A method for inversion of lake water quality parameters

Abstract

The present invention relates to a method for inverting lake water quality parameters, including: collecting water quality monitoring data of each station of the lake and hydrological and water quality monitoring data of rivers entering and leaving the lake upstream and downstream of the lake; driving a water ecological model based on boundary conditions to simulate the spatiotemporal dynamics of lake water quality parameters and perform model calibration verification; obtaining remote sensing image data corresponding to the lake and determining the center coordinates of each pixel, and extracting water quality parameter simulation data at corresponding points in the water ecological model based on the center coordinates of the pixel. The simulated values ​​of water quality parameters in the water ecological model whose reflectance of each band of each pixel matches its position constitute a training data set, and a machine learning model is used to learn the relationship between the reflectance of each band of the pixel and the simulated water quality parameters of the corresponding position; extracting the reflectance of each band of the pixel corresponding to each station of the lake, and inputting it into the trained machine learning model, obtaining the inverted values ​​of water quality parameters at each station, and comparing them with the measured values ​​of water quality parameters, and analyzing the inversion effect of water quality parameters.

Description

Technical Field

[0001] The invention relates to the field of water environment remote sensing, and in particular to a lake water quality parameter inversion method. Background Art

[0002] Obtaining the change process of water quality parameters of lakes with higher frequency and larger range is of great significance for lake water resources management, water environment governance, and water ecological restoration.

[0003] At present, the conventional method for monitoring lake water quality parameters is to set up monitoring stations for sampling and conduct sample testing in the laboratory to obtain the specific values ​​of each water quality parameter. However, due to the sampling and testing costs, on the one hand, the number of points is limited, and it is impossible to accurately characterize the spatial changes of water quality parameters in the entire lake. On the other hand, the sampling frequency is relatively low, mostly once a month or even a quarter, and it is impossible to continuously characterize the temporal changes of lake water quality parameters. Water quality parameter inversion based on remote sensing images is another way to monitor lake water quality parameters. Its monitoring range is wide and the revisit cycle is relatively short, that is, the monitoring frequency is relatively high. However, due to the lack of water quality data from the actual measurement stations, it is impossible to establish a good relationship between the reflectivity of each band of remote sensing images and water quality parameters, resulting in low accuracy of water quality parameter inversion, which is difficult to meet the needs of monitoring the spatiotemporal dynamic changes of lake water quality. Summary of the invention

[0004] The purpose of the present invention is to provide a lake water quality parameter inversion method, which drives the water ecological model by using water quality data from fewer monitoring stations with lower monitoring frequency, and obtains higher-frequency and higher-precision water quality parameter simulation data in the entire lake after appropriate model parameter calibration. A large amount of water quality parameter simulation data in the entire lake is corresponded to the band reflectivity of each pixel of the remote sensing image to construct a large sample data set for machine learning training to learn the relationship between the reflectivity of each band of the lake remote sensing image and the lake water quality parameters. The new remote sensing image band reflectivity is substituted into the trained machine learning model to obtain the water quality inversion parameters in the entire lake, thereby realizing the monitoring of the spatiotemporal dynamic changes of the lake water quality parameters.

[0005] To achieve the above object, the present invention provides the following technical solutions:

[0006] A lake water quality parameter inversion method, the method comprising the following steps:

[0007] Step S1: Collect water quality monitoring data from various stations in the lake and hydrological and water quality monitoring data from rivers entering and leaving the lake upstream and downstream;

[0008] Step S2: Using the water ecological model based on the hydrological and water quality monitoring data of the rivers entering and leaving the lake upstream and downstream as boundary conditions, the spatiotemporal dynamic simulation of the lake water quality parameters is performed, and the water quality monitoring data of each station in the lake is used to calibrate and verify the model parameters;

[0009] Step S3: Acquire remote sensing image data corresponding to the lake and determine the center coordinates of each pixel, and extract water quality parameter simulation data at the corresponding coordinate point in the water ecological model based on the center coordinates of the pixel;

[0010] Step S4: extracting the reflectivity of each band of each pixel of the remote sensing image, and matching it with the simulated value of the water quality parameter in the water ecological model that matches the pixel position, forming a model training set, and using a deep learning model based on this training set to learn the relationship between the reflectivity of each band of the pixel and the simulated water quality parameter at the corresponding position of the water ecological model;

[0011] Step S5: Extract the reflectance of each band corresponding to the remote sensing image pixels at the corresponding positions of each station in the lake, input the band reflectance data into the trained machine learning model, obtain the inversion value of water quality parameters at each station, and compare the inversion value of water quality parameters with the corresponding measured value of water quality parameters to analyze the inversion accuracy of water quality parameters.

[0012] As a further technical solution of the present invention, the water quality monitoring data of each station in the lake in step S1 include water quality parameters such as dissolved oxygen, ammonia nitrogen and chlorophyll a monitored monthly, and the hydrological and water quality monitoring data include hydrological data of the upstream inflow of the lake, hydrological data of the downstream outflow of the lake and water quality data of rivers entering and leaving the lake. The hydrological data of the upstream inflow of the lake are the upstream flow of the lake monitored daily, the hydrological data of the downstream outflow of the lake are the downstream water level monitored daily, and the water quality data of the rivers entering and leaving the lake are water quality parameters such as dissolved oxygen, ammonia nitrogen and chlorophyll a monitored monthly.

[0013] As a further technical solution of the present invention, the water ecological model used in the lake in step S2 is an EFDC model. By inputting the boundary conditions into the EFDC model, the EFDC model can be driven to obtain the spatiotemporal dynamic simulation results of various water quality parameters in the lake on a daily or even hourly basis under the boundary conditions of monthly scale water quality data.

[0014] As a further technical solution of the present invention, when the EFDC model is driven by the hydrological and water quality monitoring data of the rivers entering and leaving the lake upstream and downstream of the lake in step S2 to perform spatiotemporal dynamic simulation of water quality parameters in the lake, the water quality monitoring data of each station in the lake monitored monthly are further used to calibrate and verify the model parameters, wherein the simulated values ​​of the water quality parameters of the EFDC model at each station are compared with the measured values ​​of the corresponding station, and the Nash efficiency coefficient NSE and the root mean square error RMSE are used to calculate the simulation error of the water quality parameters of the EFDC model at the measured station, wherein the calculation formulas of NSE and RMSE are expressed as Formula 1 and Formula 2:

[0015] Formula 1: ;

[0016] Formula 2: ;

[0017] In the formula, Indicates the test set i The predicted value of the flood inflow to the reservoir for each model with a step length; Indicates the test set i The measured value of flood inflow in steps of n Indicates the number of steps of the test set, that is, the length of the test set; Represents the average value of all step lengths measured on the test set;

[0018] By continuously adjusting the relevant model parameters of EFDC, the NSE and RMSE at each station can reach a high accuracy. It can be considered that the EFDC model parameter settings are reasonable and can accurately simulate the spatiotemporal dynamic changes of water quality parameters in the entire lake.

[0019] As a further technical solution of the present invention, the remote sensing image in step S4 selects the Sentinel-2A satellite, wherein the Sentinel-2A satellite has a total of 13 bands, and 10m, 20m and 60m respectively include 4 bands (B2, B3, B4, B8), 6 bands (B5, B6, B7, B8). 8A , B 11 , B 12 ) and 3 bands (B1, B8, B9); determine the pixel size to be 60m, that is, each pixel is a square with a side length of 60m, which contains 36 10m resolution B2, B3, B4, B8 bands, 9 20m resolution B5, B6, B7, B9. 8A , B 11 , B 12Bands. For pixels with 10m and 20m resolution, each band in a 60m pixel corresponds to the average value of all its 10m and 20m resolution pixel bands. After averaging, there are 13 band values ​​in each 60m pixel, of which 4 bands (B2, B3, B4, B8) are the average values ​​of 36 10m resolution pixels, and 6 bands (B5, B6, B7, B8) are the average values ​​of 36 10m resolution pixels. 8A , B 11 , B 12 ) is the average value of 9 20m resolution pixels, and the three bands (B1, B8, B9) are the band values ​​corresponding to 60m resolution pixels.

[0020] As a further technical solution of the present invention, after extracting each 60m resolution lake remote sensing image pixel in step S4, the center coordinates of each pixel are calculated, and then the simulated values ​​of the lake water quality parameters at the corresponding positions of the center coordinates of each pixel are extracted in the water ecological model. The 13-band reflectance of each pixel is used as a feature, and the simulated value of the water quality parameter in the water ecological model corresponding to each pixel is used as a label to construct a feature and label training data set.

[0021] As a further technical solution of the present invention, after obtaining a sample data set consisting of simulated values ​​of water quality parameters of a water ecological model based on multiple band reflectances of each pixel and its corresponding position in step S4, a machine learning model XGBoost is used to learn the relationship between features and labels, that is, to train the XGBoost model.

[0022] As a further technical solution of the present invention, in step S5, the reflectance of each band in the remote sensing image pixel corresponding to the measured site in the lake is input as a feature into the trained XGBoost model, and the water quality parameter inversion result corresponding to the measured site in the lake is obtained based on the XGBoost model, and it is compared with the water quality monitoring data of the measured site, and the water quality parameter inversion accuracy is calculated using NSE. When the water quality parameter inversion accuracy evaluation index NSE is below 0.5, the XGBoost model parameters are further adjusted until NSE reaches the preset accuracy, which can be 0.7. At this time, it indicates that the water quality parameter inversion accuracy is good and the effect is good.

[0023] Compared with the prior art, the present invention has the following beneficial effects:

[0024] The present invention provides a lake water quality parameter inversion method, which aims to obtain high-frequency, continuous and large-scale lake high-precision water quality parameter simulation data by driving a water ecological model with low-frequency water quality monitoring data from a small number of monitoring stations, and train a machine learning model based on the simulation data and remote sensing image data to obtain the correspondence between the reflectivity of each pixel band of the remote sensing image and the lake water quality parameters, and then use remote sensing images with high monitoring frequency and wide monitoring range to achieve relatively continuous and large-scale high-frequency spatiotemporal dynamic change monitoring of lake water quality parameters based on the learned correspondence. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 The figure is a flow chart of a lake water quality parameter inversion method. DETAILED DESCRIPTION

[0026] The technical solution of the present invention is further described in detail below in conjunction with specific implementation methods.

[0027] See also Figure 1 As an embodiment of the present invention, a lake water quality parameter inversion method is provided, the method comprising the following steps:

[0028] Step S1: Collect water quality monitoring data from various stations in the lake and hydrological and water quality monitoring data from rivers entering and leaving the lake upstream and downstream;

[0029] Step S2: Using the water ecological model based on the hydrological and water quality monitoring data of the rivers entering and leaving the lake upstream and downstream as boundary conditions, the spatiotemporal dynamic simulation of the lake water quality parameters is performed, and the water quality monitoring data of each station in the lake is used to calibrate and verify the model parameters;

[0030] Step S3: Acquire remote sensing image data corresponding to the lake and determine the center coordinates of each pixel, and extract water quality parameter simulation data at the corresponding coordinate point in the water ecological model based on the center coordinates of the pixel;

[0031] Step S4: extracting the reflectivity of each band of each pixel of the remote sensing image, and matching it with the simulated value of the water quality parameter in the water ecological model that matches the pixel position, forming a model training set, and using a deep learning model based on this training set to learn the relationship between the reflectivity of each band of the pixel and the simulated water quality parameter at the corresponding position of the water ecological model;

[0032] Step S5: Extract the reflectance of each band corresponding to the remote sensing image pixels at the corresponding positions of each station in the lake, input the band reflectance data into the trained machine learning model, obtain the inversion value of water quality parameters at each station, and compare the inversion value of water quality parameters with the corresponding measured value of water quality parameters to analyze the inversion accuracy of water quality parameters.

[0033] In this embodiment, the lake in step S1 is a lake in Jiujiang City, which includes 8 water quality monitoring stations. The water quality monitoring data of each station include water quality parameters such as dissolved oxygen, ammonia nitrogen, and chlorophyll a monitored monthly. The upstream of the lake includes 2 rivers entering the lake, and the hydrological boundary condition data of the inflow into the lake is the flow monitored daily. The downstream of the lake includes 1 river leaving the lake, and the hydrological boundary condition data of the outflow from the lake is the water level monitored daily. The water quality boundary condition data of the rivers entering and leaving the lake are water quality parameters such as dissolved oxygen, ammonia nitrogen, and chlorophyll a monitored monthly.

[0034] In this embodiment, the water ecological model used in the lake in step S2 is the EFDC model. The boundary conditions of hydrological and water quality monitoring data of the upstream flow and water quality monitoring data of the lake and the downstream water level and water quality monitoring data of the rivers entering and leaving the lake, such as the upstream flow and water quality monitoring data of the lake, are input into the EFDC model. The EFDC model can be driven to obtain the spatiotemporal dynamic simulation results of various water quality parameters in the lake on a daily or even hourly basis under the boundary conditions of the monthly scale water quality data. The simulated water quality parameters are dissolved oxygen, ammonia nitrogen and chlorophyll a.

[0035] In this embodiment, when the EFDC model is driven by the hydrological and water quality monitoring data of the rivers flowing in and out of the lake upstream and downstream in step S2 to perform spatiotemporal dynamic simulation of water quality parameters in the lake, the water quality monitoring data of each station in the lake monitored monthly are further used to calibrate and verify the model parameters, wherein the simulated value of the water quality parameter of EFDC at each station is compared with the measured value of the corresponding station, and the Nash efficiency coefficient NSE and the root mean square error RMSE are used to calculate the simulation error of the water quality parameter of EFDC at the measured station, wherein the calculation formulas of NSE and RMSE are expressed as Formula 1 and Formula 2:

[0036] Formula 1: ;

[0037] Formula 2: ;

[0038] In the formula, Indicates the test set i The predicted value of the flood inflow to the reservoir for each model with a step length; Indicates the test set i The measured value of flood inflow in steps of n Indicates the number of steps of the test set, that is, the length of the test set; Represents the average value of all step lengths measured on the test set;

[0039] By continuously adjusting the relevant model parameters of EFDC, the NSE and RMSE at each station can reach a high accuracy. It can be considered that the EFDC model parameter settings are reasonable and can accurately simulate the spatiotemporal dynamic changes of water quality parameters in the entire lake.

[0040] In this embodiment, by continuously adjusting the relevant model parameters of EFDC, the NSE and RMSE of the simulated results and measured results of three water quality parameters of dissolved oxygen, ammonia nitrogen and chlorophyll a at 8 stations in the lake are shown in Table 1:

[0041] Table 1 Simulation accuracy of dissolved oxygen, ammonia nitrogen and chlorophyll a in a lake in Jiujiang City

[0042] Simulated water quality parameters NSE RMSE (mg / L) Dissolved oxygen 0.873 1.86 Ammonia nitrogen 0.776 0.22 Chlorophyll a 0.689 32.27

[0043] In this embodiment, the remote sensing image in step S4 is selected from the Sentinel-2A satellite, wherein the Sentinel-2A satellite has a total of 13 bands, and 10m, 20m and 60m include 4 bands (B2, B3, B4, B8), 6 bands (B5, B6, B7, B8) and 10m, 20m and 60m, respectively. 8A , B 11 , B 12 ) and 3 bands (B1, B8, B9); determine the pixel size to be 60m, that is, each pixel is a square with a side length of 60m, which contains 36 10m resolution B2, B3, B4, B8 bands, 9 20m resolution B5, B6, B7, B9. 8A , B 11 , B 12 Bands. For pixels with 10m and 20m resolution, each band in a 60m pixel corresponds to the average value of all its 10m and 20m resolution pixel bands. After averaging, there are 13 band values ​​in each 60m pixel, of which 4 bands (B2, B3, B4, B8) are the average values ​​of 36 10m resolution pixels, and 6 bands (B5, B6, B7, B8) are the average values ​​of 36 10m resolution pixels. 8A , B 11 , B 12 ) is the average value of 9 20m resolution pixels, and the three bands (B1, B8, B9) are the band values ​​corresponding to 60m resolution pixels.

[0044] In this embodiment, the lake in the study area in step S4 is generalized to be about 3000m long and 1200m wide, containing a total of 1000 pixels with a resolution of 60m. The central coordinates of each pixel are calculated, and then the simulated values ​​of the lake water quality parameters at the positions corresponding to the central coordinates of the 1000 pixels are extracted in the water ecological model. The 13-band reflectance of the 1000 pixels are used as features, and the simulated values ​​of the water quality parameters in the water ecological model at the positions corresponding to the 1000 pixels are used as labels to construct a feature and label training data set with a data set dimension of 1000×16, including 13-band reflectance features of 1000 sample points and corresponding labels of 3 water quality parameters (dissolved oxygen, ammonia nitrogen and chlorophyll a) of 1000 samples.

[0045] In this embodiment, after obtaining the sample data set consisting of 13 band reflectances of 1000 pixels and their combinations and 3 water quality parameter simulation values ​​at their corresponding positions in step S4, the machine learning model XGBoost is used to learn the relationship between the band reflectance characteristics and the water quality parameter labels, that is, to train the XGBoost model.

[0046] In this embodiment, in step S5, the reflectance of each band in the remote sensing image pixel at the corresponding position of the 8 measured stations in the lake is input as a feature into the trained XGBoost model, and the dissolved oxygen, ammonia nitrogen and chlorophyll a inversion results corresponding to the 8 measured stations in the lake are obtained based on the XGBoost model, and the inversion results are compared with the water quality monitoring data of the measured stations, and the water quality parameter inversion accuracy is calculated using NSE. After continuously adjusting the XGBoost model parameters, the inversion accuracy NSE of the three water quality parameters (dissolved oxygen, ammonia nitrogen and chlorophyll a) at the 8 stations is shown in Table 2:

[0047] Table 2 Calculated values ​​of the inversion accuracy NSE of dissolved oxygen, ammonia nitrogen and chlorophyll a at eight stations

[0048] Site 1 Site 2 Site 3 Site 4 Site 5 Site 6 Site 7 Site 8 Dissolved oxygen 0.80 0.79 0.85 0.83 0.77 0.74 0.84 0.80 Ammonia nitrogen 0.77 0.69 0.77 0.68 0.72 0.70 0.67 0.71 Chlorophyll a 0.62 0.63 0.66 0.65 0.64 0.61 0.67 0.66

[0049] It should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, an element defined by the sentence "comprises a ..." does not exclude the existence of other identical elements in the process, method, article or device including the element.

[0050] The above are only preferred embodiments of the present invention, and are not intended to limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made using the contents of the present invention specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A lake water quality parameter inversion method, characterized in that: The method comprises the following steps: Step S1: Collect water quality monitoring data from various stations in the lake and hydrological and water quality monitoring data from rivers entering and leaving the lake upstream and downstream; Step S2: Using the water ecological model based on the hydrological and water quality monitoring data of the rivers entering and leaving the lake upstream and downstream as boundary conditions, the spatiotemporal dynamic simulation of the lake water quality parameters is performed, and the water quality monitoring data of each station in the lake is used to calibrate and verify the model parameters; Step S3: Acquire remote sensing image data corresponding to the lake and determine the center coordinates of each pixel, and extract water quality parameter simulation data at the corresponding coordinate point in the water ecological model based on the center coordinates of the pixel; Step S4: extracting the reflectivity of each band of each pixel of the remote sensing image, and matching it with the simulated value of the water quality parameter in the water ecological model that matches the pixel position, forming a model training set, and using a deep learning model based on this training set to learn the relationship between the reflectivity of each band of the pixel and the simulated water quality parameter at the corresponding position of the water ecological model; Step S5: extracting the reflectance of each band corresponding to the remote sensing image pixel at the corresponding position of each station of the lake, inputting the band reflectance data into the trained machine learning model, obtaining the inversion value of the water quality parameter at each station, and comparing the inversion value of the water quality parameter with the corresponding measured value of the water quality parameter to analyze the inversion accuracy of the water quality parameter; The water ecological model used in the lake in step S2 is the EFDC model. By inputting the boundary conditions into the EFDC model, the EFDC model can be driven to obtain the spatiotemporal dynamic simulation results of various water quality parameters in the lake on a daily basis under the boundary conditions of the monthly scale water quality data. In step S2, when the EFDC model is driven by the hydrological and water quality monitoring data of the rivers entering and leaving the lake upstream and downstream to simulate the spatiotemporal dynamics of the water quality parameters in the lake, the water quality monitoring data of each station in the lake monitored monthly are further used to calibrate and verify the model parameters, wherein the simulated values ​​of the water quality parameters of the EFDC model at each station are compared with the measured values ​​of the corresponding station, and the Nash efficiency coefficient NSE and the root mean square error RMSE are used to calculate the simulation error of the water quality parameters of the EFDC model at the measured station, wherein the calculation formulas of NSE and RMSE are expressed as formula 1 and formula 2: Formula 1: Formula 2: In the formula, y_pred i represents the flood inflow prediction value of each model at the i-th step in the test set; y i represents the measured value of the inflow flood at the i-th step in the test set, and n represents the number of steps in the test set, that is, the length of the test set; Represents the average of all step length measurements on the test set.

2. A lake water quality parameter inversion method according to claim 1, characterized in that: The water quality monitoring data of each station in the lake in step S1 include dissolved oxygen, ammonia nitrogen and chlorophyll a monitored monthly, and the hydrological and water quality monitoring data of the rivers entering and leaving the lake upstream and downstream of the lake include hydrological data of the upstream entering the lake, hydrological data of the downstream leaving the lake and water quality data of the rivers entering and leaving the lake. The hydrological data of the upstream entering the lake are the upstream flow of the lake monitored daily, the hydrological data of the downstream leaving the lake are the downstream water level monitored daily, and the water quality data of the rivers entering and leaving the lake are dissolved oxygen, ammonia nitrogen and chlorophyll a monitored monthly.

3. A lake water quality parameter inversion method according to claim 1, characterized in that: The remote sensing image in step S4 is selected from the Sentinel-2A satellite, where the Sentinel-2A satellite has a total of 13 bands, 10m, 20m and 60m respectively contain 4 bands, 6 bands and 3 bands; the pixel size is determined to be 60m, that is, each pixel is a square with a side length of 60m, which contains 36 10m resolution B2, B3, B4, B8 bands, 9 20m resolution B5, B6, B7, B8 bands 8A , B 11 , B 12 Bands. For pixels with 10m and 20m resolution, each band in a 60m pixel corresponds to the average value of all its 10m and 20m resolution pixel bands. After averaging, there are 13 band values ​​in each 60m pixel, of which 4 bands are the average values ​​of 36 10m resolution pixels, 6 bands are the average values ​​of 9 20m resolution pixels, and 3 bands are the band values ​​corresponding to 60m resolution pixels.

4. A lake water quality parameter inversion method according to claim 3, characterized in that: After extracting each 60m resolution lake remote sensing image pixel in step S4, the center coordinates of each pixel are calculated, and then the simulated values ​​of the lake water quality parameters at the corresponding positions of the center coordinates of each pixel are extracted in the water ecological model. The 13-band reflectance of each pixel is used as a feature, and the simulated value of the water quality parameter in the water ecological model corresponding to each pixel is used as a label to construct a feature and label training data set.

5. A lake water quality parameter inversion method according to claim 4, characterized in that: After obtaining a sample data set consisting of the simulated values ​​of water quality parameters of the water ecological model based on multiple band reflectances of each pixel and its corresponding position in step S4, the relationship between features and labels is learned using the machine learning model XGBoost.

6. A lake water quality parameter inversion method according to claim 5, characterized in that: In step S5, the reflectance of each band in the remote sensing image pixel corresponding to the measured site in the lake is input as a feature into the trained XGBoost model. Based on the XGBoost model, the water quality parameter inversion result corresponding to the measured site in the lake is obtained, and it is compared with the water quality monitoring data of the measured site. The NSE is used to calculate the water quality parameter inversion accuracy. When the water quality parameter inversion accuracy evaluation index NSE is below 0.5, the XGBoost model parameters are further adjusted until NSE reaches the preset accuracy.

Images