A paleorunoff reconstruction method based on LASSO logarithmic regression

Abstract

The present invention discloses a paleo-runoff reconstruction method based on LASSO logarithmic regression and a paleo-runoff reconstruction method based on tree rings. Aiming at the length and accuracy problems of runoff reconstruction, the present invention creates a forward hierarchical nesting method and a LASSO logarithmic regression nonlinear reconstruction model with tree ring information optimization capability. First, the reconstructed nesting layers are divided according to the starting year of the tree ring chronology, and the tree ring information with a longer time span is fully utilized. Then, the LASSO logarithmic regression nonlinear model is used to reconstruct the paleo-runoff on the basis of determining the global optimal tree ring principal component subsets of each nested layer, thereby improving the reconstruction accuracy.

Description

Technical Field

[0001] The invention relates to the technical field of paleorunoff reconstruction, and in particular to a paleorunoff reconstruction method based on LASSO logarithmic regression. Background Art

[0002] The development and utilization of river runoff requires a sufficiently long runoff series to reveal the long-term evolution of runoff and to discover the relationship between hydrological changes and related climate changes, so as to make accurate and reliable runoff forecasts and provide an accurate scientific basis for production practice. Existing runoff record data are often short, which limits the comprehensive understanding of runoff evolution, especially on a time scale of ten to a hundred years.

[0003] Long-series runoff reconstruction data can not only overcome the limitation of decadal change research caused by the short time span of runoff record data, but also contain a large amount of evolution law and statistical characteristic information. The longer the reconstruction sequence, the more helpful it is to grasp the evolution process of different years, and provide a basis for analyzing the occurrence law and evolution characteristics of extreme events, grasping the prediction factors and their driving laws, and reducing forecast uncertainty. Tree rings and runoff have the same environmental factors and long-recorded data. Tree ring hydrology research based on tree rings can accurately reconstruct long-series runoff data and is a hot spot in runoff reconstruction research.

[0004] At present, the research methods for runoff reconstruction are relatively simple, generally using simple regression methods. At the same time, the available tree ring data are relatively scarce, which means that the length and accuracy of the reconstructed sequence need to be improved. Summary of the invention

[0005] The purpose of the present invention is to provide a paleorunoff reconstruction method based on LASSO logarithmic regression, make full use of tree rings of different time spans to extend the reconstruction time span, establish a nonlinear reconstruction model with the ability to optimize tree ring information, and reconstruct paleorunoff based on the global optimal tree ring principal component subset.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0007] A paleorunoff reconstruction method based on LASSO logarithmic regression, the steps of which include:

[0008] According to the time span and geographical location of the tree rings, tree ring data that meet the research time length and are near the research area are initially selected, and then the Pearson correlation coefficient is used to measure the correlation between the tree rings and annual runoff for secondary tree ring screening;

[0009] The secondary screened tree-ring sequences were de-trended and a standard tree-ring chronology was established for each sampling point.

[0010] The nesting layers were divided according to the standard chronology of tree rings at each sampling site;

[0011] The least angle regression algorithm and cross validation method are used to solve the regression model parameters of the LASSO logarithmic regression model. ;

[0012] According to the regression model parameters solved Determine the LASSO logarithmic regression model; in the LASSO logarithmic regression model, tree rings are used as reconstruction factors, and the runoff sequence to be reconstructed Logarithmic transformation of as the response variable;

[0013] Each nested layer is regressed using the LASSO logarithmic regression model, and each regressed nested layer is nested to obtain the reconstructed runoff sequence.

[0014] Based on the LASSO method, a regression equation with a regression coefficient that can be strictly 0 is established. At the same time, a characteristic function that conforms to the influence of climate and runoff on tree rings is used to perform nonlinear transformation on the regression equation to simulate the nonlinear relationship between tree rings and climate factors and runoff. The LASSO method sets a penalty term to minimize the residual sum of squares under the constraint that the sum of the absolute values ​​of the regression coefficients is less than a constant, so that the parameters of some tree ring reconstruction factors in the reconstruction model can be compressed to 0, and the optimal tree ring subset is screened out.

[0015] The detrending method includes negative exponential function method, 67% spline function method, "Signal-free" detrending method and empirical mode decomposition method. The detrending of tree ring sequence is carried out to obtain the climate information contained in the tree ring more realistically.

[0016] According to the above technical solution, the steps of constructing the tree-ring standard chronology include:

[0017] Four detrending methods, namely negative exponential function method, 67% spline function method, "Signal-free" detrending method and empirical mode decomposition method, were used to fit the growth trend of tree rings at different sampling points.

[0018] The corresponding growth trends were eliminated from the growth trend fitting results of tree rings at different sampling points according to the four detrending methods, and four tree ring standard chronologies corresponding to the four detrending methods were constructed for each sampling point.

[0019] The correlation between the tree ring standard chronology obtained based on different detrending methods at the same sampling point and the runoff of nearby hydrological stations was compared, and the detrending method with the highest average correlation value was selected as the optimal detrending method for the tree rings of the sampling point. The optimal detrending method was used to establish the tree ring standard chronology of each sampling point.

[0020] According to the above technical solution, a series of nested periods that extend continuously into the past are divided according to the year when the tree-ring chronology starts; the nested periods have different starting years but the same ending years; and the runoff sequence corresponding to each nested period is called a nested layer.

[0021] According to the above technical solution, the regression model parameters It is expressed as:

[0022] ;

[0023] Where: represents the tree-ring chronology, , represents sequence 1 in the dendrochronology, represents sequence 2 in the tree-ring chronology, represents the sequence m in the tree-ring chronology, represents the runoff sequence to be reconstructed, Indicates the value when the cost function is minimum in the real number range, Expressed as The L2-norm of is the penalty parameter and , and Corresponding and , for The L1-norm of is the number of runoff observations.

[0024] When dividing the nested time periods, two adjacent time periods are usually at least 20 years apart to avoid a large number of redundant reconstruction processes. Among them, the reconstruction layer with a later starting year is more reliable than the nested layer with an earlier starting year because there are more available tree ring chronologies. Therefore, the nested runoff reconstruction sequence is obtained by continuously adding the nested layer with an earlier starting year to the nested layer with a later starting year.

[0025] According to the above technical solution, the steps of determining the parameters of the LASSO logarithmic regression model using the cross-validation method include:

[0026] The original runoff data set is evenly divided into five subsets. One of the five subsets is selected as the test set, and the other four subsets are used as training sets. It is ensured that each subset is used as a test set once. Each training set and test set uses the minimum angle regression algorithm to solve the regression model parameters. , and the regression model parameters Conduct ratings;

[0027] The LASSO logarithmic regression model scoring results of the five tests were averaged using the 5-fold crossover error calculation formula to obtain the final performance evaluation of the LASSO logarithmic regression model;

[0028] The regression model parameters corresponding to the minimum mean value obtained by calculating the 5-fold cross error The parameters of the LASSO logarithmic regression model are determined. The minimum angle regression algorithm is used to find parameters based on different combinations, and the combination parameters with the smallest error are selected as the final result.

[0029] According to the above technical solution, the minimum angle regression algorithm is used to take the bisector direction of the angle formed by the projection direction and the residual direction of the forward gradient algorithm to move and solve the LASSO logarithmic regression model. The specific steps include:

[0030] 1) Initialize weight coefficients , initialized to zero vector;

[0031] 2) Initialize the residual vector as the target label vector ; Because at this time is a zero vector, so the residual vector is equal to the target label vector ;

[0032] 3) Select a feature vector that has the greatest correlation with the residual vector , along the eigenvector Find a set of weight coefficients in the direction of , another eigenvector with the greatest correlation with the residual vector appears Make the correlation between the new residual vector and the two eigenvectors equal, and recalculate the residual vector; the eigenvector is the sequence in the tree-ring chronology in the LASSO logarithmic regression model;

[0033] 4) Repeat step 3) and continue to find a set of weight coefficients , so that the third eigenvector with the greatest correlation with the residual vector The correlation between the new residual vector and the three eigenvectors is made equal, until the residual vector is less than or equal to the set threshold or all eigenvectors have been selected, and the iteration ends.

[0034] According to the above technical solution, the steps of determining the optimal LASSO logarithmic regression model using the cross-validation method include:

[0035] The original runoff dataset is evenly divided into five subsets. One of the five subsets is selected as the test set, and the other four subsets are used as training sets. It is ensured that each subset is used as a test set once. The model parameters are regressed using each training set and test set. , and the regression model parameters Conduct ratings;

[0036] The scoring results of the LASSO logarithmic regression model from the five tests were averaged using the 5-fold cross error calculation formula to obtain the final performance evaluation of the LASSO logarithmic regression model.

[0037] According to the above technical solution, the cross error calculation formula is:

[0038] ;

[0039] Where: represents the 5-fold crossover error, For the The residual sum of squares corresponding to the group LASSO logistic regression model.

[0040] The accuracy of the LASSO logarithmic regression model is verified based on the test set observations.

[0041] According to the above technical solution, the nesting layer nesting step includes:

[0042] Normalize the tree-ring sequences in the selected standard tree-ring chronology;

[0043] Taking each nesting period as the reconstruction time span, the tree-ring chronology covering the entire period was selected, and the normalized nesting layer corresponding to each nesting period was regressed using the LASSO logarithmic regression model to obtain the regression nesting layer;

[0044] Sort all regression nesting layers from small to large based on the starting year;

[0045] In the adjacent nested regression layers, the starting year of the larger ranking is used as the ending year of the smaller ranking regression reconstruction layer, and then the smaller ranking regression reconstruction layer is truncated to obtain the secondary reconstruction layer;

[0046] All sub-reconstructed layers are connected from small to large according to the starting year to obtain the reconstructed runoff sequence.

[0047] The LASSO method minimizes the residual sum of squares by setting a penalty term under the constraint that the sum of the absolute values ​​of the regression coefficients is less than a constant, so that the parameters of some tree-ring reconstruction factors in the reconstruction model can be compressed to 0, and the optimal tree-ring subset can be screened out. In the nested method, reconstructions with earlier starting years are usually more reliable than reconstructions with later starting years because more tree-ring chronologies are involved.

[0048] Compared with the prior art, the beneficial effects achieved by the present invention are: ancient runoff can be reconstructed using tree-ring information, the reconstruction nested layers can be divided according to the starting year of the tree-ring chronology, tree-ring information with a longer time span can be fully utilized, and a nonlinear reconstruction model with the ability to optimize tree-ring information can be established using the LASSO regression equation to improve the reconstruction accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0050] Figure 1 It is a flowchart of the steps of a paleoflow reconstruction method based on LASSO logarithmic regression of the present invention;

[0051] Figure 2 It is a schematic diagram of the nested refactoring method. DETAILED DESCRIPTION

[0052] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0053] The present invention provides a technical solution: a paleorunoff reconstruction method based on LASSO logarithmic regression first obtains a tree ring standard chronology, screens the tree rings in the tree ring standard chronology according to the time length and the correlation between the tree rings and the annual runoff, and then removes the growth trend of the screened tree rings and divides them into nested layers. After determining the LASSO logarithmic regression model, a regression calculation is performed on the nested layers, and each nested layer after the regression calculation is nested to achieve runoff reconstruction. In the above steps, the obtained tree rings can also be screened and the growth trend removed before constructing the tree ring standard chronology.

[0054] Taking the construction of a standard tree-ring chronology after screening the acquired tree rings and removing the growth trend as an example, a paleorunoff reconstruction method based on LASSO logarithmic regression was used to reconstruct the runoff of a certain area. The specific steps (such as Figure 1 )include:

[0055] S1. Based on the time span and geographical location of the tree rings, tree ring data that meet the research time length and are near the study area are initially selected, and then the correlation between the tree rings and annual runoff measured by the Pearson correlation coefficient is used for secondary tree ring screening.

[0056] S2. Detrending the tree-ring sequence after secondary screening, and establishing the standard tree-ring chronology of each sampling point. The specific steps include: fitting the growth trend of tree rings at different sampling points using four detrending methods: negative exponential function method, 67% spline function method, "Signal-free" detrending method, and empirical mode decomposition method;

[0057] The corresponding growth trends were eliminated from the growth trend fitting results of tree rings at different sampling points according to the four detrending methods, and four tree ring standard chronologies corresponding to the four detrending methods were constructed for each sampling point.

[0058] The correlation between the tree ring standard chronology obtained based on different detrending methods at the same sampling point and the runoff of nearby hydrological stations was compared, and the detrending method with the highest average correlation value was selected as the optimal detrending method for the tree rings of the sampling point. The optimal detrending method was used to establish the tree ring standard chronology of each sampling point.

[0059] S3. Dividing nested layers according to the standard tree-ring chronology of each sampling point, the specific steps of which include: dividing a series of nested periods that extend continuously into the past according to the year when the tree-ring chronology starts; the nested periods have different starting years but the same ending years, and the runoff sequence corresponding to each nested period is called a nested layer.

[0060] S4. Use the minimum angle regression algorithm and cross-validation method to solve the regression model parameters of the LASSO logarithmic regression model Specifically, the original runoff data set is evenly divided into five subsets, one of which is selected as the test set, and the other four subsets are selected as training sets, ensuring that each subset is used as a test set once. Each training set and test set uses the minimum angle regression algorithm to solve the regression model parameters. , and the regression model parameters Conduct ratings;

[0061] The LASSO logarithmic regression model scoring results of the five tests were averaged using the 5-fold cross error calculation formula to obtain the final performance evaluation of the LASSO logarithmic regression model; among them, the 5-fold cross error calculation formula is:

[0062] ;

[0063] Where: represents the 5-fold crossover error, For the The residual sum of squares corresponding to the group LASSO logistic regression model.

[0064] The regression model parameters corresponding to the minimum mean value obtained by calculating the 5-fold cross error Determine the parameters for the LASSO logistic regression model.

[0065] Among them, the regression model parameters It is expressed as:

[0066] ;

[0067] Where: represents the tree-ring chronology, , represents the runoff sequence to be reconstructed, Indicates the value when the cost function is minimum in the real number range, Expressed as The L2-norm of is the penalty parameter and , and Corresponding and , for The L1-norm of is the number of runoff observations.

[0068] The minimum angle regression algorithm is used to take the bisector direction of the angle formed by the projection direction and the residual direction of the forward gradient algorithm for the regression model parameters. Perform mobile solution, the specific steps include:

[0069] 1) Initialize weight coefficients ;

[0070] 2) Initialize the residual vector as the target label vector ;

[0071] 3) Select a feature vector that has the greatest correlation with the residual vector , along the eigenvector Find a set of weight coefficients in the direction of , another eigenvector with the greatest correlation with the residual vector appears Make the correlation between the new residual vector and the two eigenvectors equal and recalculate the residual vector;

[0072] 4) Repeat step 3) and continue to find a set of weight coefficients , so that the third eigenvector with the greatest correlation with the residual vector Make the correlation between the new residual vector and the three eigenvectors equal, until the residual vector is less than or equal to the set threshold or all eigenvectors have been selected, the iteration ends, and the final set of weight coefficients is obtained. .

[0073] S5, according to the regression model parameters solved Determine the LASSO logarithmic regression model; in the LASSO logarithmic regression model, tree rings are used as reconstruction factors, and the runoff sequence to be reconstructed Logarithmic transformation of As the response variable; each nested layer is regressed using the LASSO logarithmic regression model, and each nested layer of regression is nested to obtain the reconstructed runoff series.

[0074] Among them, the nesting level nesting step ( Figure 2 )include:

[0075] The tree ring sequence in the selected tree ring standard chronology is normalized, and the normalization formula is:

[0076] ;

[0077] Where: It is the original data of tree-ring chronology; It is a normalized tree-ring chronology; and Represent the mean and standard deviation of the original tree-ring chronology within the rate period, respectively.

[0078] The normalization process includes the normalization of the tree ring sequence and the normalization of the runoff reconstruction value. This procedure can avoid the artificial variability of the runoff reconstruction sequence of different nesting layers in the connection process due to the difference in variance of different nesting layers. At the same time, restoring the variance loss caused by regression also reduces the bias for the comparison of current runoff with past runoff.

[0079] Taking each nesting period as the reconstruction time span, the tree-ring chronology covering the entire period was selected, and the normalized nesting layer corresponding to each nesting period was regressed using the LASSO logarithmic regression model to obtain the regression nesting layer;

[0080] After each nested layer completes the regression, the standard deviation of the reconstructed runoff value needs to be adjusted against the standard deviation of the recorded runoff data to recover the lost variance. The calculation method is as follows:

[0081] ;

[0082] Where: and Respectively represent The original runoff reconstruction value and the adjusted runoff reconstruction value of the nested layer; and Respectively represent The mean and standard deviation of the original runoff reconstruction value rate of the nested layer within the period; represents the standard deviation of the recorded runoff values, It represents the average of the rate during the regular and validation periods.

[0083] All adjusted regression nesting layers are sorted from small to large based on the starting year;

[0084] In the adjacent nested regression layers, the starting year of the larger ranking is used as the ending year of the smaller ranking regression reconstruction layer, and then the smaller ranking regression reconstruction layer is truncated to obtain the secondary reconstruction layer;

[0085] All sub-reconstructed layers are connected from small to large according to the starting year to obtain the reconstructed runoff sequence.

[0086] S6. The runoff history data were divided into a model calibration period and a validation period to calibrate the model parameters and evaluate the model performance, respectively. The LASSO logarithmic regression model parameters and the optimal tree-ring subset were calculated using the data from the calibration period, while the LASSO logarithmic regression model performance was evaluated using the data from the validation period, which gave a less biased performance estimate.

[0087] In order to evaluate the proposed nested principal component regression runoff reconstruction model, a series of statistical validation measures are calculated for the results of each nested subset model. Widely used validation indicators include the rate period multiple determination coefficient, the square of the Pearson correlation coefficient during the validation period, the validation period error reduction, and the validation period effectiveness coefficient. They can show the error between the reconstructed value and the actual value. The larger the value, the more similar the reconstruction result is to the recorded data, and they are equal to 0, which means that the reconstructed runoff estimate is not as good as the direct use of the rate period runoff average value. For a data series with n years of rate period and m years of validation period, these three indicators can be calculated as follows:

[0088] ;

[0089] ;

[0090] ;

[0091] ;

[0092] Where: Expressed as the coefficient of determination corresponding to the rate-periodic runoff reconstruction, represents the determination coefficient corresponding to the runoff reconstruction during the validation period, represents the relative error corresponding to the runoff reconstruction during the validation period, represents the cross entropy corresponding to the reconstruction of runoff during the validation period; Representative rate regular runoff record data, Represents the runoff record data during the validation period; represents the reconstructed runoff value during the validation period, Representative rate regular runoff reconstruction value; and represent the mean of runoff record data during the regular period and the validation period, represents the mean value corresponding to the regular runoff reconstruction value, Represents the mean corresponding to the reconstructed runoff value during the validation period.

[0093] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0094] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art can still modify the technical solutions described in the aforementioned embodiments or replace some of the technical features therein by equivalents. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A paleorunoff reconstruction method based on LASSO logarithmic regression, characterized in that: The steps include: According to the time span and geographical location of the tree rings, tree ring data that meet the research time length and are near the research area are initially selected, and then the Pearson correlation coefficient is used to measure the correlation between the tree rings and annual runoff for secondary tree ring screening; The secondary screened tree-ring sequence is de-trended, and a standard tree-ring chronology for each sampling point is established; wherein the steps of constructing the standard tree-ring chronology include: Four detrending methods, namely negative exponential function method, 67% spline function method, "Signal-free" detrending method and empirical mode decomposition method, were used to fit the growth trend of tree rings at different sampling points. The corresponding growth trends were eliminated from the growth trend fitting results of tree rings at different sampling points according to the four detrending methods, and four tree ring standard chronologies corresponding to the four detrending methods were constructed for each sampling point. The correlation between the tree ring standard chronology obtained based on different detrending methods at the same sampling point and the runoff of the nearby hydrological station was compared, and the detrending method with the highest average correlation was selected as the optimal detrending method for the tree rings of the sampling point, and the tree ring standard chronology of each sampling point was established using the optimal detrending method. The nesting layers were divided according to the standard chronology of tree rings at each sampling site; The least angle regression algorithm and cross validation method are used to solve the regression model parameters of the LASSO logarithmic regression model. ; According to the regression model parameters solved Logarithmic regression model; In the LASSO logarithmic regression model, tree rings are used as reconstruction factors, so the runoff series to be reconstructed is Logarithmic transformation of as the response variable; Each nested layer is regressed using the LASSO logarithmic regression model, and each regressed nested layer is nested to obtain the reconstructed runoff sequence.

2. The paleorunoff reconstruction method based on LASSO logarithmic regression according to claim 1 is characterized in that: According to the year when the tree-ring chronology starts, a series of nested periods that extend continuously into the past are divided; the nested periods have different starting years but the same ending years; the runoff sequence corresponding to each nested period is called a nested layer.

3. The paleoflow reconstruction method based on LASSO logarithmic regression according to claim 1 is characterized in that: The regression model parameters It is expressed as: ； Where: represents the tree-ring chronology, , represents sequence 1 in the dendrochronology, represents sequence 2 in the dendrochronology, represents the sequence m in the tree-ring chronology, represents the runoff sequence to be reconstructed, Indicates the value when the cost function is minimum in the real number range, Expressed as The L2-norm of is the penalty parameter and , and Corresponding and , for The L1-norm of is the number of runoff observations.

4. The paleorunoff reconstruction method based on LASSO logarithmic regression according to claim 1 is characterized in that: The steps of determining the parameters of the LASSO logarithmic regression model using the cross-validation method include: The original runoff data set is evenly divided into five subsets. One of the five subsets is selected as the test set, and the other four subsets are used as training sets. It is ensured that each subset is used as a test set once. Each training set and test set uses the minimum angle regression algorithm to solve the regression model parameters. , and the regression model parameters Conduct ratings; The LASSO logarithmic regression model scoring results of the five tests were averaged using the 5-fold crossover error calculation formula to obtain the final performance evaluation of the LASSO logarithmic regression model; The regression model parameters corresponding to the minimum mean value obtained by calculating the 5-fold cross error Determine the parameters for the LASSO logistic regression model.

5. The paleorunoff reconstruction method based on LASSO logarithmic regression according to claim 1 is characterized in that: The minimum angle regression algorithm is used to take the bisector direction of the angle formed by the projection direction and the residual direction of the forward gradient algorithm for the regression model parameters. Perform mobile solution, the specific steps include: 1) Initialize weight coefficients ; 2) Initialize the residual vector as the target label vector ; 3) Select a feature vector that has the greatest correlation with the residual vector , along the eigenvector Find a set of weight coefficients in the direction of , another eigenvector with the greatest correlation with the residual vector appears Make the correlation between the new residual vector and the two eigenvectors equal, and recalculate the residual vector; the eigenvector is the sequence in the tree-ring chronology in the LASSO logarithmic regression model; 4) Repeat step 3) and continue to find a set of weight coefficients , so that the third eigenvector with the greatest correlation with the residual vector Make the correlation between the new residual vector and the three eigenvectors equal, until the residual vector is less than or equal to the set threshold or all eigenvectors have been selected, the iteration ends, and the final set of weight coefficients is obtained. .

6. The paleorunoff reconstruction method based on LASSO logarithmic regression according to claim 4 is characterized in that: The 5-fold crossover error calculation formula is: ； Where: represents the 5-fold crossover error, For the The residual sum of squares corresponding to the group LASSO logistic regression model.

7. The paleorunoff reconstruction method based on LASSO logarithmic regression according to claim 1 is characterized in that: The nesting layer nesting step comprises: Normalize the tree-ring sequences in the selected standard tree-ring chronology; Taking each nesting period as the reconstruction time span, the tree-ring chronology covering the entire period was selected, and the normalized nesting layer corresponding to each nesting period was regressed using the LASSO logarithmic regression model to obtain the regression nesting layer; Sort all regression nesting layers from small to large based on the starting year; In the adjacent nested regression layers, the starting year of the larger ranking is used as the ending year of the smaller ranking regression reconstruction layer, and then the smaller ranking regression reconstruction layer is truncated to obtain the secondary reconstruction layer; All sub-reconstructed layers are connected from small to large according to the starting year to obtain the reconstructed runoff sequence.

Images