A method for regionalization of hydrological model parameters by automatically identifying basin characteristic weights

Abstract

The present invention discloses a method for regionalizing hydrological model parameters based on feature selection technology and basin physical similarity principle, comprising the following steps: automatically calibrating the parameters to be estimated of the hydrological model in multiple basins with rich hydrological and meteorological data; extracting characteristic values ​​such as the area, average slope and forest coverage of the above basins based on digital elevation data, soil data and land use and cover data; quantitatively calculating the importance weight of each basin feature to the parameters to be estimated of the hydrological model; based on the basin feature weight and rank cumulative similarity matrix, transferring the hydrological model parameters from the basin with rich hydrological and meteorological data to the basin without hydrological data. The present invention has the advantages of stable and reliable data source, high calculation efficiency, objective and reasonable results, etc., which is conducive to determining the hydrological model parameters of the basin without data and improving the accuracy of flood simulation in the basin.

Description

Technical Field

[0001] The invention relates to hydrological technology, and in particular to a method for regionalizing hydrological model parameters that automatically identifies basin characteristic weights. Background Art

[0002] The 10-year project of predictions in ungauged basins (PUB) initiated by the International Hydrological Association (IAHS) advocates the use of basin hydrological similarity to conduct hydrological forecasts in data-free areas. Although the project has ended, IAHS admits that there has been no breakthrough in the research of hydrological forecasts in data-free basins, so hydrological forecasts in data-free basins remain a global challenge. Whether this challenge can be solved depends largely on whether hydrological information can be successfully extrapolated from basins with rich hydrological and meteorological data to data-free basins, a process usually referred to as regionalization. Among the various hydrological information, the most convenient and feasible is to regionalize the parameters of the hydrological model. How to transfer the parameters of the hydrological model from the basin with rich data to the basin with no data is an important aspect of the current research on hydrological forecasts in data-free basins, and it is also an inevitable key and difficult problem when constructing a reasonable and reliable hydrological model in data-free areas.

[0003] In order to further promote the development of hydrological models in data-free areas, it is necessary to conduct in-depth research on the regionalized estimation method of hydrological model parameters in data-free areas. At present, in the parameter regionalization research, the importance weights of each basin feature are mainly set by expert experience or simply set to equal weights, which is highly subjective and difficult to reasonably quantify the importance of different basin features in the parameter transfer scheme, which increases the difficulty and subjectivity of constructing parameter regionalization schemes, and is not conducive to the promotion, application and research and development of hydrological models in data-free areas.

[0004] In view of the above shortcomings, how to objectively quantify the weight of basin characteristics, objectively and accurately transfer the hydrological model parameters from the basin with rich hydrological and meteorological data to the basin without data, and achieve reasonable estimation of the hydrological model parameters in the basin without data, is exactly the problem that the inventor needs to solve. Summary of the invention

[0005] Purpose of the invention: The purpose of the present invention is to solve the deficiencies in the prior art and to provide a method for regionalizing hydrological model parameters for automatic identification of basin characteristic weights. The method has the advantages of stable and reliable data source, high calculation efficiency, and objective and reasonable results. The method is beneficial to the determination of hydrological model parameters for basins without data and to improve the accuracy of flood simulation in the basin, and is worthy of promotion.

[0006] Technical solution: A method for regionalizing hydrological model parameters for automatically identifying basin characteristic weights of the present invention comprises the following steps:

[0007] Step 1: First, select multiple (at least 5) basins with hydrological and meteorological data in the same hydrological and meteorological area as the basin without data to be tested UC, and record them as [C1, C2, C3, ..., Ck] in sequence. Then, automatically calibrate the estimated parameter set Par = [p1, p2, p3, ..., pP] of any hydrological model M (.) in the basins [C1, C2, C3, ..., Ck], and then obtain the optimal parameter set for all basins. pP represents the Pth parameter to be estimated of the hydrological model M(.); represents the optimal parameter set of the watershed Ck;

[0008] Step 2, first extract the corresponding watershed characteristic values ​​of the watershed [C1, C2, C3, …, Ck] based on the watershed underlying surface information. The 7 watershed characteristics extracted from each watershed are: watershed area X1, watershed average elevation X2, river channel average slope X3, area slope X4, river network density X5, forest area ratio X6, watershed average saturated permeability coefficient X7; obtain the watershed characteristic group {[X11, X12, …, X17], …, [Xk1, Xk2, …, Xk7]};

[0009] Then, the basin characteristic values ​​of the data-free basin UC to be tested are extracted, which are recorded as basin area Y1, basin average elevation Y2, river channel average slope Y3, area slope Y4, river network density Y5, forest area ratio Y6, and basin average saturated permeability coefficient Y7;

[0010] Step 3: quantitatively calculate the importance weight of each basin characteristic value obtained in step 2 to the estimated parameters of the hydrological model M(.); the specific method is as follows:

[0011] Step 3.1, for the watershed [C1, C2, C3, ..., Ck], first form the watershed area characteristics of each watershed into a feature vector X1_all, and then randomly shuffle the order of the elements in the vector X1_all to generate a shadow vector X1_shadow of X1_all; at the same time, do the same process for the other six watershed characteristics to obtain feature vectors X2_all, X3_all, ..., X7_all and shadow vectors X2_shadow, X3_shadow, ..., X7_shadow; the shadow vector is obtained here to facilitate the subsequent distinction between whether the watershed characteristics are important or unique, thereby affecting the weight calculation;

[0012] Step 3.2, take the feature vectors X1_all, X2_all, ..., X7_all and the shadow vectors X1_shadow, X2_shadow, ..., X7_shadow as independent variables, and the first parameter to be estimated p1 as the dependent variable, train the random forest model, and calculate the feature importance score ΔI(Xi) of each feature vector and shadow vector based on impurity;

[0013] Step 3.3, record the maximum value of the feature importance score among the shadow vectors X1_shadow, X2_shadow, ..., X7_shadow as ShadowMax;

[0014] Step 3.4: Repeat steps 3.1 to 3.3 for 100 times, calculate the median of ShadowMax, recorded as ShadowMax__med, and compare it with the median of the feature importance scores of the feature vectors X1_all, X2_all, …, X7_all. If the median of the feature importance score of the i-th feature vector ΔI_med(Xi)≥ShadowMax__med, then the watershed feature Xi corresponding to the feature vector is important to the estimated parameter p1 and is marked as Important; otherwise, it is marked as Unimportant; i∈[1,7];

[0015] Step 3.5, calculate the feature weights ω of the watershed features X1 to X7 respectively according to whether the parameter p1 is important or not;

[0016] Step 3.6, repeat steps 3.2 to 3.5 for the remaining parameters to be estimated p2, p3, ..., pP until the characteristic weights of all the parameters to be estimated for each watershed characteristic are calculated, and the final weight of the watershed characteristic Xi ​​is taken as the average value of the characteristic weights of each parameter to be estimated. All characteristic weights form a characteristic weight vector W = [W(X1), W(X2), ..., W(X7)];

[0017] Step 4: Based on the basin feature weights and rank-accumulated similarity matrix, the estimated parameters of the hydrological model M(.) are transferred from the basins [C1, C2, C3, …, Ck] with rich hydrological and meteorological data to the basin UC without data. The specific method is as follows:

[0018] Step 4.1, calculate the absolute value of the difference between the basin feature X1 of the first basin feature of basins C1 to Ck and the basin feature Y1 of the basin without data UC, and select the basin with the smallest absolute value of the difference. The rank of its watershed feature X1 is recorded as For the basin Cτ with the second smallest absolute value of the difference, the rank of its basin feature X1 is recorded as R(Cτ,X1)=2, and so on, until the rank of the basin feature X1 of all basins is calculated;

[0019] At the same time, the same process is performed on the other six watershed features X2 to X7 and Y2 to Y7 until the ranks of all watershed attributes of all watersheds are calculated, and the rank matrix R is obtained. The data in the mth row and nth column of the R matrix is ​​(Cm, Xn);

[0020] Step 4.2, multiply the rank matrix R by the transpose of the feature weight vector W obtained in step 3 to obtain the rank accumulation matrix RW with the watershed feature weights;

[0021] Step 4.3: Select the watershed Ci corresponding to the position of the minimum element in the rank accumulation matrix RW, and set the optimal parameter group of watershed Ci Directly transfer it to the data-free basin UC and run the hydrological model M(.) to obtain the outlet flow of basin UC.

[0022] Furthermore, the detailed method of step 1 is:

[0023] First, for the basins [C1, C2, C3, ..., Ck], the hydrological and meteorological data sets (including rainfall, evaporation and flow) corresponding to each basin are obtained and recorded as [U1, U2, U3, ...Uk]; and any hydrological model M(.) is selected, and its parameters to be estimated are marked. All the parameters to be estimated form the parameter group to be estimated Par = [p1, p2, p3, ...pP], and the non-estimated parameters are set to the default values;

[0024] Then, for the basin C1, the corresponding hydrological and meteorological data U1 and the estimated parameter group Par1 are input into the hydrological model M(.), and the hydrological model M(U1,Par1) of the basin C1 is constructed. The shuffled complex evolution algorithm SCE-UA is used to automatically calibrate the estimated parameters of the hydrological model M(U1,Par1) with the maximum Nash efficiency coefficient as the objective function, and the optimal parameter group of the basin C1 is obtained.

[0025] Traverse the basins [C1, C2, C3, ..., Ck] until the optimal parameter set for all basins in step 1.1 is obtained

[0026] Furthermore, the detailed process of step 2 is as follows:

[0027] First, for the watershed C1, the following characteristic values ​​of the watershed C1 are extracted based on the watershed digital elevation map: watershed area X1, watershed average elevation X2, river channel average slope X3, area slope X4, river network density X5; the forest area ratio X6 of the watershed C1 is extracted based on the watershed land use and cover map; the watershed average saturated permeability coefficient X7 of the watershed C1 is extracted based on the watershed soil distribution map;

[0028] Traverse the watershed [C1, C2, C3, …, Ck] and extract the above seven watershed features of the remaining watersheds;

[0029] The same method is used to extract seven characteristics of the watershed UC without hydrological and meteorological data, which are respectively recorded as watershed area Y1, watershed average elevation Y2, river channel average slope Y3, area slope Y4, river network density Y5, forest area ratio Y6, and watershed average saturated permeability coefficient Y7.

[0030] Furthermore, the calculation formula of the feature importance score ΔI(Xi) of the feature vector and the shadow vector based on impurity in step 3.2 is as follows:

[0031]

[0032] In the above formula, Xi represents the characteristic vector or shadow vector obtained from all watersheds, ΔI t (Xi) represents the feature importance score of Xi in the t-th tree of the random forest, T represents the total number of trees in the random forest, S j represents the set of all nodes Xi is used for segmentation, I(s) is the Gini index of node s; sl and sr are the left and right child nodes of node s, respectively, pl and pr are the weights of the left and right child nodes (the proportion of node samples), respectively; ΔI(Xi) is the arithmetic mean of the importance scores of all trees in the random forest;

[0033] In step 3.5, the watershed characteristic weight ω is calculated according to whether the current parameter to be estimated is important. The specific calculation formula is as follows:

[0034]

[0035] Where ω(Xi) is the weight of the ith watershed feature, ΔI_med(Xi) is the median of the feature importance score of the ith watershed feature, ΔI_med(Xz) is the median of the importance score of the zth watershed feature marked as Important, and Z is the number of watershed features marked as Important.

[0036] The calculation formula of the final weight W(Xi) of the watershed feature Xi in step 3.6 is:

[0037]

[0038] Where P is the number of parameters to be estimated, and all feature weights constitute the feature weight vector W = [W(X1), W(X2), …, W(X7)].

[0039] Beneficial effects: The present invention not only considers the similarity of basin attributes, but also considers the weight of basin characteristics, and transfers the parameter calibration results of basins with rich hydrological and meteorological data to basins without data. On the basis of ensuring the objectivity and reliability of the calculation results, it solves the problem of difficulty in calibrating the parameters of hydrological models in basins without data. In addition, this method mainly uses remote sensing observation data such as basin digital elevation models, soil maps, land use and vegetation coverage. The data source is stable and reliable, and the functional relationship between variables in the method is clear, which is conducive to the computer automation execution of parameter estimation of hydrological models in basins without data, ensuring the objectivity and rationality of the results, and is conducive to the in-depth development of digital hydrology research in areas without data. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 It is a schematic diagram of the process of the present invention;

[0041] Figure 2 For the study of the basin distribution map in the embodiment;

[0042] Figure 3 Schematic diagram of the importance score of watershed characteristics to parameter CS in the embodiment;

[0043] Figure 4 Schematic diagram of importance scores of watershed characteristics to parameter Lag in the embodiment;

[0044] Figure 5 This is a schematic diagram of the flood simulation results in the Tunxi River Basin in the embodiment;

[0045] Table 1 Schematic diagram of parameter calibration results for data-rich basins. DETAILED DESCRIPTION

[0046] The technical solution of the present invention is described in detail below, but the protection scope of the present invention is not limited to the embodiments.

[0047] like Figure 1 As shown, the method for regionalizing hydrological model parameters for automatically identifying basin characteristic weights of the present invention comprises the following steps:

[0048] Step 1: Select multiple basins with data in the same hydrological and meteorological area as the basin without data to be measured UC, and record them as [C1, C2, C3, ..., Ck] in sequence. Automatically calibrate the estimated parameter set Par = [p1, p2, p3, ..., pP] of any hydrological model M (.) in the basins [C1, C2, C3, ..., Ck] respectively, and then obtain the optimal parameter set for all basins. pP represents the Pth parameter to be estimated of the hydrological model M(.); represents the optimal parameter set of the watershed Ck;

[0049] Step 2: First, extract the corresponding watershed characteristic values ​​of the watershed [C1, C2, C3, ..., Ck] based on the watershed underlying surface information. The seven watershed characteristics extracted from each watershed are: watershed area X1, watershed average elevation X2, river channel average slope X3, area slope X4, river network density X5, forest area ratio X6, and watershed average saturated permeability coefficient X7;

[0050] Then, the basin characteristic values ​​of the data-free basin UC to be tested are extracted, which are recorded as basin area Y1, basin average elevation Y2, river channel average slope Y3, area slope Y4, river network density Y5, forest area ratio Y6, and basin average saturated permeability coefficient Y7;

[0051] Step 3: quantitatively calculate the importance weight of each basin characteristic value obtained in step 2 to the estimated parameters of the hydrological model M(.); the specific method is as follows:

[0052] Step 3.1, for the watershed [C1, C2, C3, …, Ck], first combine the watershed area characteristics of each watershed into a feature vector X1_all, and then randomly shuffle the order of the elements in the vector X1_all to generate a shadow vector X1_shadow of X1_all;

[0053] At the same time, the same processing is performed on the other six watershed features to obtain feature vectors X2_all, X3_all, ..., X7_all and shadow vectors X2_shadow, X3_shadow, ..., X7_shadow;

[0054] Step 3.2, take the feature vectors X1_all, X2_all, ..., X7_all and the shadow vectors X1_shadow, X2_shadow, ..., X7_shadow as independent variables, and the first parameter to be estimated p1 as the dependent variable, train the random forest model, and calculate the feature importance score ΔI(Xi) of each feature vector and shadow vector based on impurity;

[0055] Step 3.3, record the maximum value of the feature importance score among the shadow vectors X1_shadow, X2_shadow, ..., X7_shadow as ShadowMax;

[0056] Step 3.4: Repeat steps 3.1 to 3.3 for 100 times, calculate the median of ShadowMax, recorded as ShadowMax__med, and compare it with the median of the feature importance scores of the eigenvectors X1_all, X2_all, …, X7_all. If the median of the feature importance score of the i-th eigenvector ΔI_med(Xi)≥ShadowMax__med, the watershed feature Xi corresponding to the eigenvector is important to the parameter to be estimated p1 and is marked as Important; otherwise, it is marked as Unimportant.

[0057] Step 3.5, calculate the feature weights ω of the watershed features X1 to X7 respectively according to whether the parameter p1 is important or not;

[0058] Step 3.6, repeat steps 3.2 to 3.5 for the remaining parameters to be estimated p2, p3, ..., pP until the characteristic weights of all the parameters to be estimated for each watershed characteristic are calculated. The final weight of the watershed characteristic Xi ​​is the average of the characteristic weights of each parameter to be estimated. All characteristic weights form a characteristic weight vector W = [W(X1), W(X2), ..., W(X7)];

[0059] Step 4: Based on the basin feature weights and rank-accumulated similarity matrix, the parameters of the hydrological model M(.) are transferred from the basins with rich hydrological and meteorological data [C1, C2, C3, …, Ck] to the basins without data UC. The specific method is as follows:

[0060] Step 4.1, respectively calculate the absolute value of the difference between the basin feature X1 of the first basin feature of basins C1 to Ck and the basin feature Y1 of the basin without data UC, and for the basin Ci with the smallest absolute value of the difference, record the rank of its basin feature X1 as R(Ci,X1)=1, and for the basin Cj with the second smallest absolute value of the difference, record the rank of its basin feature X1 as R(Cj,X1)=2, and so on, until the rank of X1 of all basins is calculated;

[0061] At the same time, the same process is performed on the other six watershed features X2 to X7 and Y2 to Y7 until the ranks of all watershed attributes of all watersheds are calculated, and the rank matrix R is obtained. The data in the mth row and nth column of the R matrix is ​​(Cm, Xn);

[0062] Step 4.2, multiply the rank matrix R by the transpose of the feature weight vector W obtained in step 3 to obtain the rank accumulation matrix RW with the watershed feature weights;

[0063] Step 4.3: Select the watershed Ci corresponding to the position of the minimum element in the rank accumulation matrix RW, and set the optimal parameter group of watershed Ci Directly transfer it to the data-free basin UC and run the hydrological model M(.) to obtain the outlet flow of basin UC.

[0064] Example

[0065] This embodiment takes the Tunxi River Basin as an example, and regards the Tunxi River Basin as a river basin with no data. The other 29 river basins in the humid area are selected as river basins with rich hydrological data, numbered C1 to C29, and the corresponding hydrological and meteorological data sets of the river basins are marked as U1 to U29. The hydrological data of each river basin are shown in Table 1.

[0066] The hydrological model M(.) of this embodiment selects the Xin'anjiang hydrological model as an example model, and takes the river network water recession coefficient CS and the river network confluence lag time Lag as the parameters to be estimated. The estimated parameter group Par=[CS, Lag]. According to the instruction manual of the Xin'anjiang model, the remaining parameters except CS and Lag are set to the default values.

[0067] Table 1

[0068]

[0069] This example is based on the hydrological and meteorological data of 29 river basins, uses the SCE-UA algorithm, and takes the maximum Nash efficiency coefficient between the observed flow and the hydrological model simulated flow as the objective function. The parameters of Par = [CS, Lag] of river basins C1 to C29 are automatically calibrated to obtain the optimal parameter combination for each river basin. See Appendix 1.

[0070] The process of extracting watershed features based on the watershed underlying surface information in step 2 of this embodiment is as follows:

[0071] Step 2.1, based on the digital elevation map of the watershed, extract the following characteristic values ​​of the watershed C1: watershed area X1, watershed average elevation X2, river channel average slope X3, area slope X4, river network density X5;

[0072] Step 2.2, based on the watershed land use and cover map, extract the following characteristic values ​​of watershed C1: forest area ratio X6;

[0073] Step 2.3, based on the watershed soil distribution map, extract the following characteristic values ​​of watershed C1: the average saturated permeability coefficient of the watershed X7;

[0074] Step 2.4: Repeat steps 2.1 to 2.3 for each basin until the characteristic values ​​of 29 basins from C1 to C29 are extracted;

[0075] Step 2.5, repeat steps 2.1 to 2.3 until the characteristic values ​​of all watersheds in the Tunxi UC watershed without hydrological and meteorological data are extracted, which are recorded as watershed area Y1, watershed average elevation Y2, river channel average slope Y3, area slope Y4, river network density Y5, forest area ratio Y6, and watershed average saturated permeability coefficient Y7.

[0076] The process of calculating the importance weight of each basin characteristic value to the parameter to be estimated in the hydrological model in step 3 of this embodiment is as follows:

[0077] Step 3.1, for the basins [C1, C2, C3, …, C29], first combine the basin area characteristics of each basin into a feature vector X1_all, and then randomly shuffle the order of the elements in the vector X1_all to generate a shadow vector X1_shadow of X1_all;

[0078] At the same time, the other six watershed features are processed in the same way to obtain feature vectors X2_all, X3_all, ..., X7_all and shadow vectors X2_shadow, X3_shadow, ..., X7_shadow; Step 3.2, the feature vectors X1_all, X2_all, ..., X7_all and shadow vectors X1_shadow, X2_shadow, ..., X7_shadow are used as independent variables, and the estimated parameter CS is used as the dependent variable. The random forest model is trained to calculate the feature importance score ΔI(Xi) of each feature vector and shadow vector based on impurity according to the following formula:

[0079]

[0080] Step 3.3: The maximum value of the feature importance score among the shadow vectors X1_shadow, X2_shadow, ..., X7_shadow is recorded as ShadowMax.

[0081] Step 3.4: Repeat steps 3.1 to 3.4 for 100 times, calculate the median of ShadowMax, recorded as ShadowMax_med, and compare it with the median of the feature importance scores of the feature vectors X1_all, X2_all, …, X7_all. If the median of the feature importance score of the i-th feature vector ΔI_med(Xi)≥ShadowMax__med, then the watershed feature Xi corresponding to the feature vector is important to the estimated parameter CS and is marked as Important. Figure 3 The green box in the figure; otherwise it is marked as Uniimportant, see Figure 3 The red box in the

[0082] Step 3.5: Calculate the feature weights ω of the watershed features X1 to X7 according to whether the parameter CS is important or not. The calculation formula is as follows:

[0083]

[0084] Step 3.6: Repeat steps 3.2 to 3.5 for parameter Lag. The result is shown in Figure 4 The final weight of the watershed feature Xi is the average of the feature weights of the parameters CS and Lag, denoted as W(Xi):

[0085]

[0086] Finally, all feature weights in this embodiment form a feature weight vector W = [0.12, 0.11, 0.25, 0.14, 0.28, 0.00, 0.10].

[0087] The details of step 4 of this embodiment are as follows:

[0088] Step 4.1, respectively calculate the absolute value of the difference between the basin attribute X1 of basins C1 to C29 and the basin attribute Y1 of basin UC without hydrological and meteorological data. For basin C21 with the smallest absolute value of difference, the rank of attribute X1 is recorded as R(C21,X1)=1. For basin C23 with the second smallest absolute value of difference, the rank of attribute X1 is recorded as R(C23,X1)=2. This process is repeated until the rank of X1 of all basins is calculated.

[0089] Step 4.2, repeat step 4.1 for the watershed attributes X2 to X7 and Y2 to Y7 until the ranks of all watershed attributes of all watersheds are calculated, and the rank matrix R is obtained. The data in the mth row and nth column of the R matrix is ​​(Cm, Xn);

[0090] Step 4.3, multiply the rank matrix R by the transpose of the feature weight vector W to obtain the rank accumulation matrix RW with the watershed feature weights.

[0091] Step 4.4, select the watershed C23 corresponding to the position of the minimum element in the rank accumulation matrix RW, and set the optimal parameter group of watershed C23 Directly transfer to the Tunxi Basin UC without hydrological and meteorological data to be calculated, run the hydrological model, and obtain the basin outlet flow, see Figure 5 .

Claims

1. A method for regionalizing hydrological model parameters for automatically identifying basin characteristic weights, characterized in that: The following steps are involved: Step 1: Select k basins with data in the same hydrological and meteorological zone as the basin without data to be tested UC, and record them as [C1, C2, C3, ..., Ck] in sequence. Automatically calibrate the estimated parameter set Par = [p1, p2, p3, ..., pP] of any hydrological model M(.) in the basins [C1, C2, C3, ..., Ck] respectively, and then obtain the optimal parameter set for all basins. pP represents the Pth parameter to be estimated of the hydrological model M(.); represents the optimal parameter set of the watershed Ck; Step 2: First, extract the corresponding watershed characteristic values ​​of the watershed [C1, C2, C3, ..., Ck] based on the watershed underlying surface information. The seven watershed characteristics extracted from each watershed are: watershed area X1, watershed average elevation X2, river channel average slope X3, area slope X4, river network density X5, forest area ratio X6, and watershed average saturated permeability coefficient X7; Then, the basin characteristic values ​​of the data-free basin UC to be tested are extracted, which are recorded as basin area Y1, basin average elevation Y2, river channel average slope Y3, area slope Y4, river network density Y5, forest area ratio Y6, and basin average saturated permeability coefficient Y7; Step 3: quantitatively calculate the importance weight of each basin characteristic value obtained in step 2 to the estimated parameters of the hydrological model M(.); the specific method is as follows: Step 3.1, for the watershed [C1, C2, C3, …, Ck], first combine the watershed area characteristics of each watershed into a feature vector X1_all, and then randomly shuffle the order of the elements in the vector X1_all to generate a shadow vector X1_shadow of X1_all; At the same time, the same processing is performed on the other six watershed features to obtain feature vectors X2_all, X3_all, ..., X7_all and shadow vectors X2_shadow, X3_shadow, ..., X7_shadow; Step 3.2, take the feature vectors X1_all, X2_all, ..., X7_all and the shadow vectors X1_shadow, X2_shadow, ..., X7_shadow as independent variables, and the first parameter to be estimated p1 as the dependent variable, train the random forest model, and calculate the feature importance score ΔI(Xi) of each feature vector and shadow vector based on impurity; Step 3.3, record the maximum value of the feature importance score among the shadow vectors X1_shadow, X2_shadow, ..., X7_shadow as ShadowMax; Step 3.4, repeat steps 3.1 to 3.3 for a total of 100 times, calculate the median of ShadowMax, recorded as ShadowMax__med, and compare it with the median of the feature importance scores of the feature vectors X1_all, X2_all, …, X7_all. If the median of the feature importance score of the i-th watershed feature vector ΔI_med(Xi)≥ShadowMax__med, then the watershed feature Xi corresponding to the feature vector is important to the estimated parameter p1 and is marked as Important; otherwise, it is marked as Unimportant; i∈[1,7]; Step 3.5, calculate the feature weights corresponding to the watershed features X1 to X7 respectively according to whether the parameter p1 to be estimated is important; Step 3.6, repeat steps 3.2 to 3.5 for the remaining parameters to be estimated p2, p3, ..., pP until the characteristic weights of all the parameters to be estimated for each watershed characteristic are calculated, and the final weight of the watershed characteristic Xi ​​is taken as the average value of the characteristic weights of each parameter to be estimated. All characteristic weights form a characteristic weight vector W = [W(X1), W(X2), ..., W(X7)]; Step 4: Based on the basin feature weights and rank-accumulated similarity matrix, the estimated parameters of the hydrological model M(.) are transferred from the basins [C1, C2, C3, …, Ck] with rich hydrological and meteorological data to the basin UC without data. The specific method is as follows: Step 4.1, calculate the absolute value of the difference between the basin feature X1 of the first basin feature of basins C1 to Ck and the basin feature Y1 of the basin without data UC, and select the basin with the smallest absolute value of the difference. The rank of its watershed feature X1 is recorded as For the basin Cτ with the second smallest absolute value of the difference, the rank of its basin feature X1 is recorded as R(Cτ,X1)=2, and so on, until the rank of the basin feature X1 of all basins is calculated; At the same time, the same process is performed on the other six watershed features X2 to X7 and Y2 to Y7 until the ranks of all watershed features of all watersheds are calculated, and the rank matrix R is obtained. The data in the mth row and nth column of the R matrix is ​​(Cm, Xn); Step 4.2, multiply the rank matrix R by the transpose of the feature weight vector W obtained in step 3 to obtain the rank accumulation matrix RW with the watershed feature weights; Step 4.3: Select the watershed Ci corresponding to the position of the minimum element in the rank accumulation matrix RW, and set the optimal parameter group of watershed Ci Directly transfer it to the data-free basin UC and run the hydrological model M(.) to obtain the outlet flow of basin UC.

2. The method for regionalizing hydrological model parameters for automatically identifying basin characteristic weights according to claim 1, characterized in that: The detailed method of step 1 is: First, for the basins [C1, C2, C3, ..., Ck], the hydrological and meteorological data sets corresponding to each basin are obtained and recorded as [U1, U2, U3, ...Uk]; and any hydrological model M(.) is selected, and its parameters to be estimated are marked. All the parameters to be estimated form the parameter group to be estimated Par = [p1, p2, p3, ...pP], and the non-estimated parameters are set to the default values; Then, for the basin C1, the corresponding hydrological and meteorological data U1 and the estimated parameter group Par1 are input into the hydrological model M(.), and the hydrological model M(U1,Par1) of the basin C1 is constructed. The shuffled complex evolution algorithm SCE-UA is used to automatically calibrate the estimated parameters of the hydrological model M(U1,Par1) with the maximum Nash efficiency coefficient as the objective function, and the optimal parameter group of the basin C1 is obtained. Traverse the basins [C1, C2, C3, ..., Ck] until the optimal parameter set for all basins in step 1 is obtained 3. The method for regionalizing hydrological model parameters for automatically identifying basin characteristic weights according to claim 1, characterized in that: The detailed process of step 2 is as follows: First, for the watershed C1, the following characteristic values ​​of the watershed C1 are extracted based on the watershed digital elevation map: watershed area X1, watershed average elevation X2, river channel average slope X3, area slope X4, river network density X5; the forest area ratio X6 of the watershed C1 is extracted based on the watershed land use and cover map; the watershed average saturated permeability coefficient X7 of the watershed C1 is extracted based on the watershed soil distribution map; Traverse the watershed [C1, C2, C3, …, Ck] and extract the above seven watershed features of the remaining watersheds; The same method is used to extract seven characteristics of the watershed UC without hydrological and meteorological data, which are respectively recorded as watershed area Y1, watershed average elevation Y2, river channel average slope Y3, area slope Y4, river network density Y5, forest area ratio Y6, and watershed average saturated permeability coefficient Y7.

4. The method for regionalizing hydrological model parameters for automatically identifying basin characteristic weights according to claim 1, characterized in that: The calculation formula of the feature importance score ΔI(Xi) of the feature vector and the shadow vector based on impurity in step 3.2 is as follows: In the above formula, Xi represents the characteristic vector or shadow vector of the entire basin, ΔI t (Xi) represents the feature importance score of Xi in the t-th tree of the random forest, T represents the total number of trees in the random forest, S j represents the set of all nodes Xi is used for segmentation, I(s) is the Gini index of node s; sl and sr are the left and right child nodes of node s, pl and pr are the weights of the left and right child nodes, respectively; ΔI(Xi) is the arithmetic mean of the importance scores of all trees in the random forest; The specific calculation formula of the characteristic weight ω of the watershed characteristics in step 3.5 is as follows: Where ω(Xi) is the weight of the ith watershed feature, ΔI_med(xi) is the median of the feature importance score of the ith watershed feature, ΔI_med(xz) is the median of the importance score of the zth watershed feature marked as Important, and Z is the number of watershed features marked as Important. The calculation formula of the final weight W(Xi) of the watershed feature Xi in step 3.6 is as follows: Where P is the number of parameters to be estimated, and all feature weights constitute the feature weight vector W = [W(X1), W(X2), …, W(X7)].

Images