Satellite remote sensing lake algal bloom extraction method under complex observation conditions

Abstract

The invention relates to the technical field of water body detection, in particular to a satellite remote sensing lake water bloom extraction method under complex observation conditions, which comprises the following steps: firstly, acquiring a maximum water body area satellite image of a target lake in a wet season, and determining a maximum water body area; an image to be recognized in a preset time period is obtained, the occurrence frequency of each image NDWI, clear sky pixels and water body pixels is calculated, the water body frequency is calculated according to the frequency, and a water body coverage area is determined; and then, based on NDVI and FBAs indexes, for a turbid water body and various non-water-bloom pixels influenced by clouds / flares and the like, a distinguishing threshold value is set, a decision tree model is constructed through combination, the water bloom pixels are extracted from a water body coverage area, water blooms can be accurately extracted, and lake ecological monitoring and treatment are assisted.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of water body detection, and more particularly to a satellite remote sensing lake bloom extraction method under complex observation conditions. BACKGROUND

[0002] At present, for the evaluation of the occurrence of water bloom, scholars at home and abroad have proposed a variety of algorithms, which can be divided into quantitative evaluation methods based on chlorophyll, phycocyanin and other monitoring indicators, and qualitative monitoring methods based on water bloom area. Among them, chlorophyll a is a pigment common to all algae, which has unique spectral characteristics: strong absorption at blue light band 443 nm and red light band 665 nm, forming a "wave trough" of remote sensing reflectance; at green light band 560 nm and near-infrared 709 nm, it shows "wave peak" characteristics; and at red light band 681 nm, chlorophyll a emits "fluorescence" after receiving excitation light, forming a distinct fluorescence peak. Based on these characteristics, scholars have developed a number of chlorophyll a remote sensing inversion algorithms.

[0003] Phycocyanin, as a photosynthetic auxiliary pigment protein widely existing in cyanobacteria, red algae and cryptophytes, its remote sensing estimation algorithm is similar to chlorophyll a, covering band ratio algorithm, reflection peak baseline method, three-band factor method, etc., and when water bloom occurs, its near-infrared reflectance will be significantly lifted like terrestrial vegetation. In the practice of water bloom monitoring, NDVI and EVI indexes originally used for vegetation growth monitoring are widely used due to their applicability, and almost all sensors are equipped with visible and near-infrared bands. In addition, scholars have also constructed FAI index, color algorithm and other spectral models specially used for water bloom extraction.

[0004] However, the current water bloom remote sensing monitoring technology still has obvious limitations. For the quantitative method of chlorophyll a and phycocyanin, its implementation depends on the high-precision remote sensing reflectance product provided by the atmospheric correction model and the water quality monitoring model suitable for the water bloom area. Although domestic and foreign scholars have proposed various atmospheric correction model assumptions such as near-infrared dark pixel method, short-wave infrared dark pixel method, ultraviolet dark pixel method and the like, these assumptions are almost not suitable for water bloom water body, and at present there is no special atmospheric correction algorithm that can serve the water quality monitoring of water bloom area. At the same time, when water bloom occurs, a large number of algae float on the water surface, and factors such as wind speed and direction change, ship disturbance and the like during the sampling process will interfere with the floating state of the algae, resulting in that it is difficult to obtain high-quality reflectance-water quality synchronous observation data, so that the inversion model lacks sufficient sample support. And most of the existing chlorophyll a or phycocyanin remote sensing inversion models are only suitable for non-dense water bloom water body. These two problems directly lead to the difficulty of remote sensing monitoring of algal biomass in the water bloom area, and the existing lake water quality inversion product often directly masks the water bloom area with uncertain water quality parameter concentration data, resulting in the dilemma of result missing in the water bloom area which most needs water quality parameter concentration data.

[0005] In contrast, water bloom area extraction and monitoring based on satellite images is the most widely used and trusted means by environmental business departments, which does not require completely atmospheric corrected reflectance products, but only needs Rayleigh scattering products with partial atmospheric correction to achieve, but the existing water bloom remote sensing extraction model has insufficient ability in distinguishing water bloom from turbid water body, thin cloud, cloud shadow, flare and the like, which seriously limits the business application of remote sensing technology in water bloom monitoring, therefore, developing a new method for accurately obtaining the area of water bloom under complex conditions has become a problem to be solved at present. SUMMARY

[0006] Therefore, the present application provides a satellite remote sensing lake water bloom extraction method under complex observation conditions to solve the problem that the existing water bloom extraction based on the water bloom remote sensing extraction model has insufficient ability in distinguishing water bloom from turbid water body, thin cloud, cloud shadow, flare and the like, and cannot accurately obtain the area of water bloom under complex conditions.

[0007] A satellite remote sensing lake water bloom extraction method under complex observation conditions, comprising: obtaining a satellite image of the target lake with the largest water body area in the wet season to determine the maximum water body area of the target lake; obtaining the to-be-identified image of the target lake in a preset time period, and calculating the normalized water body index NDWI of each to-be-identified image and the frequency of occurrence of clear sky pixels in the to-be-identified image, for determining the frequency of occurrence of water body pixels in the to-be-identified image; wherein the water body pixel is a clear sky pixel and the corresponding NDWI is greater than zero; According to the occurrence frequency of the clear sky pixel and the occurrence frequency of the water body pixel in the to-be-identified image, a water body frequency of the corresponding pixel as the water body existing in a preset time period is determined, and a pixel with a water body frequency greater than a preset threshold is regarded as a stable water body, which is used to determine a water body coverage area of the target lake. Based on the normalized vegetation index NDVI and the four-band spectral index FBA s A distinguishing threshold of the algal bloom and different types of non-algal bloom pixels is set, the non-algal bloom pixels including turbid water body pixels, water body pixels affected by thin clouds, water body pixels affected by glints, highly turbid water body pixels, thick cloud water body pixels, cloud shadow water body pixels, water body pixels affected by glints and thin clouds together, and water body pixels affected by glints and thick clouds together. The distinguishing thresholds are merged to construct a decision tree model for extracting algal bloom pixels from the water body coverage area.

[0008] In one possible implementation, after obtaining a satellite image with the largest water body area of the target lake in a wet season, the method further includes performing Rayleigh scattering atmosphere correction on the obtained satellite image, and identifying and excluding interference pixels.

[0009] In one possible implementation, the clear sky pixel in the to-be-identified image is identified based on a pixel type classification identification band.

[0010] In one possible implementation, a normalized water body index NDWI of each to-be-identified image is calculated and expressed as: (1) In the formula, R560 represents the Rayleigh correction reflectivity at 560 nm, R1012 represents the Rayleigh correction reflectivity at 1012 nm.

[0011] In one possible implementation, according to the occurrence frequency of the clear sky pixel and the occurrence frequency of the water body pixel in the to-be-identified image, a water body frequency of the corresponding pixel as the water body existing in a preset time period is determined and expressed as: ; In the formula, Rsky represents the occurrence frequency of the clear sky pixel, Rwater represents the occurrence frequency of the water body pixel.

[0012] In one possible implementation, the distinguishing threshold includes: The distinguishing threshold of the algal bloom and the turbid water body pixel is that the NDVI corresponding to the algal bloom pixel is greater than 0.015; The distinguishing threshold of the algal bloom and the water body pixel affected by thin clouds is that the NDVI corresponding to the algal bloom pixel is greater than 0.02 and the FBA s corresponding to the algal bloom pixel is greater than 0.12. The threshold for distinguishing between water blooms and water pixels affected by solar flares and water pixels under cloud shadows is NDVI > 0.001 for water bloom pixels; The threshold for distinguishing between algal bloom and highly turbid water pixels is NDVI > -0.04 for algal bloom pixels; The threshold for distinguishing between algal blooms and thick cloud water pixels is the FBA corresponding to the algal bloom pixel. s >0.145; The threshold for distinguishing algal blooms from water body pixels affected by both solar flares and thin clouds is NDVI > -0.115 for algal bloom pixels. The threshold for distinguishing algal blooms from water body pixels affected by both solar flares and thick clouds is the FBA corresponding to the algal bloom pixel. s >0.07.

[0013] In one possible implementation, the decision tree model is used to sequentially determine whether a pixel belongs to a non-flood bloom pixel. After excluding all non-flood bloom pixels, the remaining pixels are considered flood bloom pixels. The decision tree model is represented as follows: (2) In the formula, This represents the Rayleigh-corrected reflectance at 400 nm. This represents the Rayleigh-corrected reflectance at 865 nm. This represents the Rayleigh-corrected reflectance at 674 nm. This represents the Rayleigh-corrected reflectance at 560 nm. This represents the Rayleigh-corrected reflectance at 709 nm. This represents the Rayleigh-corrected reflectance at 779 nm.

[0014] Compared with the prior art, the technical solution provided in this application has the following beneficial effects: This application solves the problem that fixed boundaries are not suitable for seasonal and interannual hydrological fluctuations by obtaining the largest water body area during the high-water season and calculating water body frequency using multi-time period images, dynamically adapting to changes in lake boundaries, and accurately delineating the range of stable water bodies. For complex observation conditions, it subdivides various non-algal bloom pixel types and sets distinction thresholds, constructs a decision tree model, effectively eliminates interference from turbid water, clouds, flares, etc., and significantly improves the accuracy of algal bloom identification. From pixel-level frequency statistics to decision tree classification extraction, a complete technical chain is formed, which is applicable to different climate zones and different types of lakes, providing reliable technical support for dynamic monitoring of lake algal blooms, ecological research, and water environment management, and helping to efficiently and accurately grasp the distribution and evolution patterns of algal blooms. Attached Figure Description

[0015] Figure 1 This is a flowchart illustrating a method for extracting algal blooms from lakes under complex observation conditions, provided in Embodiment 1 of this application.

[0016] Figure 2 The method provided in the embodiment one of the present application is used for extracting a lake bloom under a complex observation condition. The index can be used to capture the verification diagram of the red wave band absorption characteristics of the water bloom.

[0017] Figure 3 The FBA-based method provided in the embodiment one of the present application is used for extracting a lake bloom under a complex observation condition. s The diagram for distinguishing thick cloud water body from water bloom.

[0018] Figure 4 The flow chart of the method for extracting a lake bloom under a complex observation condition provided in the embodiment two of the present application.

[0019] Figure 5 The different water bloom extraction diagrams under different clear sky conditions provided in the embodiment two of the present application.

[0020] Figure 6 The water bloom extraction result diagrams under different lake cloudy and cloud shadow observation conditions provided in the embodiment two of the present application.

[0021] Figure 7 The water bloom extraction result diagrams under the sun glint observation conditions of Taihu Lake and Xingkai Lake provided in the embodiment two of the present application.

[0022] Figure 8 The water bloom extraction result diagrams under the sun glint and cloud coexistence observation conditions of Taihu Lake provided in the embodiment two of the present application.

[0023] Figure 9 The water bloom extraction result diagrams under the highly turbid water body observation conditions provided in the embodiment two of the present application. DETAILED DESCRIPTION

[0024] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the protection scope of the present application.

[0025] Embodiment one Referring to Figure 1 The flow chart of the method for extracting a lake bloom under a complex observation condition provided in the embodiment one of the present application. As shown in Figure 1 The specific implementation steps of the above method include: Step 101, acquiring a satellite image of the maximum water body area of the target lake in the flood season, and determining the maximum water body region of the target lake.

[0026] In the embodiments of the present application, the digital boundary of the maximum water body image of the target lake in the flood season is used as the basic framework to ensure that the subsequent water body mask does not miss any possible water body area, especially the expanded shore area in the flood season, and to avoid the water body pixels being misjudged as land due to the too small boundary range.

[0027] In step 102, the to-be-identified image of the target lake in a preset time period is acquired, and the normalized water body index (NDWI) of each to-be-identified image and the frequency of occurrence of clear sky pixels in the to-be-identified image are calculated.

[0028] Specifically, the OLCI sensor (Ocean and Land Color Instrument) lacks a short-wave infrared band, which causes the traditional water body index to be unable to be directly used, such as the MNDWI which depends on the short-wave infrared. The NDWI which depends on the green band and the near-infrared band will misjudge the algal bloom as land due to the strong reflection characteristics of the algal bloom in the near-infrared band, which directly leads to the failure of algal bloom extraction. At the same time, the lake shore boundary is affected by hydrology and climate, and there are seasonal and interannual variations, so it is impossible to use a fixed boundary as a mask. Through research, it is found that even in the lakes where algal blooms frequently occur (such as Dianchi Lake, Taihu Lake, and Chaohu Lake), the frequency of occurrence of algal blooms in a single water body pixel does not exceed 60%. This key feature becomes a breakthrough - normal water body pixels are correctly identified as water bodies by the NDWI in most of the time (> 60%), and the algal bloom only interferes with the identification in a small amount of time. Therefore, the stable water body pixels can be screened out by counting the frequency of water body identification by the NDWI, and a dynamic water body mask can be constructed.

[0029] The clear sky pixel mentioned above refers to a pixel that is not blocked by clouds, cloud shadows, fog, haze and other atmospheric phenomena in the image, and can clearly reflect the true state of the ground or water body. In the embodiments of the present application, only the clear sky pixel data is used for subsequent analysis to exclude the interference of clouds, fog, haze and the like. As a special case, for the lakes with a freezing period, only the non-freezing period images, such as the non-freezing period images from May to October, are used in the embodiments of the present application. Because the ice surface in the freezing period will interfere with the water body identification, it needs to be excluded to ensure the effectiveness of the data.

[0030] In step 103, the frequency of occurrence of water body pixels in the to-be-identified image is determined according to the normalized water body index NDWI and the frequency of occurrence of clear sky pixels in the to-be-identified image.

[0031] wherein the water body pixel is a clear-sky pixel and the corresponding NDWI is greater than zero. Specifically, NDWI>0 is a typical feature of normal water body, and water bloom may make NDWI≤0 due to strong near-infrared reflection, and is not counted. In this application, the total number of times that each pixel is determined as a water body in a preset time period, i.e., the water body pixel frequency, is counted based on the above technical solution, which reflects the frequency of the pixel as a normal water body. Optionally, the above preset time period is a specific year or a specific month.

[0032] Specifically, the calculation method of the above normalized water body index NDWI is represented as: (1) In the formula, R560 represents the Rayleigh-corrected reflectance at 560 nm, R1012 represents the Rayleigh-corrected reflectance at 1012 nm.

[0033] Step 104, determining the water body frequency of the corresponding pixel as a water body in the preset time period according to the clear-sky pixel frequency and the water body pixel frequency in the image to be identified.

[0034] wherein the water body frequency is represented as: In the formula, Fsky represents the clear-sky pixel frequency, Fwater represents the water body pixel frequency.

[0035] Step 105, regarding the pixel with a water body frequency greater than a preset threshold as a stable water body, and determining the water body coverage area of the target lake.

[0036] In the embodiment of this application, the pixel with a high probability of being a stable water body is screened based on the above preset threshold, so as to obtain the water body boundary of the target lake in the preset time period. For example, the above preset threshold of this application is taken as 60%, and >60% is taken as the screening condition, so as to obtain the water body boundary of a specific month or a specific month.

[0037] Further, the intersection of the water body boundary of a specific year and the water body boundary of a specific month is taken, and finally the water body mask that meets the interannual and seasonal stability is obtained, which adapts to the seasonal difference of the lake shore boundary and takes into account the interannual difference, so as to realize dynamic adaptation of water-land separation.

[0038] The method ingeniously uses the characteristics of "water bloom frequency≤60%", avoids the defects of the OLCI sensor lacking short-wave infrared, and the misjudgment problem of NDWI to water bloom. At the same time, by dynamically adapting the seasonal and interannual changes of the lake shore boundary, an accurate water-land separation mask is finally provided for water bloom extraction of OLCI image, which solves the limitations of traditional methods.

[0039] Step 106, setting a threshold for distinguishing the water bloom from different types of non-water bloom pixels, the non-water bloom pixels including turbid water pixels, water pixels affected by thin clouds, water pixels affected by glint, highly turbid water pixels, thick cloud water pixels, cloud shadow water pixels, water pixels affected by glint and thin clouds, and water pixels affected by glint and thick clouds.

[0040] The present application considers that the water bloom and different types of non-water bloom water pixels, such as turbid water, cloud-polluted water, glint-affected water, etc., have different spectral characteristics on the normalized vegetation index (Normalized Difference Vegetation Index, hereinafter referred to as NDVI) and the four-band spectral index FBA s , and the differences are quantified by setting a threshold to gradually exclude non-water bloom pixels and finally retain water bloom pixels.

[0041] Specifically, the calculation formula of the four-band spectral index FBA s is as follows:

[0042] In the formula, represents the Rayleigh-corrected reflectance at 674 nm, represents the Rayleigh-corrected reflectance at 560 nm, represents the Rayleigh-corrected reflectance at 709 nm, and represents the Rayleigh-corrected reflectance at 779 nm.

[0043] Compared with the four-band spectral index FBA constructed in the prior art, represented as: , the 674 nm proposed in the present application is closer to the absorption peak of algae in the red band (~670 nm), and the average is 0.17 m -1 higher than the average , and the average value is about 8% lower than that at 665 nm.

[0044] The higher and the relatively lower proportion show that 674 nm is more suitable for separating and than 665 nm. Therefore, the first improvement of the FBA index in the present application is to use 674 nm instead of 665 nm.

[0045] Using 674 nm instead of 665 nm requires re-parameterization of . The minimum mean absolute percentage deviation (MAPD) estimated by is used as the evaluation index, and an iterative method is used to estimate the parameters. ​The range is 0.0-1.0, with a step size of 0.01), to determine the optimal value. Value. This study obtained 253 absorption data samples from five lakes, of which 126 were used for parameterization. 127 for use on Perform verification. For example... Figure 2 As shown, when When equal to 0.19, estimate The best results were achieved, with an average relative error of only 14.63%. This was calculated based on measured reflectance. The index assesses its ability to estimate Chla. The high correlation (R²=0.75) between the spectral index and the measured Chla indicates that the spectral index provided in this application... The index can be used to capture the absorption characteristics of the red band of algal blooms.

[0046] The formula for calculating the Normalized Difference Vegetation Index (NDVI) is as follows:

[0047] In the formula, This represents the Rayleigh-corrected reflectance at 865 nm. This represents the Rayleigh-corrected reflectance at 674 nm.

[0048] In this embodiment of the application, for different types of non-bloom pixels, the distinction threshold between bloom and different types of non-bloom pixels is determined one by one.

[0049] For turbid water bodies, the core difference between algal blooms and turbid water bodies lies in NDVI. Algal blooms have a higher NDVI due to their strong near-infrared reflection. Therefore, this application uses NDVI > 0.015 as the distinguishing factor, that is, NDVI exceeding this value is considered algal bloom, otherwise it is considered turbid water body.

[0050] Furthermore, after being identified as an algal bloom, based on the four-band spectral index FBA s NDVI can further distinguish the corresponding types of algal blooms, including algal bloom pixels under clear skies (NB), algal bloom pixels under thin clouds (CB), algal bloom pixels under flares (GB), and algal bloom pixels under the combined influence of flares and thin clouds (GCB).

[0051] Specifically, for water bodies affected by thin clouds, thin clouds may reduce NDVI and affect FBA. s Therefore, the water bloom pixel must simultaneously satisfy NDVI>0.02 and FBA. s A value >0.12 is needed to ensure high reflectivity of algal blooms and eliminate low reflectivity interference from thin clouds; only by combining both can they be distinguished.

[0052] For the water body affected by flare, flare can enhance the near-infrared reflection, but the intensity is not as high as that of algal bloom, so a lower NDVI threshold NDVI>0.001 can be used for distinction. The NDVI of algal bloom is higher, and the NDVI of water body affected by flare is lower.

[0053] For the highly turbid water body, the NDVI of the highly turbid water body can be lower, so NDVI>-0.04 is used as the distinction, and the NDVI of algal bloom is significantly higher than this value.

[0054] For thick cloud water body, as shown in Figure 3 , thick cloud has a more significant impact on FBA s , and the FBA s of algal bloom is higher, so FBA s >0.145 is directly distinguished.

[0055] For cloud shadow water body, cloud shadow can reduce reflectivity, resulting in lower NDVI, and the NDVI of algal bloom is higher, so NDVI>0.001 is used as the distinction.

[0056] For water body affected by flare and thin cloud, NDVI can be further reduced under the double interference, so the NDVI threshold is relaxed to NDVI>-0.115, and even if the algal bloom is slightly affected, the NDVI is still higher than this value.

[0057] For water body affected by flare and thick cloud, thick cloud mainly affects FBA s , so FBA s >0.07 is used as the distinction, and the FBA s of algal bloom is higher than that of the water body affected by this interference.

[0058] Step 107, combine the above distinction thresholds to construct a decision tree model for extracting algal bloom pixels from the above water body coverage area.

[0059] In the embodiment of the application, the above 10 types of distinction conditions are combined to finally obtain a decision tree model for extracting algal bloom and non-algal bloom water bodies, which is represented as: (2) In the formula, represents the Rayleigh-corrected reflectance at 400 nm, represents the Rayleigh-corrected reflectance at 865 nm.

[0060] Based on the technical solutions provided in the present application, first, the difference points of water bloom and each interference, i.e., the threshold, are found, and then all interferences are sequentially investigated by a decision tree, and finally the water bloom pixels are left. For example, first, it is judged whether it is a certain type of non-water bloom pixel, such as turbid water, which is judged by NDVI≤0.015. If yes, it is excluded; if not, the next step is entered. It is judged whether it is a water body affected by thin clouds, which is judged by NDVI≤0.02 or FBAs≤0.12. If yes, it is excluded; if not, the above steps are continued to sequentially exclude all non-water bloom pixel types. Finally, the pixels that are not excluded are the water bloom pixels.

[0061] Compared with the prior art, the technical solutions provided in Embodiment One of the present application have the following beneficial effects: By obtaining the maximum water body area in the wet season and combining the water body frequency to determine the stable water body coverage area, the dynamic changing lake water body boundary is accurately outlined, and the problem that the traditional fixed boundary is not suitable is solved; the water body frequency is calculated by using NDWI and clear sky, water body pixel frequency, non-stable water body interference is effectively filtered, and the water body recognition accuracy is improved; for various complex non-water bloom pixel types, the threshold is set and the decision tree model is constructed, which can accurately distinguish water bloom from various interferences under complex observation conditions such as clouds and flares, greatly improves the water bloom extraction accuracy, provides reliable technical support for lake water bloom dynamic monitoring, ecological environment evaluation and governance decision, and helps to accurately master the distribution and evolution law of water bloom.

[0062] Embodiment Two Reference Figure 4 A flowchart of a satellite remote sensing lake water bloom extraction method under complex observation conditions provided in Embodiment Two of the present application is shown in FIG. 2. As shown in FIG. 2, the specific implementation steps of the above method include: Figure 4 (1) Data preparation. In the present application, the first-level image of the Taihu Lake region obtained by the OLCI sensor carried by the Sentinel-3 satellite of the European Space Agency is selected, which covers the whole Taihu Lake and the surrounding area.

[0063] (2) Data preprocessing. In the present application, Rayleigh scattering correction is performed on each obtained OLCI first-level image. Specifically, based on the atmospheric molecular scattering characteristics, the short-wave band signals such as blue light and green light are corrected by using the radiation transfer model and algorithm matched with the OLCI sensor, so as to eliminate the interference of Rayleigh scattering on the spectral characteristics of water body and water bloom, and obtain the surface reflectance that can be used for spectral analysis.

[0064] (3) In the preprocessed image, two types of typical samples are selected, including water bloom samples and non-water bloom samples.

[0065] (4) The water body index, vegetation index, and four-band index are calculated for the selected sample pixels, and the index difference between the water bloom and non-water bloom samples is compared.​

[0066] (5) Calculate water frequency and spectral threshold to perform water bloom extraction algorithm under complex observation conditions. As shown in FIG. 6, the water bloom extraction result example provided by the embodiment of the present application is shown. Figures 5-9

[0067] It should be noted that the above technical solutions of the second embodiment of the present application can further refer to the description of the technical solutions in the first embodiment, and the second embodiment of the present application will not be repeated.

[0068] Compared with the prior art, the technical solutions provided by the second embodiment of the present application have the following beneficial effects: The present application proposes a water body boundary remote sensing extraction algorithm considering seasonal and interannual differences, and a lake water bloom extraction algorithm under complex observation conditions based on NDVI and FBA index. The algorithm can accurately and stably extract water bloom under complex observation conditions including cloud, cloud shadow, turbid water body, solar flare, etc., and can serve the monitoring of water bloom occurrence of eutrophic lakes in China. s

[0069] Although the embodiments of the present application have been shown and described, it can be understood by those skilled in the art that various changes, modifications, replacements and variations can be made to the embodiments without departing from the principles and spirits of the present application, and the scope of the present application is defined by the appended claims and their equivalents.​​

Claims

1. A method for extracting satellite remote sensing lake bloom under complex observation conditions, characterized in that, The method comprises the following steps: acquire satellite images of the target lake in the wet season with the largest water area, and determine the maximum water area of the target lake; acquire the to-be-identified images of the target lake in a preset time period, and calculate the normalized water index (NDWI) of each to-be-identified image and the frequency of occurrence of clear sky pixels in the to-be-identified image, which is used to determine the frequency of occurrence of water pixels in the to-be-identified image; the water pixel is a clear sky pixel and the corresponding NDWI is greater than zero; determine the water frequency of the corresponding pixel as water existing in the preset time period according to the frequency of occurrence of clear sky pixels and the frequency of occurrence of water pixels in the to-be-identified image, and regard the pixel with a water frequency greater than a preset threshold as stable water, which is used to determine the water coverage area of the target lake; Based on normalized difference vegetation index (NDVI) and four-band spectral index (FBA) s Setting the threshold for distinguishing the water bloom from different types of non-water bloom pixels, including turbid water body pixels, water body pixels affected by thin clouds, water body pixels affected by glint, highly turbid water body pixels, thick cloud water body pixels, cloud shadow water body pixels, water body pixels affected by both glint and thin clouds, water body pixels affected by both glint and thick clouds; merge the classification threshold to construct a decision tree model, which is used to extract algal bloom pixels from the water coverage area.

2. The method according to claim 1, wherein, After acquiring the satellite image of the target lake in the wet season with the largest water area, the method further comprises performing Rayleigh scattering atmospheric correction on the acquired satellite image, and identifying and excluding interference pixels.

3. The method according to claim 1, wherein, The clear sky pixels in the to-be-identified image are determined based on the identification of pixel type classification bands.

4. The method according to claim 1, wherein, The normalized water index (NDWI) of each to-be-identified image is calculated and expressed as: （1） wherein R560 represents the Rayleigh corrected reflectance at 560 nm, R1012 represents the Rayleigh corrected reflectance at 1012 nm.

5. The method according to claim 1, wherein, The water frequency of the corresponding pixel as water existing in the preset time period is determined according to the frequency of occurrence of clear sky pixels and the frequency of occurrence of water pixels in the to-be-identified image, and expressed as: ； In the formula, represents the frequency of occurrence of clear-sky pixels, represents the frequency of occurrence of water body pixels.

6. The method according to claim 1, wherein, The classification threshold comprises: the classification threshold of algal bloom and turbid water pixels is that the NDVI corresponding to the algal bloom pixel is greater than 0.015; The threshold for distinguishing water bloom from water body pixels affected by thin clouds is that the NDVI of water bloom pixels is greater than 0.02 and the FBA of water bloom pixels is greater than 0.

12. s >0.12; the classification threshold of algal bloom and water pixels affected by sun glint and cloud shadow water pixels is that the NDVI corresponding to the algal bloom pixel is greater than 0.001; the classification threshold of algal bloom and highly turbid water pixels is that the NDVI corresponding to the algal bloom pixel is less than -0.04; The distinguishing threshold of water bloom and thick cloud water body pixel is FBA corresponding to water bloom pixel s > 0.145; the classification threshold of algal bloom and water pixels affected by sun glint and thin cloud is that the NDVI corresponding to the algal bloom pixel is less than -0.115; The threshold for distinguishing the water bloom and the water body pixel affected by the flare and the thick cloud is FBA corresponding to the water bloom pixel s > 0.

07.

7. The method according to claim 1, wherein, The decision tree model is used to sequentially determine whether the pixel belongs to a non-algal bloom pixel, and after excluding all non-algal bloom pixels, the remaining pixels are algal bloom pixels, and the decision tree model is expressed as: （2） wherein R(400) represents the Rayleigh corrected reflectance at 400 nm, R(865) represents the Rayleigh corrected reflectance at 865 nm, R(674) represents the Rayleigh corrected reflectance at 674 nm, R(560) represents the Rayleigh corrected reflectance at 560 nm, R(709) represents the Rayleigh corrected reflectance at 709 nm, R(779) represents the Rayleigh corrected reflectance at 779 nm.

Images