Tree-wheel-based climate and hydrological signal maximization mining method in cold and arid region

Abstract

The invention relates to the field of tree wheel climate-hydrology-ecology, and particularly discloses a tree wheel-based climate and hydrological signal maximization mining method in a cold and arid region, which comprises the following steps: S1, removing the growth trend of a tree by adopting a detrending method NEF-GA optimized by a genetic algorithm based on a deterministic function; s2, fusing linear and nonlinear functions, coupling single-index and multi-index comprehensive annual tables of multiple tree wheel indexes and climate and hydrological data in a measurement period, and retaining high, medium and low-frequency climate and hydrological signals; and S3, a linear function response model and a nonlinear function response model are adopted, a relation between the climate and hydrological elements and the annual table is established based on the detrended data, the correlation between the annual table and the key climate and hydrological elements is improved, and signal reconstruction and prediction are realized. According to the method, high-medium and low-frequency climate hydrological signals are effectively reserved, the association between the annual table and key climate elements is remarkably enhanced, and more reliable theoretical and method support is provided for high-precision inversion of historical climate hydrological data in cold and arid regions.

Description

Technical Field

[0001] The present invention relates to the technical field of dendroclimatology-hydrology-ecology, and in particular to a method for maximizing mining of climatic and hydrological signals based on tree rings in cold and arid regions. Background Art

[0002] As one of the most sensitive regions to global climate change, cold and arid regions, due to their unique geographical and climatic conditions, are particularly susceptible to global warming. In recent years, rising temperatures in these regions, coupled with the frequent occurrence of extreme climate events such as droughts, heavy rains, and heat waves, have exacerbated the uneven spatial and temporal distribution of water resources, resulting in a significant decrease in surface runoff, a drop in groundwater levels, and increasingly severe water shortages.

[0003] At present, using tree rings to invert long-term climate and hydrological data is an important means of global climate and hydrological research. However, there are problems in China's cold and arid regions such as insufficient systematic tree ring data and low accuracy in climate and hydrological signal mining.

[0004] Whether the tree's growth trend can be completely and accurately removed is the first crucial step in determining the extent of the hydroclimatic signal mining in tree rings. The traditional conservative detrending method is by far the most widely used growth trend removal method worldwide. Among them, the negative exponential function method has a more prominent advantage in cold arid and semi-arid regions. At the same time, as a deterministic function, its shortcomings should be avoided as much as possible.

[0005] Based on this, this study proposed a method for maximizing the mining of climate and hydrological signals based on tree rings in cold and arid areas, laying the foundation for maximizing the mining of hydrological and climate signals contained in tree rings. Summary of the Invention

[0006] In response to the above-mentioned problems of the prior art, the present invention provides a method for maximizing the mining of climate and hydrological signals based on tree rings in cold and arid areas. By removing the interference of the tree's own growth trend, it effectively retains the high, medium and low-frequency climate and hydrological signals, significantly enhances the correlation between the chronology and key climate elements, and provides more reliable theoretical and methodological support for the high-precision inversion of historical climate and hydrological data in cold and arid areas.

[0007] To achieve the above objectives, the present invention proposes a method for maximizing the mining of climate and hydrological signals based on tree rings in cold and arid regions, comprising:

[0008] S1. Using the detrending method NEF-GA based on deterministic function optimized by genetic algorithm to remove the growth trend of trees;

[0009] S2. Integrate linear and nonlinear functions and couple single- and multi-indicator comprehensive chronologies of various tree-ring indices with instrumental climate and hydrological data to preserve high, medium, and low-frequency climate and hydrological signals.

[0010] S3. Using linear and nonlinear function response models, based on detrended data, the relationship between climate-hydrological elements and chronology is established, the correlation between chronology and key climate-hydrological elements is enhanced, and signal reconstruction and prediction are achieved.

[0011] Preferably, in S1, the detrending method NEF-GA includes using the Fritts model and genetic algorithm to perform exponential function curve fitting of tree ring growth trends.

[0012] Preferably, the Fritts model is used to describe the relationship between annual ring width and tree age, and is corrected for the situation where tree growth tends to be balanced in the late stage. The calculation formula is:

[0013] y=ae -bx +K;

[0014] Where x is the age of the tree, y is the expected growth value corresponding to x, a and b are the coefficients of the original sequence solution, e is the base of the natural logarithm, and K is a constant.

[0015] Preferably, in the specific calculation, Fritts solves the values of a, b, and K. In the fitting process, an estimated value is first given, and then the solution is gradually approximated. The calculation process is:

[0016] S111. The slope between M1 and M2 is recorded as the average value of the first 10 years and the last 10 years of the original sequence. An initial estimate of b:

[0017]

[0018] Where n is the total number of years in the original sequence;

[0019] S112, 5×10 -5 As the initial Δ value, Add or subtract Δ from:

[0020]

[0021] S113, calculate three groups of a from the corresponding b1, b2 and b3 j and K j value, and calculate the sum of squares S j , where j = 1, 2, 3, the calculation formula is:

[0022]

[0023] Where i = 1, 2, ..., n;

[0024] Among them, S j When is the minimum, solving the simultaneous equations yields:

[0025]

[0026] During the calculation process, if S1 or S2 is the minimum, Δ is doubled to obtain new b1, b2, b3, and recalculate; when S2 is the minimum, Δ is halved and the calculation is repeated until 0≤5×10 -7 When , stop the calculation and get the last set of a, b and K, substitute them into the specified equation y = ae -bx +K, and obtain the growth trend curve of the original annual ring width sequence.

[0027] Preferably, a genetic algorithm is used to optimize the model parameters, the objective function is set as the RSME root mean square error of the residual, the model parameters are used as the optimization object, and the traditional algorithm is combined to constrain the optimization parameters to obtain the growth curve of the tree ring width sequence. The calculation formula is:

[0028]

[0029] Where a j 、b j and K j are the optimized parameters.

[0030] Preferably, in S3, the linear function response model includes a univariate linear regression model and a multivariate linear regression model; the univariate linear regression model uses the least squares method to determine the model parameters and establishes a simple functional relationship between the hydrological and climate elements and the chronology, and the functional relationship is:

[0031] y=ax+b;

[0032] In the formula, y is the dependent variable, representing the hydrological and meteorological elements, x is the independent variable, representing the tree-ring width chronology, a is the coefficient value in the equation, and b is the intercept. Both a and b are constants.

[0033] Preferably, the multiple linear regression model is based on the hysteresis of the impact of hydroclimatic factors on the radial growth of trees. The hydroclimatic data of the previous year or even the previous two years are introduced to construct a multiple regression model to comprehensively capture the climate and hydrological signals. The expression is:

[0034] Y t =aI t +bI t+1 +cI t+2 +d;

[0035] Where Y t is the hydroclimatic element value in year t, ℃, 1.0×10 8 m3;I t is the tree ring index in year t, dimensionless; I t+1is the tree ring index in year t+1, dimensionless; I t+2 is the tree ring index in year t+2, dimensionless; a, b, c, d are the coefficients in the equation, all of which are constants determined by the least squares method.

[0036] Preferably, in S3, the nonlinear function response model includes an MGF-OSR-PCA-BP / LSTM hydroclimatic regression model and a seasonal difference autoregressive moving average SARIMA model; the MGF-OSR-PCA-BP / LSTM hydroclimatic regression model adopts a combination of a generating function, principal component analysis PCA, a BP neural network or a long short-term memory network LSTM method to establish a high-precision hydroclimatic prediction model; the seasonal difference autoregressive moving average SARIMA model uses a seasonal difference method to remove seasonal factors and estimate seasonal parameters.

[0037] Preferably, the process of reconstructing hydrological and climate elements using the MGF-OSR-PCA-BP / LSTM hydroclimatic regression model using the mean generating function-optimal subset regression-principal component analysis-BP neural network MGF-OSR-PCA-BP prediction model includes:

[0038] S311. Calculate the mean of the time series x(t)={x(1), x(2), ..., x(n)} And define the generating function of the mean, whose expression is:

[0039]

[0040] In the formula, n is the total number of samples, or INT is the rounding function;

[0041] S312, based on the small randomness of the short-period average generating function, take Calculate the mean generating function of the x(t) sequence

[0042] S313. Fit the high-frequency part of the original sequence, perform first-order and second-order difference operations on the original sequence x(t), and obtain:

[0043] Δx(t)=x(t+1)-x(t), (t=1, 2,..., n-1);

[0044] Δ 2 x(t)=Δx(t+1)-Δx(t), (t=1, 2,..., n-1);

[0045] S314. Calculate the mean generating function of the first-order difference sequence and the mean generating function of the second-order difference sequence

[0046] S315. Perform periodic extension and extend the domain of the obtained mean generating function to the entire number axis, and construct the extension matrices of the mean generating function of the original sequence, first-order and second-order difference sequences, i.e., f l (0) (t), f l (1) (t), f l (2) (t), and the calculation formula for its periodic expansion is:

[0047]

[0048] In the formula, mod represents the congruence number;

[0049] S316, establish a cumulative extension sequence, fit the upward increasing and downward decreasing trends in the time series, and obtain 4m mean generating function extension sequences f l (0) (t), f l (1) (t), f l (2) (t), f l (3) (t), l = 1, 2, ..., m, as independent variables for screening, the calculation formula for the cumulative extension sequence is:

[0050]

[0051] In the formula, when t=1, f l (3) (t) = x(l);

[0052] S317, perform univariate regression on all extended sequences and original sequences, calculate CSC value, and make judgment, and set CSC>χ 2 The sequence of is roughly selected as the reconstruction factor, and it is assumed that p extension sequences are selected;

[0053] S318. Use PCA to reduce the dimension of the rough selection sequence, use the least data to reflect the most information, input the data after PCA dimension reduction into the BP neural network, establish a regression model, and reconstruct the hydrological and climate elements.

[0054] Preferably, the expression of the seasonal difference autoregressive moving average SARIMA model is:

[0055]

[0056] Among them, D (L S )=(1-L S ) D , Φp (L S )=1-Φ1L S -Φ2(L S ) 2 -…-Φ p (L S ) p ,Θ Q (L S )=1-Θ1L S -Θ2(L S ) 2 -…-Θ Q (L S ) Q ;

[0057] Where S is the cycle length, D is the seasonal split order, and L S represents the seasonal lag operator, and the model is abbreviated as SARIMA(p,d,q)×(P,D,Q).

[0058] Therefore, the present invention proposes a method for maximizing the mining of climate and hydrological signals based on tree rings in cold and arid regions, which has the following beneficial effects:

[0059] (1) This invention maximizes the retention of high, medium, and low-frequency climate and hydrological signals while removing the growth trends of trees themselves, significantly improving the correlation between the chronology and key climate and hydrological elements, and significantly improving the reliability and credibility of the inversion results, providing a basic theory and method for the high-precision inversion of historical climate and hydrological data in cold and arid areas.

[0060] (2) The NEF-GA detrending method first removes the growth trend of the trees themselves to provide a clean data source for the subsequent response model. The response model then establishes the relationship between climate and hydrological elements and the chronology based on the detrended data to achieve signal reconstruction and prediction.

[0061] (3) Linear function response models can capture simple linear relationships, while nonlinear function response models can handle complex nonlinear relationships. The two complement each other and jointly improve the accuracy and reliability of predictions. For climate and hydrological data with obvious seasonal changes, the SARIMA model can remove seasonal factors, improve the accuracy of the model, and more comprehensively capture the changing characteristics of climate and hydrological signals.

[0062] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] Figure 1 This is a diagram showing the program flow and processing results of the NEF-GA detrending method for maximizing the mining of climate and hydrological signals based on tree rings in cold and arid regions of the present invention;

[0064] Figure 2 This is a schematic diagram of the model architecture of a method for maximizing the mining of climate and hydrological signals based on tree rings in cold and arid areas according to the present invention. DETAILED DESCRIPTION

[0065] To make the technical solutions, advantages, and purposes of the present invention more clear, the technical solutions of the embodiments of the present invention will be clearly and completely described below. The described embodiments are part of the embodiments of the present invention, not all of them. Based on the described 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 this application.

[0066] Unless otherwise defined, technical or scientific terms used in the present invention shall have the same meaning as commonly understood by one of ordinary skill in the art to which the present invention belongs.

[0067] like Figure 1-Figure 2 As shown, a method for maximizing mining of climate and hydrological signals based on tree rings in cold and arid regions according to one embodiment of the present invention specifically includes the following steps:

[0068] S1. Using a deterministic function-based detrending method (NEF-GA) optimized by a genetic algorithm to remove the tree's growth trend. The NEF-GA detrending method includes fitting the exponential function curve of the tree ring growth trend using the Fritts model and genetic algorithm.

[0069] The Fritts model is used to describe the relationship between annual ring width and tree age, and is corrected for the situation where tree growth tends to be balanced in the late stage. The calculation formula is:

[0070] y=ae -bx +K;

[0071] Where x is the age of the tree, y is the expected growth value corresponding to x, a and b are the coefficients of the original sequence solution, e is the base of the natural logarithm, and K is a constant.

[0072] In the specific calculation, Fritts first gives the estimated value to solve the a, b, K values in the fitting process, and then gradually approximates the solution. The calculation process is:

[0073] S111. The slope between M1 and M2 is recorded as the average value of the first 10 years and the last 10 years of the original sequence. An initial estimate of b:

[0074]

[0075] Where n is the total number of years in the original sequence;

[0076] S112, 5×10-5 As the initial Δ value, Add or subtract Δ from:

[0077]

[0078] S113, calculate three groups of a from the corresponding b1, b2 and b3 j and K j value, and calculate the sum of squares S j , where j = 1, 2, 3, the calculation formula is:

[0079]

[0080] Where i = 1, 2, ..., n;

[0081] Among them, S j When is the minimum, solving the simultaneous equations yields:

[0082]

[0083] During the calculation process, if S1 or S2 is the minimum, Δ is doubled to obtain new b1, b2, b3, and recalculate; when S2 is the minimum, Δ is halved and the calculation is repeated until Δ≤5×10 -7 When , stop the calculation and get the last set of a, b and K, substitute them into the specified equation y = ae -bx +K, and obtain the growth trend curve of the original annual ring width sequence.

[0084] The genetic algorithm is used to optimize the model parameters, and the objective function is defined as the RSME root mean square error of the residual. The model parameters are used as the optimization object, and the traditional algorithm is combined to constrain the optimization parameters to obtain the growth curve of the tree ring width series. The calculation formula is:

[0085]

[0086] Where a j 、b j and K j are the optimized parameters.

[0087] S2. Integrate linear and nonlinear functions and couple single- and multi-indicator comprehensive chronologies of various tree-ring indices with instrumental climate and hydrological data to preserve high, medium, and low-frequency climate and hydrological signals.

[0088] S3. Using linear and nonlinear function response models, based on detrended data, the relationship between climate-hydrological elements and chronology is established, the correlation between chronology and key climate-hydrological elements is enhanced, and signal reconstruction and prediction are achieved.

[0089] The linear function response model includes a univariate linear regression model and a multivariate linear regression model. The univariate linear regression model uses the least squares method to determine the model parameters and establishes a simple functional relationship between the hydrological and climate elements and the chronology. The functional relationship is:

[0090] y=ax+b;

[0091] In the formula, y is the dependent variable, representing the hydrological and meteorological elements, x is the independent variable, representing the tree-ring width chronology, a is the coefficient value in the equation, and b is the intercept. Both a and b are constants.

[0092] The multiple linear regression model is based on the hysteresis of the impact of hydroclimatic factors on the radial growth of trees. The hydroclimatic data of the previous year or even the previous two years are introduced to construct a multiple regression model to comprehensively capture the climate and hydrological signals. Its expression is:

[0093] Y t =aI t +bI t+1 +cI t+2 +d;

[0094] Where Y t is the hydroclimatic element value in year t, ℃, 1.0×10 8 m3;I t is the tree ring index in year t, dimensionless; I t+1 is the tree ring index in year t+1, dimensionless; I t+2 is the tree ring index in year t+2, dimensionless; a, b, c, d are the coefficients in the equation, all of which are constants determined by the least squares method.

[0095] The nonlinear function response model includes an MGF-OSR-PCA-BP / LSTM hydroclimatic regression model and a seasonal difference autoregressive moving average SARIMA model; the MGF-OSR-PCA-BP / LSTM hydroclimatic regression model adopts a combination of a generating function, principal component analysis (PCA), a BP neural network, or a long short-term memory (LSTM) network method to establish a high-precision hydroclimatic prediction model; the seasonal difference autoregressive moving average SARIMA model uses a seasonal difference method to remove seasonal factors and estimate seasonal parameters.

[0096] The MGF-OSR-PCA-BP / LSTM hydroclimatic regression model uses the mean generation function-optimal subset regression-principal component analysis-BP neural network MGF-OSR-PCA-BP prediction model to reconstruct hydroclimatic elements. The process includes:

[0097] S311. Calculate the mean of the time series x(t)={x(1), x(2), ..., x(n)} And define the generating function of the mean, whose expression is:

[0098]

[0099] In the formula, n is the total number of samples, or INT is the rounding function;

[0100] S312, based on the characteristics of small randomness of short-period average generating function, take Calculate the mean generating function of the x(t) sequence

[0101] S313. Fit the high frequency part of the original sequence, perform first-order and second-order difference operations on the original sequence x(t), and obtain:

[0102] Δx(t)=x(t+1)-x(x), (t=1, 2,..., n-1);

[0103] Δ 2 x(t)=Δx(t+1)-Δx(t), (t=1, 2,..., n-1);

[0104] S314. Calculate the mean generating function of the first-order difference sequence and the mean generating function of the second-order difference sequence

[0105] S315. Perform periodic extension and extend the domain of the obtained mean generating function to the entire number axis, and construct the extension matrices of the mean generating function of the original sequence, first-order and second-order difference sequences, i.e., f l (0) (t), f l (1) (t), f l (2) (t), and the calculation formula for its periodic expansion is:

[0106]

[0107] In the formula, mod represents the congruence number;

[0108] S316, establish a cumulative extension sequence, fit the upward increasing and downward decreasing trends in the time series, and obtain 4m mean generating function extension sequences f l (0) (t), f l (1) (t), f l (2) (t), f l (3)(t), l=1,2,...,m, as independent variables for screening, the calculation formula of the cumulative extension sequence is:

[0109]

[0110] In the formula, when t=1, f l (3) (t) = x(l);

[0111] S317, perform univariate regression on all extended sequences and original sequences, calculate CSC value, and make judgment, and set CSC>χ 2 The sequence of is roughly selected as the reconstruction factor, and it is assumed that p extension sequences are selected;

[0112] S318. Use PCA to reduce the dimension of the rough selection sequence, use the least data to reflect the most information, input the data after PCA dimension reduction into the BP neural network, establish a regression model, and reconstruct the hydrological and climate elements.

[0113] The expression of the seasonal difference autoregressive moving average SARIMA model is:

[0114]

[0115] Among them, D (L S )=(1-L S ) D , Φ p (L S )=1-Φ1L S -Φ2(L S ) 2 -…-Φ p (L S ) p ,Θ Q (L S )=1-Θ1L S -Θ2(L S ) 2 -…-Θ Q (L S ) Q ;

[0116] Where S is the cycle length, D is the seasonal split order, and L s represents the seasonal lag operator, and the model is abbreviated as SARIMA(p,d,q)×(P,D,Q).

[0117] Therefore, the present invention provides a method for maximizing the mining of climate and hydrological signals based on tree rings in cold and arid areas. While removing the growth trends of trees themselves, it retains the high, medium and low-frequency climate and hydrological signals to the maximum extent, significantly improves the correlation between the chronology and key climate and hydrological elements, and significantly improves the reliability and credibility of the inversion results, providing a basic theory and method for high-precision inversion of historical climate and hydrological data in cold and arid areas.

[0118] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit the same. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that they can still modify or replace the technical solutions of the present invention with equivalents, and these modifications or equivalent replacements cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for maximizing the mining of climate and hydrological signals based on tree rings in cold and arid regions, characterized by: include: S1. Using the detrending method NEF-GA based on deterministic function optimized by genetic algorithm to remove the growth trend of trees; S2. Integrate linear and nonlinear functions and couple single- and multi-indicator comprehensive chronologies of various tree-ring indices with instrumental climate and hydrological data to preserve high, medium, and low-frequency climate and hydrological signals. S3. Using linear and nonlinear function response models, based on detrended data, the relationship between climate-hydrological elements and chronology is established, the correlation between chronology and key climate-hydrological elements is enhanced, and signal reconstruction and prediction are achieved.

2. The method for maximizing the mining of climate and hydrological signals based on tree rings in cold and arid regions according to claim 1 is characterized in that: In S1, the detrending method NEF-GA includes using the Fritts model and genetic algorithm to perform exponential function curve fitting of tree ring growth trends.

3. The method for maximizing the mining of climate and hydrological signals based on tree rings in cold and arid regions according to claim 2 is characterized in that: The Fritts model is used to describe the relationship between annual ring width and tree age, and is corrected for the situation where tree growth tends to be balanced in the late stage. The calculation formula is: y=ae -bx +K; Where x is the age of the tree, y is the expected growth value corresponding to x, a and b are the coefficients of the original sequence solution, e is the base of the natural logarithm, and K is a constant.

4. The method for maximizing the mining of climate and hydrological signals based on tree rings in cold and arid regions according to claim 2 is characterized in that: In the specific calculation, Fritts solves the values of a, b, and K. In the fitting process, the estimated values are first given and then gradually approximated. The calculation process is as follows: S111. The slope between M1 and M2 is recorded as the average value of the first 10 years and the last 10 years of the original sequence. An initial estimate of b: Where n is the total number of years in the original sequence; S112, 5×10 -5 As the initial Δ value, Add or subtract Δ from: S113, calculate three groups of a from the corresponding b1, b2 and b3 j and K j value, and calculate the sum of squares S j , where j = 1, 2, 3, the calculation formula is: Where i = 1, 2, ..., n; Among them, S j When is the minimum, solving the simultaneous equations yields: During the calculation process, if S1 or S2 is the minimum, Δ is doubled to obtain new b1, b2, b3, and recalculate; when S2 is the minimum, Δ is halved and the calculation is repeated until Δ≤5×10 -7 When , stop the calculation and get the last set of a, b and K, substitute them into the specified equation y = ae -bx +K, and obtain the growth trend curve of the original annual ring width sequence.

5. The method for maximizing the mining of climate and hydrological signals based on tree rings in cold and arid regions according to claim 2 is characterized in that: The genetic algorithm is used to optimize the model parameters, and the objective function is defined as the RSME root mean square error of the residual. The model parameters are used as the optimization object, and the traditional algorithm is combined to constrain the optimization parameters to obtain the growth curve of the tree ring width series. The calculation formula is: Where a j 、b j and K j are the optimized parameters.

6. The method for maximizing the mining of climate and hydrological signals based on tree rings in cold and arid regions according to claim 1, characterized in that: In S3, the linear function response model includes a univariate linear regression model and a multivariate linear regression model; the univariate linear regression model uses the least squares method to determine the model parameters and establishes a simple functional relationship between the hydrological and climate elements and the chronology, and the functional relationship is: y=ax+b; In the formula, y is the dependent variable, representing the hydrological and meteorological elements, x is the independent variable, representing the tree-ring width chronology, a is the coefficient value in the equation, and b is the intercept. Both a and b are constants.

7. The method for maximizing the mining of climate and hydrological signals based on tree rings in cold and arid regions according to claim 6, characterized in that: The multiple linear regression model is based on the hysteresis of the impact of hydroclimatic factors on the radial growth of trees. The hydroclimatic data of the previous year or even the previous two years are introduced to construct a multiple regression model to comprehensively capture the climate and hydrological signals. The expression is: Y t =aI t +bI t+1 +cI t+2 +d; Where Y t is the hydroclimatic element value in year t, ℃, 1.0×10 8 m3;I t is the tree ring index in year t, dimensionless; I t+1 is the tree ring index in year t+1, dimensionless; I t+2 is the tree ring index in year t+2, dimensionless; a, b, c, d are the coefficients in the equation, all of which are constants determined by the least squares method.

8. The method for maximizing the mining of climate and hydrological signals based on tree rings in cold and arid regions according to claim 1, characterized in that: In S3, the nonlinear function response model includes the MGF-OSR-PCA-BP / LSTM hydroclimatic regression model and the seasonal difference autoregressive moving average SARIMA model; the MGF-OSR-PCA-BP / LSTM hydroclimatic regression model adopts a combination of the mean generation function, principal component analysis PCA, BP neural network or long short-term memory network LSTM method to establish a high-precision hydroclimatic prediction model; the seasonal difference autoregressive moving average SARIMA model uses the seasonal difference method to remove seasonal factors and estimate seasonal parameters.

9. The method for maximizing the mining of climate and hydrological signals based on tree rings in cold and arid regions according to claim 8, characterized in that: The MGF-OSR-PCA-BP / LSTM hydroclimatic regression model uses the mean generation function-optimal subset regression-principal component analysis-BP neural network MGF-OSR-PCA-BP prediction model to reconstruct hydroclimatic elements. The process includes: S311. Calculate the mean of the time series x(t)={x(1), x(2), ..., x(n)} And define the generating function of the mean, whose expression is: In the formula, n is the total number of samples, or INT is the rounding function; S312, based on the characteristics of small randomness of short-period average generating function, take Calculate the mean generating function of the x(t) sequence S313. Fit the high frequency part of the original sequence, perform first-order and second-order difference operations on the original sequence x(t), and obtain: Δx(t)=x(t+1)-x(t), (t=1,2,...,n-1); Δ 2 x(t)=Δx(t+1)-Δx(t),(t=1,2,...,n-1); S314. Calculate the mean generating function of the first-order difference sequence and the mean generating function of the second-order difference sequence S315. Perform periodic extension and extend the domain of the obtained mean generating function to the entire number axis, and construct the extension matrices of the mean generating function of the original sequence, first-order and second-order difference sequences, i.e., f l (0) (t), f l (1) (t), f l (2) (t), and the calculation formula for its periodic expansion is: In the formula, mod represents the congruence number; S316, establish a cumulative extension sequence, fit the upward increasing and downward decreasing trends in the time series, and obtain 4m mean generating function extension sequences f l (0) (t), f l (1) (t), f l (2) (t), f l (3) (t), l = 1, 2, ..., m, as independent variables for screening, the calculation formula for the cumulative extension sequence is: In the formula, when t=1, f l (3) (t) = x(l); S317, perform univariate regression on all extended sequences and original sequences, calculate CSC value, and make judgment, and set CSC>χ 2 The sequence of is roughly selected as the reconstruction factor, and it is assumed that p extension sequences are selected; S318. Use PCA to reduce the dimension of the rough selection sequence, use the least data to reflect the most information, input the data after PCA dimension reduction into the BP neural network, establish a regression model, and reconstruct the hydrological and climate elements.

10. The method for maximizing the mining of climate and hydrological signals based on tree rings in cold and arid regions according to claim 8, characterized in that: The expression of the seasonal difference autoregressive moving average SARIMA model is: Among them, Ps D (L S )=(1-L S ) D ,F p (L S )=1-Φ1L S -Φ2(L S ) 2 -…-F p (L S ) p ,I Q (L S )=1-Θ1L S -Θ2(L S )2-…-Θ Q (L S ) Q ; Where S is the cycle length, D is the seasonal split order, and L S represents the seasonal lag operator, and the model is abbreviated as SARIMA(p,d,q)×(P,D,Q).

Images