Runoff reanalysis data simulation precision influence factor evaluation method and system

Abstract

The invention relates to the technical field of hydrological analysis, and provides a runoff reanalysis data simulation precision influence factor evaluation method and system, and the method comprises the following steps: obtaining runoff reanalysis data, observation runoff data and target drainage basin attributes; performing simulation effect evaluation on the runoff reanalysis data on the basis of a Klinlin-Cuprit efficiency coefficient; constructing a random forest regression model; wherein the random forest regression model constructs a nonlinear relationship according to a simulation effect evaluation result of the runoff reanalysis data and features of target drainage basin attributes, and important features of the drainage basin attributes are extracted through a recursive elimination method to obtain simulation precision influence factors of the runoff reanalysis data; and explaining the random forest regression model by using an explainable machine learning method to obtain a runoff reanalysis data simulation precision influence factor evaluation result.

Description

Technical Field

[0001] The present invention relates to the technical field of hydrological analysis, and more specifically, to a method and system for evaluating factors affecting simulation accuracy of runoff reanalysis data. Background Art

[0002] Runoff reanalysis data is a dataset that combines ground observations, satellite remote sensing, and numerical weather prediction models. It generates consistent spatiotemporal maps of land, ocean, and atmospheric variables by assimilating these observational data into global numerical models. Runoff reanalysis data can provide hydrological data such as temperature, precipitation, evapotranspiration, soil moisture, and runoff. It usually includes surface runoff, groundrunoff, and total runoff, where total runoff is the sum of surface runoff and groundrunoff. Runoff reanalysis data has become an important basis for simulating and evaluating water resources at global, continental, and river basin scales. It is very important for hydrological applications such as driving hydrological models, flood and drought monitoring and forecasting.

[0003] At present, using machine learning methods to construct the mapping relationship between hydrological input variables and output variables for basin runoff simulation has become one of the current research hotspots. As one of the popular machine learning methods, long short-term memory neural network (LSTM) has been widely studied and applied in the field of runoff simulation. For example, basin runoff simulation and forecasting research based on LSTM model. Most of these simulation methods directly input prediction factors into the model. However, due to the multi-source of runoff reanalysis data, the uncertainty of the model is increased, which affects the accuracy and computational efficiency of runoff simulation. In order to improve the accuracy of runoff simulation, it is particularly important to evaluate its influencing factors. At present, it is proposed to use statistical evaluation indicators such as relative bias (Bias), root mean square error (RMSE), mean absolute error (MAE), correlation coefficient (CC) and other indicators to quantitatively evaluate the accuracy of runoff simulation. It is also proposed to use LSTM to capture the association between long time series data and extract the influencing factors in runoff reanalysis data. However, the above methods do not consider the nonlinear relationship between basin attributes and simulation accuracy, resulting in low accuracy of the evaluation results of the factors affecting the simulation accuracy of runoff reanalysis data. Summary of the invention

[0004] In order to overcome the defect that the above-mentioned method for evaluating influencing factors in runoff reanalysis data does not consider the nonlinear relationship between watershed attributes and simulation accuracy, resulting in low evaluation accuracy, the present invention provides a method and system for evaluating influencing factors of runoff reanalysis data simulation accuracy.

[0005] In order to solve the above technical problems, the technical solution of the present invention is as follows:

[0006] A method for evaluating factors affecting simulation accuracy of runoff reanalysis data includes the following steps:

[0007] Obtaining runoff reanalysis data, observed runoff data and target watershed properties, and evaluating the simulation effect of the runoff reanalysis data based on the Kling-Gupta efficiency coefficient;

[0008] Constructing a random forest regression model; wherein the random forest regression model constructs a nonlinear relationship based on the simulation effect evaluation results of the runoff reanalysis data and the characteristics of the target watershed attributes, extracts important characteristics of the watershed attributes through a recursive elimination method, and obtains the factors affecting the simulation accuracy of the runoff reanalysis data;

[0009] The random forest regression model was interpreted using an interpretable machine learning method to obtain the evaluation results of factors affecting the simulation accuracy of runoff reanalysis data.

[0010] Furthermore, the present invention also proposes a system for evaluating factors affecting the simulation accuracy of runoff reanalysis data, and applies the method for evaluating factors affecting the simulation accuracy of runoff reanalysis data proposed in the present invention. The system comprises:

[0011] Reanalysis data evaluation model, used to evaluate the simulation effect of input runoff reanalysis data and observed runoff data based on the Kling-Gupta efficiency coefficient;

[0012] An influencing factor screening module is equipped with a random forest regression model; wherein the random forest regression model constructs a nonlinear relationship based on the simulation effect evaluation results of the runoff reanalysis data and the characteristics of the target watershed attributes, extracts important characteristics of the watershed attributes through a recursive elimination method, and obtains the influencing factors of the simulation accuracy of the runoff reanalysis data;

[0013] The influencing factor assessment module is used to interpret the random forest regression model using an interpretable machine learning method to obtain the assessment results of the influencing factors of the runoff reanalysis data simulation accuracy.

[0014] Furthermore, the present invention also proposes a device, including a memory and a processor, wherein the memory stores computer-readable instructions, wherein when the computer-readable instructions are executed by the processor, the processor executes all or part of the steps of the method for evaluating factors affecting the simulation accuracy of runoff reanalysis data as described in the present invention.

[0015] Furthermore, the present invention also proposes a storage medium having computer-readable instructions stored thereon, wherein the computer-readable instructions, when executed by a processor, implement all or part of the steps of the method for evaluating factors affecting the simulation accuracy of runoff reanalysis data as described in the present invention.

[0016] Compared with the prior art, the technical solution of the present invention has the following beneficial effects:

[0017] The present invention identifies key watershed attributes from multiple watershed attributes in a large sample watershed data set through a random forest model, establishes a nonlinear relationship between watershed attributes and simulation accuracy by constructing a random forest regression model of watershed attributes and simulation accuracy, and finally interprets the nonlinear relationship between watershed attributes and simulation effects in the random forest regression model through an interpretable machine learning method. The partial dependence diagram of each key watershed attribute and the simulation effect can be obtained, so as to clearly know how the watershed attributes affect the simulation accuracy of the reanalysis data.

[0018] The present invention can compare the seasonal differences in the simulation accuracy of runoff reanalysis data caused by basin attributes in each season by analyzing the nonlinear relationship between basin attributes and simulation accuracy in each season. On the one hand, it provides support for the application of runoff reanalysis data in runoff simulation in basins without data; on the other hand, it can provide a certain reference basis for the improvement of global hydrological models, thereby improving the effect of runoff simulation of global hydrological models. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 The present invention is a flowchart of a method for evaluating factors affecting runoff reanalysis data simulation accuracy according to one embodiment of the present invention.

[0020] Figure 2 It is a schematic diagram of the simulation effect of runoff reanalysis data according to an embodiment of the present invention.

[0021] Figure 3 A Spearman correlation heat map of the simulation effect of runoff reanalysis data and watershed attributes shown in accordance with an embodiment of the present invention.

[0022] Figure 4 FIG. 4 is a schematic diagram of a ten-fold cross validation according to an embodiment of the present invention.

[0023] Figure 5 This is a diagram showing the importance ranking of watershed attributes affecting KGE according to an embodiment of the present invention.

[0024] Figure 6 It is a schematic diagram showing the importance of watershed attributes to the correlation coefficient, the coefficient of variation and the coefficient of deviation according to one embodiment of the present invention.

[0025] Figure 7 The figure is a spatial cluster distribution diagram of the SHAP values ​​of watershed attributes according to one embodiment of the present invention.

[0026] Figure 8 This is a schematic diagram showing the seasonal importance of watershed attributes according to an embodiment of the present invention.

[0027] Fig. 9 FIG. 4 is a seasonal partial dependence diagram of watershed attributes according to an embodiment of the present invention.

[0028] Fig.10 The present invention is a schematic diagram of an architecture of a system for evaluating factors affecting runoff reanalysis data simulation accuracy according to an embodiment of the present invention. DETAILED DESCRIPTION

[0029] Exemplary embodiments will be described in detail herein, examples of which are shown in the accompanying drawings. When the following description refers to the drawings, the same numbers in different drawings represent the same or similar elements unless otherwise indicated. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present invention. Instead, they are merely examples of devices and methods consistent with some aspects of the present invention as detailed in the appended claims.

[0030] The terms used in the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention. The singular forms "a", "the" and "the" used in the present invention and the appended claims are also intended to include plural forms unless the context clearly indicates other meanings. It should also be understood that the term "and / or" used herein refers to and includes any or all possible combinations of one or more associated listed items.

[0031] It should be understood that although the terms first, second, third, etc. may be used in the present invention to describe various information, these information should not be limited to these terms. These terms are only used to distinguish the same type of information from each other. For example, without departing from the scope of the present invention, the first information may also be referred to as the second information, and similarly, the second information may also be referred to as the first information. Depending on the context, the word "if" as used herein may be interpreted as "at the time of" or "when" or "in response to determining".

[0032] The present invention is described in detail below with reference to the accompanying drawings and specific embodiments.

[0033] Example 1

[0034] This embodiment proposes a method for evaluating factors affecting the simulation accuracy of runoff reanalysis data, such as Figure 1 , which is a flow chart of the method for evaluating factors affecting the simulation accuracy of runoff reanalysis data of this embodiment.

[0035] The method for evaluating factors affecting the simulation accuracy of runoff reanalysis data proposed in this embodiment includes the following steps:

[0036] S1. Obtaining runoff reanalysis data, observed runoff data and target watershed attributes, and evaluating the simulation effect of the runoff reanalysis data based on the Kling-Gupta efficiency coefficient;

[0037] S2. Constructing a random forest regression model; wherein the random forest regression model constructs a nonlinear relationship based on the simulation effect evaluation results of the runoff reanalysis data and the characteristics of the target watershed attributes, extracts important characteristics of the watershed attributes through a recursive elimination method, and obtains the factors affecting the simulation accuracy of the runoff reanalysis data;

[0038] S3. Use interpretable machine learning methods to interpret the random forest regression model and obtain the evaluation results of factors affecting the simulation accuracy of runoff reanalysis data.

[0039] In this embodiment, the key watershed attributes are identified from multiple watershed attributes in a large sample watershed data set through a random forest model, and a random forest regression model is used to establish a nonlinear relationship between watershed attributes and simulation accuracy. Finally, the nonlinear relationship between watershed attributes and simulation effects in the random forest regression model is explained through an interpretable machine learning method. The partial dependence diagram of each key watershed attribute and the simulation effect can be obtained, which can clearly show how the watershed attributes affect the simulation accuracy of the reanalysis data.

[0040] This embodiment can compare the seasonal differences in the simulation accuracy of runoff reanalysis data due to basin attributes in each season by analyzing the nonlinear relationship between basin attributes and simulation accuracy in each season. On the one hand, it provides support for the application of runoff reanalysis data in runoff simulation in basins without data. On the other hand, it can provide a certain reference basis for the improvement of the global hydrological model, thereby improving the effect of runoff simulation of the global hydrological model.

[0041] In an optional embodiment, in step S1, the simulation effect evaluation of the runoff reanalysis data is performed based on the Kling-Gupta efficiency coefficient, comprising the following steps:

[0042] S1.1. Extract the runoff reanalysis simulation data of each station according to the latitude and longitude coordinates of the hydrological station;

[0043] S1.2. Calculate the correlation r between the observed runoff data and the runoff reanalysis simulation data, the average error β and the variation error γ of the runoff reanalysis data, which are further used to calculate the Kling-Gupta efficiency coefficient (KGE); its expression is:

[0044]

[0045]

[0046]

[0047]

[0048] Where r is the Pearson linear correlation coefficient, which represents the correlation between the observed runoff data and the runoff reanalysis data; β is the mean error, which is used to quantify the ability of the runoff reanalysis data to simulate the water balance; γ is the variation error, which is used to quantify the ability of the runoff reanalysis data to replicate the observed runoff variability; sim t represents the runoff reanalysis data at time t, obs t represents the observed runoff data at time t; μ sim represents the average value of runoff reanalysis data, μ obs represents the average value of observed runoff data; σ sim represents the standard deviation of runoff reanalysis data, σ obs Represents the standard deviation of observed runoff data.

[0049] In this embodiment, KGE is combined with three statistical metrics: correlation coefficient, mean error, and variance error to provide a comprehensive model performance evaluation index. Among them, for the three components of KGE, the optimal value of the Pearson linear correlation coefficient r is 1, indicating that the trend of the runoff reanalysis is completely consistent with the observed data; the optimal value of the mean error β is 1, indicating that the mean values ​​of the runoff reanalysis and the observed data are the same; the optimal value of the variance error γ is 1, indicating that the coefficient of variation of the runoff reanalysis and the observed data is equal. The optimal value of the Kling-Gupta efficiency coefficient KGE is also 1, indicating that the model prediction is completely consistent with the observed data.

[0050] Further optionally, in step S1.1, based on the latitude and longitude coordinates of the hydrological station, the nearest grid point of the hydrological station and the other 8 grid points around it are found from the runoff reanalysis data, and the runoff reanalysis data time series of these 9 grid points are extracted; the KGE between the runoff reanalysis data time series and the observed runoff data time series is calculated, and the grid point position with the largest KGE is selected as the coordinate of the runoff reanalysis of the hydrological station, thereby determining the runoff reanalysis simulation time series of the hydrological station.

[0051] Furthermore, in an optional embodiment, the following steps are also included:

[0052] S1.3. Use the Spearman rank correlation coefficient and P value to test the linear correlation between the simulation effect of runoff reanalysis data and the watershed attributes. s , and generate a Spearman rank correlation coefficient heat map. Its expression is:

[0053]

[0054] Where ρ is the Spearman rank correlation coefficient; R(sim t ) represents the rank of runoff reanalysis data; R(obs t) represents the rank of observed runoff data; cov(R(sim t ),R(obs t )) represents the covariance between two ranks; It represents the product of the standard deviation of the runoff reanalysis data rank and the standard deviation of the observed runoff data rank.

[0055] In this example, the results of the Spearman rank correlation coefficient heat map and the P-value test correlation show that the relationship between the watershed attributes and the simulation accuracy of the runoff reanalysis data is nonlinear, indicating the necessity of using a random forest regression model to construct a nonlinear relationship between the watershed attributes and the simulation accuracy of the reanalysis data.

[0056] In an optional embodiment, in step S2, the step of constructing a random forest regression model includes:

[0057] S2.1, screening predictors from target watershed attributes;

[0058] S2.2, using the simulation effect evaluation results of runoff reanalysis data as dependent variables and the normalized prediction factors as predictor variables to construct a random forest regression model;

[0059] S2.3. Calculate the importance score of each predictor, and select the predictors that affect the prediction accuracy of the model as important features of watershed attributes through recursive elimination method.

[0060] Furthermore, in an optional embodiment, step S2 further includes the following steps:

[0061] Perform C2M conversion on the simulation effect index; its expression is:

[0062]

[0063] Among them, KGE * Usually limited to the range of [-1,1]; β * represents the transformed deviation coefficient, γ * Represents the coefficient of variation after transformation. After C2M transformation, β * and γ * The maximum value is 1 for best performance.

[0064] In this example, in order to facilitate the interpretation of large sample data sets and avoid the bias introduced by KGE with large negative values, the C2M (Cloud-to-Mesh) conversion method proposed by Mathevet et al. (2006) is used to further construct a regression model. * As the dependent variable, the normalized predictors were used as predictor variables to build a random forest regression model.

[0065] Among them, the C2M conversion of the simulation effect indicators will not change the simulation performance rankings among different watersheds.

[0066] Further, in an optional embodiment, step S2.3 includes the following steps:

[0067] The runoff reanalysis data and the observed runoff data are input into the random forest regression model, and the random forest regression model obtains the importance score of each predictor by calculating the contribution of each predictor to the splitting node of the decision tree.

[0068] Exemplarily, the importance score includes a Gini index or information gain.

[0069] Among them, in the random forest model, the importance of features can be evaluated by calculating the contribution of each feature to the splitting node of the decision tree; the importance score of the feature (such as the Gini index or information gain) is usually used to judge the importance of each feature.

[0070] Furthermore, based on the feature importance scores, the least important features are removed from the dataset; the remaining features are used to train the random forest model again and its performance is evaluated. 2 ; When the features are eliminated, the performance of the model R 2 When it stops improving or starts to decrease, the recursive process stops and the features left are considered the most important features for model prediction.

[0071] Furthermore, in an optional embodiment, the step S2 further includes the following steps:

[0072] S2.4. Use ten-fold cross validation to train the random forest regression model to obtain the optimal model parameters.

[0073] The specific steps include:

[0074] Obtain a training data set, and randomly divide the training data set into 10 subsets of equal size;

[0075] Perform 10 cross validations. In each iteration, use 9 subsets as training data and the remaining 1 subset as test data.

[0076] The average performance results of 10 cross-validations were taken as the evaluation index of model performance, and the model parameters with the best performance were saved to obtain the trained random forest regression model.

[0077] In an optional embodiment, in step S3, the random forest regression model is explained using an interpretable machine learning method, including the following steps:

[0078] Calculate the SHAP values ​​of the important features selected in the random forest regression model;

[0079] Sort in descending order according to the size of the SHAP value to obtain the importance ranking of each important feature;

[0080] The SHAP importance of the important features to the Kling-Gupta efficiency coefficient was visualized using partial dependence plots, and the nonlinear trends in the scatter parameter regression fitting plots were used to obtain the evaluation results of the factors affecting the simulation accuracy of the runoff reanalysis data.

[0081] Among them, the interpretable machine learning method SHAP uses the principle of Shapley value to quantify the contribution of each feature. Its core idea is to attribute the changes in the model prediction results to the contribution of each feature. Specifically, it first obtains the "baseline value" of the model, which is calculated as the average of all predicted values. Subsequently, SHAP decomposes the importance of each input feature by evaluating the contribution of each independent variable to the final prediction, and then obtains the evaluation results of the factors affecting the simulation accuracy of the runoff reanalysis data.

[0082] Its expression is:

[0083]

[0084]

[0085] Among them, g(z′) is the explanatory model; φ0 is the mean of the predicted values ​​output by the random forest regression model, z′ i ∈{0,1} M is a binary variable, indicating the identifier of the predictor i. A value of 1 indicates that the information of the predictor is retained, and a value of 0 indicates that the information of the predictor is removed; φ i is the SHAP value of predictor i; M is the number of predictors; S is the set of non-zero indexes of predictors; f x is the expected value of the function under the condition of the input feature subset.

[0086]

[0087] Among them, I j represents the SHAP feature importance, j and i represent the input variables and data samples respectively, and n is the total number of data samples.

[0088] In this embodiment, the SHAP values ​​of the important features selected in the random forest regression model are calculated and sorted according to the size of the SHAP values ​​to obtain the importance of each indicator. The SHAP importance of the important features to the Kling-Gupta efficiency coefficient is further visualized using a partial dependence graph, in which the x-coordinate represents the size of the watershed attribute value, and the y-coordinate represents the SHAP value of the watershed attribute.

[0089] This embodiment uses partial dependence diagrams for visualization, which can reveal whether the relationship between the predictor variables and the independent features is linear, monotonic, or more complex. Furthermore, the nonlinear trends in these scatter plots are fitted using local weighted scatter plot smoothing (LOWESS), making the nonlinear relationship between the simulation effect evaluation results and the characteristics of the target watershed attributes more intuitive.

[0090] Further optionally, the method further includes:

[0091] Self-organizing map clustering was applied to the SHAP values ​​of each important feature, and the silhouette coefficient was used to determine the number of clusters. The spatial distribution map of the driving factors was obtained and output as the evaluation result of the factors affecting the simulation accuracy of runoff reanalysis data.

[0092] Further optionally, the method further includes:

[0093] The impact of key watershed attributes on seasonal KGE is analyzed, as are the seasonal changes in attribute importance and the similarities and differences in seasonal trends, and the results are output as evaluation results of factors affecting the simulation accuracy of runoff reanalysis data.

[0094] Example 2

[0095] This embodiment applies the method for evaluating factors affecting simulation accuracy of runoff reanalysis data proposed in Embodiment 1 to evaluate factors affecting simulation accuracy of runoff simulation of GloFAS runoff reanalysis on the CAMELS large-sample measured data set in the continental United States.

[0096] The implementation process includes:

[0097] Step 1: Extract GloFAS runoff reanalysis data.

[0098] In a specific implementation process, optionally, create a .py file, use def() to define source data reading, runoff reanalysis site data extraction and storage, reanalysis runoff data and observed runoff data are stored in sim_flow, obs_flow variables respectively, and basin attributes are stored in atributes variable.

[0099] Step 2: Evaluate the simulation effect of the runoff reanalysis data based on the Kling-Gupta efficiency coefficient.

[0100] In a specific implementation process, optionally, a class is defined using a class statement, and four indicator calculation functions are defined in the class using def(): KGE, correlation coefficient, deviation coefficient, and coefficient of variation; the four indicators are calculated and a drawing function is created to obtain the following: Figure 2 Schematic diagram of the simulation effect of runoff reanalysis data shown.

[0101] Furthermore, the correlation between the above four indicators and the watershed attributes and whether they are linearly correlated are diagnosed, and the following results are obtained: Figure 3 The Spearman correlation heat map of the simulated effect of the runoff reanalysis data and the watershed attributes is shown. The shaded grid points indicate that the P value is <0.05, and the heat grid points greater than 0.1 are not shaded, indicating a non-significant linear correlation.

[0102] Step 3: Build a random forest regression model.

[0103] In a specific implementation process, a random forest class function is constructed based on step 2, the RFmodel class is defined using the class() statement, and functions are defined in the class using def() including: C2M data conversion, predictor screening, and ten-fold cross-validation parameter determination.

[0104] This embodiment is implemented by the third-party library sklearn.feature_selection.RFECV. The schematic diagram of the ten-fold cross validation is as follows Figure 4 shown.

[0105] Among them, in the process of constructing the random forest function, the RandomForestRegressor function in the third-party library Sklearn package in Python is used to take the evaluation results of the runoff reanalysis data as the dependent variable, and the normalized watershed attributes as the predictor variable; the recursive elimination function REFCV() function is used to screen out important features; then the ten-fold cross validation method is used to find the optimal random forest regression model parameters to complete the construction of the random forest regression model.

[0106] In the process of predictor screening, predictors were first screened from CAMELS basin attributes, which are divided into five categories: topographic characteristics, climate index, land cover characteristics, soil characteristics, and geological characteristics. Some highly correlated attributes were excluded, for example, leaf area index difference and green vegetation coverage were highly correlated with the maximum leaf area index, and soil porosity and hydraulic conductivity were highly correlated with sand content. The selected attributes are shown in Table 1 below.

[0107] Table 1 CAMELS watershed attributes

[0108]

[0109]

[0110] Furthermore, the important features of watershed attributes are extracted by recursive feature elimination algorithm by removing the predictors in Table 1 one by one and evaluating their impact on prediction accuracy.

[0111] Step 4: Based on step 3, inherit the predictors and random forest model parameters and explain the random forest:

[0112] ① Calculate the SHAP value function of each basin attribute and sort the attributes to obtain the following Figure 5 The importance ranking diagram of watershed attributes affecting KGE is shown in Figure 6 Schematic diagram of the SHAP importance of watershed attributes on the correlation coefficient, coefficient of variation, and coefficient of deviation.

[0113] in, Figure 5 Part (a) identifies the key watershed attributes that affect the runoff simulation accuracy of runoff reanalysis data. Figure 5 Part (b) is a schematic diagram of the importance of key watershed attributes based on the average SHAP absolute value.

[0114] Figure 6 Part (a) is a schematic diagram of the SHAP importance of watershed attributes on the correlation coefficient; Figure 6 Part (b) is a schematic diagram of the SHAP importance of watershed attributes on the coefficient of variation; Figure 6 Part (c) is a schematic diagram of the SHAP importance of watershed attributes to the deviation coefficient.

[0115] ② Based on ①, the local weighted scatter plot smoothing LOWESS is used to fit the trends of the basin attribute values ​​and the basin attribute SHAP values ​​to diagnose the sensitivity of the simulation accuracy KGE to each attribute value.

[0116] ③ Based on ①, the spatial pattern of factors affecting simulation accuracy is diagnosed by clustering the SHAP values ​​of watershed attributes, and the following is obtained: Figure 7 The spatial cluster distribution map of the watershed attribute SHAP values ​​is shown.

[0117] Figure 7 Part (a) shows the five clusters obtained by clustering the SHAP values ​​of key watershed attributes using the self-organizing map method. Figure 7 Part (b) is a box plot of the SHAP value distribution of key watershed attributes of each cluster. Figure 7 It can be seen that runoff reanalysis performs better in watersheds with low precipitation seasonality and steep slopes, or in humid watersheds with low frequency of heavy precipitation events.

[0118] Optionally, the SHAP value calculation and drawing diagnosis for winter, spring, summer, and autumn are completed by importing feature_importance and partial_dependence, and the following is obtained: Figure 8 , 9 Schematic diagram of seasonal importance of watershed attributes and seasonal partial dependence diagram of watershed attributes are shown.

[0119] In a specific implementation process, the above steps are encapsulated into a ModelExplain class using class() statements and def(), including feature_importance function, partial_dependence function, and spatail_cluster function, and saved into a .py file to complete the encapsulation of the class function. When the class function needs to be called, the import function is used to call it, and the class function can be used.

[0120] Example 3

[0121] This embodiment proposes a system for evaluating factors affecting the simulation accuracy of runoff reanalysis data, and applies the method for evaluating factors affecting the simulation accuracy of runoff reanalysis data proposed in Embodiment 1. Fig.10 , which is a schematic diagram of the system for evaluating factors affecting runoff reanalysis data simulation accuracy according to the present embodiment.

[0122] The runoff reanalysis data simulation accuracy influencing factor evaluation system proposed in this embodiment includes:

[0123] Reanalysis data evaluation model, used to evaluate the simulation effect of input runoff reanalysis data and observed runoff data based on the Kling-Gupta efficiency coefficient;

[0124] An influencing factor screening module is equipped with a random forest regression model; wherein the random forest regression model constructs a nonlinear relationship based on the simulation effect evaluation results of the runoff reanalysis data and the characteristics of the target watershed attributes, extracts important characteristics of the watershed attributes through a recursive elimination method, and obtains the influencing factors of the simulation accuracy of the runoff reanalysis data;

[0125] The influencing factor assessment module is used to interpret the random forest regression model using an interpretable machine learning method to obtain the assessment results of the influencing factors of the runoff reanalysis data simulation accuracy.

[0126] It can be understood that the system of this embodiment corresponds to the method of the above-mentioned embodiment 1, and the options in the above-mentioned embodiment 1 are also applicable to this embodiment, so they will not be described repeatedly here.

[0127] Example 4

[0128] This embodiment proposes a computer device, including a memory and a processor, wherein the memory stores computer-readable instructions, and when the computer-readable instructions are executed by the processor, the processor executes all or part of the steps of the method for evaluating factors affecting the simulation accuracy of runoff reanalysis data proposed in Example 1.

[0129] Example 5

[0130] This embodiment proposes a storage medium having computer-readable instructions stored thereon, wherein the computer-readable instructions, when executed by a processor, implement all or part of the steps of the method for evaluating factors affecting runoff reanalysis data simulation accuracy proposed in Embodiment 1.

[0131] Exemplarily, the storage medium includes, but is not limited to, a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and other media that can store program codes.

[0132] Exemplarily, the instructions, programs, code sets or instruction sets may be implemented using conventional programming languages.

[0133] Exemplarily, the processor includes but is not limited to a smart phone, a personal computer, a server, a network device, etc., and is used to execute all or part of the steps of the method for evaluating factors affecting the simulation accuracy of runoff reanalysis data described in Example 1.

[0134] Each embodiment of the present invention is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the device embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment. The device embodiment described above is merely exemplary, in which the modules described as separate components may or may not be physically separated, and the functions of each module can be implemented in the same or one or more software and / or hardware when implementing the scheme of the present invention. It is also possible to select some or all of the modules according to actual needs to achieve the purpose of the scheme of this embodiment.

[0135] Obviously, the above embodiments of the present invention are merely examples for clearly illustrating the present invention, and are not intended to limit the embodiments of the present invention. For those skilled in the art, other different forms of changes or modifications can be made based on the above description. It is not necessary and impossible to list all the embodiments here. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the claims of the present invention.

Claims

1. A method for evaluating factors affecting the simulation accuracy of runoff reanalysis data, characterized in that: The following steps are involved: Obtaining runoff reanalysis data, observed runoff data and target watershed properties, and evaluating the simulation effect of the runoff reanalysis data based on the Kling-Gupta efficiency coefficient; Constructing a random forest regression model; wherein the random forest regression model constructs a nonlinear relationship based on the simulation effect evaluation results of the runoff reanalysis data and the characteristics of the target watershed attributes, extracts important characteristics of the watershed attributes through a recursive elimination method, and obtains the factors affecting the simulation accuracy of the runoff reanalysis data; The random forest regression model was interpreted using an interpretable machine learning method to obtain the evaluation results of factors affecting the simulation accuracy of runoff reanalysis data.

2. The method for evaluating factors affecting the simulation accuracy of runoff reanalysis data according to claim 1 is characterized in that: The simulation effect evaluation of the runoff reanalysis data based on the Kling-Gupta efficiency coefficient comprises the following steps: According to the longitude and latitude coordinates of the hydrological stations, the runoff reanalysis simulation data of each station is extracted; The correlation r between the observed runoff data and the runoff reanalysis simulation data, the average error β and the variation error γ of the runoff reanalysis data were calculated, which were further used to calculate the Kling-Gupta efficiency coefficient KGE; its expression is: Where r is the Pearson linear correlation coefficient, which represents the correlation between the observed runoff data and the runoff reanalysis data; β is the mean error, which is used to quantify the ability of the runoff reanalysis data to simulate the water balance; γ is the variation error, which is used to quantify the ability of the runoff reanalysis data to replicate the observed runoff variability; sim t represents the runoff reanalysis data at time t, obs t represents the observed runoff data at time t; μ sim represents the average value of runoff reanalysis data, μ obs represents the average value of observed runoff data; σ sim represents the standard deviation of runoff reanalysis data, σ obs Represents the standard deviation of observed runoff data.

3. The method for evaluating factors affecting the simulation accuracy of runoff reanalysis data according to claim 1, characterized in that: The construction of the random forest regression model includes the following steps: Screening predictors from target watershed attributes; The simulation effect evaluation results of runoff reanalysis data were used as dependent variables, and the normalized prediction factors were used as predictor variables to construct a random forest regression model. The importance score of each predictor was calculated, and the predictors that affected the prediction accuracy of the model were selected as important features of watershed attributes through recursive elimination method.

4. The method for evaluating factors affecting the simulation accuracy of runoff reanalysis data according to claim 3 is characterized in that: The calculation of the importance score of each predictor includes: Inputting the runoff reanalysis data and the observed runoff data into the random forest regression model, the random forest regression model obtains the importance score of each predictor by calculating the contribution of each predictor to the splitting node of the decision tree; The importance score includes a Gini index or information gain.

5. The method for evaluating factors affecting the simulation accuracy of runoff reanalysis data according to claim 3 is characterized in that: The method further comprises: The random forest regression model is trained using ten-fold cross validation to obtain the optimal model parameters; the steps include: Obtain a training data set, and randomly divide the training data set into 10 subsets of equal size; Perform 10 cross validations. In each iteration, use 9 subsets as training data and the remaining 1 subset as test data. The average performance results of 10 cross-validations were taken as the evaluation index of model performance, and the model parameters with the best performance were saved to obtain the trained random forest regression model.

6. The method for evaluating factors affecting simulation accuracy of runoff reanalysis data according to any one of claims 1 to 5, characterized in that: The method of using an interpretable machine learning method to interpret the random forest regression model comprises the following steps: Calculate the SHAP values ​​of the important features selected in the random forest regression model; Sort in descending order according to the size of the SHAP value to obtain the importance ranking of each important feature; The SHAP importance of the important features to the Kling-Gupta efficiency coefficient was visualized using partial dependence plots, and the nonlinear trends in the scatter parameter regression fitting plots were used to obtain the evaluation results of the factors affecting the simulation accuracy of the runoff reanalysis data.

7. The method for evaluating factors affecting simulation accuracy of runoff reanalysis data according to claim 6, characterized in that: The method further comprises: Self-organizing map clustering was applied to the SHAP values ​​of each important feature, and the silhouette coefficient was used to determine the number of clusters. The spatial distribution map of the driving factors was obtained and output as the evaluation result of the factors affecting the simulation accuracy of runoff reanalysis data.

8. A system for evaluating factors affecting simulation accuracy of runoff reanalysis data, using the method for evaluating factors affecting simulation accuracy of runoff reanalysis data according to any one of claims 1 to 7, characterized in that: include: Reanalysis data evaluation model, used to evaluate the simulation effect of input runoff reanalysis data and observed runoff data based on the Kling-Gupta efficiency coefficient; An influencing factor screening module is equipped with a random forest regression model; wherein the random forest regression model constructs a nonlinear relationship based on the simulation effect evaluation results of the runoff reanalysis data and the characteristics of the target watershed attributes, extracts important characteristics of the watershed attributes through a recursive elimination method, and obtains the influencing factors of the simulation accuracy of the runoff reanalysis data; The influencing factor assessment module is used to interpret the random forest regression model using an interpretable machine learning method to obtain the assessment results of the influencing factors of the runoff reanalysis data simulation accuracy.

9. A device comprising a memory and a processor, wherein the memory stores computer-readable instructions, characterized in that: When the computer-readable instructions are executed by the processor, the processor executes all or part of the steps of the method for evaluating factors affecting runoff reanalysis data simulation accuracy as described in any one of claims 1 to 7.

10. A storage medium having computer-readable instructions stored thereon, characterized in that: When the computer-readable instructions are executed by a processor, all or part of the steps of the method for evaluating factors affecting runoff reanalysis data simulation accuracy as described in any one of claims 1 to 7 are implemented.

Images