A basin rainstorm flood warning method and device based on improved Xin'anjiang model
The parameter rate determination of the Xin'anjiang model is simplified through satellite remote sensing data and global optimization algorithm, which solves the problems of long model calculation time and large errors, improves the accuracy and efficiency of flood warning, and is suitable for water resource management and disaster prevention and control.
Patent Information
- Application Number
- CN202410923925.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-11
- Publication Date
- 2024-11-08
- Estimated Expiration
- 2044-07-11
AI Technical Summary
In the flood warning, the Xin'anjiang model has many parameter rates and large correlations, resulting in long calculation time and large errors, which can easily lead to low traffic peak prediction, affecting the accuracy of forecast results, and missing the pre-disaster defense time.
Satellite remote sensing data combined with random forest algorithm is used to obtain the proportion of impermeable surfaces, and model parameters are determined through global optimization algorithm rate, simplify the production and convergence calculation mode, reduce the number of parameters, and improve model efficiency and accuracy.
It improves the accuracy and efficiency of flood warning, ensures the accuracy of flow peak prediction, reduces the time and error of parameter rate determination, enhances the generalization ability of the model, and is suitable for water resource management and disaster prevention and control.
Smart Images

Figure CN118467947B_ABST
Abstract
Description
Technical Field
[0001] The invention relates to a rainstorm and flood warning method in a watershed, and in particular to a rainstorm and flood warning method and equipment in a watershed based on an improved Xin'anjiang model, belonging to the technical field of satellite remote sensing and hydrological forecasting and monitoring. Background Art
[0002] In the early warning process, due to the differences in topography, climate and spatial distribution between different river basins, it is difficult to obtain the corresponding relationship between rainfall and river flow through simple formula simulation. Therefore, it is usually necessary to combine the historical data of the river basin in the region, use physical processes and mathematical formulas to abstract and summarize the water flow system, form a hydrological model, simulate the historical rainfall flow data, obtain the corresponding model parameters through iterative calibration algorithms, and finally generate the flow peak value of rainfall in the future period of the region to determine whether the region meets the conditions for flood occurrence.
[0003] As a type of lumped hydrological model, the Xin'anjiang model has been widely used in various river basins in my country since it was proposed. The model belongs to a conceptual river basin hydrological model, and its main feature is that it is mainly based on full storage and runoff. It is particularly suitable for rainfall runoff simulation in hilly and mountainous river basins in humid and semi-humid climate zones, and has certain applicable value in flood warning. However, the model involves many parameters for calibration in practical applications, and the time consumed in the model calibration process is relatively long. In addition, the correlation between the parameters is large, resulting in inaccurate actual optimization results of the parameters in actual flood warnings due to the influence of measured data errors and the complexity and incompleteness of the objective function. This easily leads to the model's low prediction results for the simulation of peak flow in the calculation of runoff generation and convergence, which usually leads to the neglect of flood peaks with low prediction values when using the model for flood disaster warning, thereby affecting the accuracy of the overall forecast results of the model, missing the best pre-disaster defense time, and exacerbating the losses caused by the disaster. Summary of the invention
[0004] The purpose of the present invention is to overcome the above-mentioned shortcomings and provide a basin rainstorm and flood warning method based on the improved Xin'anjiang model. An efficient, accurate and practical basin rainstorm and flood disaster warning method is provided based on the Xin'anjiang model. While retaining the advantages of Xin'anjiang itself, the model parameters are determined by other means, and the calculation mode of its runoff generation and confluence is optimized to improve the hydrological forecast efficiency and accuracy of the model.
[0005] The technical solution adopted by the present invention is:
[0006] A basin rainstorm flood warning method based on an improved Xin'anjiang model includes the following steps:
[0007] S1. Obtain and preprocess the remote sensing images of the watershed to be predicted;
[0008] S2. Extract texture features of preprocessed remote sensing images, annotate training samples, construct a classification decision tree using the random forest algorithm, classify objects in remote sensing images, and calculate the impervious surface percentage (IMP) based on the classification results;
[0009] S3. Use the Xin'anjiang model to simulate the runoff of the basin when it is full, and use the impervious surface percentage IMP to calculate the runoff R , the runoff calculated by Xin'an River is divided into two unit lines for confluence calculation, and the water volume of the confluence reservoir is calculated by the two unit lines respectively Q 9. Water volume at the outlet section of the basin Q 1. Introduction F As underground exchange water, the water storage change is updated again to obtain the outflow of the confluence reservoir Q r Outflow rate at the outlet section Q d , calculate the total sink flow;
[0010] S4. Input historical data of the watershed and calibrate the optimal parameters of the flow calculation model through the global optimization algorithm;
[0011] S5. Use the obtained model calibration parameters and rainfall, evaporation, and flow data to predict flow and make hydrological forecasts and warnings.
[0012] In the above method, step S1 is to obtain recent Sentinel-2 images in the area to be predicted, and perform radiometric calibration, atmospheric correction, image registration, and cropping and splicing preprocessing on the Sentinel-2 images.
[0013] In step S2, the impervious surface proportion IMP is the ratio of the impervious area to the basin area, and the sum of the high reflectivity impervious surface and the low reflectivity impervious surface predicted in the basin is taken as the impervious surface area of the basin.
[0014] The flow calculation process in step S3 is:
[0015] (1) Calculate the maximum water storage capacity WMM of each point in the basin:
[0016] WMM=WM×(1+B) / (1- IMP),
[0017] Where WM and B are parameters to be calibrated, which are the average water storage capacity of tension water in the aeration zone and the index of the tension water storage capacity curve, respectively, and IMP is the ratio of the impervious area;
[0018] (2) Calculate the critical value A of the tension water soil moisture content:
[0019] A = WMM×(1- (1- W / WM) 1 / (B+1) );
[0020] where W is the soil water content, and the calculation formula is:
[0021] W = wu + wl + wd,
[0022] where wu, wl, and wd are the soil water contents of the upper layer, lower layer, and deep layer respectively,
[0023] In the case of runoff generation, when wu0 + P - E - R < WUM (the upper layer of soil is not full), the calculation formula is as follows:
[0024] ,
[0025] where wu0, wl0, and wd0 are the custom initial values of the soil water contents of the upper layer, lower layer, and deep layer respectively, which can be obtained from the historical runoff data of the basin. If the data is missing, it can be set to 0. P, E, R are the rainfall, evaporation, and runoff respectively. WUM is the upper layer tension water storage capacity, which is obtained by the calibration algorithm;
[0026] When wu0 + P - E - R > WUM and wu0 + wl0 + P - E - R < WUM + WLM (the upper layer of soil is full and the middle layer of soil is not full), the formulas for the soil water contents of the upper layer, lower layer, and deep layer are as follows:
[0027] ,
[0028] where WLM is the lower layer tension water storage capacity, which is obtained by the calibration algorithm;
[0029] When wu0 + P - E - R > WUM and wu0 + wl0 + P - E - R > WUM + WLM (both the upper and middle layers are full and the remaining water replenishes the lower layer), the formulas for the soil water contents of the upper layer, lower layer, and deep layer are as follows:
[0030] ,
[0031] In the case of no runoff generation in the basin, if wu0 + P - E ≥ 0, the calculation formulas for the soil water contents of the upper layer, lower layer, and deep layer are:
[0032] ,
[0033] When wu0 + P - E < 0 and wl0 ≥ C × WLM, the calculation formulas for the soil water contents of the upper layer, lower layer, and deep layer are:
[0034] ,
[0035] where C is the deep evaporation and diffusion coefficient, obtained by the calibration algorithm;
[0036] If C×(E - wu0 - 2P) ≤ wl0 < C×WLM, the calculation formulas for the soil water content of the upper layer, lower layer, and deep layer are as follows:
[0037] ,
[0038] If wl0 ≤ C×(E - wu0 - 2P), the calculation formulas for the soil water content of the upper layer, lower layer, and deep layer are as follows:
[0039] ;
[0040] (3) Calculate the runoff generation:
[0041] ,
[0042] where P e is the net rainfall amount, and the calculation formula is as follows:
[0043] P e = P - K×E,
[0044] P and E are rainfall and evapotranspiration data respectively, and K is the evapotranspiration conversion coefficient, obtained by the calibration algorithm.
[0045] The calculation process of the confluence in step S3 is as follows:
[0046] (1) Divide the runoff generation into two parts for confluence calculation. The first 90% uses the unit hydrograph UH 1, and the last 10% uses the unit hydrograph UH 2. Among them, UH 1 needs to be regulated by the confluence reservoir, UH 2 needs to be collected at the outlet section of the basin;
[0047] The two unit hydrographs UH 1, UH 2 are obtained from the following formulas respectively:
[0048] ,
[0049] ,
[0050] In the formula, X 3 is the confluence time parameter, t is the time, j is an integer, and the units are all days; UH 1( j ) and UH 2( j ) are the time relationship curves for simulating the confluence, SH 1(t )and SH 2( t ) is the cumulative ratio of input time to confluence time, through SH 1( t )and SH 2( t ) UH 1( j )and UH 2( j ) to express;
[0051] Water volume of confluence reservoir calculated from unit line Q 9. Water volume at the outlet section of the basin Q 1 is shown in the following formula, R is the runoff flow,
[0052] ;
[0053] (2) Introduction F As underground exchange water, X 1 is the groundwater exchange coefficient, X 2 is the confluence reservoir capacity, R i is the water volume of the confluence reservoir, X 2×0.7, when X 1 is less than zero, indicating that runoff recharges groundwater. X 1 equals zero, indicating no water exchange.
[0054] ,
[0055] When importing UH 1 After the corresponding water volume and groundwater exchange water volume, the corresponding water storage volume R r and outflow Q r As shown in the following formula, the water storage capacity updated again after outflow R r+1 Changes to:
[0056] ,
[0057] ,
[0058] ,
[0059] For the basin outlet section, when the unit line UH After the estimated water volume of 2 is combined with the groundwater exchange volume, the outflow Q d for:
[0060] ;
[0061] The total outflow of the basin Q The outflow of the confluence reservoir Q r Outflow rate at the outlet section Q d sum.
[0062] Step S4: Parameters to be calibrated. The historical data that need to be input during the calibration process include daily rainfall (mm), daily evapotranspiration (mm), daily flow (m 3 / s). The general Xin'anjiang model needs to calibrate at least 15 parameters, while the present invention only needs to calibrate 9 parameters.
[0063] The specific implementation steps of the global optimization algorithm described in step S4 are as follows:
[0064] a. Determine the number of hydrological model parameters n, population size s and the number of population partitions p;
[0065] b. Bring the initialization parameters into the model for calculation, obtain the Nash efficiency coefficient calculation result NSE of each parameter in the model, and store each initialization parameter into array D to generate sample points;
[0066] c. Sorting the sample points, sorting the parameters in array D in ascending order according to the model calculation results (NSE), and dividing them into p partitions: C1, C2, ..., C p ;
[0067] d. Composite evolution, using the simplex method to optimize each partition, that is, performing evolution operations on the selected vertices of each partition, including generating new vertices (mutation), adjusting the shape and size of the simplex (contraction), etc., comparing the new vertices obtained after evolution with the original vertices, if the new vertices are better, replace the original vertices; otherwise, retain the original vertices; the optimal vertices after the evolution operation are the optimal parameters of the model, and then the optimal parameters of each partition are combined to regenerate the array D;
[0068] e. Repeat the above steps for multiple iterations until the model calculation value is no longer updated after reaching the preset maximum number of iterations.
[0069] Another object of the present invention is to provide a watershed rainstorm and flood warning device based on an improved Xin'anjiang model, comprising a processor and a memory storing computer program instructions; when the processor executes the computer program instructions, it implements the watershed rainstorm and flood warning method based on the improved Xin'anjiang model as described above.
[0070] The present invention classifies land objects through satellite remote sensing data combined with the Random Forest (RF) algorithm, obtains the proportion of impervious surface area in the Xin'anjiang model parameters, and uses unit lines to simulate the full storage and production flow of the Xin'anjiang model. Finally, the model parameters are calibrated and obtained through the global optimization algorithm in the process of historical data simulation. This method can more accurately determine the impervious surface parameters of the model by combining satellite remote sensing technology, and can improve the prediction value during floods by simulating the unit lines of the confluence reservoir and the outlet section. The present invention can reduce the number of parameters calibrated in the previous Xin'anjiang model, improve the calibration efficiency of the model parameters, and improve the generalization ability of the model through the description of the unit line confluence, improve its problem of low prediction at the peak flow rate, and ensure that it can obtain accurate flow prediction during the flood season of the basin. It has important practical application value and far-reaching social significance for the fields of water resources management, disaster prevention and control, and environmental protection. BRIEF DESCRIPTION OF THE DRAWINGS
[0071] Figure 1 is a flow chart of the method of the present invention;
[0072] Figure 2 It is a structural diagram of the Xin'anjiang model improved by the method of the present invention;
[0073] Figure 3 It is a flow chart of the parameters of the method of the present invention;
[0074] Figure 4 It is a scatter plot comparison of the predicted flow and the measured flow drawn by the model of the method of the present invention and other models in a river basin in Shandong Province during the period of 2000 / 3 / 30-2000 / 6 / 30;
[0075] Figure 5 It is a scatter plot comparison of the predicted flow and the measured flow drawn by the model of the method of the present invention and other models in a river basin in Shandong Province during the period of 2001 / 4 / 27 to 2001 / 7 / 27;
[0076] Figure 6 It is a scatter plot comparison of the predicted flow and the measured flow drawn by the model of the method of the present invention and other models in a river basin in Shandong Province during the period of 2004 / 4 / 1-2004 / 7 / 1;
[0077] Figure 7 It is a scatter plot comparison of the predicted flow and the measured flow drawn by the model of the method of the present invention and other models in a river basin in Shandong Province during the period of 2005 / 9 / 20 to 2005 / 12 / 20;
[0078] Figure 8 It is a scatter plot comparison of the predicted flow and the measured flow drawn by the model of the method of the present invention and other models in a river basin in Anhui during the period from 1998 / 7 / 1 to 1999 / 8 / 30. DETAILED DESCRIPTION
[0079] The present invention is further described below with reference to specific embodiments and accompanying drawings.
[0080] Embodiment 1 A method for early warning of rainstorm and flood in a river basin based on an improved Xin'anjiang model comprises the following steps (eg Figure 1 )as follows:
[0081] S1. Obtain remote sensing images of the watershed to be predicted and preprocess them:
[0082] After obtaining the recent Sentinel-2 images in the area to be predicted, preprocessing operations such as radiation calibration and atmospheric correction are performed to eliminate the influencing factors such as sensors and atmosphere. Then, the Sentinel-2 images are registered and corrected using the CDOG algorithm in combination with the Tiandi Map images to ensure their accurate geographical location. Finally, the images are cropped and spliced according to the boundary range of the watershed in the study area.
[0083] S2. Extract the texture features of the preprocessed remote sensing images, annotate the training samples, build a classification decision tree through the random forest algorithm, classify the objects in the remote sensing images, and calculate the impervious surface proportion IMP based on the classification results:
[0084] The random forest algorithm is used to construct a classification decision tree to classify Sentinel-2 image objects and obtain the percentage of impervious surface area. This algorithm is a statistical method proposed by Breiman and Cutler in 2001. The random forest consists of several integrated classification decision trees, and the trees are independently distributed. When an input classification sample is given, each tree will give its own prediction result, and then vote according to the prediction results of each tree, and the category with the most votes will be output as the final result.
[0085] The specific implementation steps are as follows:
[0086] (1) Feature extraction
[0087] The texture features of Sentinel-2 images are extracted based on the Gray-Level Co-occurrence Matrix (GLCM), and the texture information of the images is described by eight texture feature measures: mean, variance, uniformity, contrast, non-similarity, entropy, second-order moment, and correlation.
[0088] (2) Sample construction
[0089] A feature set is constructed based on the texture features of Sentinel-2 images and the spectral information of each band. The set includes a total of 300 feature point sets, including water bodies, vegetation, bare land, high reflectivity impervious surfaces, and low reflectivity impervious surfaces.
[0090] (3) Random Forest Classification
[0091] a. Use the Bootstrap method to randomly select n samples with replacement from the 300 training samples to obtain a training set of n samples. The number of samples contained in each training set is not less than 2 / 3 of the total number of samples.
[0092] b. Build a decision tree for each training set to form a random forest, then randomly select m (m<5) from the five feature categories and split the decision tree nodes by m;
[0093] c. Random forest integrates the prediction results of each tree and determines the final prediction category by voting;
[0094] (4) Determine the impervious surface
[0095] The sum of high-reflectivity impervious surfaces and low-reflectivity impervious surfaces predicted by random forest in the basin is extracted as the impervious surface area of the basin, and the ratio of impervious area to basin area is IMP.
[0096] S3. Use the Xin'anjiang model to simulate the runoff of the basin when it is full, and use the impervious surface percentage IMP to calculate the runoff R , the runoff R calculated by Xin'an River is divided into two unit lines for confluence calculation, and the water volume of the confluence reservoir is calculated by the two unit lines respectively Q 9. Water volume at the outlet section of the basin Q 1. Introduction F As underground exchange water, the water storage change is updated again to obtain the outflow of the confluence reservoir Q r Outflow rate at the outlet section Q d , calculate the total sink flow (such as Figure 2 shown):
[0097] The impervious surface ratio parameter of the Xin'anjiang model used in the present invention is obtained through satellite remote sensing images, and the runoff generation of the basin in a full state is simulated through two unit lines in the confluence stage, which omits the water source division and evapotranspiration calculation parts in the Xin'anjiang model, and the number of parameters involved is significantly reduced, as shown in Table 1.
[0098] Table 1 Parameters required for calibration of the present invention
[0099]
[0100] The runoff yield calculation of the Xin'anjiang model is derived from the theory of full storage runoff generation. The core concept of full storage runoff generation lies in "full storage", that is, the water content in the soil vadose zone reaches its field capacity state. Before the soil moisture reaches this field capacity, all rainfall will be absorbed by the soil and converted into tensiometer water, and no runoff will be generated at this time. When the soil moisture reaches or exceeds the field capacity, all rainfall (minus the evaporation during the same period) will be converted into runoff.
[0101] The process of runoff yield calculation is as follows:
[0102] (1) Calculate the maximum value of the storage capacity at each point in the basin, WMM:
[0103] WMM = WM×(1 + B) / (1 - IMP),
[0104] WM and B are parameters to be calibrated and obtained through the calibration algorithm. They represent the average storage capacity of the tensiometer water in the vadose zone and the exponent of the tensiometer water storage capacity curve respectively. IMP is the ratio of the impervious area;
[0105] (2) Calculate the critical value A of the tensiometer water soil moisture:
[0106] A = WMM×(1 - (1 - W / WM) 1 / (B+1) );
[0107] Where W is the soil moisture content, and the calculation formula is:
[0108] W = wu + wl + wd,
[0109] wu, wl, and wd are the soil moisture contents of the upper layer, lower layer, and deep layer respectively,
[0110] In the case of its runoff generation, when wu0 + P - E - R < WUM (the upper layer of soil is not fully stored), the calculation formula is as follows:
[0111] ,
[0112] Where wu0, wl0, and wd0 are the custom initial values of the soil moisture contents of the upper layer, lower layer, and deep layer respectively, which can be obtained according to the historical runoff generation data of the basin. If the data is missing, it can be set to 0. P, E, R are rainfall, evaporation, and runoff respectively. WUM is the upper layer tensiometer water storage capacity, which is obtained by the calibration algorithm;
[0113] When wu0 + P - E - R > WUM and wu0 + wl0 + P - E - R < WUM + WLM (the upper layer of soil is fully stored and the middle layer of soil is not fully stored), the formulas for the soil moisture contents of the upper layer, lower layer, and deep layer are as follows:
[0114] ,
[0115] where WLM is the storage capacity of the lower-layer tension water, obtained by the calibration algorithm;
[0116] When wu0 + P - E - R > WUM and wu0 + wl0 + P - E - R > WUM + WLM (the upper and middle layers are both full, and the remaining water replenishes the lower layer), the formulas for the soil water contents of the upper layer, lower layer, and deep layer are as follows:
[0117] ,
[0118] In the case of no runoff generation in the watershed, if wu0 + P - E ≥ 0, the calculation formulas for the soil water contents of the upper layer, lower layer, and deep layer are:
[0119] ,
[0120] When wu0 + P - E < 0 and wl0 ≥ C × WLM, the calculation formulas for the soil water contents of the upper layer, lower layer, and deep layer are:
[0121] ,
[0122] where C is the deep evaporation diffusion coefficient, obtained by the calibration algorithm;
[0123] If C × (E - wu0 - 2P) ≤ wl0 < C × WLM, the calculation formulas for the soil water contents of the upper layer, lower layer, and deep layer are:
[0124] ,
[0125] If wl0 ≤ C × (E - wu0 - 2P), the calculation formulas for the soil water contents of the upper layer, lower layer, and deep layer are:
[0126] ;
[0127] (3) Calculate runoff:
[0128] ,
[0129] where P e is the net rainfall amount, and the calculation formula is as follows:
[0130] P e = P - K × E,
[0131] P and E are rainfall and evapotranspiration data respectively, and K is the evapotranspiration conversion coefficient, obtained by the calibration algorithm.
[0132] In the runoff calculation, due to the difference in the runoff time of different runoff components, the runoff R calculated by Xin'an River is divided into two parts for runoff calculation. The first 90% is calculated by unit line. UH 1. The last 10% uses the unit line UH 2, where UH 1. It needs to be regulated by a confluence reservoir. UH 2 needs to be collected at the outlet section of the basin. In discrete form, the unit line UH 1 and UH 2 respectively have n and m coordinates, where n and m is greater than X 3 and 2 X 3. This means that the water flow is staggered UH 1 of n Unit line input and UH 2 of m The ordinates of the two unit lines are derived from the corresponding S Curve (the cumulative ratio of input to time), respectively SH 1 and SH 2 means.
[0133] Two unit lines UH 1. UH 2 are obtained by the following formula:
[0134] ,
[0135] ,
[0136] In the formula X 3 is the confluence time parameter, t For time, j It is an integer, and the unit is day; UH 1( j )and UH 2( j ) is the time relationship curve of simulated confluence, SH 1( t )and SH 2( t ) is the cumulative ratio of input time to confluence time, through SH 1( t )and SH 2( t ) UH 1( j )and UH 2( j ) to express;
[0137] Water volume of confluence reservoir calculated from unit line Q 9. Water volume at the outlet section of the basin Q 1 is shown in the following formula, R is the runoff flow,
[0138] ,
[0139] Introduction F As underground exchange water, X 1 is the groundwater exchange coefficient, X 2 is the confluence reservoir capacity, R i is the water volume of the confluence reservoir, X 2×0.7, when X 1 is less than zero, indicating that runoff recharges groundwater. X 1 equals zero, indicating no water exchange.
[0140] ,
[0141] When importing UH 1 After the corresponding water volume and groundwater exchange water volume, its corresponding water storage volume R r and outflow Q r As shown in the following formula, the water storage capacity updated again after outflow R r+1 Changes to:
[0142] ,
[0143] ,
[0144] ,
[0145] For the basin outlet section, when the unit line UH After the estimated water volume of 2 is combined with the groundwater exchange volume, the outflow Q d for:
[0146] ;
[0147] The total outflow of the basin Q The outflow of the confluence reservoir Q r Outflow rate at the outlet section Q d sum.
[0148] S4. Input the historical rainfall evaporation and flow data of the basin, and obtain the optimal parameters of the confluence calculation model through global optimization algorithm calibration:
[0149] The present invention uses a global optimization algorithm (Shuffled Complex Evolution Algorithm for Universal Application, SCE-UA) as a calibration algorithm for model parameters. The algorithm uses the Nash-Sutcliffe Efficiency coefficient (NSE) of the model forecast flow as an iteration basis. By constructing a simplex of model parameters and continuously evolving the simplex (including operations such as simplex generation, mutation, contraction and replacement), the model parameters in the simplex gradually approach the global optimal solution during the updating process. The specific implementation steps of the algorithm (such as Figure 3 )as follows:
[0150] a. Determine the number of hydrological model parameters n, population size s and the number of population partitions p;
[0151] b. Bring the initialization parameters into the model for calculation, obtain the Nash efficiency coefficient calculation result NSE of each parameter in the model, and store each initialization parameter into array D to generate sample points;
[0152] c. Sorting the sample points, sorting the parameters in array D in ascending order according to the model calculation results (NSE), and dividing them into p partitions: C1, C2, ..., C p ;
[0153] d. Composite evolution, using the simplex method to optimize each partition, that is, performing evolution operations on the selected vertices of each partition, including generating new vertices (mutation), adjusting the shape and size of the simplex (contraction), etc., comparing the new vertices obtained after evolution with the original vertices, if the new vertices are better, replace the original vertices; otherwise, retain the original vertices; the optimal vertices after the evolution operation are the optimal parameters of the model, and then the optimal parameters of each partition are combined to regenerate the array D;
[0154] e. Repeat the above steps for multiple iterations until the model calculation value is no longer updated after reaching the preset maximum number of iterations.
[0155] S5. Use the obtained calibration model parameters and rainfall evaporation flow data to predict flow and make hydrological forecasts and warnings.
[0156] A river basin in Shandong Province is taken as the study area. The average annual precipitation in the basin is 706 mm, and the average annual runoff depth is 156.5 mm. The upper reaches are low and medium mountainous and hilly areas in central and southern Shandong, where floods rise and fall rapidly; the lower reaches are accumulation plain areas in northwestern Shandong, which are prone to flood disasters.
[0157] In order to verify the accuracy of the improved model of the present invention, the results of the Nash efficiency coefficient (NSE), the relative error (RE) of the total runoff, and the Pearson correlation coefficient (R) of the flow forecast results of the improved model of the present invention in all periods of a river basin are shown in Table 2. According to the statistical results, the NSE, RE, and R of the flow forecast of the improved Xin'anjiang model of the present invention are better than those of the Xin'anjiang and GR4J models. This is because the impervious surface parameters extracted by Sentinel-2 do not participate in the calibration, and the water source division and evapotranspiration calculation of the Xin'anjiang model are omitted, so that the number of parameters for calibration has been greatly reduced. It can show strong adaptability and generalization ability in dealing with the runoff forecasting tasks of various basins, and the flow curve obtained by the forecast is more in line with the measured value, and the error is small. It is closer to the measured value at the flow peak, and the correlation with the measured curve is also higher. Therefore, the present invention is more sensitive to the flow change at the peak in the actual basin forecast.
[0158] Table 2 Evaluation results of the flow rate forecast by the method of the present invention
[0159]
[0160] Figure 4-Figure 7 The scatter plots of the flow predicted by the Xin'anjiang model and the measured flow in different periods of the basin are given respectively. (a) is the scatter plot of the flow predicted by the GR4J model and the measured flow, (b) is the scatter plot of the flow predicted by the GR4J model and the measured flow, and (c) is the scatter plot of the flow predicted by the improved model of the present invention and the measured flow. Figure 4 The comparison results of each model for the period 2000 / 3 / 30-20006 / 30 are as follows: Figure 5 The comparison results of each model for the period 2001 / 4 / 27-2001 / 7 / 27 are shown below. Figure 6 The comparison results of each model for the period 2004 / 4 / 1-2004 / 7 / 1 are shown below. Figure 7 The comparison results of each model for the period 2005 / 9 / 20-2005 / 12 / 20. In each scatter plot, the horizontal axis represents the predicted value of the model, and the vertical axis represents the measured value. Through the comparison between the models, it can be seen that the flow forecast results obtained by the improved model of the present invention are most consistent with the dotted line compared with other models, the distribution of points is more uniform and dense, and the relationship between them and the corresponding measured values is relatively stable, and the prediction accuracy is higher, especially at higher flow peaks, it is closer to the measured values than other models, and can better meet the flood monitoring and early warning needs in the basin.
[0161] The study area is a river basin in Anhui Province, which is located in a humid and semi-humid area with an annual rainfall of 1280-1700 mm. The rainfall in May and June accounts for the highest amount of the year, and most of them are torrential rains. The summer is easily affected by the plum rain, with heavy rainfall and frequent floods. Figure 8 The comparison results of the scatter plots of the predicted values and measured values of the improved model, Xin'anjiang model and GR4J model during the period of 1998 / 7 / 1-1999 / 8 / 30 are given, among which (a) is the predicted measured flow scatter plot of the Xin'anjiang model, (b) is the predicted measured flow scatter plot of the GR4J model, and (c) is the predicted measured flow scatter plot of the improved model of the present invention. It can be seen from the figure that the improved Xin'anjiang model of the present invention has a higher correlation with the measured values, and each scatter point is more consistent with other models, and the degree of fit with the measured flow curve is higher than that of other algorithms. Therefore, the present invention can accurately determine the flow peak value during the flood season of the basin, clearly perceive the changes in the flood peak, and realize the forecast and early warning of flood disasters in the rainy season.
[0162] Embodiment 2: A watershed rainstorm and flood warning device based on an improved Xin'anjiang model, comprising a processor and a memory storing computer program instructions; when the processor executes the computer program instructions, it implements the watershed rainstorm and flood warning method based on the improved Xin'anjiang model as described in Embodiment 1 above.
[0163] The above description is only a specific implementation of the present application, and the protection scope of the present invention is not limited thereto.
Claims
1. A basin rainstorm flood warning method based on the improved Xin'anjiang model, characterized by: It includes the following steps: S1. Obtain the remote sensing image of the basin to be predicted and preprocess it; S2. Extract the texture features of the preprocessed remote sensing image, label the training samples, construct a classification decision tree through the random forest algorithm, classify the ground objects of the remote sensing image, and calculate the impervious surface proportion IMP according to the classification results; S3. Use the Xin'anjiang model to simulate the runoff of the basin when it is full, and use the impervious surface percentage IMP to calculate the runoff R , the runoff calculated by Xin'an River is divided into two unit lines for confluence calculation, and the water volume of the confluence reservoir is calculated by the two unit lines respectively Q 9. Water volume at the outlet section of the basin Q 1. Introduction F As underground exchange water, the water storage change is updated again to obtain the outflow of the confluence reservoir Q r Outflow rate at the outlet section Q d , calculate the total sink flow; The runoff calculation process is as follows: (1) Calculate the maximum value WMM of the water storage capacity at each point in the basin: WMM = WM×(1 + B) / (1 - IMP), where WM and B are parameters to be calibrated, which are the average water storage capacity of the vadose zone tension and the tension water storage capacity curve index respectively, and IMP is the ratio of the impervious area; (2) Calculate the critical value A of the tension water soil moisture content: A= WMM×(1- (1- W / WM) 1 / (B+1) ), where W is the soil moisture content; (3) Calculation of flow rate R : , Where P e is the net rainfall, and the calculation formula is as follows: P e =P-K×E, P and E are rainfall and evapotranspiration data respectively, and K is the evapotranspiration conversion coefficient, which is obtained by the calibration algorithm; The confluence calculation process is as follows: (1) Divide the runoff into two parts for runoff calculation. The first 90% is calculated using the unit line. UH 1. The last 10% uses the unit line UH 2, where UH 1. It needs to be regulated by a confluence reservoir. UH 2. It needs to be collected at the outlet section of the basin; Two unit lines UH 1. UH 2 are obtained by the following formula: , , In the formula X 3 is the confluence time parameter, t For time, j It is an integer, and the unit is day; UH 1( j )and UH 2( j ) is the time relationship curve of simulated confluence, SH 1( t )and SH 2( t ) is the cumulative ratio of input time to confluence time, through SH 1( t )and SH 2( t ) UH 1( j )and UH 2( j ) to express; Water volume of confluence reservoir calculated from unit line Q 9. Water volume at the outlet section of the basin Q 1 is shown in the following formula, R is the runoff flow, ; (2) Introduction F As underground exchange water, X 1 is the groundwater exchange coefficient, X 2 is the confluence reservoir capacity, R i is the water volume of the confluence reservoir, X 2×0.7, when X 1 is less than zero, indicating that runoff recharges groundwater. X 1 equals zero, indicating no water exchange. , When importing UH 1 After the corresponding water volume and groundwater exchange water volume, its corresponding water storage volume R r and outflow Q r As shown in the following formula, the water storage capacity updated again after outflow R r+1 Changes to: , , , For the basin outlet section, when the unit line UH After the estimated water volume of 2 is combined with the groundwater exchange volume, the outflow Q d for: , The total outflow of the basin Q The outflow of the confluence reservoir Q r Outflow rate at the outlet section Q d sum; S4. Input the historical data of the basin and calibrate it through the global optimization algorithm to obtain the optimal parameters of the confluence calculation model; S5. Use the calibrated model parameters and rainfall, evaporation, and flow data to perform flow prediction and make hydrological forecasts and early warnings.
2. The method for early warning of rainstorm and flood in a river basin based on the improved Xin'anjiang model according to claim 1 is characterized in that:Step S1 is to obtain the recent Sentinel-2 image in the area to be predicted, and perform preprocessing such as radiometric calibration, atmospheric correction, image registration, and cropping and mosaicking on the Sentinel-2 image.
3. The method for early warning of rainstorm and flood in a river basin based on the improved Xin'anjiang model according to claim 1 is characterized in that: In step S2, the impervious surface proportion IMP is the ratio of the impervious area to the basin area, and the sum of the high-reflectivity impervious surface and the low-reflectivity impervious surface predicted in the basin is used as the impervious surface area of the basin.
4. The method for early warning of rainstorm and flood in a river basin based on the improved Xin'anjiang model according to claim 1 is characterized in that: The calculation formula for the soil moisture content W is: W = wu + wl + wd, where wu, wl, and wd are the soil moisture contents of the upper layer, lower layer, and deep layer respectively, In the case of its runoff generation, when wu0 + P - E - R <WUM, the calculation formula is as follows: , Where wu0, wl0 and wd0 are the user-defined initial values of soil moisture in the upper, lower and deep layers, respectively. R are rainfall, evaporation and runoff, respectively, WUM is the upper tension water storage capacity, obtained by the calibration algorithm; When wu0 + P - E - R > WUM and wu0 + wl0 + P - E - R < WUM + WLM, the soil water content formulas for the upper, lower, and deep layers are as follows: , where WLM is the lower layer tension water storage capacity, which is obtained by the calibration algorithm; When wu0+PE- R >WUM and wu0+wl0+PE- R When >WUM+WLM, the soil moisture content formulas for the upper, lower and deep layers are as follows: , In the case where the basin does not generate runoff, if wu0 + P - E ≥ 0, the calculation formulas for the soil moisture contents of the upper layer, lower layer, and deep layer are: , When wu0 + P - E < 0 and wl0 ≥ C×WLM, the calculation formulas for the soil moisture contents of the upper layer, lower layer, and deep layer are: , where C is the deep layer diffusion coefficient of evaporation, which is obtained by the calibration algorithm; If C×(E - wu0 - 2P) ≤ wl0 < C×WLM, the calculation formulas for the soil moisture contents of the upper layer, lower layer, and deep layer are: , If wl0 ≤ C×(E - wu0 - 2P), the calculation formulas for the soil moisture contents of the upper layer, lower layer, and deep layer are: 。 5. The method for early warning of rainstorm and flood in a river basin based on the improved Xin'anjiang model according to claim 1 is characterized in that: The specific implementation steps of the global optimization algorithm described in step S4 are as follows: a. Determine the number of hydrological model parameters n, the population size s, and the number of population partitions p; b. Substitute the initialized parameters into the model for calculation, obtain the Nash efficiency coefficient calculation result NSE of each parameter in the model, and store each initialized parameter in the array D to generate sample points; c. Sort the sample points, sort the parameters in array D in ascending order according to the model calculation results NSE, and divide them into p partitions: C1, C2, ..., C p ; d. Composite form evolution, use the simplex method to optimize each partition, that is, perform evolution operations on the selected vertices of each partition, including generating new vertices, adjusting the shape and size of the simplex, compare the new vertices obtained after evolution with the original vertices, and if the new vertices are better, replace the original vertices; Otherwise, the original vertex is retained; the optimal vertex after the evolution operation is the optimal parameter of the model, and then the optimal parameters of each partition are combined to regenerate the array D; e. Repeat the above steps for multiple iterations until the model calculation value is no longer updated after reaching the preset maximum number of iterations.
6. A basin rainstorm flood warning device based on the improved Xin'anjiang model, comprising a processor and a memory storing computer program instructions, characterized in that: When the processor executes the computer program instructions, the method for early warning of rainstorm and flood in a watershed based on the improved Xin'anjiang model as described in any one of claims 1 to 5 is implemented.
Citation Information
Patent Citations
Improved Xinanjiang model-based hydrological forecast method suitable for hill region
CN108874936A
Reservoir incoming water quantity early warning and forecasting method and system based on small and medium-sized basin flood forecasting
CN113742910A
Hydrological model parameter uncertainty analysis method based on Morris and Sobol methods
CN117787081A
Hydro-meteorological time sequence-based runoff simulation method and system
WO2021217776A1
Big data-based hydrologic forecasting method
WO2022032872A1
