Plain farmland area reservoir water area time sequence change extraction and mutation detection method

Abstract

The invention discloses a time sequence change extraction and mutation detection method for a water area of a reservoir in a plain farmland area. Acquiring a spatial position and a historical maximum water area range of the target reservoir; preprocessing the Sentinel-1 radar image covering the maximum water area of the reservoir and the surrounding land area, and then carrying out water body crude extraction; removing farmland elements by using optical land utilization data matched with the radar image, and removing mountain shadows; abnormal values are removed through sliding window detection, intra-month mean value synthesis is carried out on non-abnormal values, and a monthly water body area time sequence of the reservoir is obtained; and performing mutation detection by using a PELT algorithm, and identifying a mutation time point of water body change. The method does not depend on field observation data, efficiently monitors the water body area change of the reservoir in the plain farmland area, analyzes the response characteristics of extreme events, adapts to a complex environment, timely identifies the water body area sudden change time point, contributes to water resource management and protection, and ensures the safety of regional drinking water resources.

Description

Technical Field

[0001] The invention relates to the fields of remote sensing science and technology and mathematical statistics, and in particular to a method for extracting time series changes in the water area of ​​a reservoir in a plain farmland area and detecting mutations. Background Art

[0002] As the core area of ​​agricultural production and densely populated areas, plain farmland areas undertake important tasks of food supply and socio-economic development. Due to the dual pressures of farmland irrigation and residents' drinking water demand, the stable supply of water resources is particularly critical in these areas. As an important infrastructure for water resource supply, reservoirs not only undertake the function of supplying water for residents' daily life, but also provide important water resources support for extensive farmland irrigation. The drastic fluctuations in reservoir water storage may lead to an increase in the risk of eutrophication of water bodies and deterioration of water quality, thereby threatening the stability of drinking water supply and posing a potential threat to the sustainability of farmland irrigation water. The water area of ​​a reservoir is closely related to the water storage capacity of the reservoir. Therefore, changes in the water area of ​​a reservoir not only directly affect the availability of regional water resources, but may also lead to challenges to the sustainable development of agricultural production and socio-economy (Zang et al., 2022). Dynamic monitoring of the water area of ​​reservoirs in plain farmland areas and timely detection of sudden changes are of great practical significance for maintaining the health of the reservoir water environment and the safe supply of agricultural irrigation water resources.

[0003] Traditional surface water monitoring methods mainly rely on hydrological observation stations and manual ground measurements, which have significant limitations in large-scale and continuous monitoring. These methods are not only time-consuming and labor-intensive, but also limited by technical and weather conditions, and it is difficult to obtain accurate and comprehensive data. In contrast, remote sensing technology provides an effective solution. Optical and radar satellite images have been widely used in water resources management, environmental assessment and global change research (Liu Jiyuan et al., 1997). Optical satellite images, with their high resolution and rich color advantages, are suitable for the identification of surface features and the accurate extraction of water area (Guo.etal., 2018). However, optical images are easily affected by weather changes and cloud cover, which limits their effective application in cloudy or severe weather conditions. In contrast, radar images, with their unique imaging technology, have the advantages of working all-weather and not being restricted by severe meteorological conditions, and can penetrate clouds to provide accurate ground information.

[0004] However, radar images also face significant challenges in their application. Since they do not contain color information, it is difficult for human eyes to effectively distinguish the types of objects. Especially in plain farmland areas with complex object types, the backscattering characteristics of objects such as water bodies, wetlands, and farmland are often similar, and the backscattering coefficients of different objects are close. This makes it difficult for a single radar image data source and traditional threshold segmentation algorithms to accurately distinguish these objects, especially water body identification, which is easily confused with surrounding farmland or wetlands. In addition, farmland areas have significant seasonal changes. The growth, maturity, harvest and irrigation cycles of crops will affect the backscattering characteristics of radar images, further increasing the complexity and challenge of water body identification. Common speckle noise and system noise also exacerbate this problem, not only reducing image quality, but also making the boundaries between water bodies and other types of objects more blurred, making it difficult to accurately extract them. Summary of the invention

[0005] The purpose of the present invention is to overcome the defects of the above-mentioned prior art and provide a method for extracting the time series change and detecting the mutation of the water area of ​​reservoirs in plain farmland areas based on multi-source data. The method comprehensively utilizes Sentinel-1 radar images, SRTM DEM data and Dynamic World data sets based on optical remote sensing images, combined with water Volume Extraction Algorithms and mathematical statistics models have achieved cost-effective and easy-to-operate water monitoring, effectively identified and monitored water changes in reservoirs in plain farmland areas, and timely issued warnings of significant water area change time points, ensuring continuous monitoring and safety of reservoirs in plain farmland areas.

[0006] In order to achieve the above technical objectives, the present invention adopts the following technical solutions: A method for extracting time series changes in the water area of ​​a reservoir in a plain farmland area and detecting mutations, comprising: Obtain the spatial location of the target reservoir and the historical maximum water area; Acquire Sentinel-1 radar images covering the largest water area of ​​the reservoir and the surrounding land area, perform noise reduction and slope correction on the Sentinel-1 radar images, and then extract water bodies; Obtaining optical land use data that matches the Sentinel-1 radar image, obtaining a farmland area based on the land use data, removing the farmland area from the water body extraction result as a mask, and removing the mountain shadow; Set the time series data of water area extracted from scene-by-scene images N The window length is one day, and the time series outliers are identified and removed through sliding window detection; Based on the constructed time series of water area of ​​each reservoir scene, the water area data at different time points are integrated, the data mean is synthesized within a month to generate monthly water area data. The PELT algorithm is used to detect mutations in the monthly time series of water area, and the time points of significant changes in water area are identified and analyzed.

[0007] As a preferred implementation, the Sentinel-1 radar image is a Sentinel-1 VV polarimetric radar image.

[0008] As a preferred implementation, the Sentinel-1 radar image is subjected to four-directional filtering by a Perona-Malik anisotropic diffusion filter to achieve noise reduction.

[0009] As a preferred implementation, slope correction is performed on the Sentinel-1 radar image based on a radiometric terrain correction algorithm.

[0010] As a preferred implementation, when extracting water bodies, the Canny edge detector is used to preliminarily detect the edge of the water body, and then a buffer zone is established. All water and land pixels in the buffer zone are used as inputs of the Otsu algorithm histogram statistics to obtain the segmentation threshold for water body extraction; The buffer zone distance is twice the area of ​​the study area divided by the perimeter; the study area is the area enclosed by the edge of the water body.

[0011] As a preferred implementation, the optical land use data uses the Dynamic World land use data set; The Sentinel-1 radar data and the Dynamic World dataset were time-matched, and the Dynamic World data closest to the Sentinel-1 radar data of each scene was selected. The farmland features were selected as masks and removed from the Sentinel-1 radar data of that scene. Since the backscatter coefficients of farmland and water bodies in SAR images are similar, it is difficult to accurately distinguish these two types of features by relying solely on the threshold segmentation algorithm. Combined with mask processing, the farmland interference of the reservoir in the original farmland area can be effectively eliminated, thereby ensuring the high accuracy of water body extraction.

[0012] As a preferred implementation, the slope information generated by SRTM DEM data is used to remove mountain shadows.

[0013] As a preferred implementation, the method further includes, after removing the mountain shadow, filling the holes by dilation-erosion using morphological filtering; the processing unit of dilation-erosion is determined according to the resolution of the radar image.

[0014] As a preferred embodiment, theN is 60.

[0015] As a preferred implementation, the identification and removal of time series outliers through sliding window detection includes: identifying and removing abnormal fluctuations by comparing the deviation of each observation point in the window with the average of the previous three observation points.

[0016] The present invention has the following beneficial effects: (1) In the complex terrain environment of plain farmland areas, accurate extraction of water body information requires the application of more sophisticated radar image processing algorithms. This invention aims to solve the problem of large errors in the automated extraction of reservoirs in plain farmland areas by comprehensively utilizing multi-source remote sensing image data, including Sentinel-1 radar images and optical land use data. By combining the all-weather and all-time observation capabilities of radar images with the rich spectral information of optical images, the advantages of both are complemented. By dynamically selecting optical data with a time series close to that of radar images for farmland feature mask removal, combined with fine preprocessing of radar images, the recognition difficulties caused by the similar backscatter coefficients between different terrain objects are effectively overcome, and the accuracy of water body detection and the robustness of the algorithm are significantly improved, especially in complex scenes, showing better adaptability.

[0017] (2) The algorithm of the present invention is easy to implement and is applicable to reservoirs in plain farmland areas of different shapes, sizes and geomorphic environments. This method can be extended to reservoir resource research on a large regional or even global scale, providing technical support for the efficient management and sustainable utilization of water resources, and has important significance and application value for improving reservoir operation efficiency and optimizing regional water security strategies. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] The accompanying drawings are not intended to be drawn to scale. In the accompanying drawings, each identical or nearly identical component shown in various figures may be represented by the same reference numeral. For the sake of clarity, not every component is labeled in each figure. Now, the various steps of the present invention will be described by studying a case and referring to the accompanying drawings, in which: Figure 1 It is the spatial distribution map of reservoirs selected in the embodiment of the present invention.

[0019] Figure 2 It is a flow chart of the method of the present invention.

[0020] Figure 3 This is the effect diagram of removing farmland using Dynamic World data in Example 1.

[0021] Figure 4 It is a time series and trend mutation analysis diagram of the reservoir water area in Example 1. DETAILED DESCRIPTION

[0022] The specific implementation of the present invention is further described in detail below in conjunction with the accompanying drawings and examples. The following implementation cases are used to illustrate the present invention, but are not used to limit the scope of the present invention.

[0023] In the embodiment, the data acquisition platform is as follows: CRD data: https: / / doi.org / 10.5281 / zenodo.6984619; Google Earth Engine (GEE) platform: https: / / earthengine.google.com.

[0024] Example 1 This example selects the Yuecheng Reservoir located in Hebei Province (such as Figure 1 ), the reservoir is located in the farmland area of ​​the North China Plain. Its geographical location has relatively complex spatial characteristics, widely covering plain farmland and diverse land use types. It not only plays an important role in local agricultural irrigation, but also undertakes key water supply and flood control functions. The relevant spatial information of the reservoir is directly matched through the National Refined Reservoir Spatial Dataset (CRD). The CRD dataset provides precise spatial location, vector boundaries and historical maximum water area, allowing us to accurately obtain detailed information on the Yuecheng Reservoir and conduct effective analysis and management. Figure 2 FIG. 1 is a flow chart of an embodiment, and this embodiment includes the following steps: Step 1: Obtain the required data from the Google Earth Engine (GEE) platform, including SRTM DEM data, Sentinel-1 radar image data from May to October each year from 2018 to 2023, and Dynamic World land use data to support the analysis and processing of the present invention.

[0025] Step 2: Data preprocessing: For the Sentinel-1 VV polarimetric image, the Perona-Malik anisotropic diffusion filter is used for denoising, and the slope correction based on the Radiometric Terrain Correction (RTC) algorithm is combined as a preprocessing step.

[0026] In this embodiment, convolution kernels in four directions are first defined to calculate the gradient changes of the image in different directions. The more common two-way (horizontal and vertical) expands the ability to capture gradient information, and these gradient information provides a key basis for the adaptive diffusion coefficient adjustment of the filter. Then, the parameters such as the number of iterations, the diffusion intensity coefficient and the diffusion rate are initialized. In each iteration, the image is convolved with these convolution kernels to calculate the gradients in each direction. Based on these local gradient information, the filter adaptively selects an appropriate diffusion strategy: for high-gradient areas, the method of using the gradient square and exponential function is selected to strictly control the diffusion to prevent edge blurring; while in low-gradient areas, the method of using the gradient square and the reciprocal function is used to emphasize the effect of smoothing noise. Through this dynamic adjustment based on local gradient features, the filter can achieve optimal denoising and detail retention in different areas. Finally, the processed image will be added to the original image as a new band to form the final output result. In this way, the Perona-Malik filter can effectively remove noise while accurately retaining important structures in the image, and is particularly suitable for processing satellite images under complex noise backgrounds.

[0027] Step 3: Preliminary extraction of water area. For the preprocessed Sentinel-1 VV polarimetric image, a dynamic buffer was first established at the vector boundary of the reservoir. The size of the buffer was set to twice the area of ​​the vector boundary divided by its perimeter to ensure that even if the water area is close to the maximum water area of ​​the study area, sufficient differences between water and land pixels can be maintained to enhance the effectiveness of subsequent processing. Next, the initial segmentation threshold was set to -14 dB, and the image was roughly divided into water and non-water areas. Subsequently, the boundary of the water body was extracted using the Canny edge detector, which accurately identified the boundary of the water body through gradient calculation, and further refined the edge using non-maximum suppression technology to ensure the continuity and accuracy of the boundary, making the water boundary clearer. Afterwards, based on the extracted water body pixels, a special buffer was constructed according to the water body information. The size of the buffer was set to twice the area of ​​the water body divided by its perimeter, and the Otsu algorithm was used to calculate the optimal threshold of the water body pixel separately. Although the Otsu algorithm can directly perform global segmentation on the pixels in the buffer zone, the combination with Canny edge detection further improves the segmentation accuracy in complex scenes, especially in environments with blurred boundaries or high noise, effectively reducing the segmentation error and ensuring the accuracy and reliability of water extraction results.

[0028] Step 4. Use the Dynamic World dataset to assist in removing farmland features. In order to solve the identification difficulties caused by the similar backscatter coefficients of water bodies and farmlands in Sentinel-1 radar images, this study performs time series matching on Sentinel-1 radar data and the Dynamic World dataset. First, a difference analysis is performed on the timestamps of the Sentinel-1 radar image and the Dynamic World dataset, and the Dynamic World data with the closest time is selected as the matching object. Next, using the land use information marked in the Dynamic World dataset, these farmland areas are removed from the radar image as masks to improve the accuracy of water body extraction and ensure more reliable water body distinction and extraction in radar images. The results are shown below. Figure 3 shown.

[0029] Step 5: Post-processing of water area extraction. This study optimized the removal of slope shadows by using the SRTM digital elevation model to generate a slope dataset. For 30-meter resolution SRTM data, in order to avoid errors introduced by resolution mismatch and reduce the interference of mountain shadows, areas with slopes greater than 20° are judged as slope shadows and removed. At the same time, morphological filtering technology is used to solve the problem of internal holes in water bodies. The holes are filled by 10-meter expansion, and then 10-meter erosion is performed to eliminate redundant pixels to ensure the smoothness and continuity of the boundaries. Compared with the traditional single expansion, this study uses multiple iterative expansions to gradually expand the boundaries, avoiding excessive blurring and loss of details. In addition, the use of connectivity detection combined with mask updating to intelligently remove isolated noise pixels is superior to the traditional erosion method that only relies on fixed morphological kernels. While ensuring the integrity of the main structure, it significantly improves the accuracy and reliability of water body extraction.

[0030] Step 6: Time series outlier detection. In this study, we use the sliding window detector provided by the ADTK algorithm to process time series outliers. The specific operation is to set a window length of 60 days on the water area time series data of the scene-by-scene image extraction results and perform sliding window detection. This method effectively identifies persistent abnormal fluctuations by comparing the deviation of each observation point from the average of the three previous observation points. The identified outliers will be removed to ensure the accuracy and reliability of data analysis.

[0031] Step 7, accuracy evaluation of water body extraction results. In order to verify the accuracy of the above method in extracting water bodies in Sentinel-1 images, this embodiment randomly selected images at five time points and generated 300 independent sample points in each image (a total of 1500 samples). The water body range interpreted by manual visual interpretation was used as ground truth data for comparison, a confusion matrix was constructed, and a detailed classification accuracy evaluation was performed in combination with producer accuracy (PA), user accuracy (UA), overall classification accuracy (OA), Kappa coefficient and F1 score. The evaluation results show that the overall classification accuracy (OA) exceeds 0.95, the Kappa coefficient exceeds 0.90, and the F1 score and other indicators also indicate the efficiency of the algorithm. These results demonstrate the effectiveness of the algorithm in accurately extracting water body area, although there are some deviations at small water bodies and land-water body boundaries. The high-precision evaluation indicators show that the method has high feasibility and accuracy.

[0032] Table 1 Accuracy verification results

[0033] Step 8: Monthly time series construction. After identifying and removing outliers, this embodiment averages the water area results of each scene image by month, thereby constructing a time series of monthly average water area, such as Figure 4 .

[0034] Step 9. In this study, the PELT mutation test method was used to detect mutations in the water area time series of Yuecheng Reservoir. Figure 4 . Through the set RBF kernel function, the nonlinear relationship of water area changing over time is captured, which has strong local sensitivity and adaptability, thereby improving the accuracy and efficiency of detection. The research results show that the PELT method can accurately identify the mutation points of the water area of ​​Yuecheng Reservoir, demonstrating the superiority of this method in processing complex water area time series. This method provides strong technical support for water resources management and emergency warning of reservoirs.

[0035] Through the above method, the water area time series change extraction and mutation detection of reservoirs in plain farmland areas are realized. This method is not limited to the size of the study area and reservoir range, but can also extract the water area time series change and mutation detection on a long time series and multiple time scales. This provides key data support for the scientific assessment and management of water resources, further ensures the safety and reliability of drinking water, and helps to improve the level of water resources management and respond to the challenges brought by climate change.

Claims

1. A method for extracting time series changes in the water area of ​​a reservoir in a plain farmland area and detecting mutations, characterized in that: include: Obtain the spatial location of the target reservoir and the historical maximum water area; Acquire Sentinel-1 radar images covering the largest water area of ​​the reservoir and the surrounding land area, perform noise reduction and slope correction on the Sentinel-1 radar images, and then extract water bodies; Obtaining optical land use data that matches the Sentinel-1 radar image, obtaining a farmland area based on the land use data, removing the farmland area from the water body extraction result as a mask, and removing the mountain shadow; Set the time series data of water area extracted from scene-by-scene images N The window length is one day, and the time series outliers are identified and removed through sliding window detection; Based on the constructed time series of water area of ​​each reservoir scene, the water area data at different time points are integrated, the data mean is synthesized within a month to generate monthly water area data. The PELT algorithm is used to detect mutations in the monthly time series of water area, and the time points of significant changes in water area are identified and analyzed.

2. The method according to claim 1, characterized in that The Sentinel-1 radar image is a Sentinel-1VV polarimetric radar image.

3. The method according to claim 1, characterized in that The Sentinel-1 radar image is filtered in four directions using the Perona-Malik anisotropic diffusion filter to achieve noise reduction.

4. The method according to claim 1, characterized in that Slope correction of Sentinel-1 radar images is performed based on the radiometric terrain correction algorithm.

5. The method according to claim 1, characterized in that When extracting water bodies, the Canny edge detector is used to preliminarily detect the edge of the water body, and then a buffer zone is established. All water and land pixels in the buffer zone are used as inputs for the histogram statistics of the Otsu algorithm to obtain the segmentation threshold for water body extraction. The buffer zone distance is twice the area of ​​the study area divided by the perimeter; the study area is the area enclosed by the edge of the water body.

6. The method according to claim 1, characterized in that The optical land use data uses the DynamicWorld land use data set; The Sentinel-1 radar data and the Dynamic World dataset were time-series matched, and the Dynamic World data closest in time to the Sentinel-1 radar data of each scene was selected. The farmland features were selected as masks and removed from the Sentinel-1 radar data of that scene.

7. The method according to claim 1, characterized in that The slope information generated by SRTM DEM data is used to remove mountain shadows.

8. The method according to claim 1, characterized in that: It also includes filling holes through morphological filtering dilation-erosion after removing the mountain shadow; the processing unit of dilation-erosion is determined according to the resolution of the radar image.

9. The method according to claim 1, characterized in that: Said N is 60.

10. The method according to claim 1, characterized in that The method of detecting, identifying and removing abnormal values ​​of the time series by sliding window includes: identifying and removing abnormal fluctuations by comparing the deviation of each observation point in the window with the average value of the previous three observation points.

Images