Lake water volume monitoring method based on satellite image and CatBoost-XGBoost model

Abstract

The invention discloses a lake water volume monitoring method based on a satellite image and a CatBoost-XGBoost model, and relates to the technical field of remote sensing application and hydrological monitoring, and the method comprises the steps: S1, obtaining and processing multi-source data: obtaining a Sentinel-2A satellite multispectral image, carrying out atmospheric correction, synchronously collecting actually measured water depth data, carrying out space-time matching, and constructing a training data set; s2, a CatBoost-XGBoost combination model is constructed and optimized, the CatBoost-XGBoost combination model is constructed, the output of the CatBoost model and the output of the XGBoost model are fused through weight distribution, and water depth inversion is achieved; s3, water depth inversion and precision verification: outputting pixel-level water depth based on the combined model, and verifying the inversion precision by adopting multiple indexes; and S4, water body range extraction and water volume calculation: extracting a water body boundary based on a water body index, and calculating the water volume of the water body through spatial integration in combination with the inversion water depth. By adopting the method of the steps, a machine learning method is introduced into water volume calculation, the lake water depth is inverted through the machine learning method, and the water volume is further calculated by utilizing double integration of the lake area and the water depth.

Description

Technical Field

[0001] The present invention relates to the field of remote sensing application and hydrological monitoring technology, and in particular to a lake water volume monitoring method based on satellite imagery and a CatBoost-XGBoost model. Background Art

[0002] Lakes are important surface water carriers, regulating water volume, improving water quality, and modulating local climate, playing a crucial role in the global water cycle. Furthermore, lakes are a vital resource for life on Earth, including humans, providing food and habitat. Water volume is a key characteristic of lakes, and a reduction in water volume can directly reduce the area of ​​surrounding wetlands, impacting aquatic populations and threatening the ecological balance of the region.

[0003] In previous studies, monitoring lake characteristics often relied on traditional manual observation methods, which required a significant amount of manpower, material resources, and time. Furthermore, due to the involvement of human factors, the observation results may contain a certain degree of subjective error, affecting the accuracy and credibility of the data. Furthermore, manual observation is also limited by factors such as the natural environment and weather conditions, making it difficult to achieve continuous and stable monitoring in complex and changing environments. With the continuous development and advancement of science and technology, lake monitoring methods based on satellite remote sensing technology have gradually become the mainstream of research. They can remotely monitor multiple characteristics such as lake area, water level, and water volume, without being restricted by geographical location and weather conditions, and have achieved continuous observation and recording of large-scale lakes.

[0004] Most existing remote sensing-based lake water volume calculation methods use measured lake characteristic data to calculate empirical parameters. These parameters then exploit the relationship between water volume and other characteristics to construct empirical relationships. This approach relies heavily on measured data, requiring a large amount of high-precision data to establish the relationships. Furthermore, these relationships are extremely limited in scope, generally applicable only to the study area or areas with similar geographical conditions. Consequently, constructing these empirical relationships is not only laborious but also inherently lacks rigor. Therefore, it is necessary to develop a more accurate, rigorous, and widely applicable method for calculating lake water volume. Summary of the Invention

[0005] The purpose of the present invention is to provide a lake water volume monitoring method based on satellite imagery and the CatBoost-XGBoost model, introduce machine learning methods into water volume calculation, invert lake water depth through machine learning methods, and further calculate water volume using the double integral of lake area and water depth.

[0006] To achieve the above objectives, the present invention provides a lake water volume monitoring method based on satellite imagery and the CatBoost-XGBoost model, which includes the following steps:

[0007] S1. Multi-source data acquisition and processing: Acquire Sentinel-2A satellite multispectral images and perform atmospheric correction, simultaneously collect measured water depth data and perform spatiotemporal matching to construct a training dataset;

[0008] S2. CatBoost-XGBoost combination model construction and optimization: Construct a CatBoost-XGBoost combination model, fuse the outputs of the CatBoost model and the XGBoost model through weight distribution, and realize water depth inversion;

[0009] S3. Water depth inversion and accuracy verification: Output pixel-level water depth based on the combined model, and use multiple indicators to verify the inversion accuracy;

[0010] S4. Water body range extraction and water volume calculation: Extract the water body boundary based on the water body index, and calculate the water volume of the water body through spatial integration combined with the inverted water depth.

[0011] Preferably, in S1, the multi-source data acquisition and processing specifically includes:

[0012] S1.1. Obtain multispectral images from the Sentinel-2A satellite covering multiple spectral bands of the target waters. Resample the images to a uniform spatial resolution based on the research scale, perform atmospheric correction using the Sen2cor plug-in, and generate surface reflectance data.

[0013] S1.2. Collect measured water depth data, including point data from different years and different sources. Use water level difference to adjust the measured water depth data according to the satellite imaging time and use it as the measured data at the satellite imaging time. Randomly divide 70%-80% of the data into a training set and 20%-30% of the data into a validation set.

[0014] Preferably, in S2, CatBoost-XGBoost combination model construction and optimization construction specifically include:

[0015] S2.1. The radiance values ​​of all bands of the multispectral image and the measured water depth data are used as training data, which are input into the CatBoost model and the XGBoos model for model training, and the inverted water depth is calculated and output;

[0016] S2.2. Combined model fusion: Combine the outputs of the CatBoost model and the XGBoos model in a weighted linear manner so that the output of the combined model is:

[0017] Z i =w1z 1i +w2z2i

[0018] Among them, Z i is the water depth value inverted by the combined model; z 1i 、z 2i are the inverted water depth values ​​of the CatBoost and XGBoost models respectively; w1 and w2 are the weight coefficients of the CatBoost model and the XGBoost model in the combined model respectively, and they satisfy w1+w2=1;

[0019] The coefficient of determination R 2 Maximization is the optimization goal, and the weight optimization strategy is implemented.

[0020] Preferably, the weight optimization strategy in S2 includes:

[0021] Set the initial weights w1 = 1, w2 = 0, traverse w1∈[0,1] with a step size of 0.05, and calculate R under each weight combination. 2 , finally choose R 2 The largest combination.

[0022] Preferably, in S3, the water depth inversion and accuracy verification specifically include:

[0023] R 2 The coefficient, root mean square error (RMSE), mean absolute error (MAE) and mean relative error (MRE) are used as accuracy indicators to evaluate the inversion accuracy. The calculation formula is as follows:

[0024]

[0025] Where n is the number of test points; ΔZ i is the difference between the actual water depth value and the inverted water depth value at the i-th test point, is the difference between the actual water depth value at the test point and its average value, is the average value of the actual water depth at the test point, R 2 The larger it is, the smaller the RMSE, MAE, and MRE are, indicating that the inversion accuracy is higher.

[0026] Preferably, S4, water body range extraction and water volume calculation specifically includes:

[0027] S4.1. Extract water body boundaries based on the normalized difference water index (NDMI). The formula is:

[0028]

[0029] Among them, B1 represents the pixel value of the green band, and B2 represents the pixel value of the near-infrared band;

[0030] The image of the main area of ​​Hulun Lake was cropped with a rectangular frame, and the surrounding non-main lake water bodies were cropped. The NDWI value of the cropped image was calculated, and the area with NDWI>0 was determined to be water body, thereby determining the scope of the water body.

[0031] S4.2. Count the number of valid pixels within the water body and calculate the water body area S. The formula is:

[0032] S=n×S cell

[0033] Among them, S cell is the area of ​​a single pixel, and n is the number of pixels;

[0034] S4.3. Combine the inverted water depth grid data with the water body range and calculate the water volume V by spatial integration. The formula is:

[0035]

[0036] Among them, S cell is the area of ​​a single pixel, h i is the pixel water depth value, and the water volume changes are dynamically calculated on an interannual and seasonal scale.

[0037] According to the specific embodiments provided by the present invention, the present invention discloses the following technical effects:

[0038] (1) This application integrates data acquisition, model inversion, accuracy verification, and water volume calculation into a coherent process. A high-quality training set is constructed by processing satellite imagery and measured data. Water body indices are used to automatically extract water boundaries, and water volume is calculated using depth data. This entire process, from data input to result output, can be automated without frequent human intervention, providing a standardized technical framework for dynamic monitoring of lake water volume, effectively reducing operational costs and human error.

[0039] (2) A combined CatBoost and XGBoost model was used, combining the advantages of both: the former excels at processing classification features and reducing prediction bias, while the latter can fit data with high precision and improve the inversion effect of local details. By optimizing the model weights, the combined model significantly improves the water depth inversion accuracy compared to a single model, especially in areas with complex terrain and variable spectral characteristics. This effectively solves the problem of single models being prone to overfitting or underfitting, making the inversion results closer to the actual water depth distribution.

[0040] (3) The preprocessing phase ensures the temporal and spatial consistency and reliable quality of the input data by correcting the measured data and unifying the image resolution. Water body extraction uses the water index method, which can accurately distinguish between water bodies and land and adapt to the changes in water boundaries in different seasons and environments. Water volume calculation is based on the principle of pixel accumulation, without relying on empirical formulas for specific areas. It can be flexibly applied to various types of lakes and reservoirs, especially in areas with complex lake bottom topography. Multi-dimensional precision verification indicators ensure the reliability of the results and provide a scientific and quantitative basis for water resource management and ecological protection.

[0041] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0043] Figure 1 Schematic diagram of a flow chart of an embodiment of a lake water volume monitoring method based on satellite imagery and a CatBoost-XGBoost model according to the present invention;

[0044] Figure 2 Schematic diagram of the positions of measured points according to an embodiment of the present invention;

[0045] Figure 3 This is a flowchart of the CatBoost-XGBoost combined model modeling according to an embodiment of the present invention;

[0046] Figure 4 This is a result map of the Hulun Lake range extracted from 2016 to 2023 according to an embodiment of the present invention;

[0047] Figure 5 This is a graph showing inter-annual and intra-annual changes in the area of ​​Hulun Lake according to an embodiment of the present invention;

[0048] Figure 6 This is a depth map of Hulun Lake from 2016 to 2023 according to an embodiment of the present invention;

[0049] Figure 7 A graph showing the inter-annual and intra-annual variations of the average water depth of Hulun Lake according to an embodiment of the present invention;

[0050] Figure 8 This is a graph showing the inter-annual and intra-annual variations in the water volume of Hulun Lake according to an embodiment of the present invention. DETAILED DESCRIPTION

[0051] 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.

[0052] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0053] Example

[0054] This example uses the Hulun Lake area as an example. Using Sentinel-2A full-band data and a portion of measured water depth data, the water body index method is used to extract the water range of the lake. Machine learning methods are used to invert the water depth of Hulun Lake within this range. The water volume of Hulun Lake is calculated by integrating the lake area and water depth. Finally, a summary analysis is conducted to determine the interannual variation characteristics and trends of Hulun Lake's water volume from 2016 to 2023 and within 2019.

[0055] like Figure 1 As shown in FIG, the lake water volume monitoring method based on satellite imagery and the CatBoost-XGBoost model proposed in the present invention is applied, and the steps are as follows:

[0056] S1. Multi-source data acquisition and processing: Acquire Sentinel-2A satellite multispectral images and perform atmospheric correction. Simultaneously collect measured water depth data and perform spatiotemporal matching to construct a training dataset. Specifically, this includes:

[0057] S1.1. Obtain multispectral images from the Sentinel-2A satellite covering multiple spectral bands of the target waters. Resample the images to a uniform spatial resolution based on the research scale, perform atmospheric correction using the Sen2cor plug-in, and generate surface reflectance data.

[0058] The Sentinel-2A satellite was launched in 2015 and has been in operation for 10 years. It orbits at an altitude of 768 km and has a revisit period of 10 days. Sentinel-2A carries a multispectral imager (MSI) that covers 13 spectral bands, from the visible and near-infrared to the shortwave infrared. Each band has a different spatial resolution: four visible and near-infrared bands have a resolution of 10 meters, six red-edge and shortwave infrared bands have a resolution of 20 meters, and three atmospheric correction bands have a resolution of 60 meters. The swath width reaches 290 km. With high-precision geometric positioning and high-fidelity data quality, it provides accurate observation data. Its observation capabilities in multiple spectral bands and timescales enable it to obtain spectral information on various ground objects at different timescales. Sentinel-2A includes three bands in the red-edge range, enabling effective vegetation monitoring. Tables 1 and 2 show the specifications and spectral information of the Sentinel-2A satellite's multispectral imager.

[0059] Table 1 Sentinel-2A satellite multispectral imager indicators

[0060]

[0061] Table 2 Sentinel-2A spectral information

[0062]

[0063]

[0064] This example uses Sentinel-2A satellite multispectral imagery data for the Hulun Lake region. Taking into account factors such as the coverage of the study area, cloud cover, and consistent imaging time, the images used to observe interannual water volume variations were acquired during the summer flood season (June to August) from 2016 to 2023, specifically June 2, 2016, August 19, 2017, July 25, 2018, July 17, 2019, June 1, 2020, June 6, 2021, July 1, 2022, and June 26, 2023. The images used for intra-annual variations were acquired in the middle of each month in May, June, July, August, September, and October 2019, when Hulun Lake was not frozen. To facilitate water volume calculations, all images were resampled to a spatial resolution of 30 meters. Atmospheric correction was performed using the Sen2cor plugin to generate L2A data.

[0065] S1.2. Collect measured water depth data, including point data from different years and different sources. Use water level difference to adjust the measured water depth data according to the satellite imaging time and use it as the measured data at the satellite imaging time.

[0066] The measured water depth data from 54 sample points in 1988, 1991, 2008 and 2021 were collected, such as Figure 2 The data for 1988 and 1991 were obtained from field measurements at local fisheries, while the data for 2008 and 2021 were obtained from scientific research measurements at universities. Because there are too few measured depth points in a single year to be used for training the depth model, and because they are not synchronized with the satellite imagery data used, the water levels of Hulun Lake for each year were obtained. The actual measured data were then adjusted to the satellite imaging time using the water level difference, and then used as the measured data for the satellite imaging time. Fifty of these adjusted points were randomly selected as the training set, and 20 of these points were randomly selected as the validation set.

[0067] S2. CatBoost-XGBoost combination model construction and optimization: Construct a CatBoost-XGBoost combination model, and fuse the outputs of the CatBoost model and the XGBoost model through weight distribution to achieve water depth inversion.

[0068] S2.1. The radiance brightness values ​​of all bands of the multispectral image and the measured water depth data are used as training data, which are input into the CatBoost model and the XGBoos model for model training respectively, and the model accuracy is evaluated on the test data.

[0069] The CatBoost model is a new gradient boosting algorithm with unique advantages. It can manage Cat logical variables, effectively prevent overfitting, and address issues such as inability to directly handle categorical variables and prediction bias. During CatBoost model training, a set of decision trees is built sequentially to correct the gradient bias of each subsequent tree. In other words, each decision tree learns from the previous one and influences the next, thereby improving model performance. Unlike other gradient boosting algorithms, the CatBoost model is more efficient and has superior performance in prediction and classification. Therefore, it is widely used in various machine learning tasks.

[0070] The XGBoost (Extreme Gradient Boosting) algorithm is a powerful tool for solving nonlinear regression problems. It processes data very quickly, significantly outperforming traditional boosted tree algorithms. It offers exceptional performance for both classification and regression problems, and represents a new milestone for the GBDT (Gradient Boosting Decision Tree) algorithm. It provides new insights for various scientific research based on machine learning models.

[0071] The XGBoost model fits the residual between the existing model and the result by continuously absorbing new learners, thereby improving the accuracy of the model. The XGBoost model integrates multiple evaluators with weaker performance and shows good performance when solving regression problems. Unlike traditional machine learning algorithms such as artificial neural networks and support vector machines, XGBoost focuses on integrating multiple weak learners. The modeling process of XGBoost is to gradually improve the performance of the model through iteration, and its effectiveness in practice has been widely verified.

[0072] S2.2, combined model fusion, the process is as follows Figure 3 To improve the inversion capabilities of machine learning models and avoid the bias of a single model, we combined the CatBoost and XGBoost models to create a combined model, called the CatBoost-XGBoost combined model. This combined model combines the inversion results of the CatBoost model with those of the XGBoost model. This combined model not only inherits the high accuracy of the XGBoost model, but also incorporates the high generalization performance of the CatBoost model.

[0073] The outputs of the CatBoost model and the XGBoos model are combined in a weighted linear manner so that the output of the combined model is:

[0074] Z i =w1z 1i +w2z 2i

[0075] Among them, Z i is the water depth value inverted by the combined model; z 1i 、z 2i are the inverted water depth values ​​of the CatBoost and XGBoost models respectively; w1 and w2 are the weight coefficients of the CatBoost model and the XGBoost model in the combined model respectively, and they satisfy w1+w2=1;

[0076] The coefficient of determination R 2 Maximization and minimization of the root mean square error (RMSE) are the optimization objectives, and a weight optimization strategy is implemented. The weight optimization strategy includes:

[0077] Set the initial weights w1 = 1, w2 = 0, traverse w1∈[0,1] with a step size of 0.05, and calculate R under each weight combination. 2 , finally choose R 2 The largest combination.

[0078] The measured data of 50 randomly selected points and the corresponding band values ​​were combined as training data to train the CatBoost-XGBoost combination model, and the combination model formula with the best accuracy in each year was obtained, as shown in Table 3.

[0079] Table 3 The best combination model formula in each year

[0080]

[0081] On the GEE platform, the band values ​​of all pixels within the Hulun Lake are obtained as the input of the combined model, and the water depth values ​​of the geographical locations of the pixels are inverted and output, thereby obtaining the water depth of all points within the lake. Use Python to output the water depth map, such as Figure 6 As shown in (a)-(h) in the figure.

[0082] Figure 6 The darker areas represent the lake's water body. Darker colors indicate greater depth, while lighter colors indicate shallower depth. Water depth values ​​are generally concentrated within the 0-8m range. It can be seen that the colors in each image are generally lighter at the edges and darker at the center, consistent with the common sense that lake depth increases from the periphery to the center. However, in some years, such as 2019, the lake edges are not very light, or the light-colored areas at the edges are very small. Besides this fact, it is speculated that this may also occur because the training sample points are mainly located in the center of the lake, resulting in insufficient depth inversion capabilities at the lake edges.

[0083] The central area of ​​the lake shows an uneven grayscale distribution, indicating that the lakebed terrain is rugged and the water depth changes rapidly and not gently. The grayscale distribution is inconsistent across the images, indicating that the lakebed terrain is different and undergoes significant changes each year. A review of the data revealed that the lakebed of Hulun Lake is primarily composed of clay soil, which refers to soil with a plasticity index greater than 10 and is very plastic. Therefore, the lakebed of Hulun Lake is subject to scouring and erosion by the lake water, and the lakebed topography is highly susceptible to significant changes. Overall, the color of the water depth map gradually became lighter from 2016 to 2019, and then deepened significantly from 2020 to 2023, indicating that the lake depth has also undergone corresponding changes. This area of ​​change is highly consistent with the trend of lake area changes summarized above.

[0084] The average water depth in each year is obtained by statistics. Figure 7As shown in (a), it shows a trend of first decreasing and then increasing. The water depth continued to decrease from 2016 to 2019, and then continued to increase until 2022, and then decreased slightly in 2023. This trend is consistent with the color change trend presented by the water depth grayscale map. The minimum average water depth occurred in 2019, which was 4.957m; the maximum occurred in 2022, which was 6.381m; the average value was 5.54m. In order to avoid the long freezing period of Hulun Lake affecting the quality of water depth inversion, the eight years of images used were concentrated in the summer (June to August). This period is the flood season of Hulun Lake. After checking official news materials, it was found that the average water depth during the flood season from 2016 to 2023 was about 5.5m, and the inversion results were roughly consistent with the facts.

[37] .

[0085] The average water depth for each month in 2019 is as follows Figure 7 As shown in (b) of the figure, the water depth first increases and then decreases. From May to July 2019, the water depth increases continuously, and then decreases continuously until October. The minimum average water depth is 4.8148 meters in May 2019, and the maximum is 4.957 meters in July 2019, with an average of 4.559 meters. The changing trend of water depth in 2019 is similar to that of area.

[0086] S3. Water depth inversion and accuracy verification: Based on the combined model, pixel-level water depth is output and multiple indicators are used to verify the inversion accuracy, including:

[0087] R 2 The coefficient, root mean square error (RMSE), mean absolute error (MAE) and mean relative error (MRE) are used as accuracy indicators to evaluate the inversion accuracy. The calculation formula is as follows:

[0088]

[0089] Where n is the number of test points; ΔZ i is the difference between the actual water depth value and the inverted water depth value at the i-th test point, is the difference between the actual water depth value at the test point and its average value, is the average value of the actual water depth at the test point, R 2 The larger it is, the smaller the RMSE, MAE, and MRE are, indicating that the inversion accuracy is higher.

[0090] The inverted water depth values ​​at 20 verification points were calculated and compared with the measured water depth values ​​to calculate the accuracy index. The accuracy index calculation results for 2016-2023 are shown in the following table.

[0091] Table 4 Accuracy indicators

[0092]

[0093] As can be seen from the table, the accuracy coefficient R of the inverted water depth in each year is 2 The coefficients are all above 0.67, with an average of 0.76, indicating that there is a strong correlation between the inverted water depth and the measured water depth, and the combined model has a good inversion capability in the Hulun Lake waters.

[0094] S4. Water body range extraction and water volume calculation: Extract water body boundaries based on water body index, and calculate water volume through spatial integration combined with inverted water depth. Specifically, it includes:

[0095] S4.1. Extract water body boundaries based on the normalized difference water index (NDMI).

[0096] This embodiment uses the NDWI water index method to extract the lake area. NDWI, or Normalized Difference Witness Index, uses satellite remote sensing data to highlight water information in images, thereby achieving the purpose of identifying water bodies and assessing water body types. It is usually used to extract water body area or observe hydrological changes. The NDWI index calculates the green and near-infrared bands of the image. The principle is that the reflectivity of water bodies from the visible light band to the mid-infrared band will continue to decrease until it is almost zero. Therefore, using the green and near-infrared bands of visible light for normalized calculation can effectively highlight the water area in the image. In addition, the reflectivity of vegetation in the near-infrared band is relatively strong, and the calculation method of the Normalized Difference Witness Index can effectively eliminate the interference of vegetation information. The main land cover type around Hulun Lake is grassland, and the calculation principle of NDWI is consistent with this. It can maximize the highlighting of the lake part from the grassland background with high accuracy.

[0097] The calculation formula of NDWI is:

[0098]

[0099] Among them, B1 represents the pixel value of the green band, and B2 represents the pixel value of the near-infrared band;

[0100] The NDWI value ranges from -1 to 1. A value between -1 and 0 indicates that the pixel is not a water body, and a value between 0 and 1 indicates that the pixel is a water body. The higher the NDWI value, the higher the probability that the area is a water body and the more stable the water body type.

[0101] The image of the main area of ​​Hulun Lake was cropped with a rectangular frame, and the surrounding non-main lake water bodies were cropped. The NDWI value of the cropped image was calculated, and the area with NDWI>0 was determined to be water body. In this way, the scope of Hulun Lake on the image can be determined.

[0102] Taking into account factors such as coverage of the study area, image cloud coverage, and uniformity of imaging time, Sentinel-2A images of the Hulun Lake region on June 2, 2016, August 19, 2017, July 25, 2018, July 17, 2019, June 1, 2020, June 6, 2021, July 1, 2022, and June 26, 2023 were selected for interannual variation research, and one image of the Hulun Lake region per month from May to October 2019 was used for intra-annual variation research. Atmospheric correction was performed on the 13 images. For the convenience of subsequent water volume calculations, the images were resampled to 30m resolution and then cropped to the approximate range of the main Hulun Lake to prevent the inclusion of other water around Hulun Lake. The NDWI of the images was calculated, and the areas with NDWI values ​​greater than zero were selected as the range of Hulun Lake during that period. The results are as follows: Figure 4 As shown in (a)-(h) in the figure.

[0103] S4.2. Count the number of valid pixels within the water body. After extracting the Hulun Lake area using the NDWI threshold method, count the number of pixels within the lake area and multiply it by the pixel area to obtain the Hulun Lake area S. The formula is:

[0104] S=n×S cell

[0105] Among them, S cell is the area of ​​a single pixel, and n is the number of pixels. Since the resolution of Sentinel-2A images is resampled to 30m, S cell =900 square meters.

[0106] S4.3. Combine the inverted water depth grid data with the water body extent and calculate the water volume V by spatial integration.

[0107] Changes in lake water volume are driven by two factors: changes in lake area and lake depth. As lake area increases, water volume increases; as lake depth increases, water volume also increases. Based on the concept of differentials, a lake body can be abstracted as an irregular solid composed of multiple small rectangular blocks with a single pixel as the base and the pixel's water depth as the height. Therefore, integrating the two characteristics of area and water depth allows us to construct a relationship between lake water volume:

[0108] dV=dS dh

[0109] That is:

[0110] V=∫∫dS dh

[0111] Among them, V represents the volume of the lake, that is, the total water volume, and h is the water depth of the pixel point.

[0112] Consider Hulun Lake as a whole composed of small cuboids of different heights, where the base of the small cuboid is a 30m×30m square and the height is the water depth of the area. Therefore, the volume formula of Hulun Lake can be transformed into:

[0113]

[0114] Among them, S cell is the area of ​​a single pixel, h i is the pixel water depth value, and the water volume changes are dynamically calculated on an interannual and seasonal scale.

[0115] The water volume calculation formula can be used to calculate the water volume of Hulun Lake in each image. The calculation results are as follows: Figure 8 Analysis Figure 8 As shown in (a), with 2019 as the dividing point, the water volume of Hulun Lake shows a general trend of first decreasing and then increasing. From 2016 to 2019, the water volume of Hulun Lake continued to decline, by 1.1648 billion cubic meters. From 2020 to 2022, the water volume continued to rise, by 3.5437 billion cubic meters. There was a slight decrease in 2023, by 955.4 million cubic meters. The eight-year average water volume is 12.9516 billion cubic meters. The peak water volume occurred in 2022 at 15.3175 billion cubic meters, and the valley water volume occurred in 2019 at 11.4153 billion cubic meters, a difference of 3.9023 billion cubic meters. Between 2016 and 2023, the water volume of Hulun Lake generally increased, with a total increase of 1.7821 billion cubic meters in eight years, with an average annual increase of 222.8 million cubic meters.

[0116] like Figure 8 As shown in (b) of Figure 2, in 2019, with July as the dividing point, the water volume of Hulun Lake showed a general trend of first increasing and then decreasing. From May to July 2019, the water volume of Hulun Lake continued to rise, increasing by 320.7 million cubic meters; from July to October 2019, the water volume continued to decline, decreasing by 262.3 million cubic meters. The six-month average water volume was 11.21854 billion cubic meters. The water volume peaked at 11.4153 billion cubic meters in July 2019, and troughed at 11.0946 billion cubic meters in May 2019. Overall, the water volume of Hulun Lake increased from May to October 2019, totaling 58.4 million cubic meters, with an average monthly increase of 11.7 million cubic meters.

[0117] In summary, this example uses Sentinel-2A satellite remote sensing data and measured water depth data from the Hulun Lake area from 2016 to 2023 to extract the water body range of the lake and invert the water depth of Hulun Lake within this range. Finally, the water volume is calculated and the changing characteristics and trends of the water volume of Hulun Lake are analyzed. The following conclusions are drawn:

[0118] (1) The CatBoost-XGBoost combination model has strong inversion capabilities in the Hulun Lake waters. This example uses Python to establish a CatBoost-XGBoost combination model based on the relationship between measured water depth data and Sentinel-2A image data. The combined model is used to invert the water depth of the verification points and compare them with the measured points. It is found that the precision coefficient R2 of the inverted water depth in each year is above 0.67, with an average of 0.76, indicating that the model has strong inversion capabilities in the Hulun Lake waters.

[0119] (2) The water volume of Hulun Lake fluctuated significantly over the eight years from 2016 to 2023. Taking 2019 as the dividing point, the water volume of Hulun Lake generally showed a trend of first decreasing and then increasing, with an average water volume of 12.9516 billion cubic meters over the eight years. Among them, the water volume showed a downward trend from 2016 to 2019, an upward trend from 2020 to 2022, and a slight decrease in 2023. It is expected that the water volume will show a fluctuating upward trend in the next few years.

[0120] (3) Changes in the water boundary of Hulun Lake are mainly reflected in the northeast and southwest. From 2016 to 2019, the northeastern waters of Hulun Lake shrank slightly year by year, and a nearby tributary of the Hailar River gradually dried up. In addition, the lake boundary near Xinbarag Right Banner in the southwest also degraded slightly. Since 2020, the northeastern waters of Hulun Lake have expanded significantly, and a group of small lakes has gradually developed. At the same time, the lake boundary in the south of Hulun Lake has also expanded, and several small lakes have developed.

[0121] (4) The change trend of Hulun Lake's water volume is basically consistent with that of its lake area and average water depth. In the eight years from 2016 to 2023, the three elements of Hulun Lake's water volume, area, and average water depth all showed a trend of first decreasing and then increasing. Specifically, from 2016 to 2019, the water volume, area, and average water depth all showed a slow downward trend; from 2020 to 2022, all three elements showed an upward trend; and in 2023, all decreased slightly. All three reached their minimum value in 2019 and their maximum value in 2022, with an overall significant increase in eight years. In 2019, the change trend of Hulun Lake's water volume was also highly correlated with the change trend of its area and water depth.

[0122] (5) Hulun Lake has been effectively managed in recent years. Hulun Lake is a large lake in northeastern my country. However, due to global warming since around 2000, its area has been shrinking year by year, and its water volume has also decreased significantly, which has profoundly affected the surrounding ecological environment. Since 2013, Hulun Lake has carried out comprehensive ecological and environmental management. Within ten years, this measure has been very effective, so the water volume of Hulun Lake has shown an overall upward trend during the eight years of the experimental study.

[0123] The remaining technical features of the above embodiments can be flexibly selected by those skilled in the art to meet specific practical needs. However, it will be apparent to those skilled in the art that these specific details are not required to practice the present invention. In other instances, to avoid obscuring the present invention, well-known components, structures, or parts are not described in detail, and are therefore within the scope of protection of the technical solutions claimed in the claims.

[0124] Modifications and variations made by those skilled in the art without departing from the spirit and scope of the present invention are intended to be within the scope of the claims appended hereto. In the foregoing description, numerous specific details are set forth to provide a thorough understanding of the present invention. However, it will be apparent to those skilled in the art that these specific details are not necessarily required to practice the present invention. In other instances, well-known techniques, such as specific construction details, operating conditions, and other technical requirements, are not described in detail to avoid obscuring the present invention.

[0125] This document uses specific examples to illustrate the principles and implementation methods of the present invention. The above examples are only intended to help understand the method and core concept of the present invention. At the same time, those skilled in the art will find that the specific implementation methods and application scopes may vary based on the concept of the present invention. In summary, the contents of this specification should not be construed as limiting the present invention.

Claims

1. A lake water volume monitoring method based on satellite imagery and the CatBoost-XGBoost model is characterized by: Here are the steps: S1. Multi-source data acquisition and processing: Acquire Sentinel-2A satellite multispectral images and perform atmospheric correction, simultaneously collect measured water depth data and perform spatiotemporal matching to construct a training dataset; S2. CatBoost-XGBoost combination model construction and optimization: Construct a CatBoost-XGBoost combination model, fuse the outputs of the CatBoost model and the XGBoost model through weight distribution, and realize water depth inversion; S3. Water depth inversion and accuracy verification: Output pixel-level water depth based on the combined model, and use multiple indicators to verify the inversion accuracy; S4. Water body range extraction and water volume calculation: Extract the water body boundary based on the water body index, and calculate the water volume of the water body through spatial integration combined with the inverted water depth.

2. The lake water volume monitoring method based on satellite imagery and the CatBoost-XGBoost model according to claim 1, characterized in that: In S1, the multi-source data acquisition and processing specifically includes: S1.

1. Obtain multispectral images from the Sentinel-2A satellite covering multiple spectral bands of the target waters. Resample the images to a uniform spatial resolution based on the research scale, perform atmospheric correction using the Sen2cor plug-in, and generate surface reflectance data. S1.

2. Collect measured water depth data, including point data from different years and different sources. Use water level difference to adjust the measured water depth data according to the satellite imaging time and use it as the measured data at the satellite imaging time. Randomly divide 70%-80% of the data into a training set and 20%-30% of the data into a validation set.

3. The lake water volume monitoring method based on satellite imagery and the CatBoost-XGBoost model according to claim 1, characterized in that: In S2, the CatBoost-XGBoost combined model construction and optimization construction specifically include: S2.

1. The radiance values ​​of all bands of the multispectral image and the measured water depth data are used as training data, which are input into the CatBoost model and the XGBoos model for model training, and the inverted water depth is calculated and output; S2.

2. Combined model fusion: Combine the outputs of the CatBoost model and the XGBoos model in a weighted linear manner so that the output of the combined model is: WITH i =w1z 1i +w2z 2i Among them, Z i is the water depth value inverted by the combined model; z 1i 、z 2i are the inverted water depth values ​​of the CatBoost and XGBoost models respectively; w1 and w2 are the weight coefficients of the CatBoost model and the XGBoost model in the combined model respectively, and they satisfy w1+w2=1; The coefficient of determination R 2 Maximization is the optimization goal, and the weight optimization strategy is implemented.

4. The lake water volume monitoring method based on satellite imagery and the CatBoost-XGBoost model according to claim 3 is characterized in that: The weight optimization strategies in S2 include: Set the initial weights w1 = 1, w2 = 0, traverse w1∈[0,1] with a step size of 0.05, and calculate R under each weight combination. 2 , finally choose R 2 The largest combination.

5. The lake water volume monitoring method based on satellite imagery and the CatBoost-XGBoost model according to claim 3 is characterized in that: In S3, water depth inversion and accuracy verification specifically include: R 2 The coefficient, root mean square error (RMSE), mean absolute error (MAE) and mean relative error (MRE) are used as accuracy indicators to evaluate the inversion accuracy. The calculation formula is as follows: Where n is the number of test points; ΔZ i is the difference between the actual water depth value and the inverted water depth value at the i-th test point, is the difference between the actual water depth value at the test point and its average value, is the average value of the actual water depth at the test point, R 2 The larger it is, the smaller the RMSE, MAE, and MRE are, indicating that the inversion accuracy is higher.

6. The lake water volume monitoring method based on satellite imagery and the CatBoost-XGBoost model according to claim 3, characterized in that: S4, water range extraction and water volume calculation specifically include: S4.

1. Extract water body boundaries based on the normalized difference water index (NDMI). The formula is: Among them, B1 represents the pixel value of the green band, and B2 represents the pixel value of the near-infrared band; The image of the main area of ​​Hulun Lake was cropped with a rectangular frame, and the surrounding non-main lake water bodies were cropped. The NDWI value of the cropped image was calculated, and the area with NDWI>0 was determined to be water body, thereby determining the scope of the water body. S4.

2. Count the number of valid pixels within the water body and calculate the water body area S. The formula is: S=n×S cell Among them, S cell is the area of ​​a single pixel, and n is the number of pixels; S4.

3. Combine the inverted water depth grid data with the water body range and calculate the water volume V by spatial integration. The formula is: Among them, S cell is the area of ​​a single pixel, h i is the pixel water depth value, and the water volume changes are dynamically calculated on an interannual and seasonal scale.

Images