A Method for Reconstructing Annual Precipitation Based on Nested Principal Component LASSO Regression

By using the nested principal component LASSO regression method to screen the optimal tree-ring combination and establishing a PCR-LASSO regression model, the problems of multicollinearity and redundancy in tree-ring data reconstruction were solved, achieving high-precision precipitation reconstruction over long time scales and improving the stability of the model and the reliability of the reconstruction results.

CN123116202APending Publication Date: 2026-11-17CHINA YANGTZE POWER
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611167901.9
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-08-03
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

Existing precipitation reconstruction methods suffer from multicollinearity when using tree-ring data, making it difficult to balance long time spans and reconstruction accuracy. They also lack a screening mechanism for the optimal tree-ring combination, resulting in insufficient model stability and generalization ability.

Method used

A nested principal component LASSO regression method was adopted to screen the optimal combination of tree-ring predictors through stepwise optimal subset selection and principal component analysis, and a PCR-LASSO regression model was established to reconstruct annual precipitation.

Benefits of technology

It achieves high-precision precipitation reconstruction over long time scales, improves model stability and the reliability of reconstruction results, reduces the impact of redundant information, and enhances the model's generalization ability.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123116202A_ABST
    Figure CN123116202A_ABST
Patent Text Reader

Abstract

An annual precipitation reconstruction method based on nested principal component LASSO regression is proposed, belonging to the field of hydroclimate reconstruction and paleoclimatology technology. This method, executed by computer equipment, first collects tree-ring representative year sequence data and corresponding observed annual precipitation sequences from the target area and surrounding areas. Tree-ring sequences significantly correlated with precipitation are screened as candidate predictors through correlation analysis and significance testing. Then, multiple nested time periods with the same end year but different start years are constructed based on the start year of each tree-ring sequence, and the tree-ring predictors in each nested time period are standardized. Subsequently, stepwise optimal subset selection is performed on the candidate tree-ring predictors to determine the optimal subset. Principal component analysis is performed on the optimal subset to extract principal component variables. Using the principal component variables as input variables and observed annual precipitation as output variables, a PCR-LASSO regression model is established, and the optimal regularization parameter is determined through cross-validation. Precipitation is reconstructed for each nested time period using the PCR-LASSO regression model, and variance recovery is performed on the reconstruction results. The reconstruction results of each nested time period are nested and stitched together in chronological order to obtain a continuous historical annual precipitation reconstruction sequence. Finally, the reconstruction model is validated and evaluated using data from the rate period and validation period.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention belongs to the field of hydro-climate reconstruction and paleoclimatology technology, and specifically relates to an annual precipitation reconstruction method based on nested principal component LASSO regression. Background Technology

[0002] Precipitation is one of the most important components of the climate system, and its long-term variations directly affect regional ecological environment, water resource allocation, agricultural production, and socio-economic development. However, modern meteorological records typically only span a few decades to a hundred years, which is insufficient to meet the needs of studying millennial-scale climate change patterns. Therefore, reconstructing historical precipitation sequences using proxy data such as tree rings, ice cores, and lake sediments has become an important method for paleoclimate research.

[0003] Among numerous proxies, tree-ring data offers advantages such as accurate dating, high temporal resolution, good continuity, and sensitivity to climate changes. Tree radial growth is influenced by multiple climatic factors, including precipitation, temperature, and solar radiation. Soil moisture is a crucial limiting factor for tree growth, and it is primarily controlled by precipitation. Therefore, a strong correlation usually exists between tree-ring width sequences and precipitation variations, providing a reliable basis for reconstructing historical precipitation using tree-ring data.

[0004] Existing precipitation reconstruction methods mainly employ statistical methods such as linear regression, multiple regression, and traditional principal component regression. However, in practical applications, the number of tree-ring predictors is vast, and different tree-ring sequences often exhibit strong correlations and redundant information, easily leading to multicollinearity and reducing model stability and reconstruction accuracy. Furthermore, as historical periods extend forward, the number of usable tree-ring sequences gradually decreases, making it difficult for traditional reconstruction models to balance the long time span with reconstruction accuracy. In addition, existing methods typically use all tree-ring sequences directly as predictors, lacking a mechanism for selecting the optimal tree-ring combination, which can easily introduce irrelevant information and affect the model's generalization ability.

[0005] Therefore, there is an urgent need to propose a method that can make full use of tree-ring data of different time lengths, automatically select the optimal tree-ring combination, reduce the dimensionality of variables and the influence of collinearity, and achieve high-precision precipitation reconstruction over long time scales, so as to improve the reliability and stability of historical precipitation reconstruction results. Summary of the Invention

[0006] The technical problem to be solved by this invention is to provide an annual precipitation reconstruction method based on nested principal component LASSO regression. This method utilizes the statistical relationship between tree-ring proxies and precipitation, selects the optimal combination of tree-ring predictive factors through a stepwise optimal subset selection method, and establishes an annual precipitation reconstruction model by combining a nested principal component regression model, thereby achieving high-precision reconstruction of long-term historical annual precipitation sequences.

[0007] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows: A method for reconstructing annual precipitation based on nested principal component LASSO regression, comprising the following steps: S1. Collect tree-ring representative year sequence data and corresponding observed annual precipitation sequences of the target area and surrounding areas, and screen tree-ring sequences that are significantly related to precipitation as candidate predictive factors through correlation analysis and significance test; S2. Construct multiple nested time periods based on the starting year of each tree ring sequence. Each nested time period has the same ending year but different starting years. S3. Standardize the tree-ring prediction factors in each nested time period; S4. Perform stepwise optimal subset selection on the candidate tree-ring predictors in each nested time period to determine the optimal subset of tree-ring predictors for the corresponding nested time period; S5. Perform principal component analysis on the optimal tree-ring prediction factor subset and extract principal component variables; S6. Using the principal component variables as input variables and the observed annual precipitation as output variables, establish a PCR-LASSO regression model, and determine the optimal regularization parameter through cross-validation. S7. The established PCR-LASSO regression model was used to reconstruct precipitation for each nested time period, and the variance recovery was performed on the reconstruction results. S8. The reconstruction results of each nested time period are nested and spliced ​​in chronological order to obtain a continuous historical annual precipitation reconstruction sequence. S9. Use periodic and validation period data to validate and evaluate the reconstruction model, and output model evaluation indicators.

[0008] Preferably, in step S1, when collecting tree-ring representative annual sequence data, preliminary screening is performed based on the time span and spatial location of the tree-ring sequence to retain tree-ring sequences that can cover regular periods and have sufficient length; the observed annual precipitation sequence is constructed from annual precipitation data of one or more meteorological stations in the target area.

[0009] Preferably, in step S1, the correlation between the tree-ring sequence and the observed annual precipitation sequence is evaluated using the Pearson correlation coefficient, which is calculated according to the following formula:

[0010] in, The Pearson correlation coefficient is used. For the first Tree-ring sequence values ​​for each sample year For the first Annual precipitation values ​​for each sample year The mean of the tree-ring sequence. The average annual precipitation series. The sample length is given; the significance of the Pearson correlation coefficient is tested, and tree-ring sequences that meet the preset significance level are retained as candidate predictors.

[0011] Preferably, in step S1, when screening candidate predictive factors, the correlation between the current year's tree ring index and the tree ring index lagged by one year and the observed annual precipitation sequence is analyzed simultaneously, and the current year's tree ring sequence and / or the tree ring sequence lagged by one year that meet the preset significance level requirements with the observed annual precipitation sequence are selected as candidate predictive factors.

[0012] Preferably, in step S2, a nested time structure is constructed from the most recent to the oldest starting year of each candidate predictor, with the multiple nested time periods having the same ending year; as the nested time periods extend into historical periods, the number of available tree-ring predictors gradually decreases, forming multiple nested layer structures with different time spans. The nested layer refers to the data layer and model layer corresponding to a nested time period.

[0013] Preferably, in step S2, the starting year interval between adjacent nested time periods is not less than 20 years; in one embodiment, the starting year interval between adjacent nested layers constructed based on the starting year of the tree ring representative year sequence is not less than 30 years.

[0014] Preferably, in step S3, in each nested layer, the tree-ring data is standardized using the mean and standard deviation of the tree-ring sequence within the periodic period. The standardization process is performed according to the following formula:

[0015] in, This is the original tree-ring data. For standardized tree-ring data, For the periodic average of the rate, The periodic standard deviation is the rate.

[0016] Preferably, in step S4, when performing stepwise optimal subset selection on the candidate tree-ring predictors in each nested layer, the sum of squared residuals is first used as an evaluation index to determine the optimal tree-ring combination under different dimensions. The sum of squared residuals is calculated according to the following formula:

[0017] in, For the sum of squared residuals, For the first The observed annual precipitation for each sample year For the first Reconstructed annual precipitation for each sample year This represents the sample length.

[0018] Preferably, in step S4, after determining the optimal tree-ring combination under different dimensions, the Akaike Information Criterion is used to evaluate the overall performance of the model. The Akaike Information Criterion is calculated according to the following formula:

[0019] in, For the first The Akaike Information Criterion value corresponding to each candidate tree-ring combination. In model parameters and observed annual precipitation series The likelihood function under the given conditions, For the first The number of parameters to be estimated in the model corresponding to each candidate tree-ring combination; the optimal tree-ring combination is searched by progressively increasing the dimension. When the Akaike Information Criterion value no longer decreases after adding a new predictor, the variable addition process is stopped, and the optimal subset of tree-ring predictor factors for the current nested layer is determined.

[0020] Preferably, in step S5, a standardized tree-ring matrix is ​​established for the optimal subset of tree-ring predictors in each nested layer:

[0021] in, The standardized tree-ring matrix, These are the first to the second elements in the current nested level. The optimal tree-ring predictor This represents the optimal number of tree-ring predictors.

[0022] Preferably, in step S5, the covariance matrix is ​​constructed based on the standardized tree ring matrix, and the eigenvalues ​​and eigenvectors are solved, specifically according to the following formula:

[0023]

[0024] in, Let covariance matrix be the variance matrix. The standardized tree ring matrix The transpose of the matrix, For sample length, For the first 1 eigenvector To and The corresponding number Each feature value.

[0025] Preferably, in step S5, the principal component variables are obtained based on the feature vectors, specifically according to the following formula:

[0026] in, For the first One principal component variable, The standardized tree-ring matrix, For the first One feature vector; retain the first one according to the cumulative variance contribution rate threshold or significance test results. The principal component variables are used as input variables for the PCR-LASSO regression model. The number of principal component variables to be retained.

[0027] Preferably, in step S6, a PCR-LASSO regression model is established using the retained principal component variables as independent variables and the observed annual precipitation as the dependent variable. The PCR-LASSO regression model is expressed as follows:

[0028] in, To observe annual precipitation, For the intercept term, For the first One principal component variable, For the first The regression coefficients corresponding to the principal component variables. To retain the number of principal component variables, This is random error.

[0029] Preferably, in step S6, LASSO regularization is used to solve the parameters of the PCR-LASSO regression model, and the parameter solution objective is expressed as follows:

[0030] in, This is the regression coefficient estimation vector. Let be the vector of regression coefficients to be solved. For the first The observed annual precipitation for each sample year For the first The first sample year corresponding to the first The values ​​of the principal component variables, is the regularization parameter; the LASSO regularization is used to compress the coefficients of unimportant variables, causing some regression coefficients to shrink to zero.

[0031] Preferably, in step S6, the regularization parameter is determined by cross-validation. The optimal value is found; in one implementation, the leave-one-out cross-validation method is used to calculate the cross-validation error under different regularization parameters, and the regularization parameter corresponding to the minimum cross-validation error is selected as the optimal regularization parameter.

[0032] Preferably, in step S7, the variance recovery process specifically involves adjusting the standard deviation of the precipitation reconstruction results obtained from each nested layer, and recovering the variance lost during the regression process using the statistical characteristics of the observed annual precipitation series. The variance recovery process is performed according to the following formula:

[0033] in, For the first Original reconstructed precipitation for each sample year For the first Adjusted reconstructed precipitation for each sample year To reconstruct the average precipitation, To reconstruct the standard deviation of precipitation, To observe the average precipitation, To observe the standard deviation of precipitation.

[0034] Preferably, in step S8, the reconstruction results of each nested layer are extended forward layer by layer according to the order of the starting year from the nearest to the furthest. The long-term span nested layer is used to supplement the missing historical precipitation information before the short-term span nested layer, so as to obtain a continuous long-term scale historical annual precipitation reconstruction sequence.

[0035] Preferably, in step S9, the observed annual precipitation data is divided into a calibration period and a validation period. The model parameters are calibrated using the calibration period data, and the model performance is evaluated using the validation period data. The evaluation uses model evaluation indicators, including the Pearson correlation coefficient. The coefficient of determination (VRSQ) and the efficiency coefficient (CE) were verified. When the model evaluation index reached the preset threshold, it was determined that the PCR-LASSO regression model had an effective precipitation reconstruction capability.

[0036] Preferably, in step S9, Bootstrap resampling and / or Jackknife resampling methods are used to evaluate the stability of the model; when the statistical parameters obtained by resampling are consistent with the statistical parameters of the original PCR-LASSO regression model, the PCR-LASSO regression model is determined to be stable.

[0037] Preferably, the output of the continuous historical annual precipitation reconstruction sequence includes the historical year, the reconstructed annual precipitation, the starting year of each nested layer, the number of tree-ring predictors involved in the reconstruction, the number of principal component variables retained, the regularization parameter, the precipitation reconstruction results after variance recovery, and the model evaluation index, and is stored or output in the form of electronic data files.

[0038] The present invention can achieve the following beneficial effects: 1. This invention collects tree-ring representative year sequence data and corresponding observed annual precipitation sequences from the target area and surrounding areas, and filters tree-ring sequences that are significantly correlated with precipitation through correlation analysis and significance tests. This reduces the number of tree-ring sequences with weak precipitation response entering the modeling process and improves the effectiveness of the predictive factor source.

[0039] 2. When screening candidate predictive factors, this invention simultaneously analyzes the correlation between the current year's tree-ring index and the tree-ring index lagged by one year and the observed annual precipitation sequence. This can take into account the response of tree growth to the current year's precipitation and previous water conditions, and improve the completeness of tree-ring data in representing annual precipitation changes.

[0040] 3. This invention constructs multiple nested time periods based on the start year of each tree-ring sequence, so that each nested time period has the same end year but different start years. This can make full use of tree-ring data of different time lengths, use more predictive factors in more recent historical periods, and use long-sequence predictive factors that can cover the period in earlier historical periods, thereby balancing reconstruction accuracy and time length.

[0041] 4. This invention standardizes the tree-ring predictor factors in each nested time period and uses the mean and standard deviation of the tree-ring sequence within the period to unify the variable scale, which can reduce the impact of differences in the dimensions and fluctuation amplitude of different tree-ring sequences on the estimation of model parameters.

[0042] 5. This invention performs stepwise optimal subset selection of candidate tree-ring predictors in each nested time period, which can screen out tree-ring combinations that are more suitable for precipitation reconstruction in each nested layer, avoiding the direct use of all tree-ring sequences, which would lead to irrelevant and redundant information entering the model.

[0043] 6. This invention combines the sum of squared residuals and the Akaike information criterion for evaluation in the stepwise optimal subset selection, which can balance the model fitting ability and model complexity, and reduce the risk of overfitting caused by too many predictors.

[0044] 7. This invention performs principal component analysis on the optimal subset of tree-ring predictors, converting multiple related tree-ring sequences into principal component variables. This can extract the main common variation information in tree-ring data and reduce the impact of multicollinearity among different tree-ring sequences on the stability of the regression model.

[0045] 8. This invention retains principal component variables as input variables for the PCR-LASSO regression model based on the cumulative variance contribution rate threshold or significance test results. This can reduce the dimensionality of input variables while retaining the main climate signals, thereby improving the model's computational efficiency and generalization ability.

[0046] 9. This invention establishes a PCR-LASSO regression model using principal component variables as input variables and observed annual precipitation as output variables. By compressing the coefficients of unimportant variables through LASSO regularization, the model can further screen principal component variables that contribute significantly to the reconstruction of annual precipitation, thereby reducing the interference of weakly contributing variables on the reconstruction results.

[0047] 10. This invention determines the optimal value of the LASSO regularization parameter through cross-validation, which avoids the subjectivity caused by manually setting the regularization parameter and maintains a balance between the model's fitting ability on calibration samples and its predictive ability on validation samples.

[0048] 11. This invention performs variance recovery processing on the precipitation reconstruction results obtained from each nested layer, and uses the statistical characteristics of the observed annual precipitation series to recover the variance lost during the regression process. This can reduce the problem of low amplitude of the reconstructed series caused by principal component regression and LASSO regularization, and make the reconstructed precipitation series closer to the statistical variation characteristics of observed precipitation.

[0049] 12. This invention nests and splices the reconstruction results of each nested time period in chronological order, which can supplement the missing historical precipitation information before the short time span nested layer by using the long time span nested layer, thereby obtaining a continuous long-term scale historical annual precipitation reconstruction sequence.

[0050] 13. In the process of nested splicing, the present invention can perform mean, standard deviation and trend consistency checks on the overlapping time periods of adjacent nested layers, which can reduce the inter-layer mutation caused by the change in the number of prediction factors between different nested layers and improve the connection stability of the continuous historical annual precipitation reconstruction sequence.

[0051] 14. This invention uses data from the rate-period and validation periods to validate and evaluate the reconstruction model, and uses indicators such as Pearson correlation coefficient, validation determination coefficient VRSQ, and efficiency coefficient CE to judge the model performance. It can evaluate the reliability of precipitation reconstruction results from aspects such as fit, validation ability, and prediction effectiveness.

[0052] 15. The present invention can also use Bootstrap resampling and / or Jackknife resampling methods to evaluate model stability, which can identify the impact of single sample year or sample sampling changes on model results and improve the uncertainty assessment capability of annual precipitation reconstruction results.

[0053] 16. When outputting the continuous historical annual precipitation reconstruction sequence, this invention can simultaneously output the historical year, the reconstructed annual precipitation, the starting year of each nested layer, the number of tree-ring predictors involved in the reconstruction, the number of principal component variables retained, the regularization parameter, the variance recovery result, and the model evaluation index, which facilitates subsequent regional hydrological and climate change analysis, water resource assessment, and long-term dry and wet change research. Attached Figure Description

[0054] The present invention will be further described below with reference to the accompanying drawings and embodiments: Figure 1 This is a schematic diagram showing the spatial distribution of the research area, meteorological stations, and tree-ring sampling points in the Yangtze River source region; Figure 2 It is a heatmap showing the correlation between tree-ring chronology and annual precipitation sequence; Figure 3 This is a schematic diagram showing the changes in the starting year, the number of tree-ring predictors, and the number of principal components for each nested layer; Figure 4 It is a comparison diagram of the observed annual precipitation series and the reconstructed annual precipitation series during the overlap period; Figure 5 This is a graph showing the changes in evaluation metrics r, VRSQ, and CE for models with different nesting levels; Figure 6 This is a map showing the reconstructed annual precipitation sequence and its 11-year moving average in the Yangtze River source region from 1584 to 1999.

[0055] In the picture: Figure 2 In this context, “NDC107M_raw”, “QML005A_raw”, “QML007A_raw”, “QML016A_raw”, “QML025A_raw”, “QMD77B_raw”, “XDC110M_raw”, “XDC191M_raw”, “XDC210M_raw”, “XDC214M_raw”, “XHC145M_raw”, and “XHC157M_raw” are the names of the selected tree-ring sequences, while “Yushu”, “Tuotuohe”, “Zaduo”, “Wudaoliang”, “Qumalai”, and “Qingshuihe” are the names of the meteorological stations. Figure 3 In this context, "StartYear" indicates the starting year, "Number of tree rings" indicates the number of tree ring predictors, and "Number of LASSO variables" indicates the number of variables retained after LASSO regularization. Figure 4 In this context, "Estimated data" refers to reconstructed annual precipitation series, while "Observed data" refers to observed annual precipitation series. Figure 6 In this context, "Wet" indicates the wet season, and "Dry" indicates the dry season. Detailed Implementation

[0056] The present invention will be further described below with reference to the accompanying drawings and embodiments. Without departing from the concept of the present invention, any equivalent substitutions or conventional adjustments made by those skilled in the art regarding the source of tree-ring chronologies, the number of meteorological stations, the nesting layer division method, the principal component variable retention threshold, the cross-validation method, and the model evaluation index thresholds should fall within the protection scope of the present invention.

[0057] like Figures 1 to 6 As shown, this invention provides a method for reconstructing annual precipitation based on nested principal component LASSO regression, applicable to reconstructing long-term historical annual precipitation sequences using tree-ring representative year sequences from a target area and surrounding regions. This method is executed by a computer device. The tree-ring representative year sequence data and observed annual precipitation sequences are stored in the computer device's memory in the form of electronic spreadsheets or data matrices. The processor performs data processing operations such as correlation analysis, standardization, stepwise optimal subset selection, principal component analysis, PCR-LASSO regression modeling, variance recovery, and nested splicing. The method first collects tree-ring representative year sequence data and corresponding observed annual precipitation sequences from the target area and surrounding regions. Tree-ring sequences significantly correlated with precipitation are screened as candidate predictors through correlation analysis and significance testing. Then, multiple nested time periods are constructed based on the starting year of each tree-ring sequence. The candidate tree-ring predictors in each nested time period are standardized, stepwise optimal subset selection is performed, principal component analysis is conducted, and PCR-LASSO regression modeling is performed. Subsequently, precipitation reconstruction, variance recovery, and nested splicing are performed on each nested time period to obtain a continuous historical annual precipitation reconstruction sequence. Finally, the reconstruction model is validated and evaluated using data from the rate-setting and validation periods.

[0058] In this embodiment, the tree-ring representative year sequence data refers to a standardized tree-ring chronology sequence that can characterize the annual growth changes of trees in the target area or surrounding areas; the observed annual precipitation sequence refers to the annual precipitation sequence formed by quality control and regional averaging of annual precipitation data from one or more meteorological stations in the target area; the candidate predictor refers to the current year's tree-ring sequence and / or the tree-ring sequence lagged by one year that have a significant statistical correlation with the observed annual precipitation sequence; the nested time period refers to a time interval with the same ending year but different starting years; the nested layer refers to the data layer and model layer corresponding to a nested time period; and the PCR-LASSO regression model refers to a principal component regression model that takes principal component variables as input, observes annual precipitation as output, and uses LASSO regularization to solve the regression coefficients.

[0059] S1, Tree-ring predictor screening like Figure 1 and Figure 2As shown, tree-ring representative annual sequence data and corresponding observed annual precipitation sequences were collected from the target area and surrounding areas. Tree-ring sequences that are significantly correlated with precipitation were selected as candidate predictors through correlation analysis and significance tests.

[0060] Specifically, preliminary screening is conducted based on the temporal span and spatial location of the tree-ring sequences, retaining those with regular coverage and sufficient length. The observed annual precipitation sequence is constructed from annual precipitation data from one or more meteorological stations within the target area. When multiple meteorological stations exist in the target area, the annual precipitation data of each station can first undergo checks for missing data, outliers, and homogeneity. Then, the observed annual precipitation sequence for the target area is constructed using regional averaging, area-weighted averaging, or a weighted averaging method based on station representativeness.

[0061] The correlation between tree-ring sequences and observed annual precipitation sequences is evaluated using the Pearson correlation coefficient, which is calculated according to the following formula:

[0062] In the formula, Pearson correlation coefficient; The year of the sample is used as the year number; For the first Tree-ring sequence values ​​for each sample year; For the first Annual precipitation values ​​for each sample year; The mean of the tree-ring sequence; This represents the annual precipitation series mean. This represents the sample length.

[0063] When screening candidate predictive factors, the correlation between the current year's tree-ring index and the tree-ring index lagged one year later and the observed annual precipitation series is analyzed simultaneously. For any tree-ring series, if its current year's tree-ring series and the observed annual precipitation series meet the preset significance level requirement, then the current year's tree-ring series is used as a candidate predictive factor; if its tree-ring series lagged one year later and the observed annual precipitation series meet the preset significance level requirement, then the tree-ring series lagged one year later is used as a candidate predictive factor. The preset significance level can be 0.05 or 0.01, or it can be set according to the sample length and the climate response characteristics of the study area.

[0064] like Figure 2 As shown, the correlation heatmap can be used to display the correlation strength between each tree-ring chronology and the annual precipitation series of each meteorological station. This map can be used to identify tree-ring sequences with a strong response to precipitation, and, combined with significance test results, to determine candidate predictors.

[0065] S2. Construct a nested time structure like Figure 3As shown, multiple nested time periods are constructed based on the starting year of each tree-ring sequence. Each nested time period has the same ending year but different starting years.

[0066] Specifically, nested time structures are constructed from the most recent to the oldest based on the start year of each candidate predictor. The end year of each nested time period coincides with the end year of the period covered by the observed annual precipitation sequence or tree-ring data, while the start year is determined stratified based on the earliest start year of the available tree-ring sequence. As the nested time periods extend into historical periods, the number of available tree-ring predictors gradually decreases, forming multiple nested layer structures with different time spans.

[0067] To avoid model structure duplication due to excessively short time spans between adjacent nested layers, the starting year interval between adjacent nested time periods is no less than 20 years; in one implementation, the starting year interval between adjacent nested layers constructed based on the starting year of the tree-ring representative year sequence is no less than 30 years. Each nested layer corresponds to a set of available candidate predictors, a rate period, a PCR-LASSO regression model, and a nested layer reconstruction result.

[0068] S3, Tree-ring sequence normalization The tree-ring predictor factors in each nested time period are standardized.

[0069] Specifically, within each nested layer, the tree-ring data are standardized using the mean and standard deviation of the tree-ring sequence within the rate period to eliminate dimensional and scale differences between different tree-ring sequences. The standardization process is performed according to the following formula:

[0070] In the formula, This is the original tree-ring data; This is the standardized tree-ring data; The mean of the tree-ring sequence within the rate period; The standard deviation of the tree-ring sequence within the rate period; The year number is the sample year.

[0071] For any nested layer, only the mean and standard deviation within the rate period of that nested layer are used as standardized parameters, and the same standardized parameters are applied to the historical reconstruction period of that nested layer. This approach avoids the influence of historical data on the rate period statistical parameters and ensures that the variable scale remains consistent between the model training and historical reconstruction phases.

[0072] S4. Stepwise optimal subset selection A stepwise optimal subset selection is performed on the candidate tree-ring predictors in each nested time period to determine the optimal subset of tree-ring predictors for the corresponding nested time period.

[0073] Specifically, when performing stepwise optimal subset selection of candidate tree-ring predictors in each nested layer, the sum of squared residuals is first used as an evaluation metric to determine the optimal tree-ring combination under different dimensions. The sum of squared residuals is calculated according to the following formula:

[0074] In the formula, It is the sum of squared residuals; The year of the sample is used as the year number; For the first The observed annual precipitation for each sample year; For the first The reconstructed annual precipitation for each sample year; This represents the sample length.

[0075] After determining the optimal tree-ring combinations under different dimensions, the Akaike Information Criterion is used to evaluate the overall performance of the model. The Akaike Information Criterion is calculated according to the following formula:

[0076] In the formula, For the first The Akaike information criterion values ​​corresponding to each candidate tree-ring combination; In model parameters and observed annual precipitation series The likelihood function under the given conditions; For the first The number of parameters to be estimated in the model corresponding to each candidate tree-ring combination; Number the candidate tree-ring combinations.

[0077] The optimal tree-ring combination is searched using a stepwise increasing dimensionality approach. For each candidate dimension, the sum of squared residuals of all available tree-ring combinations or those obtained through the stepwise search is calculated, and the tree-ring combination with the smaller sum of squared residuals is selected for that dimension. Subsequently, the optimal tree-ring combination for different dimensions is comprehensively evaluated using the Akaike Information Criterion. When the Akaike Information Criterion value no longer decreases after adding a new predictor, the variable addition process is stopped, and the optimal subset of tree-ring predictors for the current nested layer is determined. This step is used to reduce irrelevant and redundant tree-ring sequences from entering subsequent principal component analysis and PCR-LASSO regression models.

[0078] S5, Principal Component Analysis Principal component analysis was performed on the optimal subset of tree-ring predictors to extract principal component variables.

[0079] Specifically, for the optimal subset of tree-ring predictors in each nested layer, a standardized tree-ring matrix is ​​constructed:

[0080] In the formula, This is the standardized tree-ring matrix; These are the first to the second elements in the current nested level. One optimal tree-ring predictor; This represents the optimal number of tree-ring predictors in the current nested layer.

[0081] Construct the covariance matrix based on the standardized tree ring matrix, and solve for the eigenvalues ​​and eigenvectors, specifically according to the following formula:

[0082]

[0083] In the formula, It is the covariance matrix; The standardized tree ring matrix The transpose of the matrix; The sample length; For the first 1 eigenvector; To and The corresponding number One eigenvalue; Principal component variable number.

[0084] The principal component variables are obtained from the eigenvectors, specifically according to the following formula:

[0085] In the formula, For the first One principal component variable; This is the standardized tree-ring matrix; For the first 1 eigenvector; Principal component variable number.

[0086] Retain the results according to the cumulative variance contribution rate threshold or significance test results. The principal component variables are used as input variables in the PCR-LASSO regression model. The cumulative variance contribution rate threshold can be 80%, 85%, or 90%, or it can be determined based on the model validation results; the significance test results can be determined based on the correlation between the principal component variables and the observed annual precipitation series, cross-validation error, or model evaluation indicators.

[0087] S6. Establish a PCR-LASSO regression model Using the principal component variables as input variables and the observed annual precipitation as output variables, a PCR-LASSO regression model was established, and the optimal regularization parameter was determined through cross-validation.

[0088] Specifically, a PCR-LASSO regression model was established using the retained principal component variables as independent variables and observed annual precipitation as the dependent variable. The PCR-LASSO regression model is expressed as follows:

[0089] In the formula, To observe annual precipitation; For the intercept term; Principal component variable numbering; For the first One principal component variable; For the first The regression coefficients corresponding to the principal component variables; The number of principal component variables to be retained; This is random error.

[0090] The parameters of the PCR-LASSO regression model were solved using LASSO regularization. The objective for parameter solving is expressed as follows:

[0091] In the formula, This is the regression coefficient estimation vector; Let be the vector of regression coefficients to be solved; The year of the sample is used as the year number; For the first The observed annual precipitation for each sample year; Principal component variable numbering; For the first The first sample year corresponding to the first The values ​​of each principal component variable; For the intercept term; For the first The regression coefficients corresponding to the principal component variables; The number of principal component variables to be retained; The sample length; This is the regularization parameter.

[0092] The LASSO regularization is used to compress coefficients of unimportant variables, shrinking some regression coefficients to zero. The optimal value of the regularization parameter is determined through cross-validation. In one embodiment, leave-one-out cross-validation is used to calculate the cross-validation error under different regularization parameters, and the regularization parameter corresponding to the minimum cross-validation error is selected as the optimal regularization parameter. The cross-validation error can be evaluated using metrics such as mean squared error, mean absolute error, or correlation coefficient during the validation period.

[0093] S7. Variance Recovery of Precipitation Reconstruction Results The established PCR-LASSO regression model was used to reconstruct precipitation for each nested time period, and the variance recovery was performed on the reconstruction results.

[0094] Specifically, for each nested layer, the annual precipitation within the corresponding historical period is reconstructed using the PCR-LASSO regression model of that nested layer, yielding the original reconstructed precipitation sequence for that nested layer. Since principal component regression and LASSO regularization may compress the variance of the reconstructed sequence, variance recovery processing is performed on the precipitation reconstruction results obtained from each nested layer. This variance recovery processing includes standard deviation adjustment and using the statistical characteristics of the observed annual precipitation sequence to recover the variance lost during the regression process. The variance recovery processing is performed according to the following formula:

[0095] In the formula, For the first Original reconstructed precipitation for each sample year; For the first Adjusted reconstructed precipitation for each sample year; To reconstruct the average precipitation; To reconstruct the standard deviation of precipitation; To observe the average precipitation; To observe the standard deviation of precipitation; The year number is the sample year.

[0096] After variance recovery processing, the reconstructed precipitation sequences within the same nested layer maintain consistent statistical characteristics in terms of mean and variance with the observed annual precipitation sequences, thereby reducing the amplitude underestimation problem caused by model regularization. For different nested layers, variance recovery can be performed using statistical parameters of their rate period or common overlap period to ensure the continuity of the nested and spliced ​​sequences on a statistical scale.

[0097] S8, Nested splicing and continuous historical annual precipitation reconstruction By nesting and splicing the reconstruction results of each nested time period in chronological order, a continuous historical annual precipitation reconstruction sequence is obtained.

[0098] Specifically, following the order of the most recent starting year to the oldest, the reconstruction results of each nested layer are extended forward layer by layer. Long-term nested layers are used to supplement the missing historical precipitation information before short-term nested layers, thus obtaining a continuous long-term historical annual precipitation reconstruction sequence. When there are overlapping periods between adjacent nested layers, the results of nested layers with shorter time spans and more predictors can be used as the output for the overlapping period. For earlier historical periods that cannot be covered by short-term nested layers, the results of nested layers with earlier starting years are used to supplement them.

[0099] During the stitching process, the overlapping periods of adjacent nested layers are checked for mean, standard deviation, and trend consistency. When systematic bias exists in the overlapping periods of the reconstructed sequences of adjacent nested layers, linear correction or mean-variance matching can be performed based on the statistical characteristics of the overlapping periods to ensure that there are no obvious abrupt changes at the junctions between layers in the stitched continuous historical annual precipitation reconstruction sequence. The output of the continuous historical annual precipitation reconstruction sequence includes the historical year, reconstructed annual precipitation, the starting year of each nested layer, the number of tree-ring predictors involved in the reconstruction, the number of principal component variables retained, regularization parameters, precipitation reconstruction results after variance recovery, and model evaluation indicators.

[0100] S9. Model Validation and Evaluation The reconstruction model was validated and evaluated using periodic and validation data.

[0101] Specifically, the observed annual precipitation data is divided into a calibration period and a validation period. The model parameters are calibrated using data from the calibration period, and the model performance is evaluated using data from the validation period. Model evaluation indices are used, including the Pearson correlation coefficient, the validation determination coefficient (VRSQ), and the efficiency coefficient (CE). When the model evaluation indices reach a preset threshold, the PCR-LASSO regression model is deemed to have effective precipitation reconstruction capabilities. The preset threshold can be determined based on the sample length, the hydro-climatic background of the study area, and the purpose of the model application.

[0102] like Figure 4 and Figure 5 As shown, the fit and validation capabilities of the reconstruction results can be judged by comparing the observed annual precipitation series and the reconstructed annual precipitation series during the overlap period, as well as by observing the changes in evaluation indicators for different nested models. When the reconstructed series can well reflect the interannual variation of the observed annual precipitation series, and the correlation coefficients, VRSQ, and CE of each nested layer meet the preset requirements, it indicates that the established PCR-LASSO regression model has precipitation reconstruction capabilities.

[0103] Furthermore, Bootstrap resampling and / or Jackknife resampling methods can be used to evaluate model stability. For Bootstrap resampling, sampling with replacement can be performed on the rate period sample, and the modeling can be repeated to statistically analyze the regression coefficients, correlation coefficients, and the uncertainty range of the reconstructed sequences. For Jackknife resampling, individual sample years can be successively removed and the model rebuilt to evaluate the impact of individual sample years on the model results. When the statistical parameters obtained from the resampling are consistent with the statistical parameters of the original PCR-LASSO regression model, the PCR-LASSO regression model is considered stable.

[0104] like Figure 6As shown, the final output of the continuous historical annual precipitation reconstruction sequence can be combined with moving average processing to display the characteristics of long-term precipitation changes. The moving average window can be 11 years, or other odd-numbered year windows can be selected according to research needs to analyze the characteristics of wet periods, dry periods, and interdecadal variations in precipitation in historical periods.

[0105] Example 1: Reconstruction of Annual Precipitation in the Yangtze River Source Region This embodiment uses the reconstruction of annual precipitation in the Yangtze River source region as an example to illustrate the implementation process of the present invention. The study area is the Yangtze River source region, ranging from 90°43′E to 96°45′E and from 32°30′N to 35°35′N. Figure 1 As shown, the study area includes multiple meteorological stations and tree-ring sampling points. The meteorological stations are used to construct the observed annual precipitation series, and the tree-ring sampling points are used to provide candidate predictive factors.

[0106] Annual precipitation data from 1959 to 1999 were collected from six national meteorological stations in the Yangtze River source region: Wudaoliang, Tuotuohe, Zhiduo, Qumalai, Yushu, and Qingshuihe. After a consistency check of the annual precipitation data from each station, an annual precipitation series for the study area was constructed using the regional averaging method, which served as the target variable for model calibration.

[0107] Tree-ring data were obtained from the International Tree-Ring Database, collecting 48 tree-ring chronologies covering the study area and surrounding regions, spanning from 1227 to 2011. All tree-ring sequences underwent initial screening, retaining those with sufficient length and covering the period from 1959 to 1999. Subsequently, using the Pearson correlation coefficient formula in S1, the correlation coefficients between the current year's tree-ring index and the tree-ring index lagged by one year and the observed annual precipitation sequence were calculated, and significance tests were performed. Figure 2 As shown, the correlation heatmap illustrates the correlation strength between different tree-ring chronologies and annual precipitation sequences at various meteorological stations. After significance testing, 15 tree-ring chronologies were selected as candidate predictors, with the starting years of the selected chronologies ranging from 1584 to 1914.

[0108] A nested structure was established based on the starting years of 15 tree-ring chronologies. To ensure sufficient time span differences between adjacent nested layers, the starting year interval for each nested layer was set to be no less than 30 years, ultimately constructing 5 nested layers: the first nested layer from 1914 to 1999; the second nested layer from 1791 to 1999; the third nested layer from 1754 to 1999; the fourth nested layer from 1714 to 1999; and the fifth nested layer from 1584 to 1999. Figure 3 As shown, as the nesting layers extend into historical periods, the number of available tree-ring predictors gradually decreases, while the time coverage continuously increases.

[0109] For each nested layer, the tree-ring data is first standardized using the mean and standard deviation of the tree-ring sequence within the rate period. Then, a stepwise optimal subset selection is performed on the candidate tree-ring predictors. In the stepwise optimal subset selection, the best tree-ring combination under different dimensions is first determined by the sum of squared residuals, and then the comprehensive performance of the models under different dimensions is evaluated by the Akaike Information Criterion. When the Akaike Information Criterion value no longer decreases after adding a new predictor, the optimal subset of tree-ring predictors for that nested layer is determined.

[0110] Principal component analysis was performed on the optimal subset of tree-ring predictors for each nested layer to extract principal component variables that reflect the common variation characteristics of tree rings. Subsequently, a PCR-LASSO regression model was established using the retained principal component variables as input variables and observed annual precipitation as the output variable. Leave-one-out cross-validation was used to determine the optimal regularization parameters for the LASSO model. In different nested layers, the LASSO model automatically selects the principal component variables that contribute significantly to precipitation changes and establishes corresponding precipitation reconstruction equations.

[0111] The nested models were tested using data from the rate-up and validation periods. The stability of the models was evaluated using Bootstrap resampling (1000 times) and Jackknife resampling methods. Figure 4 As shown, the observed annual precipitation series and the reconstructed annual precipitation series are consistent during the overlap period; as Figure 5 As shown, the correlation coefficients of each nested layer model The validation results show that both the coefficient of determination (VRSQ) and the efficiency coefficient (CE) remain positive, indicating that the constructed model has predictive power and stability. The statistical parameters obtained by Bootstrap and Jackknife resampling are consistent with the results of the original model, further demonstrating the stability of the method presented in this invention.

[0112] After recalibrating the model parameters using all overlapping period data, historical annual precipitation in the Yangtze River source region was reconstructed. Based on the length of the tree-ring chronology contained in the earliest nested layer, the annual precipitation sequence in the Yangtze River source region was extended back to 1584, ultimately obtaining a continuous annual precipitation reconstruction sequence from 1584 to 1999, with a total length of 416 years. Figure 6 As shown, the reconstructed sequence and its moving average results can reflect the long-term variation characteristics of annual precipitation in the Yangtze River source area, providing data support for regional climate change research and water resource assessment.

[0113] The above embodiments use the reconstruction of annual precipitation in the Yangtze River source region as an example. For other target areas, the candidate predictor screening threshold, the starting year of the nested layer, the number of principal component variables retained, the search range of regularization parameters, and the model evaluation index threshold can be adjusted according to the distribution of meteorological stations, the spatial location of tree-ring sampling points, the starting year of the tree-ring sequence, and the length of the observed annual precipitation sequence. As long as it still adopts the technical route of tree-ring predictor screening, nested time structure construction, principal component analysis, PCR-LASSO regression modeling, variance recovery, and nested splicing, it belongs to the implementation method of this invention.

Claims

1. A method for reconstructing annual precipitation based on nested principal component LASSO regression, executed by computer equipment, characterized in that, Includes the following steps: S1. Collect tree-ring representative year sequence data and corresponding observed annual precipitation sequences of the target area and surrounding areas, and screen tree-ring sequences that are significantly related to precipitation as candidate predictive factors through correlation analysis and significance test; S2. Construct multiple nested time periods based on the starting year of each tree ring sequence. Each nested time period has the same ending year but different starting years. S3. Standardize the candidate tree-ring predictors in each nested time period; S4. Perform stepwise optimal subset selection on the candidate tree-ring predictors in each nested time period to determine the optimal subset of tree-ring predictors for the corresponding nested time period; S5. Perform principal component analysis on the optimal tree-ring prediction factor subset and extract principal component variables; S6. Using the principal component variables as input variables and the observed annual precipitation as output variables, establish a PCR-LASSO regression model, and determine the optimal regularization parameter through cross-validation. S7. The established PCR-LASSO regression model was used to reconstruct precipitation for each nested time period, and the variance recovery was performed on the reconstruction results. S8. The reconstruction results of each nested time period are nested and spliced ​​in chronological order to obtain a continuous historical annual precipitation reconstruction sequence. S9. Use periodic and validation period data to validate and evaluate the reconstruction model, and output model evaluation indicators.

2. The method for reconstructing annual precipitation based on nested principal component LASSO regression according to claim 1, characterized in that, In step S1, when collecting tree-ring representative annual sequence data, preliminary screening is performed based on the time span and spatial location of the tree-ring sequence to retain tree-ring sequences that can cover regular periods and have sufficient length; the observed annual precipitation sequence is constructed from annual precipitation data of one or more meteorological stations in the target area.

3. The method for reconstructing annual precipitation based on nested principal component LASSO regression according to claim 1, characterized in that, In step S1, the correlation between the tree-ring sequence and the observed annual precipitation sequence is evaluated using the Pearson correlation coefficient, which is calculated according to the following formula: ; in, The Pearson correlation coefficient is used. For the first Tree-ring sequence values ​​for each sample year For the first Annual precipitation values ​​for each sample year The mean of the tree-ring sequence. The average annual precipitation series. The sample length is given; the significance of the Pearson correlation coefficient is tested, and tree-ring sequences that meet the preset significance level are retained as candidate predictors.

4. The method for reconstructing annual precipitation based on nested principal component LASSO regression according to claim 1, characterized in that, In step S1, when screening candidate predictive factors, the correlation between the current year tree ring index and the tree ring index lagged one year later and the observed annual precipitation series is analyzed simultaneously. The current year tree ring series and / or the tree ring series lagged one year later that meet the preset significance level requirements with the observed annual precipitation series are selected as candidate predictive factors.

5. The method for reconstructing annual precipitation based on nested principal component LASSO regression according to claim 1, characterized in that, In step S2, nested time structures are constructed from the most recent to the oldest according to the starting year of each candidate predictor. The multiple nested time periods have the same ending year. As the nested time periods extend to historical periods, the number of available tree-ring predictors gradually decreases, forming multiple nested layer structures with different time spans. The nested layer is a data layer and a model layer corresponding to a nested time period.

6. The method for reconstructing annual precipitation based on nested principal component LASSO regression according to claim 5, characterized in that, In step S2, the starting year interval between adjacent nested time periods is not less than 20 years; in one embodiment, the starting year interval between adjacent nested layers constructed based on the starting year of the tree ring representative year sequence is not less than 30 years.

7. The method for reconstructing annual precipitation based on nested principal component LASSO regression according to claim 1, characterized in that, In step S3, within each nested layer, the tree-ring data is standardized using the mean and standard deviation of the tree-ring sequence over a periodic period. The standardization process is performed according to the following formula: ; in, This is the original tree-ring data. For standardized tree-ring data, For the periodic average of the rate, The periodic standard deviation is the rate.

8. The method for reconstructing annual precipitation based on nested principal component LASSO regression according to claim 1, characterized in that, In step S4, when performing stepwise optimal subset selection on the candidate tree-ring predictors in each nested layer, the sum of squared residuals is first used as an evaluation index to determine the optimal tree-ring combination under different dimensions. The sum of squared residuals is calculated according to the following formula: ; in, For the sum of squared residuals, For the first The observed annual precipitation for each sample year For the first Reconstructed annual precipitation for each sample year This represents the sample length.

9. The method for reconstructing annual precipitation based on nested principal component LASSO regression according to claim 8, characterized in that, In step S4, after determining the optimal tree-ring combination under different dimensions, the Akaike Information Criterion is used to evaluate the overall performance of the model. The Akaike Information Criterion is calculated according to the following formula: ; in, For the first The Akaike Information Criterion value corresponding to each candidate tree-ring combination. In model parameters and observed annual precipitation series The likelihood function under the given conditions, For the first The number of parameters to be estimated in the model corresponding to each candidate tree-ring combination; the optimal tree-ring combination is searched by progressively increasing the dimension. When the Akaike Information Criterion value no longer decreases after adding a new predictor, the variable addition process is stopped, and the optimal subset of tree-ring predictor factors for the current nested layer is determined.

10. The method for reconstructing annual precipitation based on nested principal component LASSO regression according to claim 1, characterized in that, In step S5, a standardized tree-ring matrix is ​​constructed for the optimal subset of tree-ring predictors in each nested layer: ; in, The standardized tree-ring matrix, These are the first to the second elements in the current nested level. The optimal tree-ring predictor This represents the optimal number of tree-ring predictors.

11. The method for reconstructing annual precipitation based on nested principal component LASSO regression according to claim 10, characterized in that, In step S5, the covariance matrix is ​​constructed based on the standardized tree ring matrix, and the eigenvalues ​​and eigenvectors are solved, specifically according to the following formula: ; ; in, Let covariance matrix be the variance matrix. The standardized tree ring matrix The transpose of the matrix, For sample length, For the first 1 eigenvector To and The corresponding number Each feature value.

12. The method for reconstructing annual precipitation based on nested principal component LASSO regression according to claim 11, characterized in that, In step S5, the principal component variables are obtained based on the feature vectors, specifically according to the following formula: ; in, For the first One principal component variable, The standardized tree-ring matrix, For the first One feature vector; retain the first one according to the cumulative variance contribution rate threshold or significance test results. The principal component variables are used as input variables for the PCR-LASSO regression model. The number of principal component variables to be retained.

13. The method for reconstructing annual precipitation based on nested principal component LASSO regression according to claim 1, characterized in that, In step S6, a PCR-LASSO regression model is established using the retained principal component variables as independent variables and the observed annual precipitation as the dependent variable. The PCR-LASSO regression model is expressed as follows: ; in, To observe annual precipitation, For the intercept term, For the first One principal component variable, For the first The regression coefficients corresponding to the principal component variables. To retain the number of principal component variables, This is random error.

14. The method for reconstructing annual precipitation based on nested principal component LASSO regression according to claim 13, characterized in that, In step S6, LASSO regularization is used to solve for the parameters of the PCR-LASSO regression model. The objective of parameter solving is expressed as follows: ; in, This is the regression coefficient estimation vector. Let be the vector of regression coefficients to be solved. For the first The observed annual precipitation for each sample year For the first The first sample year corresponding to the first The values ​​of the principal component variables, is the regularization parameter; the LASSO regularization is used to compress the coefficients of unimportant variables, causing some regression coefficients to shrink to zero.

15. The method for reconstructing annual precipitation based on nested principal component LASSO regression according to claim 14, characterized in that, In step S6, the regularization parameter is determined through cross-validation. The optimal value is found; in one implementation, the leave-one-out cross-validation method is used to calculate the cross-validation error under different regularization parameters, and the regularization parameter corresponding to the minimum cross-validation error is selected as the optimal regularization parameter.

16. The method for reconstructing annual precipitation based on nested principal component LASSO regression according to claim 1, characterized in that, In step S7, the variance recovery process specifically involves adjusting the standard deviation of the precipitation reconstruction results obtained from each nested layer, and recovering the variance lost during the regression process using the statistical characteristics of the observed annual precipitation series. The variance recovery process is performed according to the following formula: ; in, For the first Original reconstructed precipitation for each sample year For the first Adjusted reconstructed precipitation for each sample year To reconstruct the average precipitation, To reconstruct the standard deviation of precipitation, To observe the average precipitation, To observe the standard deviation of precipitation.

17. The method for reconstructing annual precipitation based on nested principal component LASSO regression according to claim 1, characterized in that, In step S8, the reconstruction results of each nested layer are extended forward layer by layer according to the order of the starting year from the most recent to the oldest. The long-term span nested layer is used to supplement the missing historical precipitation information before the short-term span nested layer, so as to obtain a continuous long-term scale historical annual precipitation reconstruction sequence.

18. The method for reconstructing annual precipitation based on nested principal component LASSO regression according to claim 1, characterized in that, In step S9, the observed annual precipitation data is divided into a calibration period and a validation period. The model parameters are calibrated using the calibration period data, and the model performance is evaluated using the validation period data. The evaluation uses model evaluation indicators, including the Pearson correlation coefficient. The coefficient of determination (VRSQ) and the efficiency coefficient (CE) were verified. When the model evaluation index reached the preset threshold, it was determined that the PCR-LASSO regression model had an effective precipitation reconstruction capability.

19. The method for reconstructing annual precipitation based on nested principal component LASSO regression according to claim 18, characterized in that, In step S9, Bootstrap resampling and / or Jackknife resampling methods are used to evaluate the stability of the model; when the statistical parameters obtained by resampling are consistent with the statistical parameters of the original PCR-LASSO regression model, the PCR-LASSO regression model is determined to be stable.

20. The method for reconstructing annual precipitation based on nested principal component LASSO regression according to claim 1, characterized in that, The output of the continuous historical annual precipitation reconstruction sequence includes historical years, reconstructed annual precipitation, starting years of each nested layer, number of tree-ring predictors involved in the reconstruction, number of principal component variables retained, regularization parameters, precipitation reconstruction results after variance recovery, and model evaluation indicators, and is stored or output in the form of electronic data files.