阅读说明：本技术 天空地一体化高光谱水质监测分析方法 (Sky-ground integrated hyperspectral water quality monitoring and analyzing method ) 是由 郭锋 赵兴圆 周淑媛 吕薇 洪平 李德为 张延敏 宋勇军 邢璐 夏之雨 于 2019-11-29 设计创作，主要内容包括：本发明提供了一种天空地一体化高光谱水质监测分析方法,包括以下步骤：S1、水质监测数据采集,包括高光谱数据采集和水体水样数据采集；S2、水质监测数据处理,针对采集到的多源水质监测数据进行数据处理；S3、水质参数反演模型建立,根据多源高光谱监测数据建立水质参数反演模型,通过基于人工智能的反演模型对水质参数建模；S4、水质参数反演分析。本发明的有益效果是：既可以满足大范围水质监测的需要,也可以反映水质在空间和时间上的分布和变化情况,弥补了单一采用水面采样的不足,同时还能发现一些常规方法难以揭示的污染源分布以及污染物的迁移特性和影响范围,为科学监测水环境提供依据,具有高动态、低成本和宏观性的显著特点。(The invention provides a sky-ground integrated hyperspectral water quality monitoring and analyzing method, which comprises the following steps: s1, collecting water quality monitoring data, including hyperspectral data collection and water sample data collection; s2, processing water quality monitoring data, and processing the data according to the collected multi-source water quality monitoring data; s3, establishing a water quality parameter inversion model according to multi-source hyperspectral monitoring data, and establishing a water quality parameter model through an inversion model based on artificial intelligence; and S4, performing inversion analysis on the water quality parameters. The invention has the beneficial effects that: the method can meet the requirement of large-range water quality monitoring, can reflect the distribution and change conditions of water quality in space and time, makes up for the defect of single water surface sampling, can find pollution source distribution and pollutant migration characteristics and influence ranges which are difficult to disclose by a conventional method, provides a basis for scientifically monitoring the water environment, and has the remarkable characteristics of high dynamic, low cost and macroscopic property.)

1. A sky-ground integrated hyperspectral water quality monitoring and analyzing method is characterized by comprising the following steps:

s1, collecting water quality monitoring data, including hyperspectral data collection and water sample data collection;

s2, processing water quality monitoring data, and processing the data according to the collected multi-source water quality monitoring data;

s3, establishing a water quality parameter inversion model according to multi-source hyperspectral monitoring data, and establishing a water quality parameter model through an inversion model based on artificial intelligence;

s4, performing water quality parameter inversion analysis, obtaining inversion results of various water quality parameters based on the water quality parameter inversion model, and providing a basis for the distribution situation of the water quality parameters and the analysis of pollution sources by combining spatial information.

2. The sky-ground integrated hyperspectral water quality monitoring and analysis method according to claim 1, characterized in that: in step S1, the water quality hyperspectral data collection includes: the method comprises the steps of collecting satellite hyperspectral remote sensing data, airborne hyperspectral remote sensing data and ground hyperspectral instrument data, wherein the satellite hyperspectral data mainly adopts a domestic high-resolution five-number satellite, and the satellite has the advantages that a visible short wave infrared hyperspectral camera greatly improves performance indexes, image definition is greatly improved, distortion of image spectra is forcibly eliminated, abundant calibration means are configured to ensure high precision and high stability of the data, so that urban water body components are detected accurately, an unmanned aerial vehicle which is independently developed and produced in China is used for airborne hyperspectral data collection, a hyperspectral imager is carried to acquire image data in a test area, a flight mode is acquired through automatic flight routes/manual hovering, and ASD-FR is synchronously used for acquiring standard white boards and ground measured spectral data of various targets; then utilize high spectrum appearance supporting software to carry out the blackboard and rectify, obtain the high spectral radiance image data after rectifying, ground high spectral data adopts the LCTF high spectrum imager of domestic autonomous research and development, compare with traditional grating push away the type imaging spectrometer of sweeping, realize fast spectrum through automatically controlled and in succession or be interrupted the tune, do not have the process of sweeping, the light path is simple, have characteristics such as small, light in weight, easily installation and area, scanning speed is fast, the wavelength can freely be selected, can use with tripod, operation panel cooperation, also can carry on the small-size machine carries the platform.

3. The sky-ground integrated hyperspectral water quality monitoring and analysis method according to claim 1, characterized in that: in step S1, the water sample data collection includes: selecting a proper sampling point in an experimental area to collect water sample data, sampling in an area with large river water quality change, putting the collected sample in a brown bottle, sealing, freezing and storing, sending to a laboratory for measurement within 5-8 hours, and taking pictures of the water body and the periphery of the sampling point, wherein the field spectrum is required to be synchronously measured during sampling.

4. A sky-ground integrated hyperspectral water quality monitoring and analysis method according to claim 1, wherein step S2 comprises the following substeps:

s21, processing the satellite hyperspectral data;

preprocessing satellite remote sensing images on satellite hyperspectral data, wherein the preprocessing comprises radiation correction, atmospheric correction, geometric correction and spectral feature analysis, the atmospheric correction is a key problem of water quality inversion, and the premise of accurately acquiring water color element remote sensing information is to perform atmospheric correction and remove the influence of atmospheric molecules and aerosol to obtain corrected satellite hyperspectral data;

(2) processing airborne hyperspectral data;

the airborne hyperspectral data processing mainly comprises wave band classification, single-wave-band image data embedding, wave band registration and standard white board correction, and when the flying mode is hovering, only the wave band registration and the standard white board correction are needed; firstly, performing band classification on hyperspectral radiant quantity image data subjected to blackboard correction preprocessing, then performing embedding processing on each single-band image, and then performing band registration on the embedded images; finally, carrying out white board correction by utilizing ground actual measurement standard white board reflectivity data to finally obtain RGB pseudo color synthetic reflectivity image data;

(3) processing ground imaging spectrum data;

the method comprises the steps of conducting file name standardization processing on data collected by a ground imaging spectrometer, simultaneously conducting screening on original data of the spectrometer through unqualified curves and data elimination, then utilizing a reflectivity calculation equation to calculate target reflectivity, conducting arithmetic averaging on multiple pieces of spectral data of the same target, and finally conducting smoothing processing on the data through smoothing filtering.

5. The sky-ground integrated hyperspectral water quality monitoring and analysis method according to claim 1, characterized in that: in step S3, a random forest algorithm model is used, and based on N times of replaced random sampling, data with the same number as that of original samples are extracted to generate a plurality of training sets, i.e., a boottrap algorithm, and the basic classification unit of the random forest is a decision tree { h (X, θ)_k) K1, 2.. K }, the algorithm is a classifier comprising a plurality of decision trees, and the output class of the algorithm is determined by the mode of the output class of the decision trees.

6. A sky-ground integrated hyperspectral water quality monitoring and analysis method according to claim 5, characterized in that the detailed steps of establishing a random forest algorithm model are as follows:

the first step is as follows: resampling an original data sample set X by using a Bootstrap algorithm, and randomly generating K training sample sets X1, X2 and X3 … … … Xk;

the second step is that: generating corresponding decision trees T1, T2.. Tk by using each generated training set, and selecting the attribute of the optimal splitting mode in mtry attributes on each intermediate node as the splitting attribute of the current node to classify on the node;

the third step: each decision tree grows completely;

the fourth step: testing and classifying the original data sample set X by each decision tree;

the fifth step: and (3) taking the category which is most output by the K decision trees as the category of the original data sample set X by adopting a voting mode, and finally classifying the decision into the following formula:

in the above formula: h (x) represents a classification combination model, h_iIs a single decision tree classification model, I (-) is an illustrative function, and Y represents a target variable or an output variable;

and a sixth step: and evaluating the water quality according to the trained random forest evaluation model.

7. The sky-ground integrated hyperspectral water quality monitoring and analysis method according to claim 1, characterized in that: and step S4 comprises sensor channel conversion, model precision evaluation, space mapping based on remote sensing water quality parameters and space analysis of pollution sources.

8. The sky-ground integrated hyperspectral water quality monitoring and analysis method according to claim 7, characterized in that: the sensor channel switching comprises: the reflectivity actually measured in the field is collected to the corresponding sensor channel through convolution operation, and the specific calculation formula is as follows:

in the above formula, λ is the wavelength, λ min is the starting wavelength of the channel, λ max is the ending wavelength of the channel, r (λ) is the surface reflectivity corresponding to the λ wavelength, and f (λ) is the spectral response function.

9. The sky-ground integrated hyperspectral water quality monitoring and analysis method according to claim 8, characterized in that: the model precision evaluation comprises the following steps: randomly extracting 1/3 ground actual measurement data of not less than 10 point positions, and evaluating the precision of a water quality parameter inversion model and the precision of a sensor water quality parameter space mapping by adopting a Root Mean Square Error (RMSE), an average relative error (MRE) and an average absolute error (MAE);

RMSE is calculated as follows:

RMSE＝[∑(measurement_i-predicted_i)²/(N)]^0.5

middle measure of the formula_iPredicted for measured value_iIs a model estimate, N is a degree of freedom;

the accuracy of the model is evaluated by the MRE, and the calculation formula is as follows:

MRE＝[100×|measurement-predicted|/measurement]％

the accuracy of the MAE evaluation model is calculated according to the following formula:

10. a sky-ground integrated hyperspectral water quality monitoring analysis method according to claim 9, characterized in that: the space mapping based on the remote sensing water quality parameters and the space analysis of the pollution source comprise the following steps: the constructed water quality parameter model is used for data of satellite, airborne and ground hyperspectral sensors to obtain inversion results of TSS, Chla and CDOM parameters, water quality parameter space mapping is carried out, quantitative analysis is carried out on the water quality parameter space mapping, and meanwhile, the main cause of water quality pollution can be analyzed by combining the land utilization condition and the water quality pollution condition.

Technical Field

The invention relates to a water quality monitoring and analyzing method, in particular to a sky-ground integrated hyperspectral water quality monitoring and analyzing method.

Background

The increasingly worsening of urban water quality and the shortage of fresh water resources in China have led to high attention of the nation and the society. The water body pollution phenomenon is more and more common in urban riverways, becomes one of the common problems in urban water environment, and seriously affects urban image, ecological environment and physical and psychological health of citizens. The water quality of inland water affects the production and life of people, so accurate and efficient water quality monitoring is particularly important, and the water quality monitoring is the main basis of water quality evaluation and water pollution prevention and treatment, so the significance is more and more significant.

The conventional water environment monitoring method is to collect water samples from a water body to be monitored and send the water samples to a laboratory for water quality analysis or establish water quality monitoring network points, and although the methods can accurately analyze and evaluate a plurality of water quality parameters, the methods are time-consuming, labor-consuming and uneconomical, the quantity of water sample collection and analysis is limited, and for the whole water body, the data of the measuring points only have local and typical representative meanings and are not enough to reflect the time-space change of water pollution.

Disclosure of Invention

In order to solve the problems in the prior art, the invention provides a sky-ground integrated hyperspectral water quality monitoring and analyzing method.

The invention provides a sky-ground integrated hyperspectral water quality monitoring and analyzing method, which comprises the following steps:

s1, collecting water quality monitoring data, including hyperspectral data collection and water sample data collection;

s2, processing water quality monitoring data, and processing the data according to the collected multi-source water quality monitoring data;

s3, establishing a water quality parameter inversion model according to multi-source hyperspectral monitoring data, and establishing a water quality parameter model through an inversion model based on artificial intelligence;

s4, performing water quality parameter inversion analysis, obtaining inversion results of various water quality parameters based on the water quality parameter inversion model, and providing a basis for the distribution situation of the water quality parameters and the analysis of pollution sources by combining spatial information.

As a further improvement of the present invention, in step S1, the water quality hyperspectral data collection includes: the method comprises the steps of collecting satellite hyperspectral remote sensing data, airborne hyperspectral remote sensing data and ground hyperspectral instrument data, wherein the satellite hyperspectral data mainly adopts a domestic high-resolution five-number satellite, and the satellite has the advantages that a visible short wave infrared hyperspectral camera greatly improves performance indexes, image definition is greatly improved, distortion of image spectra is forcibly eliminated, abundant calibration means are configured to ensure high precision and high stability of the data, so that urban water body components are detected accurately, an unmanned aerial vehicle which is independently developed and produced in China is used for airborne hyperspectral data collection, a hyperspectral imager is carried to acquire image data in a test area, a flight mode is acquired through automatic flight routes/manual hovering, and ASD-FR is synchronously used for acquiring standard white boards and ground measured spectral data of various targets; then utilize high spectrum appearance supporting software to carry out the blackboard and rectify, obtain the high spectral radiance image data after rectifying, ground high spectral data adopts the LCTF high spectrum imager of domestic autonomous research and development, compare with traditional grating push away the type imaging spectrometer of sweeping, realize fast spectrum through automatically controlled and in succession or be interrupted the tune, do not have the process of sweeping, the light path is simple, have characteristics such as small, light in weight, easily installation and area, scanning speed is fast, the wavelength can freely be selected, can use with tripod, operation panel cooperation, also can carry on the small-size machine carries the platform.

As a further improvement of the present invention, in step S1, the water sample data collection includes: selecting a proper sampling point in an experimental area to collect water sample data, sampling in an area with large river water quality change, putting the collected sample in a brown bottle, sealing, freezing and storing, sending to a laboratory for measurement within 5-8 hours, and taking pictures of the water body and the periphery of the sampling point, wherein the field spectrum is required to be synchronously measured during sampling.

As a further improvement of the present invention, step S2 includes the following sub-steps:

s21, processing the satellite hyperspectral data;

preprocessing satellite remote sensing images on satellite hyperspectral data, wherein the preprocessing comprises radiation correction, atmospheric correction, geometric correction and spectral feature analysis, the atmospheric correction is a key problem of water quality inversion, and the premise of accurately acquiring water color element remote sensing information is to perform atmospheric correction and remove the influence of atmospheric molecules and aerosol to obtain corrected satellite hyperspectral data;

(2) processing airborne hyperspectral data;

the airborne hyperspectral data processing mainly comprises wave band classification, single-wave-band image data embedding, wave band registration and standard white board correction, and when the flying mode is hovering, only the wave band registration and the standard white board correction are needed; firstly, performing band classification on hyperspectral radiant quantity image data subjected to blackboard correction preprocessing, then performing embedding processing on each single-band image, and then performing band registration on the embedded images; finally, carrying out white board correction by utilizing ground actual measurement standard white board reflectivity data to finally obtain RGB pseudo color synthetic reflectivity image data;

(3) processing ground imaging spectrum data;

the method comprises the steps of conducting file name standardization processing on data collected by a ground imaging spectrometer, simultaneously conducting screening on original data of the spectrometer through unqualified curves and data elimination, then utilizing a reflectivity calculation equation to calculate target reflectivity, conducting arithmetic averaging on multiple pieces of spectral data of the same target, and finally conducting smoothing processing on the data through smoothing filtering.

As a further improvement of the invention, in step S3, a random forest algorithm model is adopted, and data with the same number as that of original samples are extracted after N times of replaced random sampling to generate a plurality of training sets, namely Bootstrap algorithm, wherein the basic classification unit of the random forest is a decision tree { h (X, theta)_k) K1, 2.. K }, the algorithm is a classifier comprising a plurality of decision trees, and the output class of the algorithm is determined by the mode of the output class of the decision trees.

As a further improvement of the invention, the detailed steps of establishing the random forest algorithm model are as follows:

the first step is as follows: resampling an original data sample set X by using a Bootstrap algorithm, and randomly generating K training sample sets X1, X2 and X3 … … … Xk;

the second step is that: generating corresponding decision trees T1, T2.. Tk by using each generated training set, and selecting the attribute of the optimal splitting mode in mtry attributes on each intermediate node as the splitting attribute of the current node to classify on the node;

the third step: each decision tree grows completely;

the fourth step: testing and classifying the original data sample set X by each decision tree;

the fifth step: and (3) taking the category which is most output by the K decision trees as the category of the original data sample set X by adopting a voting mode, and finally classifying the decision into the following formula:

in the above formula: h (x) represents a classification combination model, h_iIs a single decision tree classification model, I (-) is an illustrative function, and Y represents a target variable or an output variable;

and a sixth step: and evaluating the water quality according to the trained random forest evaluation model.

As a further improvement of the invention, step S4 includes sensor channel conversion, model precision evaluation, spatial mapping based on remote sensing water quality parameters and spatial analysis of pollution sources.

As a further improvement of the present invention, the sensor channel switching comprises: the reflectivity actually measured in the field is collected to the corresponding sensor channel through convolution operation, and the specific calculation formula is as follows:

in the above formula, λ is the wavelength, λ min is the starting wavelength of the channel, λ max is the ending wavelength of the channel,

r (λ) is the surface reflectivity at the corresponding λ wavelength, and f (λ) is the spectral response function.

As a further improvement of the present invention, the model accuracy evaluation includes: randomly extracting 1/3 ground actual measurement data of not less than 10 point positions, and evaluating the precision of a water quality parameter inversion model and the precision of a sensor water quality parameter space mapping by adopting a Root Mean Square Error (RMSE), an average relative error (MRE) and an average absolute error (MAE);

RMSE is calculated as follows:

RMSE＝[∑(measurement_i-predicted_i)²/(N)]^0.5

middle measure of the formula_iPredicted for measured value_iIs a model estimate, N is a degree of freedom;

the accuracy of the model is evaluated by the MRE, and the calculation formula is as follows:

MRE＝[100×|measurement-predicted|/measurement]％

the accuracy of the MAE evaluation model is calculated according to the following formula:

as a further improvement of the invention, the spatial mapping based on the remote sensing water quality parameters and the spatial analysis of the pollution source comprise: the constructed water quality parameter model is used for data of satellite, airborne and ground hyperspectral sensors to obtain inversion results of TSS, Chla and CDOM parameters, water quality parameter space mapping is carried out, quantitative analysis is carried out on the water quality parameter space mapping, and meanwhile, the main cause of water quality pollution can be analyzed by combining the land utilization condition and the water quality pollution condition.

The invention has the beneficial effects that: by the scheme, the requirement of large-range water quality monitoring can be met, the distribution and change conditions of water quality in space and time can be reflected, the defect of single water surface sampling is overcome, pollution source distribution and the migration characteristic and influence range of pollutants which are difficult to disclose by a conventional method can be found, a basis is provided for scientifically monitoring the water environment, and the method has the remarkable characteristics of high dynamic, low cost, macroscopic property and the like.

Detailed Description

The present invention will be further described with reference to the following embodiments.

A sky-ground integrated hyperspectral water quality monitoring and analyzing method mainly comprises the steps of water quality monitoring data acquisition, water quality monitoring data processing, water quality parameter inversion model establishment and water quality parameter inversion analysis.

The following is a detailed description:

1. water quality monitoring data acquisition

The data acquisition comprises hyperspectral data acquisition and water sample data acquisition.

(1) Collecting water hyperspectral data: the water quality hyperspectral data acquisition covers an integrated acquisition means of the sky, the air and the ground, and comprises the acquisition of satellite hyperspectral remote sensing data, airborne hyperspectral remote sensing data and ground hyperspectral instrument data. The satellite hyperspectral data mainly adopts a domestic high-resolution five-number satellite, and the satellite has the advantages that the visible short-wave infrared hyperspectral camera greatly improves the image definition while improving the performance index, focuses on eliminating the distortion of an image spectrum, and is provided with abundant calibration means to ensure the high precision and high stability of the data, so that the detection of urban water body components is very accurate. The airborne hyperspectral data acquisition utilizes a domestic independently-developed unmanned aerial vehicle to carry a hyperspectral imager to acquire image data in a test area, the flight mode is acquired by automatic flight route/manual hovering, and ASD-FR is synchronously used to acquire standard white boards and ground actual measurement spectral data of each target; and then, carrying out blackboard correction by using software matched with the hyperspectral meter to obtain corrected hyperspectral radiant quantity image data. The ground hyperspectral data adopts a domestic LCTF hyperspectral imager which is independently researched and developed, and compared with a traditional grating push-broom type imaging spectrometer, the fast spectrum continuous or discontinuous tuning is realized through electric control, the push-broom process is avoided, the light path is simple, the device has the characteristics of small size, light weight, easiness in installation and belt, high scanning speed, free wavelength selection and the like, can be matched with a tripod and an operation platform for use, and can also be carried on a small airborne platform.

(2) Collecting water sample data: selecting proper sampling points in an experimental area to collect water sample data, sampling in an area with large river water quality change (collecting one point position about 1-2 km), sealing and freezing the collected sample in a brown bottle for storage, and sending the sample to a laboratory for measurement within 5-8 hours. Sampling requires synchronous measurement of field spectrum, and photographs of water bodies and surrounding areas at sampling points are taken.

2. Water quality monitoring data processing

The method is used for processing the collected multi-source water quality monitoring data and specifically comprises the following steps:

(1) satellite hyperspectral data processing

The method comprises the steps of performing common preprocessing of satellite remote sensing images such as radiation correction, atmospheric correction, geometric correction, spectral feature analysis and the like on satellite hyperspectral data, wherein the atmospheric correction is a key problem of water quality inversion, and the premise of accurately acquiring water color element remote sensing information is to perform atmospheric correction, remove the influence of atmospheric molecules and aerosol and obtain corrected satellite hyperspectral data.

(2) Airborne hyperspectral data processing

The onboard hyperspectral data processing mainly comprises wave band classification, single-wave-band image data embedding, wave band registration and standard white board correction (when the flying mode is hovering, only wave band registration and standard white board correction are needed). Firstly, performing band classification on hyperspectral radiant quantity image data subjected to blackboard correction preprocessing, then performing embedding processing on each single-band image, and then performing band registration on the embedded images; and finally, carrying out white board correction by utilizing the ground actual measurement standard white board reflectivity data to finally obtain RGB pseudo color synthetic reflectivity image data.

(3) Ground imaging spectral data processing

The method comprises the steps of conducting file name standardization processing on data collected by a ground imaging spectrometer, simultaneously conducting screening on original data of the spectrometer through unqualified curves and data elimination, then utilizing a reflectivity calculation equation to calculate target reflectivity, conducting arithmetic averaging on multiple pieces of spectral data of the same target, and finally conducting smoothing processing on the data through smoothing filtering.

3. Water quality parameter inversion model establishment

And (2) establishing a water quality parameter (including TSS, Chla and CDOM) inversion model according to the multi-source hyperspectral monitoring data, and modeling the water quality parameter through the inversion model based on artificial intelligence. Here, a random forest algorithm model is used, and a plurality of training sets are generated based on N times of replaced random sampling and then data with the same number as that of original samples are extracted (Bootstrap algorithm). The basic classification unit of the random forest is a decision tree { h (X, theta)_k) K1, 2.. K }, the algorithm is a classifier comprising a plurality of decision trees, and the output class of the algorithm is determined by the mode of the output class of the decision trees. The following steps are detailed for establishing the random forest model:

the first step is as follows: original data sample set X is resampled by using a Bootstrap method, and K training sample sets X1, X2 and X3 … … … Xk are randomly generated.

The second step is that: generating corresponding decision trees T1, T2.. Tk by using each generated training set, and selecting the attribute of the optimal splitting mode in mtry attributes on each intermediate node as the splitting attribute of the current node to classify on the node;

the third step: each decision tree grows in its entirety.

The fourth step: testing and classifying the original data sample set X by each decision tree;

the fifth step: and adopting a voting mode to take the category with the most output of the K decision trees as the category of the original data sample set X. The final decision is classified as follows:

in the above formula: h (x) represents a classification combination model, h_iIs a single decision tree classification model, I (-) is an illustrative function, and Y represents a target variable or an output variable.

And a sixth step: and evaluating the water quality according to the trained random forest evaluation model.

4. Water quality parameter inversion analysis

Inversion results of various water quality parameters can be obtained based on the water quality parameter inversion model, and a basis is provided for analysis of distribution conditions and pollution sources of the water quality parameters by combining spatial information. The method comprises the specific steps of sensor channel conversion, model precision evaluation, space mapping based on remote sensing water quality parameters and space analysis of a pollution source.

(1) Sensor channel switching

Because the spectral resolution of data acquired on the ground is 1nm and is inconsistent with the spectral resolution of each channel of a remote sensing image acquired by a space sensor, in order to keep the consistency of the spectral resolution of the ground spectrum and the remote sensing image and the applicability of an inversion model constructed on the basis of the ground to a high-altitude sensor, before the model is corrected and verified, the reflectivity actually measured in the field is acquired to the corresponding sensor channel through convolution operation. The specific calculation formula is as follows:

in the above formula, λ is the wavelength, λ min is the starting wavelength of the channel, λ max is the ending wavelength of the channel, r (λ) is the surface reflectivity corresponding to the λ wavelength, and f (λ) is the spectral response function.

(2) Model accuracy evaluation

And randomly extracting 1/3 ground actual measurement data (no less than 10 point positions) and evaluating the accuracy of the water quality parameter inversion model and the accuracy of the sensor water quality parameter space mapping by using Root Mean Square Error (RMSE), average relative error (MRE) and average absolute error (MAE).

RMSE is calculated as follows:

RMSE＝[∑(measurement_i-predicted_i)²/(N)]^0.5

measurement in formula 2_iMeasured value, predicted_iFor model estimation, N is the degree of freedom.

The accuracy of the model is evaluated by the MRE, and the calculation formula is as follows:

MRE＝[100×|measurement-predicted|/measurement]％

the accuracy of the MAE evaluation model is calculated according to the following formula:

(3) water quality space mapping analysis based on remote sensing

The constructed water quality parameter model is used for data of satellite, airborne and ground hyperspectral sensors to obtain inversion results of TSS, Chla and CDOM parameters, water quality parameter space mapping is carried out, quantitative analysis is carried out on the water quality parameter space mapping, and meanwhile, the main cause of water quality pollution can be analyzed by combining the land utilization condition and the water quality pollution condition.

The spectral resolution of the hyperspectrum is less than 10nm, a continuous spectrum curve can be generated, water body spectrum curves of different water qualities are carved in detail, and the hyperspectral remote sensing technology is expected to solve the problem of urban water quality parameter inversion. With the development of domestic remote sensing, remote sensing and hyperspectral technologies in recent years, mature earth observation image series products are formed, compared with imported data sources, domestic data has a national industry supporting policy, the cost performance is better, and the data construction cost can be greatly reduced.

The invention provides a scheme for establishing a space-air-ground integrated hyperspectral water quality monitoring analysis based on the most advanced domestic hyperspectral remote sensing technology and a high-precision business water environment remote sensing inversion analysis algorithm, analyzes the test result of a river water quality water sample, selects key parameters such as suspended matters (TSS), chlorophyll (Chla), soluble organic matters (CDOM) and other data, establishes a remote sensing inversion model, performs space distribution mapping and quantitative analysis of water environment indexes in a large area range by utilizing space-ground-air integrated hyperspectral water quality information, overcomes the defects and difficulties of large space interval, time and labor consumption during water surface sampling observation to a certain extent, can discover the characteristics of pollutant discharge sources, migration diffusion directions, influence ranges and the like which are difficult to reveal by some conventional methods, is favorable for finding the coming and going veins of pollutants, and provides basis for scientifically laying ground monitoring points, provides necessary reference for effective treatment of rivers.

The invention provides a sky-ground integrated hyperspectral water quality monitoring and analyzing method, which is used for monitoring and analyzing water quality parameters such as chlorophyll, suspended matters and CDOM. The technology has the remarkable characteristics of high dynamic, low cost, macroscopic property and the like, and has the advantage that conventional monitoring can not be replaced in the aspect of river and lake water quality monitoring research. The method can meet the requirement of large-range water quality monitoring, can reflect the distribution and change conditions of water quality in space and time, makes up for the defect of single water surface sampling, can find pollution source distribution and pollutant migration characteristics and influence ranges which are difficult to disclose by some conventional methods, and provides a basis for scientifically monitoring the water environment.

The sky-ground integrated hyperspectral water quality monitoring and analyzing method integrates sky-ground acquisition hardware equipment such as satellite hyperspectral remote sensing, airborne hyperspectral remote sensing, ground hyperspectral remote sensing and the like, combines the traditional Internet of things with spatial information technology in a sky-ground integrated water quality three-dimensional sensing mode, enables the sensing to be more comprehensive and the mode to be more flexible, realizes the comprehensive sensing of a point-line surface body, and can quickly acquire the full coverage of the water quality parameter conditions of each point and each area of a fixed water area.

The foregoing is a more detailed description of the invention in connection with specific preferred embodiments and it is not intended that the invention be limited to these specific details. For those skilled in the art to which the invention pertains, several simple deductions or substitutions can be made without departing from the spirit of the invention, and all shall be considered as belonging to the protection scope of the invention.