A method for constructing a mountain flood hydrological and hydrodynamic model and calibrating parameters in data-deficient areas based on SWOT remote sensing data

Abstract

The present application discloses a method for constructing and parameter calibration of a mountain flood hydrological and hydrodynamic model in data-deficient areas based on SWOT remote sensing data, which relates to the technical fields of hydrological and hydrodynamic simulation and mountain flood disaster warning. The method includes: calculating the observed water depth of each observation point in the target area according to the SWOT satellite remote sensing data and the soil static environment data; constructing an initial mountain flood hydrological and hydrodynamic model based on cellular automata; constructing a differentiable parameter learning framework based on the initial mountain flood hydrological and hydrodynamic model and a preset machine learning model; training the differentiable parameter learning framework to optimize the model parameters of the preset machine learning model and determine the corresponding optimal hydrological and hydrodynamic parameters; and determining the final mountain flood hydrological and hydrodynamic model according to the optimal hydrological and hydrodynamic parameters and the initial mountain flood hydrological and hydrodynamic model. The present application can achieve efficient, accurate, and automated parameter calibration, thereby improving the accuracy, timeliness, and reliability of flood modeling.

Description

Technical Field

[0001] This application relates to the technical fields of hydrological and hydrodynamic simulation and flash flood disaster warning, and particularly to a method for constructing a mountain flood hydrological and hydrodynamic model and calibrating parameters in data-deficient areas based on SWOT remote sensing data. Background Art

[0002] Flash flood disaster is a natural disaster that seriously threatens the safety of life and property in mountainous areas and along rivers, characterized by strong suddenness, great destructive power, and short duration. In the prediction and prevention of flash flood disasters, the construction of hydrological and hydrodynamic models is of crucial significance. However, the existing hydrological and hydrodynamic models are computationally complex and consume a large amount of resources, making it difficult to meet the rapid simulation requirements of large-scale regions. In the emergency situation of flash flood outbreaks, traditional models often fail to provide real-time and accurate flash flood simulations. On the other hand, most of the existing hydrological model parameter calibration methods rely on traditional optimization algorithms, trial-and-error methods, or empirical formulas. These methods usually optimize model parameters based on single-point observation data and do not fully consider the spatial correlation of hydrological processes, resulting in low simulation accuracy for non-gauge areas. Moreover, traditional methods often rely on manual adjustment strategies, with a cumbersome optimization process and difficult to achieve automation, and it is difficult to use large-scale satellite remote sensing data for global calibration.

[0003] In addition, there are also many challenges in obtaining hydrometeorological data. Traditional hydrological stations are sparsely distributed and have limited observation coverage. Especially in mountainous and remote areas, due to complex terrain and environmental conditions, long-term continuous observation data is extremely lacking, it is difficult to obtain long-time series data for a single station, and the data missing problem is very serious. Traditional hydrological model parameter calibration methods usually rely on long-time series data of a single station and optimize the calibration by historical observation records of variables such as water level and flow velocity. However, in most non-gauge or data-deficient areas, due to the difficulty of obtaining continuous and stable long-time series data, traditional calibration methods are difficult to apply, resulting in a significant increase in the uncertainty of model parameters. In addition, even in some areas with observation data, the data records often have discontinuities, measurement errors, or insufficient time spans, further limiting the reliability of traditional calibration methods.

[0004] To make up for the above deficiencies, data-deficient areas usually try to introduce parameter calibration methods based on machine learning, but traditional optimization algorithms also require a large amount of computing resources, are prone to falling into unstable calibration results, and it is difficult to ensure the rationality of parameters. The lack of these data directly limits the applicability and accuracy of traditional parameter calibration methods, so innovative technical means are urgently needed to address this challenge. Summary of the Invention

[0005] The objective of this application is to provide a method for constructing and parameter calibration of a mountain flood hydrological and hydrodynamic model in data-deficient areas based on SWOT remote sensing data, which can achieve efficient, accurate, and automated parameter calibration, thereby improving the accuracy, timeliness, and reliability of flood modeling.

[0006] To achieve the above objective, this application provides the following solutions:

[0007] This application provides a method for constructing and parameter calibration of a mountain flood hydrological and hydrodynamic model in data-deficient areas based on SWOT remote sensing data, including:

[0008] Obtain the soil static environment data, environmental driving sample data, and corresponding SWOT satellite remote sensing data of the target area;

[0009] According to the SWOT satellite remote sensing data and the soil static environment data, calculate the observed water depth of each observation point in the target area; the soil static environment data, the environmental driving sample data, and the observed water depth of all corresponding observation points form a training sample, and multiple training samples form a training sample set;

[0010] Based on cellular automata, construct an initial mountain flood hydrological and hydrodynamic model according to the soil static environment data; the initial mountain flood hydrological and hydrodynamic model includes hydrological and hydrodynamic parameters;

[0011] Based on the initial mountain flood hydrological and hydrodynamic model and a preset machine learning model, construct a differentiable parameter learning framework;

[0012] Use the training sample set to train the differentiable parameter learning framework to optimize the model parameters of the preset machine learning model and determine the corresponding optimal hydrological and hydrodynamic parameters;

[0013] According to the optimal hydrological and hydrodynamic parameters and the initial mountain flood hydrological and hydrodynamic model, determine the final mountain flood hydrological and hydrodynamic model.

[0014] According to the specific embodiments provided in this application, the following technical effects are achieved: In this application, SWOT satellite remote sensing data is used for data processing. The high-resolution characteristics of SWOT satellite remote sensing data can be utilized to obtain accurate water depth and water surface change information in remote areas and mountainous areas where traditional observation methods are difficult to cover, thus making up for the problem of data shortage. Based on cellular automata, an initial mountain flood hydrological and hydrodynamic model is constructed according to soil static environment data, and spatial distribution data such as water level and water depth within the entire basin can be obtained. These data can be highly matched with the SWOT satellite remote sensing observation data. The data obtained by SWOT satellite remote sensing is data on the basin surface, which is different from the time-series data at traditional observation stations. For this reason, this application constructs a differentiable parameter learning framework, combines machine learning algorithms, and uses the method of "exchanging space for time" to effectively calibrate the hydrological and hydrodynamic parameters in the initial mountain flood hydrological and hydrodynamic model using SWOT satellite remote sensing data. Moreover, through the differentiable parameter learning method, the repeated trial-and-error process in traditional model parameter calibration can be avoided, and the spatial similarity of the hydrological process can be comprehensively considered, thereby improving the efficiency and accuracy of model parameter calibration.

[0015] In summary, this application uses cellular automata and a differentiable parameter learning framework to provide a new method for constructing a flood simulation model and automatic parameter calibration, which has higher efficiency, accuracy, and flexibility, improves the accuracy, timeliness, and reliability of flood modeling, and can be adapted to flood simulation in mountain flood areas lacking conventional monitoring data. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the following will briefly introduce the drawings required to be used in the embodiments. Obviously, the drawings described below are only some embodiments of this application. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0017] Figure 1 It is a schematic flowchart of a method for constructing a mountain flood hydrological and hydrodynamic model and parameter calibration in data-deficient areas based on SWOT remote sensing data in an embodiment of this application.

[0018] Figure 2 It is a schematic diagram of grid runoff calculation based on infiltration capacity in an embodiment of this application.

[0019] Figure 3 It is a schematic diagram of the structure and parameter calibration of a differentiable parameter learning framework in an embodiment of this application.

[0020] Figure 4 It is a schematic diagram of the structure of a computer device provided in an embodiment of this application. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0021] Next, in combination with the accompanying drawings in the embodiments of the present application, the technical solutions in the embodiments of the present application will be clearly and completely described. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present application.

[0022] With the rapid development of remote sensing technology and artificial intelligence technology, it provides a new technical idea for flood modeling and model parameter calibration. These emerging technologies can not only solve the computational bottleneck in traditional methods, but also provide more and more accurate parameter calibration ideas. Model parameter calibration will no longer be limited by traditional manual parameter adjustment and data scarcity problems.

[0023] Based on this, the present application provides a technical solution that combines intelligent algorithms and high-resolution remote sensing data, realizing efficient, accurate, and automated parameter calibration, thereby greatly improving the accuracy, timeliness, and reliability of flood modeling. This will provide stronger technical support for flood prediction and disaster prevention and control.

[0024] To make the above objects, features, and advantages of the present application more obvious and understandable, the present application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0025] In an exemplary embodiment, as Figure 1 shown, a method for constructing and calibrating a mountain flood hydrological and hydrodynamic model in data-deficient areas based on SWOT remote sensing data is provided. This method can be executed by a computer device, specifically, it can be executed independently by a computer device such as a terminal or a server, or jointly executed by a terminal and a server. In the embodiments of the present application, the following steps 201 to step 206 are included.

[0026] Step 201, obtain the soil static environment data, environmental driving sample data, and corresponding SWOT satellite remote sensing data of the target area.

[0027] Among them, the soil static environment data includes DEM data, and may also include soil thickness data, land use data, and soil type data; the DEM data includes the actual terrain elevation of each observation point.

[0028] The SWOT satellite remote sensing data is obtained through the NASA official website. The corresponding SWOT remote sensing data in the name includes features such as the water level height and range of large-scale water bodies. The obtained high-precision observation data can provide important support for the calibration of the subsequent mountain flood hydrological and hydrodynamic model. In an application example of the present application, the SWOT satellite remote sensing data includes the radar measurement elevation of each observation point (x, y).

[0029] Step 202: Calculate the observed water depth at each observation point in the target area according to the SWOT satellite remote sensing data and the soil static environment data; the soil static environment data, the environmental driving sample data, and the observed water depths of all corresponding observation points constitute a training sample, and multiple such training samples constitute a training sample set.

[0030] In an application example of the present application, the SWOT satellite can provide the water body elevation at each observation point in the basin area. Its basic principle is based on radar elevation measurement plus a series of correction factors to eliminate errors in radar measurement. Based on this, the calculation process of the observed water depth at each observation point in the target area includes:

[0031] (21) For any observation point, calculate the observed water body elevation according to the radar elevation measurement. The calculation formula is as follows: . Where is the radar elevation measurement; corrections are correction data including earth curvature, atmospheric delay, orbit error, etc.; is the observed water body elevation of the SWOT satellite at time t.

[0032] (22) Calculate the difference between the observed water body elevation and the actual terrain elevation to obtain the observed water depth. That is, the inundation depth at each observation point is the difference between the water body elevation observed by the satellite and the actual terrain elevation of the observation point. The observed water depth at each observation point is: ; where is the actual terrain elevation of the observation point; is the inundation depth of the observation point, that is, the observed water depth.

[0033] After obtaining the observed water depths of all observation points, these points can be mapped to a regular grid through spatial interpolation, so as to obtain the spatial distribution of the entire water surface elevation. Then, in subsequent steps, it is compared with the predicted value output by the differentiable parameter learning framework to adjust the hydrological and hydrodynamic parameters in the initial mountain flood hydrological and hydrodynamic model and reduce errors.

[0034] Step 203: Based on Cellular Automaton (CA), construct an initial mountain flood hydrological and hydrodynamic model according to the soil static environment data; the initial mountain flood hydrological and hydrodynamic model contains hydrological and hydrodynamic parameters. Cellular Automaton, as the core tool for simulating the flood evolution process, divides the research area into two-dimensional grid cells (cells) to simulate the dynamic evolution process of floods in a highly flexible manner. Each cell can independently carry geographical attributes (such as terrain elevation, water depth, flow velocity, etc.) and state change rules, so as to realize the integrated simulation of the entire hydrological and hydrodynamic process.

[0035] In an application example, step 203 includes the following (31)-(36).

[0036] (31) Use the HAND method to process the DEM data to obtain the terrain elevation data to be used. To optimize the hydrological and hydrodynamic simulation, the present application uses the HAND (Height Above Nearest Drainage) method to process the DEM data. The HAND method simplifies complex terrains by calculating the relative elevation of any point on the surface relative to the nearest neighboring river channel, can effectively identify the rainwater gathering areas, and provides the head information of the slope runoff process at the same time. Its mathematical definition is as follows: ; where H HAND is the HAND value of the current observation point, that is, the terrain elevation data to be used; Z C is the altitude of the current observation point; Z ND is the altitude of the river channel nearest to the current observation point. Here, "nearest neighbor" means the nearest along the flow path, rather than the nearest in a straight-line distance.

[0037] The DEM data processed by the HAND method can accurately represent the flow path and terrain features. Through the above processing, it not only helps to improve the flood inundation simulation and the accuracy of the basin runoff generation and concentration process, but also provides high-quality terrain data input for the cellular automata model.

[0038] (32) Based on the terrain elevation data to be used, divide the target area into a cellular network; the cellular network includes multiple cells. Specifically, divide the target area into regular grids (cells) in meters, and assign a unique identifier or coordinates to each cell for dynamically recording the changes in its hydrological state.

[0039] (33) Define the state attributes corresponding to each cell; the state attributes include terrain elevation and soil characteristics; thus, key geographical attributes such as elevation are assigned to each cell. The state attributes corresponding to different target areas are different, and the initial conditions such as whether the terrain of the target area is steep, whether there is grass, and whether the soil is dry or wet can also be included.

[0040] (34) Use the Moore-type eight-neighborhood relationship to define the water flow direction and flow rules between the cells; specifically, use the flow rule driven by the height difference to simulate the water flowing from high to low.

[0041] (35) Set the initial conditions and boundary conditions for each cell; specifically, the initial conditions include the initial water depth, initial soil moisture content, outflow, etc., and the boundary conditions include drainage points, impermeable boundaries, etc.

[0042] The above settings provide an accurate spatial framework and physical basis for dynamically simulating the flood evolution process, that is, the water flow direction and flow rules in the cellular network, the state attributes of each cell, the initial conditions, and the boundary conditions constitute a two-dimensional cellular scenario;

[0043] (36) Based on the two-dimensional cellular scenario, perform integrated runoff and confluence simulation to determine a set of water flow change functions; the two-dimensional cellular scenario and the set of water flow change functions constitute an initial mountain flood hydrological and hydrodynamic model; the set of water flow change functions is used to dynamically calculate the water flow direction, water flow velocity, water flow rate, and water depth of the cell; the hydrological and hydrodynamic parameters are determined by the set of water flow change functions.

[0044] Specifically, the cellular automaton uses a grid-based spatial discretization method to divide the entire target area into interacting cells. Each cell can independently store and update the hydrological state and simulate the flood evolution process through local rules. Such discretization not only enables the fine calculation of the runoff generated by rainfall (runoff generation process) in each cell but also spontaneously forms the convergence of water flow (confluence process) through the interaction between cells. Through integrated runoff and confluence calculation and analysis, dynamically calculate the water flow direction, velocity, and water volume exchange to update the water level of each cell. Among them, the set of water flow change functions includes a cell soil infiltration capacity function, a cell-to-cell slope function, a water flow velocity and flow rate function, and a water balance function.

[0045] Rainfall is the main input of the surface hydrological cycle. A part of it will be lost, including infiltration, evaporation, vegetation interception, etc., and cannot be directly converted into surface runoff. The remaining part is called net rain, that is, the precipitation effectively participating in runoff generation. In this application, the infiltration capacity curve is used to calculate runoff generation to obtain the water depth of each cell, as Figure 2 shown, where the cell soil infiltration capacity function is:

[0046] .

[0047] Among them, r is the soil infiltration rate, used to calculate the water depth of the cell; r f is the constant infiltration rate after soil saturation, that is, the stable infiltration rate; r0 is the infiltration rate at the initial rainfall, that is, the initial infiltration rate; β is the decay constant, indicating the rate at which the infiltration rate decreases with time; t is the time.

[0048] For the confluence part, by comparing the total water levels of the central cell and the neighboring cells, the path of water flow from the central cell to the cell with lower potential energy can be determined. The cell-to-cell slope function is:

[0049] .

[0050] Among them, is the slope from the central cell c to the neighboring cell n at time t, representing the water flow direction; and are the water depths of the central cell c and the neighboring cell n at time t, respectively; and are the ground elevations of the central cell c and the neighboring cell n, respectively; is the horizontal distance from the central cell c to the neighboring cell n.

[0051] Manning's equation is an empirical formula used to calculate water flow velocity and discharge, widely applied in the fields of hydrology and hydraulic engineering. By calculating the flow velocity using Manning's formula, the discharge can then be calculated. Therefore, the water flow velocity and discharge function is:

[0052] .

[0053] .

[0054] Where is the water flow velocity from the central cell c to the neighboring cell n, is the Manning roughness coefficient; is the water flow discharge from the central cell c to the neighboring cell n; b is the cell width.

[0055] The water balance formula is the core formula in hydrology based on the law of conservation of mass. In the simulation of cellular automata, the water balance is reflected in that the change in water volume in each cell is equal to the net change in the inflow water volume minus the outflow water volume, and the dynamic update of the water depth is calculated in combination with the area of the cell. The water balance function is:

[0056] .

[0057] Where is the water depth of the i-th cell at time, is the duration of water volume change, is the water depth of the i-th cell at time t; is the water flow discharge into the central cell c; is the water flow discharge out of the central cell c; A is the cell area.

[0058] Step 204: Based on the initial mountain flood hydrological and hydrodynamic model and a preset machine learning model, construct a differentiable Parameter Learning (dPL) framework. Among them, the differentiable parameter learning framework, as a large-scale parameter learning method different from the traditional point-to-point parameter calibration method, will directly establish the global mapping relationship between the observed data and the model parameters, and at the same time use the model with physical mechanisms as the constraint conditions to achieve the effect of adaptive learning of the model parameters, as Figure 3 shown.[[ID=No.3]] [[ID=No.4]]

[0059] [[ID=No.5]]The prediction of the initial mountain flood hydrological and hydrodynamic model based on the cellular automaton is related to the external input conditions, static attributes, and parameters to be optimized. Therefore, the dynamic update of the cellular automaton simulation value is based on the following formula:[[ID=No.6]] [[ID=No.7]]

[0060] [[ID=No.8]] [[ID=No.9]].[[ID=No.10]] [[ID=No.11]]

[0061] [[ID=No.12]]Among them,[[ID=No.13]] [[ID=No.14]]is the simulated water depth at the i-th cell at time t, and is the predicted value of the differentiable parameter learning framework;[[ID=No.15]] [[ID=No.16]]is the environmental driving sample data at the i-th cell at time t, generally an external driving variable related to time and location, such as meteorological forcing data (precipitation, temperature, etc.);[[ID=No.17]] [[ID=No.18]]is the static environmental data of the soil at the i-th cell, including static information related to spatial location such as topographic DEM, soil type, and land cover;[[ID=No.19]] [[ID=No.20]]is the hydrological and hydrodynamic parameter of the i-th cell, which is an unobservable parameter specific to the location and needs to be determined separately through optimization. The hydrological and hydrodynamic parameters include the constant infiltration rate after soil saturation, the infiltration rate at the initial rainfall, the decay constant, and the Manning roughness coefficient. [[ID=No.21]] [[ID=No.22]]

[0062] [[ID=No.23]]Compared with the traditional mountain flood hydrological and hydrodynamic model, in this application, the cellular automaton uses local interaction rules to simulate the water flow evolution. Therefore, it is necessary to calibrate the parameters for each cell in detail to improve the simulation accuracy and stability. Among them, a large number of measured data required for calibration come from the training sample set in Step 202, and the parameters to be calibrated are hydrological and hydrodynamic parameters, as shown in Table 1 below.[[ID=No.24]] [[ID=No.25]]

[0063] [[ID=No.26]]Table 1[[ID=No.27]] [[ID=No.28]]

[0064] [[ID=No.29]] [[ID=No.30]] [[ID=No.31]]

[0065] [[ID=No.32]]The predicted value of the differentiable parameter learning framework[[ID=No.33]] [[ID=No.34]]and the observed value[[ID=No.35]] [[ID=No.36]]The relationship can be expressed by a mapping function, and the function formula is as follows:[[ID=No.37]] [[ID=No.38]]

[0066] [[ID=No.39]] [[ID=No.40]].[[ID=No.41]] [[ID=No.42]]

[0067] [[ID=No.43]]Among them, the observed value is the observed water depth determined from the SWOT satellite remote sensing data in the training sample set, is a mapping function used to convert the predicted value of the model into the same unit or scale as the observed value; is the error between the predicted value and the observed value, and the closer it is to 0, the better.

[0068] Step 205: Use the training sample set to train the differentiable parameter learning framework to optimize the model parameters of the preset machine learning model and determine the corresponding optimal hydrological and hydrodynamic parameters. In an application example, Step 205 includes the following steps (51)-(54).

[0069] (51) In one training iteration process, input the soil thickness data, the land use data, and the soil type data into the differentiable parameter learning framework, and the preset machine learning model generates predicted hydrological and hydrodynamic parameters.

[0070] In a specific application, the preset machine learning model uses an LSTM (Long Short-Term Memory) network. LSTM was initially developed in the field of artificial intelligence for learning sequence data and has now become a common choice for processing hydrological time series data. Different from ordinary recurrent neural networks (such as RNN), LSTM has two states (cell state and hidden state) and three gating mechanisms (input gate, forget gate, output gate). By finding a set of parameters , make the output of the physical model as close as possible to the observed value . Therefore, by capturing the time series characteristics through the LSTM network, mapping the dynamic input to the state update of the hydrological model, and backpropagating the error of the model output value to adjust the model parameters, automatic parameter calibration can be achieved.

[0071] (52)Input the predicted hydrological and hydrodynamic parameters into the initial mountain flood hydrological and hydrodynamic model in the differentiable parameter learning framework for parameter update, and then, driven by the environmental driving sample data, perform mountain flood hydrological and hydrodynamic simulation to obtain the simulated water depths at all observation points.

[0072] (53)When the preset training stop condition is not reached, calculate the loss function based on the observed water depths at all observation points corresponding to the environmental driving sample data and the simulated water depths at all observation points, and use the loss function to calculate the gradient to optimize the model parameters of the preset machine learning model, and then enter the next iteration training; different iteration trainings correspond to different training samples.

[0073] The loss function is used to measure the error between the model prediction value and the actual observation value. It also combines physical constraints to optimize the model performance to ensure that the model conforms to physical laws. The formula of the loss function is:

[0074] .

[0075] in, L total is the value of the loss function; L data is the data loss value, which is used to quantify the error between the observed water depth and the simulated water depth; is the loss value of the physical constraint term, which is used to ensure conservation of mass and momentum; is a weight parameter used to balance the priority between data fitting and physical consistency.

[0076] By adjusting the parameters , so that the model predicts With observational data The best match, thereby optimizing the model performance, correspondingly, the function formula of the data loss value is:

[0077] .

[0078] Where t is the time, i is the index of the cell, is the simulated water depth of the i-th cell at time t, To observe the water depth, is the mapping function between simulated water depth and observed water depth; is the environmental driving sample data of the i-th cell at time t, is the static environmental data of the soil of the i-th cell, is the hydrological and hydrodynamic parameter of the i-th cell; is the squared Euclidean distance, which measures the difference between the predicted and observed values.

[0079] The function formula of the loss value of the physical constraint term is:

[0080] .

[0081] Among them, q is the flow rate; y is the water depth; is the divergence of flow, which indicates the degree of change of flow with space; is the rate of change of water depth with time.

[0082] In practical applications, when performing gradient calculations, the weights of the neural network are optimized through gradient backpropagation, which indirectly optimizes the parameters of the model. The gradient of the parameters is expressed as:

[0083] .

[0084] Among them, is the gradient of the loss function L with respect to the parameter θ.

[0085] When updating the parameters, the gradient descent method is used to iteratively adjust the parameters: . Among them, is the learning rate, which is used to control the step size during each parameter update.

[0086] (54) When the preset training stop condition is reached, the predicted hydro - meteorological and hydrodynamic parameters are marked as the optimal hydro - meteorological and hydrodynamic parameters. That is, through the differentiable parameter learning framework, the parameters can be optimized. By repeating the training process, the value of the loss function is gradually reduced until the objective function converges or reaches the preset threshold, and the optimal hydro - meteorological and hydrodynamic parameters are obtained.

[0087] In addition, the present application selects the Nash - Sutcliffe efficiency coefficient (NSE) as the main error measurement index. NSE measures the fitting degree of the model prediction value to the observed value. The closer it is to 1, the better the model performance. The model performance is verified through the error index until the error converges. The calculation formula of NSE is as follows:

[0088] .

[0089] Among them, is the mean value of the observed values at all time steps t of the i - th cell.

[0090] Step 206, determine the final mountain flood hydro - meteorological and hydrodynamic model according to the optimal hydro - meteorological and hydrodynamic parameters and the initial mountain flood hydro - meteorological and hydrodynamic model.

[0091] Compared with the prior art, the present application has the following advantages:

[0092] (1) The SWOT satellite is based on the interferometric synthetic aperture radar technology, breaking through the limitation of traditional satellites that only rely on point - based observations. It can provide large - scale, grid - based hydrological data and accurately measure the height, width, and slope of floods in three dimensions. Compared with other satellites, SWOT has higher spatial and temporal resolutions and can obtain synchronous water level information over a wide area in a short time, forming a continuous planar data set. Therefore, the present application uses SWOT data, providing a new idea of "exchanging space for time". Although it cannot provide long - time - series observation data at a single point, through synchronous observations of a large - scale area, relying on spatially distributed data to make up for the lack of time data.

[0093] (2) As a hydrological and hydrodynamic simulation method, the computational units of cellular automata are essentially grid-based, which can highly match the grid-based remote sensing observation data (such as SWOT satellite data), enabling the model to provide more refined hydrodynamic state simulations at different spatial positions and showing higher computational efficiency and flexibility when dealing with discrete dynamic hydrological processes.

[0094] (3) In the process of parameter calibration of cellular automata, the core challenge of traditional methods lies in the need for complete grid-level data to match it. However, the parameter calibration of traditional flood models is mainly based on single-point observation data such as water levels and flow velocities at hydrological stations, usually using overall or local adjustment methods, which are difficult to meet the refined calibration requirements of the CA model at the grid scale. In addition, since the evolution mechanism of CA depends on the interaction between grids, simply relying on limited point observation data for interpolation or approximate optimization is difficult to accurately characterize the spatial heterogeneity of the flood process, resulting in limited calibration accuracy.

[0095] In response to this, this application conducts mountain flood hydrological and hydrodynamic process simulations based on the cellular automata model, uses the discrete characteristics of cellular automata to carry out flood process calculations, and dynamically updates the cell states to simulate the basin runoff generation and concentration processes under different hydrodynamic conditions. Among them, the differentiable parameter learning framework is mainly based on deep learning and gradient optimization mechanisms, which can optimize the parameters of all grid cells simultaneously at the global scale, thus effectively improving the efficiency and accuracy of parameter calibration. Its core advantage is that the high-resolution grid-based hydrological data provided by SWOT satellites highly matches the grid cells of the CA model, enabling the model to directly calibrate using the observation data of each cell without relying on limited point observations for interpolation or approximate optimization. In addition, this application regards the model parameters as trainable variables, uses gradient descent optimization and combines the backpropagation algorithm for dynamic adjustment, so that the model prediction results can match the SWOT observation data to the greatest extent, achieving more refined and efficient parameter optimization.

[0096] Considering the time-varying nature of the flood process, a long short-term memory network is further introduced to capture the spatio-temporal variation patterns of parameters, ensuring that the calibrated parameters can adapt to the hydrodynamic characteristics under different time steps and different basin conditions. This method not only breaks through the limitation of traditional calibration methods relying on limited observation points, but also greatly improves the spatial resolution, time adaptability and computational efficiency of calibration, providing a more accurate and physically constrained solution for flood process simulation.

[0097] In summary, this application is different from the traditional point-to-point parameter calibration method. It directly establishes a global mapping relationship between the observed data and the model parameters, and uses a model with physical mechanisms as a constraint condition to achieve adaptive adjustment of the parameters, thereby improving the accuracy and automation of the model. At the same time, it can also fully consider the similarity between geographical processes, that is, applying the knowledge learned in one region to other regions, so as to effectively utilize large-scale observed data. This application constructs a differentiable parameter learning framework, effectively making up for the computational bottleneck of the traditional model calibration method, enabling the calibration of flash flood model parameters to no longer be limited by traditional manual parameter adjustment and data scarcity problems. At the same time, through the combination of intelligent algorithms and high-resolution remote sensing data, efficient, accurate, and automated parameter calibration is achieved, thus greatly improving the accuracy, timeliness, and reliability of flood modeling, which will provide stronger technical support for flood prediction and disaster prevention and control.

[0098] In an exemplary embodiment, a computer device is provided. The computer device can be a server or a terminal, and its internal structure diagram can be as Figure 4 shown. The computer device includes a processor, a memory, an input / output interface (Input / Output, abbreviated as I / O), and a communication interface. Among them, the processor, the memory, and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The input / output interface of the computer device is used to exchange information between the processor and external devices. The communication interface of the computer device is used to communicate with external terminals through a network connection. When the computer program is executed by the processor, it realizes a method for constructing and calibrating a mountain flood hydrological and hydrodynamic model in a data-deficient area based on SWOT remote sensing data.

[0099] Those skilled in the art can understand that Figure 4 the structure shown in

[0100] is only a block diagram of some structures related to the solution of this application, and does not constitute a limitation on the computer device to which the solution of this application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine some components, or have different component arrangements.

[0101] In an exemplary embodiment, a computer-readable storage medium is provided, storing a computer program which, when executed by a processor, implements the steps in the above method embodiments.

[0102] In an exemplary embodiment, a computer program product is provided, including a computer program which, when executed by a processor, implements the steps in the above method embodiments.

[0103] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use, and processing of relevant data need to comply with relevant regulations.

[0104] Those of ordinary skill in the art can understand that all or part of the processes in the above method embodiments can be completed by instructing relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the above method embodiments. Among them, any reference to a memory, database, or other medium used in the embodiments provided in this application can include at least one of non-volatile and volatile memories. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc.

[0105] In each of the embodiments provided in this application, the databases involved may include at least one of a relational database and a non-relational database. The non-relational database may include a distributed database based on blockchain, etc., and is not limited thereto. In each of the embodiments provided in this application, the processor may be a general-purpose processor, a central processing unit, a graphics processing unit, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., and is not limited thereto.

[0106] The technical features of the above embodiments can be combined arbitrarily. For the sake of concise description, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, it should be considered as the scope described in this specification.

[0107] Specific examples are used in this article to elaborate on the principles and implementation manners of this application. The description of the above embodiments is only used to help understand the method and its core idea of this application; at the same time, for those of ordinary skill in the art, according to the idea of this application, there will be changes in the specific implementation manners and application scopes. In summary, the content of this specification should not be construed as a limitation to this application.

Claims

1. A method for constructing a mountain flood hydrological and hydrodynamic model and calibrating parameters in data-deficient areas based on SWOT remote sensing data, characterized in that, The method includes: Obtaining soil static environment data, environmental driving sample data, and corresponding SWOT satellite remote sensing data of the target area; the soil static environment data includes soil thickness data, land use data, and soil type data; [[ID=?]]Calculating the observed water depth of each observation point in the target area according to the SWOT satellite remote sensing data and the soil static environment data; the soil static environment data, the environmental driving sample data, and the observed water depths of all corresponding observation points form a training sample, and multiple said training samples form a training sample set; Based on the cellular automaton, constructing an initial mountain flood hydrological and hydrodynamic model according to the soil static environment data; the initial mountain flood hydrological and hydrodynamic model includes hydrological and hydrodynamic parameters; Based on the initial mountain flood hydrological and hydrodynamic model and a preset machine learning model, constructing a differentiable parameter learning framework; the differentiable parameter learning framework is expressed as: ； Among them, is the simulated water depth of the i-th cell at time t; is the environmental driving sample data of the i-th cell at time t; is the soil static environmental data of the i-th cell; is the hydrological and hydrodynamic parameter of the i-th cell; Using the training sample set to train the differentiable parameter learning framework to optimize the model parameters of the preset machine learning model and determine the corresponding optimal hydrological and hydrodynamic parameters; including: In one training iteration process, inputting the soil thickness data, the land use data, and the soil type data into the differentiable parameter learning framework, and generating predicted hydrological and hydrodynamic parameters by the preset machine learning model; Inputting the predicted hydrological and hydrodynamic parameters into the initial mountain flood hydrological and hydrodynamic model in the differentiable parameter learning framework for parameter update, and then performing mountain flood hydrological and hydrodynamic simulation under the drive of the environmental driving sample data to obtain the simulated water depths of all observation points; When the preset training stop condition is not reached, calculating a loss function based on the observed water depths of all observation points corresponding to the environmental driving sample data and the simulated water depths of all observation points, and using the loss function to calculate the gradient to optimize the model parameters of the preset machine learning model, and then entering the next iteration training; different iteration trainings correspond to different training samples; the formula of the loss function is: ； Among them, L total is the value of the loss function; L data is the data item loss value, which is used to quantify the error between the observed water depth and the simulated water depth; is the physical constraint item loss value, which is used to ensure mass conservation and momentum conservation; is the weight parameter; When the preset training stop condition is reached, marking the predicted hydrological and hydrodynamic parameters as the optimal hydrological and hydrodynamic parameters; Determining the final mountain flood hydrological and hydrodynamic model according to the optimal hydrological and hydrodynamic parameters and the initial mountain flood hydrological and hydrodynamic model.

2. The method for constructing and parameter calibration of a mountain flood hydrological hydrodynamic model in data - scarce areas based on SWOT remote sensing data according to claim 1, wherein, The SWOT satellite remote sensing data includes the radar measurement elevation of each observation point; the soil static environment data includes DEM data, and the DEM data includes the actual terrain elevation of each observation point; The calculation process of the observed water depth of each observation point in the target area includes: For any observation point, calculating the observed water body elevation according to the radar measurement elevation; Calculating the difference between the observed water body elevation and the actual terrain elevation to obtain the observed water depth.

3. A method for constructing and calibrating parameters of a mountain flood hydrological hydrodynamic model in data - scarce areas based on SWOT remote sensing data according to claim 1, characterized in that, The soil static environment data includes DEM data; The construction process of the initial mountain flood hydrological and hydrodynamic model includes: Using the HAND method to process the DEM data to obtain the terrain elevation data to be used; Based on the terrain elevation data to be used, dividing the target area into a cellular network; the cellular network includes multiple cells; Define the state attributes corresponding to each of the cells; the state attributes include terrain elevation and soil properties; Define the water flow direction and flow rules between the cells using the Moore-type eight-neighborhood relationship; Set the initial conditions and boundary conditions for each of the cells; the water flow direction and flow rules, the state attributes of each cell, the initial conditions and boundary conditions in the cell network constitute a two-dimensional cell scenario; Based on the two-dimensional cell scenario, conduct an integrated simulation of runoff generation and concentration to determine a set of water flow change functions; the two-dimensional cell scenario and the set of water flow change functions constitute an initial mountain flood hydrological and hydrodynamic model; the set of water flow change functions is used to dynamically calculate the water flow direction, water flow velocity, water flow discharge, and water depth of the cells; the hydrological and hydrodynamic parameters are determined by the set of water flow change functions.

4. A method for constructing and calibrating parameters of a mountain flood hydrological and hydrodynamic model in data-deficient areas based on SWOT remote sensing data according to claim 3, characterized in that The set of water flow change functions includes a cell soil infiltration capacity function, an inter-cell slope function, a water flow velocity and discharge function, and a water balance function; The cell soil infiltration capacity function is: ； where r is the soil infiltration rate, which is used to calculate the water depth of the cell; r f is the constant infiltration rate after the soil is saturated; r0 is the infiltration rate at the initial rainfall; β is the decay constant; t is the time; The inter-cell slope function is: ； Among them, is the slope from the central cell c to the neighboring cell n at time t, representing the water flow direction; , are the water depths of the central cell c and the neighboring cell n at time t, respectively; , are the ground elevations of the central cell c and the neighboring cell n, respectively; is the horizontal distance from the central cell c to the neighboring cell n; The water flow velocity and discharge function is: ； ； Among them, is the water flow velocity from the central cell c to the neighboring cell n, is the Manning roughness coefficient; is the water flow rate from the central cell c to the neighboring cell n; b is the cell width. The water balance function is: ； Among them, is the water depth of the i-th cell at time, is the water volume change duration, is the water depth of the i-th cell at time t; is the water flow rate flowing from the j-th cell into the central cell c; is the water flow rate from the central cell c to the neighboring cell n; A is the cell area.

5. A method for constructing and calibrating parameters of a mountain flood hydrological and hydrodynamic model in data-deficient areas based on SWOT remote sensing data according to claim 4, characterized in that, The hydrological and hydrodynamic parameters include the constant infiltration rate after soil saturation, the infiltration rate during initial rainfall, the decay constant, and the Manning roughness coefficient.

6. A method for constructing and parameter calibration of a mountain flood hydrological and hydrodynamic model in data-deficient areas based on SWOT remote sensing data according to claim 1, characterized in that, The function formula for the data item loss value is: ； where t is the time, i is the index of the cell, is the simulated water depth of the i-th cell at time t, is the observed water depth, is the mapping function between the simulated water depth and the observed water depth; is the environmental driving sample data of the i-th cell at time t, is the soil static environmental data of the i-th cell, is the hydro-hydraulic parameter of the i-th cell; is the square of the Euclidean distance.

7. A method for constructing and parameter calibration of a mountain flood hydrological and hydrodynamic model in data - scarce areas based on SWOT remote - sensing data according to claim 6, characterized in that, The function formula for the physical constraint item loss value is: ； where q is the flow rate; y is the water depth; is the divergence of the flow rate; is the rate of change of the water depth with time.

Images