Method and system for calculating dynamic surface water reserves in combination with SWOT satellite

Abstract

The invention relates to the technical field of water resource investigation, in particular to a method and system for calculating dynamic surface water reserves in combination with an SWOT satellite.The method comprises the steps that water surface height data are collected, water surface area calculation is conducted, dynamic weight distribution sensitive to environmental factors and multi-source fusion are conducted, and the water surface area of a target surface water body is obtained; performing water level measurement; fusing and correcting all the water level measurement data obtained by calculation to obtain a corrected water level elevation; acquiring underwater topographic data, and performing feature-level fusion and error minimization optimization on the underwater topographic data by adopting a multi-modal deep learning framework to obtain the underwater topographic data; and calculating the water reserves of the target surface water body. The method solves the problems of single observation data source and susceptibility to environmental interference in the prior art, and has the characteristics of high robustness and all-weather adaptability.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of water resource investigation, in particular to a method and system for calculating dynamic surface water storage combined with a SWOT satellite. BACKGROUND

[0002] Surface water is a strategic resource indispensable to support social and economic development and maintain the balance of the ecological system. Accurately grasping the change of water storage of lakes, reservoirs and other surface water bodies is of great significance for water resource optimization, flood and drought disaster warning, water environment management and global climate change research.

[0003] Traditional methods for calculating dynamic surface water storage mainly rely on field measurement of ground hydrological stations and remote sensing technology of a single data source. Although ground stations such as water level gauges and flow cross sections can provide high-precision time series data, their spatial representativeness is limited and it is difficult to reflect the overall and spatial heterogeneity of the water storage change of large water bodies such as Poyang Lake and Dongting Lake.

[0004] With the development of remote sensing technology, satellite data is used for water storage estimation. This method usually includes three key links: water area extraction, water level measurement and underwater topography (depth) data acquisition. However, there are many bottlenecks in these links in the prior art: In the aspect of water area extraction: Although the commonly used optical remote sensing method (such as normalized difference water index NDWI) is mature, it is easily disturbed by weather conditions such as cloud, fog and rain, resulting in data loss or precision reduction. At the same time, in complex terrain areas, mountain shadows and vegetation shading often lead to misclassification or omission of water bodies. Although synthetic aperture radar (SAR) technology has all-weather observation capability, its identification accuracy of water shore boundaries is easily affected by surface roughness (such as wind waves and wet soil), and the data processing is relatively complex. Therefore, it is difficult to obtain stable and reliable water area under all conditions by relying on a single remote sensing data source.

[0005] In the aspect of water level measurement: Although traditional satellite altimeters (such as Jason series) can measure water level, their spatial resolution is low and they are mainly suitable for oceans and large lakes. The new generation of SWOT (Surface Water and Ocean Topography) satellite greatly improves the observation capability, but its measurement signal is still affected by atmospheric delay (especially tropospheric wet delay), resulting in systematic errors. Although the ground-based GNSS reflection measurement technology (GNSS-R / IR) can monitor the water level at a fixed point and high frequency, its accuracy is easily disturbed by complex environmental factors such as water waves and multipath effects. How to integrate the observation advantages of different platforms and eliminate their respective systematic errors is a difficulty in achieving centimeter-level accurate measurement.

[0006] In terms of underwater topographic data acquisition: This is the most challenging part of water storage estimation. Traditional shipborne sonar sounding (single beam / multi-beam) has high precision, but is costly, time-consuming and labor-intensive, and cannot fully cover vast water areas. The water depth remote sensing inversion method based on optical images is low-cost and widely covers, but its accuracy will decrease sharply in areas with deep or low transparency. The results of the interpolation method based on land surface topography have great uncertainty in the lake center far from the shore.

[0007] In terms of water storage calculation and prediction: The existing calculation model is mostly a simple grid accumulation method of "volume = area x water depth". This model regards the water body as a static rigid body, ignoring important physical processes such as vertical seepage, thermal expansion and contraction of water, and salinity changes, resulting in "pseudo storage changes" in the calculation results, which cannot truly reflect the hydrological physical processes. In addition, the existing methods focus on the estimation of current storage, and lack the dynamic prediction ability of future trends.

[0008] In summary, the existing surface water storage investigation technology has the problems of single observation data source and susceptibility to environmental interference, and an integrated calculation method that can systematically fuse multi-source data, overcome their limitations, and consider fine physical processes is urgently needed. SUMMARY

[0009] The present application provides a dynamic surface water storage calculation method and system combined with SWOT satellite, which has the characteristics of robustness and strong all-weather adaptability, to solve the problems of single observation data source and susceptibility to environmental interference in the prior art.

[0010] To achieve the above-mentioned purposes of the present application, the technical solutions adopted are as follows: A dynamic surface water storage calculation method combined with SWOT satellite, the method comprising the following steps: Collecting water surface height data and calculating the water surface area respectively; performing dynamic weight distribution and multi-source fusion on all the calculated water surface area results according to environmental factors, to obtain the water surface area of the target surface water body; Performing water level measurement; fusing and correcting all the calculated water level measurement data to obtain the corrected water surface elevation; Obtaining underwater topographic data, and using a multi-modal deep learning framework to perform feature-level fusion and error minimization optimization on the underwater topographic data, to obtain the underwater topographic data; Integrating the water surface area, water surface elevation and underwater topographic data to calculate the water storage of the target surface water body.

[0011] Preferably, the water surface height data is collected and the water surface area is calculated, and the specific steps are as follows: Acquire four-band optical remote sensing images and calculate the NDWI index to distinguish between water bodies and non-water bodies:

[0012] in For green band reflectivity, Near-infrared reflectance, segmented by threshold. The cumulative pixel area of ​​the region is obtained ; Acquire three-band optical remote sensing images, and apply the SAM model to perform zero-sample segmentation of the RGB images to obtain water mask images, thus obtaining the segmented water regions:

[0013] in For the segmentation function, Input three-band RGB optical remote sensing images; right Water body index Spectral screening was performed, among which , , The reflectances of the green, red, and blue bands are respectively. If the area exceeds a preset threshold, the water area is confirmed, and the total area of ​​the mask pixels is calculated by integration. ; Acquire synthetic aperture radar imagery based on SAR echo intensity Calculate OTSU automatic threshold ,Will The pixels are classified as water areas, and the pixel areas are accumulated to obtain... ; Obtain water surface height data acquired by SWOT satellite Dynamically monitor water body boundaries and calculate the pixel area within the boundaries:

[0014] in Based on water level Edge detection function, This is the dynamic water height threshold.

[0015] Furthermore, considering environmental impact, the water surface area is calculated based on the aforementioned multi-source water surface observation data; the calculated water surface area results are then subjected to dynamic weight allocation and multi-source fusion sensitive to environmental factors to obtain the water surface area of ​​the target surface water body: the specific steps are as follows: Considering the influence of clouds and fog, construct the weights corresponding to the NDWI water index method:

[0016] wherein is a cloud cover error term, is a NDWI-specific adjustment parameter considering sensitive cloud cover interference; a terrain impact factor is considered to build the weight corresponding to the SAM model segmentation method:

[0017] wherein is a shadow interference error term, is a SAM-specific adjustment parameter considering sensitive vegetation shadow; a meteorological impact factor is considered to build the weight corresponding to the OTSU threshold segmentation method:

[0018] wherein is a rainfall interference error term, is a SAR-specific adjustment parameter considering sensitive all-weather rough data; a satellite acquisition error impact factor is considered to build the weight corresponding to the SWOT satellite:

[0019] wherein is a spatial resolution error term, is a SWOT-specific adjustment parameter considering monitoring scale and sensitivity; the water surface area of the target ground water body is calculated; .

[0020] Further, water level measurement is performed, and the specific steps are as follows: GNSS signal data is acquired, and GNSS-R technology is used to extract the preliminary water level :

[0021] wherein is the speed of light, is the reflection signal time delay; the water level is extracted through GNSS-IR technology ;

[0022] wherein, is the phase difference, is the signal frequency, is the satellite elevation angle; the water surface DEM is generated by using the LiDAR sensor, and is scanned synchronously in the SWOT satellite overpass period, and the satellite data is aligned through the ICP algorithm:

[0023] wherein is the SWOT point cloud, is the UAV point cloud, is the rotation matrix, is the translation vector; the average water surface height from the aligned DEM ; acquire the time delay of SWOT satellite radar echo , extract the water level : .

[0024] Further, all the calculated water level measurement data are fused and corrected to obtain the corrected water surface elevation, and the specific steps are as follows: construct the dynamic coefficient corresponding to GNSS-R technology:

[0025] wherein is the wave interference error term, is the unified adjustment parameter; construct the dynamic coefficient corresponding to GNSS-IR:

[0026] wherein is the phase drift error term; construct the dynamic coefficient corresponding to the UAV water level gauge:

[0027] wherein is the positioning error term; construct the dynamic coefficient corresponding to SWOT satellite technology:

[0028] wherein is the atmospheric delay error term; perform dynamic weight distribution and multi-source fusion to obtain the water surface elevation:

[0029] further optimize through Kalman filter iteration :

[0030] wherein is the Kalman gain based on variance, is the current measurement, is the predicted value, and the coefficients are iteratively optimized by least squares method to realize closed-loop correction, and As the corrected water surface elevation.

[0031] Further, the underwater topographic data is acquired, and the specific steps are as follows: The water depth is extracted by using single / multi-beam sounding technology :

[0032] Wherein is the sound speed, is the echo time delay, and the underwater area is covered by multi-beam fan scanning; Based on remote sensing image data , the water depth is extracted by using random forest method :

[0033] Wherein, is a pre-trained random forest model; The land surface topographic interpolation is acquired, and the terrain is estimated by using Kriging method :

[0034] Wherein, is the minimum variance weight, is the measured value of the surrounding land surface known point, is a global average constant.

[0035] Further, a multi-modal deep learning framework is used to learn the underwater topographic data, and the underwater topographic data is obtained, and the specific steps are as follows: The fusion weight corresponding to single / multi-beam sounding is constructed:

[0036] Wherein is a depth resolution error term, is a special parameter; The fusion weight corresponding to remote sensing inversion is constructed:

[0037] Wherein is a spectral noise error term, is a special parameter; The fusion weight corresponding to land surface topographic interpolation is constructed:

[0038] Wherein is an interpolation sparse error term calculated based on the known point density, is a special parameter; The multi-modal deep learning framework is used to learn the topographic data 、 、 Feature fusion is performed to obtain final bottom terrain data wherein represents a convolutional neural network including multiple convolutional layers and attention layers.

[0039] Further, the water storage of the target surface water body is calculated by synthesizing the water surface area, water surface elevation, and underwater terrain data, specifically by the steps of: building a four-dimensional hydrological coupling model; inputting the water surface area , the water surface elevation , and the underwater terrain data into the four-dimensional hydrological coupling model to calculate the water storage of the target surface water body:

[0040] wherein, is a seepage coefficient, is a vertical seepage gradient, is a volumetric thermal expansion coefficient, is a temperature change based on fused water temperature data, is a salinity expansion coefficient, is a change based on water salinity data, is a time series adjustment term.

[0041] Further, after obtaining the water storage, a water level-water storage model and an area-water storage model are built for subsequent dynamic monitoring of the target surface water body, specifically by the steps of: inputting the multi-temporal final water level , the final area , the final terrain , the seepage loss term , and the thermal expansion term as input sequences to build the water level-water storage model:

[0042] wherein the LSTM includes a gating mechanism (forget gate, input gate, and output gate) to capture long-term dependencies, and the network weights are optimized by training historical water storage data; building the area-water storage model:

[0043] wherein , , are fitting coefficients based on historical data, and the coefficients are updated in real time in combination with SWOT satellite dynamic monitoring data.

[0044] A kind of dynamic ground water storage survey system combined with SWOT satellite, including water surface area extraction module, water surface elevation extraction module, underwater topography extraction module, water storage calculation module; The water surface area extraction module is used to collect water surface height data and calculate the water surface area respectively;The dynamic weight distribution and multi-source fusion of environmental factors sensitive to all water surface area results obtained by calculation are carried out to obtain the water surface area of the target ground water body; The water surface elevation extraction module is used to measure water level;All water level measurement data obtained by calculation are fused and corrected to obtain the corrected water surface elevation; The underwater topography extraction module is used to obtain underwater topography data, and a multi-modal deep learning framework is used to perform feature-level fusion and error minimization optimization on the underwater topography data to obtain the underwater topography data; The water storage calculation module is used to integrate the water surface area, water surface elevation and underwater topography data to calculate the water storage of the target ground water body.

[0045] The beneficial effects of the present application are as follows: The dynamic ground water storage calculation method combined with SWOT satellite proposed in the present application effectively suppresses the influence of environmental factors on optical images using a dynamic weight fusion model, and combines the all-weather and wide coverage advantages of SAR and SWOT, so that the extraction accuracy and reliability of the final water surface area far exceed any single data source, achieving high-precision water surface area calculation. GNSS-R, GNSS-IR, unmanned aerial vehicle water level gauge and SWOT satellite monitoring technology are used for water level measurement, and all water level measurement data obtained by calculation are fused and corrected to effectively eliminate major error sources and achieve ultra-high precision water level measurement;A multi-modal deep learning framework is used to perform feature-level fusion and error minimization optimization on the underwater topography data to obtain the underwater topography data; The present application does not rely on any single technology, and through multi-source complementation, even when part of the data is of poor quality due to weather reasons, the system can still rely on other data sources such as SAR and SWOT to output reliable results, ensuring the continuity and stability of the survey work, solving the problem of single observation data source and environmental interference in the prior art, and having BRIEF DESCRIPTION OF DRAWINGS Figure 1 It is a flowchart of a dynamic ground water storage calculation method combined with SWOT satellite.

[0046] Figure 2 It is a block diagram of a dynamic ground water storage survey system combined with SWOT satellite. DETAILED DESCRIPTION

[0047] As Figure 1As shown, a method for calculating the dynamic surface water storage of the SWOT satellite, the method comprising the following steps: Collecting water surface height data and calculating water surface area respectively; performing dynamic weight distribution and multi-source fusion on the calculated water surface area results according to environmental factors to obtain the water surface area of the target surface water body; Performing water level measurement; fusing and correcting all calculated water level measurement data to obtain the corrected water surface elevation; Obtaining underwater topographic data, and using a multi-modal deep learning framework to perform feature-level fusion and error minimization optimization on the underwater topographic data to obtain the underwater topographic data; Integrating the water surface area, water surface elevation, and underwater topographic data to calculate the water storage of the target surface water body.

[0048] In one specific embodiment, water surface height data is collected and water surface area is calculated, and the specific steps are as follows: Obtain four-band optical remote sensing images, calculate the NDWI index to distinguish water bodies and non-water bodies:

[0049] wherein is the green band reflectance, is the near-infrared band reflectance, and the area is segmented by threshold to obtain ; Obtain three-band optical remote sensing images, apply the SAM model to the RGB images to obtain the water mask image and obtain the segmented water area:

[0050] wherein is the segmentation function, is the input three-band RGB optical remote sensing image; The is spectrally screened by the water index , wherein , , are the green, red, and blue band reflectances, respectively, and if is greater than a predetermined threshold, the water area is confirmed, and the total area of the mask pixels is calculated by integration to obtain ; Obtain synthetic aperture radar images, calculate the OTSU automatic threshold based on the SAR echo intensity , classify the pixels of as water areas, and accumulate the pixel area to obtain ; Acquiring water surface height data collected by SWOT satellite Dynamic monitoring of water body boundary, calculating the area of pixels within the boundary:

[0051] wherein is an edge detection function based on water level, is a dynamic water body height threshold.

[0052] In this embodiment, the specific process of using TSI terrain shading index to optimize SAM model segmentation includes calculating before SAM segmentation, wherein is the terrain slope angle (calculated based on DEM), is the solar zenith angle (obtained based on image time), is the solar azimuth angle (obtained based on image metadata), is the terrain azimuth angle (calculated based on DEM), and the TSI value ranges from 0 to 1, indicating from full shadow to full illumination; then the TSI is used as a correction factor for the SAM mask, and if TSI<0.5, the correction formula is to improve the pixel weight in low illumination area, thereby reducing the influence of error rate exceeding 30% in shadow interference area, and inputting the optimized into the dynamic weight fusion model to realize accurate correction of water body segmentation under vegetation coverage and complex terrain.

[0053] In one specific embodiment, considering environmental impact, water surface area calculation is performed based on the multi-source water surface observation data; dynamic weight distribution and multi-source fusion are performed on the calculated water surface area results based on environmental factor sensitivity, to obtain the water surface area of the target surface water body: the specific steps are as follows: Considering the cloud and fog impact factor, the weight corresponding to the NDWI water body index method is constructed:

[0054] wherein is a cloud and fog coverage error term ranging from 0 to 1, calculated in real time based on satellite cloud images, is a default value of 0.2, used as a special adjustment parameter for NDWI sensitive to cloud and fog interference; Considering the terrain impact factor, the weight corresponding to the SAM model segmentation method is constructed:

[0055] wherein is a shadow interference error term ranging from 0 to 1, calculated in combination with terrain DEM, is a default value of 0.15, used as a special adjustment parameter for SAM sensitive to vegetation shadow; ​Considering the meteorological influence factor, the weight corresponding to the OTSU threshold segmentation method is constructed:

[0056] wherein is the rainfall interference error term based on meteorological data, ranging from 0 to 1, is the default 0.25, which is a special adjustment parameter for SAR sensitive to all-weather but rough data; Considering the satellite acquisition error influence factor, the weight corresponding to the SWOT satellite is constructed:

[0057] wherein is the spatial resolution error term calculated based on the SWOT pixel size of 100-250 meters, ranging from 0 to 1, is the default 0.3, which is a special adjustment parameter for SWOT sensitive to large-scale monitoring; The water surface area of the target surface water body is calculated; .

[0058] In one specific embodiment, water level measurement is performed, and the specific steps are: GNSS signal data is acquired, and GNSS-R technology is used to extract the preliminary water level :

[0059] wherein is the speed of light, is the reflected signal time delay; The water level is extracted by GNSS-IR technology ;

[0060] wherein, is the phase difference, is the signal frequency, is the satellite elevation angle; The water surface DEM is generated by the LiDAR sensor, which is scanned synchronously during the SWOT satellite transit period, and the satellite data is aligned by the ICP algorithm:

[0061] wherein is the SWOT point cloud, is the unmanned aerial vehicle point cloud, is the rotation matrix, is the translation vector; the average water surface height is obtained from the aligned DEM ; The SWOT satellite radar echo time delay is acquired , extraction water level : .

[0062] In one specific embodiment, all the calculated water level measurement data are fused and corrected to obtain the corrected water surface elevation, and the specific steps are as follows: Construct the dynamic coefficient corresponding to the GNSS-R technology:

[0063] Wherein is a wave interference error term based on real-time wind speed data, ranging from 0 to 1, is a uniform adjustment parameter of 0.1-0.5, default 0.3; Construct the dynamic coefficient corresponding to the GNSS-IR:

[0064] Wherein is a phase drift error term based on signal SNR calculation, ranging from 0 to 1; Construct the dynamic coefficient corresponding to the unmanned aerial vehicle water level gauge:

[0065] Wherein is a positioning error term based on GPS / IMU fusion, ranging from 0 to 1; Construct the dynamic coefficient corresponding to the SWOT satellite technology:

[0066] Wherein is an atmospheric delay error term based on wet delay model calculation, ranging from 0 to 1; Perform dynamic weight distribution and multi-source fusion to obtain the water surface elevation:

[0067] Further optimize through Kalman filter iteration :

[0068] Wherein is the Kalman gain based on variance, is the current measurement, is the predicted value, and the least squares method is used to iteratively optimize the coefficient to realize closed-loop correction, and is taken as the corrected water surface elevation.

[0069] In one specific embodiment, the underwater terrain data is obtained, and the specific steps are as follows: Extracting water depth using single / multi-beam bathymetry :

[0070] wherein is the sound wave velocity, is the echo time delay, covering the seabed area by multi-beam fan scanning; Based on remote sensing image data , water depth is extracted using random forest method :

[0071] wherein, is a pre-trained random forest model; Obtain land surface terrain interpolation, estimate terrain using Kriging method :

[0072] wherein, is the minimum variance weight, is the measured value of the known point of the surrounding land surface, is the global average constant.

[0073] In one specific embodiment, a multi-modal deep learning framework is used to learn underwater terrain data, and the specific steps are as follows: Construct the fusion weight corresponding to single / multi-beam bathymetry:

[0074] wherein is the depth resolution error term based on water depth range, ranging from 0 to 1, is a special parameter of 0.2 by default; Construct the fusion weight corresponding to remote sensing inversion:

[0075] wherein is the spectral noise error term based on image quality, ranging from 0 to 1, is a special parameter of 0.15 by default; Construct the fusion weight corresponding to land surface terrain interpolation:

[0076] wherein is the interpolation sparsity error term based on known point density, ranging from 0 to 1, is a special parameter of 0.25 by default; Adopting a multi-modal deep learning framework on terrain data 、 、 Performing feature fusion to obtain final bottom terrain data , represents a convolutional neural network including multiple layers of convolution and attention layers.

[0077] In one specific embodiment, the water surface area, water surface elevation, and underwater terrain data are integrated to calculate and calculate the water storage of the target surface water body, specifically as follows: Building a four-dimensional hydrological coupling model; Inputting the water surface area , water surface elevation , and underwater terrain data into the four-dimensional hydrological coupling model to calculate the water storage of the target surface water body:

[0078] wherein, is a seepage coefficient calculated based on a riverbed sediment type database, such as clay , gravel , is a vertical seepage gradient derived from the water level gradient based on Darcy's law, is a volumetric thermal expansion coefficient of 0.00021 / °C by default, which is a standard value for water at around 20°C, is a temperature change based on fused water temperature data, is a salinity expansion coefficient, is a change based on water body salinity data, is a timing adjustment term based on seasonal fluctuations calculated from historical SWOT data.

[0079] In one specific embodiment, after obtaining the water storage, a water level-water storage model and an area-water storage model are also constructed for subsequent dynamic monitoring of the target surface water body, with the specific steps being: Inputting the multi-temporal final water level , final area , final terrain , seepage loss term , and thermal expansion term as input sequences to construct a water level-water storage model:

[0080] wherein the LSTM includes a gating mechanism (forget gate, input gate, and output gate) to capture long-term dependencies, and the network weights are optimized through training of historical water storage data; Constructing an area-water storage model:

[0081] wherein 、 、 are fitting coefficients calculated from the measured water storage samples by the least squares method based on historical data, and the coefficients are updated in real time in combination with the dynamic monitoring data of the SWOT satellite.

[0082] Embodiment 2 In this embodiment, a surface water storage survey is carried out for a real reservoir, and the method is implemented in combination with multi-source data (such as optical remote sensing images, SAR images, SWOT satellite data, GNSS signals, unmanned aerial vehicle LiDAR and depth sounder data). The survey time is the flood period in 2024, the environmental conditions include cloud cover of 40%, shadow interference of 30% and rainfall interference of 20%, and the resolution of the SWOT satellite is about 150 m.

[0083] Firstly, four-band optical remote sensing images (resolution 10 m) are collected, and the NDWI water body index method is used to extract the preliminary water surface area A1. The calculation formula is wherein G is the green band reflectivity (average 0.12), and NIR is the near-infrared band reflectivity (average 0.08). The region with NDWI>0 is segmented by threshold, and the cumulative pixel area is obtained as A1≈104.25 km² (error about 4.25% affected by cloud and fog).

[0084] Secondly, three-band RGB images (resolution 5 m) are segmented using the SAM model to extract the preliminary water surface area A2. The SAM model is applied to the RGB image for zero sample segmentation to obtain the water body mask image M=SAM(I), and further combined with the water body index (G≈0.15, R≈0.10, B≈0.08) for spectral screening, if WaterIndex>0.1, the water body region is confirmed, and the total area of the mask pixels is obtained by integration to obtain A2≈105.21 km².

[0085] Then, the SAR image (resolution 20 m) is segmented using the OTSU threshold segmentation method to extract the preliminary water surface area A3. Based on the SAR echo intensity Pixel (average -15 dB), the OTSU automatic threshold T_{OTSU}≈-12 dB is calculated, and the pixels with Pixel>T_{OTSU} are classified as water body regions, and the cumulative pixel area is obtained as A3≈112.20 km² (all-weather but affected by rainfall roughness).

[0086] Finally, the SWOT satellite data (high resolution 10 cm) is segmented using the SWOT technology to extract the preliminary water surface area A4. The water surface height data h(t) (average 4.8 m) is used to dynamically monitor the water body boundary, and the contour extraction formula Compute the area of pixels within the boundary, where h_{threshold}=mean(h(t))-std(h(t))≈4.5m, we get A4≈120.06km² (large scale but 15% small water body missing rate).

[0087] Dynamic weight fusion model is used to adjust the weight dynamically according to the weather conditions: Cloud cover error =0.40 (σ1=0.2), shadow interference =0.30 (σ2=0.15), rainfall interference =0.20 (σ3=0.25), spatial resolution =0.15 (σ4=0.3), we get w1≈0.18, w2≈0.22, w3≈0.30, w4≈0.30. The final water area A_{final}=w1·A1+w2·A2+w3·A3+w4·A4≈114.27km² is obtained by fusion, with an error of 14.27%, which is better than the average of ~20% of single method.

[0088] GNSS-R technology is used to analyze the reflected signal time delay τ≈33.4ns (light speed c=3×10^8m / s), formula ≈4.77m (wave interference error 25%). GNSS-IR technology measures the phase difference Δφ≈1.2rad (frequency f=1.575GHz, elevation angle e=30°), formula ≈4.61m (phase shift error 18%). The unmanned aerial vehicle water level gauge uses LiDAR to generate water surface DEM, aligns SWOT point cloud through ICP algorithm, minimizes , and the average height is ≈4.81m (positioning error 8%). The time delay τ≈33.4ns of SWOT satellite radar echo, formula ≈4.01m (atmospheric delay error 12%).

[0089] A four-source data closed-loop correction system is constructed, and the dynamic coefficient uses softmax form: =0.25, =0.18, =0.08, =0.12 (κ=0.3), ≈0.20, β≈0.25, δ≈0.35, γ≈0.20. Preliminary fusion ≈4.55m, further iteration optimization through Kalman filter, formula (K=0.5 based on variance, h_pred=5.10m), resulting in a final water level height h_corrected≈4.82m, with an error of 3.6%, see solving centimeter level waves and delay effects.

[0090] Using single / multi-beam bathymetry to emit sound waves, the echo time delay t≈4ms, formula ≈3.21m.

[0091] Using remote sensing inversion using random forest method to predict from image data X (feature vector [reflectance, texture]), formula z2=RF(X)≈2.94m.

[0092] Using land surface terrain interpolation based on kriging method, formula , weight λ_i based on autocorrelation, μ=3.0, resulting in z3≈3.33m.

[0093] Using multi-modal deep learning framework fusion, weight ( =0.10, η1=0.2)≈0.35, λ2≈0.30, λ3≈0.35. Fusion results in final bottom terrain data ≈3.19m, error 6.3%, see improving complex environment accuracy; Using four-dimensional hydrological coupling model to calculate water storage, , where γ=0.10 (gravel type), ∂q / ∂z=0.05 (Darcy's law), ε=0.00021 / °C, ΔT=5°C (SWOT water temperature data), β=0.0008 / psu, ΔS=2psu (salinity sensor), ΔV_time=10 (historical SWOT seasonal fluctuations). Calculate seepage loss≈0.01, thermal / salinity correction≈0.999, V_base≈114.27×(4.82-3.19)≈186.26, V≈195.34 million cubic meters, error 2.33%, see solving the pseudo-change problem.

[0094] Further construct water level-water storage model, through hidden layer 64, LSTM neural network processing time series data with MSE loss, input includes final water level , final area , final terrain , seepage loss term and thermal expansion term , formula , predict flood period trend (area-water storage model uses polynomial regression , linkage 4D-HCM optimization coefficient.

[0095] The results of this embodiment show that the method of the present invention significantly improves the accuracy of water storage estimation and is suitable for water resource management in complex environments.

[0096] Example 3 like Figure 2 As shown, a dynamic surface water storage survey system combining SWOT satellite data includes a water surface area extraction module, a water surface elevation extraction module, an underwater topography extraction module, and a water storage calculation module. The water surface area extraction module is used to collect water surface height data and calculate the water surface area separately; it performs dynamic weight allocation and multi-source fusion sensitive to environmental factors on all the calculated water surface area results to obtain the water surface area of ​​the target surface water body; The water surface elevation extraction module is used for water level measurement; it fuses and corrects all the calculated water level measurement data to obtain the corrected water surface elevation. The underwater terrain extraction module is used to acquire underwater terrain data and uses a multimodal deep learning framework to perform feature-level fusion and error minimization optimization on the underwater terrain data to obtain underwater terrain data. The water storage calculation module is used to calculate the water storage of the target surface water body by integrating water surface area, water surface elevation, and underwater topographic data.

[0097] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

Claims

1. A method of calculating dynamic surface water storage in combination with SWOT satellites, characterized in that: The method comprises the following steps: Collecting water surface height data and calculating water surface area respectively; performing dynamic weight distribution and multi-source fusion on all water surface area results obtained by calculation according to environmental factor sensitivity to obtain the water surface area of the target surface water body; Performing water level measurement; performing fusion and correction on all water level measurement data obtained by calculation to obtain the corrected water surface elevation; Obtaining underwater topographic data, and performing feature-level fusion and error minimization optimization on the underwater topographic data by using a multi-modal deep learning framework to obtain the underwater topographic data; Integrating the water surface area, the water surface elevation and the underwater topographic data to calculate the water storage of the target surface water body.

2. The method of claim 1, wherein: The specific steps of collecting water surface height data and calculating water surface area are as follows: Obtaining four-band optical remote sensing images, calculating the NDWI index to distinguish water bodies and non-water bodies: Wherein is the green band reflectance, is the near-infrared band reflectance, and is the area of the region accumulated by the pixels ; Obtaining three-band optical remote sensing images, applying the SAM model to the RGB images to obtain the water mask image and the segmented water area: wherein is a segmentation function, is an input RGB optical remote sensing image of three bands; To By water body index Spectral screening is performed, wherein , , Green, red, and blue band reflectance, respectively, if Is greater than a preset threshold value, a water body region is confirmed, and by integrating the total area of the mask pixels, is obtained ; Acquiring synthetic aperture radar images, based on SAR echo intensity Computing OTSU automatic threshold , the pixels of are classified as water body area, and the pixel area is accumulated to obtain ; Acquiring water surface height data collected by SWOT satellite Dynamically monitoring the water body boundary, calculating the area of pixels within the boundary: wherein is a water level based edge detection function, is a dynamic water height threshold.

3. The method of claim 2, wherein: Considering the environmental impact, the water surface area is calculated based on the multi-source water surface observation data; the water surface area results obtained by calculation are subjected to dynamic weight distribution and multi-source fusion according to environmental factor sensitivity to obtain the water surface area of the target surface water body; the specific steps are as follows: Considering the cloud and fog influence factor, the weight corresponding to the NDWI water index method is constructed: wherein is a cloud cover error term, is a NDWI-specific adjustment parameter that takes into account sensitive cloud cover interference; Considering the terrain influence factor, the weight corresponding to the SAM model segmentation method is constructed: wherein is a shadow interference error term, is a SAM-specific adjustment parameter that takes into account the shadow of sensitive vegetation; Considering the meteorological influence factor, the weight corresponding to the OTSU threshold segmentation method is constructed: wherein is a rainfall disturbance error term, is a SAR-specific adjustment parameter that takes into account sensitive all-weather roughness data; Considering the satellite acquisition error influence factor, the weight corresponding to the SWOT satellite is constructed: wherein is a spatial resolution error term, is a SWOT-specific adjustment parameter taking into account the monitoring scale and sensitivity; The water surface area of the target surface water body is calculated; .

4. The method of claim 1, wherein: The specific steps of performing water level measurement are as follows: Acquiring GNSS signal data, using GNSS-R technology to extract preliminary water level : ， wherein is the speed of light, is the time delay of the reflected signal; Extracting water level by GNSS-IR technique ; wherein, is the phase difference, is the signal frequency, is the satellite elevation angle; Generating water surface DEM by using LiDAR sensor, synchronously scanning in the SWOT satellite transit period, and aligning the satellite data by using ICP algorithm: where is the SWOT point cloud, is the drone point cloud, is the rotation matrix, is the translation vector; the average water surface height from the aligned DEM is obtained as ; Acquiring time delay of SWOT satellite radar echo , extracting water level : 。 5. The method of claim 4, wherein: The specific steps of performing fusion and correction on all water level measurement data obtained by calculation to obtain the corrected water surface elevation are as follows: Constructing the dynamic coefficient corresponding to the GNSS-R technology: wherein is a wave interference error term, is a uniform adjustment parameter; Constructing the dynamic coefficient corresponding to the GNSS-IR: wherein is a phase drift error term; Constructing the dynamic coefficient corresponding to the unmanned aerial vehicle water level gauge: wherein is a positioning error term; Constructing the dynamic coefficient corresponding to the SWOT satellite technology: wherein is the atmospheric delay error term; Performing dynamic weight distribution and multi-source fusion to obtain the water surface elevation: further iteratively optimised by Kalman filtering : wherein is the variance-based Kalman gain, is the current measurement, is the predicted value, the coefficients are iteratively optimized by least squares, a closed loop correction is achieved, and is the corrected water surface elevation.

6. The method of claim 1, wherein: The specific steps of obtaining underwater topographic data are as follows: Extracting water depth using single / multi-beam sounding technology : wherein is the sound wave velocity, is the echo time delay, covering the water bottom area by multi-beam sector scan; Based on remote sensing image data , water depth is extracted using random forest method : wherein, is a pre-trained random forest model; Obtain land surface topography interpolation, estimate topography using Kriging method : wherein, is the minimum variance weight, is the measured value of the peripheral terrestrial known point, is the global average constant.

7. The method of claim 6, wherein: Learning the underwater topographic data by using a multi-modal deep learning framework to obtain the underwater topographic data, and the specific steps are as follows: Constructing the fusion weight corresponding to the single / multi-beam sounding: wherein is a depth resolution error term, is a dedicated parameter; Constructing the fusion weight corresponding to the remote sensing inversion: wherein is a spectral noise error term, is a dedicated parameter; Constructing the fusion weight corresponding to the land surface topographic interpolation: wherein is an interpolated sparse error term based on known point density calculations, is a dedicated parameter; Adopting a multi-modal deep learning framework on terrain data 、 、 Performing feature fusion to obtain final bottom terrain data , wherein represents a convolutional neural network including multi-layer convolution and attention layer.

8. The method of claim 1, wherein: Integrating the water surface area, the water surface elevation and the underwater topographic data to calculate the water storage of the target surface water body, and the specific steps are as follows: Constructing a four-dimensional hydrological coupling model; inputting water surface area , water surface elevation , underwater topographic data into a four-dimensional hydrological coupling model, calculating water storage of a target surface water body: wherein, is a seepage coefficient, is a vertical seepage gradient, is a volumetric thermal expansion coefficient, is a temperature change based on fused water temperature data, is a salinity expansion coefficient, is a change based on water body salinity data, is a timing adjustment term.

9. The method of claim 8, wherein: After obtaining the water storage, a water level-water storage model and an area-water storage model are also constructed to monitor the target surface water body, and the specific steps are as follows: multiple time series of final water levels final areas final topography seepage loss terms and thermal expansion terms as input series, construct a water level- water storage model: The LSTM includes a gating mechanism (forget gate, input gate and output gate) to capture long-term dependencies, and the network weight is optimized by training historical water storage data; Constructing an area-water storage model: wherein , , are fitting coefficients based on historical data, which are updated in real time in combination with dynamic monitoring data of the SWOT satellite.

10. A system for integrating the dynamic surface water storage survey of the SWOT satellite, characterized in that: Including a water surface area extraction module, a water surface elevation extraction module, an underwater topographic extraction module and a water storage calculation module; The water surface area extraction module is used for collecting water surface height data and calculating water surface areas respectively; all the calculated water surface area results are subjected to dynamic weight distribution and multi-source fusion according to environmental factors, so as to obtain the water surface area of the target ground water body; The water surface elevation extraction module is used for water level measurement; all the calculated water level measurement data are subjected to fusion and correction, so as to obtain the corrected water surface elevation; The underwater topography extraction module is used for obtaining underwater topography data, and a multi-modal deep learning framework is adopted to perform feature-level fusion and error minimization optimization on the underwater topography data, so as to obtain the underwater topography data; The water storage calculation module is used for comprehensively calculating the water storage of the target ground water body according to the water surface area, water surface elevation and underwater topography data.

Images