Remote sensing retrieval method of dissolved inorganic nitrogen and silicate in estuary based on salinity synergy

Abstract

This invention belongs to the field of environmental monitoring technology and discloses a remote sensing inversion method for dissolved inorganic nitrogen and silicate in estuaries based on salinity synergy. The method includes the following steps: S1. During a field survey in the land-sea interaction zone of the estuary, remote sensing reflectance, salinity, and nutrient data are collected simultaneously. The measured hyperspectral reflectance is simulated as the equivalent reflectance of the satellite band using the spectral response function of the target satellite, which is used to construct the training and validation datasets for the model. S2. A nonlinear inversion model of remote sensing reflectance and salinity is established, and a nutrient mixture model of salinity and nutrients is constructed. The training dataset is used for model training. S3. The satellite remote sensing reflectance data of the estuarine area to be predicted is input into the trained nonlinear inversion model and combined with the nutrient mixture model to generate the spatial distribution of DIN concentration and DSi concentration in the estuarine area. This method achieves high-precision remote sensing inversion prediction of DIN concentration and dissolved silicate DSi concentration in the estuarine area.

Description

Technical Field

[0001] This invention belongs to the field of environmental monitoring, specifically relating to a remote sensing inversion method for dissolved inorganic nitrogen and silicates in estuaries based on salinity synergy. Background Technology

[0002] In the field of estuarine and nearshore marine ecological environment prediction and management, nutrients are key factors affecting the structure and function of ecosystems. Among them, dissolved inorganic nitrogen (DIN) and dissolved silicate (DSi) play an important regulatory role in phytoplankton growth and reproduction and community structure succession. Their spatiotemporal distribution characteristics are directly related to the eutrophication level and ecological environment health of estuaries.

[0003] Currently, domestic and international research on estuarine nutrient monitoring has significant shortcomings: on the one hand, traditional empirical models have poor adaptability to complex estuarine environments and lack model universality; on the other hand, existing machine learning methods are prone to overfitting in long-term applications and lack handling of core issues such as scale matching between satellite and ground data and differences in spectral response, resulting in insufficient model generalization ability. Especially in strong tidal estuaries like the Jiulong River, where hydrodynamic conditions are complex, the spatiotemporal distribution of nutrients is jointly controlled by physical mixing and biogeochemical processes, making it difficult for traditional methods to accurately reflect their dynamic changes.

[0004] Furthermore, many existing methods fail to fully utilize the intrinsic relationship between salinity and nutrients in estuaries, neglecting the role of salinity as a key indicator of terrestrial input and physical mixing processes. This results in significant limitations in our understanding of the land-sea interaction mechanisms, transport pathways, and transformation processes of estuarine nutrients, hindering the accuracy and timeliness of estuarine ecological environment management. Therefore, there is an urgent need to develop a new remote sensing inversion technology for estuarine nutrients that balances accuracy and robustness to meet the pressing needs of estuarine ecosystem health assessment and eutrophication control. Summary of the Invention

[0005] To address the aforementioned issues, this invention proposes a remote sensing inversion method for dissolved inorganic nitrogen and silicates in estuaries based on salinity synergy. This method tackles the technical challenge of quantifying nutrients in estuarine areas due to their lack of direct optical activity. It constructs a complete technical chain: "spectral response function bridging—construction and screening of multi-band spectral indices—two-step salinity inversion." Its core innovations are: 1) introducing spectral response functions to accurately match satellite and ground spectral data, eliminating spectral scale differences between different sensors; 2) systematically constructing and screening optimal multi-band spectral index combinations to enhance salinity sensitivity and suppress environmental noise; 3) comparing and optimizing various models (multivariate linear regression, ridge regression, random forest regression, XGBoost algorithm, etc.) and performing hyperparameter optimization to significantly improve inversion accuracy and generalization ability; 4) constructing a progressive inversion framework of "remotely sensed reflectance—salinity—nutrients," deeply integrating a physical model based on conservative mixture theory with a data-driven machine learning model. This invention not only achieves high-precision remote sensing inversion prediction of dissolved inorganic nitrogen (DIN) and dissolved silicate (DSi) concentrations in estuaries, but also provides important technical support for estuarine biogeochemical cycle research and integrated watershed management, and has good prospects for widespread application.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: A remote sensing inversion method for dissolved inorganic nitrogen and silicates in estuaries based on salinity synergy includes the following steps: S1. During the field survey of the estuarine land-sea interaction zone, remote sensing reflectance, salinity and nutrient data were collected simultaneously. The measured hyperspectral reflectance was simulated as the satellite band equivalent reflectance using the spectral response function of the target satellite, which was used to build the training dataset and validation dataset for the model. S2. Establish a nonlinear inversion model of remote sensing reflectance and salinity, then construct a nutrient mixture model of salinity and nutrients, and train the model using the training dataset. The specific process of step S2 is as follows: S21. Based on multiple spectral bands of equivalent reflectance of satellite bands, the system constructs 2-4 band combinations of difference index, ratio index and normalized difference index, and generates multiple spectral variables; each spectral variable is subjected to Pearson correlation analysis with synchronously measured salinity, and the top 5 spectral combinations with the highest correlation coefficient at a 95% confidence level are selected as the optimal input features of the model. S22. Based on the selected optimal spectral combination, multiple analysis models are constructed. The analysis model is one of the following: multiple linear regression model, ridge regression model, random forest regression model, and XGBoost algorithm. A nonlinear inversion model of remote sensing reflectance and salinity is established using the random forest algorithm. Then, the hyperparameters are optimized through random search and 5-fold cross-validation, and the accuracy of the nonlinear inversion model is evaluated to obtain the optimal model parameter combination. The hyperparameters include the number of trees, maximum depth, minimum number of sample splits, and minimum number of samples in the leaf nodes. The calculation formula for the nonlinear inversion model is: Salinity = RF( (λ1), (λ2),..., (λ n ), where Salinity is the inverted salinity value; RF(·) is the random forest regression model; (λ1), (λ2),..., (λ n ( ) represent the equivalent reflectivity of satellite bands for different combinations of bands; The number of samples; S23. Based on the theory of conservative nutrient mixing in estuaries, a nutrient mixing model is constructed, and the calculation formula is as follows: , Where R is the proportion of freshwater at the selected salinity; M is the ratio of freshwater to seawater at the selected salinity; V r V is the volume of fresh water; m V is the volume of seawater; S is the volume of the mixed water; r The salinity of freshwater; S m Let S be the salinity of seawater; S be the salinity of the mixed water; under a water quality model where mixing is the primary process, the formula for calculating the nutrient concentration in the mixed water is: Where C is the nutrient concentration in the mixed water; C r C represents the nutrient concentration of freshwater at the river's end. m This refers to the nutrient concentration of seawater. S3. Input the satellite remote sensing reflectance data of the estuary area to be predicted into the trained nonlinear inversion model, and combine it with the nutrient mixing model to generate the spatial distribution of DIN concentration and DSi concentration in the estuary area.

[0007] Preferably, the specific process of step S1 is as follows: S11. Measure the spectral data above the water surface using a ground-based spectrometer, and calculate the remote sensing reflectance based on the spectral data. The calculation formula is as follows: ,in, For wavelength Measured hyperspectral reflectance; Wavelength; For wavelength The radiance of the water surface; For wavelength The brightness of the sky's light radiance; For wavelength Standard gray board radiance; For wavelength The reflectivity of the gray board; The Fresnel reflection coefficient; S12. Obtain the spectral response function and extra-atmospheric solar irradiance of the target satellite, and convert the measured hyperspectral reflectance into the equivalent reflectance in the satellite band. The calculation formula is as follows: ,in, For wavelength Satellite band equivalent reflectivity; For wavelength The spectral response function of the satellite; For wavelength Solar irradiance outside the atmosphere; S13. Surface water salinity was measured in situ using a water quality analyzer, followed by water sample collection and laboratory analysis of DIN and DSi concentrations; DIN concentration was determined to be NO3-. - -N, NO2 - -N and NH4 + The sum of the concentrations of -N; all water samples were analyzed within 24 hours of collection.

[0008] Preferably, in step S11, during measurement, the angle between the observation plane of the ground object spectrometer and the solar incident plane is kept between 90° and 135°, the probe is perpendicular to the water surface and at a 40° angle to the normal direction; dark current correction is performed before each measurement, 10 spectral curves are continuously collected, and the average value is calculated after outlier removal as the spectral data of the station.

[0009] Preferably, in step S22, the accuracy of the nonlinear inversion model is evaluated using the coefficient of determination, root mean square error, and mean absolute error, to ensure that the accuracy of the nonlinear inversion model meets the application requirements. The calculation formula is as follows: , , ,in, The coefficient of determination; Mean absolute error; This is the root mean square error; Number the sample; This is the actual value; This is a predicted value; This is the average of the actual values.

[0010] Preferably, the specific process of step S3 is as follows: S31. Input the satellite remote sensing reflectance data of the estuary area to be predicted into the trained nonlinear inversion model to obtain the spatial distribution of salinity in the estuary area. S32. Substitute the salinity values ​​obtained from step S31 into the salinity-DIN mixing model and the salinity-DSi mixing model respectively, and calculate the spatial distribution of DIN concentration and the spatial distribution of DSi concentration respectively. S33. Using several years of satellite remote sensing reflectance data, generate a long-term spatiotemporal distribution dataset of DIN and DSi in the estuary area, with a time resolution of monthly or quarterly. S34. Evaluate the inversion accuracy using the validation dataset, and quantify the model performance using the coefficient of determination, root mean square error, and mean absolute error. S35. Identify and quantify the main sources of error, including atmospheric correction error, sensor noise, and model extrapolation uncertainty, and evaluate the impact of each factor on the inversion results through sensitivity analysis.

[0011] After adopting the above technical solution, the present invention has the following beneficial effects: The present invention breaks through the bottleneck of inversion without optically active parameters: it innovatively constructs an indirect inversion link of "remote sensing reflectance—salinity—nutrients," using salinity as a physical bridge to achieve high-precision remote sensing estimation of DIN and DSi. First, the method explicitly introduces a spectral response function, accurately simulating the equivalent reflectance of a specific satellite from ground hyperspectral data, fundamentally solving the problem of inconsistency between multi-source and multi-scale spectral data; by systematically constructing thousands of multi-band spectral index combinations and combining correlation screening, it maximizes the mining of spectral information, enhances salinity-sensitive signals, and suppresses environmental noise. Comparing various empirical regression and machine learning models, it selects ensemble learning algorithms such as random forest or XGBoost, and performs systematic hyperparameter optimization through random search and cross-validation, significantly improving the model's nonlinear fitting ability and generalization performance; it organically couples the physical model based on conservative mixture theory with the data-driven machine learning model (spectral-salinity nonlinear relationship), giving the model both physical interpretability and high prediction accuracy. It achieves large-scale, long-term, high spatiotemporal resolution remote sensing prediction of DIN and DSi concentrations in estuaries. This method not only significantly improves the efficiency and accuracy of estuarine nutrient monitoring, but also provides reliable technical support for understanding estuarine biogeochemical cycles, assessing the impact of human activities, and developing precise management strategies, thus possessing significant scientific value and application prospects. Attached Figure Description

[0012] Figure 1 This is a flowchart of the present invention. Detailed Implementation

[0013] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0014] like Figure 1 As shown, the remote sensing inversion method for dissolved inorganic nitrogen and silicates in estuaries based on salinity synergy includes the following steps: S1. During the field survey of the estuarine land-sea interaction zone, remote sensing reflectance, salinity and nutrient data were collected simultaneously. The measured hyperspectral reflectance was simulated as the satellite band equivalent reflectance using the spectral response function of the target satellite, which was used to build the training dataset and validation dataset for the model. The specific process of step S1 is as follows: S11. Measure the spectral data above the water surface using a ground-based spectrometer, and calculate the remote sensing reflectance based on the spectral data. The calculation formula is as follows: ,in, For wavelength Measured hyperspectral reflectance; Wavelength; For wavelength The radiance of the water surface; For wavelength The brightness of the sky's light radiance; For wavelength Standard gray board radiance; For wavelength The reflectivity of the gray board; The Fresnel reflection coefficient; S12. Obtain the spectral response function and extra-atmospheric solar irradiance of the target satellite, and convert the measured hyperspectral reflectance into the equivalent reflectance in the satellite band. The calculation formula is as follows: ,in, For wavelength Satellite band equivalent reflectivity; For wavelength The spectral response function of the satellite; For wavelength Solar irradiance outside the atmosphere; S13. Surface water salinity was measured in situ using a water quality analyzer, followed by water sample collection and laboratory analysis of DIN and DSi concentrations; DIN concentration was determined to be NO3-. - -N, NO2 - -N and NH4 + The sum of the concentrations of -N; all water samples were analyzed within 24 hours of collection; S2. Establish a nonlinear inversion model of remote sensing reflectance and salinity, then construct a nutrient mixture model of salinity and nutrients, and train the model using the training dataset. The specific process of step S2 is as follows: S21. Based on multiple spectral bands of equivalent reflectance of satellite bands, the system constructs 2-4 band combinations of difference index, ratio index and normalized difference index, and generates multiple spectral variables; each spectral variable is subjected to Pearson correlation analysis with synchronously measured salinity, and the top 5 spectral combinations with the highest correlation coefficient at a 95% confidence level are selected as the optimal input features of the model. S22. Based on the selected optimal spectral combination, multiple analysis models are constructed. The analysis model is one of the following: multiple linear regression model, ridge regression model, random forest regression model, and XGBoost algorithm. A nonlinear inversion model of remote sensing reflectance and salinity is established using the random forest algorithm. Then, the hyperparameters are optimized through random search and 5-fold cross-validation, and the accuracy of the nonlinear inversion model is evaluated to obtain the optimal model parameter combination. The hyperparameters include the number of trees, maximum depth, minimum number of sample splits, and minimum number of samples in the leaf nodes. The calculation formula for the nonlinear inversion model is: Salinity = RF( (λ1), (λ2),..., (λ n ), where Salinity is the inverted salinity value; RF(·) is the random forest regression model; (λ1), (λ2),..., (λ n ( ) represent the equivalent reflectivity of satellite bands for different combinations of bands; The number of samples; In step S22, the accuracy of the nonlinear inversion model is evaluated using the coefficient of determination, root mean square error, and mean absolute error. These are used to ensure that the accuracy of the nonlinear inversion model meets the application requirements. The calculation formula is as follows: , , ,in, The coefficient of determination; Mean absolute error; This is the root mean square error; Number the sample; This is the actual value; This is a predicted value; The average of the actual values;

[0015] S23. Based on the theory of conservative nutrient mixing in estuaries, a nutrient mixing model is constructed, and the calculation formula is as follows: , Where R is the proportion of freshwater at the selected salinity; M is the ratio of freshwater to seawater at the selected salinity; V r V is the volume of fresh water; m V is the volume of seawater; S is the volume of the mixed water; r The salinity of freshwater; S m Let S be the salinity of seawater; S be the salinity of the mixed water; under a water quality model where mixing is the primary process, the formula for calculating the nutrient concentration in the mixed water is: Where C is the nutrient concentration in the mixed water; C r C represents the nutrient concentration of freshwater at the river's end. m This refers to the nutrient concentration of seawater. The parameter optimization results of the salinity inversion model are shown in Table 1.

[0016] Table 1: Parameter optimization results of the salinity inversion model

[0018] S3. Input the satellite remote sensing reflectance data of the estuary area to be predicted into the trained nonlinear inversion model, and combine it with the nutrient mixing model to generate the spatial distribution of DIN concentration and DSi concentration in the estuary area. The specific process of step S3 is as follows: S31. Input the satellite remote sensing reflectance data of the estuary area to be predicted into the trained nonlinear inversion model to obtain the spatial distribution of salinity in the estuary area. S32. Substitute the salinity values ​​obtained from step S31 into the salinity-DIN mixing model and the salinity-DSi mixing model respectively, and calculate the spatial distribution of DIN concentration and the spatial distribution of DSi concentration respectively. S33. Using several years of satellite remote sensing reflectance data, generate a long-term spatiotemporal distribution dataset of DIN and DSi in the estuary area, with a time resolution of monthly or quarterly. S34. Evaluate the inversion accuracy using the validation dataset, and quantify the model performance using the coefficient of determination, root mean square error, and mean absolute error. The accuracy verification results of the salinity inversion model, the salinity-DIN mixture model, and the salinity-DSi mixture model are shown in Table 2.

[0019] Table 2: Accuracy verification results of the salinity inversion model, salinity-DIN mixture model, and salinity-DSi mixture model Model type R² RMSE MAE Random Forest Regression 0.812 4.47 3.215 Multiple linear regression 0.5 7.27 5.63 Ridge Return 0.63 6.26 4.91 XGboost 0.74 5.2 3.91 Salinity-DIN Mixing Model 0.764 0.485 0.372 Salinity-DSi Mixing Model 0.780 39.725 29.288 S35. Identify and quantify the main sources of error, including atmospheric correction error, sensor noise, and model extrapolation uncertainty, and evaluate the impact of each factor on the inversion results through sensitivity analysis.

[0020] The above description is merely a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A remote sensing inversion method for dissolved inorganic nitrogen and silicates in estuaries based on salinity synergy, characterized in that, Includes the following steps: S1. During the field survey of the estuarine land-sea interaction zone, remote sensing reflectance, salinity and nutrient data were collected simultaneously. The measured hyperspectral reflectance was simulated as the satellite band equivalent reflectance using the spectral response function of the target satellite, which was used to build the training dataset and validation dataset for the model. S2. Establish a nonlinear inversion model of remote sensing reflectance and salinity, then construct a nutrient mixture model of salinity and nutrients, and train the model using the training dataset. The specific process of step S2 is as follows: S21. Based on multiple spectral bands of equivalent reflectance of satellite bands, the system constructs 2-4 band combinations of difference index, ratio index and normalized difference index, and generates multiple spectral variables; each spectral variable is subjected to Pearson correlation analysis with synchronously measured salinity, and the top 5 spectral combinations with the highest correlation coefficient at a 95% confidence level are selected as the optimal input features of the model. S22. Based on the selected optimal spectral combination, multiple analysis models are constructed. The analysis model is one of the following: multiple linear regression model, ridge regression model, random forest regression model, and XGBoost algorithm. A nonlinear inversion model of remote sensing reflectance and salinity is established using the random forest algorithm. Then, the hyperparameters are optimized through random search and 5-fold cross-validation, and the accuracy of the nonlinear inversion model is evaluated to obtain the optimal model parameter combination. The hyperparameters include the number of trees, maximum depth, minimum number of sample splits, and minimum number of samples in the leaf nodes. The calculation formula for the nonlinear inversion model is: Salinity = RF( (λ1), (λ2),..., (λ n ), where Salinity is the inverted salinity value; RF(·) is the random forest regression model; (λ1), (λ2),..., (λ n ( ) represent the equivalent reflectivity of satellite bands for different combinations of bands; The number of samples; S23. Based on the theory of conservative nutrient mixing in estuaries, a nutrient mixing model is constructed, and the calculation formula is as follows: , Where R is the proportion of freshwater at the selected salinity; M is the ratio of freshwater to seawater at the selected salinity; V r V is the volume of fresh water; m V is the volume of seawater; S is the volume of the mixed water; r The salinity of freshwater; S m Let S be the salinity of seawater; S be the salinity of the mixed water; under a water quality model where mixing is the primary process, the formula for calculating the nutrient concentration in the mixed water is: Where C is the nutrient concentration in the mixed water; C r C represents the nutrient concentration of freshwater at the river's end. m This refers to the nutrient concentration of seawater. S3. Input the satellite remote sensing reflectance data of the estuary area to be predicted into the trained nonlinear inversion model, and combine it with the nutrient mixing model to generate the spatial distribution of DIN concentration and DSi concentration in the estuary area.

2. The method for remote sensing inversion of dissolved inorganic nitrogen and silicates in estuaries based on salinity synergy as described in claim 1, characterized in that, The specific process of step S1 is as follows: S11. Measure the spectral data above the water surface using a ground-based spectrometer, and calculate the remote sensing reflectance based on the spectral data. The calculation formula is as follows: ,in, For wavelength Measured hyperspectral reflectance; Wavelength; For wavelength The radiance of the water surface; For wavelength The brightness of the sky's light radiance; For wavelength Standard gray board radiance; For wavelength The reflectivity of the gray board; The Fresnel reflection coefficient; S12. Obtain the spectral response function and extra-atmospheric solar irradiance of the target satellite, and convert the measured hyperspectral reflectance into the equivalent reflectance in the satellite band. The calculation formula is as follows: ,in, For wavelength Satellite band equivalent reflectivity; For wavelength The spectral response function of the satellite; For wavelength Solar irradiance outside the atmosphere; S13. Surface water salinity was measured in situ using a water quality analyzer, followed by water sample collection and laboratory analysis of DIN and DSi concentrations; DIN concentration was determined to be NO3-. - -N, NO2 - -N and NH4 + The sum of the concentrations of -N; all water samples were analyzed within 24 hours of collection.

3. The method for remote sensing inversion of dissolved inorganic nitrogen and silicates in estuaries based on salinity synergy as described in claim 2, characterized in that: In step S11, during the measurement, the angle between the observation plane of the ground object spectrometer and the solar incident plane is kept between 90° and 135°, and the probe is perpendicular to the water surface and at a 40° angle to the normal direction. Dark current correction is performed before each measurement, and 10 spectral curves are continuously collected. After outlier removal, the average value is calculated as the spectral data of the station.

4. The method for remote sensing inversion of dissolved inorganic nitrogen and silicates in estuaries based on salinity synergy as described in claim 1, characterized in that: In step S22, the accuracy of the nonlinear inversion model is evaluated using the coefficient of determination, root mean square error, and mean absolute error. These are used to ensure that the accuracy of the nonlinear inversion model meets the application requirements. The calculation formula is as follows: , , ,in, The coefficient of determination; Mean absolute error; This is the root mean square error; Number the sample; This is the actual value; This is a predicted value; This is the average of the actual values.

5. The method for remote sensing inversion of dissolved inorganic nitrogen and silicates in estuaries based on salinity synergy as described in claim 1, characterized in that, The specific process of step S3 is as follows: S31. Input the satellite remote sensing reflectance data of the estuary area to be predicted into the trained nonlinear inversion model to obtain the spatial distribution of salinity in the estuary area. S32. Substitute the salinity values ​​obtained from step S31 into the salinity-DIN mixing model and the salinity-DSi mixing model respectively, and calculate the spatial distribution of DIN concentration and the spatial distribution of DSi concentration respectively. S33. Using several years of satellite remote sensing reflectance data, generate a long-term spatiotemporal distribution dataset of DIN and DSi in the estuary area, with a time resolution of monthly or quarterly. S34. Evaluate the inversion accuracy using the validation dataset, and quantify the model performance using the coefficient of determination, root mean square error, and mean absolute error. S35. Identify and quantify the main sources of error, including atmospheric correction error, sensor noise, and model extrapolation uncertainty, and evaluate the impact of each factor on the inversion results through sensitivity analysis.

Images