Gridding distributed hydrological simulation method

Abstract

The invention discloses a gridding distributed hydrological simulation method, and relates to the technical field of hydrological models. Comprising the following steps: acquiring historical hydrology and environmental factor data of a target drainage basin, dividing the target drainage basin into a plurality of grid units, and screening key grids; sensitive parameters which have obvious influence on hydrological simulation are determined through sensitivity analysis. Dividing the simulation time period into a plurality of sub-time periods, generating a near-optimal solution set of each sub-time period through an optimization algorithm, and realizing historical space-time two-dimensional dynamic identification of key grid sensitive parameters through a dynamic planning algorithm; non-key grid parameters are obtained through spatial interpolation. And training a machine learning model through the historical environment factor data and the historical space-time two-dimensional dynamic parameters. Inputting environmental factor data of the basin to be measured into the trained machine learning model to obtain prediction parameters, and applying the prediction parameters to the hydrological model to obtain a simulated flow process. According to the method, collaborative dynamic optimization of the parameters in the time dimension and the space dimension can be realized, and the hydrological simulation precision is effectively improved.

Description

Technical Field

[0001] The present invention relates to the technical field of hydrological models, and in particular to a grid-based distributed hydrological simulation method. Background Art

[0002] Hydrological models describe real-world hydrological processes through certain assumptions and empirical formulas and are the core of hydrological forecasting. The parameters of most conceptual hydrological models have clear physical meanings and play a crucial role in accurate hydrological simulation and forecasting. However, these parameters cannot usually be obtained through direct measurement; instead, they are commonly calibrated using historical data. However, with climate change and rapid economic and social development, constant parameters calibrated from historical data may become unsuitable for future scenarios, significantly reducing the accuracy of runoff forecasts and introducing errors and risks into water resource assessment, optimal allocation, and scheduling.

[0003] Compared with hydrological models that use constant parameters, accounting for the temporal variation of parameters (i.e., time-varying variability) can improve runoff simulations under changing environments. Methods for identifying time-varying parameters in lumped conceptual hydrological models have been extensively studied, with extensive research yielding results from the Budyko model (based on multi-year averages), the two-parameter monthly water balance model (on monthly scales), and even the Xin'anjiang model (on daily or hourly scales). However, a major drawback of lumped conceptual hydrological models is that they treat the entire watershed as a single entity, with uniform inputs and parameters across the entire basin. Consequently, when applied to identifying parameter responses to changing environments, they can only reflect macroscopic patterns. This is particularly true when applied to large-scale watersheds, where climatic conditions and underlying surface conditions are highly spatially heterogeneous. Lumped conceptual hydrological models can only simulate spatially uniform virtual states of rainfall, evaporation, topography, vegetation, and land use, yielding homogenized results.

[0004] For the identification of time-varying parameters of distributed hydrological models, existing studies mainly use simple segmented calibration methods, which ignore the spatial heterogeneity and time-varying nature of grid parameters and are insufficient to more precisely characterize the evolution process of hydrological model parameters under changing environments. Summary of the Invention

[0005] Based on this, a grid-based distributed hydrological simulation method is provided to address the problem that the time-varying and spatial heterogeneity of grid parameters in existing distributed hydrological models are not well described.

[0006] An embodiment of the present invention provides a grid-based distributed hydrological simulation method, including: Obtain historical hydrological data and historical environmental factor data for the target basin, and divide the target basin into multiple grid cells using a grid-based distributed hydrological model. Select key grids from the multiple grid cells based on the historical hydrological data for the target basin. Key grids are used to reflect the spatial inconsistency of the basin's hydrology. The sensitivity analysis method is used to evaluate the impact of different hydrological parameters on the basin hydrological simulation in the key network, and the sensitivity of the hydrological parameters is obtained. The hydrological parameters with sensitivity greater than the set threshold are regarded as sensitive parameters. The historical hydrological data is divided into multiple sub-periods. In each sub-period, an optimization algorithm is used to generate a near-optimal solution set of sensitive parameters of key grids in each sub-period. The near-optimal solution set of sensitive parameters of key grids in each sub-period is dynamically estimated in two dimensions in space and time using a dynamic programming algorithm to obtain the two-dimensional dynamic sensitive parameters of key grids. Based on the two-dimensional dynamic sensitive parameters of key grids, the two-dimensional dynamic sensitive parameters of non-key grids are determined using a spatial interpolation method to achieve the identification of the two-dimensional dynamic parameters of space and time for all grids in all periods in the target basin, and obtain the historical two-dimensional dynamic parameters of the target basin. The historical environmental factor data of the target watershed is used as the input of the machine learning model, and the historical spatiotemporal two-dimensional dynamic parameters of the target watershed are used as the output of the machine learning model to train the machine learning model and obtain a trained machine learning model; The real-time environmental factor data of the measured basin is input into the trained machine learning model to obtain the predicted spatiotemporal two-dimensional dynamic parameters, and the predicted spatiotemporal two-dimensional dynamic parameters are applied to the gridded distributed hydrological model to obtain the simulated flow process of the measured basin.

[0007] Optionally, key grids are selected from multiple grid cells based on historical hydrological data of the target basin, including: Based on the historical hydrological data of the target basin, the number and location of key grids are used as optimization variables, and the maximum Nash efficiency coefficient is used as the objective function. The dynamic dimension search method is used for calibration to obtain the optimal number, location and parameter constant values ​​of key grids. Among them, the objective function is to maximize the Nash efficiency coefficient based on the following formula: ; in, is the Nash efficiency coefficient, for t The observed value at time, for t The analog value at time n The total length of time.

[0008] Optionally, the historical hydrological data is divided into multiple sub-periods. In each sub-period, an optimization algorithm is used to generate a near-optimal solution set for sensitive parameters of key grids within each sub-period, specifically including: For each sub-period, an optimization algorithm is used to generate a set of near-optimal solutions for the sensitive parameters of the key grid that are close to the optimal objective function value; The optimal objective function value is determined based on the following formula: ; in, is the optimal objective function value, for t The observed value at time, is the average of the runoff observations.

[0009] Optionally, a spatiotemporal two-dimensional dynamic estimation is performed on a near-optimal solution set of sensitive parameters of key grids in each sub-period using a dynamic programming algorithm based on the following formula to obtain spatiotemporal two-dimensional dynamic sensitive parameters of the key grids, specifically including: ; in, is the spatiotemporal two-dimensional dynamic sensitive parameter of the key grid, for +1 runoff simulation accuracy index for the sub-period, For the Sub-period The first parameter q The estimated value of the key grid, is the maximum value of the parameter, is the minimum value of the parameter, is the number of parameters, is the total number of grids, is the weight factor, N is the total number of sub-periods.

[0010] Optionally, the spatial and temporal two-dimensional dynamic sensitive parameters of the non-critical grids are determined by a spatial interpolation method based on the spatial and temporal two-dimensional dynamic sensitive parameters of the critical grids, specifically including: The weight of each key grid is calculated based on the Euclidean distance between each non-key grid and the key grid according to the following formula: ; in, For the The weight of the key grid, is the Euclidean distance between the non-critical grid and the critical grid, is the distance attenuation coefficient, For all Key grid sum; The sensitive parameter values ​​of the key grids are weighted and averaged to obtain the estimated values ​​of the sensitive parameters of each non-key grid.

[0011] Optionally, the hydrological data includes: meteorological data, geographical data and runoff data; Meteorological data include: gridded precipitation, relative humidity, wind speed, and temperature; Geographic data include digital elevation model data, land use and cover data, soil type data, and flow direction data; Runoff data are flow data; Environmental factor data include: meteorological data, land use and cover data, leaf area index, socioeconomic factors and reservoir construction data.

[0012] The grid-based distributed hydrological simulation method provided by the embodiment of the present invention has the following beneficial effects compared with the prior art: The present invention discretizes the watershed into key grids and non-key grids through a gridded distributed hydrological model, and combines it with a dynamic programming algorithm to realize two-dimensional spatiotemporal continuous estimation of parameters. It not only captures the non-uniform distribution characteristics (spatial heterogeneity) of hydrological parameters in the spatial dimension, but also reflects the evolution law (time-varying) of parameters with changes in environmental factors such as climate and underlying surface through dynamic optimization on the time axis, realizing the coordinated dynamic optimization of hydrological parameters in the temporal and spatial dimensions, and more finely depicting the evolution process of hydrological parameters under a changing environment.

[0013] In addition, an explicit mapping relationship between parameters and space-time coordinates is established through machine learning models, which has dynamic adaptive capabilities driven by physical mechanisms, and improves the parameter identification granularity from the "section level" to the "grid-moment level", more accurately depicting the dynamic response process of hydrological parameters and enhancing the model's hydrological simulation accuracy under future environmental changes. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 The present invention is a flowchart of a grid-based distributed hydrological simulation method provided in one embodiment. DETAILED DESCRIPTION

[0015] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0016] In one embodiment, a grid-based distributed hydrological simulation method is provided, the method comprising: Historical hydrological and environmental factor data for the target basin are obtained and divided into multiple grid cells using a gridded distributed hydrological model. Key grid cells are selected from these multiple grid cells based on the historical hydrological data of the target basin. Key grid cells are used to reflect the spatial inconsistency of the basin's hydrology.

[0017] The sensitivity analysis method is used to evaluate the impact of different hydrological parameters on the basin hydrological simulation in the key network, and the sensitivity of the hydrological parameters is obtained. The hydrological parameters with sensitivity greater than the set threshold are regarded as sensitive parameters.

[0018] Historical hydrological data is divided into multiple sub-periods. Within each sub-period, an optimization algorithm is used to generate a set of near-optimal solutions for the sensitive parameters of key grids within each sub-period. A dynamic programming algorithm is then used to perform a spatiotemporal two-dimensional dynamic estimation of these near-optimal solutions for the sensitive parameters of key grids within each sub-period, obtaining the spatiotemporal two-dimensional dynamic sensitive parameters of the key grids. Based on these spatiotemporal two-dimensional dynamic sensitive parameters of the key grids, spatial interpolation is used to determine the spatiotemporal two-dimensional dynamic sensitive parameters of non-key grids. This allows the identification of the spatiotemporal two-dimensional dynamic parameters for all grids in the target basin across all time periods, ultimately obtaining the historical spatiotemporal two-dimensional dynamic parameters of the target basin.

[0019] The historical environmental factor data of the target watershed is used as the input of the machine learning model, and the historical spatiotemporal two-dimensional dynamic parameters of the target watershed is used as the output of the machine learning model to train the machine learning model and obtain a trained machine learning model.

[0020] The real-time environmental factor data of the measured basin is input into the trained machine learning model to obtain the predicted spatiotemporal two-dimensional dynamic parameters, and the predicted spatiotemporal two-dimensional dynamic parameters are applied to the gridded distributed hydrological model to obtain the simulated flow process of the measured basin.

[0021] like Figure 1 As shown, the specific implementation includes: Step 1: Select a grid-based distributed hydrological model, organize the basin's hydrological data, and collect meteorological, geographical, runoff, and environmental factor data; Step 2: Based on meteorological and hydrological data, all grids are divided into key grids and non-key grids. (For example, if a basin has 50 grids, calibrating the parameters of all 50 grids separately will result in the curse of dimensionality. The solution is to divide all grids in the basin into key grids and non-key grids. Assume that the key grid refers to the grid that best reflects the spatial inconsistency of the basin's hydrological processes. Only the number, location, and key values ​​of the key grids need to be determined. The parameters of other grids are interpolated based on the law of geographic similarity to obtain the parameter values ​​of the entire grid.) Key grids are considered to be the grids that best reflect the spatial inconsistency of the basin. Their number, location, and initial parameters are obtained by optimizing and calibrating the hydrological model using the Nash efficiency coefficient (Equation 1) as the objective function.

[0022] (1) Where, is the Nash efficiency coefficient, for t The observed value at time, for t The analog value at time n The total length of time.

[0023] Step 3: Combined with the sensitivity analysis method, the influence of each parameter of the hydrological model on the grid-based distributed hydrological model is evaluated. Specifically, the hydrological model is simulated to obtain the sensitivity of runoff simulation under different parameters. The parameters are ranked by sensitivity, and the top-ranked parameters (e.g., the top 30%) are selected as sensitive parameters. The following steps only identify time-varying effects on sensitive parameters, while non-sensitive parameters are fixed to constant values ​​(obtained in step 1), significantly reducing the parameter optimization dimension.

[0024] Step 4: Divide the hydrological model data into multiple sub-periods, and use the optimization algorithm to generate a near-optimal objective function value for each sub-period ( NSE The set of near-optimal solutions of the sensitive parameters of the key grids (e.g., containing 1000 sets of sensitive parameters of the key grids) is the largest.

[0025] (2) Where, is the optimal objective function value, for t The observed value at time, is the average of the runoff observations.

[0026] Step 5: Use a dynamic programming algorithm to perform spatiotemporal two-dimensional dynamic estimation on the near-optimal solution set of the sensitive parameters of the key grids in each sub-period to obtain the spatiotemporal two-dimensional dynamic sensitive parameters of the key grids.

[0027] A dynamic programming algorithm is used to select the sensitive parameters of the key grids in each sub-period, with the balance between model accuracy and parameter continuity as the objective function, to achieve spatiotemporal dynamic identification of the key grid sensitive parameters that takes into account both model efficiency and continuity: (3) Where, is the spatiotemporal two-dimensional dynamic sensitive parameter of the key grid, for +1 runoff simulation accuracy index for the sub-period, For the Sub-period The first parameter qThe estimated value of the key grid, is the maximum value of the parameter, is the minimum value of the parameter, is the number of parameters, is the total number of grids, is the weight factor representing the importance of parameter continuity, N is the total number of sub-periods. As increases, the model performance may decrease slightly, but the temporal continuity of the parameters will be enhanced.

[0028] Step 6: For each sub-period, the sensitive parameters of the non-critical grids are generated by a spatial interpolation method such as Inverse Distance Weight (IDW) based on the sensitive parameters of the critical grids.

[0029] Specifically, the weight of each key grid is calculated according to the Euclidean distance between each non-key grid and the key grid (Equation 4), and the sensitive parameter values ​​of the key grids are weighted averaged according to the weights to obtain the estimated value of the sensitive parameters of each non-key grid, thereby realizing the spatiotemporal two-dimensional dynamic parameter identification of all grids in all time periods.

[0030] (4) in, For the The weight of the key grid, is the Euclidean distance between the non-critical grid and the critical grid, is the distance attenuation coefficient, For all Key grid The sum of , in order to achieve weight normalization, to ensure .

[0031] Step 7: Based on a machine learning model, establish a quantitative relationship between the spatiotemporal dynamic parameters and environmental factors (such as climate, land use, and human activities). Specifically, the machine learning model (such as the random forest algorithm or XGB algorithm) uses historical environmental factor data as input and historical spatiotemporal dynamic parameters as output for model training. After model training is complete, the model uses predicted scenario data for future environmental factors as input and outputs simulated values ​​of the spatiotemporal dynamic parameters under future scenarios, providing a quantitative basis for parameter selection in future scenarios.

[0032] In step 8, the machine learning model is applied to the gridded distributed hydrological model using the parameters simulated by environmental factors to obtain the simulated flow process of the test basin, and its accuracy is evaluated with indicators.

[0033] A specific embodiment of the present invention is provided, comprising: Step 1: Select the hydrological model VIC (Variable Infiltration Capacity) as the case model. This is a grid-based distributed hydrological model that is widely used in large-scale water and energy balance calculations. The grid resolution used in this example is 0.5 0.5 Table 1 summarizes the seven key parameters used in model calibration and testing. The meteorological, geographical and runoff data of a watershed in the embodiment from January 1, 1990 to December 31, 2015 were collected and sorted for parameter calibration of the VIC model. The meteorological data included gridded precipitation, relative humidity, wind speed and temperature. The geographical data included digital elevation model (DEM) data, land use and cover data, soil type data and flow direction data. Runoff refers to flow data. It is also necessary to collect data on other environmental factors that may affect the watershed parameters, including leaf area index (LAI), socioeconomic factors and reservoir construction data.

[0034] Table 1 VIC model parameters to be calibrated Step 2: Screen the number and location of key grids. Specifically, based on the data from the rate period, the number and location of key grids are used as optimization variables. Together with all the parameters of the VIC model, they are calibrated using the dynamic dimension search (Dissertation Discovery System, DDS) method with the maximum Nash efficiency coefficient as the objective function to obtain the optimal number, location, and parameter constant values ​​of key grids. In this case, the optimization results show that there are two key grids in a certain basin, located in the south and center of the basin respectively. It is determined that the parameters that need to be dynamically estimated in this case are the parameters of these two key grids.

[0035] Step 3: Screening sensitive parameters involves performing sensitivity analysis on the parameters of the VIC model using periodic utilization data. The sensitivity analysis method uses the MCMC (Markov Chain Monte Carlo) algorithm. This method simulates random sampling, performs a large number of attempts within the range of possible parameter values, records the output of the model for each attempt, and selects the most sensitive parameters. In this case, the top 30% are selected as B and Depth2 , fix the remaining non-sensitive parameters to the constant values ​​obtained in step 2.

[0036] Step 4: Divide the hydrological model data of the rate period into multiple sub-periods, with each sliding window lasting one year. In the embodiment, the window length is set to 5 years, and the rate period is divided into 15 sub-periods, such as 1990-1994, 1991-1995, ..., 2004-2008. Use DDS to generate a near-optimal objective function value ( NSE This means that for each sensitive parameter, the optimization result is not a single optimal value, but rather multiple near-optimal parameter values. Thus, for each sub-period, a near-optimal solution set consisting of 1,000 sensitive parameter groups is generated.

[0037] In step 5, a dynamic programming algorithm is used to select the sensitive parameters of the key grid from the set of near-optimal solutions in each sub-period. The objective function is to balance the model accuracy and parameter continuity, and form a dynamic trajectory of the key grid sensitive parameters covering all time periods, thereby realizing the spatiotemporal two-dimensional dynamic identification of the key grid sensitive parameters that takes into account both model efficiency and continuity.

[0038] (5) Where, is the spatiotemporal two-dimensional dynamic sensitive parameter of the key grid, for +1 runoff simulation accuracy index for the sub-period; Indicates the Sub-period The first parameter q An estimate of the key grid; and are the maximum and minimum values ​​of the parameters respectively; is the number of parameters; is the total number of grid cells; is a weight factor indicating the importance of parameter continuity; N is the total number of sub-periods. As increases, the model performance may decrease slightly, but the temporal continuity of the parameters will be enhanced.

[0039] Step 6: For each sub-period, the sensitive parameters of the non-critical grids are generated by a spatial interpolation method such as IDW based on the sensitive parameters of the critical grids to obtain the spatiotemporal two-dimensional dynamic parameters of all grids in all periods.

[0040] Step 7: Through the spatiotemporal two-dimensional dynamic estimation of the parameters of the VIC model, the embodiment obtains 15 (the number of sub-periods) 43 (number of grids) groups of parameters are used to simulate the relationship between the parameters and the watershed environmental factors collected in step 1 using a random forest regression model.

[0041] In step 8, the parameters simulated using environmental factors from the random forest regression model for the rate period (January 1, 1990, to December 31, 2009) and the test period (January 1, 2010, to December 31, 2015) were applied to the VIC model to obtain runoff for the rate period and the test period. Compared to the traditional constant parameter method, this method can effectively improve the simulation accuracy of the distributed hydrological model (Table 2).

[0042] Table 2 Comparison of the effects of the present invention and the constant parameter method in flow simulation The above-described embodiments merely illustrate several implementations of the present invention. While the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that a person skilled in the art would be able to make numerous variations and improvements without departing from the spirit of the present invention, and all such variations and improvements fall within the scope of protection of the present invention.

Claims

1. A grid-based distributed hydrological simulation method, characterized in that: include: Obtain historical hydrological data and historical environmental factor data of the target basin, and divide the target basin into multiple grid cells using a grid-based distributed hydrological model; Selecting key grids from multiple grid cells based on historical hydrological data of the target basin, wherein the key grids are used to reflect the spatial inconsistency of the basin's hydrology; The sensitivity analysis method is used to evaluate the impact of different hydrological parameters on the basin hydrological simulation in the key network, and the sensitivity of the hydrological parameters is obtained. The hydrological parameters with sensitivity greater than the set threshold are regarded as sensitive parameters. The historical hydrological data is divided into multiple sub-periods. In each sub-period, an optimization algorithm is used to generate a near-optimal solution set of sensitive parameters of key grids in each sub-period. Then, the dynamic programming algorithm is used to perform a spatiotemporal two-dimensional dynamic estimation of the near-optimal solution set of the sensitive parameters of the key grids in each sub-period, and the spatiotemporal two-dimensional dynamic sensitive parameters of the key grids are obtained; Based on the spatiotemporal two-dimensional dynamic sensitive parameters of the key grids, the spatiotemporal two-dimensional dynamic sensitive parameters of the non-key grids are determined by spatial interpolation method to realize the spatiotemporal two-dimensional dynamic parameter identification of all grids in all time periods in the target basin, and obtain the historical spatiotemporal two-dimensional dynamic parameters of the target basin; The historical environmental factor data of the target watershed is used as the input of the machine learning model, and the historical spatiotemporal two-dimensional dynamic parameters of the target watershed are used as the output of the machine learning model to train the machine learning model and obtain a trained machine learning model; The real-time environmental factor data of the measured basin is input into the trained machine learning model to obtain the predicted spatiotemporal two-dimensional dynamic parameters, and the predicted spatiotemporal two-dimensional dynamic parameters are applied to the gridded distributed hydrological model to obtain the simulated flow process of the measured basin.

2. A grid-based distributed hydrological simulation method according to claim 1, characterized in that: The step of selecting key grids from multiple grid cells based on the historical hydrological data of the target basin specifically includes: Based on the historical hydrological data of the target basin, the number and location of key grids are used as optimization variables, and the Nash efficiency coefficient is used as the objective function. The dynamic dimension search method is used to calibrate and obtain the optimal number, location and parameter constant values ​​of key grids. Among them, the Nash efficiency coefficient is used as the objective function based on the following formula: ； in, is the Nash efficiency coefficient, for t The observed value at time, for t The analog value at time n The total length of time.

3. A grid-based distributed hydrological simulation method according to claim 1, characterized in that: The historical hydrological data is divided into multiple sub-periods. In each sub-period, an optimization algorithm is used to generate a near-optimal solution set of sensitive parameters of key grids in each sub-period, specifically including: For each sub-period, an optimization algorithm is used to generate a set of near-optimal solutions for the sensitive parameters of the key grid that are close to the optimal objective function value; The optimal objective function value is determined based on the following formula: ； in, is the optimal objective function value, for t The observed value at time, is the average of the runoff observations.

4. A grid-based distributed hydrological simulation method according to claim 1, characterized in that: Based on the following formula, the dynamic programming algorithm is used to perform a spatiotemporal two-dimensional dynamic estimation of the near-optimal solution set of the sensitive parameters of the key grids in each sub-period, and the spatiotemporal two-dimensional dynamic sensitive parameters of the key grids are obtained, specifically including: ； in, is the spatiotemporal two-dimensional dynamic sensitive parameter of the key grid, for +1 runoff simulation accuracy index for the sub-period, For the Sub-period The first parameter q The estimated value of the key grid, is the maximum value of the parameter, is the minimum value of the parameter, is the number of parameters, is the total number of grids, is the weight factor, N is the total number of sub-periods.

5. A grid-based distributed hydrological simulation method according to claim 1, characterized in that: The method of determining the spatiotemporal two-dimensional dynamic sensitive parameters of non-key grids by a spatial interpolation method based on the spatiotemporal two-dimensional dynamic sensitive parameters of the key grids specifically includes: The weight of each key grid is calculated based on the Euclidean distance between each non-key grid and the key grid according to the following formula: ； in, For the The weight of the key grid, is the Euclidean distance between the non-critical grid and the critical grid, is the distance attenuation coefficient, For all Key grid sum; The sensitive parameter values ​​of the key grids are weighted and averaged to obtain the estimated values ​​of the sensitive parameters of each non-key grid.

6. A grid-based distributed hydrological simulation method according to claim 1, characterized in that: The hydrological data include: meteorological data, geographical data and runoff data; Meteorological data include: gridded precipitation, relative humidity, wind speed, and temperature; Geographic data include digital elevation model data, land use and cover data, soil type data, and flow direction data; Runoff data are flow data; The environmental factor data include: meteorological data, land use and cover data, leaf area index, socioeconomic factors and reservoir construction data.

Images