RU2750853C1 - Method for determining the boundaries of water bodies and areas of distribution of semi-aquatic species based on multispectral data of remote sensing of the earth - Google Patents

Abstract

FIELD: environmental studies.SUBSTANCE: invention relates to the field of environmental studies and concerns a method for identifying the boundaries of water bodies and areas of distribution of semi-aquatic species based on multispectral data of remote sensing of the Earth. The method includes radiometric calibration of the image of the earth’s surface with the reduction of brightness values to energy brightness, atmospheric correction with the translation of pixel values from energy brightness to reflectivity coefficients from 0 to 1. The selection of the land surface is carried out by selecting the pixels of the image for which the reflection coefficient value exceeds 0.15 in the near infrared region. The selection of a clean water surface is carried out by selecting the pixels of the image for which the parameter determined by the ratio of the reflectivity coefficientsis greater than 1.EFFECT: increased efficiency of determining the boundaries of water bodies and the allocation of zones of distribution of semi-aquatic species.1 cl, 10 dwg

Description

The invention relates to methods for determining the location of the coastline, the boundary of a water body and zones of distribution of air-water vegetation.

For cadastral registration, as well as for the study of water bodies, especially coastal ecotone zones, there is a need to accurately identify the actual boundaries of the coastline. Ecotone zone - the zone between the present and the water is often occupied by aquatic vegetation, which complicates the selection of the boundaries of objects. At the same time, air-water vegetation often extends hundreds of meters from the coastline. Aquatic vegetation plays a significant role in the regulation of the processes inhabited by it. It protects coastlines from erosion, traps suspended solids, and emits large amounts of carbon and nitrogen. Identifying the areas of distribution of aquatic vegetation is a rather important task for assessing the state, dynamics of development and determining the stability of ecosystems of water bodies.

There are three most common methods for determining the boundary of a water body and zones of distribution of air-water vegetation using multispectral data of remote sensing of the Earth: visual interpretation, image clustering, classification based on index images.

1. Visual interpretation of space images by the operator. This is the simplest method that does not require specialized software and operator qualifications. The disadvantages of this method are: low accuracy of the results and high labor intensity with large sizes of objects.

2. Clustering a snapshot using specialized algorithms. There are two known clustering algorithms: "without training", for example, ISODATA, and "with training", for example, methods: minimum distance, maximum likelihood, spectral correlation SAM, etc. The essence of the methods consists in dividing points of a two-dimensional space into subsets - clusters connected a certain property. The results of the implementation of the known methods are thematic images, the classes of which are correlated with groups of objects on the earth's surface. The known methods allow obtaining results with a high degree of automation of their algorithms.

The disadvantages of the known methods are: uncontrollability of results when using classification without training, as well as the choice of suitable control polygons when using clustering with training.

3. Classification based on index images obtained by processing multispectral satellite images. The most popular known classification methods based on index images are: normalized vegetation index (NDVI); soil vegetation index (SAVI); normalized differential water index (NDWI), etc. In the known methods, the classification of objects is based on the pixel values of the index images of the objects. The known methods are quite convenient and are the most common. In addition, some multispectral Earth remote sensing data are supplied with calculated NDVI values. The disadvantages of the known methods are: the inability to control the classification, as well as inefficiency in the allocation of ecotone zones between the water surface and land.

The closest in technical essence analogue to the claimed invention is a method for assessing the level of pollution of water areas according to hyperspectral aerospace sensing data RU 2616716, publ. 04/18/2017 [1], which consists in the sequential implementation of the following operations:

- calculation of signs of water surface clusters related to thick films of oil products with a film thickness of more than 0.2 mm and a coarse suspension with a concentration of more than 200 mg / l;

- exclusion of areas of pure water by the integral reflection index in the entire analyzed range of the spectrum;

- the use of a semi-analytical approach to calculate the concentration of suspended minerals and chlorophyll "a" from regression dependences in spectral bands corresponding to the maximum variation in brightness values under the influence of suspension and absorption of radiation by chlorophyll;

- the formation of a feature space for the separation of thin films of oil products from areas of intensive distribution of chlorophyll "a" and segmentation of areas of suspended matter against the background of shallow water;

- classification of the water surface in the created feature space.

The method consists in determining the contours and parameters of pollution by the reflective characteristics of the water surface, characterized in that the calculation of features is carried out simultaneously in spectral channels corresponding to the maximum value of backscattering by suspended particles, absorption bands of organic impurities in the form of phytoplankton, intervals close to the maximum excitation of luminescent luminescence oil fractions in the short-wave part of the visible range of the spectrum, and having a width from several to tens of nanometers.

The disadvantages of the closest analogue are:

- limited use, due to the limited availability of hyperspectral images with a spectral resolution of at least 5-10 nm, especially in online mode;

- implementation of the method in the form of a software package with a high degree of automation of the process of thematic processing of hyperspectral data, which, in conditions of insufficient operator qualifications or poor control of the information received, creates conditions for obtaining false results,

- lack of consideration of the influence of aquatic vegetation as a factor in the heterogeneity of the water surface.

Method description

The aim of the invention is to improve the efficiency of determining the boundaries of water bodies and the allocation of zones of distribution of air-water vegetation.

The inventive method is illustrated in FIG. 1-10.

FIG. 1. Typical reflection coefficients of the water surface and land (according to the Landsat image).

FIG. 2. An example of calculating the boundaries of a reservoir based on dividing the pixels of the IR channel by the threshold value of the reflection coefficients. a) profile of the coastal zone in the Landsat image in the IR channel; b) a graph of the values of the reflection coefficients of the coastal zone in the IR channel; c) an example of the distribution of pixel values in the coastal zone in the IR channel.

FIG. 3. Reflection coefficients of water surface and aquatic vegetation.

FIG. 4. Optical parameters of the atmospheric model for summer conditions at mid-latitudes. a) The transmittance of electromagnetic radiation; b) the luminosity of the atmosphere itself

FIG. 5. Boundaries of water bodies, highlighted by Landsat image

FIG. 6. An example of identifying areas of distribution of aerial-aquatic vegetation based on the Landsat-8 satellite image (July 4, 2015, Volzhsky reach, area of the village of Staroe Melkovo).

FIG. 7. Algorithm for determining the distribution zones of air-water vegetation according to the Landsat-8 sensor data.

FIG. 8. Spectral curves in the form of "raw values" brightness (DN).

FIG. 9. Spectral curves after radiometric calibration in the form of radiance values.

FIG. 10. Spectral curves after atmospheric correction in the form of reflection coefficients.

The inventive method is based on the use of multispectral data of remote sensing of the Earth, namely multispectral images of the Landsat-8 satellite. This is free data. They are placed in the public domain on the servers of the US Geological Survey. Also, the inventive method can be used for processing data of remote sensing of the Earth from other sensors producing images of the Earth's surface in the red and near-IR spectral regions.

It is known that water absorbs electromagnetic radiation in the infrared range, therefore, this part of the spectrum is chosen to determine the boundary between land and water bodies, which are not clearly distinguishable in visible light. FIG. 1. The best contrasting of the water surface and land is observed when comparing their reflection coefficients in the near infrared range. So for multispectral images of the Landsat-8 satellite, the 5th channel 0.845-0.885 microns is selected. And to determine the boundaries (create a mask) of the reservoir, it is proposed to use the division of the image pixels according to the reflectance value in the near infrared range into groups corresponding to water and land.

The calculation of the threshold value between groups of pixels is based on the profiles of the transition zone. To calculate the threshold value, using a GIS editor, several near-infrared profiles of the coastal zone are plotted, containing the coastal and water surfaces. FIG. 2a. The use of multiple profiles allows you to increase the representativeness of the data obtained. Even more informative for determining the land-water threshold is to use the distribution of the pixel reflectivity values for the coastal polygonal portion of FIG. 2c. This approach is laborious, therefore, in the claimed method, it is used as an additional one. FIG. 2b and FIG. 2c, we observe that the pixels of the satellite image corresponding to the water surface in the near infrared range have a low reflectivity, while the pixels corresponding to the land are characterized by a higher reflectivity. It has been experimentally established that in most cases, for the European part of Russia, the threshold value of the reflection coefficient in the near infrared range between the pixels of satellite images of the corresponding land and water surface is about 0.15 in the near infrared range. In the need to use the proposed method for other territories, the threshold value can be adjusted to correspond to local conditions, but it should be borne in mind that it should be in the range from 0.1 to 0.2. Since the value of the reflection coefficient of a satellite image in the near-infrared range is below 0.1, it unambiguously relates it to a clean water surface, and a value above 0.2 - to land. In this case, pixels having a reflectivity in the near-infrared range less than the threshold value are interpreted as a surface occupied by water and water by vegetation.

Air-aquatic vegetation, as a rule, is located in the coastal shallow-water zone, and the task of determining it, according to satellite imagery data, is reduced to the separation of adjacent pixels between a clean water surface and land. On the basis of satellite images and field surveys, the individual spectral characteristics of the zones of development of air-water vegetation have been revealed. The identification of these zones in inland water bodies is complicated by the large amount of impurities in the waters, which makes the reflection spectra of water masses and the reflection of aquatic vegetation similar. Studies have shown that the presence of aquatic vegetation manifests itself in an increase in the reflectivity of the pixel of the satellite image in the visible and near infrared ranges. 3. For multispectral images of the Landsat-8 satellite, the most significant increase in the reflectivity of aerial-aquatic vegetation areas is observed in channel 5 at 845-885 nm.

But at the same time, an increase in the reflectivity in inland waters, relative to waters without a significant amount of impurities, may be due to an increase in the concentration of chlorophyll and / or the turbidity index. 3.

It has been experimentally established that the most characteristic feature of the zones of development of air-water vegetation is the excess of the reflectivity of the satellite image pixel in the near infrared range - channel 5 of multispectral images of the Landsat-8 satellite 845-885 nm over the reflectivity values in the red part of the visible range - 4 channel of multispectral images of the Landsat-8 satellite 630-680 nm. Therefore, to identify the areas of distribution of air-water vegetation, it is effective to use the estimate of the values of the pixels in the images in the red part of the spectrum relative to the values for them in the near-IR range. To distinguish the areas of aerial-aquatic vegetation, it is possible to estimate the 630 - 885 nm portion of the spectrum. The selection of a clean water surface is proposed to be carried out by means of the selection of image pixels for which the parameter is determined by the ratio of the reflectance coefficients (k), which is calculated by the following formula:

if k <1, then the pixel values refer to air-water vegetation, if k> 1, then the pixel values refer to the water surface not covered with air-water vegetation.

An example of the implementation of the method

Determination of the boundaries of the water surface of the Ivankovskoye reservoir was carried out using its multispectral image from the Landsat satellite. The data received from the satellite is recorded in the form of "raw values" of brightness (DN - Digital Number). These are relative brightness values - spectral curves in the form of DN are shown in FIG. 8. It is noteworthy that the application of the proposed method is sufficient to use any multispectral images with red and near infrared channels. It should be noted that the applied spectral zones are relatively weakly prone to errors associated with the influence of the atmosphere and organic substances dissolved in water, which is favorable for obtaining results of high quality and reliability.

At the first stage, the images are subjected to radiometric calibration. The task of radiometric calibration is to convert DN values into spectral radiance values at the upper boundary of the atmosphere, so the following formula is used for Landsat-8:

(2)

(2)

where, L _λ - energy brightness for the spectral zone;

M _L - calibration factor of scaling of a specific channel (in metadata - RADIANCE_MULT_BAND);

And _L is the additional scaling factor of a specific channel, this is a calibration constant that corresponds to the minimum value of the recorded brightness (in the metadata - RADIANCE_ADD_BAND),

Q _cal - calibrated value (DN) [2].

Spectral curves as radiance values are shown in FIG. nine.

Next, atmospheric correction is carried out based on the MODTRAN atmosphere model. The MODTRAN model allows you to set several standard atmospheres. When processing images of the Ivankovskoye reservoir, an atmospheric model for mid-latitude summer (Mid-Latitude Summer) was used, which is based on the following average parameters: the thickness of the water vapor layer is 3636 atm. cm, while the amount of water vapor for each unit of the earth's surface area is 2.92 g / cm ² , the air temperature at the earth's surface is 21 ° C. At atmospheric correction, the transmission of electromagnetic radiation and the luminosity of the atmosphere itself were taken into account. 4. At the same stage, the pixel values were converted from radiance to reflectance coefficients from 0 to 1. The reflectance for the spectral zone can be determined by the following, often used in remote sensing, formula:

where E _SUN - average solar extra-atmospheric illumination [W / (m ² nm)];

D is the distance from the Earth to the Sun in astronomical units for a specific date;

- энергетическая яркость для спектральной зоны [Вт/(ср≥м²·нм)];

- energy brightness for the spectral zone [W / (cf≥m ² · nm)];

_- угол возвышения солнца (90° - зенитный угол)

_- angle of elevation of the sun (90 ° - zenith angle)

Spectral curves as reflectance values are shown in FIG. 10.

To identify the boundaries of the reservoir, the pixels of the Landsat-8 image were divided into two subsets (land and water bodies). For this, the previously obtained threshold value of the reflection coefficient between the pixels of satellite images of the corresponding land and water surface was used, which was 0.15 for the near infrared region. Pixels with reflectivity less than the threshold were interpreted as water bodies. Based on this approach, the pixels of the image were divided into two groups (water bodies and land), thus, the boundaries of the Ivankovskoye reservoir were identified. five.

Landsat-8 satellite images obtained in July-August 2015-16 were used to identify spectral images characteristic of aquatic vegetation. According to the claimed method, the areas of development of air-water vegetation are characterized by an excess of the reflectivity of channel 5, corresponding to the near IR 845-885 nm, over channel 4, red 630 - 680 nm. To divide the surface of the reservoir into two groups (covered with air-water vegetation and not covered with air-water vegetation), the coefficient k is used, which is determined by the ratio of the reflectivity coefficients (1). The values of pixels with k <1 identify them as air-water vegetation, and with k> 1 - as a water surface not covered with air-water vegetation. On the basis of this algorithm, areas of distribution of air-aquatic vegetation in the water area of the Ivankovskoye reservoir were identified. Fig. 6.

Algorithm diagram

The general scheme of the algorithm proposed for identifying areas of distribution of air-water vegetation according to Landsat-8 data is shown in Fig. 7.

FIG. 8-10 show the change in the nature of the spectral curves when processing images according to the claimed method.

The implementation of the proposed method is not in doubt due to the availability of the required multispectral data for remote sensing of the Earth and the simplicity of the implementation of computational calculations on a standard computer.

Used Books

1. A method for assessing the level of pollution of water areas according to hyperspectral data of aerospace sounding Patent RU No. 2 616 716 G01N 21/55 2014

2. Landsat 8 (L8) Data Users Handbook Version 5.0. USGS, 2019.106 p.

Claims (1)

A method for identifying the boundaries of water bodies and areas of distribution of air-water vegetation according to multispectral data of remote sensing of the Earth, including: radiometric calibration of an image of the earth's surface with bringing brightness values to energy brightness, atmospheric correction with converting pixel values from energy brightness to reflectance coefficients from 0 to one; highlighting the land surface by highlighting image pixels for which the reflectance value exceeds 0.15 in the near infrared region; selection of a clean water surface due to the selection of image pixels, for which a parameter determined by the ratio of the reflectance coefficients

, greater than 1.

Images