Remote sensing estimation method for chlorophyll concentration of river

Abstract

The invention provides a remote sensing estimation method for the chlorophyll a concentration of a river, and belongs to the technical field of hydrological remote sensing, and the method comprises the following steps: during a satellite transit period, sampling at a plurality of point locations of the river to obtain measured data, obtaining satellite data, carrying out Rayleigh scattering correction on the satellite data to obtain reflectivity after Rayleigh correction, and calculating the chlorophyll a concentration of the river; the method comprises the following steps: screening measured data, establishing an XGBoost model, training the model data, and calculating the chlorophyll a concentration of the river by using the trained XGBoost model. The invention relates to a river chlorophyll a concentration remote sensing estimation method by using Landsat and machine learning. According to the method, based on actually measured chlorophyll a concentration data of the Pearl River basin, a machine learning model capable of estimating the river chlorophyll a concentration is established by using Landsat data after Rayleigh scattering correction, a new thought is provided for monitoring river water quality, and powerful technical support is provided for river eutrophication monitoring and water environment protection and treatment.

Description

Technical Field

[0001] The invention relates to the technical field of hydrological remote sensing, and in particular to a remote sensing estimation method for chlorophyll a concentration in a river. Background Art

[0002] As one of the most important aquatic ecosystems in the world (Grill et al., 2019), rivers are the product of the earth's evolutionary process and are closely related to the survival and development of human society (Best, 2019). Rivers play important hydrological, geological and ecological roles, such as power generation, navigation, water supply and sediment transport (Tang, 2020). They also constitute one of the most diverse habitats in the world (Palmer et al., 2015). Rivers around the world are significantly affected by natural and anthropogenic influences (especially human activities such as the construction of reservoirs and dams, and the discharge of agricultural and industrial wastewater), resulting in serious degradation of river ecosystems (Best, 2019; Butman et al., 2015; Feng et al., 2014a; Hou et al., 2018; Ranasinghe et al., 2019; Sabate et al., 2018; Wu et al., 2023; Yan et al., 2022a), such as the Amazon River

[0003] (Latrubesse et al., 2017; Park and Latrubesse, 2017), Yangtze River (Dai et al., 2014), Yellow River (Lvet al., 2019; Wang et al., 2016) and Seine River (Yane et al., 2021), these rivers have reached a worldwide consensus on this issue. The construction of reservoirs usually changes the ecological environment of natural rivers (Maavara et al., 2020, 2017, 2015; 2017; Wang, 2020), raising the upstream water level and hindering the transport of sediments and algae, regulating dynamic biogeochemistry. This may lead to changes in water quality (Ranetal., 2016), with algae accumulation leading to eutrophication in the upstream waters, while potentially improving the water quality of the mid- and downstream water bodies.

[0004] Chlorophyll-a (Chl-a) is an important indicator of changes in the ecological environment of rivers (Cao et al., 2020). In water bodies, phytoplankton is affected by temperature and tends to gather on the surface of the water body to receive more sunlight (Jiang et al., 2019). Within a certain threshold range, the photosynthetic activity of phytoplankton is positively correlated with temperature, driving them to produce more chlorophyll a concentration, thereby accelerating their reproduction (Kholssi et al., 2023), which is the main reason for the increase in chlorophyll a concentration in water bodies. In addition, human activities introduce nutrients such as nitrogen (N) and phosphorus (P) into rivers (Ma et al., 2011; Zhou et al., 2020), resulting in higher chlorophyll a concentrations near the coast and lower concentrations in the central region. This phenomenon is more obvious in wider river sections. Changes in chlorophyll a concentration reflect the nutrient level and eutrophication of water bodies. Excessive chlorophyll a concentration in water bodies often indicates the presence of excessive nutrients, which can trigger algal blooms (Davidson et al., 2012), leading to deterioration of water quality, reduced dissolved oxygen levels, and death of aquatic organisms (Huang et al., 2020). Regular monitoring of chlorophyll a concentration can timely detect and warn of water quality problems, so that effective management measures can be taken to control pollution sources and reduce the impact of eutrophication on the water environment (Qin et al., 2015). Large-scale monitoring can identify and analyze macro-environmental problems, such as large-scale algal blooms and declining water quality trends, which may not be discovered or accurately assessed through small-scale monitoring (Guan et al., 2020).

[0005] Traditional field surveys for chlorophyll a concentration assessment are costly, time-consuming, and have limited spatial coverage. In contrast, remote sensing has unique advantages in performing such tasks, providing a cost-effective way to monitor changes in chlorophyll a concentrations over large areas (Duan et al., 2012a). Satellites such as MODIS (Chen et al., 2014; Duan et al., 2017; Shi et al., 2020, 2015) and Landsat (Renet al., 2018; Yin et al., 2023; Zheng et al., 2016) provide extensive and continuous coverage, facilitating comprehensive monitoring over a long period of time. Their high temporal and spatial resolution enables the detection of changes in chlorophyll a concentrations at different scales and supports long-term trend analysis. Using these data, previous studies have developed many algorithms for estimating chlorophyll a concentrations (Duan et al., 2012b), thereby enhancing our understanding of the dynamics of chlorophyll a concentrations and their ecological significance. One obstacle to the application of chlorophyll a concentration inversion algorithms is that the inversion models are mainly based on the optical properties of natural lakes and rivers. Previous studies generally used band combinations to monitor chlorophyll a concentrations (Hu, 2009; Wang and Chen, 2024a; Watanabe et al., 2018). However, inland waters present more complex factors, such as water quality and optical properties, which can vary greatly. Even in the same lake, the composition and proportion of optically active substances in water can vary at different times. Therefore, the accuracy and applicability of these models are relatively low in this context, and the estimation performance in highly turbid water bodies is generally insufficient (Neil et al., 2020). As a result, these models often show obvious regional and seasonal characteristics, limiting their applicability. Addressing these gaps remains a current challenge. In recent years, machine learning algorithms (Saganet al., 2020), including deep neural networks (Cao et al., 2019), convolutional neural networks (Pyo et al., 2019), and random forests (Chen et al., 2019), have been used to analyze algal pigments in water bodies. Li et al. (Li et al., 2024) developed a practical method to retrieve chlorophyll a concentrations in fish ponds using an XGBoost model with Landsat OLI images, which showed higher accuracy than previous machine learning methods. Their results also highlighted the improvement in the applicability and accuracy of machine learning algorithms for chlorophyll a retrieval using multi-band Landsat images. In addition, atmospheric conditions have a significant impact on the retrieval of water color parameters. Current atmospheric correction models do not consider regional differences in atmospheric composition, limiting their applicability.

[0006] Reservoirs are common artificial water bodies on rivers. They are usually formed by building dams in the upper or middle reaches of rivers to accumulate water (Guo et al., 2021). Through large-scale hydropower generation and increased irrigation water, they can significantly improve related energy and environmental issues (Lu et al., 2020). More than 70,000 large dams have been built around the world. Reservoirs and dams can regulate and control river flows, especially during flood seasons, effectively reducing the damage caused by floods to downstream areas (Zarfl et al., 2015). More importantly, reservoirs have a certain retention function, which can precipitate and filter sediments, algae residues and other substances brought from upstream, thereby improving the water quality of downstream rivers and slowing down the spread and accumulation of pollutants (Maavara et al., 2020). Most studies focus on specific river sections, lakes, and estuaries (Duane et al., 2012c; Feng et al., 2014b; Lee et al., 2009; Qi et al., 2014), while relatively few studies have examined the longitudinal variation of chlorophyll a concentrations in large river basins. Differences in geographic location, satellite data, and study time spans hinder the applicability of inversion models. In addition, the mechanisms by which cascade reservoirs and artificial dams affect chlorophyll a concentrations in rivers remain unclear.

[0007] Traditional field surveys for assessing chlorophyll a concentrations are costly, time-consuming, and have limited spatial coverage. In contrast, remote sensing has unique advantages in performing such tasks, providing a cost-effective way to monitor changes in chlorophyll a concentrations over large areas. Satellites such as MODIS and Landsat provide extensive and continuous coverage, facilitating comprehensive monitoring over a long period of time. Their high temporal and spatial resolution enables the detection of chlorophyll a concentration changes at different scales and supports long-term trend analysis. One obstacle to the application of chlorophyll a concentration inversion algorithms is that the inversion models are mainly based on the optical properties of natural lakes and rivers. Previous studies have typically used band combinations to monitor chlorophyll a concentrations. However, inland water bodies present more complex factors, such as water quality and optical properties, which can vary greatly. Even in the same water body, the composition and proportion of optically active substances in water can vary from period to period. Therefore, the accuracy and applicability of these models are relatively low in this context, and the estimation performance in highly turbid water bodies is generally insufficient. As a result, these models often exhibit significant regional and seasonal characteristics, limiting their applicability. Addressing these gaps remains a current challenge. Existing technologies are mostly applicable to water bodies such as lakes, estuaries and reservoirs, but the error is relatively high for natural rivers with fast flow and large water exchange, and the uncertainty is high in the extraction of narrow river water surface ranges. Differences in geographical location, satellite data and research time spans hinder the applicability of inversion models. Summary of the invention

[0008] The purpose of the present invention is to provide a remote sensing estimation method for chlorophyll a concentration in rivers to solve the technical problems mentioned in the background technology. The present invention establishes an XGBoost model to calculate the chlorophyll a concentration in rivers, and studies its spatiotemporal variation characteristics using the Pearl River as an example. The research results will help explore the drinking water safety and ecological stability in the middle and lower reaches of the Pearl River, support the healthy development of industries such as ecological environment protection, and verify the accuracy of this method for calculating chlorophyll a concentration in rivers.

[0009] In order to achieve the above object, the technical solution adopted by the present invention is as follows:

[0010] A remote sensing estimation method for chlorophyll a concentration in a river, the method comprising the following steps:

[0011] Step 1: During the satellite transit, sample several points on the river to obtain measured data and obtain satellite data;

[0012] Step 2: Perform Rayleigh scattering correction on the satellite data to obtain the Rayleigh corrected reflectivity.

[0013] Step 3: Screen the measured data;

[0014] Step 4: Build the XGBoost model and train the model data;

[0015] Step 5: Use the trained XGBoost model to calculate the chlorophyll a concentration of the river.

[0016] Further, in step 1, the sampling points are evenly distributed longitudinally, and a portable multi-parameter instrument is used to measure pH, dissolved oxygen and water temperature. Before measurement, the pH meter is calibrated with a standard pH buffer to ensure an accuracy of ±0.01 pH unit. TA is determined by titrating 20 mL of the water sample with HCl three times, and the titration end point is indicated by a mixed indicator of bromocresol green and methyl red. The river width is determined by a laser rangefinder, and the water depth is measured by a sonar depth sounder. 500 mL of water samples are collected from 0-50 cm on the surface of the center line of the river and stored at 4°C immediately before analysis. For the analysis of chlorophyll a, the water samples are filtered using a pre-combusted 0.7 μm glass fiber membrane, then extracted with 30 mL of acetone, and measured at 665 nm using a UV-1700 ultraviolet-visible spectrophotometer with an accuracy of approximately ±1.2 μgL-1. After filtering through a 0.45 μm nitrocellulose membrane, total phosphorus and total nitrogen are measured using an automatic flow fraction analyzer.

[0017] Furthermore, in step 2, in SeaDAS 8.4.0, the Level-1 product of the Landsat satellite data is corrected for Rayleigh scattering to obtain the Rayleigh-corrected reflectance, a global surface water dataset is used to depict river water bodies, and then the normalized difference water index is used as a threshold filter to exclude shallow water areas to reduce the impact of bottom reflection.

[0018] Furthermore, in step 3, the measured data were screened including: 1. Any sampling points covered by clouds in the RGB image were excluded, 2. Tests showed that samples with abnormally higher chlorophyll a levels than the set value would affect the establishment of the model, and samples with extreme chlorophyll a concentrations were excluded, 3. Samples covered by algal blooms were removed based on FAI>-0.04, 4. When obtaining Rrc data, reflectance values ​​were extracted from a 3×3 pixel window centered on each sampling point and averaged to ensure the consistency of surrounding pixel values. After screening, the field data sites were matched with the OLI images, and then they were randomly divided into training and validation datasets.

[0019] Furthermore, in step 4, when establishing the XGBoost model, the XGBoost package in python is used to implement it, and the hyperparameters are adjusted. The hyperparameters include learning rate, maximum tree depth, subsampling rate, number of estimators and regularization parameters, and the optimal hyperparameters are gradually determined using grid search.

[0020] Furthermore, 5-fold cross validation is used in step 4 to ensure that the training dataset is randomly divided into different parts, and 2 / 3 of the data is selected as the model training dataset and 1 / 3 as the validation dataset.

[0021] Furthermore, in step 4, the same dataset is used to train two machine learning models, GBDT and RF, based on R 2 , RMSE, MAPE and MAE were used to compare and evaluate the accuracy in calculating chlorophyll a, and then the XGBoost model with the highest accuracy was selected as the final chlorophyll a concentration estimation model.

[0022] Remote sensing reflectance data of each band after Rayleigh scattering correction, including: Rrc(443), Rrc(482), Rrc(561), Rrc(655), Rrc(865), Rrc(1609), Rrc(2201); and their different forms of combined data, including BGI(Rrc(443) / Rrc(561)), RGI(Rrc(655) / Rrc(561)), NIR-RI(Rrc(865) / Rrc(655)) and FAI; as well as pixel spatiotemporal information.

[0023] The above reflectivity data was combined with the measured chlorophyll a concentration data on the day of the satellite pass as the training data set to train the XGBoost model.

[0024] Landsat OLI level-1 image data was used. This study excluded the traditional atmospheric correction method and performed atmospheric correction on the first-level remote sensing image products in SeaDAS8.4.0 to achieve Rayleigh corrected reflectance (Rrc). After correction, the image showed excellent results in identifying water bodies.

[0025] , Rrc(443), Rrc(482), Rrc(561), Rrc(655), Rrc(865), and Rrc(1609) are subtracted from Rrc(2201) in order to eliminate the atmospheric effect. In addition, the SWIR reflectivity threshold is set to Rrc(2201)>0.018 to exclude pixels covered by clouds.

[0026] The measured data used for modeling were screened, including: 1. Any sampling points covered by clouds in the RGB image were excluded. 2. Extremely high chlorophyll a concentrations are rare. Tests have shown that samples with abnormally high chlorophyll a levels will significantly affect the establishment of the model. Therefore, this study excluded samples with extreme chlorophyll a concentrations. 3. Samples covered by algal blooms were removed based on FAI>-0.04. 4. When obtaining Rrc data, this study extracted reflectance values ​​from a 3×3 pixel window centered on each sampling point and averaged them to ensure consistency in surrounding pixel values.

[0027] The present invention uses Landsat data corrected for Rayleigh scattering, combined with the synchronous satellite-ground measured data of rivers, and compares the traditional band combination algorithm and machine learning method to develop an estimation model for river chlorophyll a concentration suitable for Landsat sensors. This study uses measured data to fully evaluate the performance of the model, which can not only accurately estimate the chlorophyll a concentration of rivers in a large range, but also analyze long time series. This invention breaks through the limitations of remote sensing estimation of river chlorophyll a concentration over a large range and for a long time, and provides a new idea for monitoring river water quality. It helps to understand the drinking water safety and ecological stability in the middle and lower reaches of the river, support the healthy development of industries such as fisheries, agriculture, and tourism, thereby generating significant economic and social benefits, and providing strong technical support for river eutrophication monitoring and water environment protection and management.

[0028] From the above technical solutions, it can be seen that the remote sensing estimation method of river chlorophyll a concentration using Landsat and machine learning invented in this study. This method is based on the measured chlorophyll a concentration data of the Pearl River Basin, and uses the Landsat data corrected by Rayleigh scattering to establish a machine learning model that can estimate the chlorophyll a concentration of rivers, which provides a new idea for monitoring river water quality and provides strong technical support for river eutrophication monitoring and water environment protection and management.

[0029] The present invention has the following beneficial effects due to the adoption of the above technical solution:

[0030] (1) The present invention uses a remote sensing estimation method for river chlorophyll a concentration using Landsat and machine learning. This method is based on the measured chlorophyll a concentration data of the Pearl River Basin and uses Rayleigh scattering corrected Landsat data to establish a machine learning model that can estimate the chlorophyll a concentration of rivers, providing a new idea for monitoring river water quality and providing strong technical support for river eutrophication monitoring and water environment protection and management.

[0031] (2) The temporal and spatial variation characteristics of the Pearl River were studied using the Pearl River as an example. The research results will help to explore the drinking water safety and ecological stability in the middle and lower reaches of the Pearl River, support the healthy development of industries such as ecological environment protection, and verify the accuracy of this method in calculating chlorophyll a concentration in rivers. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1 It is a collection diagram of an embodiment of the present invention;

[0033] Figure 2 This is the main process of the present invention using the XGBoost model to predict chlorophyll a concentration;

[0034] Figure 3 This is a performance graph of the three models XGBoost, GBDT and RF evaluated in estimating the chlorophyll a concentration in the Pearl River Basin using a validation data set (N=35) in the present invention;

[0035] Figure 4 The present invention is a distribution diagram of the annual average chlorophyll a concentration of Xijiang River from the source to the outlet from 2013 to 2022;

[0036] Figure 5 This is a seasonal variation diagram of chlorophyll a concentration in different sections of the Pearl River in the past ten years;

[0037] Figure 6 This is a longitudinal multi-year average variation diagram of chlorophyll a concentration in the main tributaries of the Pearl River of the present invention;

[0038] Figure 7 It is the annual and seasonal characteristic diagram of chlorophyll a concentration in different sections (R1 to R5) of Xijiang River of the present invention;

[0039] Figure 8 It is the annual and seasonal characteristic diagram of chlorophyll a concentration in different sections (R1-R19) of the Pearl River of the present invention;

[0040] Fig. 9 It is a graph showing the changes in chlorophyll a concentrations of four representative large reservoirs of the present invention. DETAILED DESCRIPTION

[0041] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below with reference to the accompanying drawings and preferred embodiments. However, it should be noted that many details listed in the specification are only for the purpose of enabling the reader to have a thorough understanding of one or more aspects of the present invention, and these aspects of the present invention can be implemented even without these specific details.

[0042] A remote sensing estimation method for chlorophyll a concentration in a river, the method comprising the following steps:

[0043] Step 1. During the Landsat transit, field sampling was carried out in the Pearl River to obtain measured data. From July 19 to August 3, 2021 (flood season) and December 7 to 22 (dry season), field sampling was carried out at 62 points, and a total of 124 field data sets were collected. The sampling points were carefully selected and evenly distributed longitudinally to reflect the spatial characteristics of chlorophyll a concentration in the Pearl River Basin. At the same time, field confirmation was carried out to ensure that the sampling points were far away from artificial discharge outlets and reflected the true characteristics of the surrounding waters. At each sampling point, a portable multi-parameter instrument (Multi3,420, WTWGmbH, Germany) was used to measure pH, dissolved oxygen (DO), and water temperature (WT). Before measurement, the pH meter was calibrated with a standard pH buffer to ensure an accuracy of ±0.01 pH unit. TA was determined by titrating 20 mL of water sample three times with HCl (0.02 molL-1), and the titration end point was indicated by a mixed indicator of bromocresol green and methyl red. The width of the river channel was determined using a laser rangefinder, and the water depth was measured using a sonar depth sounder. 500 mL of water samples (n = 124) were collected from the surface layer (0-50 cm) at the center line of the river and stored at 4 °C immediately before analysis. For the analysis of chlorophyll a, the water samples were filtered using a pre-combusted 0.7 μm glass fiber membrane (Whatman GF / F), then extracted through 30 mL of acetone and measured using a UV-1700 UV-visible spectrophotometer (Shimadzu, Kyoto, Japan) at 665 nm with an accuracy of approximately ±1.2 μgL-1. Total phosphorus (TP) and total nitrogen (TN) were measured using an automatic flow fraction analyzer after filtration through a 0.45 μm nitrocellulose membrane.

[0044] Step 2: In SeaDAS 8.4.0, the Level-1 product of Landsat data is corrected for Rayleigh scattering to obtain the Rayleigh corrected reflectance (Rrc). Furthermore, this study uses the Global Surface Water (GSW) dataset to depict river water bodies, and then uses the Normalized Difference Water Index (NDWI) as a threshold filter to exclude shallow water areas to reduce the impact of bottom reflection. Furthermore, this study uses Rrc(443), Rrc(482), Rrc(561), Rrc(655), Rrc(865), and Rrc(1609) minus Rrc(2201) in sequence to eliminate atmospheric effects. In addition, the SWIR reflectance threshold is set to Rrc(2201)>0.018 to exclude pixels covered by clouds. Furthermore, they are combined into different forms of data, including BGI (Rrc(443) / Rrc(561)), RGI (Rrc(655) / Rrc(561)), NIR-RI (Rrc(865) / Rrc(655)) and FAI; as well as pixel spatiotemporal information.

[0045] Step 3. Before building the XGBoost model, the measured data was screened, including: 1. Any sampling points covered by clouds in the RGB image were excluded. 2. Extremely high chlorophyll a concentrations are rare. Tests have shown that samples with abnormally high chlorophyll a levels will significantly affect the establishment of the model. Therefore, this study excluded samples with extreme chlorophyll a concentrations. 3. Based on FAI>-0.04, algal bloom-covered sample points were removed. 4. When obtaining Rrc data, this study extracted reflectance values ​​from a 3×3 pixel window centered on each sampling point and averaged them to ensure consistency in surrounding pixel values. After screening, 105 field data sites were matched with the OLI image. They were then randomly divided into a training data set (N=70) and a validation data set (N=35).

[0046] Step 4: When building the XGBoost model, this study used the XGBoost package in Python and adjusted its hyperparameters, including learning rate, maximum tree depth, subsampling rate, number of estimators, and regularization parameters. This study used grid search to gradually determine the optimal hyperparameters.

[0047] In addition, this study used 5-fold cross validation to ensure that the training dataset was randomly divided into different parts, selecting 2 / 3 of the data as the model training dataset and 1 / 3 as the validation dataset.

[0048] In addition, this study used the same dataset to train two machine learning models, GBDT and RF, based on R 2, RMSE, MAPE and MAE to compare and evaluate their accuracy in calculating chlorophyll a. The XGBoost model with the highest accuracy (R 2 =0.8616, RMSE = 0.8578 μgL -1 ,MAPE=13.92%,Bias=0.0673,MAE=0.7551) was used as the final chlorophyll a concentration estimation model.

[0049] Step 5. Use the trained XGBoost model to calculate the chlorophyll a concentration of the Pearl River. In this study, a periodic trend of chlorophyll a concentration in the mainstream of the Pearl River was observed. The chlorophyll a concentration was highest at the source of the river, and the chlorophyll a concentration fluctuated significantly when the river flowed through the reservoir. This may be due to the storage of water in the cascade reservoirs. To study the impact of these reservoirs on the chlorophyll a concentration of the mainstream and major tributaries, this study divided the Pearl River into 19 sections (labeled R1-R19). According to the spatial distribution of cascade reservoirs, dams that are particularly close to each other are regarded as the same dam. In these sections, the main streams M-1 to M-5 are labeled R1 to R5, and the tributaries T-1 to T-7, NR, and ER are labeled R6-R19. This organization can more clearly show the changes in chlorophyll a concentration in each section. Figure 5 The results show that from 2013 to 2022, the chlorophyll a concentration in the Xijiang River was mainly between 2 and 10 μg / L. -1 The highest chlorophyll a concentration was found in the source section R1, and most of the chlorophyll a levels were between 6 and 10 μg / L. -1 After passing through the first larger reservoir TSQ, the chlorophyll a concentration decreased significantly. In the R2-R3 section, the chlorophyll a concentration remained at 4-8 μg L -1 range. The chlorophyll a concentration in the R4-R5 section gradually increased, exceeding the level in the R2-R3 section, but still lower than that in the R1 section. The chlorophyll a concentration in the Xijiang River water body generally showed a trend of decreasing first and then increasing, and the fluctuation was more obvious in the downstream area. This shows that although the chlorophyll a concentration in the source area of ​​the Xijiang River is the highest, after the reservoir is impounded, most of the suspended matter and nutrients are deposited in the upstream reservoir, inhibiting the growth of algae downstream. The increase in chlorophyll a concentration near the estuary is mainly due to the nutrients input from the downstream tributaries, which promotes the growth of algae. This shows that reservoirs and artificial dams have an effective regulatory effect on the transport of upstream substances, thereby affecting the longitudinal distribution of chlorophyll a in the Pearl River water body.

[0050] In addition, this study also calculated the average chlorophyll a concentrations of the Pearl River mainstream and its tributaries at different time scales during 2013-2022, and presented their characteristics and relationships ( Figure 5-8). This example applies the present invention to 718 scenes of Landsat OLI data of the Pearl River from 2013 to 2022, and simulates the spatiotemporal distribution characteristics of the chlorophyll a concentration of the Pearl River over the past decade. The chlorophyll a concentration of the Pearl River is highest in summer and lowest in autumn and winter. In the past decade, it has generally shown an upward trend, declining from 2013 to 2015, reaching the lowest value in 2015, and then gradually rising again, reaching a peak in 2020.

[0051] In order to more clearly illustrate the impact of reservoirs on the chlorophyll a concentration in the Pearl River, this study selected four large reservoirs to observe the changes in chlorophyll a concentration before, during, and after water was pumped into the reservoirs. Fig. 9 As shown in the figure, the chlorophyll a concentration in the river before the reservoir enters the river is significantly higher than that after the reservoir leaves the river, and the chlorophyll a concentration is highest in the reservoir. This phenomenon is mainly attributed to the reduction of flow in the reservoir, which leads to the sedimentation of nutrients transported by the reservoir, thus causing the chlorophyll a concentration to increase. In addition, only a small amount of algae can pass through the artificial dam and enter the downstream river section, resulting in a high chlorophyll a concentration before the reservoir enters the river, and then a decrease in the downstream river section.

[0052] Figure 1 In the figure, (a) the enlarged image of the sampling points on the high-resolution satellite image; (b) the spatial configuration of the sampling points in the mainstream and main tributaries; (c) the photo of the sampling site in the Pearl River. All photos were taken by the author. Figure 2 In the figure, the main process of using XGBoost model to predict chlorophyll a concentration. Figure 3 In the paper, a validation dataset (N=35) was used to evaluate the performance of the three models, XGBoost, GBDT and RF, in estimating the chlorophyll a concentration in the Pearl River Basin. The evaluation showed that the XGBoost model achieved the highest performance. Figure 4 In the figure, the distribution of annual average chlorophyll a concentration from the source to the outlet of the Xijiang River from 2013 to 2022. The mainstream of the Xijiang River is divided into R1-R5 according to the location of the reservoir. Longtan Reservoir is the largest reservoir in the Xijiang River Basin. Figure 5 Seasonal variations of chlorophyll-a concentrations in different reaches of the Pearl River in the past decade. Figure 6 Longitudinal multi-year average changes in chlorophyll-a concentrations in the main tributaries of the Pearl River. Figure 7 Interannual and seasonal characteristics of chlorophyll a concentration in different reaches (R1-R5) of the Xijiang River. Figure 8 Interannual and seasonal characteristics of chlorophyll a concentrations in different sections of the Pearl River (R1-R19). Fig. 9In the figure, taking four representative large reservoirs as examples, the changes in chlorophyll a concentration before entering the reservoir, in the reservoir and outflowing from the reservoir are shown (a); Figure (b) depicts in detail the fluctuations in chlorophyll a concentration in the first large reservoir in Figure (a): before entering the reservoir, the chlorophyll a concentration level is low (<5μgL-1), then reaches a peak value in the reservoir, close to 8μgL-1, and then decreases after leaving the reservoir.

[0053] This method is a cost-effective way to monitor the changes in chlorophyll a concentration over a large area and for a long time. At the same time, an obstacle to the application of chlorophyll a concentration inversion algorithms is that the inversion models are mainly based on the optical properties of natural lakes and rivers. Previous studies usually use band combinations to monitor chlorophyll a concentrations. However, inland water bodies have more complex factors, such as water quality and optical properties, which can vary greatly. Even in the same water body, the composition and proportion of optically active substances in water can vary at different times. Therefore, the accuracy and applicability of these models are relatively low in this context, and the estimation performance in highly turbid water bodies is usually insufficient. Therefore, these models usually show obvious regional and seasonal characteristics, which limits their applicability. Addressing these gaps remains a current challenge. Existing technologies are mostly applicable to water bodies such as lakes, estuaries, and reservoirs, while the errors for natural rivers with fast flow rates and large water exchange are relatively high, and the uncertainty is high in the extraction of narrow river water surface ranges. Differences in geographical location, satellite data, and research time spans hinder the applicability of inversion models. In addition, the mechanism by which cascade reservoirs and artificial dams affect the chlorophyll a concentration in rivers remains unclear. To address this problem, this study established an XGBoost model to calculate the chlorophyll a concentration in rivers, and studied the spatiotemporal variation characteristics of chlorophyll a using the Pearl River and the reservoirs built on the Pearl River as examples. The results will help explore the drinking water safety and ecological stability in the middle and lower reaches of the Pearl River, support the healthy development of industries such as ecological environmental protection, and verify the accuracy of this method for calculating chlorophyll a concentration in rivers.

[0054] Matters not covered by the present invention are known technologies.

[0055] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principle of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A remote sensing estimation method for chlorophyll a concentration in a river, characterized in that: The method comprises the following steps: Step 1: During the satellite transit, sample several points on the river to obtain measured data and obtain satellite data; Step 2: Perform Rayleigh scattering correction on the satellite data to obtain the Rayleigh corrected reflectivity. Step 3: Screen the measured data; Step 4: Build the XGBoost model and train the model data; Step 5: Use the trained XGBoost model to calculate the chlorophyll a concentration of the river.

2. The remote sensing estimation method for chlorophyll a concentration in a river according to claim 1, characterized in that: In step 1, the sampling points are evenly distributed longitudinally. A portable multi-parameter instrument is used to measure pH, dissolved oxygen and water temperature. Before measurement, the pH meter is calibrated with a standard pH buffer to ensure an accuracy of ±0.01 pH unit. TA is determined by titrating 20 mL of water sample with HCl three times, and the titration end point is indicated by a mixed indicator of bromocresol green and methyl red. The river width is determined by a laser rangefinder, and the water depth is measured by a sonar depth sounder. 500 mL of water sample is collected from 0-50 cm on the surface of the center line of the river and stored at 4°C immediately before analysis. For the analysis of chlorophyll a, the water sample is filtered using a pre-combusted 0.7 μm glass fiber membrane, then extracted with 30 mL of acetone, and measured at 665 nm using a UV-1700 ultraviolet-visible spectrophotometer with an accuracy of approximately ±1.2 μgL-1. After filtering through a 0.45 μm nitrocellulose membrane, the total phosphorus and total nitrogen are measured using an automatic flow fraction analyzer.

3. The remote sensing estimation method for chlorophyll a concentration in a river according to claim 1, characterized in that: In step 2, in SeaDAS 8.4.0, the Level-1 product of Landsat satellite data is corrected for Rayleigh scattering to obtain the Rayleigh-corrected reflectance, a global surface water dataset is used to depict river water bodies, and then the normalized difference water index is used as a threshold filter to exclude shallow water areas to reduce the impact of bottom reflection.

4. The remote sensing estimation method for chlorophyll a concentration in a river according to claim 1, characterized in that: In step 3, the measured data were screened including:

1. Any sampling points covered by clouds in the RGB image were excluded, 2. Tests showed that samples with abnormally higher chlorophyll a levels than the set value would affect the establishment of the model, and samples with extreme chlorophyll a concentrations were excluded, 3. Samples covered by algal blooms were removed based on FAI>-0.04, 4. When obtaining Rrc data, reflectance values ​​were extracted from a 3×3 pixel window centered on each sampling point and averaged to ensure the consistency of surrounding pixel values. After screening, the field data sites were matched with the OLI images, and then they were randomly divided into training and validation datasets.

5. The remote sensing estimation method for chlorophyll a concentration in a river according to claim 1, characterized in that: In step 4, when building the XGBoost model, the XGBoost package in Python is used to implement it and adjust the hyperparameters, including the learning rate, maximum tree depth, subsampling rate, number of estimators, and regularization parameters. The grid search is used to gradually determine the optimal hyperparameters.

6. The remote sensing estimation method for chlorophyll a concentration in a river according to claim 5, characterized in that: In step 4, 5-fold cross validation is used to ensure that the training dataset is randomly divided into different parts. 2 / 3 of the data is selected as the model training dataset and 1 / 3 as the validation dataset.

7. The remote sensing estimation method for chlorophyll a concentration in a river according to claim 5, characterized in that: In step 4, the same dataset is used to train two machine learning models, GBDT and RF, based on R 2 , RMSE, MAPE and MAE four indices were used to compare and evaluate the accuracy in calculating chlorophyll a, and then the XGBoost model with the highest accuracy was selected as the final chlorophyll a concentration estimation model.

Images