Remote sensing evaluation method and system for in-sea flux of dissolved organic carbon in medium and small river water body

Abstract

The invention provides a medium and small river water dissolved organic carbon into-sea flux remote sensing evaluation method and system.The method comprises the steps that firstly, a high-resolution remote sensing image of a target drainage basin and actual measurement data of permanganate index sampling points are obtained, and dissolved organic carbon concentration accurate actual measurement data corresponding to the permanganate index actual measurement data are calculated according to the linear regression relation; secondly, the remote sensing data of the sampling points are matched with the quasi-measured data of the dissolved organic carbon concentration, an XGBoost model is trained, and a dissolved organic carbon concentration prediction model is obtained; finally, acquiring a remote sensing image and sea inflow data of the target drainage basin on the same day, and determining a pixel corresponding to a sea inflow measurement section in the remote sensing image; utilizing the dissolved organic carbon concentration prediction model to obtain dissolved organic carbon concentration prediction data corresponding to pixels in the remote sensing image; and calculating the dissolved organic carbon into-sea flux of the target drainage basin according to the dissolved organic carbon concentration prediction data and the into-sea flow data, and calculating the dissolved organic carbon into-sea flux in combination with the into-sea flow data.

Description

Technical Field

[0001] The present invention belongs to the field of remote sensing, and in particular relates to a remote sensing evaluation method and system for the flux of dissolved organic carbon into the sea from small and medium-sized rivers. Background Art

[0002] Dissolved organic carbon (DOC), as a major nutrient element and potential pollutant, plays an important role in the carbon cycle of rivers and estuarine ecosystems. DOC is a product of photosynthesis by plants and autotrophic microorganisms, providing energy and substrate for the growth and reproduction of heterotrophic microorganisms. It is also an intermediate product in the mineralization of particulate organic carbon. Changes in DOC concentration have a significant impact on the water quality of rivers and near-shore waters, and can directly reflect changes in natural processes in the basin, as well as the impact of human activities such as changes in land use patterns, urban engineering construction, agricultural planting, and water conservancy engineering construction on water bodies. For example, the construction of reservoirs and dams will change the hydrodynamic structure of rivers, thereby affecting the transport and deposition of organic carbon. Studying changes in DOC concentration and its transmission process is essential for clarifying It is crucial to understand the fate of carbon emissions and the impact of human activities on the carbon cycle.

[0003] However, DOC is usually analyzed by high-temperature combustion, ultraviolet absorbance or chemical oxidation, which requires sophisticated experimental equipment and relatively complex sample pretreatment procedures, so the cost is relatively high. In addition, most DOC analysis equipment is not portable and difficult to use on site. Data acquisition requires laboratory support. In addition, DOC analysis takes a long time, which is not convenient for high-frequency, real-time monitoring.

[0004] In recent years, with the development of satellite remote sensing technology, DOC concentration data can be obtained through remote sensing inversion. Combined with the river's flow into the sea, the river's DOC flux into the sea can be calculated, thus providing a low-cost, seamless solution for the monitoring of traditional organic carbon fluxes. However, the non-optical activity characteristics of DOC make its direct relationship with the spectral signal weak, resulting in limited accuracy of the algorithm for inverting DOC concentration based on remote sensing spectral data. In addition, the existing remote sensing inversion models are mainly concentrated in large rivers such as the Yellow River, Yangtze River and Pearl River, and are relatively inadequate for small and medium-sized rivers. Summary of the invention

[0005] The embodiment of the present invention discloses a remote sensing assessment method and system for the flux of dissolved organic carbon into the sea from small and medium-sized rivers, so as to solve the problem that the algorithm for inverting DOC concentration from remote sensing spectral data in the prior art is limited in accuracy and is not suitable for small and medium-sized rivers.

[0006] In order to solve the above technical problems, the embodiments of the present invention disclose the following technical solutions: One aspect of the present invention provides a remote sensing assessment method for the flux of dissolved organic carbon into the sea from small and medium-sized rivers, the method comprising: Obtain high-resolution remote sensing images of the target watershed on multiple historical dates and measured data of all permanganate index sampling points; According to the preset linear regression relationship, the quasi-measured data of dissolved organic carbon concentration corresponding to each measured data of permanganate index are calculated respectively; For each sampling point, the remote sensing data of the corresponding pixel in the remote sensing image is matched with the corresponding quasi-measured data of dissolved organic carbon concentration to construct a matching data set; The XGBoost model was trained based on the matching data set to obtain a prediction model for dissolved organic carbon concentration; Obtain remote sensing images and sea discharge data of the target basin on the same day, and determine the pixels in the remote sensing images corresponding to the sea discharge measurement section; The prediction model of dissolved organic carbon concentration is used to obtain the prediction data of dissolved organic carbon concentration corresponding to the pixel in the remote sensing image; The dissolved organic carbon flux into the sea in the target basin is calculated based on the predicted dissolved organic carbon concentration data and the sea flow data.

[0007] Optionally, the method further includes: A linear regression relationship between the dissolved organic carbon concentration and the permanganate index was constructed based on the measured values ​​in the target watershed.

[0008] Optionally, constructing a linear regression relationship between the dissolved organic carbon concentration and the permanganate index measured in the target watershed includes: Obtain the measured data of dissolved organic carbon concentration and permanganate index in the target watershed in different seasons; Match the measured data of dissolved organic carbon concentration and the measured data of permanganate index to establish a measured data set; A linear regression relationship model between the two is established based on the measured data set.

[0009] Optionally, after executing the step of obtaining high-resolution remote sensing images of the target watershed on multiple historical dates and measured data of all permanganate index sampling points, the following steps are used to pre-process each remote sensing image: Atmospheric correction is performed using Rayleigh scattering lookup tables and exponential extrapolation; The normalized difference vegetation index was used for land masking; Perform cloud mask processing based on Rayleigh-corrected reflectivity and a preset reflectivity threshold; All bands were resampled using nearest neighbor interpolation, and the remote sensing image resolution was adjusted to 10 meters.

[0010] Optionally, after executing the step of matching the remote sensing data of the corresponding pixel in the remote sensing image with the corresponding quasi-measured data of dissolved organic carbon concentration for each sampling point to construct a matching data set, the method further includes: Determine whether there are pixels in the matching dataset that are affected by clouds, flares, tiny ships, and ship wakes; If so, the remote sensing data of the pixel and the corresponding quasi-measured data of dissolved organic carbon concentration are removed from the matching data set.

[0011] Optionally, the method further includes: Based on the matching data set, the correlation between each spectral characteristic parameter in the remote sensing data and the dissolved organic carbon concentration was calculated respectively; The spectral feature parameters whose correlation is greater than a preset correlation threshold are used as training parameters.

[0012] Optionally, before executing the step of training the XGBoost model based on the matching data set to obtain a dissolved organic carbon concentration prediction model, the method further includes: Optimizing the matching data set based on the training parameters so that the matching data set only includes the data of the training parameters and the quasi-measured data of dissolved organic carbon concentration; The optimized matching dataset is divided into a training set and a validation set.

[0013] Optionally, the training of the XGBoost model based on the matching data set to obtain a dissolved organic carbon concentration prediction model includes: The training parameter data in the training set is used as input features, and the quasi-measured data of dissolved organic carbon concentration is used as output features to construct the XGBoost model; The grid search method was used to determine the optimal parameter combination of the XGBoost model, and a five-fold cross-validation was used to train a prediction model for dissolved organic carbon concentration. The validation set was used to evaluate the accuracy of the dissolved organic carbon concentration prediction model.

[0014] Optionally, the calculation of the dissolved organic carbon flux into the sea in the target watershed based on the predicted dissolved organic carbon concentration data and the flux into the sea data includes: Obtain the predicted dissolved organic carbon concentration data of all pixels corresponding to the sea flow measurement section in the remote sensing image and calculate the average value; The product of the average value and the sea discharge data is taken as the dissolved organic carbon flux into the sea in the target watershed.

[0015] Another aspect of the present invention provides a remote sensing assessment system for the flux of dissolved organic carbon into the sea from small and medium-sized rivers, and the system applies the remote sensing assessment method for the flux of dissolved organic carbon into the sea from small and medium-sized rivers provided in the aforementioned aspects.

[0016] The present invention discloses a method and system for remote sensing assessment of dissolved organic carbon flux into the sea in small and medium-sized rivers. First, high-resolution remote sensing images of target watersheds and measured data of permanganate index sampling points are obtained, and quasi-measured data of dissolved organic carbon concentration corresponding to the measured data of permanganate index are calculated according to a linear regression relationship; then, the remote sensing data of the sampling points are matched with the quasi-measured data of dissolved organic carbon concentration, and the XGBoost model is trained to obtain a prediction model for dissolved organic carbon concentration; finally, the prediction data of dissolved organic carbon concentration corresponding to each pixel in the remote sensing image is obtained using the prediction model for dissolved organic carbon concentration, and the flux of dissolved organic carbon into the sea is calculated in combination with the flow rate into the sea. The present invention can effectively improve the problems of high cost, large time and space restrictions, and neglect of self-sourced organic carbon in the river in traditional methods. At the same time, it can also solve the problems of low spatial resolution and low algorithm accuracy in conventional quantitative inversion of dissolved organic carbon remote sensing.

[0017] This Summary is provided to introduce a selection of concepts in a simplified form that are further described below in the Detailed Description. This Summary is not intended to identify key features or essential features of the disclosure, nor is it intended to limit the scope of the disclosure. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] The above and other objects, features and advantages of the present disclosure will become more apparent through a more detailed description of exemplary embodiments of the present disclosure in conjunction with the accompanying drawings, wherein like reference numerals generally represent like components throughout the exemplary embodiments of the present disclosure.

[0019] Figure 1 A schematic diagram of a flow chart of a remote sensing assessment method for the flux of dissolved organic carbon into the sea from small and medium-sized rivers provided in an embodiment of the present invention; Figure 2 A schematic flow chart of another remote sensing assessment method for the flux of dissolved organic carbon into the sea from small and medium-sized rivers provided in an embodiment of the present invention; Figure 3 An implementation provided for an embodiment of the present invention Figure 2 Schematic diagram of the process of step S010; Figure 4 An implementation provided for an embodiment of the present invention Figure 1 Schematic diagram of the process of step S400. DETAILED DESCRIPTION

[0020] Embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided to make the present disclosure more thorough and complete, and to fully convey the scope of the present disclosure to those skilled in the art.

[0021] As used herein, the term "including" and its variations mean open inclusion, i.e., "including but not limited to". Unless otherwise stated, the term "or" means "and / or". The term "based on" means "based at least in part on". The terms "an example embodiment" and "an embodiment" mean "at least one example embodiment". The term "another embodiment" means "at least one additional embodiment". The terms "first", "second", etc. may refer to different or the same objects. Other explicit and implicit definitions may also be included below.

[0022] Figure 1 Schematic diagram of a flow chart of a remote sensing assessment method for the flux of dissolved organic carbon into the sea from small and medium-sized rivers disclosed in an embodiment of the present invention. Figure 1 As shown, the method includes the following steps.

[0023] Step S100: obtaining high-resolution remote sensing images of the target watershed on multiple historical dates and measured data of all permanganate index sampling points.

[0024] Acquire Sentinel-2 high-resolution multispectral remote sensing satellite data of the target watershed on multiple historical dates, as well as the measured data of all permanganate index sampling points on the same historical dates.

[0025] In one embodiment disclosed in the present invention, Figure 2 As shown, before executing step S100, the following steps are also included: Step S010: construct a linear regression relationship between the dissolved organic carbon concentration and the permanganate index measured in the target watershed.

[0026] The concentration of dissolved organic carbon (DOC) is usually analyzed by high-temperature combustion, ultraviolet absorbance or chemical oxidation, which requires sophisticated experimental equipment (such as TOC analyzer) and relatively complex sample pretreatment procedures, which are costly. In addition, DOC measurement has strict requirements on the experimental environment, and samples are easily contaminated, resulting in large errors. It requires precise control of environmental conditions, which increases the difficulty of analysis.

[0027] The permanganate index (hereinafter referred to as COD Mn The determination method of COD is usually potassium permanganate oxidation method, which is relatively simple and can be completed quickly in the laboratory or on site. Mn The reagents and instruments required for the test are low in cost and easy to operate.

[0028] For example, DOC concentration and COD Mn Take measurements.

[0029] In the process of measuring DOC concentration, the collected water samples were filtered by glass fiber membrane (Whatman GF / F), and the filter membrane was treated by baking at 450℃ for 5 hours. The water samples were stored in treated Pyrex glass sample bottles. The glass bottles were soaked in 10% hydrochloric acid solution for 24 hours before use and baked at 450℃ for 5 hours again to remove the residual carbon in the bottle. The collected water samples were measured in the laboratory by a total organic carbon analyzer (TOC-VCPH, Shimadzu). The analysis method was based on the national standard "Water Quality-Determination of Total Organic Carbon-Combustion Oxidation Non-Dispersive Infrared Absorption Method" (HJ 501-2009).

[0030] Measuring COD Mn The water samples were stored in polyethylene sample bottles, 25% (v / v) sulfuric acid was added to adjust the pH to 1-2, and stored in a light-proof environment at 1-5°C. The collected water samples were measured in the laboratory using an automatic permanganate index analyzer (APA-500, Anjie), and the analysis method was based on the national standard "Water Quality-Determination of Permanganate Index" (ISO 8467-1986).

[0031] Due to the complexity of equipment and analysis, DOC measurements are rarely implemented in high-frequency, long-term monitoring programs. Mn The detection method is relatively mature and can achieve long-term, high-frequency monitoring. Therefore, in the disclosed embodiment of the present invention, a linear regression relationship between the dissolved organic carbon concentration and the permanganate index is established to convert the more easily obtained permanganate index measured data into quasi-measured data of the dissolved organic carbon concentration.

[0032] In one embodiment disclosed in the present invention, Figure 3 As shown, the following sub-steps may be used to implement step S010: Step S0101: Obtain the measured data of dissolved organic carbon concentration and permanganate index in the target watershed in different seasons.

[0033] Obtain DOC concentration and COD at different sampling points in the main stream in the target watershed in four seasons Mn The measured data covers different hydrological and climatic conditions. Mn It fluctuates with the change of seasons and basin conditions, so the robustness of the model can be enhanced by using data from different seasons.

[0034] The DOC concentration and COD concentration were measured at the same sampling point and within the same time range. Mn Water samples to ensure DOC and COD Mn The data are from the same environmental conditions.

[0035] Step S0102: Match the measured data of dissolved organic carbon concentration and the measured data of permanganate index to establish a measured data set.

[0036] The measured data of DOC concentration and COD Mn The measured data are aligned according to the sampling time and sampling point location to ensure that the sampling time and spatial location of the two are consistent. For time points with missing values, interpolation methods can be considered or incomplete data records can be removed to ensure data consistency.

[0037] The measured data of DOC concentration and the corresponding COD Mn The measured data are paired and included in the data set as a complete record. For example, for each sampling period and sampling point location, the record is (sampling period, sampling point location, DOC concentration measured data, COD Mn Measured data), ensuring that the sample has one-to-one corresponding DOC concentration measured data and COD Mn Measured data.

[0038] Step S0103: Establish a linear regression relationship model between the two based on the measured data set.

[0039] In a specific embodiment disclosed in the present invention, a linear regression relationship model can be constructed in the following manner: The measured data of DOC concentration were used as the independent variable (X), COD Mn The measured data is used as the dependent variable (Y) to establish a simple linear regression model: Among them, a is the regression coefficient and b is the intercept.

[0040] The least squares method was used to fit the model. After fitting, the regression coefficient a and intercept b were obtained, thereby obtaining the DOC concentration and COD Mn The linear regression relationship model between .

[0041] In one embodiment disclosed in the present invention, after executing step S100, each remote sensing image is preprocessed using the following method: (1) Atmospheric correction is performed using the Rayleigh scattering lookup table and exponential extrapolation method.

[0042] The purpose of atmospheric correction of remote sensing images is to eliminate the influence of atmospheric components on the images and obtain a reflectivity close to the actual surface of the object.

[0043] Through the atmospheric correction algorithm, the Rayleigh scattering lookup table is used to eliminate the radiation effect caused by Rayleigh scattering, which is a part of atmospheric scattering and mainly affects the shortwave band.

[0044] Assuming that the water-off radiation in the shortwave infrared band (1610nm and 2190nm) is zero, the exponential extrapolation method is used to calculate the aerosol contribution in the visible and near-infrared bands. By estimating the aerosol concentration, the atmospheric correction can be performed on the reflectance in the visible and near-infrared bands to obtain the remote sensing reflectance. This method is suitable for highly turbid waters and inland waters, especially when there are a lot of suspended matter in the water body.

[0045] (2) Use the normalized difference vegetation index to perform land masking.

[0046] After atmospheric correction of the remote sensing images, the land mask processing of the remote sensing images was performed using the Normalized Difference Vegetation Index (NDVI) calculated based on the reflectance of the near-infrared (865nm) and red light (665nm) bands.

[0047] Land masking separates water bodies from land areas through the Normalized Difference Vegetation Index (NDVI). The difference between the near-infrared and red bands in remote sensing images can be used to effectively distinguish vegetation from water bodies. Based on the threshold of the NDVI value, the vegetation and land pixels in the image can be masked out, leaving only the spectral information of the water area. This process simplifies the analysis of water remote sensing data and makes subsequent processing more accurate.

[0048] (3) Perform cloud mask processing based on the Rayleigh corrected reflectivity and the preset reflectivity threshold.

[0049] The 2190nm band is in the near infrared (NIR) range and has good ability to distinguish clouds from other surface features. In this band, the reflectivity of clouds is generally high, while the reflectivity of water bodies and vegetation is low.

[0050] A reflectivity threshold is set in advance, and a threshold judgment is performed on each pixel in the image. When the Rayleigh-corrected reflectivity of a pixel in the 2190nm band in the remote sensing image is higher than the reflectivity threshold, the pixel can be considered to contain clouds, otherwise it is marked as non-cloud, thereby generating a cloud mask map to distinguish clouds from other ground objects.

[0051] (4) Perform nearest neighbor interpolation resampling on all bands and adjust the remote sensing image resolution to 10 meters.

[0052] Nearest neighbor interpolation can redistribute pixel values ​​into a grid of target resolution (e.g., 10 meters). For each new pixel, nearest neighbor interpolation directly uses the closest original pixel value without weighted calculation, thus maintaining the discreteness of the image and being suitable for processing classified or discrete value data.

[0053] Interpolation is performed for each band separately. The nearest neighbor interpolation process recalculates the pixel coordinates of each band according to the target resolution and matches them to the nearest pixel in the new resolution grid. If the original projection coordinates of the image are different from the grid corresponding to the target resolution, a projection transformation is first performed to ensure geographic alignment at the target resolution.

[0054] In one embodiment disclosed in the present invention, the measured data of all permanganate index sampling points in the target watershed on multiple historical dates can be obtained in the following manner: Collect COD from all automatic monitoring stations (sampling points) in the target watershed every day Mn Data, select the historical date that is the same as the date of acquisition of remote sensing images from the collected data, so that COD Mn The data and remote sensing image data can correspond to each other in terms of date.

[0055] Step S200: Calculate the quasi-measured data of dissolved organic carbon concentration corresponding to each measured data of permanganate index according to a preset linear regression relationship.

[0056] Using the preset linear regression relationship disclosed in the above embodiment, each COD Mn The measured data are converted into corresponding DOC concentration prediction values. DOC concentration prediction value = a × COD Mn Measured value + b In the disclosed embodiment of the present invention, COD Mn The data are high-frequency, long-time series measured data (such as daily or hourly observations), which are converted through linear regression to generate quasi-measured DOC concentration data of the same frequency and time series, thus solving the problem of not being able to obtain DOC concentration data at high frequency and long time series.

[0057] Step S300: For each sampling point, the remote sensing data of the corresponding pixel in the remote sensing image is matched with the corresponding quasi-measured data of dissolved organic carbon concentration to construct a matching data set.

[0058] In the embodiment disclosed in the present invention, first, the position coordinates of each sampling point are determined, and then, according to the spatial resolution of the remote sensing image, the position coordinates of the sampling point are converted to the remote sensing image grid, the pixel closest to the sampling point position is found to correspond to the sampling point, and the remote sensing data of the corresponding pixel is obtained.

[0059] Sentinel-2 high-resolution multi-spectral remote sensing satellite data is used in the disclosed embodiments of the present invention, so the remote sensing data includes spectral characteristic data (i.e., reflectivity) of multiple bands. At the same time, the disclosed embodiments of the present invention also construct data of multiple different band combinations based on the spectral characteristic data of each band, and use all bands and all band combinations as spectral characteristic parameters in the remote sensing data.

[0060] It is assumed that data are collected once a day at the permanganate index sampling points and a remote sensing image is taken on each selected historical date.

[0061] Taking any permanganate index sampling point as an example, COD Mn The data is matched with the remote sensing data of the pixel at the sampling point in the remote sensing image taken on day M. As a sample in the matching data set, COD is converted into Mn After matching the data with the pixel remote sensing data, a matching data set is obtained.

[0062] After completing step S300, the embodiment of the present invention further uses the following method to perform quality control on the sample data in the matching data set: Determine if there are pixels in the matching dataset that are affected by clouds, tiny vessels, ship wakes, and flares.

[0063] For example, cloud mask algorithms such as Fmask (Function of mask) or machine learning-based cloud detection models can be applied to identify cloud and cloud shadow pixels using the reflectivity characteristics of specific bands (such as 2.19 μm).

[0064] Alternatively, the reflectance value of a specific band in the hyperspectral remote sensing data can be used to set the threshold, such as the infrared or shortwave infrared band (such as 1.38μm and 2.2μm). Clouds usually have higher reflectance in these bands, and setting a reasonable threshold can effectively identify cloud pixels.

[0065] Flares usually manifest as an abnormal increase in reflectivity in a specific band. By calculating the abnormal values ​​of the reflectivity in each band, possible flare pixels can be preliminarily screened.

[0066] Flares can cause abnormal peaks in the spectral curve in local bands, and discontinuous abnormal changes in the spectral curve can also be detected through spectral smoothness checks (such as local standard deviation or deviation analysis).

[0067] The remote sensing data and the corresponding quasi-measured data of DOC concentration of pixels affected by clouds and flares in the matching dataset are removed.

[0068] In the embodiments disclosed in the present invention, visual interpretation can also be used, for example, receiving data of affected pixels input by a user to determine whether there are pixels in the matching data set that are affected by clouds, tiny ships, ship wakes, and flares.

[0069] In one embodiment disclosed in the present invention, after establishing the matching data set, the spectral characteristic parameters are further screened in the following manner, that is, the spectral characteristic parameters that are sensitive to the DOC concentration are screened, wherein the spectral characteristic parameters include various different bands and a plurality of set band combinations. For example, the remote sensing image includes Band1-Band9 bands, and the spectral characteristic parameters include not only these 9 bands, but also a plurality of preset band combinations such as Band4+Band5, Band4+Band5+Band7, (Band4+Band5) / Band2, Band7-Band9, etc.

[0070] (1) Based on the matching data set, the correlation between each spectral characteristic parameter in the remote sensing data and the dissolved organic carbon concentration is calculated.

[0071] In the embodiments disclosed in the present invention, the correlation value between the data value of each spectral characteristic parameter and the DOC concentration may be calculated by using a Pearson Correlation Coefficient or a Spearman Rank Correlation Coefficient.

[0072] (2) The spectral feature parameters whose correlation is greater than the preset correlation threshold are used as training parameters.

[0073] The absolute values ​​of the correlation values ​​are analyzed, and spectral feature parameters whose correlation is greater than a preset correlation threshold (eg, 0.6 or 0.7) are selected as training parameters.

[0074] In a specific embodiment disclosed in the present invention, the selected training parameters are: 1. 2. 3. 4. 5. Where N is the band number from 1 to 9. Represents the band with serial number i.

[0075] In one embodiment disclosed in the present invention, before executing step S400, the following steps need to be performed: (1) The matching data set is optimized based on the training parameters so that the matching data set only includes the data of the training parameters and the quasi-measured data of dissolved organic carbon concentration.

[0076] The matching data set is optimized, and only the data values ​​of the training parameters are retained in the remote sensing data of each sample, and the data of the spectral characteristic parameters with low correlation are removed, so that the optimized matching data set only contains the training parameter data with high correlation and the corresponding quasi-measured data of DOC concentration, so as to reduce the dimension and size of the data set and improve the training efficiency and accuracy of the model.

[0077] (2) Divide the optimized matching dataset into a training set and a validation set.

[0078] 80% of the data were randomly selected from the matching data group as the training data set for building the prediction model, and the remaining 20% ​​of the data were used as the validation data set to evaluate the accuracy of the prediction model.

[0079] Step S400: training the XGBoost model based on the matching data set to obtain a dissolved organic carbon concentration prediction model.

[0080] In one embodiment disclosed in the present invention, Figure 4 As shown, the training of the prediction model can be completed in the following sub-steps: Step S401: construct an XGBoost model using the training parameter data in the training set as input features and the quasi-measured data of dissolved organic carbon concentration as output features.

[0081] The input features are the training parameter data in the training set, and the output features are the quasi-measured data of dissolved organic carbon concentration in the training set. The model is built using the XGBoost (Extreme Gradient Boosting) algorithm. XGBoost is an efficient gradient boosting decision tree algorithm that can predict the target variable through a weighted combination of a series of decision trees.

[0082] In the training process, the input features (spectral features) of the training data set are fitted with the output features (DOC concentration) to learn the relationship between features and targets. Initial training is performed using the default XGBoost parameters to ensure that the model can learn the relationship between spectral features and DOC concentration.

[0083] Step S402: A grid search method is used to determine the optimal parameter combination of the XGBoost model, and a five-fold cross validation is used to train a dissolved organic carbon concentration prediction model.

[0084] Grid search is a hyperparameter optimization technique that mainly searches through all possible hyperparameter combinations to find the parameter combination that makes the model perform best. Grid search will traverse different hyperparameter settings and select the combination that performs best in the XGBoost model.

[0085] Cross-validation divides the training data set into several subsets (folds), trains the model multiple times and verifies it on different data subsets, and finally obtains the average performance of the model. Five-fold cross-validation means dividing the data set into 5 subsets (or folds), training with 4 subsets each time, and verifying with the remaining 1 subset, repeating 5 times, with each subset taking turns as the verification set, and finally averaging the results of the 5 verifications.

[0086] The evaluation results of the five rounds of cross-validation are averaged to obtain the performance of the model on the entire dataset to reduce the bias that may be generated on a single validation set and obtain a more robust evaluation.

[0087] The hyperparameter combination determined in a specific embodiment disclosed in the present invention is: n_estimators is 110, learning_rate is 0.03, max_depth is 4, min_child_weight is 4, gamma is 0.24, colsample_bytree is 0.8, subsample is 0.5, reg_alpha is 0.01, and reg_lambda is 0.01.

[0088] Through the above method, a dissolved organic carbon concentration prediction model was trained.

[0089] Step S403: using the validation set to evaluate the accuracy of the dissolved organic carbon concentration prediction model.

[0090] The purpose of using the validation set to evaluate the accuracy of the dissolved organic carbon concentration prediction model is to understand the generalization ability and prediction accuracy of the model through a comprehensive evaluation of the model performance. The comprehensive evaluation can be: 2 , root mean square error RMSE, mean absolute percentage error MAPD and bias deviation.

[0091] Step S500: Acquire the remote sensing image and sea discharge data of the target watershed on the same day, and determine the pixels in the remote sensing image corresponding to the sea discharge measurement section.

[0092] The sea flow data of the target basin on the target date is obtained based on the river flow data measured by the hydrological station or the SWAT model simulation, where the sea flow data refers to the water flow data of the water body passing through the measurement section, and the flow measurement section refers to a cross-section selected in the water body.

[0093] A multispectral remote sensing image of the target watershed on the target date is obtained, and preprocessed using the method in the aforementioned embodiment.

[0094] According to the position coordinates of the flow measurement section, the corresponding pixels are determined in the remote sensing image.

[0095] Step S600: using a dissolved organic carbon concentration prediction model to obtain predicted dissolved organic carbon concentration data corresponding to pixels in a remote sensing image.

[0096] After completing the training of the dissolved organic carbon concentration prediction model, it is applied to the prediction of dissolved organic carbon concentration. The training parameter data values ​​corresponding to the pixels of the sea flow measurement section in the remote sensing image taken on the target date are input into the dissolved organic carbon concentration prediction model, and the predicted dissolved organic carbon concentration data corresponding to these pixels can be obtained from the output data.

[0097] Step S700: Calculate the dissolved organic carbon flux into the sea in the target watershed based on the predicted dissolved organic carbon concentration data and the flux into the sea data.

[0098] (1) Obtain the predicted dissolved organic carbon concentration data for each pixel in the remote sensing image corresponding to the sea flow measurement section and calculate the average value of these predicted data .

[0099] (2) The product of the average value and the sea discharge data is taken as the dissolved organic carbon flux into the sea in the target basin.

[0100] in, represents the flux of dissolved organic carbon into the sea, represents the average value of the predicted dissolved organic carbon concentration data of all pixels in the corresponding flow measurement section into the sea, Represents the flow data into the sea.

[0101] The embodiment of the present invention also discloses a remote sensing assessment system for the flux of dissolved organic carbon into the sea in small and medium-sized rivers. The system applies the remote sensing assessment method for the flux of dissolved organic carbon into the sea in small and medium-sized rivers disclosed in the above-mentioned embodiment.

[0102] The embodiments of the present disclosure have been described above, and the above description is exemplary, not exhaustive, and is not limited to the disclosed embodiments. Many modifications and variations will be apparent to those of ordinary skill in the art without departing from the scope and spirit of the described embodiments. The selection of terms used herein is intended to best explain the principles of the embodiments, practical applications, or technical improvements to the technology in the market, or to enable other persons of ordinary skill in the art to understand the embodiments disclosed herein.

Claims

1. A remote sensing assessment method for the flux of dissolved organic carbon into the sea from small and medium-sized rivers, characterized in that: The method comprises: Obtain high-resolution remote sensing images of the target watershed on multiple historical dates and measured data of all permanganate index sampling points; According to the preset linear regression relationship, the quasi-measured data of dissolved organic carbon concentration corresponding to each measured data of permanganate index are calculated respectively; For each sampling point, the remote sensing data of the corresponding pixel in the remote sensing image is matched with the corresponding quasi-measured data of dissolved organic carbon concentration to construct a matching data set; The XGBoost model was trained based on the matching data set to obtain a prediction model for dissolved organic carbon concentration; Obtain remote sensing images and sea discharge data of the target basin on the same day, and determine the pixels in the remote sensing images corresponding to the sea discharge measurement section; The prediction model of dissolved organic carbon concentration is used to obtain the prediction data of dissolved organic carbon concentration corresponding to the pixel in the remote sensing image; The dissolved organic carbon flux into the sea in the target basin is calculated based on the predicted dissolved organic carbon concentration data and the sea flow data.

2. The evaluation method according to claim 1, characterized in that: The method further comprises: A linear regression relationship between the dissolved organic carbon concentration and the permanganate index was constructed based on the measured values ​​in the target watershed.

3. The evaluation method according to claim 2, characterized in that: The method of constructing a linear regression relationship between the dissolved organic carbon concentration and the permanganate index measured in the target watershed includes: Obtain the measured data of dissolved organic carbon concentration and permanganate index in the target watershed in different seasons; Match the measured data of dissolved organic carbon concentration and the measured data of permanganate index to establish a measured data set; A linear regression relationship model between the two is established based on the measured data set.

4. The evaluation method according to claim 1, characterized in that: After executing the steps of obtaining the high-resolution remote sensing images of the target watershed on multiple historical dates and the measured data of all permanganate index sampling points, the following steps are used to pre-process each remote sensing image: Atmospheric correction is performed using Rayleigh scattering lookup tables and exponential extrapolation; The normalized difference vegetation index was used for land masking; Perform cloud mask processing based on Rayleigh-corrected reflectivity and a preset reflectivity threshold; All bands were resampled using nearest neighbor interpolation, and the remote sensing image resolution was adjusted to 10 meters.

5. The evaluation method according to claim 1, characterized in that: After executing the step of matching the remote sensing data of the corresponding pixel in the remote sensing image with the corresponding quasi-measured data of dissolved organic carbon concentration for each sampling point to construct a matching data set, the method further includes: Determine whether there are pixels in the matching dataset that are affected by clouds, flares, tiny ships, and ship wakes; If so, the remote sensing data of the pixel and the corresponding quasi-measured data of dissolved organic carbon concentration are removed from the matching data set.

6. The evaluation method according to claim 1, characterized in that: The method further comprises: Based on the matching data set, the correlation between each spectral characteristic parameter in the remote sensing data and the dissolved organic carbon concentration was calculated respectively; The spectral feature parameters whose correlation is greater than a preset correlation threshold are used as training parameters.

7. The evaluation method according to claim 6, characterized in that: Before executing the step of training the XGBoost model based on the matching data set to obtain a dissolved organic carbon concentration prediction model, the method further includes: Optimizing the matching data set based on the training parameters so that the matching data set only includes the data of the training parameters and the quasi-measured data of dissolved organic carbon concentration; The optimized matching dataset is divided into a training set and a validation set.

8. The evaluation method according to claim 7, characterized in that: The XGBoost model is trained based on the matching data set to obtain a dissolved organic carbon concentration prediction model, including: The training parameter data in the training set is used as input features, and the quasi-measured data of dissolved organic carbon concentration is used as output features to construct the XGBoost model; The grid search method was used to determine the optimal parameter combination of the XGBoost model, and a five-fold cross-validation was used to train a prediction model for dissolved organic carbon concentration. The validation set was used to evaluate the accuracy of the dissolved organic carbon concentration prediction model.

9. The evaluation method according to claim 1, characterized in that: The calculation of dissolved organic carbon flux into the sea in the target watershed based on the predicted dissolved organic carbon concentration data and the sea flow data includes: Obtain the predicted dissolved organic carbon concentration data of all pixels corresponding to the sea flow measurement section in the remote sensing image and calculate the average value; The product of the average value and the sea discharge data is taken as the dissolved organic carbon flux into the sea in the target watershed.

10. A remote sensing assessment system for the flux of dissolved organic carbon into the sea from small and medium-sized rivers, characterized in that: The system applies the remote sensing assessment method for the flux of dissolved organic carbon into the sea from small and medium-sized rivers as described in any one of claims 1 to 9.

Images