observations.

3. The upscaling method for floodplain wetland carbon flux integrating a hydrodynamic model and integrated machine learning according to claim 1, characterized in that, Step four includes: Based on the previously determined optimal ensemble model scheme, and taking the upscaling calculation unit as the basis, conduct refined NEE spatial upscaling analysis for the entire floodplain wetland research area; On the basis of spatial upscaling, use the differences in the predictions of each single model during the model integration process to calculate the standard error of the predicted values of the multi-model integration scheme, so as to quantitatively characterize the model prediction accuracy; based on the uncertainty assessment results, produce the NEE spatio-temporal distribution uncertainty map.

4. A floodplain wetland carbon flux upscaling system integrating a hydrodynamic model and integrated machine learning, characterized in that, It includes a memory, a processor, and a computer program stored on the memory and executable on the processor. When the computer program is loaded into the processor, it implements the floodplain wetland carbon flux upscaling method that fuses the hydrodynamic model and integrated machine learning as described in any one of claims 1-3.

Images