Pit-pond water storage amount measuring and calculating method based on dynamic grid sampling

Abstract

The invention discloses a pit-pond water storage amount measuring and calculating method based on dynamic grid sampling, and relates to the technical field of water resource monitoring, and the method comprises the steps: based on an optical image and an improved normalized water body index, extracting a pit-pond water surface, and calculating a pit-pond area and a pit-pond number in a task area; a grid sampling method is used for carrying out classified sampling on the pit ponds; based on the territorial area, the pit-pond area and the pit-pond number in the task area, the pit-pond density and the grid width are calculated, dynamic grid generation is conducted, and batch sampling is conducted on pits and ponds through programming software; according to urban territorial space monitoring road network data, big data analysis is utilized, the quality of pit-pond samples is evaluated and replaced, the pit-pond samples are coded, and geographic information software is utilized to construct a sample database; and constructing a water depth inversion model by utilizing programming software and a machine learning algorithm to obtain a water depth inversion result, and calculating the water storage amount of the pit pond by combining the area of the pit pond. According to the invention, the space coverage balance of the samples can be optimized.

Description

Technical Field

[0001] The present invention relates to the technical field of water resource monitoring. Specifically, it relates to a method for calculating the water storage capacity of ponds based on dynamic grid sampling. Background Art

[0002] Traditional water resource investigation methods, such as fixed-site observation, full-area census, or simple random sampling, have significant limitations when dealing with pond resource investigations. Ponds usually have characteristics such as highly uneven spatial distribution, large quantity (large in number and wide in area), and complex terrain environment. However, existing methods are difficult to effectively adapt to this spatial heterogeneity: fixed-site observation has limited coverage; full-area census is costly and unrealistic; simple random sampling completely ignores the spatial distribution characteristics of ponds, resulting in difficulty in covering key areas such as water depth changes, and the training data provided for subsequent remote sensing inversion models is biased, ultimately increasing the estimation error of water storage capacity.

[0003] Although grid sampling has been applied in forestry and agricultural surveys, it generally uses grids of fixed size. This rigid design lacks adaptability and cannot be dynamically adjusted according to the local density difference of ponds and the complexity of the terrain. It is easy to cause sample redundancy in sparse pond areas, increasing the cost of ineffective surveys; while in dense pond areas or areas with complex terrain, it may lead to sample omission and fail to fully represent key spatial variation information. Therefore, for calculating the water storage capacity of ponds in large areas, there is an urgent need to develop an adaptive sampling method that can dynamically adjust the grid size and sampling strategy according to spatial characteristics (such as pond density, terrain complexity) to overcome the inherent defects of fixed grids and pure random sampling and ensure the spatial representativeness and estimation accuracy of samples.

[0004] No effective solution has been proposed for the problems in the related art. Summary of the Invention

[0005] In view of the problems in the related art, the present invention proposes a method for calculating the water storage capacity of ponds based on dynamic grid sampling to overcome the above-mentioned technical problems existing in the existing related art.

[0006] To this end, the specific technical solution adopted by the present invention is as follows: A method for calculating the water storage capacity of ponds based on dynamic grid sampling, comprising the following steps: S1. Based on the pre-acquired optical image and the improved normalized difference water index, extract the pond water surface, calculate the pond area and the number of ponds in the task area; and use the grid sampling method to classify and sample the ponds to obtain the number of pond samples; S2. Based on the land area, pond area, and number of ponds in the task area, calculate the pond density and grid width, generate a dynamic grid, and use programming software to batch sample the ponds; S3. Optimize the quality of the sampled pond samples based on the pre-acquired urban territorial space monitoring road network data, and use big data analysis to evaluate and replace the quality of the pond samples; S4. Encode the optimized pond samples, and use geographic information software to construct a sample database; use programming software and machine learning algorithms to construct a water depth inversion model, obtain the water depth inversion results, and calculate the water storage capacity of the ponds in combination with the pond area.

[0007] Optionally, based on the pre-acquired optical images and the improved normalized water body index, extract the pond water surfaces, calculate the pond areas and the number of ponds in the task area; and use the grid sampling method to classify and sample the ponds. The steps for obtaining the pond sampling quantity include: S11. Based on the pre-acquired optical images and the improved normalized water body index, extract the water body boundaries of the pond water surfaces in the task area, and use the spatial relationship analysis method to calculate the pond areas; S12. Based on the sampling ratio in the classified ponds, use the grid sampling method to extract the largest sample in each grid, and calculate the pond sampling quantity.

[0008] Optionally, based on the pre-acquired optical images and the improved normalized water body index, extract the water body boundaries of the pond water surfaces in the task area, and use the spatial relationship analysis method to calculate the pond areas, including the following steps: S111. Based on the pre-acquired optical images, combine the improved normalized water body index to extract the water body boundaries of the pond water surfaces in the task area, and set the threshold of the improved normalized water body index; S112. After smoothing the extracted water surface boundaries to remove serrations, and masking and removing rivers, lakes, reservoirs and ditches, obtain the pond patches; S113. Divide a single pond patch into multiple patches, use the spatial relationship analysis method to merge the patches, correct the patch boundaries, and use geographic information software to calculate the pond areas.

[0009] Optionally, based on the territorial area, pond areas and the number of ponds in the task area, calculate the pond density and grid width, perform dynamic grid generation, and use programming software to batch sample the ponds, including the following steps: S21. Based on the pond areas and the number of ponds in the task area, initialize the grid width, and combine the terrain complexity and different pond distribution densities to perform dynamic grid generation; S22. According to the dynamically adjusted grids, use programming software to batch sample the ponds, and for the ponds that span multiple grids, use overlapping sampling.

[0010] Optionally, based on the pond area and the number of ponds in the task area, initialize the grid width, and combine the terrain complexity and different pond distribution densities to perform dynamic grid generation, including the following steps: S211. Calculate the pond density in the task area based on the pond area and the number of ponds in the task area to obtain the pond density; S212. Calculate the grid width according to the obtained pond density and in combination with the number of ponds; S213. Perform dynamic grid generation according to the terrain complexity and different pond distribution densities.

[0011] Optionally, performing dynamic grid generation according to the terrain complexity and different pond distribution densities includes the following steps: S2131. Divide the grid into a basic sampling grid and a fine sampling grid according to different sizes according to the pre-set grid size, where the size of the basic sampling grid is larger than that of the fine sampling grid; S2132. When the terrain in the task area is complex or the water area is dense, select the fine sampling grid; when the terrain in the task area is simple or the ponds are sparse, select the basic sampling grid.

[0012] Optionally, optimize the quality of the sampled pond samples according to the pre-obtained urban territorial space monitoring road network data, and use big data analysis to evaluate and replace the quality of the pond samples, including the following steps: S31. According to the pre-obtained urban territorial space monitoring road network data, use geographic information software to calculate the straight-line distance from the centroid of each pond patch to the nearest road, and perform accessibility analysis on the pond samples according to the pre-set distance threshold; S32. Calculate the shape index of the patch according to the patch perimeter and area, and replace the patches that exceed the pre-set long and narrow threshold according to the pre-set long and narrow threshold; S33. Based on the pond samples after distance and shape optimization, use big data analysis to calculate the multi-dimensional indicators of the pond samples, perform secondary optimization on the pond samples, and identify and replace the pond samples that do not meet the expected quality after evaluating the quality of the samples.

[0013] Optionally, according to the pre-obtained urban territorial space monitoring road network data, use geographic information software to calculate the straight-line distance from the centroid of each pond patch to the nearest road, and perform accessibility analysis on the pond samples according to the pre-set distance threshold, including the following steps: S311. When the straight-line distance from the centroid of the pond patch to the nearest road does not exceed the pre-set distance threshold, retain the pond sample; S312. When the straight-line distance from the centroid of the pond patch to the nearest road exceeds a pre-set distance threshold, mark the pond sample as an inaccessible sample, and replace it in the grid sample pool with the sample that has road access and the largest area.

[0014] Optionally, encode based on the optimized pond samples, and use geographic information software to construct a sample database; use programming software and machine learning algorithms to construct a water depth inversion model to obtain the water depth inversion result, and combine with the pond area to calculate the water storage capacity of the pond, including the following steps: S41. Encode based on the optimized pond samples according to the pre-set encoding rules to construct a sample database; S42. Based on the sample database, combine with optical images to construct a sample dataset, and use programming software to construct a water depth inversion model to obtain the average water depth of the pond, and combine with the pond area to calculate the water storage capacity of the pond.

[0015] Optionally, based on the sample database, combine with optical images to construct a sample dataset, and use programming software to construct a water depth inversion model to obtain the average water depth of the pond, and combine with the pond area to calculate the water storage capacity of the pond, including the following steps: S421. For the pond samples obtained by the grid sampling method, use the single-beam and water level gauge methods to measure the water depth actually, and use it as the sample data for remote sensing inversion of the pond water depth; S422. Preprocess the optical images, calculate the feature importance of the spectral reflectance, and use it as the sample data of the inversion factor for remote sensing inversion of the water depth; S423. According to the sample data of remote sensing inversion of the pond water depth and the sample data of the inversion factor for remote sensing inversion of the water depth, form a sample dataset, and divide it into a training set and a validation set. Use the root mean square error and the coefficient of determination as evaluation indicators to verify the inversion accuracy; S424. Use programming software to perform mask extraction on the inversion result to obtain an inversion spatial distribution map that only contains the pond area, and calculate the average water depth; S425. According to the obtained average water depth of the pond, combine with the water surface area of the pond to calculate the water storage capacity of the pond.

[0016] The beneficial effects of the present invention are as follows: The present invention provides a dynamically adjustable grid sampling method. This sampling method takes into account the spatial distribution of ponds, the characteristics of area and shape, and terrain differences, which can optimize the spatial coverage balance of samples, such as automatically adapting to sample gaps in high / low density areas, and is superior to the traditional fixed grid random sampling method; in addition, embedding big data evaluation into the dynamic grid sampling process can optimize the sample quality, avoid model bias caused by turbid pond samples, and can improve the accuracy of remote sensing water depth inversion and the accuracy of water storage capacity. Description of the Drawings

[0017] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the accompanying drawings required for use in the embodiments. Obviously, the accompanying drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other accompanying drawings can be obtained based on these drawings.

[0018] Figure 1 is the overall technical roadmap of a method for measuring the water storage capacity of pit ponds based on dynamic grid sampling according to an embodiment of the present invention; Figure 2 is the dynamic grid sampling flowchart of a method for measuring the water storage capacity of pit ponds based on dynamic grid sampling according to an embodiment of the present invention; Figure 3 is the sample optimization flowchart of a method for measuring the water storage capacity of pit ponds based on dynamic grid sampling according to an embodiment of the present invention; Figure 4 is the schematic diagram of dynamic grid generation of a method for measuring the water storage capacity of pit ponds based on dynamic grid sampling according to an embodiment of the present invention; Figure 5 is the spectral curve comparison diagram of a method for measuring the water storage capacity of pit ponds based on dynamic grid sampling according to an embodiment of the present invention; Figure 6 is the spectral cosine similarity comparison diagram of a method for measuring the water storage capacity of pit ponds based on dynamic grid sampling according to an embodiment of the present invention; Figure 7 is the comparison diagram before and after sample optimization of a method for measuring the water storage capacity of pit ponds based on dynamic grid sampling according to an embodiment of the present invention; Figure 8 is the remote sensing water depth inversion spatial distribution diagram of a method for measuring the water storage capacity of pit ponds based on dynamic grid sampling according to an embodiment of the present invention; Figure 9 is the verification diagram of the remote sensing water depth inversion result of a method for measuring the water storage capacity of pit ponds based on dynamic grid sampling according to an embodiment of the present invention. Detailed implementation manners

[0019] To further illustrate the embodiments, the present invention provides accompanying drawings. These accompanying drawings are part of the disclosure of the present invention. They are mainly used to illustrate the embodiments and can be used in conjunction with the relevant descriptions in the specification to explain the operation principle of the embodiments. With reference to these contents, those of ordinary skill in the art should be able to understand other possible implementation manners and the advantages of the present invention. The components in the figures are not drawn to scale, and similar component symbols are usually used to represent similar components.

[0020] According to an embodiment of the present invention, a method for calculating the water storage capacity of ponds based on dynamic grid sampling is provided.

[0021] The present invention will be further described below in conjunction with the accompanying drawings and specific embodiments. As Figure 1 shown, according to an embodiment of the present invention, a method for calculating the water storage capacity of ponds based on dynamic grid sampling is provided. The method for calculating the water storage capacity of ponds includes the following steps: S1. Based on the pre-acquired optical image and the improved normalized difference water index, extract the pond water surface, calculate the pond area and the number of ponds in the task area; and use the grid sampling method to classify and sample the ponds to obtain the pond sampling quantity.

[0022] In one embodiment, based on the pre-acquired optical image and the improved normalized difference water index, extract the pond water surface, calculate the pond area and the number of ponds in the task area; and use the grid sampling method to classify and sample the ponds to obtain the pond sampling quantity, which includes the following steps: S11. Based on the pre-acquired optical image and the improved normalized difference water index, extract the water body boundary of the pond water surface in the task area, and use the spatial relationship analysis method to calculate the pond area.

[0023] In one embodiment, based on the pre-acquired optical image and the improved normalized difference water index, extract the water body boundary of the pond water surface in the task area, and use the spatial relationship analysis method to calculate the pond area, which includes the following steps: S111. Based on the pre-acquired optical image, combine the improved normalized difference water index to extract the water body boundary of the pond water surface in the task area, and set the threshold of the improved normalized difference water index; S112. After smoothing the extracted water surface boundary to remove the serrations and masking to remove rivers, lakes, reservoirs and ditches, obtain the pond patches; S113. Divide a single pond patch into multiple patches, use the spatial relationship analysis method to merge the patches, correct the patch boundary, and calculate the pond area using geographic information software.

[0024] S12. Based on the sampling ratio in the classified ponds, use the grid sampling method to extract the sample with the largest area in each grid, and calculate the pond sampling quantity.

[0025] It should be explained that for identifying ponds based on the MNDWI index, first, for pond extraction, the optical image (such as Sentinel-2) and the threshold method of the corrected normalized difference water index (MNDWI) are used to preliminarily extract the water body boundary. Based on experience and experiments, the MNDWI threshold is set to 0.3, and binary segmentation is performed to preliminarily extract the area where MNDWI is greater than 0.3 as the water body.

[0026] ; In the formula, MNDWI represents the corrected normalized difference water index, represents the green band, represents the short-wave infrared band.

[0027] Among them, the extracted water body boundary needs to be smoothed. Using the smooth surface tool of geographic information software, select the PAEK algorithm, set the tolerance to 20 meters, and smooth the water surface boundary to remove the jagged shape; then, combined with the data of land use change survey results, mask and remove the data of rivers, lakes, reservoirs, ditches, etc., and extract the pond patches.

[0028] Then, patch merging. For the problem of pond patch fragmentation, such as a single pond being divided into multiple patches, use the spatial relationship analysis method to merge the fragmented patches and correct the patch boundaries. The specific method is to use geographic information software for buffer zone fusion, generate buffer zones for adjacent patches, set the threshold to 1 pixel width, and merge the patches if the buffer zones overlap, systematically solve the problem of patch fragmentation, and ensure that the patch boundaries are consistent with the real ground objects; finally, area calculation, use the geometric calculation tool of geographic information software to automatically calculate the pond area.

[0029] The classification sampling design is as follows: 1) For aquaculture ponds, use the grid sampling method to extract the sample with the largest area in each grid, a total of pieces, calculate the sampling quantity of aquaculture ponds, and its calculation formula is as follows: ; In the formula, represents the total number of aquaculture ponds, k represents the sampling ratio, represents the sampling quantity of the extracted aquaculture ponds.

[0030] 2) For natural ponds, use the grid sampling method to extract the sample with the largest area in each grid, a total of pieces, calculate the sampling quantity of natural ponds, and its calculation formula is as follows: ; In the formula, represents the total number of natural ponds, k represents the sampling ratio, represents the sampling quantity of the extracted natural ponds.

[0031] S2. Based on the land area, pond area and pond quantity in the task area, calculate the pond density and grid width, generate dynamic grids, and use programming software to conduct batch sampling of ponds.

[0032] In one embodiment, based on the land area, pond area, and number of ponds in the task area, calculate the pond density and grid width, perform dynamic grid generation, and batch sample the ponds using programming software, including the following steps: S21. Initialize the grid width based on the pond area and number of ponds in the task area, and perform dynamic grid generation in combination with the terrain complexity and different pond distribution densities.

[0033] In one embodiment, initializing the grid width based on the pond area and number of ponds in the task area and performing dynamic grid generation in combination with the terrain complexity and different pond distribution densities includes the following steps: S211. Calculate the pond density in the task area based on the pond area and number of ponds in the task area to obtain the pond density. S212. Calculate the grid width according to the obtained pond density in combination with the number of ponds. S213. Perform dynamic grid generation according to the terrain complexity and different pond distribution densities.

[0034] In one embodiment, performing dynamic grid generation according to the terrain complexity and different pond distribution densities includes the following steps: S2131. Divide the grid into basic sampling grids and fine sampling grids according to different sizes according to the pre-set grid size, where the size of the basic sampling grid is larger than that of the fine sampling grid. S2132. When the terrain in the task area is complex or the water area is dense, select the fine sampling grid; when the terrain in the task area is simple or the ponds are sparse, select the basic sampling grid.

[0035] S22. According to the dynamically adjusted grid, batch sample the ponds using programming software, and for ponds spanning multiple grids, use overlapping sampling.

[0036] It should be noted that for dynamic adjustable grid generation, extract the land area of each district and county according to the land use change survey data, with the unit of square kilometers; then initialize the grid width and calculate the pond density; evaluate the pond density in the task area to obtain the number of ponds within each square kilometer. The calculation formula is as follows: ; In the formula, represents the pond density in the working area, represents the total number of ponds in the working area, represents the land area in the working area.

[0037] Determine the grid width according to the pond density, and calculate the grid width. The calculation method is as follows: ; In the formula, represents the width of the regular grid, represents the number of samples within the grid, represents the sampling ratio.

[0038] Among them, The unit of is square kilometers, The unit of

[0039] In addition, considering the terrain complexity of different regions and the distribution density of different ponds, in areas with complex terrain or dense water areas, smaller grids can be selected to more accurately capture water depth changes; while in areas with relatively simple terrain or sparse ponds, larger grids can be selected, as Figure 4 shown.

[0040] According to the latest land use change survey data, the calculation results of the grid width parameters are given. Taking the ponds in a certain southern province as an example, the expression of the classification result of the pond density in this province is as follows: ; In the formula, represents the pond density.

[0041] Use a Python program for batch sampling in large areas (more than 2 counties). When sampling, remove the ponds that cross the county administrative boundaries and keep the sampling quantity unchanged. Input the pond boundary vector data and pond type labels as the data to be sampled. Among them, the process of batch sampling using a Python program includes the following steps: 1) Calculate the sampling quantity of ponds in each county according to the determined sampling ratio.

[0042] 2) Calculate the grid width of each county and generate grids with different widths.

[0043] 3) Eliminate the empty grids, sort them in descending order according to the number of ponds within the grids, and give priority to sampling from the high-density grids to ensure coverage of the hot spots; secondly, sort the ponds in descending order according to the area of ponds within the grids; then, loop through the grids and sequentially extract the largest pond from each grid until the sampling quantity requirement is met.

[0044] 4) Termination condition: If the samples are insufficient after traversal, expand the grid width and resample.

[0045] 5) For the ponds that span multiple grids, use overlapping sampling to avoid omission; judge whether the patch x intersects with the grid. When the pond patch intersects with both grids A and B, classify the patch into the sampling base of grid A. At this time, it is determined that the patch x is already in the sampling base and will not be repeatedly counted into the sampling base of grid B.

[0046] S3. Optimize the quality of the sampled pond samples according to the pre-acquired urban territorial space monitoring road network data, and use big data analysis to evaluate and replace the quality of the pond samples.

[0047] In one embodiment, optimizing the quality of the sampled pond samples according to the pre-acquired urban territorial space monitoring road network data, and using big data analysis to evaluate and replace the quality of the pond samples includes the following steps: S31. According to the pre-acquired urban territorial space monitoring road network data, use geographic information software to calculate the straight-line distance from the centroid of each pond patch to the nearest road, and perform accessibility analysis on the pond samples based on a pre-set distance threshold.

[0048] In one embodiment, according to the pre-acquired urban territorial space monitoring road network data, use geographic information software to calculate the straight-line distance from the centroid of each pond patch to the nearest road, and perform accessibility analysis on the pond samples based on a pre-set distance threshold includes the following steps: S311. When the straight-line distance from the centroid of the pond patch to the nearest road does not exceed the pre-set distance threshold, retain the pond sample; S312. When the straight-line distance from the centroid of the pond patch to the nearest road exceeds the pre-set distance threshold, mark the pond sample as an inaccessible sample, and replace it with the sample with the largest accessible area and the largest area in the grid sample pool.

[0049] S32. Calculate the shape index of the patch according to the patch perimeter and area, and replace the patches that exceed the pre-set narrow and long threshold.

[0050] S33. Based on the pond samples optimized in terms of distance and shape, use big data analysis to calculate the multi-dimensional indicators of the pond samples, perform secondary optimization on the pond samples, and after evaluating the quality of the samples, identify and replace the pond samples that do not meet the expected quality.

[0051] It should be noted that by collecting urban territorial space monitoring road network data, performing accessibility analysis, and replacing ponds more than 1000 meters away from the nearest road, the specific method of accessibility analysis is as follows: input the pond boundary data and road network vector data, use the proximity analysis tool of geographic information software to calculate the straight-line distance from the centroid of each pond patch to the nearest road, set the threshold to 1000 meters, when the distance is less than 1000 meters, retain the sample; otherwise, mark it as an inaccessible sample and replace it with the sample with the largest accessible area and the largest area in the grid sample pool; avoid narrow and long samples, calculate the patch shape index, such as the ratio of perimeter to area, that is, the narrow and long ratio, set the threshold to 0.5, and samples exceeding this threshold are also regarded as narrow and long samples and replaced.

[0052] In addition, on the basis of optimizing parameters such as distance and shape in the previous step, further utilize big data analysis to calculate multi-dimensional indicators (water quality, historical changes, surrounding environment) for secondary optimization of the samples, evaluate the representativeness and data quality of the samples, identify and replace low-quality samples. Figure 3 The following is the technical roadmap for sample optimization and its implementation steps: 1) Extract multi-dimensional features and calculate the following three types of features (NDTI variance, factory density, spectral similarity) for the candidate samples.

[0053] For NDTI variance, obtain images from more than 2 periods, perform preprocessing such as radiometric calibration, atmospheric correction, and geometric registration, calculate the NDTI variance of the samples, set the threshold to 0.1, and exclude samples with unstable water quality whose NDTI variance is higher than the threshold; NDTI characterizes the turbidity of water bodies through the reflectance ratio of the red and green bands as follows: ; Calculate the average NDTI and NDTI variance of the pond areas in the images as follows: ; ; In the formula, represents the degree of fluctuation that can reflect the turbidity of the water body NDTI variance, represents the t th period of the image of a certain pond NDTI mean value, represents the average value of multiple periods of images of, represents the reflectance of the red band, represents the reflectance of the green band, represents the i th red band reflectance of a pixel, represents the i th green band reflectance of a pixel.

[0054] Among them, t = 1, 2,... T , T is the number of image periods, i = 1, 2,... N , N represents the total number of pixels of a certain pond.

[0055] In addition, for factory density, factory data with quantity and location are collected, and the factory density within the 2000m buffer of the sample is calculated. The threshold is set to 3 factories per square kilometer, and samples with factory density greater than the threshold are replaced to avoid interference from surrounding pollution sources; for spectral similarity, the spectral cosine similarity with the spectral characteristics of the sample is calculated to ensure the consistency of model input; the similarity value range is [-1, 1], where 1 indicates completely identical, 0 indicates irrelevant, and -1 indicates completely opposite; candidate samples with similarity > 0.95 are retained to ensure the adaptability of water depth inversion, and its expression is as follows: ; In the formula, represents the cosine similarity, represents the reflectance value of the candidate sample in the i th band, represents the reflectance value of the high-quality sample in the i th band, represents the number of bands involved in the calculation.

[0056] 2) Decision on replacement of low-quality samples Input candidate samples, calculate three parameters: NDVI variance, factory density, and spectral similarity. Determine the sample quality according to the set threshold, and mark the samples as binary categories, where 1 represents "high-quality samples" and 0 represents "low-quality samples"; for samples marked as 0, select eligible alternative samples from the candidate pool (NDTI variance less than 0.1; spectral similarity greater than 0.95; factory density ≤ 3 factories / km 2 ); for samples with missing features, adjust the samples through manual review.

[0057] S4. Encode the optimized pond samples, and use geographic information software to construct a sample database; use programming software and machine learning algorithms to construct a water depth inversion model, obtain the water depth inversion result, and combine with the pond area to calculate the water storage capacity of the pond.

[0058] In one embodiment, encoding the optimized pond samples, using geographic information software to construct a sample database; using programming software and machine learning algorithms to construct a water depth inversion model, obtaining the water depth inversion result, and combining with the pond area to calculate the water storage capacity of the pond includes the following steps: S41. Based on the optimized pond samples, encode them according to the pre-set encoding rules to construct a sample database.

[0059] S42. Based on the sample database, combine with optical images to construct a sample dataset, and use programming software to construct a water depth inversion model to obtain the average water depth of the pond, and combine with the pond area to calculate the water storage capacity of the pond.

[0060] In one embodiment, based on the sample database, a sample dataset is constructed by combining optical images, and a water depth inversion model is constructed using programming software to obtain the average water depth of the pond. Combining the pond area, calculating the water storage capacity of the pond includes the following steps: S421. For the pond samples obtained by the grid sampling method, the water depth is actually measured using a single-beam and water level gauge method, and used as the sample data for remote sensing inversion of the pond water depth; S422. Preprocess the optical image, calculate the feature importance of the spectral reflectance, and use it as the sample data of the inversion factor for remote sensing inversion of the water depth; S423. According to the sample data for remote sensing inversion of the pond water depth and the sample data of the inversion factor for remote sensing inversion of the water depth, a sample dataset is formed and divided into a training set and a validation set. The root mean square error and the coefficient of determination are used as evaluation indicators to verify the inversion accuracy; S424. Use programming software to perform mask extraction on the inversion result, obtain an inversion spatial distribution map containing only the pond area, and calculate the average water depth; S425. According to the obtained average water depth of the pond, combine it with the water surface area of the pond to calculate the water storage capacity of the pond.

[0061] It should be explained that for the standardization processing of the sample data, first, unify the data spatial reference benchmark and the pond ID to facilitate large-area data integration and identification of field measurement patches; then retain the land use patch number and land use patch name, add the county administrative region name and remarks field, add the pond coding field, and code the sample patches according to certain rules, ensuring the uniqueness of the coding. For example: Pond coding rule: XZQBMMSYLXNNNNNN; Among them, XZQBMM represents the 6-digit county administrative region code, SYLX represents the 4-digit water area type (1104 for ponds), and NNNNNN represents the 6-digit pond serial number.

[0062] Finally, batch create folders for the names of cities and counties, export vector files to form a sample database; generate a statistical table and calculate the number of each county and sample patches.

[0063] In the construction of the water depth inversion model, first, for the pond samples obtained by the grid sampling method, conduct field water depth measurement using methods such as single-beam and water level gauge as the sample data for remote sensing inversion of the pond water depth; then collect Sentinel-2 images and perform preprocessing such as radiometric calibration, atmospheric correction, and orthorectification, and calculate the feature importance of the spectral reflectance as the sample data of the inversion factor for remote sensing inversion of the water depth.

[0064] In addition, for radiometric calibration, the calibration coefficients of each band corresponding to the image at the corresponding time are correctly obtained, and the original pixel brightness value recorded in the image is converted into the actual radiance value through the following formula; it can be realized by using the "RadiometricCalibration" tool of geographic information software, and the expression is as follows: ; In the formula, represents the radiance, represents the pixel brightness value of the remote sensing image, represents the increment of the radiometric calibration coefficient, represents the offset of the radiometric calibration coefficient.

[0065] For atmospheric correction, using the spectral response function of the image, information such as the orbital altitude, observation angle, imaging time, solar azimuth angle at the time of imaging, aerosol type, and geographical location of the imaging area, and adopting a model-based atmospheric correction method, selecting a suitable radiative transfer model (6S, MODTRAN, LOWTRAN, ATCOR, FLAASH, etc.) or using a lookup table (LUT), accurately removing the atmospheric influence, completing the correction of the image data, and realizing the conversion from radiance to the actual surface reflectance; specifically, it can be realized by using the "FLAASH Atmospheric Correction" tool of geographic information software.

[0066] For orthorectification, when orthorectifying, clearly, clearly, and uniformly select ground control points on the image that do not change significantly over time; resample the image in an appropriate manner (nearest neighbor method, bilinear interpolation method, cubic convolution interpolation method, etc.) to obtain the corrected image; specifically, it can be realized by using the "RPC Orthorectification Workflow" tool of geographic information software.

[0067] For the feature importance of inversion factors, extract the sample reflectance of the preprocessed image according to the spatial position of the measured water depth sample points, calculate the feature importance of inversion factors (ratio, logarithmic ratio, etc.), and calculate the feature importance of each inversion factor. The higher the feature importance value, the higher the correlation between the factor and the water depth. The expression is as follows: ; ; In the formula, represents the ratio inversion factor, represents the logarithmic ratio inversion factor, represents the reflectance in the blue band, represents the reflectance in the green band.

[0068] The inversion factors and measured water depth data form a sample data set, which is divided into a training set and a validation set according to 8:2. Training models such as BP neural network, random forest, and support vector machine are selected, and the root mean square error (RMSE) and coefficient of determination (R²) are used as evaluation indicators to verify the inversion accuracy. Their expressions are as follows: ; ; In the formula, represents the i th true value, represents the i th estimated value, represents the average value of the true values, represents the total number of true values.

[0069] Using the mask extraction tool of geographic information software and the pond vector polygons, the inversion results of the entire image are masked and extracted to obtain an inversion spatial distribution map that only contains the pond area, and the average water depth is calculated. Its expression is as follows: ; In the formula, represents the average water depth value of the i th pond, represents the measured water depth value of the i th pond at the u th measurement point.

[0070] The water storage capacity of the pond is calculated by multiplying the water surface area by the average water depth. Its expression is as follows: ; In the formula, represents the total water storage capacity of the area, represents the i th pond area.

[0071] The specific steps of the implementation are as follows: 1) Data preparation: Collect GF1 images with a resolution of 2 meters and Sentinel-2 image data with a resolution of 10 meters that are consistent with the date of the measured data, as well as the data of land use change surveys; the water depth values of the measured sample ponds.

[0072] 2) MNDWI water body extraction: Set the threshold to 0.3, use the GF1 image and the index method to extract the water body boundary, and combine the data of land use change surveys to mask and remove data such as rivers, lakes, and reservoirs to extract the pond boundary.

[0073] 3) Dynamic grid generation: Select a high-density area (5 km × 5 km) and a low-density area (6 km × 6 km) as the test areas in the task area. The pond density in the high-density area is 40 per km2 , the grid width is calculated to be 0.83 km and rounded up to 1 km; the pond density in the low-density area is 2 per km 2 , the grid width is calculated to be 4 km.

[0074] Among them, Figure 4 The division effects of the traditional fixed 2-km grid and the dynamic grid are compared. It can be found that the dynamic grid can better take into account the characteristics of water body distribution, and the fixed grid is suitable for single-district sampling; the dynamic grid is suitable for large-area sampling, which can capture local changes according to the characteristics of water body distribution and ensure the spatial uniform coverage of samples. Among them, Figure 4 (a) is the fixed grid in the low-density area with a = 2 km; Figure 4 (b) is the fixed grid in the high-density area with a = 2 km; Figure 4 (c) is the dynamic grid in the low-density area with a = 4 km; Figure 4 (d) is the dynamic grid in the high-density area with a = 1 km.

[0075] 4) Sample extraction and optimization. Based on the traditional fixed-grid sampling method, a pond is randomly selected as a sample in each grid. A total of 25 samples are extracted in the high-density area, and a total of 29 samples are extracted in the low-density area; based on the dynamic-grid sampling method, the pond with the largest area is extracted as a sample in each grid (including ponds across grids), and then optimization processes such as accessibility analysis, water quality stability analysis, and spectral adaptability analysis are carried out. Finally, a total of 28 samples are replaced in the high-density area, and a total of 107 samples are extracted; 3 samples are replaced in the low-density area, and a total of 11 samples are extracted.

[0076] Among them, Figure 5 and Figure 6 show the spectral curve graphs and spectral cosine similarities of different samples. It is found that the cosine similarity between low-quality samples and high-quality samples caused by the water body itself is poor. Through this index, pond samples can be screened and optimized; Figure 7 show the sample comparison graphs of the fixed grid and the dynamic grid. It can be found that the samples optimized based on the dynamic grid and big data are more typical and representative, avoiding samples with small areas, narrow shapes, poor water quality, land use changes, and traffic difficulties. The circles mark the key samples missed by the traditional method, and the squares highlight the sample hot spots newly added by the dynamic method. Among them, Figure 7 (a) is the random sampling of the fixed grid in the low-density area; Figure 7 (b) is the random sampling of the fixed grid in the high-density area; Figure 7 (c) is the optimized sampling of the dynamic grid in the low-density area; Figure 7 (d) is the optimized sampling of the dynamic grid in the high-density area.

[0077] 5) Water depth inversion and result verification. Taking the high-density area as an example, the measured water depth data and Sentinel-2 spectral data of the samples are obtained, and a water depth inversion model is constructed using the random forest algorithm.

[0078] Among them, Figure 8 The inversion results based on the traditional fixed grid and dynamic grid are shown. Figure 8 (a) is the inversion result based on the samples sampled from the fixed grid. Figure 8 (b) is the inversion result based on the samples sampled from the dynamic grid. Figure 9 The scatter plot for verifying the water depth inversion result is shown. Among them, Figure 9 (a) is the scatter plot for verifying the water depth inversion result in the "fixed grid + random sampling" mode. Figure 9 (b) is the scatter plot for verifying the water depth inversion result in the "dynamic grid + optimized sampling" mode.

[0079] The results show that in the "fixed grid + random sampling" mode, there are problems such as fewer sample data and poor sample quality, and the inversion effect is poor, with RMSE = 0.42 and R 2 = 0.37; in the "dynamic grid + optimized sampling" mode, the sample data is increased, the sample quality is optimized, the inversion effect is better, with RMSE = 0.22 and R 2 = 0.56. Compared with the traditional method, the correlation is increased by 51% and the accuracy is increased by 46%.

[0080] The above are only the preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for calculating the storage volume of pond water based on dynamic grid sampling, characterized in that, The method includes the following steps: S1. Based on the pre-acquired optical images and the improved normalized difference water index, extract the pond water surface, calculate the pond area and the number of ponds in the task area; and use the grid sampling method to conduct classified sampling on the ponds to obtain the pond sampling quantity; S2. Based on the national land area, pond area and the number of ponds in the task area, calculate the pond density and grid width, generate dynamic grids, and use programming software to conduct batch sampling on the ponds; S3. Optimize the quality of the sampled pond samples according to the pre-acquired urban national land space monitoring road network data, and use big data analysis to evaluate and replace the quality of the pond samples; S4. Encode based on the optimized pond samples, use geographic information software to construct a sample database; use programming software and machine learning algorithms to construct a water depth inversion model, obtain the water depth inversion result, and combine with the pond area to calculate the water storage capacity of the ponds.

2. The method for calculating the storage volume of pond water based on dynamic grid sampling according to claim 1, wherein Based on the pre-acquired optical images and the improved normalized difference water index, extract the pond water surface, calculate the pond area and the number of ponds in the task area; And using the grid sampling method to conduct classified sampling on the ponds to obtain the pond sampling quantity includes the following steps: S11. Based on the pre-acquired optical images and the improved normalized difference water index, extract the water body boundary of the pond water surface in the task area, and use the spatial relationship analysis method to calculate the pond area; S12. Based on the sampling ratio in the classified ponds, use the grid sampling method to extract the sample with the largest area in each grid, and calculate the pond sampling quantity.

3. A method for calculating the storage capacity of pond water based on dynamic grid sampling according to claim 2, characterized in that, Based on the pre-acquired optical images and the improved normalized difference water index, extract the water body boundary of the pond water surface in the task area, and use the spatial relationship analysis method to calculate the pond area includes the following steps: S111. Based on the pre-acquired optical images, combine with the improved normalized difference water index to extract the water body boundary of the pond water surface in the task area, and set the threshold of the improved normalized difference water index; S112. After smoothing the extracted water surface boundary to remove the serrations, and masking to remove rivers, lakes, reservoirs and ditches, obtain the pond patches; S113. Divide a single pond patch into multiple patches, use the spatial relationship analysis method to merge the patches, correct the patch boundaries, and use geographic information software to calculate the pond area.

4. A method for calculating the storage volume of pond water based on dynamic grid sampling according to claim 1, characterized in that, Based on the national land area, pond area and the number of ponds in the task area, calculate the pond density and grid width, generate dynamic grids, and use programming software to conduct batch sampling on the ponds includes the following steps: S21. Based on the pond area and the number of ponds in the task area, initialize the grid width, and combine with the terrain complexity and different pond distribution densities to generate dynamic grids; S22. According to the dynamically adjusted grids, use programming software to conduct batch sampling on the ponds, and for the ponds that span multiple grids, use overlapping sampling.

5. A method for calculating the storage capacity of pond water based on dynamic grid sampling according to claim 4, characterized in that, Based on the pond area and the number of ponds in the task area, initialize the grid width, and combine with the terrain complexity and different pond distribution densities to generate dynamic grids includes the following steps: S211. Calculate the pond density in the task area based on the pond area and the number of ponds in the task area to obtain the pond density; S212. Calculate the grid width according to the obtained pond density and in combination with the number of ponds; S213. Generate dynamic grids according to the terrain complexity and the distribution density of different ponds.

6. The method for calculating the storage capacity of pond water based on dynamic grid sampling according to claim 5, characterized in that, The generating dynamic grids according to the terrain complexity and the distribution density of different ponds includes the following steps: S2131. Divide the grids into basic sampling grids and fine sampling grids according to different sizes according to the pre-set grid size, where the size of the basic sampling grid is larger than that of the fine sampling grid; S2132. When the terrain in the task area is complex or the water area is dense, select the fine sampling grid; when the terrain in the task area is simple or the ponds are sparse, select the basic sampling grid.

7. A method for calculating the storage volume of pond water based on dynamic grid sampling according to claim 1, characterized in that The optimizing the quality of the sampled pond samples according to the pre-obtained urban territorial space monitoring road network data, evaluating the quality of the pond samples by using big data analysis and replacing them includes the following steps: S31. According to the pre-obtained urban territorial space monitoring road network data, use geographic information software to calculate the straight-line distance from the centroid of each pond patch to the nearest road, and conduct accessibility analysis on the pond samples according to the pre-set distance threshold; S32. Calculate the shape index of the patch according to the patch perimeter and area, and replace the patches that exceed the pre-set long and narrow threshold according to the pre-set long and narrow threshold; S33. Based on the pond samples after distance and shape optimization, use big data analysis to calculate multi-dimensional indicators of the pond samples, conduct secondary optimization on the pond samples, and identify and replace the pond samples that do not meet the expected quality after evaluating the quality of the samples.

8. A method for calculating the storage capacity of pond water based on dynamic grid sampling according to claim 7, characterized in that, The calculating the straight-line distance from the centroid of each pond patch to the nearest road according to the pre-obtained urban territorial space monitoring road network data, using geographic information software, and conducting accessibility analysis on the pond samples according to the pre-set distance threshold includes the following steps: S311. When the straight-line distance from the centroid of the pond patch to the nearest road does not exceed the pre-set distance threshold, retain the pond sample; S312. When the straight-line distance from the centroid of the pond patch to the nearest road exceeds the pre-set distance threshold, mark the pond sample as an inaccessible sample and replace it with the sample with the largest accessible area and the largest area from the sample pool within the grid.

9. The method for calculating the storage volume of pond water based on dynamic grid sampling according to claim 1, wherein, The encoding based on the optimized pond samples, constructing a sample database by using geographic information software; constructing a water depth inversion model by using programming software and machine learning algorithms, obtaining the water depth inversion result, and calculating the water storage capacity of the pond in combination with the pond area includes the following steps: S41. Encode the optimized pond samples according to the pre-set encoding rules to construct a sample database; S42. Based on the sample database, construct a sample dataset in combination with optical images, and use programming software to construct a water depth inversion model to obtain the average water depth of the pond, and calculate the water storage capacity of the pond in combination with the pond area.

10. The method for calculating the storage capacity of pond water based on dynamic grid sampling according to claim 1, wherein Based on the sample database, a sample dataset is constructed by combining optical images, and a water depth inversion model is constructed using programming software to obtain the average water depth of the pond. Combining with the pond area, calculating the water storage capacity of the pond includes the following steps: S421. For the pond samples obtained by the grid sampling method, the water depth is measured actually using the single-beam and water level gauge methods, and used as the sample data for remote sensing inversion of the pond water depth; S422. Preprocess the optical image, calculate the feature importance of the spectral reflectance, and use it as the sample data of the inversion factor for remote sensing inversion of the water depth; S423. A sample dataset is composed of the sample data for remote sensing inversion of the pond water depth and the sample data of the inversion factor for remote sensing inversion of the water depth, and is divided into a training set and a validation set. The root mean square error and the coefficient of determination are used as evaluation indicators to verify the inversion accuracy; S424. Use programming software to perform mask extraction on the inversion result, obtain an inversion spatial distribution map containing only the pond area, and calculate the average water depth; S425. According to the obtained average water depth of the pond, calculate the water storage capacity of the pond in combination with the water surface area of the pond.

Images