A method for flood inundation prediction based on rainfall

Abstract

The present invention relates to the technical field of flood disaster prediction, and specifically to a rainfall-based flood inundation prediction method. The present invention comprehensively addresses the deficiencies in existing flood prediction methods in situations where passive inundation processes such as water accumulation in low-lying areas due to high-intensity, large-scale rainfall are the cause. By performing inundation analysis based on DEM data, statistically analyzing the relationship between water level-inundated area and water level-inundated volume, a flood inundation prediction model is constructed in which the inundated area changes with water level. On this basis, the principle of total water volume equality is utilized to develop a rapid inundation range prediction method based on total rainfall data, thereby further delineating the scope of flood monitoring areas.

Description

Technical Field

[0001] The present invention relates to the technical field of flood disaster prediction, and in particular to a rainfall-based flood inundation prediction method. Background Art

[0002] The mechanism of flooding is that there is a channel (such as a breach or sluice gate release) between the source area and the flooded area, and a water level difference exists. Inundation is a dynamic and equilibrium process. However, the process of water accumulation in low-lying areas caused by high-intensity, large-scale rainfall is more similar to a passive inundation process. Existing methods do not yet exist for predicting flooding caused by passive inundation processes such as rainfall.

[0003] Therefore, this application proposes a rainfall-based flood inundation prediction method, which uses the water level-water volume relationship in the flood inundation area to quickly estimate the inundated area. It can quickly determine the monitoring range based on the determination of the target area, and provide a more detailed scope delineation basis for the subsequent download of flood monitoring data and extraction of disaster information. Summary of the Invention

[0004] The purpose of the present invention is to overcome the shortcomings of the existing technology and provide a rainfall-based flood inundation prediction method. It uses the water level-water volume relationship in the flood inundation area to quickly estimate the inundated area. On the basis of determining the target area, it can quickly determine the monitoring range, providing a more detailed scope delineation basis for the subsequent downloading of flood monitoring data and extraction of disaster information.

[0005] In order to achieve the above object, the present invention provides a rainfall-based flood inundation prediction method, comprising:

[0006] 1.1 Analysis of lake water volume changes;

[0007] 1.2 Analysis of historical rainfall accumulation;

[0008] 1.3 Lake volume change prediction model;

[0009] 1.4 Prediction of flooding range;

[0010] 1.1 The specific steps for analyzing lake water volume changes are:

[0011] S1, Submergence analysis:

[0012] Active flooding:

[0013] S1-1, create a new empty inundation map: Based on the input DEM raster file, generate an empty file with the same raster information as the inundation map file. The inundation map raster file has the same row and column values, four corner coordinates, grid size, and coordinate system as the input DEM raster file, and all grid values ​​are empty;

[0014] S1-2, processing input source points:

[0015] S1-2-1, based on the input source point coordinates, calculate its corresponding grid on the DEM, and take its row number r, column number c, and elevation value e;

[0016] S1-2-2, create a new source point list to store source points. The source point type is a tuple, that is, the source point list is written as [(r, c), ...], where r is the row number and c is the column number;

[0017] S1-2-3, store the row and column numbers of the source point grid as the first source point in the source point list;

[0018] S1-2-4, define the current flooding elevation E and assign it a value of e as the flooding elevation of the first step of flooding;

[0019] S1-3, taking E as the inundation elevation, complete the active inundation expansion and spread algorithm at water level E according to the source point list, and update the inundation map:

[0020] S1-3-1, create a new source point deletion list: used to store source point information that is no longer located at the boundary and needs to be deleted from the source point list;

[0021] S1-3-2, traverse the source point list: for the current source point, take the 8 neighboring grids of the source point to form a neighborhood list; set the boundary flag and initialize it to False;

[0022] S1-3-3, traverse the neighborhood list, for the current neighborhood point, determine whether the value of the neighborhood point in the flood map is empty; if the value is not empty, it means that the point has been flooded, and the current neighborhood point is not processed; if the value is empty, proceed to the next step; determine the elevation value e' of the neighborhood point, if e'>E, if the neighborhood point is not flooded, do not process the neighborhood point, and set the boundary flag of the current source point to True; if e'≤E, the point is flooded, add the neighborhood point to the source point list, and update the grid value of the point in the flood map to E;

[0023] S1-3-4, after traversing all neighboring points, if the boundary point flag is still False, it means that all neighbors of the source point have been submerged and are no longer on the boundary. The source point is added to the source point deletion list;

[0024] S1-3-5, after traversing all source points, delete the source points recorded in the source point deletion list from the source point list; the flooded elevation increases by one step, that is, E=E+E step ; If the source point list is not empty, repeat steps 1-3; otherwise, the inundation map calculation is completed and the program exits;

[0025] S1-3-6, using the inundation map algorithm, obtain the "water level-inundated area relationship" and "water level-inundated volume relationship". The "water level-inundated volume relationship" is used as a conversion basis to link rainfall and inundation range. Passive inundation: By calculating the DEM volume below the water level, the "water level-inundated volume relationship" in S1-3-6 can be obtained.

[0026] S2, measured water level data:

[0027] By collecting the water level data of each station in the existing relevant hydrological data, and using the closest water level data between different stations as the modeled water level;

[0028] S3, water level-volume conversion model:

[0029] S3-1: Use the DEM model to calculate the relationship between water level and submerged volume. The statistical area covers the entire water area, and the data from historical water level changes and a certain period during the high flood season are used as the basis for conversion.

[0030] S3-2, fitting the water level-volume statistical results:

[0031] The results in S3-1 are fitted and the water level-volume direct conversion method is used to retain the volume of the target water level e under the influence of terrain. The calculation formula is:

[0032] Formula 1: V e =V n +(V n+1 -V n )×(en);

[0033] Where n is the integer part of the water level e, V n V is the submerged volume corresponding to the queried water level n, n+1 is the submerged volume corresponding to the queried water level n+1;

[0034] S4, water level volume change:

[0035] The water level-submerged volume conversion based on DEM is used to calculate the water level volume change. The daily lake volume V is obtained by conversion based on the daily water level measured data. n , the change in water level volume on day n is ΔV n , and its calculation formula is:

[0036] Formula 2: ΔV n =V n+1 -V n ;The water level volume variable is the change in water volume in the next 24 hours;

[0037] 1.2 The specific steps of historical rainfall accumulation analysis are:

[0038] S10: Rainfall data collection:

[0039] Collect hourly rainfall data sets over land areas over the years through ERA5;

[0040] S20: Rainfall data normalization:

[0041] Normalize hourly rainfall into 24-hour cumulative rainfall, normalize rainfall data by distance and trim data outside the basin;

[0042] S30: Historical cumulative rainfall:

[0043] The highest historical water level point is used as the statistical object, and the rainfall data is accumulated from 0:00 to 24:00. The spatial range is based on the preset radius of the target area covered by the main body as the starting point. According to the preset step size, the daily cumulative rainfall data of the annular area around the main body center is calculated.

[0044] 1.3 The specific steps of the lake volume change prediction model are:

[0045] S100, prediction model variable selection:

[0046] Based on rainfall data, a statistical relationship model between measured water level changes and accumulated rainfall is constructed to comprehensively consider the overall impact of outflow, evaporation, and groundwater exchange factors;

[0047] S200, variable correlation analysis:

[0048] S200-1, by accumulating the annular area, the cumulative variable of the circular area rainfall is constructed, and the bivariate analysis of the double annular range and the circular range is performed;

[0049] S200-2, taking into account the time factor, conducts a time series characteristic analysis of the water level and volume changes and the accumulated rainfall at different distances;

[0050] S200-3, based on the circular range accumulated rainfall, compare the correlation coefficient between the water level volume change and the accumulated rainfall in different ranges to further determine the spatial range of rainfall accumulation;

[0051] S300, preliminary model fitting:

[0052] S300-1, multiple rounds of principal component analysis were performed on rainfall data at different distance ranges;

[0053] S300-2, fitting the cumulative rainfall and water level volume variables synthesized by the principal components of each round, and comparing them with the fitting results of the cumulative rainfall and water level volume variables in the circular area;

[0054] S400, sample screening: screening by water level volume change to exclude samples of water body change caused by leakage and increased emission as the main factors;

[0055] The samples of the whole year were screened in multiple groups according to different water levels, and the screened samples were subjected to correlation analysis; the correlation and fitting between the water level volume change and the rainfall accumulation in different ranges in different screened data sets were calculated;

[0056] S500, rainfall accumulation duration comparison: Analyze and compare the correlation between the accumulated rainfall of different accumulation durations and the change in water level volume, and determine the rainfall accumulation duration used in the model;

[0057] S600, model and fitting results: Compare the fitting results of water level screening only and water level + volume change screening;

[0058] S600-1, comparison of linear, second-order polynomial, and third-order polynomial fitting effects;

[0059] S600-2, comparison of the fitting effect of nonlinear neural network method;

[0060] S600-3, using the 24-hour rainfall accumulation time and water level volume change after water level screening 10 9 The above sample subsets are fitted and modeled, and the calculation formula is:

[0061] Formula 3: ΔV = b × P + c;

[0062] Where ΔV is the change in water level volume, in m 3 ; P is the 24-hour cumulative rainfall within the circular range of the basin, in m 3 ; c and b are the constant term and coefficient respectively;

[0063] 1.4 The specific steps for flood range prediction are:

[0064] S1000, calculation of accumulated rainfall in the next 24 hours:

[0065] S1000-1, download the next 24 hours forecast rainfall data in raster format;

[0066] S1000-2, create a rainfall accumulation range file, extract the part within the basin, and obtain the cumulative range file of the forecast rainfall;

[0067] S1000-3, project the accumulated rainfall, the horizontal unit of the projection result is m; then use the rainfall accumulation range file to extract the rainfall file after projection, and extract the rainfall within the accumulation range; finally, extract the rainfall file.

[0068] The extracted rainfall file is used for surface volume calculation;

[0069] S2000, water level forecast:

[0070] S2000-1, calculate the volume change of water bodies based on the accumulated rainfall, and substitute the accumulated value of the 24-hour forecast rainfall into the formula

[0071] 3. Calculate and obtain the predicted water level volume change ΔV;

[0072] S2000-2, obtain the current water level data measurement value E, substitute the current water level value into formula 1 to obtain the current water level corresponding

[0073] The volume of lake water V e , and obtain the predicted lake volume V′ based on the predicted water level volume change e , the formula is:

[0074] Formula 4: V′ e =V e +ΔV;

[0075] Rewrite formula 1 to obtain V′ e The corresponding predicted water level E′ is:

[0076] Formula 5: E′=n+(V′ e -V n ) / (V n+1 -V n );

[0077] Among them, V n is the volume of lake water when the water level is an integer n, V n+1 When the water level is an integer n+1,

[0078] The volume of the lake, in the calculation formula, first find the water level n, so that V n+1 <V′ e <V n+1 ;

[0079] S3000, flooding range prediction:

[0080] The prepared inundation map is interpolated to make it consistent with the accuracy of the measured water level data.

[0081] The plane resolution of ERA5 in S10 is 0.1°, which is equivalent to 10 km.

[0082] The multi-round principal component analysis in S300-1 includes four rounds:

[0083] In the first round, the cumulative rainfall in the annular area within 180 km was used as the factor;

[0084] The second round uses the cumulative rainfall in the circular area of ​​all distances as a factor;

[0085] The third round uses the accumulated rainfall in a circular area between 90 and 180 km as the factor;

[0086] The fourth round uses the accumulated rainfall in a circular area between 120 and 180 km as a factor.

[0087] c and b in S600-3 are 2062612888.591 and 0.100611 respectively.

[0088] Compared with the existing technology, the present invention conducts inundation analysis based on DEM data, statistics the relationship between water level-inundation area and water level-inundation volume, and constructs a flood inundation prediction model in which the inundation area changes with the water level. On this basis, the principle of equal total volume of water is used to develop a rapid prediction method for the inundation range based on total rainfall data, and further delineate the scope of the flood monitoring area. BRIEF DESCRIPTION OF THE DRAWINGS

[0089] Figure 1 Schematic diagram of the overall method of the present invention.

[0090] Figure 2 Schematic diagram of the active flooding algorithm flow of the present invention.

[0091] Figure 3 This is a comparison diagram of the active and passive submerged plane effects of the present invention.

[0092] Figure 4 This is a comparison diagram of the water level-area and water level-volume curves of active and passive flooding of the present invention.

[0093] Figure 5 This is a curve diagram of water level changes at various measuring stations in the Poyang Lake area in 2020 in an embodiment of the present invention.

[0094] Figure 6 Schematic diagram of the water level-flooded volume statistical area in an embodiment of the present invention.

[0095] Figure 7 Schematic diagram of water level-volume fitting in an embodiment of the present invention.

[0096] Figure 8 Schematic diagram of ERA5 dataset selection in an embodiment of the present invention.

[0097] Figure 9 Schematic diagram of the spatial range of rainfall accumulation in an embodiment of the present invention.

[0098] Figure 10 This is a time series curve diagram of water level volume change and daily cumulative rainfall in an embodiment of the present invention.

[0099] Figure 11 This is a comparison chart of the bivariate correlation analysis between the water level volume change and the accumulated rainfall in the embodiment of the present invention.

[0100] Figure 12 Schematic diagram of time series analysis of water level volume change and rainfall in different ranges in an embodiment of the present invention.

[0101] Figure 13 Schematic diagram of the correlation coefficient between the water level volume change and the accumulated rainfall in different ranges in an embodiment of the present invention.

[0102] Figure 14 Schematic diagram of the fitting results of the main components, 120km and 180km ranges in an embodiment of the present invention.

[0103] Figure 15 Schematic diagram of the fitting results after screening of the water level volume change in an embodiment of the present invention.

[0104] Figure 16 Schematic diagram of the change of correlation coefficient after water level screening in an embodiment of the present invention.

[0105] Figure 17 is the fitting R of different screening data sets in the embodiment of the present invention 2 And the correlation coefficient comparison chart.

[0106] Figure 18 This is a correlation analysis diagram between volume change and accumulated rainfall under different rainfall accumulation durations in an embodiment of the present invention.

[0107] Figure 19 Schematic diagram of different screening and fitting comparisons in the embodiments of the present invention.

[0108] Figure 20 Schematic diagram comparing different curve fitting effects in the embodiments of the present invention.

[0109] Figure 21 Schematic diagram of the fitting effect of the neural network method in an embodiment of the present invention.

[0110] Figure 22 Schematic diagram of the calculation result of the cumulative forecast rainfall in an embodiment of the present invention.

[0111] Figure 23 Schematic diagram of active flooding map and predicted flooding range in an embodiment of the present invention. DETAILED DESCRIPTION

[0112] The present invention will now be further described with reference to the accompanying drawings.

[0113] See also Figures 1 to 23The present invention provides a method for flood inundation prediction based on rainfall. Taking the Poyang Lake area as an example, the invention conducts inundation analysis based on DEM data, statistically analyzes the relationship between water level-inundated area and water level-inundated volume, and constructs a flood inundation prediction model in which the inundated area changes with the water level. On this basis, the principle of total volume equality of water volume is used to develop a rapid inundation range prediction method based on total rainfall data, and further delineate the scope of the flood monitoring area. The detailed process is as follows: Figure 1 shown.

[0114] 1. Analysis of lake water volume changes:

[0115] Without considering factors such as the discharge of Poyang Lake into the Yangtze River, the exchange of lake water with groundwater, evapotranspiration, interception by water conservancy facilities, and water withdrawal for domestic and agricultural use, it would be reasonable to assume that the increase in lake water volume corresponding to the rainfall volume and water level rise is exactly the same. However, in reality, these factors have a significant impact on the lake level.

[0116] Therefore, the prediction still needs to model the relationship between rainfall and lake volume growth caused by water level changes, quantify the impact of factors other than rainfall, and quantitatively evaluate the reliability of predicting lake volume based on rainfall.

[0117] Submergence analysis

[0118] Inundation analysis is one of the important contents of hydrological analysis in GIS spatial analysis.

[0119] Currently, there are two main types of inundation analysis: active inundation and passive inundation. Passive inundation disregards the source of inundation and assumes that all water comes from rainfall, resulting in inundation caused by a flat, planar water source. This directly uses the DEM (Demonstration and Inundation Analysis) algorithm, disregarding the path of inundation expansion. Active inundation uses a specific water source for inundation. For low-lying areas enclosed by high ground, inundation occurs only when the inundation water level exceeds the edge of the low-lying area.

[0120] Passive inundation can be calculated simply by counting the elevation distribution in a DEM elevation model, making it a type of inundation analysis that can be easily implemented in most GIS software. Active inundation, on the other hand, requires setting a spreading seed and applying a spreading algorithm. However, the lack of mature commercial software makes it difficult to apply.

[0121] Traditional active flooding methods set a target flooding level and a source point. Starting from the source point, they continuously traverse eight neighborhoods to determine the flooding state, ultimately determining the flooding range at the target water level. This process must be repeated each time the target water level is set, making it unable to meet the rapid prediction requirements of emergency applications.

[0122] Therefore, in view of the particularity of the lake area, this embodiment proposes the concept of a flooding map. Any point that is still submerged during the dry season is selected as the source point in the lake area, and with a step size of 1 m, the minimum flooding water level corresponding to each grid point is calculated to form a flooding map.

[0123] In the obtained flooding map, the meaning of the elevation Ecell of a certain grid point cell is that when the water level of the seed point exceeds Ecell, the current cell will be submerged. Based on the flooding map, when the determined flooding water level elevation is E, all cells on the flooding map where Ecell < E are set as the flooded area, and thus the flooded range at the target water level can be obtained.

[0124] The production of the flooding map for flood range prediction can be carried out for the entire basin or a local area, and it is only necessary to select a point with a relatively low elevation within the water area as the source point. For the flooding map used to statistically analyze the relationship between flooding water level - area - volume, it is necessary to select a point with a relatively low elevation within the water area of the dry - season flooding range of the entire basin as the source point, and it should include the maximum lake area range during the wet season. For example, in the case of Poyang Lake, the center position of the Hukou Waterway can be selected as the source point.

[0125] As Figure 2 shown, the algorithm flow of the flooding map is as follows:

[0126] 1. Create an empty flooding map.

[0127] According to the input DEM grid file, an empty file with the same grid information is generated as the flooding map file;

[0128] The flooding map grid file has the same row and column values, four - corner coordinates, grid size, coordinate system, etc. as the input DEM grid file, but all grid values are empty.

[0129] 2. Process the input source point.

[0130] According to the input source - point coordinates, calculate its corresponding grid on the DEM, obtain its row number r, column number c, and elevation value e;

[0131] Create a source - point list to store the source points. The source - point type is a tuple, that is, the source - point list is written as [(r, c), ……], where r is the row number and c is the column number;

[0132] Store the row and column numbers of the source - point grid as the first source point in the source - point list;

[0133] Define the current flooding elevation E and assign it the value of e as the flooding elevation for the first step of flooding.

[0134] 3. Taking E as the flooding elevation, according to the source - point list, complete the active flooding expansion and spread algorithm at the E water level and update the flooding map:

[0135] Create a new source point deletion list to store source point information that is no longer located at the boundary and needs to be deleted from the source point list.

[0136] Traverse the source point list. For the current source point, take the source point's eight neighboring rasters to form a neighborhood list; set the boundary flag and initialize it to False.

[0137] Traverse the neighborhood list, for the current neighborhood point:

[0138] Determine whether the value of the neighboring point in the flooding map is empty. If the value is not empty, it means that the point has been flooded and the current neighboring point will not be processed; if the value is empty, proceed to the next step;

[0139] Determine the elevation value e' of the neighboring point. If e'>E, the neighboring point is not submerged and is not processed. The boundary flag of the current source point is set to True. If e'<=E, the point is submerged and the neighboring point is added to the source point list. The grid value of the point in the submerged map is updated to E.

[0140] After traversing all neighboring points, if the boundary point flag is still False, it means that all neighbors of the source point have been submerged and are no longer on the boundary. The source point is added to the source point deletion list.

[0141] After traversing all source points, delete the source points recorded in the source point deletion list from the source point list; the flooded elevation increases by one step, E=E+E step ; If the source point list is not empty, repeat step 3; otherwise, the inundation map calculation is completed and exit the program.

[0142] The flooding map calculated using the active flooding algorithm is as follows Figure 3 As shown in the red oval box, the active flooding removes the dark red local pits.

[0143] Based on the inundation map, the "water level-inundated area relationship" and "water level-inundated volume relationship" can be obtained respectively, of which the latter is an important conversion basis for linking rainfall and inundation range.

[0144] For passive inundation, the water level-inundation volume relationship can be obtained by counting the DEM volume below the water level. For active inundation, it is necessary to first determine the inundation area under a specific water level through the inundation map, and then further determine the inundation volume through the original DEM. Figure 4As shown in the figure, the dotted line represents passive flooding, and the solid line represents active flooding. The difference in the submerged area between active and passive flooding is large at low water levels, but not obvious above 10m. The volume change between active and passive flooding is not large at all, and the difference is even smaller above 10m. The water level in the Poyang Lake area is above 10m most of the time, and above 15m during flood season (such as Figure 5 shown).

[0145] When the effect difference is not significant, this embodiment selects the passive submergence method which is more convenient to calculate and constructs the water level-submergence volume model.

[0146] Measured water level data

[0147] In order to predict floods caused by heavy rains, the year 2020, when the water level of Poyang Lake was relatively high, was selected for modeling. The water levels of relevant stations in the lake area of ​​'Hydrological Yearbook of the People's Republic of China - Volume 6 (2020), Volume 17, Volume 6 (2020), No. 17-Hydrological Data of the Yangtze River Basin, Poyang Lake Area (Ganjiang River System)' were collected for modeling. Poyang Lake is huge and there are many water level measuring stations. This embodiment selected the daily average water level data of Hukou Waterway-Hukou Station, Pingfeng (II) Station, Xingzi Station; Poyang Lake-Duchang Station, Kangshan Station, Longkou Station, Nanfeng Station, Tangyin (II) Station and other stations in 2020 for modeling. The water level curves of each station are as follows: Figure 5 shown.

[0148] During low water periods, water levels at the eight stations in the lake area and the Hukou waterway vary significantly. During the flood season from July to November, when water levels exceed 15 meters, water level differences between stations are less pronounced. The average water level at Hukou and Xingzi stations in the Hukou waterway is relatively consistent, indicating minimal impact from water conservancy facilities. Therefore, the average water level of these two stations was selected as the modeling water level.

[0149] Water level-volume conversion model

[0150] Since the difference between the volume of passive submergence and active submergence is very small, the calculation time cost of active submergence is too high and the effect is very small. Therefore, when calculating the water level-submergence volume, the idea of ​​passive submergence is still adopted, and DEM is directly used to calculate the relationship between water level and submergence volume.

[0151] The statistical scope of water level-flooded volume must at least cover the entire lake area. The coverage of the statistical area is as follows: Figure 5 The statistical results of the water level-submerged volume are shown in Table 1 below. The 2020 water level fluctuation range and the high flood season predicted in this embodiment are combined, and a certain range is used as the basis for the water level-submerged volume conversion.

[0152] Table 1 Water level-area-volume statistics based on DEM passive submergence

[0153]

[0154]

[0155] The water level-volume statistical results were fitted, and the third-order polynomial fitting result was the best, such as Figure 7 As shown, the fitted R 2 The value is almost close to 1 ( Figure 7 The reason is that the relationship between water level and volume should be infinitely close to the volume of the column if the influence of terrain is deducted. Although the fitting effect is very good, there are still residuals in the fitting. The absolute value of the residual is generally 1-2 orders of magnitude smaller than the measured rainfall data ( Figure 7 b) in the above example.

[0156] In order to better preserve the influence of terrain, a direct water level-volume conversion method is considered. Based on Table 1 above, for the target water level e, the corresponding volume is calculated as follows:

[0157] Formula 1: V e =V n +(V n+1 -V n )×(en);

[0158] Where: n is the integer part of the water level e, V n is the submerged volume corresponding to the water level n found in Table 1, V n+1 is the submerged volume corresponding to the water level n+1 found in Table 1.

[0159] The direct conversion method uses the flooded volume corresponding to the integer part and, according to a linear model, controls the volume values ​​within the interval. This ensures that the predictions for integer water levels are consistent with the statistical results and that the predictions for decimal water levels do not conflict with the statistical results. The resulting prediction curve is equivalent to the curve obtained by directly connecting the measurement points.

[0160] Water level volume change:

[0161] The water volume change corresponding to the water level (hereinafter referred to as the water level volume change) is obtained by converting the water level to the submerged volume based on the DEM in the previous section. The daily lake volume V is obtained by converting the daily water level measured data. n The change in water level volume on day n is △V n :

[0162] Formula 2: ΔV n =V n+1 -V n ;

[0163] Therefore, the water level volume variable is actually the change in water volume after the next 24 hours.

[0164] 2. Historical rainfall accumulation analysis:

[0165] Rainfall data collection

[0166] The rainfall data uses the ERA5 hourly rainfall data, which comes from the ERA5 land hourly dataset since 1950 ( Figure 8 ERA5 has a 0.1° resolution, which translates to approximately 10 km in this area, with an average grid area of ​​109,664,196 square meters. When downloading data from this website, batch downloads using Python are often rejected, while web-based orders are more likely to be approved. After submitting an order, there's a wait time of approximately two hours, but all orders can be submitted at once. Download hourly rainfall data for the entire year of 2020.

[0167] Rainfall data regularization

[0168] The temporal and spatial range of rainfall accumulation is a key parameter in the modeling of this embodiment. Chen Sanming et al. (2014) directly used three days of rainfall from the Statistical Yearbook for their calculations, with a rain intake area of ​​15 km². However, the paper did not provide any experimental evidence to confirm the appropriateness of this choice.

[0169] The rainfall data obtained are hourly rainfall in m. Since the water level data can only obtain the daily average rainfall, the rainfall must first be normalized into 24-hour cumulative rainfall.

[0170] In terms of space, the location of the Hukou waterway near the Yangtze River is taken as the center point (near the Xingzi, Hukou and other hydrological stations), and the rainfall data are regularized by distance, and the data outside the basin are cut off. The distance partition of the rainfall data sampling points is as follows: Figure 9 shown.

[0171] Historical accumulated rainfall

[0172] After the flood season began in July 2020, the water level of Poyang Lake quickly reached its highest point, which is also the typical situation to be predicted in this embodiment. Therefore, the water level data and rainfall data for July are used as the embodiment object.

[0173] The starting radius and step size parameters for rainfall statistics covering the target area are primarily used to batch generate rainfall amounts in different ranges. These rainfall amounts are then compared and selected as the most appropriate in subsequent steps. Therefore, the starting radius and step size are not fixed and are gradually explored based on the study area.

[0174] For example, we first take the radius of the target area as the starting point, and use a certain distance as the step length to count rainfall data, and then preliminarily determine the rainfall impact radius range; then we use the minimum radius of the determined rainfall impact range as the starting point, and use a smaller distance as the step length to count rainfall data, so as to more accurately determine the rainfall impact radius. Rainfall data is accumulated from 0:00 to 24:00 on the same day. In terms of space, starting from the main range of the lake area with a radius of 60km, with a step length of 30km, we calculate the daily cumulative rainfall data of the annular areas of 60, 90, 120, 150, 180, 210, 240, and 270 from the center point of the lake mouth. Figure 10 ).from Figure 10 From the time series curve, the change in water volume corresponding to the water level is basically synchronized with the change in accumulated rainfall.

[0175] 3. Lake Volume Change Prediction Model

[0176] Predictive model variable selection

[0177] Predictions based on rainfall should not predict absolute water levels. Absolute water level changes in lakes are often influenced by the broader climate and fall under the purview of time-series forecasting. All water level forecasts should predict incremental water levels. This means simulating the increase in water levels after a given rainfall, given a known initial water level.

[0178] If equal volumes are used, then the outflow of lake water must also be considered, as well as the water consumption due to infiltration, evaporation, and groundwater exchange. However, in actual emergency water level forecasts, the only available data is rainfall, and timely data on outflow and evaporation are not readily available. Therefore, this model, based on rainfall data, comprehensively considers the overall impact of other factors, such as outflow, evaporation, and groundwater exchange, by constructing a statistical relationship between measured water level changes and accumulated rainfall.

[0179] Variable correlation analysis

[0180] By accumulating the annular area, the cumulative rainfall variable of the circular area is constructed. The results of the bivariate correlation analysis show that the cumulative rainfall in the annular range has a correlation coefficient of more than 0.8 within a range of 180 km. Figure 11 a); the correlation coefficients of the accumulated rainfall in the circular area are all over 0.8 ( Figure 11 b in the figure). The comparison between the two shows that ( Figure 11 In (c), the correlation between the cumulative rainfall in the circular range and the water level volume change is higher than that in the annular range with the same cumulative distance, and the correlation coefficient in the 120 km range reaches the highest value.

[0181] Taking into account the time factor, the time series characteristics of the water level and volume changes and the accumulated rainfall at different distances are analyzed. The analysis results are as follows: Figure 12 As shown in the figure. Regardless of distance, the most relevant factor for water level volume change is still the rainfall on that day. While there is still a significant correlation with rainfall before and after the day, the correlation is significantly weaker than on that day. The significant correlation with rainfall after the day may be due to the fact that rainfall processes generally last for several days, not just a short time. Therefore, the water level volume change is primarily predicted based on the rainfall on that day.

[0182] In order to further determine the appropriate spatial range of rainfall accumulation, based on the circular range of rainfall accumulation, starting from 120km, the step size was reduced to 10km, and the correlation coefficient between the water level volume change and the accumulated rainfall in different ranges was compared. The results show that ( Figure 13 ), the highest correlation coefficient is still the cumulative rainfall within a range of 120 km.

[0183] Preliminary model fitting

[0184] According to the correlation results, the cumulative rainfall in different distance ranges has a significant correlation with the water level volume variable, and the correlation coefficient is as high as above 0.8.

[0185] First, principal component analysis was performed on rainfall data at different distance ranges. Four rounds of principal component analysis were conducted, and those with an explanatory power greater than 1.0 had only one principal component. The first round used the cumulative rainfall in the annular area within 180 km as the factor, and the second round used the cumulative rainfall in the annular area within 180 km as the factor. Figure 13 The cumulative rainfall in the circular area of ​​all distances shown in the figure is used as the factor, the third round uses the cumulative rainfall in the circular area between 90-180km as the factor, and the fourth round uses the cumulative rainfall in the circular area between 120-180km as the factor.

[0186] The cumulative rainfall and water level volume variables synthesized from the principal components of each round were fitted and compared with the fitting results of the cumulative rainfall and water level volume variables in the circular areas of 120km and 180km.

[0187] The fitting results show that principal component analysis is not very meaningful, and the optimal cumulative distance can be directly selected for analysis.

[0188] The fitting results show that the best linear fitting result is the cumulative rainfall in the circular area within 120 km, and the best quadratic polynomial fitting result is the principal component analysis result of the cumulative rainfall in the annular area within 180 km. However, the fitting R 2 Therefore, the subsequent analysis will no longer consider the principal component analysis, and directly use the cumulative rainfall in the circular area for fitting.

[0189] Although the current fitting R2 The performance is acceptable, but most of the points with more water are seriously underestimated, which increases the fitting R 2 The points are mostly concentrated in the lower part, where the water level is dropping. The primary cause of the water level drop is clearly not rainfall, but rather the discharge of lake water into the Yangtze River, evaporation, and infiltration in the absence of rainfall. Therefore, using rainfall to simulate this situation is not reasonable. During periods of low rainfall, changes in water level and volume are more dispersed, with most samples showing a drop in water level, while a few samples show a significant rise. This may be related to the varying flood storage and discharge conditions of upstream water conservancy facilities.

[0190] Therefore, the data should be filtered to exclude the period of low water levels before refitting. If only July data is used, the sample size may be too small. Therefore, the full year data should be reprocessed to test the sample filtering rules.

[0191] Sample screening

[0192] By screening the water level volume change, we can eliminate water body change samples caused by discharge and increased emission as the main factors. The accumulated rainfall within a 120km circular range is selected and calculated for the annual samples.

[0193] First, consider filtering the water level volume change value. The water level volume change is above 0 and 10 9 The fitting results after the above screening are as follows Figure 14 As shown. The fitting R above 0 2 All of them are less than 0.5, the three-time fitting is better, and the fitting R 2 It is only 0.484. The fitting effect above 109 has improved, and the R2 of the cubic fitting reaches 0.627.

[0194] During high water levels, the possibility of large amounts of water being stored in upstream and downstream water conservancy facilities is eliminated, and the impact of rainfall on the water level in the entire lake area can be fully reflected. Figure 15 Judging from the consistency of water level data at various hydrological stations in 2020, the water level is above 15m, and the water levels at different hydrological stations throughout the lake area are basically uniform.

[0195] Therefore, three groups of samples were screened for the whole year, namely water levels above 15m, above 18m, and above 20m. Correlation analysis was performed on the screened samples, and the analysis results are as follows: Figure 16 ,Water level-based screening can effectively improve the correlation coefficient. The higher the starting water level, the greater the correlation coefficient.

[0196] Correlation and fitting R between water level volume change and rainfall accumulation in different ranges in different screening data sets 2 Calculations are performed and the results are compared. Figure 17For different digitally screened sample subsets, the correlation coefficient and fitting R between water level volume change and cumulative rainfall in different ranges are 2 The trend of change is basically the same. The rainfall accumulation range with good fitting for the subset above 15m water level is 140-170km, while the fitting effect for the subset above 18m water level gradually deteriorates with increasing distance. The rainfall accumulation range with good fitting for the subset above 20m water level is 150-190km. The R2 of the fitting of 15m is generally low, while the fitting of 20m water level and above is better with increasing distance. This phenomenon may be due to the deviation caused by too few sample points (see Figure 17 ,There are only 34 samples above 20 m). In summary, the screening subset with water level above 18 m and water level volume change above 109 was finally selected, and the rainfall accumulation range of 120 km was used.

[0197] Comparison of rainfall accumulation duration

[0198] Time series analysis of water level volume change and rainfall in different ranges ( Figure 12 ) shows that the rainfall of the previous day also has a high correlation with the change in water level volume. Therefore, we further analyze and compare the correlation between the accumulated rainfall of different accumulation periods and the change in water level volume to determine the rainfall accumulation period used in the model. The correlation statistics show that ( Figure 18 ), the cumulative rainfall in one day has the highest correlation with the change in water volume, so 24 hours is selected as the rainfall accumulation time for water level prediction in the model.

[0199] Model and fitting effect

[0200] Compare the fitting results of water level screening only and water level + volume change screening, such as Figure 19 If only water level screening is used, the fitted R 2 The large value is due to the fact that a large number of samples with decreasing water levels are concentrated, which increases the goodness of fit. However, these samples are of less interest in emergency observations. Due to the presence of these samples, the fitted model underestimates the majority of samples with increasing water levels, which are precisely the cases of greater concern in emergency observations.

[0201] Comparing the effects of linear, second-order polynomial, and third-order polynomial fitting, R 2 The values ​​show that the third-order polynomial has a higher degree of fit, but from the curve shape, the linear fit is more reasonable.

[0202] In addition, the fitting effect of the nonlinear neural network method is compared. Taking the whole year sample as the embodiment object, adding the water level value as a covariate, the R 2 The value is 0.596958442, and the R of the cubic fit under the same sample set is 2The value is 0.59672262, and the improvement effect is limited. The R value of the neural network method fitting the 18m water level screening sample set is 0.59672262, and the improvement effect is limited. 2 The value is 0.666372999, and the R of the cubic fit is 2 The R of the straight line fitting is 0.733077365. 2 It is 0.717888969. It can be seen that compared with the annual sample set, the fitting relationship between water level volume and rainfall in the sample set after high water level screening is closer to a linear relationship.

[0203] Finally, we use the water level screening (above 18m), the cumulative rainfall time in 24 hours, the water level volume change 10 9 The above sample subset is fitted and modeled. The fitted model is recorded as:

[0204] Formula 3: ΔV = b × P + c;

[0205] Where △V is the volume change of water level, unit is m 3 , P is the 24-hour cumulative rainfall within a circular area with a radius of 120 km in the basin (the next 24 hours' forecast rainfall can be used in the prediction), unit is m 3 , constant term c = 2062612888.591, coefficient b = 0.100611.

[0206] 4. Prediction of flooding range based on rainfall:

[0207] Based on the above model, when the rainy season comes and the water level rises, the emergency workflow for predicting the inundation range includes calculation of the accumulated rainfall in the next 24 hours, water level prediction, and inundation range prediction.

[0208] Calculation of accumulated rainfall in the next 24 hours

[0209] First, download the data. After downloading, the rainfall data is converted to raster format. Next, create a rainfall accumulation range mask file. Centered on the Hukou waterway and the location near the Yangtze River, with a radius of 120 km, extract the basin area and obtain the cumulative rainfall range (shp file).

[0210] The forecast rainfall accumulation calculation is completed based on arcpy. First, the accumulated rainfall is projected, and the horizontal unit of the projection result is m. Then, the mask file is used to extract the projected rainfall file and extract the rainfall raster within the 120km accumulation range. Finally, the surface volume of the extracted rainfall file is calculated, and the reference_plane parameter is set to "ABOVE" and the base_z parameter is set to 0.0. The statistical accumulated rainfall volume value is stored in the specified txt file, such as Figure 22As shown in the figure in the red box. The above function is implemented in flood_source.Ac24Rainfall().

[0211] Water level forecast

[0212] First, prepare the current water level data. On the website http: / / xxfb.mwr.cn / sq_dtcx.html?v=1.0, you can use the map search function to enter "Xingzi" to view the water level information at the lake mouth waterway and obtain the current water level value E.

[0213] Then, the volume change of the water body is calculated based on the accumulated rainfall. Substitute the 24-hour forecast rainfall accumulated value into formula (3) to calculate the predicted water level volume change △V.

[0214] Finally, the water level value after 24 hours is predicted. Substitute the current water level value into formula (1) to obtain the lake volume V corresponding to the current water level. e ; and obtain the predicted lake volume V′ based on the predicted water level volume change e , and then obtain the predicted lake volume V after 24 hours e ':

[0215] Formula 4: V′ e =V e +ΔV

[0216] Finally, rewrite formula (1) to obtain V e The corresponding predicted water level E' is:

[0217] Formula 5: E′=n+(V′ e -V n ) / (V n+1 -V n );

[0218] Where: V n is the volume of lake water when the water level is an integer n, V n+1 is the volume of the lake when the water level is an integer n+1. When calculating, we need to first find the water level n so that V n+1 <V e ' <V n+1 .

[0219] The above functionality is implemented in flood_source.predictWaterLevel() . Selecting the water level on July 6, 2020, and the historical cumulative rainfall for that day, the predicted water level for the next day is 20.045, while the actual value is 20.005. The water level prediction error is not large.

[0220] Flooding range prediction

[0221] Satellite-derived DEM data typically contains elevation fields as integers in meters, and the data's reliable accuracy typically struggles to reach 1 meter. However, water level fluctuations within a single day are often much lower than the meter level. Measured water level data is reported in meters, with three decimal places. Therefore, inundation maps must be interpolated to ensure they match the accuracy of measured water level data.

[0222] If dem is interpolated before the inundation map is created, the computation time will be too long if the inundation step size is set to 0.001 when creating an inundation map with active inundation. Therefore, the interpolation process is performed on the created inundation map.

[0223] When predicting inundation based on water levels, there's no need to use an existing inundation map. In ArcGIS, use the Raster to TIN tool to convert the DEM into a terrain model. Set the Z Tolerance to 1 to ensure the maximum elevation difference between the input raster and the output TIN model does not exceed 1 meter. Then, use TIN to Raster to convert the TIN model into raster data.

[0224] In conjunction with the test data of other parts of the project, the water level on July 16, 2020 (measured water level) was selected, and the water coverage area of ​​the local area of ​​Poyang Lake on that day was predicted based on the inundation map of active inundation after interpolation. The prediction results are as follows Figure 23 shown.

[0225] The above are merely preferred embodiments of the present invention and are intended to help understand the method and core concepts of this application. The scope of protection of the present invention is not limited to the above embodiments. All technical solutions based on the principles of the present invention are within the scope of protection of the present invention. It should be noted that improvements and modifications that do not depart from the principles of the present invention, which are within the scope of protection of the present invention, are within the scope of protection of the present invention.

[0226] The present invention comprehensively solves the lack of flood prediction methods in the existing technology in the case of passive inundation processes such as water accumulation in low-lying areas due to high-intensity and large-scale rainfall. It conducts inundation analysis based on DEM data, statistically analyzes the relationship between water level-inundation area and water level-inundation volume, and constructs a flood inundation prediction model in which the inundation area changes with the water level. On this basis, it uses the principle of equal total volume of water to develop a rapid prediction method for the inundation range based on total rainfall data, and further delineates the scope of the flood monitoring area.

Claims

1. A method for flood inundation prediction based on rainfall, characterized in that: include: 1.1 Analysis of lake water volume changes; 1.2 Analysis of historical rainfall accumulation; 1.3 Lake volume change prediction model; 1.4 Prediction of flooding range; The specific steps of the analysis of lake water volume changes in 1.1 are as follows: S1, Submergence analysis: Active flooding: S1-1, create a new empty inundation map: Based on the input DEM raster file, generate an empty file with the same raster information as the inundation map file. The inundation map raster file has the same row and column values, four corner coordinates, grid size, and coordinate system as the input DEM raster file, and all grid values ​​are empty; S1-2, processing input source points: S1-2-1, based on the input source point coordinates, calculate its corresponding grid on the DEM, and take its row number r, column number c, and elevation value e; S1-2-2, create a new source point list for storing source points. The source point type is a tuple, that is, the source point list is written as [(r, c), ...], where r is the row number and c is the column number; S1-2-3, store the row and column numbers of the source point grid as the first source point in the source point list; S1-2-4, define the current flooding elevation E and assign it a value of e as the flooding elevation of the first step of flooding; S1-3, taking E as the inundation elevation, complete the active inundation expansion and spread algorithm at water level E according to the source point list, and update the inundation map: S1-3-1, create a new source point deletion list: used to store source point information that is no longer located at the boundary and needs to be deleted from the source point list; S1-3-2, traverse the source point list: for the current source point, take the 8 neighboring grids of the source point to form a neighbor list; Set the boundary flag and initialize it to False; S1-3-3, traverse the neighborhood list, for the current neighborhood point, determine whether the value of the neighborhood point in the flood map is empty; if the value is not empty, it means that the point has been flooded, and the current neighborhood point is not processed; if the value is empty, proceed to the next step; determine the elevation value e' of the neighborhood point, if e'>E, if the neighborhood point is not flooded, do not process the neighborhood point, and set the boundary flag of the current source point to True; if e'≤E, the point is flooded, add the neighborhood point to the source point list, and update the grid value of the point in the flood map to E; S1-3-4, after traversing all neighboring points, if the boundary point flag is still False, it means that all neighbors of the source point have been submerged and are no longer on the boundary. The source point is added to the source point deletion list; S1-3-5, after traversing all source points, delete the source points recorded in the source point deletion list from the source point list; the flooded elevation increases by one step, that is, E=E+E step If the source point list is not empty, repeat the steps S1-3; otherwise, the flood map calculation is completed and the program exits. S1-3-6, using the inundation map algorithm, obtain the "water level-inundation area relationship" and the "water level-inundation volume relationship". The "water level-inundation volume relationship" is used as a basis for converting rainfall to inundation range; Passive inundation: By counting the DEM volume below the water level, the "water level-submerged volume relationship" in S1-3-6 is obtained; S2, measured water level data: By collecting the water level data of each station in the existing relevant hydrological data, and using the closest water level data between different stations as the modeled water level; S3, water level-volume conversion model: S3-1: Use the DEM model to calculate the relationship between water level and submerged volume. The statistical area covers the entire water area, and the data from historical water level changes and a certain period during the high flood season are used as the basis for conversion. S3-2, fitting the water level-volume statistical results: The results in S3-1 are fitted and the water level-volume direct conversion method is used to retain the volume of the target water level e under the influence of the terrain. The calculation formula is: Formula 1: V e =V n +(V n+1 -V n )×(en); Where n is the integer part of the water level e, V n V is the submerged volume corresponding to the queried water level n, n+1 is the submerged volume corresponding to the queried water level n+1; S4, water level volume change: The water level-submerged volume conversion based on DEM is used to calculate the water level volume change. The daily lake volume V is obtained by conversion based on the daily water level measured data. n , the change in water level volume on day n is ΔV n , and its calculation formula is: Formula 2: ΔV n =V n+1 -V n ;The water level volume variable is the change in water volume in the next 24 hours; The specific steps of the historical rainfall accumulation analysis in 1.2 are as follows: S10: Rainfall data collection: Collect hourly rainfall data sets over land areas over the years through ERA5; S20: Rainfall data normalization: Normalize hourly rainfall into 24-hour cumulative rainfall, normalize rainfall data by distance and trim data outside the basin; S30: Historical cumulative rainfall: The highest historical water level point is used as the statistical object, and the rainfall data is accumulated from 0:00 to 24:

00. The spatial range is based on the preset radius of the target area covered by the main body as the starting point. According to the preset step size, the daily cumulative rainfall data of the annular area around the main body center is calculated. The specific steps of the lake volume change prediction model in 1.3 are as follows: S100, prediction model variable selection: Based on rainfall data, a statistical relationship model between measured water level changes and accumulated rainfall is constructed to comprehensively consider the overall impact of outflow, evaporation, and groundwater exchange factors; S200, variable correlation analysis: S200-1, by accumulating the annular area, the cumulative variable of the circular area rainfall is constructed, and the bivariate analysis of the double annular range and the circular range is performed; S200-2, taking into account the time factor, conducts a time series characteristic analysis of the water level and volume changes and the accumulated rainfall at different distances; S200-3, based on the circular range accumulated rainfall, compare the correlation coefficient between the water level volume change and the accumulated rainfall in different ranges to further determine the spatial range of rainfall accumulation; S300, preliminary model fitting: S300-1, multiple rounds of principal component analysis were performed on rainfall data at different distance ranges; S300-2, fitting the cumulative rainfall and water level volume variables synthesized by the principal components of each round, and comparing them with the fitting results of the cumulative rainfall and water level volume variables in the circular area; S400, sample screening: screening by water level volume change to exclude samples of water body change caused by leakage and increased emission as the main factors; The samples throughout the year were screened in multiple groups according to different water levels, and the screened samples were subjected to correlation analysis; The correlation and fitting between the water level volume change and rainfall accumulation in different ranges in different filtered data sets were calculated; S500, rainfall accumulation duration comparison: Analyze and compare the correlation between the accumulated rainfall of different accumulation durations and the change in water level volume, and determine the rainfall accumulation duration used in the model; S600, model and fitting results: Compare the fitting results of water level screening only and water level + volume change screening; S600-1, comparison of linear, second-order polynomial, and third-order polynomial fitting effects; S600-2, comparison of the fitting effect of nonlinear neural network method; S600-3, using the 24-hour rainfall accumulation time and water level volume change after water level screening 10 9 The above sample subsets are fitted and modeled, and the calculation formula is: Formula 3: ΔV = b × P + c; Where ΔV is the change in water level volume, in m 3 ; P is the 24-hour cumulative rainfall within the circular range of the basin, in m 3 ; c and b are the constant term and coefficient respectively; The specific steps for flood range prediction in 1.4 are as follows: S1000, calculation of accumulated rainfall in the next 24 hours: S1000-1, download the next 24 hours forecast rainfall data in raster format; S1000-2, create a rainfall accumulation range file, extract the part within the basin, and obtain the cumulative range file of the forecast rainfall; S1000-3, project the accumulated rainfall, and the horizontal unit of the projection result is meter. Then, use the rainfall accumulation range file to extract the rainfall within the accumulation range from the projected rainfall file. Finally, calculate the surface volume of the extracted rainfall file. S2000, water level forecast: S2000-1, calculate the volume change of the water body based on the accumulated rainfall. Substitute the 24-hour forecast rainfall accumulated value into Formula 3 to calculate the predicted water level volume change ΔV; S2000-2, obtain the current water level data measurement value E, substitute the current water level value into formula 1 to obtain the lake volume V corresponding to the current water level e , and obtain the predicted lake volume V′ based on the predicted water level volume change e , the formula is: Formula 4: V′ e =V e +ΔV; Rewrite the formula 1 to obtain V′ e The corresponding predicted water level E′ is: Formula 5: E′=n+(V′ e -V n ) / (V n+1 -V n ); Among them, V n is the volume of lake water when the water level is an integer n, V n+1 is the volume of the lake when the water level is an integer n+1. In the calculation formula, first find the water level n so that V n+1 <V′ e <V n+1 ; S3000, flood range prediction: The prepared inundation map is interpolated to make it consistent with the accuracy of the measured water level data.

2. The rainfall-based flood inundation prediction method according to claim 1, characterized in that: The plane resolution of ERA5 in S10 is 0.1°, which is equivalent to 10 km.

3. The rainfall-based flood inundation prediction method according to claim 1, characterized in that: The multi-round principal component analysis in S300-1 includes four rounds: In the first round, the cumulative rainfall in the annular area within 180 km was used as the factor; The second round uses the cumulative rainfall in the circular area of ​​all distances as a factor; The third round uses the accumulated rainfall in a circular area between 90 and 180 km as the factor; The fourth round uses the accumulated rainfall in a circular area between 120 and 180 km as a factor.

4. The rainfall-based flood inundation prediction method according to claim 1, characterized in that: In the S600-3, c and b are 2062612888.591 and 0.100611 respectively.

Images