Water body classification and chlorophyll concentration segmented inversion method based on dynamic threshold value

Abstract

The invention discloses a water body classification and chlorophyll a concentration segmented inversion method based on a dynamic threshold value, which comprises the following steps of: constructing a water body nutrition state classification system by using a ratio index and an improved floating algae index (AFAI) according to the difference of remote sensing reflectivity of spectra of different types of water bodies in different wavebands, and constructing a regression model based on measured data, and the original static classification threshold is dynamic, so that self-adaptive adjustment of water body classification is realized. And finally, respectively constructing Chla inversion algorithms on the oligotrophic water body and the moderate algal bloom water body. Compared with a traditional fixed threshold value classification method, the dynamic threshold value can automatically adjust the classification boundary according to the environmental change, and the stability of the model under the seasonal change is enhanced. Meanwhile, compared with a single inversion algorithm, the segmented inversion method can significantly improve the estimation accuracy of Chla, and effectively improves the adaptability of an inversion model to water bodies in different nutritional states.

Description

Technical Field

[0001] The present invention relates to the technical field of water environment remote sensing monitoring, in particular to a water body classification and chlorophyll a concentration segmented inversion method based on dynamic thresholds. Background Art

[0002] Inland water bodies, such as lakes, rivers, and reservoirs, play a key role in water resource supply, biodiversity maintenance, and climate regulation. However, with the intensification of industrialization and agricultural activities, large amounts of nutrients such as nitrogen and phosphorus are discharged into these bodies, promoting the proliferation of algae in the water, leading to increasingly serious eutrophication and, in turn, algal blooms, which pose a serious threat to aquatic ecological security and human health. Chlorophyll a concentration, as a representative indicator of phytoplankton biomass in water, is an important parameter for measuring the degree of eutrophication and algal growth. Therefore, efficient and accurate monitoring of Chla in lakes is a key tool for water environment management.

[0003] Traditional water quality monitoring methods rely primarily on manual sampling and laboratory analysis. While highly accurate, they suffer from long monitoring cycles, limited spatial coverage, and significant labor and resource consumption, making them inadequate for monitoring the continuous dynamics of lake Chla. With the advancement of remote sensing technology, inversion models based on the correlation between various water quality parameters and remote sensing spectral reflectance can accurately determine the frequency of algal blooms and their spatial distribution. The Landsat series and Sentinel-2 satellites offer high spatial resolution and multispectral bands. Their combination provides a high revisit frequency, providing a rich data source for monitoring the spatial and temporal dynamics of algal blooms in inland waters.

[0004] Common methods for inverting Chla using remote sensing images include empirical models, semi-empirical and semi-analytical models, and machine learning models. However, in practical applications, due to the complexity of the optical characteristics of water bodies and the significant differences in water quality between lakes, a single model is difficult to apply to water bodies in different concentration ranges, which will lead to a decrease in inversion accuracy and even systematic deviations. Therefore, it is an effective solution to establish a segmented inversion model that is applicable to multi-source remote sensing data and can model different concentration ranges. In previous studies, the main methods for classifying water bodies are as follows: ① Water quality parameter threshold classification method. This method divides the types into 5 levels according to the thresholds set by the water quality parameters. Among them, when the algae density is greater than 1*10 7When the concentration of chlorophyll cells / L is less than 1%, the water body is considered to have an algal bloom. ② Spectral feature analysis method. This method is based on the differences in spectral characteristics of different water types and is usually classified by constructing water body classification indicators such as the normalized difference chlorophyll index (NDCI) and the maximum chlorophyll index (MCI). ③ Optical feature classification method. Based on the differences in scattering coefficients of different water types, water bodies are divided into three categories: phytoplankton-dominated, inorganic particulate-dominated, and both. Based on this, a relationship model between the scattering coefficient and the concentration of inorganic suspended solids is constructed. ④ Machine learning method. Supervised classification methods such as support vector machines (SVM) and maximum likelihood classification (MLC) based on training samples are suitable for areas with a large number of water body type samples. Unsupervised clustering methods such as K-means and ISODATA do not require sample support and can automatically identify water body types with different spectral characteristics in remote sensing data, but they suffer from poor interpretability. In addition, deep learning methods such as convolutional neural networks (CNN) have also shown good performance in water body classification, but the large amount of training data and time limit their widespread application. The above methods are usually based on fixed thresholds, ignoring the changes in spectral shape caused by changes in water color during seasonal changes. In addition, the metabolic levels of algae at different temperatures will also affect the changes in spectral emissivity. Therefore, fixed thresholds will still lead to a certain degree of misclassification, especially during seasonal changes. Summary of the Invention

[0005] The purpose of the present invention is to provide a water body classification and chlorophyll a concentration segmented inversion method based on dynamic thresholds, which can effectively identify water bodies with different nutrient states and their spatial distribution according to temperature and monthly information, and at the same time establish differentiated inversion algorithms for different types of water bodies, significantly improving the Chla inversion accuracy in complex water environments, so as to address the shortcomings of the existing technology.

[0006] To achieve the above object, the present invention provides the following technical solutions:

[0007] The water body classification and chlorophyll a concentration segmented inversion method based on dynamic thresholds includes the following steps:

[0008] S1: Data collection and preprocessing: Cloud-free L1 products from LandSat 8 and 9 and cloud-free L1C products from Sentinel-2 in the study area were selected. Radiometric calibration, atmospheric correction, and super-resolution synthesis were performed on the images to obtain water reflectance information. Measured data were obtained through field sampling and laboratory analysis, with sampling occurring on average once a month, excluding freezing periods.

[0009] S2: Construct water body classification index: Based on the spectral information obtained in S1, construct the Rrs(Blue) / Rrs(Green) ratio index as the classification index for oligotrophic water bodies and eutrophic water bodies, and then construct the AFAI index to classify eutrophic water bodies into moderate algal bloom water bodies and severe algal bloom water bodies;

[0010] S3: Construct a dynamic threshold function: Based on the ratio index and AFAI index value calculated in S2, use a regression model to convert the threshold into a function of temperature and month, and then classify the water body type in each image according to the dynamic threshold obtained by this function;

[0011] S4: Chla segmented inversion: Chla inversion algorithms are constructed for oligotrophic water bodies and moderate algal bloom water bodies respectively, and a linear weighting function is introduced to solve the problem of spatial discontinuity of inversion results in water type transition areas.

[0012] Furthermore, the L1 level images of Landsat 8 & 9 in S1 were first calibrated using the RadiometricCalibration tool in ENVI 5.3.1 to convert the DN value into apparent radiance, as shown in Equation (1). Then, the MODTRAN 4+ radiation transfer model in the FLAASH module was used to perform atmospheric correction on the images to remove the atmospheric effects of water vapor and aerosols, and obtain remote sensing reflectance data.

[0013]

[0014] Where L is the radiant brightness [W / (m 2 ·sr·μm)], DN is the grayscale value, A is the gain, and L0 is the offset.

[0015] Furthermore, Sen2cor was used in S1 to perform atmospheric correction on the Sentinel-2 data. Then, the super-resolution synthesis function in SNAP was used to increase the spatial resolution of the 20m and 60m band data to 10m. The band synthesis function of ENVI 5.3.1 was used to synthesize the 12 band data except the B10 band, which was used for atmospheric correction, to obtain the remote sensing reflectance data of all bands.

[0016] Furthermore, the formula for constructing the AFAI index in S2 is as follows:

[0017]

[0018] Where R rs (λ) is the remote sensing reflectivity of the λ band, Order λ is the band sequence number of the λ band in each image.

[0019] Furthermore, the formula for constructing the dynamic threshold function in S3 is as follows:

[0020]

[0021] Where, T is the average temperature of the day (℃), M is the current month, a~f are the parameters obtained by regression fitting, R rs (Blue) / R rs (Green)_threshold and AFAI_threshold are the optimal water body classification thresholds for the current date.

[0022] Furthermore, the Chla segmented inversion algorithm constructed in S4 is as follows:

[0023]

[0024] Where Thr_1 is R rs (Blue) / R rs (Green)_threshold, Thr_2 is AFAI_threshold, Thr_3 and Thr_4 are the thresholds of the water type transition section.

[0025] Compared with the prior art, the present invention has the following beneficial effects:

[0026] The water body classification and chlorophyll a concentration segmented inversion method based on dynamic thresholds of the present invention utilizes the excellent adaptability of Landsat 8&9 and Sentinel-2 images in small and medium-sized water bodies, constructs a dynamic threshold function for more detailed water body division, and builds a segmented differentiated inversion system on this basis, effectively improving the adaptability and accuracy of remote sensing inversion, meeting the dynamic monitoring needs of Chla in inland water bodies under complex water quality conditions. It not only realizes water body classification that varies with time, but also overcomes the limitations of traditional inversion models under conditions of large concentration spans or significant regional water body differences. At the same time, it has strong versatility and technical innovation. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1 This is the Chla segmented inversion flow chart of the present invention;

[0028] Figure 2 This is the geographical structure distribution map of Zhukui Reservoir, the study area of ​​the present invention;

[0029] Figure 3 Spectral reflectance curves of water bodies in different nutrient states according to the present invention;

[0030] Figure 4 This is the Chla inversion model diagram of Zhukuima Reservoir under different water body types of the present invention;

[0031] Figure 5 This is the water body classification result map of Zhukuma Reservoir based on Landsat 8&9 and Sentinel-2 images in the present invention;

[0032] Figure 6 This is the Chla segmented inversion result diagram of Zhukuima Reservoir on Landsat 8&9 and Sentinel-2 images in the present invention;

[0033] Figure 7 This is a scatter plot of the inversion accuracy of the Zhukui Reservoir of the present invention. DETAILED DESCRIPTION

[0034] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0035] See also Figure 1 The embodiment of the present invention provides a method for water body classification and chlorophyll a concentration segmented inversion based on dynamic thresholds, comprising the following steps:

[0036] S1: Data collection and preprocessing: Cloud-free L1 products from LandSat 8 and 9 and cloud-free L1C products from Sentinel-2 in the study area were selected. The images were preprocessed for radiometric calibration, atmospheric correction, and super-resolution synthesis to obtain water remote sensing reflectance information. Measured data were obtained through field sampling and laboratory analysis, with sampling occurring on average once a month, excluding the freezing period.

[0037] In this step, the remote sensing products used in the embodiment of the present invention are Landsat 8&9 and Sentinel-2. Landsat 8&9 images can be downloaded free of charge from the website of the United States Geological Survey (http: / / earthexplorer.usgs.gov), and Sentinel-2 L1C image data can be downloaded from the website of the Copernicus Open Access Center (https: / / scihub.copernicus.eu / dhus / # / home). In order to ensure the availability of remote sensing data, only cloudless or lightly clouded images with less than 10% cloud cover above the study area are selected. The field sampling frequency is once a month (except during the freezing period), and the sampling points evenly cover the entire water area, with an interval of about 1,000 meters, in order to fully reflect the spatiotemporal variation characteristics of Lake Chla. The specific sampling point distribution is shown in Figure 2 .

[0038] Secondly, the L1 level images of Landsat 8 & 9 were first calibrated using the RadiometricCalibration tool in ENVI 5.3.1 to convert the DN value into apparent radiance, as shown in Equation (1). Then, the MODTRAN 4+ radiation transfer model in the FLAASH module was used to perform atmospheric correction on the images to remove atmospheric effects such as water vapor and aerosols, and obtain remote sensing reflectance data.

[0039]

[0040] Where L is the radiant brightness [W / (m 2 ·sr·μm)], DN is the grayscale value, A is the gain, and L0 is the offset.

[0041] It should be noted that Sen2cor in this example is an official European Space Agency atmospheric correction model with high accuracy for inland water bodies. Therefore, this example uses Sen2cor to perform atmospheric correction on Sentinel-2 data. The super-resolution synthesis function in SNAP is then used to increase the spatial resolution of the 20m and 60m bands to 10m. The band synthesis function in ENVI 5.3.1 is used to synthesize the data of the 12 bands, excluding the B10 band (which is used for atmospheric correction), to obtain remote sensing reflectance data for all bands.

[0042] S2: Construct water body classification index: Based on the spectral information obtained in S1, construct the Rrs(Blue) / Rrs(Green) ratio index as the classification index for oligotrophic water bodies and eutrophic water bodies, and then construct the AFAI index to classify eutrophic water bodies into moderate algal bloom water bodies and severe algal bloom water bodies; for details, please refer to Figure 3 , (a) and (b) show the spectral reflectance curves of oligotrophic water bodies and eutrophic water bodies, (c) and (d) show the spectral reflectance curves of moderate algal bloom water bodies and severe algal bloom water bodies. The concentration of algae and suspended matter in oligotrophic water bodies is low, and the clear water absorbs most of the light, so the overall reflectivity is low. Eutrophic water bodies are strongly absorbed by chlorophyll, carotenoids and yellow substances, and their reflectivity is low in the blue light band of 400-500nm, showing a reflection valley. Since the content of algae and organic matter in oligotrophic water bodies is small, the reflection effect on green light is weak, so its reflectivity in the green light band of 500-600nm is low. Eutrophic water bodies form a reflection peak near 560nm due to the weak absorption of algae and the scattering effect of cells, and as the concentration of solid suspended matter such as sediment particles in the water body increases, the reflection shifts to the red light channel. Based on the above-mentioned spectral reflectance differences, the present invention constructs a ratio index R rs (Blue) / R rs(Green), which can effectively distinguish oligotrophic water bodies from eutrophic water bodies.

[0043] Due to the strong absorption of phycocyanin and chlorophyll in the red light band, eutrophic water bodies form reflection valleys around 620-680nm ( Figure 3 (c) and (d)). In addition, when algal blooms occur, the water body has a strong reflection in the near-infrared band, which makes the algal bloom water body have a red-edge reflection characteristic similar to terrestrial vegetation. Therefore, eutrophic water bodies will form a reflection peak around 700nm, and the peak value increases with the increase of Chla concentration. Utilizing this feature, the present invention uses the AFAI index (Formula (2)) to distinguish between moderate algal bloom water bodies and severe algal bloom water bodies.

[0044]

[0045] Where R rs (λ) is the remote sensing reflectivity of the λ band, Order λ is the band sequence number of the λ band in each image.

[0046] S3: Construct a dynamic threshold function: Based on the ratio index and AFAI index value calculated in S2, use a regression model to convert the threshold into a function of temperature and month, and then classify the water body type in each image according to the dynamic threshold obtained by this function;

[0047] Specifically, the daily average temperature and month information for each remote sensing image are extracted. At the same time, for points with measured data, the specific values ​​of the ratio index and AFAI index are calculated, and the optimal static classification threshold of the day obtained by the visual interpretation method is used to classify the image into water body types. The static threshold, daily average temperature, and current month corresponding to the multi-day water body classification results are collected, and then these data are regressed and fitted to obtain a dynamic threshold function. The specific formulas are shown in Formula (3) and Formula (4):

[0048]

[0049] Where, T is the average temperature of the day (℃), M is the current month, a~f are the parameters obtained by regression fitting, R rs (Blue) / R rs (Green)_threshold and AFAI_threshold are the optimal water body classification thresholds for the current date.

[0050] S4: Chla Segmented Inversion: Chla inversion algorithms are constructed for oligotrophic water bodies and moderate algal bloom water bodies, respectively, and a linear weighting function is introduced to address the spatial discontinuity of inversion results in transitional regions between water types. Specifically, after an algal bloom forms, cyanobacteria can alter buoyancy, causing their vertical distribution within the water body to become uneven. However, the signal received by satellites only reflects the surface layer of the water. When the bloom is severe, Chla typically exceeds 100 μg / L and is difficult to accurately measure. Estimating an inaccurate Chla concentration using remote sensing imagery when the water body is severely eutrophic is completely unnecessary. Therefore, this embodiment only marks water bodies with severe algal blooms without calculating their specific values. Differentiated inversion models are established for oligotrophic water bodies and moderate algal bloom water bodies.

[0051] The two-band ratio algorithm constructs the reflectance ratio of the Chla sensitive band, effectively enhancing the Chla spectral response signal while suppressing interference from water background noise. This algorithm also has clear physical meaning and is compatible with a variety of medium- and high-resolution satellite data, including Landsat 8&9 and Sentinel-2, demonstrating excellent adaptability.

[0052] Specifically, for oligotrophic water bodies, the reflectivity changes are relatively weak, and it is preferred to select a band combination that has a highly sensitive response to trace Chla fluctuations. This embodiment constructs a B2 (blue light, 483nm) / B4 (red light, 655nm) ratio model in Landsat 8&9 images, and constructs B8 (near infrared, 865nm) / B6 (red edge, 740nm) on Sentinel-2 images; for moderate algal bloom water bodies, the Chla concentration is at a medium-high level, the phytoplankton density on the water surface is large, the spectral response is strong, and there is more clutter interference. The model needs to have stronger anti-interference and nonlinear fitting capabilities. The present invention constructs AFAI for Landsat 8&9 images and B5 (near infrared, 705nm) / B3 (green light, 560nm) algorithms for Sentinel-2 images. The scatter plot of the inversion model is shown in Figure 2. Figure 4 ((a) and (b) are Landsat 8&9 images, (c) and (d) are Sentinel-2 images).

[0053] In order to solve the common spatial mutation and discontinuity problems in water body type segmented modeling, the present invention further introduces a linear weighted fusion function in this step. In the model transition area, according to the proximity of the water body spectral characteristics, a continuously changing weight coefficient is set, and the output results of the two types of models are smoothly transitioned, thereby achieving continuous transition and physical consistency of the inversion results in space, significantly improving the engineering usability and ecological interpretation value of the model. The present invention uses a linear weighted function to calculate the transition segment. Combined with S2-S4, the present invention finally forms a set of Chla segmented inversion methods as follows:

[0054]

[0055]

[0056] Where Thr_1 is R rs (Blue) / R rs (Green)_threshold, Thr_2 is AFAI_threshold, Thr_3 and Thr_4 are the thresholds of the water type transition section.

[0057] In order to further better explain the applicability of the present invention in practical applications, the embodiment of the present invention also provides a case study of Zhukui Reservoir for verification:

[0058] Zhukui Reservoir (39°50′-39°52′N, 122°45′-122°53′E) is located on the western tributary of Zhuanghe River ( Figure 2 ), a large reservoir mainly used for flood control and urban water supply, with irrigation, fish farming, power generation and other comprehensive uses, with a water area of ​​about 90km 2 Nutrient enrichment from agricultural activities around the basin creates conditions for algal blooms year-round in the reservoir. In the summer, when temperatures rise or air pressure drops, severe algal blooms can occur. Therefore, the area offers excellent conditions for water quality inversion, enabling verification of the accuracy of the Chla segmented inversion algorithm.

[0059] The present invention was tested on all cloud-free Landsat 8&9 and Sentinel-2 images of Zhukui Reservoir from April to November 2023. The results are as follows: Figure 5 and Figure 6 As shown in the figure, the water body classification and Chla segment inversion results of the two images on similar dates are highly consistent in spatial pattern and can verify each other, indicating that the proposed method has good adaptability and consistency and has high spatial recognition accuracy.

[0060] In addition, images near the measured date are selected to construct a scatter plot of the inversion results as shown in Figure 7 The scattered points are evenly distributed near the 1:1 line, and the overall inversion accuracy of Landsat 8&9 images reaches R 2 =0.83, RMSE = 19.58 μg / L, and the inversion accuracy of Sentinel-2 image is R 2=0.76, RMSE =24.51 μg / L, both demonstrating strong Chla inversion capabilities. Regarding the inversion performance for different water types, for oligotrophic waters, RMSE_L = 0.87 μg / L, RMSE_S = 0.28 μg / L, and for waters with moderate algal blooms, RMSE_L = 34.70 μg / L, RMSE_S = 29.40 μg / L. This demonstrates not only high overall inversion accuracy for both image types, but also stable and high-precision inversion performance for waters with different nutrient levels, demonstrating the practical and scalable value of this invention.

[0061] In summary: The water body classification and chlorophyll a concentration segmented inversion method based on dynamic thresholds of the present invention introduces a dynamic threshold adjustment mechanism based on temperature and monthly factors, realizes accurate identification and inversion modeling of the nutrient status of inland complex water bodies, helps to complete low-cost, high-precision water quality monitoring tasks, and thus effectively reduces the risk of monitoring personnel conducting field measurements. Secondly, the present invention mainly constructs a segmented inversion algorithm based on LandSat 8&9 (30m) and Sentinel-2 (10m, 30m, 60m) images with high spatial resolution. The results show that the spatial distribution of water body classification and Chla inversion results obtained by this method has high consistency and high inversion accuracy. In addition, the combined use of the two images can make the results mutually confirm each other, improve the cross-validation ability of the results, and can also improve the temporal resolution of the inversion to a certain extent. With the replacement of optical satellites carrying imaging spectrometers and the emergence of high temporal and spatial resolution remote sensing data sets, the practicality and reliability of the present invention are expected to be further improved, so the present invention has high application potential.

[0062] The above description is only a preferred specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any technician familiar with the technical field, within the technical scope disclosed by the present invention, who makes equivalent replacements or changes based on the technical solution and inventive concept of the present invention, should be covered by the scope of protection of the present invention.

Claims

1. The water body classification and chlorophyll a concentration segmented inversion method based on dynamic threshold is characterized by: The following steps are involved: S1: Data collection and preprocessing: Cloud-free L1 products from LandSat 8 and 9 and cloud-free L1C products from Sentinel-2 in the study area were selected. Radiometric calibration, atmospheric correction, and super-resolution synthesis were performed on the images to obtain water reflectance information. Measured data were obtained through field sampling and laboratory analysis, with sampling occurring on average once a month, excluding freezing periods. S2: Construct water body classification index: Based on the spectral information obtained in S1, construct the Rrs(Blue) / Rrs(Green) ratio index as the classification index for oligotrophic water bodies and eutrophic water bodies, and then construct the AFAI index to classify eutrophic water bodies into moderate algal bloom water bodies and severe algal bloom water bodies; S3: Construct a dynamic threshold function: Based on the ratio index and AFAI index value calculated in S2, use a regression model to convert the threshold into a function of temperature and month, and then classify the water body type in each image according to the dynamic threshold obtained by this function; S4: Chla segmented inversion: Chla inversion algorithms are constructed for oligotrophic water bodies and moderate algal bloom water bodies respectively, and a linear weighting function is introduced to solve the problem of spatial discontinuity of inversion results in water type transition areas.

2. The method for water body classification and chlorophyll a concentration segmented inversion based on dynamic threshold according to claim 1, characterized in that: The L1 level images of Landsat 8 & 9 in S1 were first calibrated using the Radiometric Calibration tool in ENVI 5.3.1 to convert the DN value into apparent radiance, as shown in Equation (1). Then, the MODTRAN4+ radiation transfer model in the FLAASH module was used to perform atmospheric correction on the images to remove the atmospheric effects of water vapor and aerosols, and obtain remote sensing reflectance data. Where L is the radiant brightness [W / (m 2 ·sr·μm)], DN is the grayscale value, A is the gain, and L0 is the offset.

3. The method for water body classification and chlorophyll a concentration segmented inversion based on dynamic threshold according to claim 1, characterized in that: In S1, Sen2cor was used to perform atmospheric correction on the Sentinel-2 data. The super-resolution synthesis function in SNAP was then used to increase the spatial resolution of the 20m and 60m bands to 10m. The band synthesis function in ENVI 5.3.1 was used to synthesize the 12 bands except the B10 band for atmospheric correction to obtain the remote sensing reflectance data of all bands.

4. The method for water body classification and chlorophyll a concentration segmented inversion based on dynamic threshold according to claim 1, characterized in that: The formula for constructing the AFAI index in S2 is as follows: Where R rs (λ) is the remote sensing reflectivity of the λ band, Order λ is the band sequence number of the λ band in each image.

5. The method for water body classification and chlorophyll a concentration segmented inversion based on dynamic threshold according to claim 1, characterized in that: The formula for constructing the dynamic threshold function in S3 is as follows: Where, T is the average temperature of the day (℃), M is the current month, a~f are the parameters obtained by regression fitting, R rs (Blue) / R rs (Green)_threshold and AFAI_threshold are the optimal water body classification thresholds for the current date.

6. The method for water body classification and chlorophyll a concentration segmented inversion based on dynamic threshold according to claim 1, characterized in that: The Chla segmented inversion algorithm constructed in S4 is as follows: Where Thr_1 is R rs (Blue) / R rs (Green)_threshold, Thr_2 is AFAI_threshold, Thr_3 and Thr_4 are the thresholds of the water type transition section.

Images