Drainage basin hydrological prediction method and system based on meteorological driving and causal reconstruction

Abstract

The invention relates to a watershed hydrological prediction method and system based on meteorological driving and causal reconstruction, and belongs to the technical field of watershed hydrological prediction. The method comprises the following steps: constructing a causal structure diagram between meteorological driving variables and between meteorological driving variables and rainfall to form an interpretable meteorological and rainfall causal diagram, explicitly converting a functional relationship in a sub-module into a series of structural equations by taking a traditional hydrological model as a construction basis, and uniformly embedding the structural equations into a causal model framework to form a meteorological and rainfall causal diagram; and a unified structural causal graph is constructed through a graph structure merging technology, and the whole process description of a causal path from external climate factors to drainage basin runoff and runoff response is realized. According to the method, a physical mechanism, statistical learning and causal reasoning are fused, the defect that a traditional model lacks explanatory ability and generalization ability when coping with non-stationary meteorological conditions is overcome, and a system modeling method with causal support is provided for flood forecasting, water resource management and extreme climate response simulation.

Description

TECHNICAL FIELD

[0001] The present application relates to a method and system for predicting basin hydrology based on meteorological driving and causal reconstruction, and belongs to the technical field of basin hydrology prediction and model construction. BACKGROUND

[0002] In the field of flood control, commonly used hydrological models are mainly divided into physical (process) driving and data driving. Physical driving is to analyze the variables in the hydrological process and the causal relationship between variables, collect hydrological process knowledge, and model the hydrological process knowledge. Data driving is to collect hydrological elements (historical rainfall data, historical flow data, historical evaporation data, etc.) for prediction training modeling.

[0003] The main problems of the prior art are: 1. Insufficient model explanation: existing hydrological models lack mechanism explanation of the causal relationship between meteorological driving and hydrological response; 2. Limited model stability and generalization ability: fixed structure empirical models are prone to structural failure or performance decline when dealing with climate change or unconventional meteorological conditions; 3. Unable to perform scientific intervention reasoning and counterfactual prediction: under extreme weather or regulatory change scenarios, traditional models cannot provide causal response mechanisms at the variable level, limiting their application in decision support. Current models generally cannot simulate the intervention effect between system variables, nor can they conduct policy simulation and hypothesis testing under non-observed scenarios, which does not meet the new requirements of flood regulation, urban resilience construction, agricultural irrigation optimization, etc. on decision controllability, hydrological response sensitivity and counterfactual simulation; 4. Unclear error sources, lack of system fault diagnosis capability: when existing models have simulation errors, it is difficult to determine whether the problem is input, structural assumption or parameter setting. SUMMARY

[0004] The purpose of the present application is to overcome the above-mentioned deficiencies and provide a method for predicting basin hydrology based on meteorological driving and causal reconstruction. By combining a structural causal model (Structural Causal Model) with a traditional hydrological model such as the Xin'anjiang model, a method for constructing a meteorological-hydrological causal model with strong explanation, intervention, and system integration is proposed.

[0005] The technical solution adopted by the present application is: The method for predicting basin hydrology based on meteorological driving and causal reconstruction comprises the following steps: S1. Collecting multiple key meteorological factor data in the basin; S2. Using the structural causal modeling method, the precipitation causal path is first identified, and all meteorological factor variable nodes that have a causal impact on precipitation are identified. Their influence intensity and direction are quantified, so as to intervene in the simulation calculation of the direct and indirect causal impact on precipitation under the change of a certain meteorological factor variable. Finally, an interpretable meteorological causal map and structural path are obtained. S3. Transform all modules in the traditional hydrological model into structural equations to describe the causal structure; S4. Using historical observation data of the watershed, the parameters of each structural equation are calibrated using the structural equation modeling method to realize the numerical reconstruction of the structural equation of the hydrological model and transform the modules of the traditional hydrological model into the causal structure of hydrological elements. S5. Connect the causal structure of hydrological elements with the meteorological causal diagram to obtain a joint causal model structure, forming a causal path from external meteorological drivers to hydrological responses; S6. Use joint causal models for hydrological prediction.

[0006] The above method also includes steps such as intervention response simulation and counterfactual path tracing based on the obtained causal structure for specific scenarios. The intervention response simulation is performed through intervention operations. The simulation of changes in variables affects intermediate and runoff output variables; counterfactual path tracing simulates the system response under hypothetical scenarios based on obtained measured results, in order to support post-hoc evaluation and early warning optimization.

[0007] The above methods also include introducing error tracking and model optimization mechanisms into the joint causal model based on the causal structure: This includes path sensitivity calculation: simulating the amplification effect of model variable perturbations on causal paths; Error attribution decomposition: Based on the actual deviation, it is decomposed into three categories: input error, structural error, and estimation error, and traced back to their respective paths; Model improvement mechanism: Based on the maximum path deviation, it is recommended to reconstruct the structure function curve, reduce the dimensionality of nodes, or add auxiliary variables.

[0008] In the above method, the meteorological driving factors mentioned in step S1 include, but are not limited to, precipitation, temperature, humidity, wind speed, atmospheric pressure, and radiation.

[0009] In step S2, causal path identification uses a Directed Acyclic Graph (DAG) constructed from variables based on the Additive Noise Model (ANM) and the Peter-Clark PC algorithm.

[0010] In step S3, the Xin'anjiang model is preferred as the traditional hydrological model, and its modules include runoff generation module, evapotranspiration module, and runoff confluence module.

[0011] Another object of the present application is to provide a basin hydrological prediction system based on meteorological driving and causal reconstruction, comprising A data acquisition module acquires multiple key meteorological factor data in the basin. A meteorological causal diagram construction module: using structural causal modeling method, first performing precipitation causal path identification, identifying all meteorological factor variable nodes having causal influence on precipitation, quantifying the action strength and direction, and performing intervention simulation calculation on the direct and indirect causal influence of a certain meteorological factor variable change on precipitation, finally obtaining an interpretable meteorological causal diagram and structural path. A hydrological structural function reconstruction module: all modules in the traditional hydrological model are converted into structural equations, using historical observation data of the basin, using structural equation modeling method to calibrate the parameters of each structural equation, realizing numerical reconstruction of the structural equation of the hydrological model, and analyzing each module in the model into a structural causal form. A joint causal model structure generation module: used to connect the hydrological element causal structure with the meteorological causal diagram with the same element, obtain a joint causal model structure, and form a causal path from external meteorological driving to hydrological response. A hydrological prediction module uses the joint causal model to perform hydrological prediction.

[0012] The basin hydrological prediction system based on meteorological driving and causal reconstruction further comprises a causal reasoning and counterfactual evaluation module, which realizes intervention simulation reasoning, scenario evolution simulation and counterfactual tracking analysis.

[0013] The basin hydrological prediction system based on meteorological driving and causal reconstruction further comprises an error attribution and sensitivity analysis module: identifying the source of deviation, realizing path vector attribution and model optimization suggestion.

[0014] The present application has the following beneficial effects: 1. Improve the causal interpretability and process transparency of the model: The present application breaks through the limitations of traditional hydrological models based on empirical formula or statistical correlation modeling, and for the first time introduces structural causal modeling method in the whole process of meteorological driving-precipitation-runoff response. By clearly describing the causal relationship path between variables, the physical consistency and causal explanation of the model are enhanced, and the cognitive ability of the model to the response mechanism of complex environmental systems is improved.

[0015] 2. Support multi-level intervention simulation and counterfactual reasoning ability: Compared with the traditional model which only has prediction function, the present method can perform intervention simulation calculation based on the causal diagram, support scenario simulation and causal evaluation of external intervention (such as temperature change, land use change, water conservancy project scheduling, etc.), and expand the policy auxiliary decision function of the model.

[0016] 3. Realize the structural attribution and path correction of model error: By constructing a hierarchical causal graph model, the propagation process of model prediction error in the causal path can be tracked, and high sensitivity variables or structures can be identified, thereby realizing the directional analysis and optimization of model precision problems and improving the systematicness and efficiency of model debugging and optimization.

[0017] 4. Enhance the adaptability and cross-region migration ability of the model: Compared with traditional fixed structure parameter models, the present application effectively improves the generalization ability of the model under different regions and different hydrological and meteorological conditions by describing the general mechanism between driving variables and response variables through causal structure. This method is particularly suitable for areas where data is scarce or model migration is difficult, and has a significant advantage in regional model customization.

[0018] 5. Improve the fusion ability of the model to multi-source heterogeneous data: The causal structure modeling framework provided by the present application has high structural flexibility, can effectively integrate multi-source data such as remote sensing data, meteorological observation data, artificial activity information and expert knowledge, adapt to the development trend of modern hydrological monitoring data diversification and unstructured, and significantly enhance the adaptability and response ability of the model to big data environment.

[0019] 6. Promote the fusion and innovation of physical models and data-driven models: Through the organic combination of causal modeling path and traditional Xinanjiang model structure, the effective fusion between physical process constraint and causal structure learning is realized, the robustness and expansibility of the model are improved, and a theoretical basis and technical path are provided for constructing the next generation of intelligent hydrological model with interpretability, generalization and reasoning ability.

[0020] The present application combines physical mechanism, statistical learning and causal reasoning, overcomes the defects of traditional models in dealing with non-stationary weather conditions, such as lack of explanatory power and generalization ability, and provides a causal support system modeling method for flood prediction, water resources management and extreme climate response simulation. The method has good generalizability and practicality, and is suitable for complex watershed hydrological model construction and evolution research. BRIEF DESCRIPTION OF DRAWINGS

[0021] Figure 1 The flowchart of the method of the present application; Figure 2 The meteorological causal structure diagram obtained by the PC algorithm of the present application; Figure 3 The joint causal model structure diagram obtained by the present application. DETAILED DESCRIPTION

[0022] The following will be further illustrated by specific embodiments.

[0023] The embodiment 1, a watershed hydrological prediction method based on weather driving and causal reconstruction, comprises steps (such as Figure 1 ) as follows: S1. Collecting data of multiple key weather factors in the watershed, including but not limited to precipitation, temperature, humidity, wind speed, atmospheric pressure and radiation; S2. Using structural causal modeling method, first performing causal path identification of precipitation, identifying all weather factor variable nodes that have causal influence on precipitation, quantifying the action strength and direction, and simulating the direct and indirect causal influence of a weather factor variable change on precipitation, finally obtaining an interpretable weather causal graph and structural path: (1) Data preprocessing and standardization; (2) Application of causal identification technology: based on structural discovery algorithms such as additive noise model (ANM) and PC (Peter-Clark) algorithm, combined with expert heuristics to construct directed acyclic graph DAG (Directed Acyclic Graph) for variables; (3) Precipitation causal path identification: identifying all upstream variable nodes that have causal influence path on precipitation, quantifying the action strength and direction; (4) Intervention simulation calculation: calculating the direct and indirect causal influence of a weather variable change on precipitation by intervention simulation do(X).

[0024] Taking the Yishu River Basin as an example, the hydrological causal structure model of the Yishu River Basin is constructed by using era5 reanalysis meteorological data, and meteorological data of the watershed from 2015 to 2020, including precipitation, evaporation, runoff, wind speed, 2-meter dew point temperature, 2-meter air temperature, surface soil temperature, surface latent heat flux, surface sensible heat flux, and surface soil water, are collected. In this embodiment, era5 meteorological data is selected, and the causal structure graph is obtained by PC algorithm and additive noise algorithm as Figure 2 , wherein the meanings and units of the variables are shown in Table 1.

[0025] Table 1 Causal model variable table .

[0026] Further, the front door criterion and the back door criterion can be used to analyze the causal effect between variables, and the linear regression model can be used to calculate the average treatment effect and the conditional treatment effect, which are used to quantitatively analyze the influence of one variable on another variable in the causal graph.

[0027] S3. All modules in the traditional hydrological model are converted into structural equations to describe the causal structure: Take the Xin'anjiang model as an example to reconstruct the model structure equation. The Xin'anjiang model has a clear module division and multiple core intermediate variables (evaporation coefficient, runoff coefficient, free water storage, etc.).

[0028] Convert all module formula forms in the Xin'anjiang model, including the runoff module, evaporation module, and confluence module, into structural equation representation: , where, represents the target variable, is all its causal parent nodes, is the disturbance term.

[0029] For the Xin'anjiang model, specifically for the runoff module, the transformed structural equation is , For the evaporation module, it is: , For the water source module, it is: , For the confluence module, it is , where is the effective rainfall, is the soil moisture content, is the maximum soil water storage capacity, is the water storage capacity curve shape parameter, is the potential evapotranspiration, is the infiltration or deep seepage, K is the confluence coefficient, is the surface runoff, is the soil flow, is the groundwater runoff, , represents the error term.

[0030] S4. Using historical observation data of the basin, the structural equation modeling method is used to calibrate the parameters of each structural equation, realize the numerical reconstruction of the structural equation of the hydrological model, and convert each module of the traditional hydrological model into a causal structure of hydrological elements: The parameter calibration method generally uses genetic algorithm, ant colony optimization algorithm, and SCEUA algorithm.

[0031] S5. Connect the hydrological element causal structure with the meteorological causal graph with the same elements to obtain a joint causal model structure, forming a causal path from external meteorological driving to hydrological response: The hydrological model modules such as runoff generation and confluence are converted into nested causal subgraphs, and are connected with the meteorological causal graph, so that the entire system graph can depict the complete causal path level from external driving to hydrological response.

[0032] After the meteorological causal graph and the hydrological model structure equation are reconstructed, the application constructs a joint causal model structure through a causal graph merging method, wherein: each variable node has a clear physical semantic; all arrows in the causal graph show the influence direction and eliminate the circular structure; the response calculation can be performed through intervention simulation operation between paths; the causal comparison and path decomposition are supported for all process variables.

[0033] The output is a joint causal graph , wherein: represents all variable nodes participating in modeling (including meteorological and hydrological variables); represents a causal edge with a clear direction between variables, such as .

[0034] Taking the Yi-Shu River Basin as an example, based on the meteorological element causal graph obtained in the foregoing steps and the hydrological element causal structure obtained by modifying the Xin'anjiang model, a meteorological-hydrological element causal graph is obtained as Figure 3 , wherein k is the confluence coefficient, PET is the potential evapotranspiration, Q is the basin outflow, and the meanings of the remaining variables are the same as before.

[0035] S6. Hydrological prediction using a joint causal model.

[0036] Based on the obtained causal structure, the application proposes the following causal analysis mechanism for scenarios such as potential extreme weather, policy regulation or land use change: (1) Intervention response simulation, through intervention operation , simulate the response of intermediate variables and runoff output variables after variable change; (2) Counterfactual path tracking, on the basis of the obtained measured results, simulate the system response under the hypothetical situation to support post-evaluation and early warning optimization.

[0037] Further, the application introduces an error tracking and model optimization mechanism based on the causal structure to avoid error amplification, and the key mechanisms are as follows: (1) Path sensitivity calculation: simulate the amplification effect of model variable disturbance on the causal path; (2) Error attribution decomposition: based on the actual deviation, decompose into three categories of input error, structural error and estimation error, and trace back to the corresponding path; (3) Model improvement mechanism: according to the maximum path deviation, suggest to reconstruct the curve of the structure function, reduce the dimension of the node or increase the auxiliary variable.

[0038] Embodiment 2: A basin hydrological prediction system based on meteorological driving and causal reconstruction, comprising: a data collection module for collecting data of multiple key meteorological factors in the basin; a meteorological causal diagram construction module: using structural causal modeling method, first performing causal path identification of precipitation, identifying all meteorological factor variable nodes having causal influence on precipitation, quantifying the action strength and direction, and simulating the direct and indirect causal influence of a certain meteorological factor variable change on precipitation, to finally obtain an interpretable meteorological causal diagram and structural path; a hydrological structural function reconstruction module: converting all modules in the traditional hydrological model into structural equations, using historical observation data of the basin, and using structural equation modeling method to calibrate the parameters of each structural equation, to realize numerical reconstruction of the structural equation of the hydrological model, and analyze each module in the model into structural causal form; a joint causal model structure generation module: for connecting the hydrological element causal structure with the meteorological causal diagram of the same element to obtain a joint causal model structure, forming a causal path from external meteorological driving to hydrological response; a hydrological prediction module for performing hydrological prediction using the joint causal model.

[0039] The basin hydrological prediction system based on meteorological driving and causal reconstruction further comprises a causal reasoning and counterfactual evaluation module for realizing intervention simulation reasoning, scenario evolution simulation and counterfactual tracking analysis.

[0040] The basin hydrological prediction system based on meteorological driving and causal reconstruction further comprises an error attribution and sensitivity analysis module for identifying bias sources, realizing path vector attribution and model optimization suggestions.

[0041] The above is a further description of the present application in combination with embodiments, and the protection scope of the present application is not limited thereto.

Claims

1. A watershed hydrological forecasting method based on meteorological driving forces and causal reconstruction, characterized by: The steps include the following: S1. Collect data on multiple key meteorological factors within the basin; S2. Using structural causal modeling, the causal path of precipitation is first identified, and all meteorological factor variable nodes that have a causal impact on precipitation are identified. Their influence intensity and direction are quantified, and the direct and indirect causal impacts on precipitation under the influence of a certain meteorological factor variable are simulated and calculated. Finally, an interpretable meteorological causal map is obtained. S3. Transform all modules in the traditional hydrological model into structural equations to describe the causal structure; S4. Using historical observation data of the watershed, the parameters of each structural equation are calibrated using the structural equation modeling method to realize the numerical reconstruction of the structural equation of the hydrological model and transform the modules of the traditional hydrological model into the causal structure of hydrological elements. S5. Connect the causal structure of hydrological elements with the meteorological causal diagram to obtain a joint causal model structure, forming a causal path from external meteorological drivers to hydrological responses; S6. Use joint causal models for hydrological prediction.

2. The watershed hydrological prediction method based on meteorological driving and causal reconstruction according to claim 1, characterized in that, It also includes steps such as intervention response simulation and counterfactual path tracing based on the obtained causal structure for specific scenarios. Intervention response simulation is performed through intervention operations. The response of the simulated variables to intermediate variables and runoff output variables after changes in the simulated variables; Counterfactual path tracing simulates the response if the actual situation did not occur, based on actual test results, to assist in posterior evaluation and early warning improvement.

3. The watershed hydrological prediction method based on meteorological driving and causal reconstruction according to claim 1, characterized in that, It also includes introducing error tracking and model optimization mechanisms into the joint causal model based on the causal structure: This includes path sensitivity calculation: simulating the amplification effect of model variable perturbations on causal paths; Error attribution decomposition: Based on the actual deviation, it is decomposed into three categories: input error, structural error, and estimation error, and traced back to their respective paths; Model improvement mechanism: Based on the maximum path deviation, the structure function is reconstructed as a curve, the nodes are reduced in dimensionality, or auxiliary variables are added.

4. The watershed hydrological prediction method based on meteorological driving and causal reconstruction according to claim 1, characterized in that, The meteorological driving factors mentioned in step S1 include, but are not limited to, precipitation, temperature, humidity, wind speed, atmospheric pressure, and radiation.

5. The watershed hydrological prediction method based on meteorological driving and causal reconstruction according to claim 1, characterized in that, In step S2, causal path identification uses a directed acyclic graph (DAG) constructed from variables based on an additive noise model and a PC algorithm.

6. The watershed hydrological prediction method based on meteorological driving and causal reconstruction according to claim 1, characterized in that, In step S3, the Xin'anjiang model is preferred as the traditional hydrological model, and its modules include runoff generation module, evapotranspiration module, and runoff confluence module.

7. A watershed hydrological forecasting system based on meteorological driving forces and causal reconstruction, characterized by: include: The data acquisition module collects data on multiple key meteorological factors within the watershed. Meteorological causal map construction module: Using structural causal modeling, the precipitation causal path is first identified, and all meteorological factor variable nodes that have a causal impact on precipitation are identified. Their influence intensity and direction are quantified, and the direct and indirect causal impacts on precipitation under the intervention simulation calculation are calculated. Finally, an interpretable meteorological causal map is obtained. Hydrological structure function reconstruction module: It transforms all modules in the traditional hydrological model into structural equations, uses historical watershed observation data, and adopts structural equation modeling method to calibrate the parameters of each structural equation, realizes the numerical reconstruction of the hydrological model structural equations, and parses each module in the model into a structural causal form. Joint causal model structure generation module: used to connect the causal structure of hydrological elements with the meteorological causal map and other elements to obtain a joint causal model structure, forming a causal path from external meteorological drivers to hydrological responses; The hydrological prediction module uses a joint causal model to predict hydrological conditions.

8. The watershed hydrological prediction system based on meteorological driving and causal reconstruction according to claim 7, characterized in that, It also includes a causal reasoning and counterfactual assessment module, enabling intervention simulation reasoning, scenario evolution simulation, and counterfactual tracking analysis.

9. The watershed hydrological prediction system based on meteorological driving and causal reconstruction according to claim 7, characterized in that, It also includes an error attribution and sensitivity analysis module: identifying the sources of deviation, enabling path vector attribution and model optimization suggestions.

Images