River total nitrogen concentration remote sensing estimation method based on multi-source data fusion

Abstract

The invention relates to the technical field of water ecological environment remote sensing monitoring, in particular to a river total nitrogen concentration remote sensing estimation method based on multi-source data fusion, and the method comprises the following steps: S1, obtaining and processing total nitrogen measured data, and unifying time and a coordinate reference system; s2, processing the Landsat 8 / 9 image, and extracting water body reflectivity and remote sensing indexes; s3, ERA5-Land meteorological data are acquired, and derivation indexes are calculated; s4, DEM elevation data are extracted; s5, fusing the multi-source data to generate a table type data set; s6, screening sensitive indexes based on SHAP and correlation; s7, constructing and evaluating an inversion model; s8, extracting an MNDWI water body mask; s9, generating a feature stack and inverting and outputting a total nitrogen concentration diagram; according to the method, through fusion of multi-source data, construction of a high-precision inversion model and a cloud platform automatic processing flow, high-temporal-spatial-resolution, stable and reproducible remote sensing estimation of the total nitrogen concentration of the river is realized, and the monitoring efficiency and the application value are effectively improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of water ecological environment remote sensing monitoring, and particularly relates to a river total nitrogen concentration remote sensing estimation method based on multi-source data fusion. BACKGROUND

[0002] Total nitrogen concentration is an important indicator reflecting the nutritional status of rivers, and has a significant impact on the health and function of aquatic ecosystems. High concentration of total nitrogen can cause eutrophication, leading to overpopulation of algae and phytoplankton, forming water blooms, reducing water light transmittance and dissolved oxygen content, and destroying the living environment of fish and benthic organisms, thereby affecting species diversity and community structure. At the same time, the fluctuation of total nitrogen concentration also interferes with the ecological chain of rivers, changes the nutrient cycle and benthic ecosystem service function, reduces the natural recovery ability of the ecosystem, and threatens the ecological services such as water quality purification, fishery resources and water source protection relied by human beings.

[0003] Existing water quality monitoring methods mainly rely on field sampling and laboratory analysis, which are accurate but time-consuming, labor-intensive and difficult to achieve continuous observation in a large range. Limited sampling points cannot fully reflect the water quality status, and traditional methods are difficult to track the dynamic changes of rivers in a timely manner, making it difficult to support the rapid identification and management of pollution sources in the river basin. Although remote sensing technology has been applied in the field of water quality monitoring, models based on single spectral information or single data source still have deficiencies in precision and generalization ability. SUMMARY

[0004] The present application provides a river total nitrogen concentration remote sensing estimation method based on multi-source data fusion, which comprehensively utilizes remote sensing images, meteorological data and terrain data, constructs a multi-dimensional feature database and selects sensitive indicators, uses machine learning technology to establish a high-precision inversion model, and realizes the automation of data processing, model reasoning and mapping on the remote sensing cloud platform, obtains high spatio-temporal resolution total nitrogen concentration distribution products, realizes efficient, reproducible and continuous monitoring of river total nitrogen concentration, and provides scientific basis for river basin governance and water environment management.

[0005] A river total nitrogen concentration remote sensing estimation method based on multi-source data fusion, comprising the following steps: S1, obtaining total nitrogen measured data of the target area, unifying time and coordinate reference system, and completing missing and abnormal value quality inspection; S2, processing Landsat8 / 9 image data, matching collection time and spatial coordinates, extracting water body reflectivity and calculating remote sensing indicators; S3, obtaining ERA5-Land meteorological data, matching surface runoff, air temperature and precipitation data according to collection time and spatial coordinates, and calculating time derivative indicators; S4, obtaining digital elevation model data, and obtaining elevation information through spatial coordinate matching; S5, spatiotemporal alignment and fusion of water body reflectivity, remote sensing indicators, meteorological data and elevation information with the monitoring section-observation time as the composite primary key to generate tabular data sets; S6, calculate SHAP and correlation analysis to evaluate feature importance, and select sensitive indicators related to total nitrogen concentration changes to construct a multi-dimensional sensitive indicator database; S7, divide the multi-dimensional sensitive indicator database into training set and test set to build an inversion model, and use cross-validation and error evaluation to evaluate the performance of the inversion model; S8, calculate the MNDWI index and extract the water body range using threshold segmentation method to generate a water mask image; S9, build a multi-channel feature stack consistent with the training, and apply the inversion model with the best performance within the water mask image range to output the spatial distribution map of total nitrogen concentration in rivers.

[0006] Optionally, the S1 comprises: S11, collect total nitrogen measurement data, including total nitrogen concentration (TN), monitoring time, longitude and latitude, and time resolution of 4 hours; S12, uniformly process the collected total nitrogen measurement data, including converting the time to UTC standard time and the coordinates to WGS84 spatial reference system; S13, quality control of the collected total nitrogen measurement data, including missing and outlier quality control.

[0007] Optionally, the S2 comprises: S21, according to Landsat8 / 9 image data, select image data with cloud cover not exceeding 20%, and use QA_PIXEL mask to remove abnormal pixels to form an effective image set; S22, with the monitoring section-observation time as the anchor point, select the image data with the smallest time difference within the ±1 hour time window, and extract the pixel reflectivity using the nearest neighbor method to obtain the water body reflectivity; S23, according to the water body reflectivity, calculate the remote sensing indicators, including NDWI, MNDWI, NDVI, BNDVI, CVI, GBNDVI, RBNDVI, NDCI, OC_Index and band ratio; S24, match the water body reflectivity, remote sensing indicators and total nitrogen measurement data in time and space to build a remote sensing sample set; S25, remove missing values and outliers from the remote sensing sample set.

[0008] Optionally, the S3 comprises: S31, obtain ERA5-Land meteorological data; S32, with the monitoring section-observation time as the anchor point, matching the corresponding time data within a time window of ±1 hour, extracting surface runoff, 2-meter air temperature and total precipitation; S33, calculating the 5-day sliding window cumulative amount on the precipitation sequence.

[0009] Optionally, the S6 comprises: S61, calculating the SHAP values of water reflectivity, remote sensing indicators, meteorological data, elevation information, latitude and longitude information; S62, using the Pearson correlation coefficient method to calculate the correlation coefficients between water reflectivity, remote sensing indicators, meteorological data, elevation information, latitude and longitude information and total nitrogen concentration; S63, comprehensively considering the SHAP values and the Pearson correlation coefficients, constructing a multi-dimensional sensitive index database including elevation information, latitude and longitude information, 2-meter air temperature, the ratio of blue band to green band, the ratio of coastal band to green band, the ratio of short-wave infrared band 1 to short-wave infrared band 2, the ratio of blue band to short-wave infrared band 2, the ratio of coastal band to blue band, and total precipitation.

[0010] Optionally, the correlation coefficient is represented as: ; Wherein, r is the Pearson correlation coefficient, n is the number of observation values, x and y are the observation values of two variables.

[0011] Optionally, the S7 comprises: S71, dividing the multi-dimensional sensitive index database into a training set and a test set in a ratio of 7:3; S72, performing hyperparameter tuning on the machine learning inversion model for remote sensing estimation of total nitrogen concentration; S73, using K=5-fold cross-validation and fixed random seed to evaluate the stability of the inversion model, and completing the training of the inversion model; S74, using the test set as the input variable, and using the trained inversion model to predict the total nitrogen concentration; S75, calculating the performance evaluation index of the inversion model.

[0012] Optionally, the performance evaluation index comprises the root mean square error and the determination coefficient, and is represented as: ; Wherein, is the measured value, is the predicted value of the inversion model, is the sample number, is the root mean square error; ; Wherein, is the mean value of total nitrogen concentration, is the coefficient of determination.

[0013] Optionally, the S8 comprises: S81, according to the spatial distribution requirement of total nitrogen concentration inversion, acquiring Landsat8 / 9 image data of the target phase, and removing abnormal pixels based on the QA_PIXEL quality mask; S82, calculating the improved normalized difference water index MNDWI using the green band (B3) and the short-wave infrared band (B6); S83, based on MNDWI, classifying the pixels into water body area and non-water body area according to the threshold value, wherein MNDWI is greater than or equal to the threshold value, and is marked as a water body area, and MNDWI is less than the threshold value, and is marked as a non-water body area, a water mask image is generated, and the generated water mask image is smoothed and refined to form a final river range.

[0014] The beneficial effects of the present application are: The present application, by fusing Landsat 8 / 9 remote sensing image reflectivity, remote sensing index, ERA5-Land meteorological data and DEM elevation information and other multi-source heterogeneous data, constructs a multi-dimensional feature database including spectral, hydrological, meteorological and topographic dimensions, and combines SHAP value and correlation analysis to screen sensitive indicators, effectively improves the robustness and generalization ability of the total nitrogen concentration inversion model under different regions and different climate conditions, and is superior to the traditional estimation method which only relies on spectral information.

[0015] The present application, by using the high temporal resolution remote sensing image resources provided by the Landsat 8 / 9 dual satellite network, and combining the batch processing capability of the cloud platform, realizes the spatial and temporal continuous monitoring capability of the total nitrogen concentration with a period of about 8 days, can quickly respond to water environment changes, significantly improves the timeliness and regional coverage of the total nitrogen concentration inversion results, and is suitable for dynamic monitoring and management of large-scale basins.

[0016] The present application, by constructing an integrated automatic process of image processing, feature stack generation, inversion model reasoning and result mapping on the remote sensing cloud platform, realizes the full-chain automatic processing from the original image to the inversion result, not only improves the efficiency of remote sensing estimation, but also guarantees the standardization and reproducibility of the results, provides efficient and reliable technical support for watershed water quality evaluation and ecological supervision. BRIEF DESCRIPTION OF DRAWINGS

[0017] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only for this invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a schematic diagram of the estimation method according to an embodiment of the present invention; Figure 2 This is a schematic diagram of satellite imagery data according to an embodiment of the present invention; Figure 3 This is a schematic diagram of the SHAP analysis results during the random forest modeling process in this embodiment of the invention; Figure 4 This is a schematic diagram of the SHAP analysis results during the XGBoost modeling process in this embodiment of the invention; Figure 5 This is a schematic diagram of the model scatter plot in an embodiment of the present invention; Figure 6 This is a schematic diagram of the water extraction results according to an embodiment of the present invention; Figure 7 This is a schematic diagram of the spatial distribution of total nitrogen concentration in an embodiment of the present invention. Detailed Implementation

[0019] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments. Those skilled in the art may employ other alternative methods to implement some well-known technologies; moreover, the accompanying drawings are only for more specific description of the embodiments and are not intended to specifically limit the present invention.

[0020] like Figures 1-7 As shown, a remote sensing estimation method for total nitrogen concentration in rivers based on multi-source data fusion includes the following steps: S1, obtain the historical database of total nitrogen concentration; S2, processes Landsat 8 / 9 image data, matches acquisition time and spatial coordinates to extract water reflectance and calculate relevant remote sensing indicators; S3 utilizes ERA5-Land meteorological data, matching the collection time and spatial coordinates to obtain surface runoff, temperature, and precipitation data, and calculates time-derived indices; S4: Acquire digital elevation model data and obtain elevation information through spatial coordinate matching; S5, multi-source data spatiotemporal fusion, constructing a tabular training database; S6. Perform feature importance analysis, remove redundant features, retain key variables with strong correlation, and form a multidimensional sensitive indicator database. S7 divides the multidimensional sensitive indicator database into a training set and a test set. Using the sensitive indicators as input and the total nitrogen concentration as output, an inversion model is constructed, and the model performance is verified through the test set to select the best model. S8, extracting river extent based on MNDWI and threshold segmentation method; S9 generates a multi-channel feature stack, applies the best model to the water body, and outputs the spatial distribution product of total nitrogen concentration.

[0021] Changes in total nitrogen (TNO) concentration directly impact river ecosystems, disrupting interactions between organisms, altering nutrient cycling and benthic ecosystem services, and leading to biodiversity decline. Therefore, conducting remote sensing-based TNO monitoring research is crucial. While traditional water quality monitoring is accurate, it is time-consuming and labor-intensive, and the limited number of sampling points cannot comprehensively reflect water quality conditions. Satellite remote sensing technology is increasingly used in marine and inland water quality monitoring, offering advantages such as long-range monitoring and low cost, effectively compensating for the shortcomings of traditional methods. Remote sensing technology can obtain high spatiotemporal resolution data, helping to identify the nutrient status and health level of river ecosystems, while supporting the management and control of pollution sources. Remote sensing monitoring can track river dynamics in real time, improving the timeliness and scientific rigor of responses to water environment issues. Furthermore, remote sensing data provides scientific evidence and decision support for watershed management and water resource management, effectively promoting sustainable water environment management. Therefore, using remote sensing technology to monitor river TNO concentration is not only technically feasible but also contributes to efficient and comprehensive ecological monitoring, holding significant practical importance for the protection and restoration of aquatic ecosystems.

[0022] The specific implementation method is as follows: The spatial distribution of total nitrogen concentration in the main stream of the Yellow River in a rectangular area of ​​the Kaifeng section in 2024 was estimated, combined with a flow chart ( Figure 1 (This will be explained in detail.)

[0023] Step 1: Obtain measured total nitrogen (TN) data from hydrological stations in the study area from 2020 to 2024, with a time resolution of 4 hours. Fields include TN, monitoring time, longitude, and latitude. The time is unified to UTC and the coordinate system is WGS84. Complete the quality check for missing and outlier values.

[0024] Step Two: As Figure 2Landsat 8 / 9 Level-2 surface reflectance (SR) images from 2020 to 2024 were acquired using a remote sensing cloud platform. First, images were filtered by spatial and temporal range of the study area, then initially selected based on scene-level cloud cover ≤ 20%. At the pixel level, a QA_PIXEL mask was used to remove clouds, cloud shadows, snow cover, and saturated pixels, resulting in a valid image set (a total of 2218 images). Using the "monitoring section - observation time" as the anchor point, images with the smallest time difference were selected within a ±1-hour window in the temporal dimension. In the spatial dimension, the section coordinates were reprojected onto the image coordinate system, and pixel reflectance was extracted using the nearest neighbor method to obtain water reflectance samples. Based on the acquired water reflectance data, band calculations were performed, the specific calculation method of which is as follows: Band ratio method: B ij =B i / B j (i <j,i,j=1,2,3,4,5,6,7); Remote sensing index method: NDWI = (B3 - B5) / (B3 + B5); MNDWI=(B3-B6) / (B3+B6); NDVI = (B5 - B4) / (B5 + B4); BNDVI = (B5 - B2) / (B5 + B2); CVI = B5 × B4 / B3; GBNDVI=(B5-B3+B2) / (B5+B3+B2); RBNDVI=(B5-B2+B4) / (B5+B4+B2); ; ; Among them, B1 is the Coastal band, B2 is the blue band, B3 is the green band, B4 is the red band, B5 is the near-infrared band, B6 is the short-wave infrared band 1, and B7 is the short-wave infrared band 2.

[0025] The extracted band reflectance and remote sensing index were strictly matched with TN data in terms of time and space. Successful matching records were retained and marked as missing / abnormal, and missing and abnormal samples were removed.

[0026] Step 3: Obtain ERA5-Land hourly data from 2020 to 2024; use ±1 hour time window matching according to “monitoring section - observation time” to extract total_runoff (m), 2m_temperature (K), and total_precipitation (m), and calculate the 5-day sliding window cumulative amount on the precipitation series to characterize the lag effect; Step 4: Acquire DEM data, reproject and resample to a CRS and 30m resolution consistent with Landsat, and extract elevations according to station coordinates after strict cell alignment; Step 5: The results from steps 2, 3, and 4 are spatiotemporally aligned and fused using "monitoring section - observation time" as the composite primary key to form a tabular dataset for training and validation. After matching and quality filtering, a total of 1357 valid samples were obtained. Step Six: Calculate the SHAP value for each parameter to determine its importance. The SHAP method is a technique for interpreting the output of machine learning models. Based on Shapley values ​​and game theory, it quantifies the contribution of each feature to the model's predictions, thus improving the model's interpretability. SHAP results are shown below. Figure 3 , Figure 4 As shown. Calculate the correlation coefficients between each parameter and the total nitrogen concentration. The calculation formula is as follows: ; Where: r is the Pearson correlation coefficient, n is the number of observations, and x and y are the observed values ​​of the two variables. The Pearson coefficient ranges from -1 to 1. A value less than 0 indicates a negative correlation between the parameter and total nitrogen concentration, while a value greater than 0 indicates a positive correlation. The larger the value, the better the correlation between the parameter and the total nitrogen concentration. Based on SHAP and correlation screening, a multidimensional sensitive index database was constructed, which includes DEM, longitude, temperature, B2 / B3, B1 / B3, B6 / B7, B2 / B7, B1 / B2, and precipitation.

[0027] Step 7: Divide the sensitive indicator database into a 7:3 training / test set, perform hyperparameter tuning using grid search or Bayesian optimization, and evaluate model generalization using K=5-fold cross-validation and fixed random seeds. After completing model construction, validate the model using the test set. Input the selected sensitive indicators from the test set into the inversion model to obtain the predicted total nitrogen concentration values. Use the root mean square error and coefficient of determination to evaluate the model's performance, as shown in the following formula: The formula for calculating the root mean square error (RMSE) is: ; in, These are measured values. The predicted value of the model. This represents the number of samples.

[0028] Coefficient of determination (R) 2 The formula for calculating ) is: ; in, This represents the average total nitrogen concentration.

[0029] During the validation process, the performance of multiple models was compared, and the model that performed best on the test set was selected as the final total nitrogen concentration inversion model. The model scatter plot is shown below. Figure 5 As shown. Through this series of steps, an accurate model with good generalization ability is constructed.

[0030] Step 8: Based on the spatial distribution requirements of total nitrogen concentration, acquire Landsat 8 / 9 remote sensing images for the corresponding time period. Remove anomalous pixels such as clouds, fog / shadows, and snow using QA_PIXEL to improve the accuracy of water body identification. Calculate the MNDWI index using the green light band and shortwave infrared band, using the following formula: ; B3 represents the green light band, and B6 represents the shortwave infrared band. The higher the MNDWI value, the greater the likelihood of a body of water.

[0031] After calculating the MNDWI index, an MNDWI image is obtained. The maximum threshold for MNDWI is automatically calculated using the Otsu method, dividing the MNDWI values ​​into water and non-water regions. Based on the calculated threshold, pixels above the threshold are marked as 1 (water region), and pixels below the threshold are marked as 0 (non-water region), forming a water mask image. The extracted water mask is smoothed and thinned to remove isolated small regions and noise, forming the final river extent, as shown below. Figure 6 As shown, this provides basic data support for subsequent remote sensing inversion of total nitrogen concentration.

[0032] Step Nine: Construct a multi-channel feature stack consistent with the training at coordinates identical to Landsat and a resolution of 30m. Apply the optimal model from Step Seven within the water mask obtained in Step Eight to perform pixel-level inference, generate a spatial distribution product of total nitrogen concentration, export it in the standard GeoTIFF format, and perform hierarchical color mapping. Figure 7 ).

[0033] This invention encompasses any substitutions, modifications, equivalent methods, and solutions made within the spirit and scope of this invention. To provide the public with a thorough understanding of this invention, specific details are described in detail in the following preferred embodiments; however, those skilled in the art will fully understand the invention even without these details. Furthermore, to avoid unnecessary misunderstanding of the essence of this invention, well-known methods, processes, procedures, components, and circuits are not described in detail.

[0034] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A remote sensing estimation method for total nitrogen concentration in rivers based on multi-source data fusion, characterized in that, Includes the following steps: S1, acquire the measured total nitrogen data of the target area, unify the time and coordinate reference system, and complete the quality inspection of missing and outlier values; S2, processes Landsat 8 / 9 image data, matches acquisition time and spatial coordinates to extract water reflectance and calculate remote sensing indicators; S3: Acquire ERA5-Land meteorological data, and obtain surface runoff, temperature and precipitation data based on the acquisition time and spatial coordinate matching, and calculate time-derived indices; S4: Acquire digital elevation model data and obtain elevation information through spatial coordinate matching; S5 aligns and fuses water reflectance, remote sensing indicators, meteorological data, and elevation information in a spatiotemporal manner using the monitoring section-observation time as the composite primary key, generating a tabular dataset; S6, calculate SHAP and use correlation analysis to assess the importance of features, screen out sensitive indicators related to changes in total nitrogen concentration, and construct a multidimensional sensitive indicator database; S7 divides the multidimensional sensitive index database into training and testing sets to build an inversion model, and uses cross-validation and error to evaluate the performance of the inversion model. S8, calculate the MNDWI index and use the threshold segmentation method to extract the water body range, and generate a water body mask image; S9 constructs a multi-channel feature stack consistent with the training, applies the best-performing inversion model within the water body mask image range, and outputs a spatial distribution map of total nitrogen concentration in the river.

2. The remote sensing estimation method for total nitrogen concentration in rivers based on multi-source data fusion according to claim 1, characterized in that, S1 includes: S11 collects measured total nitrogen data, including total nitrogen concentration, monitoring time, longitude and latitude, with a time resolution of 4 hours; S12, the collected total nitrogen measurement data are processed uniformly, including unifying the time to UTC standard time and the coordinates to WGS84 spatial reference system; S13, perform quality control on the collected total nitrogen measurement data, including missing and outlier checks.

3. The remote sensing estimation method for total nitrogen concentration in rivers based on multi-source data fusion according to claim 2, characterized in that, S2 includes: S21. Based on Landsat 8 / 9 image data, filter image data with cloud cover not exceeding 20%, and use QA_PIXEL mask to remove abnormal pixels to form an effective image set; S22, using the monitoring section-observation time as the anchor point, selects the image data with the smallest time difference within a ±1 hour time window, and extracts the pixel reflectance using the nearest neighbor method to obtain the water reflectance; S23. Calculate remote sensing indices based on water reflectance, including NDWI, MNDWI, NDVI, BNDVI, CVI, GBNDVI, RBNDVI, NDCI, OC_Index, and band ratios. S24, matching water reflectance, remote sensing indicators and total nitrogen measured data in time and space to construct a remote sensing sample set; S25, remove missing and outlier values ​​from the remote sensing sample set.

4. The remote sensing estimation method for total nitrogen concentration in rivers based on multi-source data fusion according to claim 3, characterized in that, S3 includes: S31, acquire ERA5-Land meteorological data; S32 uses the monitoring section-observation time as the anchor point, matches the corresponding time data within a ±1 hour time window, and extracts surface runoff, 2-meter temperature and total precipitation; S33 calculates the cumulative amount over a 5-day sliding window on a precipitation series.

5. The remote sensing estimation method for total nitrogen concentration in rivers based on multi-source data fusion according to claim 4, characterized in that, S6 includes: S61, calculate the SHAP value of water reflectance, remote sensing indicators, meteorological data, elevation information, and latitude and longitude information; S62 uses the Pearson correlation coefficient method to calculate the correlation coefficients between water reflectance, remote sensing indicators, meteorological data, elevation information, latitude and longitude information and total nitrogen concentration; S63, combining SHAP values ​​and Pearson correlation coefficients, constructs a multidimensional sensitive index database, including elevation information, latitude and longitude information, 2-meter temperature, the ratio of blue light band to green light band, the ratio of coastal band to green light band, the ratio of shortwave infrared band 1 to shortwave infrared band 2, the ratio of blue light band to shortwave infrared band 2, the ratio of coastal band to blue light band, and total precipitation.

6. The remote sensing estimation method for total nitrogen concentration in rivers based on multi-source data fusion according to claim 5, characterized in that, The correlation coefficient is expressed as: ； Where r is the Pearson correlation coefficient, n is the number of observations, and x and y are the observations of the two variables.

7. The remote sensing estimation method for total nitrogen concentration in rivers based on multi-source data fusion according to claim 6, characterized in that, S7 includes: S71, the multidimensional sensitive indicator database is divided into training set and test set in a 7:3 ratio; S72, Hyperparameter tuning of the machine learning inversion model used for remote sensing estimation of total nitrogen concentration; S73, use K=5 fold cross-validation and fixed random seed to evaluate the stability of the inversion model and complete the training of the inversion model; S74 uses the test set as input variables and the trained inversion model to predict total nitrogen concentration. S75, calculate the performance evaluation index of the inversion model.

8. The remote sensing estimation method for total nitrogen concentration in rivers based on multi-source data fusion according to claim 7, characterized in that, The performance evaluation metrics include root mean square error and coefficient of determination, expressed as follows: ； in, These are measured values. These are the predicted values ​​from the inversion model. For the sample size, This is the root mean square error; ； in, This represents the average total nitrogen concentration. The coefficient of determination.

9. The remote sensing estimation method for total nitrogen concentration in rivers based on multi-source data fusion according to claim 8, characterized in that, S8 includes: S81. Based on the spatial distribution requirements of total nitrogen concentration inversion, Landsat 8 / 9 image data of the target time phase are acquired, and abnormal pixels are removed based on QA_PIXEL quality mask. S82, using green light band and shortwave infrared band to calculate the improved normalized differential water index MNDWI; S83, based on MNDWI, classifies pixels into water areas and non-water areas according to a threshold. Where MNDWI ≥ threshold, it is marked as a water area, and MNDWI < threshold, it is marked as a non-water area. A water mask image is generated, and the generated water mask image is smoothed and thinned to form the final river range.

Images