A method for monitoring N2O greenhouse gas emissions from surface water bodies based on remote sensing and intelligent algorithms

Abstract

The application provides a kind of surface water body N2O greenhouse gas emission monitoring method based on remote sensing and intelligent algorithm, first with surface water body as research object, by measured water quality data, explore the key water quality parameters affecting dissolved N2O, using deep neural network algorithm to build dissolved N2O prediction model;Using multi-source remote sensing data, ground hyperspectral information and water quality measured data, build water temperature, pH, dissolved oxygen, inorganic nitrogen remote sensing inversion algorithm;Further combined with dissolved N2O prediction model, water-gas interface N2O emission model, finally build a kind of surface water body N2O greenhouse gas emission monitoring method based on remote sensing and intelligent algorithm, realize the spatial inversion of surface water body N2O generation and emission based on remote sensing.The design fully combines the advantages of remote sensing and intelligent algorithm, more comprehensive, systematic and accurate estimation of surface water body N2O generation and emission, provide scientific monitoring technology and data support for water ecological system greenhouse gas emission reduction.

Description

Technical Field

[0001] This invention relates to the field of greenhouse gas monitoring technology, and to a method for quantitatively assessing greenhouse gas emissions from surface water bodies, particularly a method for monitoring N2O greenhouse gas emissions from surface water bodies based on remote sensing and intelligent algorithms. Background Technology

[0002] Surface water bodies, as a major channel for the transport of substances from land to ocean, carry large amounts of nitrogen nutrients in various forms. They are sinks for non-point source pollutants in watersheds and one of the main sources of nitrogen (N2O) greenhouse gases entering the atmosphere. Typical agricultural areas in my country with numerous water bodies contain wetland landscapes such as ditches, natural aquaculture ponds, and rivers. Large amounts of nitrogen are lost to surrounding water bodies through runoff and diffuse into the atmosphere through the water-air interface, leading to severe water pollution and the greenhouse effect. Unlike point source pollution, non-point source pollution is characterized by its randomness and irregularity. Its formation mechanisms and processes are exceptionally complex, and the emission channels and quantities are difficult to determine accurately. The spatiotemporal distribution of this pollution is highly variable, posing significant challenges to monitoring, simulation, and control. Simultaneous monitoring and reduction of surface water pollution and greenhouse gas emissions face severe technical challenges.

[0003] Current research on N2O emissions from surface water is still very limited, and there is considerable uncertainty regarding the influencing factors and generation mechanisms of N2O in aquatic ecosystems. Furthermore, existing N2O emission estimation studies are mostly based on limited field sampling methods, which are costly and only provide spatially discrete data. This low spatiotemporal resolution sampling and monitoring cannot address the high variability in surface water N2O concentrations caused by dynamic production and consumption processes. Remote sensing, with its comprehensive and multi-temporal Earth observation capabilities, greatly compensates for the shortcomings of field observation methods. Domestic and international scholars mainly study river N2O generation and emissions at different spatial scales by constructing emission statistical models, semi-empirical models, and mechanistic models. Existing empirical models of environmental factors often use simple regression models, which cannot comprehensively extract data features and have low model accuracy. Intelligent algorithms can serve as a substitute or supplement for model simulation and prediction of greenhouse gas emissions. Emission mechanism models are needed. Simulated river greenhouse gas emissions yield relatively accurate results, but these models require high-precision water quality parameters. Water-air interface models, which consider atmospheric dynamics, have been well-received in previous studies, but their accuracy also relies on accurate water quality and meteorological data. Remote sensing-based water quality parameter inversion research is relatively mature; rapidly and accurately acquiring high spatiotemporal resolution water quality information for rivers can provide key water quality parameters and technical support for constructing remote sensing-based river N2O generation and emission models. By utilizing remote sensing technology and intelligent algorithms to construct models that consider the dynamic changes of N2O at the water-air interface, a more comprehensive, systematic, and accurate estimate of surface water greenhouse gas emissions can be achieved, providing a reference for developing more comprehensive climate change mitigation strategies. Summary of the Invention

[0004] In the context of my country's current low-carbon, green, and sustainable development, and in response to existing problems, this invention aims to provide a method for monitoring N2O greenhouse gases in surface water bodies based on remote sensing and intelligent algorithms.

[0005] A method for monitoring N2O greenhouse gas emissions from surface water bodies based on remote sensing and intelligent algorithms comprises the following steps:

[0006] Step 1: Obtaining water quality parameters (water temperature, pH, conductivity, dissolved oxygen content, ammonium nitrogen, nitrate nitrogen, total phosphorus, total nitrogen, suspended solids, water temperature, dissolved N2O concentration);

[0007] Step 2: Investigate the correlation between dissolved N2O and the above water quality parameters, identify the key water quality parameters (water temperature, pH, dissolved oxygen, inorganic nitrogen) that affect dissolved N2O, and use a deep neural network algorithm to build a dissolved N2O prediction model;

[0008] Step 3: Land surface temperature (including water temperature) inversion algorithm based on Landsat-8, Sentinel-2 and MODIS;

[0009] Step 4: Based on multi-point measured water quality data - ground-based measured hyperspectral and Sentinel-2 datasets, construct pH, dissolved oxygen and inorganic nitrogen inversion algorithms;

[0010] Step 5: Based on the above remote sensing inversion algorithms for water temperature, pH, dissolved oxygen, and inorganic nitrogen, the dissolved N2O prediction model is driven. Further combined with the water-air interface N2O gas exchange model, a monitoring method for N2O greenhouse gas emissions from surface water bodies based on remote sensing and intelligent algorithms is finally constructed, realizing the spatial inversion of N2O generation and emission from surface water bodies.

[0011] Step two involves constructing a prediction model for dissolved N2O using a deep neural network algorithm, and includes the following steps:

[0012] S1. The neural network uses the Rectified Linear Function (ReLU) as the activation function, contains 4 intermediate layers, and uses the Adaptive Moment Estimation Optimization Algorithm (Adam) as the optimizer for optimization. The training iterations are 500.

[0013] S2. The collected key water quality parameters (water temperature, pH, dissolved oxygen, inorganic nitrogen) data were randomly divided into 80% and 20% sets. 80% of the sample data was used as the training set and 20% of the sample data was used as the validation set. A deep neural network model was used to construct a fitting model for dissolved N2O.

[0014] S3. Use mean squared error (MSE), mean absolute error (MAE), and coefficient of determination (R²). 2 Three metrics are used to further evaluate the model's accuracy during both the training and validation periods.

[0015] The N2O emission flux formula in the water-air interface N2O gas exchange model of step five is:

[0016]

[0017] In the above formula: The N2O emission flux at the water-air interface, in μg / m³ 2 ·d; The gas exchange rate is expressed in cm / h; N2O disc The concentration of dissolved N2O in the water body is expressed in μg / L. The pH, dissolved oxygen, inorganic nitrogen, and water temperature involved are obtained from remote sensing inversion. eqc The theoretical equilibrium concentration of dissolved N2O in water is given in μg / L.

[0018] (1) The theoretical equilibrium concentration of dissolved N2O in water is calculated based on water temperature, atmospheric N2O concentration, and the Weiss equation, as shown in the following formula:

[0019] N2O eqc =M×F×N2O air

[0020] In F=-165.8806+222.8743×(100 / T)+92.0792×In(T / 100)-1.48425×(T / 100)2

[0021] In the above formula: M is the molecular mass of N2O; T is the thermodynamic temperature, obtained based on remote sensing inversion; N2O air The values ​​represent atmospheric N2O concentrations (ppm) at different times and locations.

[0022] (2) Gas exchange rate The calculation formula is as follows:

[0023]

[0024] S c =2141.2 - 152.56 × T + 5.8963 × T 2 -0.12411×T 3 +0.0010655×T 4

[0025] In the above formula, Sc is the ratio of the dynamic viscosity of water to the diffusion rate of N2O gas molecules, and n is the Schmidt number, which is physically affected by water surface fluctuations. When the wind speed at a height of 10m is greater than 3.6m / s, n = 1 / 2; when the wind speed is less than 3.6m / s, n = 2 / 3; K 600 is the gas exchange coefficient; T is the water temperature, °C, obtained from remote sensing inversion.

[0026] (3) Gas exchange coefficient K 600 The calculation formula is:

[0027] K 600 =0.251×U 10 2

[0028] In the above formula: U 10 The wind speed at an altitude of 10m is given in m / s.

[0029] Compared with the prior art, the advantages of the present invention are:

[0030] This invention provides a method for monitoring N2O greenhouse gas emissions from surface water bodies based on remote sensing and intelligent algorithms. It utilizes Python programming and the Google Earth Engine remote sensing cloud computing platform to automatically extract spectral data from image data online. Based on measured data, it explores key water quality parameters affecting N2O production and constructs a remote sensing inversion algorithm. Based on multi-source remote sensing methods for pH, dissolved oxygen, inorganic nitrogen, and water temperature inversion, it can demonstrate the spatial continuity of water quality changes and obtain large-area continuous observation results. Furthermore, it combines a dissolved N2O prediction model constructed using deep neural network algorithms and a water-air interface N2O gas exchange model to build a high spatiotemporal resolution N2O production and emission simulation method, enabling a more comprehensive, systematic, and accurate estimation of N2O greenhouse gas production and emissions from surface water bodies.

[0031] This invention presents a method for monitoring N2O greenhouse gas emissions from surface water bodies based on remote sensing and intelligent algorithms. It uses a deep neural network algorithm to construct a prediction model for dissolved N2O. Currently, there are few reports on the combination of remote sensing and deep neural network algorithms for predicting dissolved N2O. Compared with other conventional algorithms, the deep neural network algorithm can automatically extract key features from the data, has the best applicability, and the highest model accuracy. Attached Figure Description

[0032] Figure 1 This is a roadmap of the method for monitoring greenhouse gas emissions from surface water bodies based on remote sensing and intelligent algorithms, according to the present invention.

[0033] Figure 2 This is a correlation analysis diagram of water environmental factors and dissolved N2O concentration according to the present invention;

[0034] Figure 3 This is a comparison chart of the predicted and actual values ​​of dissolved N2O in the training set of this invention;

[0035] Figure 4 This is a comparison chart of the predicted and actual values ​​of dissolved N2O in the validation set of this invention;

[0036] Figure 5 A schematic diagram of the measured hyperspectral reflectance of water bodies in this invention;

[0037] Figure 6a and Figure 6b This is a comparison chart of the predicted and measured values ​​of dissolved N2O concentration and emission flux in this invention;

[0038] Figure 7 This is a time series diagram of water temperature inversion based on Landsat-8, Sentinel-2, and MODIS in this invention;

[0039] Figure 8a , Figure 8b , Figure 8c This is a comparison chart of measured and predicted values ​​of key water quality parameters (pH, dissolved oxygen, inorganic nitrogen) based on spectral information in this invention;

[0040] Figure 9a , Figure 9b , Figure 9c This is a spatial distribution map of dissolved N2O concentration in surface water based on remote sensing, as described in this invention.

[0041] Figure 10a , Figure 10b , Figure 10c This is a spatial distribution map of N2O emission flux from surface water bodies based on remote sensing, as described in this invention. Detailed Implementation

[0042] To make the objectives, technical solutions, and advantages of the present invention clearer, the technical solutions of the present invention will be clearly and completely described below in conjunction with the accompanying drawings.

[0043] See Figure 1 The method for monitoring N2O greenhouse gas emissions from surface water bodies based on remote sensing and intelligent algorithms comprises the following steps:

[0044] Step 1: Obtaining water quality parameters (water temperature, pH, conductivity, dissolved oxygen content, ammonium nitrogen, nitrate nitrogen, total phosphorus, total nitrogen, suspended solids, water temperature, dissolved N2O concentration);

[0045] The water quality parameters were obtained by using a portable YSL water quality analyzer to measure surface water temperature, pH, conductivity, and dissolved oxygen content; the instruments for measuring ammonium nitrogen and nitrate nitrogen were German Seal A33 continuous flow analyzers, and the detection method was ultraviolet spectrophotometry; suspended solids were measured using the gravimetric method; total nitrogen was measured using alkaline potassium persulfate digestion-ultraviolet spectrophotometry; total phosphorus was measured using ammonium molybdate spectrophotometry; and the dissolved N2O concentration was measured using the headspace-equilibrium method based on Henry's law and Dalton's law of partial pressures of gases.

[0046] Step 2: Investigate the correlation between dissolved N2O and the above water quality parameters, identify the key water quality parameters (water temperature, pH, dissolved oxygen, inorganic nitrogen) that affect dissolved N2O, and use a deep neural network algorithm to build a dissolved N2O prediction model;

[0047] The measured water quality parameters were analyzed using Pearson correlation coefficient, and the correlation coefficient (R²) was used to determine the water quality parameters. 2 The key water quality parameters affecting dissolved N2O are identified, and a deep neural network algorithm is used to construct a dissolved N2O prediction model based on these parameters. The correlation between dissolved N2O and the water quality parameters in step one is shown in step two. Figure 2 As shown, dissolved N2O was significantly correlated with pH, ​​dissolved oxygen, nitrate nitrogen, nitrite nitrogen, and inorganic nitrogen (p<0.001). Specifically, pH, dissolved oxygen, and water temperature were significantly negatively correlated with dissolved N2O, with coefficients of determination of -0.56, -0.40, and -0.40, respectively; while nitrate nitrogen, nitrite nitrogen, and inorganic nitrogen were significantly positively correlated with dissolved N2O, with coefficients of determination of 0.52, 0.38, and 0.53, respectively. Finally, four variables—pH, dissolved oxygen, water temperature, and inorganic nitrogen—that were significantly correlated with dissolved N2O were selected to construct a dissolved N2O prediction model.

[0048] The deep neural network algorithm constructs a prediction model for dissolved N2O, which includes the following steps:

[0049] S1. The neural network uses the Rectified Linear Function (ReLU) as the activation function, contains 4 intermediate layers, and uses the Adaptive Moment Estimation Optimization Algorithm (Adam) as the optimizer for optimization. The training iterations are 500.

[0050] The linear rectified function is a commonly used activation function in neural networks, and its definition is as follows:

[0051] ReLU(x) = max(0,x)

[0052] x: The input value of the neuron. When x>0, the output of ReLU is x, that is, ReLU(x)=x; when x≤0, the output of ReLU is 0, that is, ReLU(x)=0.

[0053] The main purpose of the intermediate layers (hidden layers) is to extract and process features from the input data, enabling the model to learn and represent complex nonlinear relationships. The neural network contains four intermediate layers, each with several neurons, using the ReLU activation function. Input layer: Input data; Intermediate layer 1: Extracts primary features from the raw input data; Intermediate layer 2: Combines and processes the primary features extracted by the first hidden layer to generate intermediate features; Intermediate layer 3: Further combines and abstracts the intermediate features from the second hidden layer to generate high-level features; Intermediate layer 4: Integrates the high-level features from the third hidden layer to generate a more abstract and comprehensive feature representation; Output layer: Outputs the prediction result.

[0054] The optimization approach of Adam as an optimizer is as follows: Adam uses first-order moment estimation and second-order moment estimation in each parameter update process, so that it uses different adaptive learning rates for different parameters. This can accelerate convergence and handle sparse gradient problems. It uses the first-order moment estimation (mean) and second-order moment estimation (variance) of the gradient to dynamically adjust the learning rate of each parameter, thereby improving optimization efficiency and performance.

[0055] First-order moment estimate: m t =β1m t-1 +(1-β1)gt

[0056] In the above formula, m t β1 is the first moment estimate for the current time t; β1 is a hyperparameter controlling the momentum decay rate, usually set to 0.9, which means the current gradient g t Will be to m t It produces a 10% impact, while the momentum m at the previous moment... t-1 It will retain 90% of the impact; m t-1 The momentum at the previous moment represents the gradient information of the previous step; g t This represents the gradient at the current moment.

[0057] Second moment estimate: v t =β2v t-1 +(1-β2)gt 2

[0058] In the above formula, v t β2 is the second moment estimate at the current time t; β2 is a hyperparameter controlling the decay rate of the second moment, usually set to 0.999; this means the current squared gradient gt 2 Will affect v tIt produces a 0.1% impact, while the second-order moment estimate v from the previous time step... t-1 It will retain 99.9% of the impact. t-1 This represents the second moment estimate from the previous time step, indicating the squared gradient information from the previous step; gt 2 This is the square of the gradient at the current time.

[0059] S2. The collected key water quality parameters (water temperature, pH, dissolved oxygen, inorganic nitrogen) data were randomly divided into 80% and 20% sets. 80% of the sample data was used as the training set and 20% of the sample data was used as the validation set. A deep neural network model was used to construct a dissolved N2O prediction model.

[0060] Among them, the predicted and true values ​​of dissolved N2O in the training set are as follows: Figure 3 As shown, the validation set of dissolved N2O shows that the predicted and actual values ​​are compared in the following ways: Figure 4 As shown, the constructed deep neural network algorithm for predicting dissolved N2O exhibits MSE, MAE, and R-values ​​during the training and validation periods. 2 As shown in Table 1, both the MSE and MAE during the training and validation periods are relatively low, indicating that the model's prediction error is small. The R-value during the training period... 2 The R value during the validation period is 0.80. 2 The value is 0.63, indicating that the model has a good fit to the data and a good interpretability of the validation data.

[0061] Table 1. Accuracy of the Deep Neural Network Model for Dissolved N2O

[0062]

[0063] Step 3: Land surface temperature (including water temperature) inversion algorithm based on Landsat-8, Sentinel-2 and MODIS;

[0064] The specific steps are as follows: Using the MODIS global land surface temperature dataset with a daily resolution of 1 km, the robust least squares (RLS) method is used to establish a relationship between MODIS-LST and Sentinel-2, Landsat-8 red, green, blue, and near-infrared bands, as well as a constant 1, to obtain the confusion matrix coefficients, thereby calculating the land surface temperature. The RLS formula is as follows:

[0065]

[0066] In the formula f i (x) is the observed independent variable. Let be the dependent variable, β(i) be the generated weight value, and ε be the intercept.

[0067] Furthermore, due to differences in the sensors used, the reflection data from Landsat-8 and Sentinel-2 differ. To ensure the uniformity and consistency of subsequent calculations of relevant parameters, the Landsat-8 data bands underwent preprocessing. The band processing used the code provided by GEE, and the formula is as follows:

[0068] 0pticalBands=(SR Band *0.0000275-0.2)*10000

[0069] In the formula SR Band It is the optical band of Landsat-8.

[0070] Step 4: Based on multi-point measured water quality data - ground-based measured hyperspectral and Sentinel-2 datasets, construct pH, dissolved oxygen and inorganic nitrogen inversion algorithms;

[0071] The construction of the pH, dissolved oxygen, and inorganic nitrogen inversion algorithms includes the following steps:

[0072] Acquisition of S1, pH, inorganic nitrogen, dissolved oxygen, and spectral data: Spectral data were obtained using an ASD FieldSpec 3 ground-based spectrometer (USA) to measure the spectral information of the river at the sampling points. The measured hyperspectral reflectance of the water body is shown in the diagram. Figure 5 As shown;

[0073] S2. Construction and accuracy evaluation of pH, dissolved oxygen, and inorganic nitrogen inversion algorithms based on spectral information and neural network algorithms: A dataset is constructed by combining the hyperspectral reflectance of water surfaces at various locations with measured data of pH, inorganic nitrogen, and dissolved oxygen. A neural network model is then built using the coefficient of determination (R²). 2 The performance of the constructed neural network models for pH, dissolved oxygen, and inorganic nitrogen was evaluated using three metrics: mean square error (MAE), mean square error (RMSE), and mean square error (RMSE). By matching the center wavelengths of each band of Sentinel-2 with the hyperspectral wavelengths, the inversion of pH, dissolved oxygen, and inorganic nitrogen in surface water based on Sentinel-2 was achieved.

[0074] The specific steps are as follows: Hyperspectral reflectance is matched with measured data (pH, inorganic nitrogen, dissolved oxygen) so that the hyperspectral reflectance at each location corresponds to its corresponding measured data, constructing a dataset containing both hyperspectral reflectance and measured data. To train or validate models using Sentinel-2 data, the hyperspectral reflectance needs to be matched to Sentinel-2 bands. The center wavelength of each Sentinel-2 band is compared with the hyperspectral wavelength to find the closest hyperspectral wavelength, and the corresponding hyperspectral reflectance is extracted. This ensures that each location of the hyperspectral reflectance can be mapped to the band features of Sentinel-2. The matched Sentinel-2 band reflectance and measured data are used to construct a neural network model using Python for training and validation. The results of the constructed neural network models for pH, dissolved oxygen, and inorganic nitrogen are shown in Table 2. The neural network models for pH, dissolved oxygen, and inorganic nitrogen showed R... 2 A high value close to 1 indicates that the MAE and RMSE values ​​are low during the training and validation periods, and the model has a high fit to the training data.

[0075] Table 2 Evaluation of Neural Network Models for Water pH, Dissolved Oxygen, and Inorganic Nitrogen

[0076]

[0077] Step 5: Based on the above remote sensing inversion algorithms for water temperature, pH, dissolved oxygen, and inorganic nitrogen, the dissolved N2O prediction model is driven. Further combined with the semi-empirical model of N2O gas exchange at the water-air interface, a monitoring method for N2O greenhouse gas emissions from surface water bodies based on remote sensing and intelligent algorithms is finally constructed, realizing the spatial inversion of N2O generation and emission from surface water bodies.

[0078] The formula for the N2O emission flux at the water-air interface in the semi-empirical model of N2O gas exchange at the water-air interface is:

[0079]

[0080] In the above formula: The N2O emission flux at the water-air interface, in μg / m³ 2 ·d; The gas exchange rate is expressed in cm / h; N2O disc The concentration of dissolved N2O in the water body is expressed in μg / L. The pH, dissolved oxygen, inorganic nitrogen, and water temperature involved are obtained from remote sensing inversion. eqc The theoretical equilibrium concentration of dissolved N2O in water is given in μg / L.

[0081] (1) The theoretical equilibrium concentration of dissolved N2O in water is calculated based on water temperature, atmospheric N2O concentration, and the Weiss equation, as shown in the following formula:

[0082] N2O eqc =M×F×N2O air

[0083] In F=-165.8806+222.8743×(100 / T)+92.0792×In(T / 100)-1.48425×(T / 100)2

[0084] In the above formula: M is the molecular mass of N2O; T is the thermodynamic temperature, obtained based on remote sensing inversion; N2O air The values ​​represent atmospheric N2O concentrations (ppm) at different times and locations.

[0085] (2) Gas exchange rate The calculation formula is as follows:

[0086]

[0087] S c =2141.2 - 152.56 × T + 5.8963 × T 2 -0.12411×T 3 +0.0010655×T 4

[0088] In the above formula, Sc is the ratio of the dynamic viscosity of water to the diffusion rate of N2O gas molecules, and n is the Schmidt number, which is physically affected by water surface fluctuations. When the wind speed at a height of 10m is greater than 3.6m / s, n = 1 / 2; when the wind speed is less than 3.6m / s, n = 2 / 3; K 600 is the gas exchange coefficient; T is the water temperature, °C, obtained from remote sensing inversion.

[0089] (3) Gas exchange coefficient K 600 The calculation formula is:

[0090] K 600 =0.251×U 10 2

[0091] In the above formula: U 10 The wind speed at an altitude of 10m is given in m / s.

[0092] In step five, the RMSE of the predicted and measured values ​​of dissolved N2O was 0.192, and the MAE was 0.155. Figure 6a As shown, the mean square error and absolute error values ​​are low, indicating high prediction accuracy of the model; the RMSE of the measured N2O emission flux fitting the predicted N2O emission flux is 5.308, and the MAE is 3.751. Figure 6bAs shown, the measured and predicted values ​​of N2O emissions from surface water bodies are largely consistent, and the model generally has a high degree of fit, except for some abrupt change points.

[0093] In the practical application of this invention, the following specific steps are included:

[0094] 1. Obtaining water quality parameters (water temperature, pH, conductivity, dissolved oxygen content, ammonium nitrogen, nitrate nitrogen, total phosphorus, total nitrogen, suspended solids, water temperature, dissolved N2O concentration);

[0095] 2. Analysis of Influencing Factors and Construction of Predictive Model for Dissolved N2O in Surface Water. The correlation between dissolved N2O and various water quality parameters was analyzed. Water quality parameters (water temperature, pH, dissolved oxygen, and inorganic nitrogen) that were highly significantly correlated with dissolved N2O were selected. A deep neural network algorithm was used to construct a predictive model for dissolved N2O.

[0096] 3. Construction and Evaluation of Water Temperature Retrieval Algorithms Based on Multi-Source Remote Sensing. Land surface temperature (including water temperature) retrieval algorithms based on Landsat-8, Sentinel-2, and MODIS are presented. The time series diagram of water temperature retrieval based on Landsat-8, Sentinel-2, and MODIS is shown below. Figure 7 As shown, this illustrates the trend of temperature change over time;

[0097] 4. Construction and Evaluation of Water Temperature, pH, Dissolved Oxygen, and Inorganic Nitrogen Retrieval Algorithms Based on Remote Sensing and Hyperspectral Imaging. Spectral information at sampling points was measured using the ASD FieldSpec3 ground object spectrometer. Hyperspectral data of the water surface at each point was combined with measured data of pH, dissolved oxygen, and inorganic nitrogen to form a data matrix, which was then used to construct a neural network model. Furthermore, by matching the center wavelength of each band of the Sentinel-2 spectrometer with the hyperspectral wavelength, water temperature, pH, dissolved oxygen, and inorganic nitrogen retrieval based on Sentinel-2 was achieved. Comparison of measured and predicted values ​​of pH, dissolved oxygen, and inorganic nitrogen based on spectral information was performed. Figure 8a , Figure 8b , Figure 8c As shown, the trends of the measured values ​​and the predicted values ​​are quite consistent, and the model has a high degree of fit overall.

[0098] 5. Construction and Evaluation of N2O Emission Model for Surface Water Based on Remote Sensing. Based on the remote sensing inversion algorithms for water temperature, pH, dissolved oxygen, and inorganic nitrogen from steps two, three, and four, a dissolved N2O prediction model is driven. Further combined with a water-air interface N2O gas exchange model, a monitoring method for N2O greenhouse gas emissions from surface water based on remote sensing and intelligent algorithms is finally constructed, achieving spatial inversion of N2O generation and emissions from surface water. The spatial distribution of dissolved N2O concentration in surface water based on remote sensing is shown below. Figure 9a , Figure 9b , Figure 9c As shown, the dissolved N2O concentration in the upstream area is relatively high, ranging from 0.58 to 1.41 μg / L, while in the midstream area it ranges from 0.31 to 0.69 μg / L. As the river flows downstream, the channel widens and flattens, diluting the inorganic nitrogen concentration and reducing the dissolved N2O content to between 0.03 and 0.33 μg / L. The spatial distribution of N2O emission flux from surface water based on remote sensing is shown below. Figure 10a , Figure 10b , Figure 10c As shown, the upstream water-air interface has the highest N2O emission flux, ranging from 70.93 to 141.6 μg / (m²). 2 Between ·d), the N2O emission flux at the water interface in the middle reaches of the river ranges from 29.80 to 100.81 μg / (m³). 2 •d), the downstream water-air interface N2O emission flux is -15.89 to 10.44 μg / (m³). 2 ·d) The N2O emission flux in the downstream region is mostly negative, indicating that the water body absorbs N2O from the atmosphere. The upstream and midstream regions are the source of atmospheric N2O, while the downstream region is the sink of atmospheric N2O.

Claims

1. A method for monitoring N2O greenhouse gas emissions from surface water bodies based on remote sensing and intelligent algorithms, characterized in that: Includes the following steps: Step 1: Obtaining water quality parameters; Step 2: Analyze the correlation between dissolved N2O and the above water quality parameters, identify the key water quality parameters affecting dissolved N2O, and use a deep neural network algorithm to construct a dissolved N2O prediction model; Step 3: Land surface temperature inversion algorithm based on Landsat-8, Sentinel-2, and MODIS; Step 4: Based on multi-point measured water quality data - ground-based measured hyperspectral data and Sentinel-2 dataset, construct pH, dissolved oxygen and inorganic nitrogen inversion algorithms; Step 5: Based on the remote sensing inversion algorithm for water temperature, pH, dissolved oxygen, and inorganic nitrogen, drive the dissolved N2O prediction model, combine it with the water-air interface N2O gas exchange model, and construct a monitoring method for N2O greenhouse gas emissions from surface water bodies based on remote sensing and intelligent algorithms to realize the spatial inversion of N2O generation and emission from surface water bodies. Specifically, in step three: using the MODIS global land surface temperature dataset with a daily resolution of 1 km, the robust least squares (RLS) method is used to establish a relationship between MODIS-LST and the Sentinel-2, Landsat-8 red, green, blue, and near-infrared bands, as well as a constant 1, to obtain the confusion matrix coefficients, thereby calculating the land surface temperature. The RLS formula is as follows: ； In the formula, For the observed independent variable, As the dependent variable, For the generated weight values, The intercept; Furthermore, due to differences in the sensors used, the reflection data from Landsat-8 and Sentinel-2 differ. To ensure the uniformity and consistency of subsequent calculations of relevant parameters, the Landsat-8 data bands underwent preprocessing. The band processing used the code provided by GEE, and the formula is as follows: ； In the formula For the optical band of Landsat-8; Specifically, in step four, the construction of the pH, dissolved oxygen, and inorganic nitrogen inversion algorithms includes the following steps: S4.1, pH, inorganic nitrogen, dissolved oxygen and spectral data acquisition. Spectral data were obtained using an ASD FieldSpec 3 ground object spectrometer to measure the spectral information of the river at the sampling point. S4.2 Construction and accuracy evaluation of pH, dissolved oxygen, and inorganic nitrogen inversion algorithms based on spectral information and neural network algorithms: A dataset was constructed by combining the hyperspectral reflectance of the water surface at each location with measured data of pH, inorganic nitrogen, and dissolved oxygen. A neural network model was then built using the coefficient of determination R0. 2 The performance of the constructed neural network models for pH, dissolved oxygen, and inorganic nitrogen was evaluated using three indicators: absolute error (MAE), mean square error (RMSE). Based on the matching of the center wavelength of each band of Sentinel-2 with the hyperspectral wavelength, the inversion of pH, dissolved oxygen, and inorganic nitrogen of surface water based on Sentinel-2 was achieved.

2. The method for monitoring N2O greenhouse gas emissions from surface water bodies based on remote sensing and intelligent algorithms according to claim 1, characterized in that: Water quality parameters include: water temperature, pH, conductivity, dissolved oxygen content, ammonium nitrogen, nitrate nitrogen, total phosphorus, total nitrogen, suspended solids, water temperature, and dissolved N2O concentration.

3. A method for monitoring N2O greenhouse gas emissions from surface water bodies based on remote sensing and intelligent algorithms, as described in claim 1 or 2, characterized in that: The water quality parameters were obtained by using a portable YSL water quality analyzer to measure surface water temperature, pH, conductivity, and dissolved oxygen content; the instruments for measuring ammonium nitrogen and nitrate nitrogen were German Seal A33 continuous flow analyzers, and the detection method was ultraviolet spectrophotometry; suspended solids were measured using the gravimetric method; total nitrogen was measured using alkaline potassium persulfate digestion-ultraviolet spectrophotometry; and total phosphorus was measured using ammonium molybdate spectrophotometry. The concentration of dissolved N2O was determined using the headspace-equilibrium method based on Henry's law and Dalton's law of partial pressures of gases.

4. The method for monitoring N2O greenhouse gas emissions from surface water bodies based on remote sensing and intelligent algorithms according to claim 1, characterized in that: In step two, specifically: S2.1 In the neural network, the linear rectified function ReLU is selected as the activation function, which contains 4 intermediate layers. The adaptive moment estimation optimization algorithm Adam is used as the optimizer for optimization, and the training times are 500. S2.

2. Collect water temperature, pH, dissolved oxygen, and inorganic nitrogen. Randomly divide the data into 80% and 20% portions. Use 80% of the sample data as the training set and 20% of the sample data as the validation set. Use a deep neural network model to construct a fitting model for dissolved N2O. S2.3, Use mean squared error (MSE), mean absolute error (MAE), and coefficient of determination (R²). 2 Three metrics are used to evaluate the model's accuracy during both the training and validation periods.

5. A method for monitoring N2O greenhouse gas emissions from surface water bodies based on remote sensing and intelligent algorithms according to claim 4, characterized in that: In step S2.1, specifically: the linear rectified function ReLU is defined as follows: ReLU(x) = max(0,x) Where x is the input value of the neuron, and when x > 0, the output of ReLU is x That is, ReLU(x) = x; when x ≤ 0, the output of ReLU is 0, that is, ReLU(x) = 0.

6. The method for monitoring N2O greenhouse gas emissions from surface water bodies based on remote sensing and intelligent algorithms according to claim 4, characterized in that: The learning rate of each parameter is dynamically adjusted by using the first and second moment estimates of the gradient, thereby improving optimization efficiency and performance. First-order moment estimation: ; In the above formula, For the current moment The first moment estimate; : This is a hyperparameter that controls the rate of momentum decay, set to 0.9, representing the current gradient. Will to It produces a 10% impact, while the momentum of the previous moment... It will retain 90% of the impact; The momentum is the value from the previous moment, and the gradient information is represented in the previous step. The gradient at the current time step; Second-order moment estimation: ; In the above formula, For the current moment The second moment estimate; The hyperparameter for controlling the decay rate of the second moment is set to 0.999; the current gradient squared... Will to It produces a 0.1% impact, while the second moment estimate of the previous time step... It will retain 99.9% of the impact; This is the second moment estimate from the previous time step, representing the squared gradient information from the previous step; This is the square of the gradient at the current time.

7. The method for monitoring N2O greenhouse gas emissions from surface water bodies based on remote sensing and intelligent algorithms according to claim 1, characterized in that: The hyperspectral reflectance is matched with measured data to ensure that the hyperspectral reflectance at each location corresponds to its corresponding measured data, thus constructing a dataset containing both hyperspectral reflectance and measured data. To train or validate models using Sentinel-2 data, the hyperspectral reflectance needs to be matched to Sentinel-2 bands. The center wavelength of each Sentinel-2 band is compared with the hyperspectral wavelength to find the closest hyperspectral wavelength, and the corresponding hyperspectral reflectance is extracted. This ensures that each location of the hyperspectral reflectance can be mapped to the band features of Sentinel-2. The matched Sentinel-2 band reflectance and measured data are then used to construct a neural network model using Python for training and validation.

8. The method for monitoring N2O greenhouse gas emissions from surface water bodies based on remote sensing and intelligent algorithms according to claim 1, characterized in that: In step five, the N2O emission flux formula in the water-air interface N2O gas exchange model is: ； In the above formula: N2O emission flux at the water-air interface, in μg / m³ 2 •d; Gas exchange rate, unit: cm / h; N2O disc The concentration of dissolved N2O in water is expressed in μg / L. The pH, dissolved oxygen, inorganic nitrogen, and water temperature were obtained from remote sensing inversion. eqc The theoretical equilibrium concentration of dissolved N2O in water, in μg / L; The theoretical equilibrium concentration of dissolved N2O in water is calculated based on water temperature, atmospheric N2O concentration, and the Weiss equation, as shown in the following formula: ； In F=-165.8806+222.8743×(100 / T)+92.0792×In(T / 100)-1.48425×(T / 100) 2 In the above formula: M is the molecular mass of N2O; T is the thermodynamic temperature, obtained based on remote sensing inversion; N2O air The values ​​represent atmospheric N2O concentrations at different times and locations, in ppm. Gas exchange rate The calculation formula is as follows: =K 600 ×(S c / 600) -n ； In the above formula, Sc is the ratio of the dynamic viscosity of water to the diffusion rate of N2O gas molecules, n is the Schmidt number, which is physically affected by water surface fluctuations. When the wind speed at a height of 10m is greater than 3.6m / s, n = 1 / 2; when the wind speed is less than 3.6m / s, n = 2 / 3; K 600 The gas exchange coefficient; Gas exchange coefficient K 600 The calculation formula is: ； In the above formula: The wind speed at an altitude of 10m is given in m / s.

Images