Hydrodynamic model and integrated machine learning fused flooding wetland carbon flux upscaling method and system

Abstract

The invention discloses a flooding wetland carbon flux upscaling method and system fusing a hydrodynamic model and integrated machine learning. The invention relates to the technical field of water conservancy and ecological environment monitoring management, and discloses a method for carrying out multi-scale segmentation on an underlying surface of a flooding wetland by collecting flooding wetland vortex covariance measurement, a digital elevation model, hydro meteorology and remote sensing image data to obtain a carbon flux upscaling calculation unit; a flooding wetland hydrodynamic model is established, and related hydrological situation index factors influencing the flooding wetland NEE are calculated; constructing a plurality of machine learning models, and optimizing the prediction weight of each model to obtain an optimal model integration prediction scheme; nEE spatial upscaling is carried out, multi-model prediction difference is utilized to quantitatively characterize prediction uncertainty, and a NEE space-time distribution diagram and an uncertainty diagram are obtained; according to the method, the wetland NEE upscaling high-precision prediction is realized, and a new technical method and reliable data support are provided for flooding wetland carbon flux research 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 in particular to a method and system for upscaling carbon flux in flooded wetlands by integrating a hydrodynamic model and integrated machine learning. Background Art

[0002] Flooded wetlands are one of the most important types of wetlands, with multiple functions such as regulating floods, purifying water bodies, harbouring plants and animals, and fixing carbon. Flooded wetlands are hotspots of global carbon flux changes due to their frequent water level fluctuations and rich carbon and nitrogen reserves. Climate change and human activities are accelerating changes in flooded wetland ecosystems, leading to deterioration of the water environment, fragmentation of the underlying surface, and interruption of the natural carbon cycle. Clarifying the spatiotemporal trends of carbon flux in flooded wetlands under a changing environment 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 carbon dioxide exchange (Net Ecosystem Exchange, NEE) of flooded wetlands at regional or global scales.

[0003] Traditional site-scale eddy covariance observations are difficult to meet the needs of regional or global scale NEE assessment. At present, pixel-scale remote sensing data and a single machine learning model are mostly used for NEE upscaling, which has problems such as difficulty in segmenting wetland patch heterogeneity, neglect of hydrological processes, and limited prediction accuracy of a single model. Therefore, it is urgent to propose a method and system that comprehensively considers the influence of wetland heterogeneity and hydrological processes and accurately upscales 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 flux in flooded wetlands that integrates hydrodynamic models and integrated machine learning, so as to solve the problems of difficulty in segmenting wetland patch heterogeneity, neglect of hydrological processes, and limited prediction accuracy of a single model in the prior art.

[0005] In a first aspect, the present invention provides a method for upscaling carbon flux in flooded wetlands by integrating a hydrodynamic model and integrated machine learning, comprising: Step 1: Collect eddy covariance measurement data, digital elevation model data, hydrological and meteorological data, and remote sensing image data of flooded wetlands, and use object-oriented computing methods to perform multi-scale segmentation of the flooded wetland underlying surface to obtain the upscaling calculation unit of carbon flux; Step 2: Based on the hydrological and meteorological data of floodplain wetlands, a hydrodynamic model of floodplain wetlands is established to obtain the water level parameters of floodplain wetlands and calculate the relevant hydrological indicator factors affecting the NEE of floodplain wetlands; Step 3: Integrate eddy covariance measurement data, hydrological index factors, and environmental variables based on remote sensing image data to build multiple machine learning models, and use the Powell algorithm to optimize the prediction weights of each model to obtain the optimal model integration prediction solution; Step 4: Based on the optimal model integration scheme, NEE spatial upscaling is carried out to characterize the spatiotemporal changes of carbon flux in flooded wetland ecosystems, and the prediction uncertainty is quantitatively characterized using multi-model prediction differences to obtain NEE spatiotemporal distribution maps and uncertainty maps.

[0006] Furthermore, step one includes: The NEE data were obtained through the eddy covariance tower installed on site in the flooded wetland, and EddyPro software was used to perform coordinate rotation correction, trend correction, data synchronization, density correction, ultrasonic virtual temperature correction, spectrum correction, angle of attack correction and data quality control processing; Obtain multi-temporal remote sensing data band reflectance data, digital elevation model data, and daily flow, water level, precipitation, temperature, wind speed, vegetation index, and vegetation photosynthetically active radiation data of flooded wetlands; The multi-resolution segmentation algorithm in the eCognition Developer software was used to perform object-oriented multi-scale segmentation of remote sensing images. By testing different combinations of scale parameters, shape weights, and compactness weights, the optimal segmentation parameters were determined to divide the spatial segmentation object units of the floodplain wetland.

[0007] Furthermore, step 2 includes: Based on the collected hydrological and meteorological data, the upper and lower boundaries of the region are set, the roughness parameters of the model are defined, and the hydrodynamic model is constructed using the two-dimensional incompressible Navier-Stokes equation. The continuity equation and momentum equation formulas of the two-dimensional incompressible Navier-Stokes average equation integrated along the water depth are as follows: Continuity equation:

[0008] Momentum equation:

[0009] In the formula, h for water depth; t is the time step; , Horizontal x , Vertical y Flow rate in direction; u , v The vertical average velocity in the horizontal direction is x , Vertical y Directional weight; z is the water level; gis the acceleration due to gravity; C is Xiecai coefficient; v t is the turbulent viscosity coefficient; Considering the hydrological indicators that are closely related to the NEE of floodplain wetlands, the hydrological indicators of each unit are obtained based on the water depth data of each scale NEE calculation unit of the floodplain wetland obtained by the hydrodynamic model.

[0010] Furthermore, step three includes: The upscaled computational units generated by object-oriented image analysis were spatially and temporally integrated with NEE data and the remote sensing bands, environmental variables, and hydrological regime indicators of the corresponding computational units to construct a unified data set; Establishing multiple machine learning models based on the integrated data set and determining the optimal model parameters through grid search, wherein the machine learning models include random forest, support vector regression, extreme gradient boosting algorithm, gradient boosting decision tree, K-nearest neighbor and multilayer perceptron; The Powell algorithm is used to integrate and optimize the prediction results of multiple machine learning models, determine the optimal weight combination of each model, and achieve the optimal model integration solution with the goal of minimizing the prediction error:

[0011] In the formula, It is the final prediction NEE; is the number of models in the ensemble; is the weight of the i-th machine learning model; It is i A machine learning model predicts NEE; For each machine learning model participating in the ensemble , Powell's algorithm finds the optimal weights by minimizing the difference between the observed NEE and the predicted NEE, subject to the following constraints:

[0012] 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:

[0013] In the formula, It is k The observed NEE of observations, is the number of observations.

[0014] Furthermore, step four includes: Based on the optimal integrated model scheme determined in the early stage, a refined NEE spatial upscaling analysis was carried out for the entire floodplain wetland study area based on the upscaling calculation unit; On the basis of spatial upscaling, the differences in predictions of each single model in the model integration process are used to calculate the standard error of the prediction value of the multi-model integration scheme, so as to quantitatively characterize the model prediction accuracy; based on the uncertainty assessment results, an uncertainty map of the spatiotemporal distribution of NEE is produced.

[0015] The second invention is to provide a flooded wetland carbon flux upscaling system that integrates a hydrodynamic model and integrated machine learning, including a memory, a processor, and a computer program stored on the memory and executable on the processor, wherein when the computer program is loaded into the processor, the flooded wetland carbon flux upscaling method that integrates a hydrodynamic model and integrated machine learning is implemented.

[0016] The present invention has the following beneficial effects: a method and system for upscaling carbon flux in flooded wetlands that integrates a hydrodynamic model and integrated machine learning collects eddy covariance measurement data, digital elevation model data, hydrological and meteorological data, and remote sensing image data of flooded wetlands, and uses an object-oriented calculation method to perform multi-scale segmentation of the underlying surface of flooded wetlands to obtain a calculation unit for upscaling carbon flux; based on the hydrological and meteorological data of flooded wetlands, a hydrodynamic model of flooded wetlands is established to obtain water level parameters of flooded wetlands and calculate relevant hydrological situation indicator factors affecting NEE of flooded wetlands; eddy covariance measurement data are integrated to obtain the hydrological and meteorological data of flooded wetlands; and the hydrodynamic model of flooded wetlands is established to obtain water level parameters of flooded wetlands and calculate relevant hydrological situation indicator factors affecting NEE of flooded wetlands. Based on the data of climate change, hydrological indicator factors and environmental variables based on remote sensing image data, multiple machine learning models were constructed, and the Powell algorithm was used to optimize the prediction weights of each model to obtain the optimal model integration prediction scheme; based on the optimal model integration scheme, NEE spatial upscaling was carried out to realize the characterization of the spatiotemporal changes of carbon flux in flooded wetland ecosystems, and the prediction uncertainty was quantitatively characterized by multi-model prediction differences, and the NEE spatiotemporal distribution map and uncertainty map were obtained; the high-precision prediction of wetland NEE upscaling was achieved, which provided new technical methods and reliable data support for the research on carbon flux and ecological management of flooded wetlands. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solution of the present invention, the drawings required for use in the embodiments are briefly introduced below. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0018] Figure 1 A flow chart of the method for upscaling carbon flux in flooded wetlands by integrating hydrodynamic model and machine learning provided by the present invention; Figure 2 This is a schematic diagram comparing the predicted NEE based on the integrated model and the NEE monitored at the vortex station. DETAILED DESCRIPTION

[0019] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described in conjunction with the specific embodiments of the present invention and the corresponding drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention. The technical solutions provided by the embodiments of the present invention are described in detail below in conjunction with the drawings.

[0020] See also Figure 1 , an embodiment of the present invention provides a method for upscaling carbon flux in flooded wetlands by integrating a hydrodynamic model and integrated machine learning, comprising: Step 1: Collect eddy covariance measurement data, digital elevation model data, hydrological and meteorological data, and remote sensing image data of flooded wetlands, and use object-oriented computing methods to perform multi-scale segmentation of the underlying surface of flooded wetlands to obtain the upscaling calculation unit of carbon flux.

[0021] This step specifically includes: obtaining continuous, high-frequency NEE data through the eddy covariance tower installed on-site in the flooded wetland, and using EddyPro software to perform coordinate rotation correction, trend correction, data synchronization, density correction, ultrasonic virtual temperature correction, spectrum correction, angle of attack correction and data quality control processing.

[0022] Acquire multi-temporal remote sensing data band reflectance data, Digital Elevation Model (DEM) data, as well as daily flow, water level, precipitation, temperature, wind speed, Enhanced Vegetation Index (EVI), and Fraction of absorbed Photosynthetically Active Radiation (Fpar) data of flooded wetlands.

[0023] The multi-resolution segmentation algorithm in the eCognition Developer software is used to perform object-oriented multi-scale segmentation of remote sensing images. By testing different combinations of scale parameters, shape weights, and compactness weights, the optimal segmentation parameters are determined to accurately divide the spatial segmentation object units of the floodplain wetland. In this example, the optimal segmentation parameters are determined to be a scale parameter of 250 meters, a shape weight of 0.1, and a compactness weight of 0.5.

[0024] Step 2: Based on the hydrological and meteorological data of floodplain wetlands, a hydrodynamic model of floodplain wetlands is established to obtain the water level parameters of floodplain wetlands and calculate the relevant hydrological situation indicator factors affecting the NEE of floodplain wetlands.

[0025] Specifically, based on the collected basic hydrological and meteorological data, the upper and lower boundaries of the region are set, the roughness parameters of the model are defined, and the two-dimensional incompressible Navier-Stokes equation is used to construct the hydrodynamic model. The continuity equation and momentum equation formulas of the two-dimensional incompressible Navier-Stokes average equation integrated along the water depth are as follows: Continuity equation:

[0026] Momentum equation:

[0027] In the formula, h for water depth; t is the time step; , Horizontal x Horizontal, vertical y Flow rate in direction; u , v The vertical average velocity in the horizontal direction is x , Vertical y Directional weight; z is the water level; g is the acceleration due to gravity; C is Xiecai coefficient; v t is the turbulent viscosity coefficient.

[0028] Considering the hydrological indicators closely related to the NEE of flooded wetlands, the average water level (WL), maximum flood level (MFL), inundation frequency (IF) and water level fluctuation (WLF) within 8 days are used as representative hydrological indicators in this embodiment, see Table 1. The hydrological indicators of each unit are obtained based on the water depth data of each scale NEE calculation unit of the flooded wetland obtained by the hydrodynamic model.

[0029] Table 1. Definition and description of hydrological regime indicators

[0030] in, T It is an 8-day interval; It is t the water level of the day; is the number of days that the water level exceeds the flood threshold; , Respectively represent the highest and lowest water levels within 8 days.

[0031] Step three, integrate eddy covariance measurement data, hydrological situation index factors and environmental variables based on remote sensing image data, build multiple machine learning models, and use the Powell algorithm to optimize the prediction weights of each model to obtain the optimal model integration prediction solution.

[0032] This step specifically includes: integrating the upscaled calculation units generated by the object-oriented computing method (OBIA) with the NEE data and the remote sensing bands, environmental variables, and hydrological indicators of the corresponding calculation units in time and space to build a unified data set. Among them, environmental variables include precipitation, temperature, EVI, and Fpar; hydrological indicators include WL, MFL, IF, and WLF.

[0033] Considering the temporal resolution characteristics of remote sensing images, this study uses an 8-day time window to solve the difference in temporal resolution between variables. Hydrological indicators are calculated using established formulas, and environmental and hydrological variables are averaged every 8 days. NEE, as the response variable, is averaged every 8 days to ensure the temporal and spatial consistency of the model input data.

[0034] Based on the integrated data set, multiple machine learning models are established, and the optimal model parameters are determined through grid search. This example selects six popular machine learning models, including random forest, support vector regression, extreme gradient boosting algorithm (XGBoost), gradient boosting decision tree, K-nearest neighbor, and multilayer perceptron.

[0035] The Powell optimization algorithm is used to integrate and optimize the prediction results of multiple machine learning models, determine the optimal weight combination of each model, and achieve the optimal model integration solution with the goal of minimizing the prediction error:

[0036] In the formula, It is the final prediction NEE; is the number of models in the ensemble; is the weight of the i-th machine learning model; is the NEE predicted by the i-th machine learning model.

[0037] For each machine learning model participating in the ensemble , Powell's method finds the optimal weights by minimizing the difference between the observed NEE and the predicted NEE, subject to the following constraints:

[0038] The Powell method 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, such as Figure 2 The objective function is as follows:

[0039] In the formula, is the observed NEE of the kth observation, is the number of observations.

[0040] Step 4: Based on the optimal model integration scheme, NEE spatial upscaling is carried out to characterize the spatiotemporal changes of carbon flux in flooded wetland ecosystems, and the prediction uncertainty is quantitatively characterized using multi-model prediction differences to obtain NEE spatiotemporal distribution maps and uncertainty maps.

[0041] This step specifically includes: Based on the optimal integrated model scheme determined in the early stage, and based on the upscaling calculation unit, a refined NEE spatial upscaling analysis is carried out for the entire floodplain wetland study area. Specifically, using an 8-day time window as a unit, a continuous dynamic NEE prediction is carried out for each spatial unit in the region, fully capturing the changing laws 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 an accurate description of the spatial heterogeneity and temporal dynamics of carbon flux in the wetland ecosystem.

[0042] On the basis of spatial upscaling, the standard error (STDE) of the prediction value of the multi-model integration scheme is calculated by using the differences in the predictions of each single model in the model integration process, so as to quantitatively characterize the model prediction accuracy. Based on the uncertainty assessment results, a high-precision NEE spatiotemporal distribution uncertainty map is produced to provide more scientific and robust data support for wetland carbon flux research and ecological environment management.

[0043] An embodiment of the present invention also provides a flooded wetland carbon flux upscaling system that integrates a hydrodynamic model and integrated machine learning, characterized in that it includes a memory, a processor, and a computer program stored in the memory and executable on the processor, and when the computer program is loaded into the processor, it implements the above-mentioned flooded wetland carbon flux upscaling method that integrates a hydrodynamic model and integrated machine learning.

[0044] The above-described embodiments of the present invention do not limit the protection scope of the present invention.

Claims

1. A method for upscaling carbon flux in flooded wetlands by integrating hydrodynamic modeling and machine learning, characterized in that: include: Step 1: Collect eddy covariance measurement data, digital elevation model data, hydrological and meteorological data, and remote sensing image data of flooded wetlands, and use object-oriented computing methods to perform multi-scale segmentation of the flooded wetland underlying surface to obtain the upscaling calculation unit of carbon flux; Step 2: Based on the hydrological and meteorological data of floodplain wetlands, a hydrodynamic model of floodplain wetlands is established to obtain the water level parameters of floodplain wetlands and calculate the relevant hydrological indicator factors affecting the NEE of floodplain wetlands; Step 3: Integrate eddy covariance measurement data, hydrological index factors, and environmental variables based on remote sensing image data to build multiple machine learning models, and use the Powell algorithm to optimize the prediction weights of each model to obtain the optimal model integration prediction solution; Step 4: Based on the optimal model integration scheme, NEE spatial upscaling is carried out to characterize the spatiotemporal changes of carbon flux in flooded wetland ecosystems, and the prediction uncertainty is quantitatively characterized using multi-model prediction differences to obtain NEE spatiotemporal distribution maps and uncertainty maps.

2. The method for upscaling carbon flux in flooded wetlands by integrating hydrodynamic modeling and machine learning as claimed in claim 1, characterized in that: Step one includes: The NEE data were obtained through the eddy covariance tower installed on site in the flooded wetland, and EddyPro software was used to perform coordinate rotation correction, trend correction, data synchronization, density correction, ultrasonic virtual temperature correction, spectrum correction, angle of attack correction and data quality control processing; Obtain multi-temporal remote sensing data band reflectance data, digital elevation model data, and daily flow, water level, precipitation, temperature, wind speed, vegetation index, and vegetation photosynthetically active radiation data of flooded wetlands; The multi-resolution segmentation algorithm in the eCognition Developer software was used to perform object-oriented multi-scale segmentation of remote sensing images. By testing different combinations of scale parameters, shape weights, and compactness weights, the optimal segmentation parameters were determined to divide the spatial segmentation object units of the floodplain wetland.

3. The method for upscaling carbon flux in flooded wetlands by integrating hydrodynamic modeling and machine learning as claimed in claim 1, characterized in that: Step 2 includes: Based on the collected hydrological and meteorological data, the upper and lower boundaries of the region are set, the roughness parameters of the model are defined, and the hydrodynamic model is constructed using the two-dimensional incompressible Navier-Stokes equation. The continuity equation and momentum equation formulas of the two-dimensional incompressible Navier-Stokes average equation integrated along the water depth are as follows: Continuity equation: Momentum equation: In the formula, h for water depth; t is the time step; , Horizontal x , Vertical y Flow rate in direction; u , v The vertical average velocity in the horizontal direction is x , Vertical y Directional weight; z is the water level; g is the acceleration due to gravity; C is Xiecai coefficient; v t is the turbulent viscosity coefficient; Considering the hydrological indicators that are closely related to the NEE of floodplain wetlands, the hydrological indicators of each unit are obtained based on the water depth data of each scale NEE calculation unit of the floodplain wetland obtained by the hydrodynamic model.

4. The method for upscaling carbon flux in flooded wetlands by integrating hydrodynamic modeling and machine learning as claimed in claim 1, characterized in that: Step three includes: The upscaled computational units generated by object-oriented image analysis were spatially and temporally integrated with NEE data and the remote sensing bands, environmental variables, and hydrological regime indicators of the corresponding computational units to construct a unified data set; Establishing multiple machine learning models based on the integrated data set and determining the optimal model parameters through grid search, wherein the machine learning models include random forest, support vector regression, extreme gradient boosting algorithm, gradient boosting decision tree, K-nearest neighbor and multilayer perceptron; The Powell algorithm is used to integrate and optimize the prediction results of multiple machine learning models, determine the optimal weight combination of each model, and achieve the optimal model integration solution with the goal of minimizing the prediction error: In the formula, It is the final prediction NEE; is the number of models in the ensemble; is the weight of the i-th machine learning model; It is i A machine learning model predicts NEE; For each machine learning model participating in the ensemble , Powell's algorithm finds the optimal weights by minimizing the difference between the observed NEE and the predicted NEE, subject to 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, It is k The observed NEE of observations, is the number of observations.

5. The method for upscaling carbon flux in flooded wetlands by integrating hydrodynamic modeling and machine learning as claimed in claim 1, characterized in that: Step 4 includes: Based on the optimal integrated model scheme determined in the early stage, a refined NEE spatial upscaling analysis was carried out for the entire floodplain wetland study area based on the upscaling calculation unit; On the basis of spatial upscaling, the differences in predictions of each single model in the model integration process are used to calculate the standard error of the prediction value of the multi-model integration scheme, so as to quantitatively characterize the model prediction accuracy; based on the uncertainty assessment results, an uncertainty map of the spatiotemporal distribution of NEE is produced.

6. A carbon flux upscaling system for flooded wetlands that integrates hydrodynamic modeling and machine learning, characterized by: The invention comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein when the computer program is loaded into the processor, the carbon flux upscaling method for flooded wetlands integrating a fusion hydrodynamic model and machine learning as described in any one of claims 1 to 5 is implemented.

Images