A method for calculating pond water storage capacity based on dynamic grid sampling

Abstract

The present invention discloses a method for calculating pond water storage capacity based on dynamic grid sampling, which relates to the field of water resource monitoring technology. The method comprises: extracting pond water surface based on optical images and an improved normalized water index, calculating the pond area and number within a mission area; and using a grid sampling method to classify and sample the ponds; calculating the pond density and grid width based on the land area, pond area, and number within the mission area, dynamically generating a grid, and batch sampling the ponds using programming software; using big data analysis based on urban land space monitoring road network data, evaluating the quality of pond samples and replacing them, encoding the pond samples, and constructing a sample database using geographic information software; and constructing a water depth inversion model using programming software and a machine learning algorithm to obtain water depth inversion results. The water storage capacity of the ponds is calculated based on the pond area. The present invention can optimize the spatial coverage balance of samples.

Description

Technical Field

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

[0002] Traditional water resource survey methods, such as fixed-site observations, full-scale surveys, or simple random sampling, have significant limitations when it comes to investigating pond resources. Ponds are often characterized by highly uneven spatial distribution, large numbers (large in number and across a wide area), and complex topography. However, existing methods struggle to effectively adapt to this spatial heterogeneity: fixed-site observations have limited coverage; full-scale surveys are costly and impractical; and simple random sampling completely ignores the spatial distribution of ponds, making it difficult to cover key areas such as water depth variations. This leads to biased training data for subsequent remote sensing inversion models, ultimately increasing errors in water storage estimates.

[0003] Although grid sampling has been used in forestry and agricultural surveys, it generally uses a fixed-size grid. This rigid design lacks adaptability and cannot dynamically adjust to local differences in pond density and terrain complexity. In areas with sparse ponds, this can lead to sample redundancy, increasing ineffective survey costs; in areas with dense ponds or complex terrain, it can lead to sample omissions and fail to fully represent key spatial variation. Therefore, for large-scale pond water storage estimation, there is an urgent need to develop an adaptive sampling method that can dynamically adjust grid size and sampling strategies based on spatial characteristics (such as pond density and terrain complexity). This method can overcome the inherent shortcomings of fixed grids and purely random sampling, ensuring the spatial representativeness of samples and the accuracy of estimates.

[0004] Currently, no effective solutions have been proposed for the problems in related technologies. Summary of the Invention

[0005] In response to the problems in the related art, the present invention proposes a method for calculating pond water storage capacity 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 solutions adopted in the present invention are as follows:

[0007] A method for calculating pond water storage capacity based on dynamic grid sampling includes the following steps:

[0008] S1. Based on pre-acquired optical images and the improved normalized water index, extract the pond water surface and calculate the area and number of ponds within the mission area. Then, use the grid sampling method to classify and sample the ponds to obtain the number of pond samples.

[0009] S2. Based on the land area, pond area, and number of ponds within the mission area, calculate the pond density and grid width, dynamically generate the grid, and use programming software to conduct batch sampling of ponds;

[0010] S3. Based on the pre-acquired urban land space monitoring road network data, optimize the quality of the sampled pit and pond samples, and use big data analysis to evaluate the quality of the pit and pond samples and replace them;

[0011] S4. Encode the optimized pond samples and build a sample database using geographic information software. Use programming software and machine learning algorithms to build a water depth inversion model to obtain water depth inversion results. Combined with the pond area, calculate the water storage capacity of the pond.

[0012] Optionally, based on the pre-acquired optical image and the improved normalized water index, the pond water surface is extracted, and the area and number of ponds in the task area are calculated; and the ponds are classified and sampled using a grid sampling method. Obtaining the number of pond samples includes the following steps:

[0013] S11. Based on the pre-acquired optical imagery and the improved normalized water index, extract the water boundaries of the ponds within the mission area and calculate the pond areas using spatial relationship analysis.

[0014] S12. Based on the sampling proportions in the classified pits and ponds, use the grid sampling method to extract the sample with the largest area in each grid and calculate the number of pit and pond samples.

[0015] Optionally, based on the pre-acquired optical image and the improved normalized water index, extracting the water boundary of the pond surface in the task area and calculating the pond area using a spatial relationship analysis method includes the following steps:

[0016] S111. Based on the pre-acquired optical image and the improved normalized water index, extract the water boundary of the pond water surface in the task area, and set the improved normalized water index threshold;

[0017] S112, smoothing the extracted water surface boundary to remove jagged edges, and masking out rivers, lakes, reservoirs, and ditches to obtain a pond patch;

[0018] S113. Split a single pond into multiple plots, merge the plots using spatial relationship analysis, correct the plot boundaries, and calculate the pond area using geographic information software.

[0019] Optionally, based on the land area, pond area, and number of ponds in the mission area, the pond density and grid width are calculated, a dynamic grid is generated, and batch sampling of ponds is performed using programming software, including the following steps:

[0020] S21. Initialize the grid width based on the area and number of ponds in the mission area, and dynamically generate the grid based on the terrain complexity and the density of different ponds.

[0021] S22. Based on the dynamically adjusted grid, use programming software to conduct batch sampling of ponds, and adopt overlapping sampling for ponds that span multiple grids.

[0022] Optionally, based on the area and number of ponds in the mission area, the grid width is initialized, and the dynamic grid generation is performed in combination with the terrain complexity and the density of different ponds. The steps include:

[0023] S211. Calculate the pond density within the mission area based on the area and number of ponds within the mission area to obtain a pond density;

[0024] S212, calculating the grid width based on the obtained pond density and the number of ponds;

[0025] S213. Dynamically generate a grid based on the complexity of the terrain and the distribution density of different ponds.

[0026] Optionally, based on the complexity of the terrain and the distribution density of different ponds, dynamic grid generation includes the following steps:

[0027] S2131. Divide the grid into a basic sampling grid and a fine sampling grid according to different sizes according to a preset grid size, wherein the size of the basic sampling grid is larger than the size of the fine sampling grid;

[0028] S2132. When the terrain in the mission area is complex or the area is densely populated with water, a fine sampling grid is selected; when the terrain in the mission area is simple or the area is sparsely populated with ponds, a basic sampling grid is selected.

[0029] Optionally, optimizing the quality of sampled pit and pond samples based on pre-acquired urban land space monitoring road network data, and utilizing big data analysis to evaluate the quality of the pit and pond samples and replace them includes the following steps:

[0030] S31. Based on the pre-acquired urban land space monitoring road network data, use geographic information software to calculate the straight-line distance from the centroid of each pothole to the nearest road, and perform accessibility analysis on the pothole samples based on a pre-set distance threshold;

[0031] S32, calculating the shape index of the patch based on its perimeter and area, and replacing patches exceeding the pre-set length threshold according to the pre-set length threshold;

[0032] S33. Based on the pit pond samples after distance and shape optimization, use big data analysis to calculate the multi-dimensional indicators of the pit pond samples, perform secondary optimization on the pit pond samples, and after evaluating the quality of the samples, identify and replace the pit pond samples that do not meet the expected quality.

[0033] Optionally, based on pre-acquired urban land space monitoring road network data, geographic information software is used to calculate the straight-line distance from the centroid of each pothole to the nearest road, and accessibility analysis of the pothole samples is performed based on a pre-set distance threshold, including the following steps:

[0034] S311: When the straight-line distance from the centroid of the pothole patch to the nearest road does not exceed a preset distance threshold, retain the pothole sample;

[0035] S312. When the straight-line distance from the centroid of the pothole patch to the nearest road exceeds a preset distance threshold, the pothole sample is marked as an unreachable sample, and is replaced by a sample with a road-reachable area and the largest area from the sample pool in the grid.

[0036] Optionally, encoding is performed based on the optimized pond samples, and a sample database is constructed using geographic information software; a water depth inversion model is constructed using programming software and a machine learning algorithm to obtain a water depth inversion result, and the water storage capacity of the pond is calculated in combination with the pond area, including the following steps:

[0037] S41. Based on the optimized pit and pond samples, encoding is performed according to pre-set encoding rules to construct a sample database;

[0038] S42. Based on the sample database, a sample data set is constructed in combination with optical images, and a water depth inversion model is constructed using programming software to obtain the average water depth of the pond. Combined with the pond area, the water storage capacity of the pond is calculated.

[0039] Optionally, based on the sample database, a sample data set is constructed in combination with optical images, and a water depth inversion model is constructed using programming software to obtain the average water depth of the pond. Combined with the pond area, the water storage capacity of the pond is calculated, including the following steps:

[0040] S421. For pond samples obtained using the grid sampling method, conduct water depth measurements using the single-beam water level gauge method and use these as sample data for pond water depth remote sensing inversion.

[0041] S422. Preprocess the optical image, calculate the characteristic importance of the spectral reflectance, and use it as the inversion factor sample data for water depth remote sensing inversion;

[0042] S423, forming a sample data set based on the sample data of pond water depth remote sensing inversion and the sample data of inversion factors of water depth remote sensing inversion, and dividing the data into a training set and a validation set, using the root mean square error and the coefficient of determination as evaluation indicators to verify the inversion accuracy;

[0043] S424, using programming software to perform mask extraction on the inversion results to obtain an inversion spatial distribution map containing only the pond area, and calculate the average water depth;

[0044] S425. Calculate the water storage capacity of the pond based on the obtained average water depth of the pond and the water surface area of ​​the pond.

[0045] The beneficial effects of the present invention are:

[0046] The present invention provides a dynamically adjustable grid sampling method that takes into account the spatial distribution of ponds, their area and shape characteristics, and terrain differences. This can optimize the spatial coverage balance of samples, such as automatically adapting to sample gaps in high / low density areas, which is superior to traditional fixed grid random sampling methods. In addition, big data evaluation is embedded in the dynamic grid sampling process to optimize sample quality and avoid model bias caused by turbid pond samples, which can improve the accuracy of remote sensing water depth inversion and water storage capacity. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0048] Figure 1 This is an overall technical roadmap for a method for calculating pond water storage capacity based on dynamic grid sampling according to an embodiment of the present invention;

[0049] Figure 2 A dynamic grid sampling flow chart of a method for calculating pond water storage capacity based on dynamic grid sampling according to an embodiment of the present invention;

[0050] Figure 3 is a sample optimization flow chart of a method for calculating pond water storage capacity based on dynamic grid sampling according to an embodiment of the present invention;

[0051] Figure 4 is a schematic diagram of dynamic grid generation of a method for calculating pond water storage capacity based on dynamic grid sampling according to an embodiment of the present invention;

[0052] Figure 5is a spectrum curve comparison diagram of a method for calculating pond water storage capacity based on dynamic grid sampling according to an embodiment of the present invention;

[0053] Figure 6 spectral cosine similarity comparison diagram of a method for calculating pond water storage capacity based on dynamic grid sampling according to an embodiment of the present invention;

[0054] Figure 7 3. This is a comparison diagram before and after sample optimization of a method for calculating pond water storage capacity based on dynamic grid sampling according to an embodiment of the present invention;

[0055] Figure 8 A spatial distribution map of remote sensing water depth inversion according to a method for calculating pond water storage capacity based on dynamic grid sampling according to an embodiment of the present invention;

[0056] Figure 9 This is a verification diagram of remote sensing water depth inversion results of a pond water storage capacity calculation method based on dynamic grid sampling according to an embodiment of the present invention. DETAILED DESCRIPTION

[0057] To further illustrate each embodiment, the present invention provides drawings, which 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 operating principles of the embodiments. By referring to these contents, ordinary technicians in this field should be able to understand other possible implementation methods and advantages of the present invention. The components in the figures are not drawn to scale, and similar component symbols are generally used to represent similar components.

[0058] According to an embodiment of the present invention, a method for calculating pond water storage capacity based on dynamic grid sampling is provided.

[0059] The present invention will now be further described with reference to the accompanying drawings and specific embodiments. Figure 1 According to one embodiment of the present invention, a method for calculating pond water storage capacity based on dynamic grid sampling is provided. The method for calculating pond water storage capacity includes the following steps:

[0060] S1. Based on the pre-acquired optical image and the improved normalized water index, the pond water surface is extracted, and the area and number of ponds in the mission area are calculated; and the ponds are classified and sampled using the grid sampling method to obtain the number of pond samples.

[0061] In one embodiment, based on pre-acquired optical images and an improved normalized water index, the pond water surface is extracted, and the area and number of ponds within the task area are calculated. The ponds are then classified and sampled using a grid sampling method to obtain the number of pond samples, including the following steps:

[0062] S11. Based on the pre-acquired optical image and the improved normalized water index, the water boundary of the pond surface in the mission area is extracted, and the pond area is calculated using the spatial relationship analysis method.

[0063] In one embodiment, based on pre-acquired optical images and the improved normalized water index, extracting the water boundary of the pond surface within the task area and calculating the pond area using a spatial relationship analysis method includes the following steps:

[0064] S111. Based on the pre-acquired optical image and the improved normalized water index, extract the water boundary of the pond water surface in the task area, and set the improved normalized water index threshold;

[0065] S112, smoothing the extracted water surface boundary to remove jagged edges, and masking out rivers, lakes, reservoirs, and ditches to obtain a pond patch;

[0066] S113. Split a single pond into multiple plots, merge the plots using spatial relationship analysis, correct the plot boundaries, and calculate the pond area using geographic information software.

[0067] S12. Based on the sampling proportions in the classified pits and ponds, use the grid sampling method to extract the sample with the largest area in each grid and calculate the number of pit and pond samples.

[0068] It should be explained that to identify ponds based on the MNDWI index, the ponds are first extracted. Optical images (such as Sentinel-2) and the modified Normalized Difference Water Index (MNDWI) threshold method are used to preliminarily extract the water body boundaries. Based on experience and experiments, the MNDWI threshold is set to 0.3, and binary segmentation is performed. Areas with MNDWI greater than 0.3 are preliminarily extracted as water bodies.

[0069] ;

[0070] Where, MNDWI represents the modified normalized water index, Indicates the green band, Indicates the shortwave infrared band.

[0071] Among them, the extracted water body boundaries need to be smoothed. Use the smooth surface tool of the geographic information software, select the PAEK algorithm, set the tolerance to 20 meters, smooth the water surface boundaries and remove the jagged edges; then, combine the land change survey results data, mask out the data of rivers, lakes, reservoirs and ditches, and extract the pond patches.

[0072] Then the patches are merged. To address the problem of fragmented patches, such as when a single pond is divided into multiple patches, the spatial relationship analysis method is used to merge the fragmented patches and correct the patch boundaries. The specific method is to use geographic information software to perform buffer fusion and generate buffers for adjacent patches. The threshold is set to 1 pixel width. If the buffers overlap, the patches are merged to systematically solve the patch fragmentation problem and ensure that the patch boundaries are consistent with the real objects. Finally, the area is calculated using the computational geometry tool of the geographic information software to automatically calculate the pond area.

[0073] The categorical sampling design is as follows:

[0074] 1) For aquaculture ponds, the grid sampling method was used to select the largest sample in each grid. The number of aquaculture pond samples is calculated using the following formula:

[0075] ;

[0076] Where, Indicates the total number of aquaculture ponds, k represents the sampling ratio, Indicates the number of aquaculture pond samples taken.

[0077] 2) For natural ponds, the grid sampling method was used to select the largest sample in each grid. The number of natural pond samples was calculated using the following formula:

[0078] ;

[0079] Where, represents the total number of natural ponds, k represents the sampling ratio, Indicates the number of natural pond samples taken.

[0080] S2. Based on the land area, pond area and number of ponds in the mission area, calculate the pond density and grid width, generate the grid dynamically, and use programming software to conduct batch sampling of ponds.

[0081] In one embodiment, based on the land area, pond area, and number of ponds within the mission area, the pond density and grid width are calculated, a dynamic grid is generated, and batch sampling of ponds is performed using programming software, including the following steps:

[0082] S21. Initialize the grid width based on the area and number of ponds in the mission area, and dynamically generate the grid based on the complexity of the terrain and the density of different ponds.

[0083] In one embodiment, based on the area and number of ponds in the mission area, the grid width is initialized, and the dynamic grid generation is performed in combination with the terrain complexity and the density of different ponds. The steps include:

[0084] S211. Calculate the pond density within the mission area based on the area and number of ponds within the mission area to obtain a pond density;

[0085] S212, calculating the grid width based on the obtained pond density and the number of ponds;

[0086] S213. Dynamically generate a grid based on the complexity of the terrain and the distribution density of different ponds.

[0087] In one embodiment, dynamic grid generation based on the complexity of the terrain and the distribution density of different ponds includes the following steps:

[0088] S2131. Divide the grid into a basic sampling grid and a fine sampling grid according to different sizes according to a preset grid size, wherein the size of the basic sampling grid is larger than the size of the fine sampling grid;

[0089] S2132. When the terrain in the mission area is complex or the area is densely populated with water, a fine sampling grid is selected; when the terrain in the mission area is simple or the area is sparsely populated with ponds, a basic sampling grid is selected.

[0090] S22. Based on the dynamically adjusted grid, use programming software to conduct batch sampling of ponds, and adopt overlapping sampling for ponds that span multiple grids.

[0091] It should be explained that the dynamically adjustable grid is generated by extracting the land area of ​​each district and county based on land change survey data, with the unit being square kilometers. The initial grid width is then used to calculate the pond density. The pond density within the mission area is evaluated to obtain the number of ponds per square kilometer. The calculation formula is as follows:

[0092] ;

[0093] Where, represents the density of ponds in the work area, Indicates the total number of ponds in the work area, Indicates the land area within the work area.

[0094] Determine the grid width based on the density of ponds and calculate the grid width as follows:

[0095] ;

[0096] Where, represents the width of the regular grid, represents the number of samples in the grid, Indicates the sampling ratio.

[0097] in, The unit is square kilometers, The unit is kilometers, , which means that at least one sample is drawn from each grid.

[0098] In addition, considering the complexity of terrain and the density of ponds in different regions, a smaller grid can be selected in areas with complex terrain or dense water areas to more accurately capture water depth changes; while a larger grid can be selected in areas with simpler terrain or sparse ponds, such as Figure 4 shown.

[0099] Based on the latest land change survey data, the grid width parameter calculation results are given. Taking the ponds in a southern province as an example, the pond density classification results of the province are expressed as follows:

[0100] ;

[0101] Where, Indicates pond density.

[0102] The Python program was used to conduct batch sampling in large areas (more than two counties). During sampling, ponds that crossed county administrative boundaries were removed to keep the sampling number unchanged. The pond boundary vector data and pond type labels were input as the data to be sampled. The batch sampling process using the Python program included the following steps:

[0103] 1) Calculate the number of ponds to be sampled in each county based on the determined sampling ratio.

[0104] 2) Calculate the grid width for each county and generate grids of different widths.

[0105] 3) Eliminate empty grids and sort the grids in descending order by the number of ponds within them, prioritizing sampling from high-density grids to ensure coverage of hotspot areas. Secondly, sort the ponds in descending order by their area within the grids. Then, loop through the grids, sampling the largest pond from each grid in turn until the sampling quantity requirement is met.

[0106] 4) Termination condition: if there are insufficient samples after traversal, the grid width will be expanded and resampled.

[0107] 5) For ponds that span multiple grids, overlapping sampling is used to avoid omissions; determine whether the spot x intersects with the grids. When the pond spot intersects with grids A and B, the spot is included in the sampling base of grid A. At this time, it is determined that the spot x is already in the sampling base and is no longer counted in the sampling base of grid B.

[0108] S3. Based on the pre-acquired urban land space monitoring road network data, optimize the quality of the sampled pit and pond samples, and use big data analysis to evaluate the quality of the pit and pond samples and replace them.

[0109] In one embodiment, the quality of sampled pit and pond samples is optimized based on pre-acquired urban land space monitoring road network data, and big data analysis is used to evaluate the quality of the pit and pond samples and replace them, including the following steps:

[0110] S31. Based on the pre-acquired urban land space monitoring road network data, use geographic information software to calculate the straight-line distance from the centroid of each pothole to the nearest road, and perform accessibility analysis on the pothole samples based on the pre-set distance threshold.

[0111] In one embodiment, based on pre-acquired urban land space monitoring road network data, geographic information software is used to calculate the straight-line distance from the centroid of each pothole patch to the nearest road, and accessibility analysis of the pothole samples is performed based on a pre-set distance threshold, including the following steps:

[0112] S311: When the straight-line distance from the centroid of the pothole patch to the nearest road does not exceed a preset distance threshold, retain the pothole sample;

[0113] S312. When the straight-line distance from the centroid of the pothole patch to the nearest road exceeds a preset distance threshold, the pothole sample is marked as an unreachable sample, and is replaced by a sample with a road-reachable area and the largest area from the sample pool in the grid.

[0114] S32. Calculate the shape index of the patch based on its perimeter and area, and replace the patch that exceeds the pre-set length threshold according to the pre-set length threshold.

[0115] S33. Based on the pit pond samples after distance and shape optimization, use big data analysis to calculate the multi-dimensional indicators of the pit pond samples, perform secondary optimization on the pit pond samples, and after evaluating the quality of the samples, identify and replace the pit pond samples that do not meet the expected quality.

[0116] It needs to be explained that by collecting urban land space monitoring road network data, performing accessibility analysis, and replacing potholes that are more than 1,000 meters away from the nearest road, the specific approach of the accessibility analysis is as follows: input pothole boundary data and road network vector data, and use the neighbor analysis tool of the geographic information software to calculate the straight-line distance from the centroid of each pothole patch to the nearest road. The threshold is set to 1,000 meters. When the distance is less than 1,000 meters, the sample is retained; otherwise, it is marked as an unreachable sample, and the sample with the largest area and accessible by road is replaced from the sample pool in the grid; avoid thin and narrow samples, calculate the patch shape index, such as the ratio of perimeter to area, that is, the length-to-narrow ratio, and set a threshold of 0.5. Samples exceeding this threshold are also considered thin and narrow samples and replaced.

[0117] In addition, based on the optimization of parameters such as distance and shape in the previous step, we further use big data analysis to calculate multi-dimensional indicators (water quality, historical changes, surrounding environment) to perform secondary optimization of samples, evaluate the representativeness and data quality of samples, identify and replace low-quality samples, Figure 3 Optimize the technical roadmap for the sample, and its implementation steps:

[0118] 1) Multi-dimensional feature extraction: Calculate the following three feature types (NDTI variance, plant density, and spectral similarity) for candidate samples.

[0119] NDTI variance: Obtain at least two images and perform preprocessing such as radiometric calibration, atmospheric correction, and geometric registration. Calculate the sample NDTI variance with a threshold of 0.1 to exclude samples with unstable water quality whose NDTI variance is higher than the threshold. NDTI represents water turbidity through the reflectance ratio of the red and green light bands as follows:

[0120] ;

[0121] The average NDTI and NDTI variance of the pothole area in the image are calculated as follows:

[0122] ;

[0123] ;

[0124] Where, Indicates the degree of fluctuation of water turbidity NDTI variance, Indicates the representative t A pond in the first image NDTI mean, Indicates multiple images The average value of represents the reflectivity of red light band, represents the reflectivity of the green light band, Indicates the iThe red band reflectance of each pixel, Indicates the i The green band reflectance of each pixel.

[0125] in, t =1, 2, … T , T is the image period, i =1, 2, … N , N Indicates the total number of pixels in a pond.

[0126] In addition, for factory density, we collected factory data with the number and location, calculated the factory density within a 2000m buffer zone of the sample, set the threshold at 3 factories / km², and replaced samples with factory density greater than the threshold to avoid interference from surrounding pollution sources. For spectral similarity, we calculated the spectral cosine similarity with the sample spectral characteristics to ensure the consistency of the model input. The similarity value range is [-1, 1], where 1 indicates exact similarity, 0 indicates irrelevant, and -1 indicates complete opposite. We retained candidate samples with a similarity greater than 0.95 to ensure the adaptability of the water depth inversion. The expression is as follows:

[0127] ;

[0128] Where, represents the cosine similarity, Indicates that the candidate sample is i The reflectivity value of each band, Indicates that the high-quality samples are i The reflectivity value of each band, Indicates the number of bands involved in the calculation.

[0129] 2) Low-quality sample replacement decision

[0130] Input candidate samples, calculate three parameters including NDVI variance, plant density and spectral similarity, determine sample quality based on the set threshold, and mark the samples as binary categories, 1 for "high-quality samples" and 0 for "low-quality samples". For samples marked as 0, select those that meet the conditions (NDVI variance less than 0.1; spectral similarity greater than 0.95; plant density ≤ 3 / km) from the candidate pool. 2 ) replacement samples; for samples with missing features, samples are adjusted through manual review.

[0131] S4. Encode the optimized pond samples and build a sample database using geographic information software. Use programming software and machine learning algorithms to build a water depth inversion model to obtain water depth inversion results. Combined with the pond area, calculate the water storage capacity of the pond.

[0132] In one embodiment, encoding is performed based on the optimized pond samples, and a sample database is constructed using geographic information software. A water depth inversion model is constructed using programming software and a machine learning algorithm to obtain water depth inversion results. The water storage capacity of the pond is calculated based on the pond area, including the following steps:

[0133] S41. Based on the optimized pit and pond samples, encoding is performed according to pre-set encoding rules to build a sample database.

[0134] S42. Based on the sample database, a sample data set is constructed in combination with optical images, and a water depth inversion model is constructed using programming software to obtain the average water depth of the pond. Combined with the pond area, the water storage capacity of the pond is calculated.

[0135] In one embodiment, a sample data set is constructed based on a sample database and optical images, and a water depth inversion model is constructed using programming software to obtain the average water depth of the pond. The water storage capacity of the pond is calculated based on the pond area, including the following steps:

[0136] S421. For pond samples obtained using the grid sampling method, conduct water depth measurements using the single-beam water level gauge method and use these as sample data for pond water depth remote sensing inversion.

[0137] S422. Preprocess the optical image, calculate the characteristic importance of the spectral reflectance, and use it as the inversion factor sample data for water depth remote sensing inversion;

[0138] S423, forming a sample data set based on the sample data of pond water depth remote sensing inversion and the sample data of inversion factors of water depth remote sensing inversion, and dividing the data into a training set and a validation set, using the root mean square error and the coefficient of determination as evaluation indicators to verify the inversion accuracy;

[0139] S424, using programming software to perform mask extraction on the inversion results to obtain an inversion spatial distribution map containing only the pond area, and calculate the average water depth;

[0140] S425. Calculate the water storage capacity of the pond based on the obtained average water depth of the pond and the water surface area of ​​the pond.

[0141] It should be explained that the standardization of sample data first unifies the data spatial reference benchmark and the pond ID to facilitate large-area data integration and field measurement patch identification. Then, the land type patch number and name are retained, and the county-level administrative district name and remarks fields are added, as well as the pond code field. The sample patches are coded according to certain rules to ensure the uniqueness of the codes. For example:

[0142] Pond coding rule: XZQBMMSYLXNNNNNN;

[0143] Among them, XZQBMM represents the 6-digit county-level administrative district code, SYLX represents the 4-digit water area type (1104 for ponds), and NNNNNN represents the 6-digit pond serial number.

[0144] Finally, batches of city and county name folders are created, vector files are exported, and a sample database is formed; statistical tables are generated to calculate the number of each county and sample patches.

[0145] In the construction of the water depth inversion model, the pond samples obtained by the grid sampling method are first measured in the field using single-beam and water level gauge methods to serve as sample data for the remote sensing inversion of pond water depth; then Sentinel-2 images are collected and preprocessed such as radiometric calibration, atmospheric correction, and orthorectification are performed, and the characteristic importance of spectral reflectance is calculated as sample data for the inversion factor of water depth remote sensing inversion.

[0146] In addition, radiometric calibration is to correctly obtain the calibration coefficients of each band of the image corresponding to the corresponding time, and convert the original pixel brightness value recorded in the image into the actual radiometric brightness value through the following formula; this can be achieved using the "RadiometricCalibration" tool of the geographic information software, and the expression is as follows:

[0147] ;

[0148] Where, represents the radiance, Represents the brightness value of the remote sensing image pixel, represents the radiation calibration coefficient increment, Indicates the radiometric calibration coefficient offset.

[0149] Atmospheric correction uses the image's spectral response function, the sensor's orbital altitude, observation angle, imaging time, the corresponding solar azimuth at the time of imaging, the aerosol type and geographic location of the imaging area, and other information. It employs a model-based atmospheric correction method, selecting an appropriate radiation transfer model (6S, MODTRAN, LOWTRAN, ATCOR, FLAASH, etc.) or using a lookup table (LUT) to accurately remove atmospheric effects, complete the correction of the image data, and achieve the conversion from radiance to actual surface reflectance. This can be achieved using the "FLAASH Atmospheric Correction" tool in geographic information software.

[0150] Orthorectification: During orthorectification, obvious, clear, and time-invariant ground control points are uniformly selected on the image; the image is resampled using appropriate methods (nearest neighbor method, bilinear interpolation, cubic convolution interpolation, etc.) to obtain a corrected image; this can be achieved using the "RPC Orthorectification Workflow" tool in geographic information software.

[0151] The characteristic importance of the inversion factor is calculated by extracting the sample reflectance of the preprocessed image according to the spatial position of the measured water depth sample point, and calculating the characteristic importance of the inversion factors such as (ratio, logarithmic ratio). The characteristic importance of each inversion factor is calculated. The higher the characteristic importance value, the higher the correlation between the factor and the water depth. The expression is as follows:

[0152] ;

[0153] ;

[0154] Where, represents the ratio inversion factor, represents the log-ratio inversion factor, represents the blue band reflectivity, Represents the green band reflectance.

[0155] The inversion factors and measured water depth data constitute the sample data set, which is divided into training set and validation set at an 8:2 ratio. Training models such as BP neural network, random forest, and support vector machine are selected. The root mean square error (RMSE) and coefficient of determination (R²) are used as evaluation indicators to verify the inversion accuracy. The expressions are as follows:

[0156] ;

[0157] ;

[0158] Where, Indicates the i A true value, Indicates the i An estimated value, represents the mean of the true values, Represents the total number of true values.

[0159] Using the mask extraction tool and pond vector map of geographic information software, we performed mask extraction on the inversion results of the entire image to obtain an inversion spatial distribution map containing only the pond area, and calculated the average water depth. The expression is as follows:

[0160] ;

[0161] Where, Indicates the i The average water depth of the ponds, Indicates the i The pond u The measured water depth value of each measuring point.

[0162] The water storage capacity of a pond is calculated by multiplying the water surface area by the average water depth. The expression is as follows:

[0163] ;

[0164] Where, represents the total water storage capacity of the region, Indicates the i The area of ​​a pond.

[0165] The specific steps of the implementation process are as follows:

[0166] 1) Data preparation: Collect 2-meter resolution GF1 imagery and 10-meter resolution Sentinel-2 imagery data consistent with the measured data date, as well as land change survey results; and measure the water depth values ​​of sample ponds.

[0167] 2) MNDWI water body extraction: A threshold of 0.3 was set, and the GF1 image and exponential method were used to extract water body boundaries. Combined with the land change survey data, the data of rivers, lakes, and reservoirs were masked out to extract the boundaries of ponds.

[0168] 3) Dynamic grid generation: A high-density area (5km×5km) and a low-density area (6km×6km) were selected as test areas within the mission area. The density of ponds in the high-density area was 40 / km. 2 The calculated grid width is 0.83 km, rounded to 1 km; the density of ponds in the low-density area is 2 / km 2 , the calculated grid width is 4km.

[0169] in, Figure 4 By comparing the division effects of the traditional fixed 2km grid and the dynamic grid, it can be found that the dynamic grid can better take into account the distribution characteristics of water bodies. The fixed grid is suitable for sampling in a single district or county; the dynamic grid is suitable for sampling in a large area and can capture local changes according to the distribution characteristics of water bodies, ensuring uniform spatial coverage of samples. Figure 4 (a) is the fixed grid a=2km in the low-density area; Figure 4 (b) is the fixed grid a=2km in the high-density area; Figure 4 (c) is the dynamic grid of low-density area with a=4km; Figure 4 (d) is the dynamic grid of high-density area a=1km.

[0170] 4) Sample selection and optimization. Based on the traditional fixed grid sampling method, one pond was randomly selected from each grid as a sample. A total of 25 samples were collected from high-density areas and 29 samples were collected from low-density areas. Based on the dynamic grid sampling method, the largest pond in each grid was selected as a sample (including ponds in the quad grid). Optimization processes such as accessibility analysis, water quality stability analysis, and spectral adaptability analysis were then performed. Ultimately, 28 samples were replaced in high-density areas, for a total of 107 samples. Three samples were replaced in low-density areas, for a total of 11 samples.

[0171] in, Figure 5 and Figure 6 The spectral curves and spectral cosine similarities of different samples were displayed. It was found that the cosine similarity between low-quality samples and high-quality samples was poor due to the water body itself. This indicator can be used to screen and optimize pond samples. Figure 7 The sample comparison chart of fixed grid and dynamic grid shows that the samples optimized based on dynamic grid and big data are more typical and representative, avoiding samples of small area, narrow shape, poor water quality, land type change, and difficult transportation. Circles mark the key samples missed by traditional methods, and boxes highlight the sample hotspots added by dynamic methods. Figure 7 (a) is a random sampling of fixed grids in low-density areas; Figure 7 (b) is random sampling of fixed grids in high-density areas; Figure 7 (c) is the dynamic grid optimization sampling in low-density areas; Figure 7 (d) is the dynamic grid optimization sampling in high-density areas.

[0172] 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 were obtained, and the random forest algorithm was used to construct a water depth inversion model.

[0173] in, Figure 8 The inversion results based on traditional fixed grid and dynamic grid are shown. Figure 8 (a) is the inversion result based on fixed grid sampling, Figure 8 (b) is the inversion result based on dynamic grid sampling. Figure 9 The verification scatter plot of the water depth inversion results is shown, where Figure 9 (a) is a scatter plot of the water depth inversion results verification under the "fixed grid + random sampling" method. Figure 9 (b) is a scatter plot verifying the water depth inversion results under the “dynamic grid + optimized sampling” method.

[0174] The results show that under the “fixed grid + random sampling” method, there are problems such as less 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, and the inversion effect is better, RMSE=0.22, R 2 =0.56, compared with the traditional method, the correlation is improved by 51% and the accuracy is improved by 46%.

[0175] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for calculating pond water storage capacity based on dynamic grid sampling, characterized in that: The method comprises the following steps: S1. Based on pre-acquired optical images and the improved normalized water index, extract the pond water surface and calculate the area and number of ponds within the mission area. Then, 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 within the mission area, calculate the pond density and grid width, dynamically generate the grid, and use programming software to conduct batch sampling of ponds; S3. Based on the pre-acquired urban land space monitoring road network data, optimize the quality of the sampled pit and pond samples, and use big data analysis to evaluate the quality of the pit and pond samples and replace them; S4. Encode the optimized pond samples and build a sample database using geographic information software. Use programming software and machine learning algorithms to build a water depth inversion model, obtain water depth inversion results, and calculate the water storage capacity of the ponds based on the pond area. The method of optimizing the quality of the sampled pit and pond samples based on the pre-acquired urban land space monitoring road network data, and using big data analysis to evaluate the quality of the pit and pond samples and replace them includes the following steps: S31. Based on the pre-acquired urban land space monitoring road network data, use geographic information software to calculate the straight-line distance from the centroid of each pothole to the nearest road, and perform accessibility analysis on the pothole samples based on a pre-set distance threshold; S32, calculating the shape index of the patch based on its perimeter and area, and replacing patches exceeding the pre-set length threshold according to the pre-set length threshold; S33. Based on the pit pond samples after distance and shape optimization, use big data analysis to calculate the multi-dimensional indicators of the pit pond samples, perform secondary optimization on the pit pond samples, and after evaluating the quality of the samples, identify and replace the pit pond samples that do not meet the expected quality.

2. The method for calculating pond water storage capacity based on dynamic grid sampling according to claim 1, characterized in that: The method extracts the pond water surface based on the pre-acquired optical image and the improved normalized water index, and calculates the pond area and number in the task area; The grid sampling method is used to classify and sample the ponds. The following steps are involved in obtaining the number of pond samples: S11. Based on the pre-acquired optical imagery and the improved normalized water index, extract the water boundaries of the ponds within the mission area and calculate the pond areas using spatial relationship analysis. S12. Based on the sampling proportions in the classified pits and ponds, use the grid sampling method to extract the sample with the largest area in each grid and calculate the number of pit and pond samples.

3. The method for calculating pond water storage capacity based on dynamic grid sampling according to claim 2 is characterized in that: The method of extracting the water boundary of the pond surface in the mission area based on the pre-acquired optical image and the improved normalized water index, and calculating the pond area using the spatial relationship analysis method includes the following steps: S111. Based on the pre-acquired optical image and the improved normalized water index, extract the water boundary of the pond water surface in the task area, and set the improved normalized water index threshold; S112, smoothing the extracted water surface boundary to remove jagged edges, and masking out rivers, lakes, reservoirs, and ditches to obtain a pond patch; S113. Split a single pond into multiple plots, merge the plots using spatial relationship analysis, correct the plot boundaries, and calculate the pond area using geographic information software.

4. The method for calculating pond water storage capacity based on dynamic grid sampling according to claim 1, characterized in that: The method of calculating the pond density and grid width based on the land area, pond area, and number of ponds in the mission area, dynamically generating a grid, and batch sampling the ponds using programming software includes the following steps: S21. Initialize the grid width based on the area and number of ponds in the mission area, and dynamically generate the grid based on the terrain complexity and the density of different ponds. S22. Based on the dynamically adjusted grid, use programming software to conduct batch sampling of ponds, and adopt overlapping sampling for ponds that span multiple grids.

5. The method for calculating pond water storage capacity based on dynamic grid sampling according to claim 4, characterized in that: The method of initializing the grid width based on the area and number of ponds in the mission area and combining the terrain complexity and the density of different ponds to dynamically generate the grid includes the following steps: S211. Calculate the pond density within the mission area based on the area and number of ponds within the mission area to obtain a pond density; S212, calculating the grid width based on the obtained pond density and the number of ponds; S213. Dynamically generate a grid based on the complexity of the terrain and the distribution density of different ponds.

6. The method for calculating pond water storage capacity based on dynamic grid sampling according to claim 5, characterized in that: The dynamic grid generation according to the complexity of the terrain and the distribution density of different ponds includes the following steps: S2131. Divide the grid into a basic sampling grid and a fine sampling grid according to different sizes according to a preset grid size, wherein the size of the basic sampling grid is larger than the size of the fine sampling grid; S2132. When the terrain in the mission area is complex or the area is densely populated with water, a fine sampling grid is selected; when the terrain in the mission area is simple or the area is sparsely populated with ponds, a basic sampling grid is selected.

7. The method for calculating pond water storage capacity based on dynamic grid sampling according to claim 1, characterized in that: The method of calculating the straight-line distance from the centroid of each pothole to the nearest road using geographic information software based on the pre-acquired urban land space monitoring road network data and performing accessibility analysis on the pothole samples based on a pre-set distance threshold includes the following steps: S311: When the straight-line distance from the centroid of the pothole patch to the nearest road does not exceed a preset distance threshold, retain the pothole sample; S312. When the straight-line distance from the centroid of the pothole patch to the nearest road exceeds a preset distance threshold, the pothole sample is marked as an unreachable sample, and is replaced by a sample with a road-reachable area and the largest area from the sample pool in the grid.

8. The method for calculating pond water storage capacity based on dynamic grid sampling according to claim 1, characterized in that: The method comprises the following steps: encoding the optimized pond samples, building a sample database using geographic information software, building a water depth inversion model using programming software and a machine learning algorithm, obtaining a water depth inversion result, and calculating the water storage capacity of the pond based on the pond area. S41. Based on the optimized pit and pond samples, encoding is performed according to pre-set encoding rules to construct a sample database; S42. Based on the sample database, a sample data set is constructed in combination with optical images, and a water depth inversion model is constructed using programming software to obtain the average water depth of the pond. Combined with the pond area, the water storage capacity of the pond is calculated.

9. The method for calculating pond water storage capacity based on dynamic grid sampling according to claim 8, characterized in that: The method of constructing a sample data set based on a sample database and optical images, and constructing a water depth inversion model using programming software to obtain the average water depth of the pond, and calculating the water storage capacity of the pond based on the pond area, includes the following steps: S421. For pond samples obtained using the grid sampling method, conduct water depth measurements using the single-beam water level gauge method and use these as sample data for pond water depth remote sensing inversion. S422. Preprocess the optical image, calculate the characteristic importance of the spectral reflectance, and use it as the inversion factor sample data for water depth remote sensing inversion; S423, forming a sample data set based on the sample data of pond water depth remote sensing inversion and the sample data of inversion factors of water depth remote sensing inversion, and dividing the data into a training set and a validation set, using the root mean square error and the coefficient of determination as evaluation indicators to verify the inversion accuracy; S424, using programming software to perform mask extraction on the inversion results to obtain an inversion spatial distribution map containing only the pond area, and calculate the average water depth; S425. Calculate the water storage capacity of the pond based on the obtained average water depth of the pond and the water surface area of ​​the pond.

Images