Method for automatic estimation and correction of reservoir storage capacity based on remote sensing image and DEM

Abstract

The application discloses a reservoir water storage capacity automatic estimation and correction method based on remote sensing images and DEM, which comprises the following steps: obtaining remote sensing images and DEM data of a target reservoir or dam, pre-processing the remote sensing images, and extracting a water body boundary; superimposing the water body boundary line on the DEM data of the target reservoir or dam, extracting the boundary elevation of each pixel at the water body boundary, and obtaining the minimum value, maximum value, average value and median value of the boundary elevation; according to the terrain slope and water body recognition confidence of each boundary pixel, a weight is assigned to each boundary pixel, and the weighted water level is calculated by using the boundary elevation and weight of each boundary pixel; the water storage capacity corresponding to the five water levels is calculated; the confidence interval width and standard deviation of the water storage capacity of the target reservoir or dam are obtained by using the Monte Carlo simulation method; the confidence interval width, standard deviation, all water storage capacities before the Monte Carlo simulation, and the corresponding water body area and water level are input into a trained machine learning model, and the corrected water storage capacity of the target reservoir or dam is obtained.

Description

Technical Field

[0001] This invention relates to a method for predicting water storage capacity, specifically to an automatic estimation and correction method for reservoir water storage capacity based on remote sensing images and DEM. Background Technology

[0002] The water storage capacity of reservoirs and ponds is a key parameter in water resource management and flood control scheduling. Traditionally, water storage capacity (i.e., reservoir capacity) is mainly obtained through on-site monitoring data such as water gauges or radar elevation measurements, as well as pre-determined water level-capacity curves. However, for small reservoirs or ponds lacking on-site monitoring, such data is often missing, resulting in the inability to obtain water storage capacity in a timely and accurate manner. With the increasing demand for refined water resource management, using remote sensing imagery combined with digital elevation models (DEMs) for water storage capacity estimation has become an important approach. However, existing technologies have shortcomings in the following aspects: Limited accuracy in water body extraction: Traditional remote sensing water body extraction often employs threshold segmentation methods such as Normalized Difference Water Index (NDWI) and modified NDWI, or supervised classification-based methods. While these methods are computationally simple and can identify water bodies on a large scale, they are often inaccurate in terms of water body boundary details, small water areas, and situations with cloud shadows or vegetation obstruction, easily resulting in edge breaks or omissions.

[0003] DEM volume estimation has significant errors: DEM-based water storage calculation methods typically assume that the water surface elevation is known, and then calculate the volume difference in height between the water surface and the terrain. A common practice is to overlay the extracted water surface area onto the DEM, assuming that the entire water surface is at the same height. For each raster cell within the water surface area, use the formula Calculate the unit volume (where This is the ground elevation of that pixel. The water storage capacity is calculated by summing the area of ​​all pixels (where the area is the pixel area). However, in practical applications... Often unknown, the results can only be approximated by the DEM elevation of the water body boundary pixels, making the results unreliable.

[0004] In summary, there is currently a lack of a method that can automatically extract the water body range and infer the water level, accurately calculate the water storage using DEM, and provide uncertainty analysis and achieve result correction and generalization through machine learning, so as to meet the needs of water storage estimation for large-scale unmonitored water bodies. Summary of the Invention

[0005] In view of the above-mentioned shortcomings in the prior art, the automatic estimation and correction method for reservoir water storage based on remote sensing images and DEM provided by the present invention solves the problem that the prior art cannot automatically and accurately estimate the water storage of small reservoirs or ponds using remote sensing images and DEM when there is a lack of on-site water level monitoring.

[0006] To achieve the above-mentioned objectives, the technical solution adopted by this invention is as follows: An automatic estimation and correction method for reservoir water storage based on remote sensing imagery and DEM is provided, comprising the following steps: S1. Acquire remote sensing images and DEM data of the area where the target reservoir or pond is located, preprocess the remote sensing images, and use the preprocessed remote sensing images to extract the water body boundary. S2. Overlay the water body boundary line onto the DEM data of the target reservoir or pond, extract the boundary elevation of each cell at the water body boundary, and then obtain the minimum, maximum, average and median values ​​of the boundary elevation. S3. Assign weights to each boundary cell based on its terrain slope and water body identification confidence, and calculate the weighted water level using the boundary elevation and weight of each boundary cell. S4. Calculate the water storage capacity corresponding to each water level based on the minimum, maximum, average, and median values ​​of the weighted water level and boundary elevation. S5. Based on the DEM pixels and water body boundaries, the confidence interval width and standard deviation of the water storage capacity of the target reservoir or pond are obtained using the Monte Carlo simulation method. S6. Input the confidence interval width, standard deviation, all water storage volumes before Monte Carlo simulation, and their corresponding water areas and water levels into the trained machine learning model to obtain the corrected water storage volume of the target reservoir or pond.

[0007] Furthermore, step S3 further includes: S31. Assign the same initial weight to each boundary cell on the water body boundary line. For each boundary cell on the water body boundary, calculate the slope within the preset area with it as the center. S32. When the slope is less than the preset threshold and the current boundary pixel is identified as a water boundary in multiple temporal images, its initial weight is increased by the preset value; otherwise, its initial weight is decreased by the preset value. S33. Normalize the updated weights of all boundary cells and make the sum of the normalized weights of all boundary cells equal to the total number of boundary cells. S34. Calculate the weighted water level using the boundary elevation of the boundary pixels and their normalized weights: in, Weighted water level; w i h is the normalized weight of the i-th boundary cell; i Let be the boundary elevation of the i-th boundary cell; n is the total number of boundary cells.

[0008] Furthermore, step S3 further includes: For each boundary cell, calculate the slope within the preset area centered on it; Confidence scores are extracted based on the slope and water body corresponding to each boundary cell, and confidence weights are calculated: in, , and These represent the confidence weight of the i-th boundary cell, the confidence score for water extraction, and the slope, respectively. Calculate the weighted water level based on the boundary elevation and confidence weight of the boundary cells: in, Weighted water level; h i Let be the boundary elevation of the i-th boundary cell; n is the total number of boundary cells.

[0009] Furthermore, in step S4, the method for calculating the water storage corresponding to each water level includes: For each DEM cell within the water body boundary area, select DEM cells whose ground elevation is less than the set water level height. Calculate the water storage volume of the selected DEM pixels: in, , and These represent the water volume, ground elevation, and pixel area of ​​the j-th selected DEM pixel, respectively; H is the set water level height, which is the minimum, maximum, average, median, or weighted water level of the boundary elevation; By summing the water storage volumes of all selected DEM pixels, the total water storage capacity of the target reservoir or pond can be obtained. .

[0010] Furthermore, step S5 further includes: S51. Determine the probability distribution of DEM elevation error based on the accuracy specifications provided by the DEM data provider or the results of on-site calibration. S52. In each Monte Carlo iteration, a random noise source selected from a probability distribution is added to each DEM cell, and spatial correlation is introduced by convolution smoothing or covariance matrix method. S53. Set a maximum disturbance distance d for each water body boundary segment, and randomly select a disturbance amount between 0 and d to expand or shrink the water body boundary based on the confidence level of the boundary pixels. S54. Based on the adjusted water body boundary and DEM elevation, execute steps S2~S4, and then calculate the average water storage based on all water storage obtained in each iteration. S55. Determine whether the Monte Carlo iteration count has reached the preset iteration count. If yes, proceed to step S56; otherwise, return to step S52. S56. Based on all the average water storage values ​​obtained through iteration, calculate the confidence interval and standard deviation: in, V is the average of all average water storage capacities; k σ is the average water storage capacity at the k-th iteration; V The standard deviation is N; the preset number of iterations is N. and Take the first one respectively The position and the first The corresponding bit As the lower and upper bounds of the 95% confidence interval; The confidence interval; The cumulative distribution function of water storage results Quantile function.

[0011] Furthermore, in step S53, when the confidence level of the boundary cell is greater than the preset confidence level, a perturbation amount between 0 and d is randomly selected to expand the water body boundary; when the confidence level of the boundary cell is less than or equal to the preset confidence level, a perturbation amount between 0 and d is randomly selected to shrink the water body boundary.

[0012] Furthermore, step S1 further includes: S11. Acquire remote sensing images of the area where the target reservoir or pond is located, and perform atmospheric correction and cloud detection on the remote sensing images to obtain preprocessed remote sensing images. S12. Based on the preprocessed remote sensing images, extract the boundary pixels of suspected water areas using the spectral index method or a deep learning segmentation model. S13. Perform depression analysis on the DEM data to identify closed depressions in the terrain and generate depression masks. Overlay the boundary pixels of the depression masks and suspected water areas. S14. Use the boundary pixels of all suspected water areas located within the depression mask as the water body boundary.

[0013] Furthermore, methods for training machine learning models include: Acquire remote sensing images of multiple reservoirs with actual water storage capacity and the actual water storage capacity at the time the remote sensing images were acquired; Steps S1 to S5 are used to obtain the confidence interval width and standard deviation for multiple reservoirs; The confidence interval width and standard deviation of each reservoir, all water storage volumes before the Monte Carlo simulation and their corresponding water areas and water levels are used as inputs to the machine learning model, and the actual water storage volume is used as the output of the machine learning model. The input and output corresponding to each reservoir are used as a training sample. The machine learning model is trained using all the training samples to obtain the trained machine learning model.

[0014] Furthermore, the machine learning model is a random forest, gradient boosting decision tree, or ensemble tree model.

[0015] Compared with the prior art, the present invention has the following outstanding advantages: 1. Fully automated remote sensing calculation, no on-site data required: This solution relies entirely on remote sensing imagery and DEM for water storage estimation, eliminating dependence on manual monitoring stations, water gauge readings, or existing reservoir capacity curves. For remote areas or small reservoirs without supervision, it enables "unattended" water storage monitoring, significantly expanding monitoring coverage and frequency.

[0016] 2. Multi-scenario water level fusion to improve volumetric accuracy: To address the uncertainty caused by unknown water levels, multiple water level scenarios, such as minimum, maximum, average, and median values, are introduced to calculate water storage capacity and obtain a range of results. A reliability-weighted mechanism is used to fully utilize reliable boundary elevations to calculate weighted water levels. This effectively reduces interference from outliers, making the water level estimate closer to the actual water level. The final output water storage range [ , Covers possible true values, while weighted results As the best estimate, it significantly improves the accuracy of reservoir capacity calculation.

[0017] 3. Monte Carlo simulation to quantify uncertainty: This method comprehensively assesses the error range of water storage estimation through Monte Carlo random simulation, providing quantitative indicators such as the 95% confidence interval. This uncertainty analysis is lacking in traditional methods, and it informs users of the reliability of the estimation results in probabilistic form. For example, it can report "the 95% confidence interval of a reservoir's water storage is ±5%", providing a scientific basis for risk assessment and decision-making.

[0018] 4. Data-Driven Correction Improves Accuracy: Utilizing a machine learning model to correct biases in remote sensing estimation results significantly reduces systematic errors. The model, combining multi-source input features (parameters input in step S6), learns the bias patterns of the remote sensing method, resulting in corrected estimates that more closely approximate the actual water storage. For example, when the DEM shows an overall overestimation, the model can automatically correct the estimate upwards; in cases of extreme drought or flooding where empirical formulas are inaccurate, the model can also provide compensatory adjustments based on features. Testing showed that applying this method to small reservoirs reduced the water storage estimation error from 10%–20% before correction to 5%–10%, a significant improvement in accuracy. This means that even for ponds and dams never monitored before, the water storage results obtained using the aforementioned process can approach the accuracy of actual measurements, providing the possibility for dynamic monitoring of large-scale, unregulated water bodies.

[0019] 5. Strong Model Generalization and Transfer Ability: Through appropriate training strategies and input uncertainty features, the machine learning calibration model exhibits good generalization performance across different regions and water bodies. Even in small reservoirs without monitoring data, the model can provide reliable estimates and will not fail due to a lack of specialized training. Furthermore, the model's adaptability can be further enhanced by continuously updating training data, introducing regional divisions, or transfer learning, ensuring the robustness of the method when it is widely deployed.

[0020] 6. Easy Integration into Real-Time Monitoring Systems: This method has a clear workflow and a high degree of modularity, making it easy to embed into existing remote sensing monitoring platforms or water conservancy information systems. Utilizing cloud computing or local servers, batch processing and analysis of multi-temporal satellite imagery can achieve real-time updates of reservoir water storage status. Managers can intuitively obtain water storage dynamics and early warning information for each reservoir through a visual interface, significantly improving the timeliness and intelligence of monitoring. Attached Figure Description

[0021] Figure 1 This is a flowchart of a method for automatically estimating and correcting reservoir water storage based on remote sensing imagery and DEM.

[0022] Figure 2 The diagram shows the calculation of water storage volume by overlaying DEM with the elevation of the reservoir boundary; the left image shows the reservoir's planar extent extracted from the remote sensing image; the right image shows a schematic diagram of the topographic elevation profile along this profile direction. Detailed Implementation

[0023] The specific embodiments of the present invention are described below to enable those skilled in the art to understand the present invention. However, it should be understood that the present invention is not limited to the scope of the specific embodiments. For those skilled in the art, various changes are obvious as long as they are within the spirit and scope of the present invention as defined and determined by the appended claims. All inventions utilizing the concept of the present invention are protected.

[0024] refer to Figure 1 , Figure 1 A flowchart illustrating an automatic estimation and correction method for reservoir water storage based on remote sensing imagery and DEM is shown; Figure 1 As shown, the method S includes steps S1 to S6.

[0025] In step S1, remote sensing images and DEM data of the area where the target reservoir or pond is located are acquired, the remote sensing images are preprocessed, and the water body boundaries are extracted using the preprocessed remote sensing images. In one embodiment of the present invention, step S1 further includes: S11. Acquire remote sensing images of the area where the target reservoir or pond is located, and perform atmospheric correction and cloud detection on the remote sensing images to obtain preprocessed remote sensing images. In this approach, ESA Sentinel-2 imagery (10m resolution) or commercial high-resolution satellite imagery is preferred to balance spatial accuracy and temporal coverage. For DEM data, the elevation accuracy (such as root mean square error) is determined based on the data provider's specifications or field calibration results, and the spatial resolution and projected coordinate system of the imagery are resampled and aligned.

[0026] S12. Based on the preprocessed remote sensing image, extract the boundary pixels of suspected water areas using a spectral index-based method or a deep learning segmentation model; the specific implementation process of step S12 may include: For spectral indices: calculate the Normalized Differential Water Index (NDWI) or an improved water body index, perform threshold segmentation on image pixels, and extract a set of suspected water body pixels; the threshold can be determined empirically or adaptively using the Otsu method, and can be adjusted according to the spectral characteristics of the image and the underlying surface conditions.

[0027] Deep learning segmentation model: A water body recognition network based on a U-Net structure is used. Processed multispectral images are input into the neural network, which outputs pixel-level water body probability maps. Then, an appropriate probability threshold is selected to obtain a binary water body mask. For the extraction of small water bodies, DEM auxiliary information can be fused to improve accuracy—for example, limiting water body prediction to low-lying areas to reduce misclassification of high-lying areas. The initial water body mask obtained through the above methods can be smoothed at its boundaries using morphological operations (such as opening and closing operations) and isolated noise points can be removed to obtain connected water surface polygons.

[0028] After executing step S12, the polygon of the target reservoir's water body and the set of boundary pixels of the suspected water area in the image coordinates can be obtained (including the DEM elevation value and confidence information of each boundary pixel).

[0029] S13. Perform depression analysis on the DEM data to identify closed depressions in the terrain and generate depression masks. Overlay the boundary pixels of the depression masks and suspected water areas. S14. Use the boundary pixels of all suspected water areas located within the depression mask as the water body boundary.

[0030] In step S2, the water body boundary line is superimposed onto the DEM data of the target reservoir or pond. The resulting schematic diagram can be found by referring to... Figure 2 The left sub-image in the image shows the reservoir's planar extent (blue polygon) extracted from the remote sensing image. Then, the boundary elevation of each pixel at the water body boundary is extracted, followed by obtaining the minimum, maximum, average, and median values ​​of the boundary elevation. In step S3, each boundary cell is weighted based on its terrain slope and water body identification confidence level, and the weighted water level is calculated using the boundary elevation and weight of each boundary cell. This scheme provides two different methods for calculating the weighted water level: The first method for calculating weighted water levels includes: S31. Assign the same initial weight to each boundary cell on the water body boundary line. For each boundary cell on the water body boundary, calculate the slope within the preset area with it as the center. S32. When the slope is less than the preset threshold and the current boundary pixel is identified as a water boundary in multiple temporal images, its initial weight is increased by the preset value; otherwise, its initial weight is decreased by the preset value. S33. Normalize the updated weights of all boundary cells and make the sum of the normalized weights of all boundary cells equal to the total number of boundary cells. S34. Calculate the weighted water level using the boundary elevation of the boundary pixels and their normalized weights: in, Weighted water level; w i h is the normalized weight of the i-th boundary cell; i Let be the boundary elevation of the i-th boundary cell; n is the total number of boundary cells.

[0031] The second method for calculating weighted water levels includes: For each boundary cell, calculate the slope within the preset area centered on it; Confidence scores are extracted based on the slope and water body corresponding to each boundary cell, and confidence weights are calculated: in, , and These represent the confidence weight of the i-th boundary cell, the confidence score for water extraction, and the slope, respectively. Calculate the weighted water level based on the boundary elevation and confidence weight of the boundary cells: in, Weighted water level; h i Let be the boundary elevation of the i-th boundary cell; n is the total number of boundary cells.

[0032] This scheme can improve the accuracy of water level estimation by introducing a water level confidence weighting mechanism, and calculate the water storage range by combining multiple water level scenarios to reduce the deviation of a single estimate.

[0033] In step S4, the water storage capacity corresponding to each water level is calculated based on the minimum, maximum, average, and median values ​​of the weighted water level and boundary elevation. In implementation, this scheme preferably includes the following method in step S4 for calculating the water storage capacity corresponding to each water level: For each DEM cell within the water body boundary area, select DEM cells whose ground elevation is less than the set water level height. Calculate the water storage volume of the selected DEM pixels: in, , and These represent the water volume, ground elevation, and pixel area of ​​the j-th selected DEM pixel, respectively; H is the set water level height, which is the minimum, maximum, average, median, or weighted water level of the boundary elevation; By summing the water storage volumes of all selected DEM pixels, the total water storage capacity of the target reservoir or pond can be obtained. The illustration can be found here. Figure 2 The right subgraph in the middle, the right subgraph is along Figure 2 The red section in the left sub-figure shows the terrain elevation profile. The black solid line represents the terrain surface, with the height measured in altitude. The blue dashed line represents the water level under different assumptions. The blue shaded area represents the volume difference between the water surface and the terrain, i.e., the water storage.

[0034] In step S5, based on the DEM pixels and water body boundaries, the confidence interval width and standard deviation of the target reservoir or pond's water storage capacity are obtained using the Monte Carlo simulation method; step S5 further includes: S51. Determine the probability distribution of DEM elevation error based on the accuracy specifications provided by the DEM data provider or the results of on-site calibration. S52. In each Monte Carlo iteration, a random noise source selected from a probability distribution is added to each DEM cell, and spatial correlation is introduced by convolution smoothing or covariance matrix method. S53. Set a maximum disturbance distance d for each water body boundary segment, and randomly select a disturbance amount between 0 and d to expand or shrink the water body boundary based on the confidence level of the boundary pixels. S54. Based on the adjusted water body boundary and DEM elevation, execute steps S2~S4, and then calculate the average water storage based on all water storage obtained in each iteration. S55. Determine whether the Monte Carlo iteration count has reached the preset iteration count. If yes, proceed to step S56; otherwise, return to step S52. S56. Based on all the average water storage values ​​obtained through iteration, calculate the confidence interval and standard deviation: in, V is the average of all average water storage capacities; k σ is the average water storage capacity at the k-th iteration; V The standard deviation is N; the preset number of iterations is N. and Take the first one respectively The position and the first The corresponding bit As the lower and upper bounds of the 95% confidence interval; The confidence interval; The cumulative distribution function of water storage results Quantile function.

[0035] In each Monte Carlo simulation, the average water storage for the five different water storage levels is calculated to represent the estimated water storage for the current simulation. This approach can comprehensively utilize information from multiple scenarios, reduce random fluctuations caused by extreme water level assumptions in a single simulation, and make the distribution of Monte Carlo output more stable and closer to reality.

[0036] In practice, this scheme preferably involves the following steps in step S53: when the confidence level of the boundary cell is greater than the preset confidence level, a perturbation amount between 0 and d is randomly selected to expand the water body boundary; when the confidence level of the boundary cell is less than or equal to the preset confidence level, a perturbation amount between 0 and d is randomly selected to shrink the water body boundary.

[0037] In step S6, the confidence interval width, standard deviation, all water storage volumes before the Monte Carlo simulation, and their corresponding water areas and levels are input into the trained machine learning model to obtain the corrected water storage volume of the target reservoir or pond. The machine learning model is a random forest, gradient boosting decision tree, or ensemble tree model XGBoost.

[0038] In implementation, the preferred methods for training the machine learning model in this scheme include: Acquire remote sensing images of multiple reservoirs with actual water storage capacity and the actual water storage capacity at the time the remote sensing images were acquired; Steps S1 to S5 are used to obtain the confidence interval width and standard deviation for multiple reservoirs; The confidence interval width and standard deviation of each reservoir, all water storage volumes before the Monte Carlo simulation and their corresponding water areas and water levels are used as inputs to the machine learning model, and the actual water storage volume is used as the output of the machine learning model. The input and output corresponding to each reservoir are used as a training sample. The machine learning model is trained using all training samples to obtain the trained model. During training, to fully utilize the data and avoid overfitting, the following techniques can be used: Cross-validation can be used to evaluate model performance, or a portion of the data can be reserved as a validation set for parameter tuning.

[0039] During training, the model iteratively updates its internal parameters (such as tree splitting decisions, leaf node values, etc.) to minimize the predicted value. Compared with the true value The loss between the two is considered. For regression problems, mean squared error (MSE) or mean absolute error (MAE) is generally used as the loss function. Taking the ensemble tree model XGBoost as an example, it gradually grows decision trees through the gradient boosting algorithm. Each new tree fits the current residual, thus gradually approaching the true value. Training is repeated until a preset stopping condition is reached (such as the number of trees, minimum loss descent, etc.).

[0040] When the trained model is used to target reservoirs or ponds, its main function is to correct the water storage volume calculated in step S4. For the water storage volume estimation result obtained from a satellite observation of the target area (the parameters input in step S6), the trained model is input, and a corrected water storage volume estimate is output. This result is an adjustment to the current remote sensing results by comprehensively considering historical statistical patterns.

[0041] If the model detects The range of water storage corresponding to the minimum and maximum boundary elevations is generally low (which may mean that the overall DEM is too high, leading to an underestimation), which will... Upward correction: If the observed water surface area is abnormally small and exceeds the training sample range, the model can also provide appropriate bias compensation using input features. Because the model considers multiple input combinations and corresponding true values ​​during training, it has a certain generalization ability for samples in different situations, can correct systematic biases, and output a more accurate estimate of water storage.

[0042] In summary, this invention innovatively combines multi-source remote sensing, physical models, and data-driven correction to form a complete automatic estimation method for the water storage capacity of small reservoirs. This method can provide relatively accurate estimates of water storage capacity and its uncertainty range even without on-site monitoring, offering a powerful tool for regional water resource management and showing broad application prospects in flood control, drought relief, and reservoir operation.

Claims

1. A method for automatic estimation and correction of reservoir water storage based on remote sensing imagery and DEM, characterized in that, Including the following steps: S1. Acquire remote sensing images and DEM data of the area where the target reservoir or pond is located, preprocess the remote sensing images, and use the preprocessed remote sensing images to extract the water body boundary. S2. Overlay the water body boundary line onto the DEM data of the target reservoir or pond, extract the boundary elevation of each cell at the water body boundary, and then obtain the minimum, maximum, average and median values ​​of the boundary elevation. S3. Assign weights to each boundary cell based on its terrain slope and water body identification confidence, and calculate the weighted water level using the boundary elevation and weight of each boundary cell. S4. Calculate the water storage capacity corresponding to each water level based on the minimum, maximum, average, and median values ​​of the weighted water level and boundary elevation. S5. Based on the DEM pixels and water body boundaries, the confidence interval width and standard deviation of the water storage capacity of the target reservoir or pond are obtained using the Monte Carlo simulation method. S6. Input the confidence interval width, standard deviation, all water storage volumes before Monte Carlo simulation, and their corresponding water areas and water levels into the trained machine learning model to obtain the corrected water storage volume of the target reservoir or pond.

2. The method for automatic estimation and correction of reservoir water storage based on remote sensing imagery and DEM according to claim 1, characterized in that, Step S3 further includes: S31. Assign the same initial weight to each boundary cell on the water body boundary line. For each boundary cell on the water body boundary, calculate the slope within the preset area with it as the center. S32. When the slope is less than the preset threshold and the current boundary pixel is identified as a water boundary in multiple temporal images, its initial weight is increased by the preset value; otherwise, its initial weight is decreased by the preset value. S33. Normalize the updated weights of all boundary cells and make the sum of the normalized weights of all boundary cells equal to the total number of boundary cells. S34. Calculate the weighted water level using the boundary elevation of the boundary pixels and their normalized weights: in, Weighted water level; w i h is the normalized weight of the i-th boundary cell; i Let be the boundary elevation of the i-th boundary cell; n is the total number of boundary cells.

3. The method for automatic estimation and correction of reservoir storage capacity based on remote sensing imagery and DEM according to claim 1, characterized in that, Step S3 further includes: For each boundary cell, calculate the slope within the preset area centered on it; Confidence scores are extracted based on the slope and water body corresponding to each boundary cell, and confidence weights are calculated: in, , and These represent the confidence weight of the i-th boundary cell, the confidence score for water extraction, and the slope, respectively. Calculate the weighted water level based on the boundary elevation and confidence weight of the boundary cells: in, Weighted water level; h i Let be the boundary elevation of the i-th boundary cell; n is the total number of boundary cells.

4. The method for automatic estimation and correction of reservoir water storage based on remote sensing imagery and DEM according to claim 1, characterized in that, In step S4, the method for calculating the water storage corresponding to each water level includes: For each DEM cell within the water body boundary area, select DEM cells whose ground elevation is less than the set water level height. Calculate the water storage volume of the selected DEM pixels: in, , and These represent the water volume, ground elevation, and pixel area of ​​the j-th selected DEM pixel, respectively; H is the set water level height, which is the minimum, maximum, average, median, or weighted water level of the boundary elevation; By summing the water storage volumes of all selected DEM pixels, the total water storage capacity of the target reservoir or pond can be obtained. .

5. The method for automatic estimation and correction of reservoir water storage based on remote sensing imagery and DEM according to claim 1, characterized in that, Step S5 further includes: S51. Determine the probability distribution of DEM elevation error based on the accuracy specifications provided by the DEM data provider or the results of on-site calibration. S52. In each Monte Carlo iteration, a random noise source selected from a probability distribution is added to each DEM cell, and spatial correlation is introduced by convolution smoothing or covariance matrix method. S53. Set a maximum disturbance distance d for each water body boundary segment, and randomly select a disturbance amount between 0 and d to expand or shrink the water body boundary based on the confidence level of the boundary pixels. S54. Based on the adjusted water body boundary and DEM elevation, execute steps S2~S4, and then calculate the average water storage based on all water storage obtained in each iteration. S55. Determine whether the Monte Carlo iteration count has reached the preset iteration count. If yes, proceed to step S56; otherwise, return to step S52. S56. Based on all the average water storage values ​​obtained through iteration, calculate the confidence interval and standard deviation: in, V is the average of all average water storage capacities; k σ is the average water storage capacity at the k-th iteration; V The standard deviation is denoted as N; the preset number of iterations is N. and Take the first one respectively The position and the first The corresponding bit As the lower and upper bounds of the 95% confidence interval; The confidence interval; The cumulative distribution function of water storage results Quantile function.

6. The method for automatic estimation and correction of reservoir water storage based on remote sensing imagery and DEM according to claim 5, characterized in that, In step S53, when the confidence level of the boundary cell is greater than the preset confidence level, a perturbation amount between 0 and d is randomly selected to expand the water body boundary; when the confidence level of the boundary cell is less than or equal to the preset confidence level, a perturbation amount between 0 and d is randomly selected to shrink the water body boundary.

7. The method for automatic estimation and correction of reservoir water storage based on remote sensing imagery and DEM according to claim 1, characterized in that, Step S1 further includes: S11. Acquire remote sensing images of the area where the target reservoir or pond is located, and perform atmospheric correction and cloud detection on the remote sensing images to obtain preprocessed remote sensing images. S12. Based on the preprocessed remote sensing images, extract the boundary pixels of suspected water areas using the spectral index method or a deep learning segmentation model. S13. Perform depression analysis on the DEM data to identify closed depressions in the terrain and generate depression masks. Overlay the boundary pixels of the depression masks and suspected water areas. S14. Use the boundary pixels of all suspected water areas located within the depression mask as the water body boundary.

8. The method for automatic estimation and correction of reservoir water storage based on remote sensing imagery and DEM according to claim 5, characterized in that, Methods for training machine learning models include: Acquire remote sensing images of multiple reservoirs with actual water storage capacity and the actual water storage capacity at the time the remote sensing images were acquired; Steps S1 to S5 are used to obtain the confidence interval width and standard deviation for multiple reservoirs; The confidence interval width and standard deviation of each reservoir, all water storage volumes before the Monte Carlo simulation and their corresponding water areas and water levels are used as inputs to the machine learning model, and the actual water storage volume is used as the output of the machine learning model. The input and output corresponding to each reservoir are used as a training sample. The machine learning model is trained using all the training samples to obtain the trained machine learning model.

9. The method for automatic estimation and correction of reservoir water storage based on remote sensing imagery and DEM according to claim 1 or 8, characterized in that, The machine learning model is a random forest, gradient boosting decision tree, or ensemble tree model.

Images