Method for estimating average regulation capacity of a floodplain lake
By constructing the maximum inundation boundary of floodplains and combining SWOT and Sentinel-1 data, and using median filtering and quantile calculation, the problem of stable quantification of floodplain storage capacity was solved, and robust assessment was achieved under conditions without ground data.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- NANJING INST OF GEOGRAPHY & LIMNOLOGY
- Filing Date
- 2026-04-17
- Publication Date
- 2026-07-14
AI Technical Summary
In the absence of surface water level and reservoir capacity data, existing technologies struggle to achieve stable quantification of the average storage capacity of floodplains at long-term, decadal scales and across lakes. Especially under conditions of multi-source error superposition and extreme events, existing methods are susceptible to outliers, leading to distorted assessments.
By constructing the maximum inundation boundary of floodplains, combining SWOT and Sentinel-1 data, filtering water level and area data using the median, calculating the multi-year average storage capacity using quantile intervals, and eliminating outliers, a robust quantification of floodplains is achieved.
It improves the accuracy and integrity of floodplain boundaries, can automatically acquire continuous storage capacity indicators at the decadal scale without ground data, reduces the impact of outliers, and achieves robust assessments for multi-year comparisons.
Smart Images
Figure CN122390330A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of lake hydrological remote sensing monitoring technology, and in particular relates to a method for estimating the average storage capacity of floodplain lakes based on SWOT + Sentinel-1. Background Technology
[0002] As off-river regulation and storage units within river-lake systems, floodplains, when connected to river channels, determine the flood storage and discharge process and the capacity to reduce flood peaks through their water level and inundated area. They are crucial indicators for characterizing regional flood and drought response and water resource regulation potential. However, globally, floodplains are mostly distributed in remote areas or regions with sparse monitoring networks. Measured water levels, reservoir capacity curves, and long-term continuous hydrological records are generally lacking or difficult to obtain, significantly limiting the quantification of floodplain storage and regulation capacity based on surface data. Therefore, developing scalable long-term monitoring technologies based on remote sensing has become an important approach.
[0003] In existing technologies, remote sensing estimation of lake water volume changes typically employs a "water level combined with area" framework: time series of water area are obtained from optical or synthetic aperture radar imagery, water level series are obtained from satellite altimetry or 2D water surface elevation products, and volume changes are calculated using geometric volume approximations or empirical relationships. While this approach can output volume change results even in the absence of ground data, it still faces bottlenecks in characterizing the long-term, decadal-scale, and cross-lake comparative regulation capacity of floodplains. Specifically, existing schemes often use a monthly time step, making it difficult to balance the rapid fluctuations within a decadal period of flooding with the sparsity of satellite observations, resulting in insufficient effective paired samples and increased sensitivity during anomalous study periods.
[0004] On the one hand, if the inundation area sequence relies on optical imagery, the critical flooding period is often obscured by clouds and rain, resulting in a lack of effective observations. Even with SAR imagery, factors such as speckle noise, vegetation cover, wetland mixed pixels, and irregular flood boundary heights can lead to systematic biases in flood zone boundary identification and area statistics. Furthermore, existing methods for determining the maximum boundary often directly use the maximum water surface area in a single period, the maximum historical water body area, or a simple buffer boundary as the statistical boundary of flooded lakes. However, when these methods are used for flooded lakes, two types of problems easily arise: First, due to the quality of historical images, missed boundary pixels, or classification errors, the directly extracted maximum historical water body area may be too small at the lake shoreline, making it difficult to fully characterize the potential inundation area of the flooded lake. Second, flooded lakes are often connected to adjacent rivers during high water periods. If the maximum water body area is directly expanded, some river water may be mistakenly included within the lake's regulation and storage unit, leading to increased deviations in subsequent area statistics and water volume change calculations. On the other hand, water level series rely on altimetry or two-dimensional water surface elevation observations. These are often affected by track coverage, revisit cycles, differences in quality indicators, and anomalous observations, resulting in missing data, inconsistent timescales, and outliers. This makes it difficult to obtain stable "water level-area" paired samples on a decadal scale over a long period. Especially under conditions of multi-source error superposition and extreme event disturbances, existing schemes tend to focus on outputting the "volume change series" itself, lacking a set of rules for automatically executing, insensitive to outliers, and applicable to multi-year comparisons across lakes for extracting average regulation capacity indicators. The traditional method of representing amplitude using extreme value differences is easily affected by extreme outliers, leading to distorted capacity assessments.
[0005] In recent years, Sentinel-1's all-weather observation capabilities have supported the construction of continuous inundation areas, while SWOT's two-dimensional water surface elevation observations have provided a new data source for water level information. However, on long-term scales, relying solely on SWOT is still insufficient to obtain sufficient decadal-scale water level inputs to cover many years. Furthermore, multi-source errors and anomalous data can affect the stability of capacity indicators. Therefore, in the absence of surface reservoir capacity relationships and long-term hydrological records, how to automatically obtain decadal-scale continuous average floodplain storage capacity indicators that can be compared across multiple lakes over many years and are insensitive to outliers is a pressing problem that needs to be solved. Summary of the Invention
[0006] The purpose of this invention is to address the characteristics of floodplains, such as the connectivity between floodplains and rivers, the dynamic changes in boundaries due to hydrological conditions, and the long-term scarcity of surface water level and reservoir capacity data. It also addresses the instability in the characterization of the multi-year average water storage capacity caused by missing water level measurements, inconsistent time scales, abnormal research periods, and the superposition of multiple source errors. The invention proposes a method for estimating the average water storage capacity of floodplains, enabling the transformation from raw observations to parameters that can be directly used for assessing the flood storage and peak reduction potential of floodplains, prioritizing water storage units, and parameterizing flood control scheduling.
[0007] To achieve the above objectives, the present invention adopts the following technical solution:
[0008] A method for estimating the average storage capacity of floodplains, the method comprising:
[0009] The initial maximum water volume range of the target floodplain is extracted based on the historical surface water frequency. A buffer zone is constructed on the boundary of the initial maximum water volume range. Excess water is removed from the buffer zone based on the connectivity between the water body and the adjacent river channel to construct the maximum inundation boundary of the floodplain.
[0010] Within the constraint of the maximum inundation boundary, SWOT water level data and water area data extracted from remote sensing images of the floodplain are extracted; the elevation values of eight basic azimuth points on the maximum inundation boundary of the target floodplain are extracted, and the median of the elevation values is used as the reference elevation value to filter the water level data.
[0011] The median of water level and area observations was selected as the representative water level and representative area of the lake. The time series of the representative water level and representative area were obtained and used to calculate the time series of water volume changes of flooded lakes relative to the baseline state.
[0012] The water volume change time series is sorted by numerical value, and the upper and lower quantile values of the water volume change time series are calculated respectively. The difference between the upper and lower quantile values is used to characterize the estimated result of the average regulation capacity of the floodplain lake.
[0013] In some embodiments of the present invention, when there is a lack of SWOT representative water levels during the study period, a water level-area correlation model is constructed during the overlapping observation period of SWOT water levels and the effective flooded area of remote sensing images, and the missing water level data is supplemented based on the correlation model.
[0014] In some embodiments of the present invention, the method of filtering water level data is as follows:
[0015] Determine the latitude and longitude of the eight basic azimuth extreme points of the maximum inundation boundary of the target floodplain lake;
[0016] Extract the elevation values corresponding to the latitude and longitude locations from the SRTM DEM, take the median of the eight elevation values as the elevation reference value, and set the elevation threshold range based on the reference elevation value.
[0017] The water level pixel values for each observation date are compared with the elevation threshold range, and water level data that exceed the elevation threshold range are removed.
[0018] In some embodiments of the present invention, the reference state is: the earliest time when both representative water level and effective flooded area are simultaneously present as the reference; or the dry season as the reference; or the time representing low water level during the study period as the reference.
[0019] In some embodiments of the present invention, the water change value corresponding to the 90th percentile and the water change value corresponding to the 10th percentile of the water change time series are calculated respectively as the upper and lower percentile values.
[0020] In some embodiments of the present invention, the method further includes setting a lower limit for effective data used for quantile distance calculation; when the number of effective data is insufficient, the average storage capacity is not output or a reliability insufficiency prompt is output.
[0021] In some embodiments of the present invention, the method of approximating the volume of a frustum is used to calculate the change in lake water volume.
[0022] In some embodiments of the present invention, the remote sensing image is selected from radar remote sensing image.
[0023] In some embodiments of the present invention, a water body index is constructed based on radar remote sensing image VH and VV polarization data to extract water bodies within the maximum submergible boundary constraint range and obtain area data.
[0024] In some embodiments of the present invention, during the study period, the SWOT water level data and water area data extracted from remote sensing images are analyzed on a ten-day time scale to obtain an estimate of the average regulation capacity of floodplains at a ten-day time scale.
[0025] The present invention has the following beneficial effects:
[0026] (1) In view of the dynamic changes in the boundary of flooded lakes and their easy connection with adjacent rivers, this invention constructs the maximum inundation boundary by constraining the frequency of historical surface water occurrence, compensating for boundary expansion, and verifying and revising the river adhesion area. This can improve the integrity and accuracy of the boundary of the storage unit and provide reliable spatial constraints for subsequent effective inundation area statistics and water volume change calculation.
[0027] (2) Relying only on publicly available SWOT water surface elevation and Sentinel-1 data, without the need for surface water level network and reservoir capacity curve, the time series of water level and area at the ten-day scale and the time series of water volume change can be constructed under the maximum submergence boundary constraint.
[0028] (3) The multi-year average storage capacity parameter is extracted by using quantile interval, which can eliminate the interference of abnormal observation time period and extreme flood and drought years in a statistical sense, making the results insensitive to abnormal data and realizing the robust quantification of the multi-year average storage capacity of flood-prone lakes. The results can be directly used for flood storage and peak reduction potential assessment, priority ranking of storage units and parameterization of flood control scheduling in flood-prone areas. Attached Figure Description
[0029] Figure 1 This is a flowchart of the method of the present invention.
[0030] Figure 2 This is a time series plot of the Sentinel-1 area of the target flooded lake in the embodiment.
[0031] Figure 3 This is a schematic diagram of the SWOT observation water level of the target flooded lake and the water level time series after constraint completion in the embodiment.
[0032] Figure 4 This is a schematic diagram of the time series of water volume changes relative to the baseline state of the target flooded lake in the embodiment.
[0033] Figure 5 This is a schematic diagram illustrating the calculation results of the multi-year average storage capacity of the target floodplain lake in the embodiment. Detailed Implementation
[0034] The specific embodiments of the present invention will be described in further detail below with reference to the accompanying drawings and examples. The following examples are for illustrative purposes only and are not intended to limit the scope of the invention.
[0035] The embodiments of this application take three floodplains in the Amazon basin as research objects. The selected floodplains are typical lakes connected to rivers, with significant seasonal variations in water level and water area, making them suitable as verification objects for the method of this invention. This method is also applicable to other areas with similar wide-swath radar altimetry data and SAR imagery data.
[0036] like Figure 1 The diagram shown is a flowchart of the method in this embodiment, which includes the following steps:
[0037] Step 1: Construct a decadal-scale floodplain water area sequence based on Sentinel-1 image data covering the study area, as detailed below:
[0038] Sentinel-1 SAR image data covering the target floodplain lake study area was acquired, and the images were preprocessed and denoised to reduce the impact of speckle noise and other interference on water boundary interpretation. For each preprocessed Sentinel-1 image, its VV and VH polarization backscattering information was extracted, and the SDWI water index was constructed based on this, with the expression as follows:
[0039]
[0040] Based on the SDWI index image, the Otsu adaptive threshold segmentation method is used to extract the water body range corresponding to each image.
[0041] The steps for constructing the maximum inundation limit of the target floodplain lake based on the extracted water body are as follows:
[0042] The overall water frequency information of the target floodplain lake from 2000 to 2020 was obtained from the JRC Global Surface Water (GSW) surface water frequency data, and the historical maximum water volume mask of the lake was extracted as the initial candidate boundary range. Considering that the poor quality of historical images may lead to missed boundary pixels, a 10 m buffer zone was established for the initial candidate boundary range to compensate for missing edge areas. To address the issue that the floodplain lake may overlap and adhere to adjacent rivers after buffer expansion, the area connected to the river was manually visually checked and revised, and excess buffer water was removed along the river edge, finally obtaining the maximum inundation boundary of the target floodplain lake's regulation and storage unit. The maximum inundation boundary serves as the spatial constraint range for area statistics.
[0043] The above method for constructing the maximum inundation boundary of the target floodplain lake storage unit can solve the problem of poor accuracy in extracting the maximum inundation boundary of floodplain lakes. On the one hand, by extracting the initial maximum water body range based on the frequency of historical surface water occurrence and expanding and compensating for the boundary, the problem of missing lake shore edges caused by poor historical image quality, missed boundary pixels, or classification errors can be reduced, thus improving the integrity of the boundary. On the other hand, by verifying and revising the area connected to adjacent rivers, it is possible to avoid mistakenly including river water bodies that do not belong to the main body of the target floodplain lake into the storage unit after expansion, thereby improving the accuracy of the maximum inundation boundary of the floodplain lake and reducing systematic biases in subsequent effective inundation area statistics, water level matching, and water volume change calculations.
[0044] The water body extent extraction results are filtered, and the water body area is extracted within the maximum inundation boundary to obtain the time series of floodplain lake water body area at the decadal scale (e.g., Figure 2 (As shown). When multiple images exist within the same ten-day period, the median area of that ten-day period is used in subsequent calculations.
[0045] Step 2: Extract the decadal-scale water level sequence from SWOT satellite data and perform constraint completion, specifically including:
[0046] Read the SWOT water level data and its corresponding quality information, perform quality screening on the water level observations, and retain only the water level pixels that meet the quality requirements;
[0047] The water level data is cropped using the maximum inundation boundary of the floodplain lake storage unit to obtain a set of water level pixels within the lake area;
[0048] This embodiment further combines SRTM DEM data to perform physical rationality filtering on obviously unreasonable water level observations. Specifically, the following steps are taken: First, determine the latitude and longitude of the extreme points in the eight directions of the maximum inundation boundary of the target flooded lake (i.e., the intersection of the azimuth lines of the eight reference directions and the maximum inundation boundary); second, extract the elevation values corresponding to the above eight locations from the SRTM DEM, and take the median elevation as the prior elevation reference value of the flooded lake; then, compare the water level pixel values of each observation date with the prior elevation reference value. When the difference between the water level pixel value and the prior elevation reference value is within a preset threshold range, it is determined as an abnormal pixel and automatically removed. In this embodiment, the preset threshold range is set to ±100 m.
[0049] After completing the above range filtering, outlier pixels remaining within the same observation date are removed using the wse qual layer, and the median of the remaining pixels is used as the representative water level for that observation date. Furthermore, the representative water levels for each valid observation date within the same ten-day period are summarized, and the median is taken as the ten-day scale representative water level, thus forming a ten-day scale water level sequence (e.g., ...). Figure 3 (As shown). Due to the significant dynamic changes in the boundaries of floodplains and lakes, and their tendency to spatially mix with river channels, wetlands, or shoals, the water level and area observations within the same ten-day period are prone to skewed distributions or a small number of outliers. Using the median as a representative value can reduce the impact of outliers on representative water levels and areas, and improve the robustness of the ten-day time series construction.
[0050] Given that the available water level observation coverage period of SWOT in this embodiment is shorter than the study period, an area and water level constraint function is established in the period when SWOT and Sentinel-1 observations overlap. This constraint relationship is then used to fill in the missing water level measurements during the study period, thereby obtaining a complete decadal-scale water level sequence covering the study period for subsequent water volume change calculations.
[0051] Step 3: Determine the baseline state and calculate the time series of water volume changes on a ten-day scale. Select the initial time of the study period as the baseline state, and read the water level and water area of the baseline state, denoted as H1 and A1 respectively; calculate the water level H1 for each ten-day period. t With water body area A t Perform ten-day scale pairing and unify units, and calculate the water volume change value relative to the baseline data for each ten-day period according to the following formula:
[0052]
[0053] Thus, a series of decadal water volume changes during the study period was obtained (e.g., Figure 4 (As shown).
[0054] Step 4: Summarize the water volume changes for all ten-day periods within the study period into a sample set and sort them by value. Determine the water volume change value P10 corresponding to the 10th percentile and the water volume change value P90 corresponding to the 90th percentile. Use the difference between the two as the multi-year average water storage capacity index. Output:
[0055]
[0056] Where R is the average storage capacity parameter (e.g., Figure 5 As shown in the figure, this is used to characterize the multi-year average available flood storage and regulation capacity of floodplains and to reduce the impact of abnormal study time and extreme values on the results. This embodiment further compares the quantile feature used in this invention with the traditional range feature. For the three target floodplains, the traditional range feature is 0.039 km. 3 0.19 km 3 and 1.51 km 3 The corresponding quantile features are 0.018 km³, 0.15 km³, and 1.38 km³, respectively. The results show that the traditional range feature is more susceptible to outliers, while the quantile feature is more robust.
[0057] Step 5: Synchronously output the number of valid samples, quality level, or reliability information for result interpretation.
Claims
1. A method for estimating the average storage capacity of floodplains, characterized in that, The method includes: The initial maximum water volume range of the target floodplain is extracted based on the historical surface water frequency. A buffer zone is constructed on the boundary of the initial maximum water volume range. Excess water is removed from the buffer zone based on the connectivity between the water body and the adjacent river channel to construct the maximum inundation boundary of the floodplain. Within the constraint of the maximum inundation boundary, SWOT water level data and water area data extracted from remote sensing images of the floodplain are extracted; the elevation values of eight basic azimuth points on the maximum inundation boundary of the target floodplain are extracted, and the median of the elevation values is used as the reference elevation value to filter the water level data. The median of water level and area observations was selected as the representative water level and representative area of the lake. The time series of the representative water level and representative area were obtained and used to calculate the time series of water volume changes of flooded lakes relative to the baseline state. The water volume change time series is sorted by numerical value, and the upper and lower quantile values of the water volume change time series are calculated respectively. The difference between the upper and lower quantile values is used to characterize the estimated result of the average regulation capacity of the floodplain lake.
2. The estimation method according to claim 1, characterized in that, When there is a lack of SWOT representative water levels during the study period, a water level-area correlation model is constructed during the overlapping observation period of SWOT water levels and the effective flooded area of remote sensing images, and the missing water level data is supplemented based on the correlation model.
3. The estimation method according to claim 1 or 2, characterized in that, The method for filtering water level data is as follows: Determine the latitude and longitude of the eight basic azimuth extreme points of the maximum inundation boundary of the target floodplain lake; Extract the elevation values corresponding to the latitude and longitude locations from the SRTM DEM, take the median of the eight elevation values as the elevation reference value, and set the elevation threshold range based on the reference elevation value. The water level pixel values for each observation date are compared with the elevation threshold range, and water level data that exceed the elevation threshold range are removed.
4. The estimation method according to claim 1, characterized in that, The benchmark state is: the earliest time when both representative water level and effective flooded area are simultaneously available; or the dry season; or the time representing low water level during the study period.
5. The estimation method according to claim 1, characterized in that, The water change values corresponding to the 90th percentile and the 10th percentile of the water change time series are calculated respectively, and used as the upper and lower quantile values.
6. The estimation method according to claim 1, characterized in that, The method further includes setting a lower limit for effective data used for quantile distance calculation; when the number of effective data is insufficient, the average storage capacity is not output or a reliability insufficiency prompt is output.
7. The estimation method according to claim 1, characterized in that, The method of approximating the volume of a frustum is used to calculate the changes in lake water volume.
8. The estimation method according to claim 1, characterized in that, The remote sensing images used are radar remote sensing images.
9. The estimation method according to claim 8, characterized in that, Water body indices were constructed based on VH and VV polarization data from radar remote sensing imagery. Water bodies within the maximum submergence boundary constraint range were extracted, and area data were obtained.
10. The estimation method according to claim 1, characterized in that, During the study period, the SWOT water level data and water area data extracted from remote sensing images were analyzed on a ten-day time scale to obtain the estimated results of the average regulation capacity of floodplains at the ten-day time scale.