High-resolution long-sequence land water reserve data generation method and device

Abstract

The invention discloses a method and a device for generating high-resolution long-sequence land water reserve data, which are used for solving the technical problem of poorer accuracy of land water reserve data provided for flood and drought disaster identification and water resource management from a watershed to a global scale due to the existing earth surface land water storage capacity inversion technology. The method comprises the following steps: generating target high-resolution land water reserve prediction data according to an acquired hydro-meteorological data set and an acquired land water reserve abnormal data set; according to the hydro-meteorological data set and the target high-resolution land water reserve prediction data, determining a monthly land water storage residual component; determining first component data and second component data according to the hydrometeorological data set, the target high-resolution land water reserve prediction data and the monthly land water storage capacity residual component; and according to the first component data, the second component data and the land water reserve abnormal data set, generating target high-resolution long-sequence land water reserve data.

Description

Technical Field

[0001] The present invention relates to the technical field of data prediction and processing, and in particular to a method and device for generating high-resolution long-sequence terrestrial water storage data. Background Art

[0002] The Gravity Recovery and Climate Experiment (GRACE) satellite mission and its follow-on mission, GRACE-FO (GRACE Follow-On), were launched on March 17, 2002, by NASA and the German Research Centre for Geosciences. This mission is crucial for studying terrestrial water reserves.

[0003] The Earth's gravitational field is constantly changing, with short-term variations primarily influenced by factors such as terrestrial water storage, atmospheric water vapor, and ocean tides. The GRACE satellite mission accurately measures gravitational field variations across different regions of Earth, providing a unique perspective on terrestrial water storage. This is because, through sophisticated data processing techniques, GRACE satellite data can be used to infer changes in Earth's surface terrestrial water storage (TWS) by removing irrelevant signals such as atmospheric water vapor and ocean tides.

[0004] However, existing technologies for retrieving terrestrial water storage on the Earth's surface can only provide data with a relatively low resolution (approximately 300 km × 300 km) and terrestrial water storage data after 2002. Furthermore, due to battery management issues with the GRACE satellite and the subsequent GRACE-FO mission being launched 11 months after the GRACE decommissioning date, the monthly GRACE Terrestrial Water Anomaly (TWSA) data are temporally discontinuous. This results in poor accuracy of the terrestrial water storage data provided for flood and drought disaster identification and water resources management at scales from river basins to the global level. Summary of the Invention

[0005] The present invention provides a method and device for generating high-resolution long-sequence terrestrial water storage data, which is used to solve the technical problem that the existing technology for inverting the terrestrial water storage on the earth's surface leads to poor accuracy of the terrestrial water storage data provided for flood and drought disaster identification and water resources management at the basin to global scale.

[0006] A first aspect of the present invention provides a method for generating high-resolution long-sequence terrestrial water storage data, comprising:

[0007] Obtain hydrometeorological datasets and terrestrial water storage anomaly datasets;

[0008] Based on the bilinear interpolation method, a plurality of target high-resolution terrestrial water storage prediction data are generated according to the hydrological and meteorological dataset and the terrestrial water storage anomaly dataset;

[0009] Based on multiple preset exponential decay parameters, a time series decomposition algorithm is used to determine multiple monthly land water storage residual components according to multiple target high-resolution land water storage prediction data and multiple daily precipitation data in the hydrological and meteorological dataset;

[0010] Based on multiple preset lag times, using the time series decomposition algorithm, the hydrological and meteorological dataset, the multiple target high-resolution land water storage prediction data, and the multiple monthly land water storage residual components, to determine multiple first component data and multiple second component data;

[0011] A preset machine learning statistical algorithm is used to generate multiple target high-resolution long-sequence terrestrial water storage data based on multiple first component data, multiple second component data and the terrestrial water storage anomaly data set.

[0012] Optionally, the hydrometeorological dataset further includes a plurality of daily temperature data, a plurality of daily land evaporation data, a plurality of soil moisture data, and a plurality of surface runoff data; and the generating of a plurality of target high-resolution terrestrial water storage prediction data based on the hydrometeorological dataset and the terrestrial water storage anomaly dataset based on the bilinear interpolation method includes:

[0013] resampling the plurality of daily air temperature data, the plurality of daily land evaporation data, the plurality of daily precipitation data, the plurality of soil moisture data, and the plurality of surface runoff data based on the time scale of the terrestrial water storage anomaly dataset to determine a plurality of resampled daily air temperature data, a plurality of resampled daily land evaporation data, a plurality of resampled daily precipitation data, a plurality of resampled soil moisture data, and a plurality of resampled surface runoff data;

[0014] performing data integration on a plurality of the resampled daily air temperature data, a plurality of the resampled daily land evaporation data, a plurality of the resampled daily precipitation data, a plurality of the resampled soil moisture data, and a plurality of the resampled surface runoff data to determine a plurality of resampled high-resolution prediction factor data;

[0015] performing spatial aggregation on a plurality of terrestrial water storage anomaly data in the terrestrial water storage anomaly dataset to determine a plurality of low-resolution terrestrial water storage anomaly data;

[0016] Based on the spatial grid corresponding to each of the low-resolution terrestrial water storage anomaly data, spatially aggregating each of the resampled high-resolution prediction factor data to determine a plurality of low-resolution prediction factor data;

[0017] Using the plurality of low-resolution prediction factor data as independent variables and the plurality of low-resolution terrestrial water storage anomaly data as dependent variables, a random forest model is constructed;

[0018] Using the plurality of resampled high-resolution prediction factor data as input to the random forest model, and outputting a plurality of initial high-resolution terrestrial water storage prediction data;

[0019] performing a subtraction operation on each of the initial high-resolution land water storage prediction data and each of the low-resolution land water storage anomaly data, and outputting a plurality of prediction residuals;

[0020] A bilinear interpolation method is used to generate a plurality of target high-resolution terrestrial water storage prediction data based on the plurality of prediction residuals and the plurality of initial high-resolution terrestrial water storage prediction data.

[0021] Optionally, the method of determining a plurality of monthly land water storage residual components based on a plurality of preset exponential decay parameters and a time series decomposition algorithm according to a plurality of target high-resolution land water storage prediction data and a plurality of daily precipitation data in the hydrometeorological dataset comprises:

[0022] Screening a plurality of daily precipitation data in the hydrological and meteorological dataset to determine a plurality of daily precipitation data in specific time periods;

[0023] Based on a plurality of preset exponential decay parameters, constructing an initial exponential decay filter corresponding to each of the preset exponential decay parameters on a spatial grid corresponding to the daily precipitation data in each of the specific time periods;

[0024] Using the daily precipitation data of each specific time period as the input of each initial exponential decay filter on the spatial grid corresponding to the daily precipitation data of each specific time period, and outputting the first month land water storage data corresponding to each initial exponential decay filter on each spatial grid;

[0025] Screening a plurality of first-month land water storage data corresponding to each spatial grid to determine a plurality of monthly land water storage data corresponding to each spatial grid for a specific time period;

[0026] The initial exponential decay filter corresponding to the monthly land water storage data of each specific time period corresponding to each spatial grid is used as the intermediate exponential decay filter;

[0027] Determine, using a time series decomposition algorithm, subseasonal residual component data corresponding to each intermediate exponential decay filter corresponding to each spatial grid based on the plurality of target high-resolution land water storage prediction data and the monthly land water storage data for a plurality of specific time periods corresponding to each spatial grid;

[0028] Calculating, based on the subseasonal residual component data corresponding to each of the intermediate exponential decay filters corresponding to each of the spatial grids, a first correlation coefficient corresponding to each of the intermediate exponential decay filters corresponding to each of the spatial grids;

[0029] Selecting the first correlation coefficient corresponding to each spatial grid and within the preset first coefficient range as the corresponding first initial correlation coefficient, and taking the largest first initial correlation coefficient corresponding to each spatial grid as the first optimal correlation coefficient;

[0030] The intermediate exponential decay filter corresponding to the first optimal correlation coefficient corresponding to each spatial grid is used as the target exponential decay filter;

[0031] Using multiple daily precipitation data in the hydrological and meteorological dataset as input to the target exponential decay filter corresponding to each spatial grid, and outputting multiple second-month land water storage data;

[0032] The time series decomposition algorithm is used to generate a plurality of monthly land water storage residual components based on each of the second monthly land water storage data.

[0033] Optionally, the determining of the plurality of first component data and the plurality of second component data based on the hydrological and meteorological dataset, the plurality of target high-resolution land water storage prediction data, and the plurality of monthly land water storage residual components using the time series decomposition algorithm based on the plurality of preset lag times includes:

[0034] performing data integration on the plurality of daily temperature data, the plurality of daily land evaporation data, the plurality of daily precipitation data, the plurality of soil moisture data, and the plurality of surface runoff data to generate a plurality of high-resolution prediction factor data;

[0035] Screening the plurality of high-resolution prediction factor data to determine a plurality of first specific time period prediction factor data;

[0036] performing time lag processing on each of the first specific time period prediction factor data using a plurality of preset lag times on the spatial grid corresponding to each of the first specific time period prediction factor data to generate initial time lag high-resolution prediction factor data associated with each of the preset lag times;

[0037] Screening the plurality of initial time-lagged high-resolution prediction factor data to determine a plurality of second specific time period prediction factor data corresponding to each spatial grid;

[0038] Using the preset lag time associated with each second specific time period prediction factor data corresponding to each spatial grid as the intermediate lag time;

[0039] Determining, using the time series decomposition algorithm, first component data associated with each intermediate lag time corresponding to each spatial grid based on the plurality of target high-resolution terrestrial water storage prediction data and the second specific time period prediction factor data corresponding to each spatial grid;

[0040] Calculating, based on the first component data of the intermediate lag time associations corresponding to the spatial grids, the second correlation coefficients of the intermediate lag time associations corresponding to the spatial grids;

[0041] Selecting the second correlation coefficient associated with each spatial grid and within the preset second coefficient range as the corresponding second initial correlation coefficient, and taking the largest second initial correlation coefficient corresponding to each spatial grid as the second optimal correlation coefficient;

[0042] The middle lag time associated with the second best correlation coefficient corresponding to each spatial grid is used as the target lag time;

[0043] On each spatial grid, performing time lag processing on each of the high-resolution prediction factor data using a target lag time corresponding to each spatial grid to generate target time lag high-resolution prediction factor data corresponding to each spatial grid;

[0044] Determining a plurality of initial prediction factor data component data based on a plurality of target time-lagged high-resolution prediction factor data using the time series decomposition algorithm;

[0045] Data integration is performed on the multiple initial prediction factor data component data and the multiple monthly land water storage residual components to generate multiple second component data.

[0046] Optionally, the preset machine learning statistical algorithm includes an initial multiple linear regression model and an initial multilayer perceptron neural network; the method of using the preset machine learning statistical algorithm to generate a plurality of target high-resolution long-sequence terrestrial water storage data based on a plurality of the first component data, a plurality of the second component data, and the terrestrial water storage anomaly data set includes:

[0047] Based on multiple preset specific time periods, using multiple first component data to respectively perform model training on an initial multiple linear regression model and an initial multilayer perceptron neural network, and determining a target multiple linear regression model and a target multilayer perceptron neural network corresponding to each of the component data sets in the specific time period;

[0048] constructing a second component data set according to the plurality of second component data;

[0049] using the second component data set as input to each of the target multiple linear regression models and each of the target multilayer perceptron neural networks, respectively, and outputting a plurality of first land water storage prediction values ​​corresponding to each of the target multiple linear regression models and a second land water storage prediction value corresponding to each of the target multilayer perceptron neural networks;

[0050] performing mean processing on each of the first land water reserve prediction values ​​and each of the second land water reserve prediction values, respectively, and outputting a first land water reserve mean value and a second land water reserve mean value;

[0051] Calculating a first error value based on the first terrestrial water storage mean value and the terrestrial water storage anomaly data set;

[0052] calculating a second error value based on the second terrestrial water storage mean value and the terrestrial water storage anomaly data set;

[0053] comparing the first error value and the second error value;

[0054] Multiple terrestrial water storage prediction values ​​corresponding to the smallest error value are selected as multiple target high-resolution long-sequence terrestrial water storage data.

[0055] Optionally, based on multiple preset specific time periods, using multiple first component data to respectively perform model training on an initial multiple linear regression model and an initial multilayer perceptron neural network, and determining a target multiple linear regression model and a target multilayer perceptron neural network corresponding to each of the component data sets in the specific time period, includes:

[0056] constructing a first component data set according to the plurality of first component data;

[0057] Dividing the first component data set based on a plurality of preset specific time periods to generate specific time period component data sets corresponding to the preset specific time periods;

[0058] The initial multiple linear regression model and the initial multilayer perceptron neural network are trained respectively using the component data sets of each specific time period to determine the target multiple linear regression model and the target multilayer perceptron neural network corresponding to the component data sets of each specific time period.

[0059] A second aspect of the present invention provides a device for generating high-resolution long-sequence terrestrial water storage data, comprising:

[0060] Acquisition module, used to obtain hydrological and meteorological data sets and land water storage anomaly data sets;

[0061] A first generating module is configured to generate a plurality of target high-resolution terrestrial water storage prediction data based on the hydrological and meteorological dataset and the terrestrial water storage anomaly dataset based on a bilinear interpolation method;

[0062] A first determination module is configured to determine, based on a plurality of preset exponential decay parameters and a time series decomposition algorithm, a plurality of target high-resolution land water storage prediction data and a plurality of daily precipitation data in the hydrological and meteorological dataset, a plurality of monthly land water storage residual components;

[0063] a second determining module, configured to determine, based on a plurality of preset lag times, a plurality of first component data and a plurality of second component data according to the hydrological and meteorological dataset, the plurality of target high-resolution land water storage prediction data, and the plurality of monthly land water storage residual components by using the time series decomposition algorithm;

[0064] The second generation module is used to generate multiple target high-resolution long-sequence terrestrial water storage data based on multiple first component data, multiple second component data and the terrestrial water storage anomaly data set using a preset machine learning statistical algorithm.

[0065] The third aspect of the present invention provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of the method for generating high-resolution long-sequence terrestrial water storage data as described in any one of the above items.

[0066] A fourth aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed, implements the steps of the method for generating high-resolution long-sequence terrestrial water storage data as described in any one of the above items.

[0067] A fifth aspect of the present invention provides a computer program product, which includes a computer program stored on a non-transitory computer-readable storage medium, and the computer program includes program instructions, wherein when the program instructions are executed by a computer, the computer executes the steps of the high-resolution long-sequence terrestrial water storage data generation method as described in any one of the above items.

[0068] It can be seen from the above technical solutions that the present invention has the following advantages:

[0069] The above technical solution of the present invention provides a method for generating high-resolution long-sequence terrestrial water storage data. First, a hydrological and meteorological data set and a terrestrial water storage anomaly data set are obtained; then, based on the bilinear interpolation method, multiple target high-resolution terrestrial water storage prediction data are generated according to the hydrological and meteorological data set and the terrestrial water storage anomaly data set; based on multiple preset exponential decay parameters, a time series decomposition algorithm is used to determine multiple monthly terrestrial water storage residual components according to the multiple target high-resolution terrestrial water storage prediction data and multiple daily precipitation data in the hydrological and meteorological data set; based on multiple preset lag times, a time series decomposition algorithm is used to determine multiple monthly terrestrial water storage residual components according to the hydrological and meteorological data set, the multiple target high-resolution terrestrial water storage prediction data and the multiple monthly terrestrial water storage residual components. Differential components are used to determine multiple first component data and multiple second component data; finally, a preset machine learning statistical algorithm is used to generate multiple target high-resolution long-sequence terrestrial water storage data based on the multiple first component data, multiple second component data and terrestrial water storage anomaly data sets; based on the above scheme, a time series decomposition algorithm is used to process the hydrological and meteorological data sets and the terrestrial water storage anomaly data sets to obtain multiple first component data and multiple second component data, and then combined with the preset machine learning statistical algorithm, long-sequence, high-spatial-resolution terrestrial water storage data can be obtained based on the multiple first component data and multiple second component data, thereby providing more accurate terrestrial water storage data for water and drought disaster identification and water resources management from river basin to global scales. BRIEF DESCRIPTION OF THE DRAWINGS

[0070] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0071] Figure 1 A flowchart of the steps of a method for generating high-resolution long-sequence terrestrial water storage data provided in Example 1 of the present invention;

[0072] Figure 2 A schematic diagram of the process of generating high-resolution long-sequence terrestrial water storage data provided in the second embodiment of the present invention;

[0073] Figure 3 This is a structural block diagram of a device for generating high-resolution long-sequence terrestrial water storage data provided in Example 3 of the present invention. DETAILED DESCRIPTION

[0074] An embodiment of the present invention provides a method and device for generating high-resolution long-sequence terrestrial water storage data, which is used to solve the technical problem that the existing technology for inverting the terrestrial water storage on the earth's surface leads to poor accuracy of the terrestrial water storage data provided for flood and drought disaster identification and water resources management at the basin to global scale.

[0075] In order to make the purpose, features, and advantages of the present invention more obvious and easy to understand, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the embodiments described below are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0076] See also Figure 1 , Figure 1 This is a flowchart of the steps of a method for generating high-resolution long-sequence terrestrial water storage data provided in Example 1 of the present invention.

[0077] The present invention provides a method for generating high-resolution long-sequence terrestrial water storage data, comprising:

[0078] Step 101: Obtain a hydrological and meteorological dataset and a land water storage anomaly dataset.

[0079] It should be noted that the hydrometeorological dataset includes multiple daily precipitation data, multiple daily temperature data, multiple daily land evaporation data, multiple soil moisture data, and multiple surface runoff data; the terrestrial water storage anomaly dataset includes multiple terrestrial water storage anomaly data. Among them, the daily precipitation data are the daily precipitation data with a resolution of 0.25° over China (CHM-PRE, China Meteorological Administration-Precipitation) from 1961 to 2022 provided by the National Qinghai-Tibet Plateau Science Data Center; the daily temperature data are the daily gridded temperature data with a resolution of 0.25° over China (CN05.1) from 1961 to 2022 provided by the National Climate Center; the daily land evaporation data are the daily global land evaporation data with a resolution of 0.25° from Amsterdam from 2003 to 2022; and the soil moisture data include the daily soil moisture data with a resolution of 0.25° from Amsterdam from 2003 to 2022 and the Global Land Data Assimilation System (GLDAS). The surface runoff data are the 3-hourly surface runoff data with a resolution of 0.25° from 1961 to 2022 obtained from the Global Land Surface Assimilation System (GLDAS); the terrestrial water storage anomaly data are the terrestrial water storage anomaly (TWSA) data inverted from the Mascon (Mass Concentration) solution with a resolution of 0.5° from 2003 to 2022 downloaded from the Jet Propulsion Laboratory (JPL).

[0080] Step 102: Generate multiple target high-resolution land water storage prediction data based on the hydrological and meteorological dataset and the land water storage anomaly dataset based on the bilinear interpolation method.

[0081] It should be noted that daily time series (time scale) of daily precipitation data, daily temperature data, daily land evaporation data, soil moisture data, and surface runoff data were extracted and resampled into irregular time series corresponding to the GRACE-TWSA data (terrestrial water storage anomaly data) from 2003 to 2022, so that the time scales of daily precipitation data, daily temperature data, daily land evaporation data, soil moisture data, and surface runoff data were the same as the time scale of the terrestrial water storage anomaly data. All resampled data were then integrated into 0.25° resolution predictor data (resampled high-resolution predictor data) from 2003 to 2022. The GRACE-TWSA 0.5° data were spatially aggregated into 1° resolution GRACE-TWSA data (low-resolution terrestrial water storage anomaly data), and then the 0.25° resolution predictor data were spatially aggregated into 1° resolution predictor data aligned with the GRACE grid (low-resolution predictor data).

[0082] Specifically, step 102 may include the following sub-steps:

[0083] Step S21: resampling a plurality of daily temperature data, a plurality of daily land evaporation data, a plurality of daily precipitation data, a plurality of soil moisture data, and a plurality of surface runoff data based on the time scale of the land water storage anomaly dataset to determine a plurality of resampled daily temperature data, a plurality of resampled daily land evaporation data, a plurality of resampled daily precipitation data, a plurality of resampled soil moisture data, and a plurality of resampled surface runoff data;

[0084] It should be noted that according to the time scale of the terrestrial water storage anomaly dataset, each data in the hydrological and meteorological dataset was resampled to obtain multiple resampled daily temperature data, resampled daily land evaporation data, resampled daily precipitation data, resampled soil moisture data, and resampled surface runoff data with a resolution of 0.25° from 2003 to 2022.

[0085] Step S22: integrating the plurality of resampled daily temperature data, the plurality of resampled daily land evaporation data, the plurality of resampled daily precipitation data, the plurality of resampled soil moisture data, and the plurality of resampled surface runoff data to determine a plurality of resampled high-resolution prediction factor data;

[0086] It should be noted that the resampled data are integrated to obtain a 0.25° resolution resampled high-resolution prediction factor dataset, which includes multiple resampled high-resolution prediction factor data.

[0087] Step S23: spatially aggregate the multiple terrestrial water reserve anomaly data in the terrestrial water reserve anomaly data set to determine multiple low-resolution terrestrial water reserve anomaly data;

[0088] It should be noted that the terrestrial water storage anomaly data are spatially aggregated to obtain 1° resolution terrestrial water storage anomaly data, that is, low-resolution terrestrial water storage anomaly data.

[0089] Step S24: Based on the spatial grids corresponding to the low-resolution terrestrial water storage anomaly data, spatially aggregate the resampled high-resolution prediction factor data to determine a plurality of low-resolution prediction factor data;

[0090] It should be noted that the 0.25° resampled high-resolution predictor data were spatially aggregated into 1° resolution predictor data (low-resolution predictor data) that were grid-aligned with the low-resolution terrestrial water storage anomaly data.

[0091] Step S25: constructing a random forest model using the multiple low-resolution prediction factor data as independent variables and the multiple low-resolution terrestrial water storage anomaly data as dependent variables;

[0092] It should be noted that each data has a corresponding resolution spatial grid. The low-resolution predictor data is used to train a decision tree model on its corresponding 0.25° spatial grid (all decision tree models constitute a random forest model) to establish a regression-based relationship between the low-resolution terrestrial water storage anomaly data and the low-resolution predictor data on each 1° spatial grid to predict TWSA at 1° resolution.

[0093] Step S26: using the multiple resampled high-resolution prediction factor data as input to the random forest model, and outputting multiple initial high-resolution terrestrial water storage prediction data;

[0094] It should be noted that after the above steps, the trained random forest model is obtained, and the 0.25° resolution prediction factor data from 2003 to 2022 (resampling high-resolution prediction factor data) is input into the trained random forest model to output the 0.25° resolution TWSA prediction data from 2003 to 2022 (initial high-resolution terrestrial water storage prediction data).

[0095] Step S27: performing a subtraction operation on each initial high-resolution land water storage prediction data and each low-resolution land water storage anomaly data, and outputting a plurality of prediction residuals;

[0096] It should be noted that the prediction residuals on each 1° modeling grid were obtained by subtracting the 1° resolution GRACE-TWSA data (low-resolution terrestrial water storage anomaly data) from the 0.25° resolution TWSA prediction data (initial high-resolution terrestrial water storage prediction data).

[0097] Step S28: Generate multiple target high-resolution land water reserve prediction data based on multiple prediction residuals and multiple initial high-resolution land water reserve prediction data using a bilinear interpolation method.

[0098] It should be noted that these prediction residuals were interpolated to the target grid with reduced resolution using the bilinear interpolation method and then added to the TWSA predicted at 0.25° resolution to obtain high-resolution TWSA data (target high-resolution terrestrial water storage prediction data).

[0099] Step 103: Based on multiple preset exponential decay parameters, a time series decomposition algorithm is used to determine multiple monthly land water storage residual components according to multiple target high-resolution land water storage prediction data and multiple daily precipitation data in the hydrological and meteorological dataset.

[0100] The time series decomposition algorithm is a signal decomposition algorithm for irregular time sampling series, also known as the improved STL algorithm (Seasonal-Trend Decomposition Procedure Based on Loess). The improved STL algorithm can decompose irregular hydrological and meteorological time series data into long-term components (X long ), seasonal cycle (X seas ) and the remaining subseasonal residuals (X sub ) three components, and then use the Theil-Sen trend estimator to extract the long-term component X long The linear trend X lin , interannual variability X inter By X long Subtract X lin The parameters defined by the STL process are the number of inner loops (N inner ) is set to 2; the number of outer loops (N outer ) is set to 3; the period length of the seasonal component (P) is set to 365.25; the polynomial fitting order D for seasonal and long-term trend estimates sea and D long Set to 1, the maximum lag effect time of seasonal estimation (λ seas ) is set to 1.5P and the long-term trend and the estimated maximum lag effect time (λ long ) is set to 60 days. The Loess program uses sequence operations to solve.

[0101] It should be noted that the processing of the monthly residual component of land-like water storage can be divided into four steps: (1) selecting the τ value (an integer from 0 to 200), that is, presetting the exponential decay parameter, and constructing an exponential decay filter, inputting the daily precipitation data to obtain the land-like water storage data; (2) applying the improved STL algorithm to decompose the sub-seasonal residual component (X sub ); (3) Calculate the correlation coefficient between the two and record the τ value corresponding to the optimal correlation coefficient (closest to 1); (4) Use the exponential decay filter associated with the τ value corresponding to the optimal correlation coefficient (closest to 1) to obtain the monthly land water storage residual component based on the 0.25° resolution daily precipitation data from 1961 to 2022.

[0102] Specifically, step 103 may include the following sub-steps:

[0103] Step S31: screening multiple daily precipitation data in the hydrological and meteorological data set to determine multiple daily precipitation data in specific time periods;

[0104] It should be noted that among the multiple daily precipitation data with a resolution of 0.25° in China from 1961 to 2022, multiple daily precipitation data with a resolution of 0.25° from 2000 to 2022 (daily precipitation data for a specific time period) were selected.

[0105] Step S32: Based on a plurality of preset exponential decay parameters, constructing an initial exponential decay filter corresponding to each preset exponential decay parameter on a spatial grid corresponding to daily precipitation data in each specific time period;

[0106] It should be noted that the value range of the preset exponential decay parameter is an integer from 0 to 200, and multiple initial exponential decay filters corresponding to the preset exponential decay parameters are constructed on the 0.25° spatial grid corresponding to the daily precipitation data in each specific time period.

[0107] Step S33: using the daily precipitation data of each specific time period as input to each initial exponential decay filter on the spatial grid corresponding to the daily precipitation data of each specific time period, and outputting the first month land water storage data corresponding to each initial exponential decay filter on each spatial grid;

[0108] It should be noted that the daily precipitation data in each specific time period are input into multiple initial exponential decay filters constructed on the 0.25° spatial grid corresponding to the daily precipitation data in each specific time period, and the corresponding first-month land water storage data output by each initial exponential decay filter are output, thereby obtaining multiple first-month land water storage data with a resolution of 0.25° from 2000 to 2022.

[0109] It is worth mentioning that the monthly land water storage data can be described as an approximation of the water storage state by converting precipitation data into the water storage state using the ensemble exponential decay filtering method. That is, the processing process of the monthly land water storage data can be expressed as:

[0110] ;

[0111] in, is the water storage status of month j (monthly land water storage data); is the weight of the impact of previous precipitation on a given time period, which is calculated based on an exponential decay function; is the time corresponding to the synthesized monthly precipitation; is the time corresponding to daily precipitation; is the daily precipitation data of the i-th day; n is the length of the input daily precipitation data.

[0112] Furthermore, the calculation process of the impact weight of previous precipitation on a given time period can be expressed as:

[0113] ;

[0114] in, is the preset exponential decay parameter, which represents the free parameter that controls the exponential decay rate. τ is calibrated on all modeling grids by comparing with the GRACE measured data; is the starting time of month j; is the end time of month j. For months with incomplete GRACE sampling, 、 corresponding to the start and end times of the incomplete monthly GRACE sampling; is the time corresponding to the synthesized monthly precipitation; is the time corresponding to daily precipitation; is the impact weight of previous precipitation on a given time period.

[0115] Step S34: screening the plurality of first monthly land water storage data corresponding to each spatial grid to determine the plurality of monthly land water storage data corresponding to each spatial grid for a specific time period;

[0116] It should be noted that, among multiple first-month land water storage data with a resolution of 0.25° from 2000 to 2022, each first-month land water storage data with a resolution of 0.25° from 2003 to 2022 was selected as the corresponding monthly land water storage data for a specific time period, and multiple monthly land water storage data with a resolution of 0.25° for a specific time period from 2003 to 2022 were obtained.

[0117] Step S35: using the initial exponential decay filter corresponding to the monthly land water storage data of each specific time period corresponding to each spatial grid as an intermediate exponential decay filter;

[0118] Step S36: using a time series decomposition algorithm to determine sub-seasonal residual component data corresponding to each intermediate exponential decay filter corresponding to each spatial grid based on multiple target high-resolution land water storage prediction data and multiple monthly land water storage data for specific time periods corresponding to each spatial grid;

[0119] It should be noted that the improved STL algorithm is applied to decompose multiple sub-seasonal residual component data X corresponding to each 0.25° spatial grid on all 0.25° spatial grids. sub The subseasonal residual component data include the subseasonal residual components of the monthly land water storage data in a specific time period and the subseasonal residual components of the target high-resolution land water storage prediction data.

[0120] Step S37: Calculate the first correlation coefficient corresponding to each intermediate exponential decay filter corresponding to each spatial grid according to the sub-seasonal residual component data corresponding to each intermediate exponential decay filter corresponding to each spatial grid;

[0121] It should be noted that the correlation coefficient (first correlation coefficient) of the 2003-2022 data between the sub-seasonal residual components of the monthly terrestrial water storage data in a specific time period and the sub-seasonal residual components of the target high-resolution terrestrial water storage forecast data was calculated, and multiple first correlation coefficients corresponding to each 0.25° spatial grid were obtained.

[0122] Step S38: Select the first correlation coefficient corresponding to each spatial grid and within the preset first coefficient range as the corresponding first initial correlation coefficient, and use the largest first initial correlation coefficient corresponding to each spatial grid as the first optimal correlation coefficient;

[0123] It should be noted that the preset first coefficient range can be set in the range of 0-1 as needed, by selecting the first correlation coefficient in the preset first coefficient range corresponding to each 0.25° spatial grid as the corresponding first initial correlation coefficient, and then recording the maximum first initial correlation coefficient corresponding to each 0.25° spatial grid as the optimal correlation coefficient, that is, the first optimal correlation coefficient (closest to 1).

[0124] Step S39: using the intermediate exponential decay filter corresponding to the first optimal correlation coefficient corresponding to each spatial grid as the target exponential decay filter;

[0125] It should be noted that the preset exponential decay parameters associated with the first optimal correlation coefficient corresponding to each 0.25° spatial grid are recorded, and the intermediate exponential decay filter corresponding to the preset exponential decay parameters is used as the target exponential decay filter.

[0126] Step S310: using multiple daily precipitation data in the hydrological and meteorological dataset as input to the target exponential decay filter corresponding to each spatial grid, and outputting multiple second-month land water storage data;

[0127] It should be noted that on each 0.25° spatial grid, the corresponding daily precipitation data of China with a resolution of 0.25° from 1961 to 2022 is input into the target exponential attenuation filter associated with each 0.25° spatial grid, and multiple monthly land water storage data with a resolution of 0.25° from 1961 to 2022 (second monthly land water storage data) are obtained.

[0128] Step S311: using a time series decomposition algorithm to generate multiple monthly land water storage residual components based on the second monthly land water storage data.

[0129] It should be noted that the time series decomposition algorithm is used to decompose the obtained second-month land water storage data to obtain multiple residual components of monthly land water storage from 1961 to 2022 (monthly land water storage residual components).

[0130] Step 104 : Based on multiple preset lag times, a time series decomposition algorithm is used to determine multiple first component data and multiple second component data according to the hydrological and meteorological dataset, multiple target high-resolution land water storage prediction data, and multiple monthly land water storage residual components.

[0131] The plurality of first component data include the interannual variation components (X inter ) and the subseasonal residual component (X sub ), the interannual variation component of multiple target high-resolution terrestrial water storage prediction data (X inter ) and the subseasonal residual component (X sub ).

[0132] The plurality of second component data include a plurality of monthly value prediction factor inter-annual variation component data and a plurality of monthly value prediction factor sub-seasonal residual component data.

[0133] It should be noted that the processing of component data obtained based on the daily scale lag calibration method can be divided into four steps: (1) selecting the lag time m (0 to 120 days), that is, presetting the lag time, and performing lag processing on the high-resolution prediction factor data, that is, modifying the time label t corresponding to the high-resolution prediction factor data to t+m; (2) applying the improved STL algorithm to decompose the interannual variation components (X inter ) and the subseasonal residual component (X sub ), i.e. the first component data; (3) calculate the correlation coefficient of the two corresponding components, and record the lag time m corresponding to the optimal correlation coefficient (closest to 1); (4) obtain the second component data based on the lag time m corresponding to the optimal correlation coefficient (closest to 1).

[0134] Specifically, step 104 may include the following sub-steps:

[0135] Step S41: integrating a plurality of daily temperature data, a plurality of daily land evaporation data, a plurality of daily precipitation data, a plurality of soil moisture data, and a plurality of surface runoff data to generate a plurality of high-resolution prediction factor data;

[0136] It should be noted that the 0.25° resolution prediction factor dataset for 1961-2022 was obtained by integrating multiple 0.25° resolution daily temperature data, daily land evaporation data, daily precipitation data, soil moisture data, and surface runoff data from 1961 to 2022. This dataset includes multiple high-resolution prediction factor data.

[0137] Step S42: screening the plurality of high-resolution prediction factor data to determine a plurality of first specific time period prediction factor data;

[0138] It should be noted that, in the 0.25° resolution prediction factor data set from 1961 to 2022, multiple high-resolution prediction factor data with a resolution of 0.25° from 2000 to 2022 are selected as the corresponding first specific time period prediction factor data, thereby obtaining multiple first specific time period prediction factor data.

[0139] Step S43: performing time lag processing on the prediction factor data of each first specific time period using a plurality of preset lag times on the spatial grid corresponding to the prediction factor data of each first specific time period, and generating initial time lag high-resolution prediction factor data associated with each preset lag time;

[0140] It should be noted that the preset lag time ranges from 0 to 120 days. On the 0.25° spatial grid corresponding to the prediction factor data of each first specific time period, multiple preset lag times are used to perform lag processing on the corresponding first specific time period prediction factor data to obtain initial lag high-resolution prediction factor data associated with each preset lag time corresponding to each 0.25° spatial grid. The initial lag high-resolution prediction factor data is the lag prediction factor data with a resolution of 0.25° from 2000 to 2022.

[0141] Step S44: screening the multiple initial time-lagged high-resolution prediction factor data to determine multiple second specific time period prediction factor data corresponding to each spatial grid;

[0142] It should be noted that, among the multiple 0.25° resolution lagged prediction factor data from 2000 to 2022, the 0.25° resolution lagged prediction factor data from 2003 to 2022 are selected as the corresponding second specific time period prediction factor data.

[0143] Step S45: using the preset lag time associated with each second specific time period prediction factor data corresponding to each spatial grid as the intermediate lag time;

[0144] Step S46: using a time series decomposition algorithm to determine first component data associated with each intermediate lag time corresponding to each spatial grid based on the multiple target high-resolution terrestrial water storage prediction data and the prediction factor data for each second specific time period corresponding to each spatial grid;

[0145] It should be noted that the improved STL algorithm is used to decompose the interannual variation component of the prediction factor data for the second specific time period (X inter ) and the subseasonal residual component (X sub ), the interannual variation component of high-resolution TWSA data (target high-resolution terrestrial water storage prediction data) (X inter ) and the subseasonal residual component (X sub ).

[0146] Step S47: Calculate the second correlation coefficient of each intermediate lag time associated with each spatial grid based on the first component data associated with each intermediate lag time corresponding to each spatial grid;

[0147] It should be noted that the interannual variation component (X inter ) and the subseasonal residual component (X sub ), the interannual variation component of high-resolution TWSA data (target high-resolution terrestrial water storage prediction data) (X inter ) and the subseasonal residual component (Xsub ), calculate the correlation coefficient (second correlation coefficient) of the corresponding components of the two.

[0148] Step S48: Select the second correlation coefficient associated with each spatial grid and within the preset second coefficient range as the corresponding second initial correlation coefficient, and use the largest second initial correlation coefficient corresponding to each spatial grid as the second optimal correlation coefficient;

[0149] It should be noted that the preset second coefficient range can be set in the range of 0-1 as needed, by selecting the second correlation coefficient in the preset second coefficient range corresponding to each 0.25° spatial grid as the corresponding second initial correlation coefficient, and then recording the maximum second initial correlation coefficient corresponding to each 0.25° spatial grid as the optimal correlation coefficient, that is, the second optimal correlation coefficient (closest to 1).

[0150] Step S49: taking the intermediate lag time associated with the second optimal correlation coefficient corresponding to each spatial grid as the target lag time;

[0151] It should be noted that the intermediate lag time m corresponding to the second optimal correlation coefficient (closest to 1) corresponding to each 0.25° spatial grid is recorded as the target lag time.

[0152] Step S410: performing time lag processing on each high-resolution prediction factor data at each spatial grid using the target lag time corresponding to each spatial grid to generate target time lag high-resolution prediction factor data corresponding to each spatial grid;

[0153] It should be noted that, according to the target lag time of each 0.25° spatial grid, the high-resolution prediction factor data with a resolution of 0.25° from 1961 to 2022 corresponding to each 0.25° spatial grid are lagged to generate multiple high-resolution prediction factor data with a target lag time of 0.25° resolution from 1961 to 2022.

[0154] Step S411: using a time series decomposition algorithm to determine a plurality of initial prediction factor data component data based on a plurality of target time-lagged high-resolution prediction factor data;

[0155] The initial predictor data component data include the interannual variation component and the subseasonal residual component of the monthly predictor data.

[0156] Specifically, multiple target lagged high-resolution prediction factor data with a resolution of 0.25° from 1961 to 2022 are synthesized into continuous monthly prediction factor data, which are then decomposed using a time series decomposition algorithm to obtain the interannual variation components and subseasonal residual components of multiple monthly prediction factor data with a resolution of 0.25° from 1961 to 2022.

[0157] Step S412: integrating the multiple initial prediction factor data components and the multiple monthly land water storage residual components to generate multiple second component data.

[0158] It should be noted that the interannual variation components and subseasonal residual components of multiple monthly value prediction factor data from 1961 to 2022 and the monthly land water storage residual components from 1961 to 2022 are integrated to obtain the interannual variation component dataset of monthly value prediction factors from 1961 to 2022 and the subseasonal residual component dataset of monthly value prediction factors from 1961 to 2022. The interannual variation component dataset of monthly value prediction factors includes multiple interannual variation component data of monthly value prediction factors, and the subseasonal residual component dataset of monthly value prediction factors includes multiple subseasonal residual component data of monthly value prediction factors.

[0159] Step 105: Generate multiple target high-resolution long-sequence terrestrial water storage data based on the multiple first component data, the multiple second component data, and the terrestrial water storage anomaly data set using a preset machine learning statistical algorithm.

[0160] Pre-built machine learning statistical algorithms include the initial multiple linear regression model and the initial multilayer perceptron neural network.

[0161] It should be noted that, using multiple first component data, the initial multiple linear regression model (MLR) and the initial multilayer perceptron neural network (MLP) were trained respectively to obtain multiple high-resolution long-term reconstruction models of land water storage. The high-resolution long-term reconstruction model of land water storage includes a target multiple linear regression model and a target multilayer perceptron neural network. Then, using multiple high-resolution long-term reconstruction models of land water storage, based on multiple second component data and land water storage anomaly data sets, multiple target high-resolution long-sequence land water storage data were obtained.

[0162] Specifically, step 105 may include the following sub-steps:

[0163] Step S51: Based on multiple preset specific time periods, the initial multiple linear regression model and the initial multilayer perceptron neural network are trained using multiple first component data to determine the target multiple linear regression model and the target multilayer perceptron neural network corresponding to the component data sets in each specific time period;

[0164] Furthermore, step S51 may include the following sub-steps:

[0165] Step S511: construct a first component data set according to a plurality of first component data;

[0166] Step S512: dividing the first component data set based on a plurality of preset specific time periods to generate specific time period component data sets corresponding to each preset specific time period;

[0167] Step S513: Use the component data sets of each specific time period to perform model training on the initial multiple linear regression model and the initial multilayer perceptron neural network respectively, and determine the target multiple linear regression model and target multilayer perceptron neural network corresponding to the component data sets of each specific time period.

[0168] It should be noted that the interannual variation components and subseasonal residual components of multiple second-specific time period prediction factor data and the interannual variation components and subseasonal residual components of multiple target high-resolution terrestrial water storage prediction data are used to construct the 2003-2022 data set (first component data set), and multiple training schemes are set up, that is, the 2003-2022 data set is divided into multiple training subsets of different time periods (specific time period component data sets), and the specific time period component data sets include training (60%), validation (20%) and test (20%) sets. Based on the preset specific time period, the specific time period component data sets are trained. The periods of the training set and the validation set are changed, wherein the preset specific time period can be set in 2003-2022 as needed. For example, assuming that the number of training schemes is 3, the training set and validation set in the component data set of the specific time period of 2003-2009, the training set and validation set in the component data set of the specific time period of 2009-2016, and the training set and validation set in the component data set of the specific time period of 2016-2022 can be divided. The test period of the test set in the component data set of the specific time period is fixed from January 2003 to July 2006, and the training scheme only changes the training period and validation period.

[0169] Furthermore, the component data sets for each specific time period obtained based on the above steps were applied to the MLP and MLR models for model training, respectively. The relationship between TWSA and prediction factors was established on all 0.25° spatial grids, and the target multiple linear regression model and target multilayer perceptron neural network corresponding to the component data sets for each specific time period were obtained.

[0170] Step S52: construct a second component data set according to the plurality of second component data;

[0171] Step S53: using the second component data set as input to each target multiple linear regression model and each target multilayer perceptron neural network, and outputting a plurality of first land water reserve prediction values ​​corresponding to each target multiple linear regression model and a second land water reserve prediction value corresponding to each target multilayer perceptron neural network;

[0172] It should be noted that the interannual variation component dataset of the monthly value prediction factors from 1961 to 2022 and the subseasonal residual component dataset of the monthly value prediction factors from 1961 to 2022 were respectively input into each target multiple linear regression model and each target multilayer perceptron neural network to obtain all 0.25° resolution spatial grid TWSA prediction values ​​corresponding to each target multiple linear regression model and each target multilayer perceptron neural network, that is, multiple first terrestrial water storage prediction values ​​and multiple second terrestrial water storage prediction values.

[0173] Step S54: performing mean processing on each first land water reserve prediction value and each second land water reserve prediction value, and outputting the first land water reserve mean value and the second land water reserve mean value;

[0174] Step S55: calculating a first error value based on the first land water storage mean value and the land water storage abnormality data set;

[0175] Step S56: Calculate a second error value based on the second land water storage mean value and the land water storage abnormality data set;

[0176] Step S57: comparing the first error value and the second error value;

[0177] Step S58: Select multiple land water storage prediction values ​​corresponding to the minimum error value as multiple target high-resolution long-sequence land water storage data.

[0178] It should be noted that by comparing the first error value and the second error value, multiple land water storage prediction values ​​corresponding to the smallest error value between the first error value and the second error value are selected as multiple target high-resolution long-sequence land water storage data, thereby obtaining a time-continuous high-resolution, long-sequence land water storage data set.

[0179] For a comparison of technical performance, existing technologies can be used as a reference. GRACE-FO is a successor mission to the GRACE mission, which orbited Earth from 2002 to 2017, and their measurements are consistent. However, GRACE only provides low-resolution terrestrial water storage inversion data, approximately 300 km × 300 km in size, starting in 2002. Furthermore, due to battery management issues with the GRACE satellite and the fact that GRACE-FO was launched 11 months after the decommissioning of GRACE, the GRACE terrestrial water storage anomaly (TWSA) monthly data are not only discontinuous in time but also have insufficient sampling days in the months with data. Therefore, obtaining a high-resolution, long-sequence, and continuous terrestrial water storage dataset has become a pressing and challenging issue. While terrestrial water storage (TWS) data beyond the GRACE period and at higher resolutions can be obtained by establishing regression relationships with hydrological variables or hydrological model outputs, no technology currently exists to invert high-resolution, long-sequence terrestrial water storage data while addressing the temporal insufficiency and discontinuity of the original data.

[0180] To address the above issues, the present invention proposes a method for generating high-resolution long-sequence terrestrial water storage data. Through machine learning and statistical modeling methods, combined with time resampling of prediction factors and an improved STL time series decomposition algorithm, the present invention solves the problems of the short and discontinuous original GRACE data series and low spatial resolution, thereby providing more accurate, detailed and high-coverage data basis for flood and drought disaster identification and water resources management at the basin to global scale.

[0181] In an embodiment of the present invention, a method for generating high-resolution long-sequence terrestrial water storage data is provided. First, a hydrological and meteorological data set and a terrestrial water storage anomaly data set are obtained. Then, based on a bilinear interpolation method, a plurality of target high-resolution terrestrial water storage prediction data are generated according to the hydrological and meteorological data set and the terrestrial water storage anomaly data set. Based on a plurality of preset exponential decay parameters, a time series decomposition algorithm is used to determine a plurality of monthly terrestrial water storage residual components according to the plurality of target high-resolution terrestrial water storage prediction data and a plurality of daily precipitation data in the hydrological and meteorological data set. Based on a plurality of preset lag times, a time series decomposition algorithm is used to determine a plurality of monthly terrestrial water storage residual components according to the hydrological and meteorological data set, the plurality of target high-resolution terrestrial water storage prediction data and the plurality of monthly terrestrial water storage residual components. Residual components are used to determine multiple first component data and multiple second component data; finally, a preset machine learning statistical algorithm is used to generate multiple target high-resolution long-sequence terrestrial water storage data based on the multiple first component data, multiple second component data and terrestrial water storage anomaly data sets; based on the above scheme, a time series decomposition algorithm is used to process the hydrological and meteorological data sets and the terrestrial water storage anomaly data sets to obtain multiple first component data and multiple second component data, and then combined with the preset machine learning statistical algorithm, long-sequence, high-spatial-resolution terrestrial water storage data can be obtained based on the multiple first component data and multiple second component data, thereby providing more accurate terrestrial water storage data for water and drought disaster identification and water resources management from river basin to global scales.

[0182] For better explanation, refer to Figure 2 , which shows a schematic flow chart of a method for generating high-resolution, long-sequence terrestrial water storage data, provided in Example 2 of the present invention. It should be noted that this embodiment briefly describes only the general flow of the method for generating high-resolution, long-sequence terrestrial water storage data. The specific implementation of each step can be understood by referring to the relevant content in the previous embodiments and will not be elaborated here. It should be understood that the present invention is not limited to this.

[0183] S1. Collect and normalize data sets, specifically including: obtaining a hydrometeorological dataset and a terrestrial water storage anomaly dataset; resampling and integrating the hydrometeorological dataset based on the time series of the terrestrial water storage anomaly dataset to obtain a prediction factor dataset sampled at the same time as the terrestrial water storage anomaly dataset; and spatially aggregating the terrestrial water storage anomaly dataset and the prediction factor dataset to obtain a 1° resolution terrestrial water storage anomaly dataset and a 1° resolution prediction factor dataset.

[0184] S2. Build a machine learning-based GRACE downscaling model to generate high-resolution terrestrial water storage. Specifically, a random forest model is constructed using a 1°-resolution terrestrial water storage anomaly dataset and a 1°-resolution predictor dataset. This random forest model is used to establish a regression-based relationship between the TWSA and the predictor at each 1° grid cell to predict TWSA at 1° resolution. The high-resolution predictor data are fed into the trained random forest model (the machine learning-based GRACE downscaling model) to output high-resolution terrestrial water storage.

[0185] S3. Calibrate the lag parameter m or the exponential decay filter parameter τ on all spatial grids, and generate the interannual variation component data of the monthly prediction factor and the subseasonal residual component data of the monthly prediction factor based on the optimal parameters. Specifically, based on the high-resolution land water storage and prediction factor dataset, use the signal decomposition algorithm suitable for irregular time sampling series to decompose the high-resolution land water storage and long-series prediction factor dataset into residual component, trend component and seasonal component; based on the high-resolution land water storage residual component and the residual component of the precipitation data in the long-series prediction factor dataset, calibrate the collective exponential decay function parameter on each spatial grid. The precipitation is converted into monthly land water storage using the calibrated ensemble exponential decay function, and the improved STL algorithm is used to decompose the monthly land water storage to obtain the residual components of the monthly land water storage; based on the residual components and trend components (interannual variation components) of high-resolution land water storage and the residual components and trend components of high-resolution prediction factor data, the daily-scale lag calibration method is used to optimize the lag time of the residual components and trend components of other prediction factors on each spatial grid, and then combined with the monthly land water storage residual components to generate the processed monthly prediction factor interannual variation component data and monthly prediction factor sub-seasonal residual component data.

[0186] S4. Use multiple linear regression (MLR) and multi-layer perceptron neural network (MLP) to train high-resolution long-term reconstruction models of terrestrial water storage, including: inter and X sub The dataset is divided into training, validation and test sets, and three training schemes are set up. The three schemes are then applied to the MLP and MLR models respectively to establish the relationship between TWSA and the predictor. The trained MLP and MLR models are used as the corresponding high-resolution long-term reconstruction models of terrestrial water storage.

[0187] S5. Generate a time-continuous high-resolution, long-sequence terrestrial water storage dataset based on a high-resolution long-term reconstruction model of terrestrial water storage. Specifically, the data on the interannual variation component of the monthly prediction factor and the sub-seasonal residual component of the monthly prediction factor are input into multiple high-resolution long-term reconstruction models of terrestrial water storage, and the output results are averaged and the error value is evaluated to finally obtain a time-continuous high-resolution, long-sequence terrestrial water storage dataset.

[0188] Among them, the present invention processes hydrological and meteorological data sets and terrestrial water storage anomaly data sets through machine learning and statistical modeling methods, combined with time resampling of prediction factors and an improved STL time series decomposition algorithm, to obtain long-sequence, high-spatial-resolution terrestrial water storage data, thereby providing more accurate terrestrial water storage data for water and drought disaster identification and water resources management from river basin to global scales.

[0189] See also Figure 3 , Figure 3 This is a structural block diagram of a device for generating high-resolution long-sequence terrestrial water storage data provided in Example 3 of the present invention.

[0190] The present invention provides a device for generating high-resolution long-sequence terrestrial water storage data, comprising:

[0191] An acquisition module 301 is used to acquire a hydrological and meteorological dataset and a land water storage anomaly dataset;

[0192] A first generating module 302 is configured to generate a plurality of target high-resolution land water storage prediction data based on a hydrological and meteorological dataset and a land water storage anomaly dataset using a bilinear interpolation method;

[0193] The first determination module 303 is configured to determine, based on multiple preset exponential decay parameters and a time series decomposition algorithm, multiple target high-resolution land water storage prediction data and multiple daily precipitation data in the hydrometeorological dataset, multiple monthly land water storage residual components;

[0194] The second determining module 304 is configured to determine a plurality of first component data and a plurality of second component data based on a hydrological and meteorological dataset, a plurality of target high-resolution land water storage prediction data, and a plurality of monthly land water storage residual components using a time series decomposition algorithm based on a plurality of preset lag times;

[0195] The second generation module 305 is used to generate multiple target high-resolution long-sequence terrestrial water storage data based on multiple first component data, multiple second component data and terrestrial water storage anomaly data sets using a preset machine learning statistical algorithm.

[0196] Furthermore, the hydrological and meteorological data set also includes a plurality of daily temperature data, a plurality of daily land evaporation data, a plurality of soil moisture data, and a plurality of surface runoff data; the first generating module 302 is specifically configured to:

[0197] Based on the time scale of the terrestrial water storage anomaly dataset, multiple daily temperature data, multiple daily land evaporation data, multiple daily precipitation data, multiple soil moisture data and multiple surface runoff data are resampled to determine multiple resampled daily temperature data, multiple resampled daily land evaporation data, multiple resampled daily precipitation data, multiple resampled soil moisture data and multiple resampled surface runoff data;

[0198] Data integration is performed on multiple resampled daily temperature data, multiple resampled daily land evaporation data, multiple resampled daily precipitation data, multiple resampled soil moisture data, and multiple resampled surface runoff data to determine multiple resampled high-resolution prediction factor data;

[0199] Perform spatial aggregation on multiple terrestrial water storage anomaly data in the terrestrial water storage anomaly dataset to determine multiple low-resolution terrestrial water storage anomaly data;

[0200] Based on the spatial grids corresponding to the low-resolution terrestrial water storage anomaly data, the resampled high-resolution prediction factor data are spatially aggregated to determine multiple low-resolution prediction factor data;

[0201] A random forest model was constructed using multiple low-resolution predictor data as independent variables and multiple low-resolution terrestrial water storage anomaly data as dependent variables.

[0202] Multiple resampled high-resolution predictor data are used as input to the random forest model, and multiple initial high-resolution terrestrial water storage prediction data are output;

[0203] Performing a subtraction operation on each initial high-resolution land water storage prediction data and each low-resolution land water storage anomaly data, and outputting a plurality of prediction residuals;

[0204] The bilinear interpolation method is used to generate multiple target high-resolution terrestrial water storage prediction data based on multiple prediction residuals and multiple initial high-resolution terrestrial water storage prediction data.

[0205] Furthermore, the first determining module 303 is specifically configured to:

[0206] Filter multiple daily precipitation data in the hydrological and meteorological dataset to determine daily precipitation data in multiple specific time periods;

[0207] Based on multiple preset exponential decay parameters, an initial exponential decay filter corresponding to each preset exponential decay parameter is constructed on the spatial grid corresponding to the daily precipitation data in each specific time period;

[0208] The daily precipitation data of each specific time period is used as the input of each initial exponential decay filter on the spatial grid corresponding to the daily precipitation data of each specific time period, and the first month land water storage data corresponding to each initial exponential decay filter on each spatial grid is output;

[0209] Screening a plurality of first-month land water storage data corresponding to each spatial grid to determine a plurality of monthly land water storage data corresponding to each spatial grid for a specific time period;

[0210] The initial exponential decay filter corresponding to the monthly land water storage data of each specific time period corresponding to each spatial grid is used as the intermediate exponential decay filter;

[0211] A time series decomposition algorithm is used to determine the subseasonal residual component data corresponding to each intermediate exponential decay filter for each spatial grid based on multiple target high-resolution terrestrial water storage prediction data and multiple monthly terrestrial water storage data for specific time periods corresponding to each spatial grid.

[0212] Calculating first correlation coefficients corresponding to the intermediate exponential decay filters corresponding to the spatial grids based on subseasonal residual component data corresponding to the intermediate exponential decay filters corresponding to the spatial grids;

[0213] Selecting the first correlation coefficient corresponding to each spatial grid and within the preset first coefficient range as the corresponding first initial correlation coefficient, and taking the largest first initial correlation coefficient corresponding to each spatial grid as the first optimal correlation coefficient;

[0214] The intermediate exponential decay filter corresponding to the first optimal correlation coefficient corresponding to each spatial grid is used as the target exponential decay filter;

[0215] Multiple daily precipitation data in the hydrological and meteorological dataset are used as input to the target exponential decay filter corresponding to each spatial grid, and multiple second-month land water storage data are output;

[0216] The time series decomposition algorithm is used to generate multiple monthly land water storage residual components based on the second-month land water storage data.

[0217] Furthermore, the second determining module 304 is specifically configured to:

[0218] Integrate multiple daily temperature data, multiple daily land evaporation data, multiple daily precipitation data, multiple soil moisture data, and multiple surface runoff data to generate multiple high-resolution prediction factor data;

[0219] Screening the plurality of high-resolution predictor data to determine a plurality of predictor data for a first specific time period;

[0220] performing time lag processing on the prediction factor data of each first specific time period using a plurality of preset lag times on the spatial grid corresponding to the prediction factor data of each first specific time period to generate initial time lag high-resolution prediction factor data associated with each preset lag time;

[0221] Screening a plurality of initial time-lagged high-resolution prediction factor data to determine a plurality of second specific time period prediction factor data corresponding to each spatial grid;

[0222] using the preset lag time associated with each second specific time period prediction factor data corresponding to each spatial grid as the intermediate lag time;

[0223] Determine, using a time series decomposition algorithm, first component data associated with each intermediate lag time corresponding to each spatial grid based on multiple target high-resolution terrestrial water storage prediction data and prediction factor data for each second specific time period corresponding to each spatial grid;

[0224] Calculating, based on the first component data associated with each intermediate lag time corresponding to each spatial grid, a second correlation coefficient associated with each intermediate lag time corresponding to each spatial grid;

[0225] Selecting the second correlation coefficient associated with each spatial grid and within the preset second coefficient range as the corresponding second initial correlation coefficient, and taking the largest second initial correlation coefficient corresponding to each spatial grid as the second optimal correlation coefficient;

[0226] The middle lag time associated with the second best correlation coefficient corresponding to each spatial grid is used as the target lag time;

[0227] On each spatial grid, each high-resolution prediction factor data is subjected to time lag processing using the target lag time corresponding to each spatial grid, thereby generating target time lag high-resolution prediction factor data corresponding to each spatial grid;

[0228] A time series decomposition algorithm is used to determine multiple initial predictor data component data based on multiple target lagged high-resolution predictor data;

[0229] Data integration is performed on multiple initial prediction factor data component data and multiple monthly land water storage residual components to generate multiple second component data.

[0230] Furthermore, the preset machine learning statistical algorithm includes an initial multiple linear regression model and an initial multilayer perceptron neural network; the second generation module 305 includes:

[0231] The first submodule is used to perform model training on the initial multiple linear regression model and the initial multilayer perceptron neural network using multiple first component data based on multiple preset specific time periods, and determine the target multiple linear regression model and target multilayer perceptron neural network corresponding to the component data sets in each specific time period;

[0232] A second submodule is configured to construct a second component data set based on the plurality of second component data;

[0233] The third submodule is used to use the second component data set as the input of each target multiple linear regression model and each target multilayer perceptron neural network, and output a plurality of first land water storage prediction values ​​corresponding to each target multiple linear regression model and a second land water storage prediction value corresponding to each target multilayer perceptron neural network;

[0234] a fourth submodule, configured to perform mean processing on each first land water reserve prediction value and each second land water reserve prediction value, and output the first land water reserve mean value and the second land water reserve mean value;

[0235] a fifth submodule, configured to calculate a first error value based on the first terrestrial water storage mean value and the terrestrial water storage anomaly data set;

[0236] a sixth submodule, configured to calculate a second error value based on the second terrestrial water storage mean and the terrestrial water storage anomaly data set;

[0237] A seventh submodule, configured to compare the first error value and the second error value;

[0238] The eighth submodule is used to select multiple land water storage prediction values ​​corresponding to the minimum error value as multiple target high-resolution long-sequence land water storage data.

[0239] Furthermore, the first submodule is specifically configured to:

[0240] constructing a first component data set according to the plurality of first component data;

[0241] Dividing the first component data set based on a plurality of preset specific time periods to generate specific time period component data sets corresponding to each preset specific time period;

[0242] The component data sets of each specific time period are used to train the initial multiple linear regression model and the initial multilayer perceptron neural network respectively, and the target multiple linear regression model and the target multilayer perceptron neural network corresponding to the component data sets of each specific time period are determined.

[0243] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described devices, modules and sub-modules can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0244] An embodiment of the present invention also provides a computer device, including a memory and a processor, wherein a computer program is stored in the memory; when the computer program is executed by the processor, the processor executes the steps of the method for generating high-resolution long-sequence terrestrial water storage data as described in any of the above embodiments.

[0245] An embodiment of the present invention also provides a computer-readable storage medium having a computer program / instruction stored thereon. When the computer program / instruction is executed by a processor, the steps of the method for generating high-resolution long-sequence terrestrial water storage data as described in any of the above embodiments are implemented.

[0246] An embodiment of the present invention also provides a computer program product, including a computer program / instruction, which, when executed by a processor, implements the steps of the method for generating high-resolution long-sequence terrestrial water storage data as described in any of the above embodiments.

[0247] In the several embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are merely schematic. For example, the division of units is only a logical function division. In actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0248] The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of these units may be selected to achieve the purpose of this embodiment according to actual needs.

[0249] As described above, 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 above embodiments, those skilled in the art should understand that the technical solutions described in the above embodiments can still be modified, or some of the technical features thereof can be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for generating high-resolution long-sequence terrestrial water storage data, characterized in that: include: Obtain hydrometeorological data sets and terrestrial water storage anomaly data sets; Based on the bilinear interpolation method, a plurality of target high-resolution land water storage prediction data are generated according to the hydrological and meteorological data set and the land water storage anomaly data set; Based on multiple preset exponential decay parameters, a time series decomposition algorithm is used to determine multiple monthly land water storage residual components according to multiple target high-resolution land water storage prediction data and multiple daily precipitation data in the hydrological and meteorological data set; Based on multiple preset lag times, using the time series decomposition algorithm according to the hydrological and meteorological data set, multiple target high-resolution land water storage prediction data and multiple monthly land water storage residual components, multiple first component data and multiple second component data are determined; A preset machine learning statistical algorithm is used to generate multiple target high-resolution long-sequence terrestrial water reserve data based on multiple first component data, multiple second component data and the terrestrial water reserve anomaly data set.

2. The method for generating high-resolution long-sequence terrestrial water storage data according to claim 1, characterized in that: The hydrological and meteorological data set also includes a plurality of daily temperature data, a plurality of daily land evaporation data, a plurality of soil moisture data and a plurality of surface runoff data; the bilinear interpolation method is based on the hydrological and meteorological data set and the land water storage anomaly data set to generate a plurality of target high-resolution land water storage prediction data, including: Based on the time scale of the land water storage anomaly data set, resampling the plurality of daily air temperature data, the plurality of daily land evaporation data, the plurality of daily precipitation data, the plurality of soil moisture data and the plurality of surface runoff data, to determine a plurality of resampled daily air temperature data, a plurality of resampled daily land evaporation data, a plurality of resampled daily precipitation data, a plurality of resampled soil moisture data and a plurality of resampled surface runoff data; Integrate the plurality of resampled daily temperature data, the plurality of resampled daily land evaporation data, the plurality of resampled daily precipitation data, the plurality of resampled soil moisture data and the plurality of resampled surface runoff data to determine a plurality of resampled high-resolution prediction factor data; Performing spatial aggregation on a plurality of terrestrial water reserve anomaly data in the terrestrial water reserve anomaly data set to determine a plurality of low-resolution terrestrial water reserve anomaly data; Based on the spatial grids corresponding to the low-resolution terrestrial water storage anomaly data, spatially aggregating the resampled high-resolution prediction factor data to determine a plurality of low-resolution prediction factor data; Using the plurality of low-resolution prediction factor data as independent variables and the plurality of low-resolution terrestrial water storage anomaly data as dependent variables, constructing a random forest model; Using the plurality of resampled high-resolution prediction factor data as inputs to the random forest model, and outputting a plurality of initial high-resolution terrestrial water storage prediction data; Performing a subtraction operation on each of the initial high-resolution land water storage prediction data and each of the low-resolution land water storage anomaly data, and outputting a plurality of prediction residuals; A bilinear interpolation method is used to generate a plurality of target high-resolution terrestrial water reserve prediction data based on the plurality of prediction residuals and the plurality of initial high-resolution terrestrial water reserve prediction data.

3. The method for generating high-resolution long-sequence terrestrial water storage data according to claim 1, characterized in that: The method uses a time series decomposition algorithm based on multiple preset exponential decay parameters to determine multiple monthly land water storage residual components according to multiple target high-resolution land water storage prediction data and multiple daily precipitation data in the hydrological and meteorological data set, including: Screening a plurality of daily precipitation data in the hydrological and meteorological data set to determine a plurality of daily precipitation data in specific time periods; Based on a plurality of preset exponential decay parameters, constructing an initial exponential decay filter corresponding to each of the preset exponential decay parameters on a spatial grid corresponding to the daily precipitation data in each of the specific time periods; Using the daily precipitation data of each specific time period as the input of each initial exponential decay filter on the spatial grid corresponding to the daily precipitation data of each specific time period, and outputting the first month land water storage data corresponding to each initial exponential decay filter on each spatial grid; Screening a plurality of first-month land water storage data corresponding to each spatial grid to determine a plurality of monthly land water storage data for a specific time period corresponding to each spatial grid; The initial exponential decay filter corresponding to the monthly land water storage data of each specific time period corresponding to each spatial grid is used as an intermediate exponential decay filter; Determine the sub-seasonal residual component data corresponding to each intermediate exponential decay filter corresponding to each spatial grid by using a time series decomposition algorithm based on a plurality of target high-resolution land water storage prediction data and a plurality of monthly land water storage data for specific time periods corresponding to each spatial grid; Calculate the first correlation coefficient corresponding to each of the intermediate exponential decay filters corresponding to each of the spatial grids according to the sub-seasonal residual component data corresponding to each of the intermediate exponential decay filters corresponding to each of the spatial grids; Selecting the first correlation coefficient corresponding to each spatial grid and within the preset first coefficient range as the corresponding first initial correlation coefficient, and taking the largest first initial correlation coefficient corresponding to each spatial grid as the first optimal correlation coefficient; The intermediate exponential decay filter corresponding to the first optimal correlation coefficient corresponding to each spatial grid is used as the target exponential decay filter; Using multiple daily precipitation data in the hydrological and meteorological data set as inputs of target exponential decay filters corresponding to each spatial grid, and outputting multiple second-month land water storage data; The time series decomposition algorithm is used to generate multiple monthly land water storage residual components based on each of the second monthly land water storage data.

4. The method for generating high-resolution long-sequence terrestrial water storage data according to claim 2, characterized in that: The method of determining a plurality of first component data and a plurality of second component data based on a plurality of preset lag times and using the time series decomposition algorithm according to the hydrological and meteorological data set, a plurality of the target high-resolution land water storage prediction data and a plurality of the monthly land water storage residual components comprises: Integrate the plurality of daily temperature data, the plurality of daily land evaporation data, the plurality of daily precipitation data, the plurality of soil moisture data and the plurality of surface runoff data to generate a plurality of high-resolution prediction factor data; Screening the plurality of high-resolution prediction factor data to determine a plurality of first specific time period prediction factor data; On the spatial grids corresponding to the prediction factor data of each first specific time period, using a plurality of preset lag times to perform lag processing on the prediction factor data of each first specific time period, and generating initial lag high-resolution prediction factor data associated with each preset lag time; Screening the plurality of initial time-lag high-resolution prediction factor data to determine a plurality of second specific time period prediction factor data corresponding to each spatial grid; Using the preset lag time associated with each of the second specific time period prediction factor data corresponding to each of the spatial grids as the intermediate lag time; Determine the first component data associated with each of the intermediate lag times corresponding to each of the spatial grids using the time series decomposition algorithm based on the plurality of target high-resolution terrestrial water storage prediction data and the prediction factor data of each of the second specific time periods corresponding to each of the spatial grids; Calculating the second correlation coefficient of each intermediate lag time associated with each spatial grid according to the first component data associated with each intermediate lag time corresponding to each spatial grid; Selecting the second correlation coefficient associated with each spatial grid and within the preset second coefficient range as the corresponding second initial correlation coefficient, and taking the maximum second initial correlation coefficient corresponding to each spatial grid as the second optimal correlation coefficient; The middle lag time associated with the second best correlation coefficient corresponding to each spatial grid is used as the target lag time; On each spatial grid, the target lag time corresponding to each spatial grid is used to perform lag processing on each of the high-resolution prediction factor data to generate the target lag high-resolution prediction factor data corresponding to each spatial grid; Determining a plurality of initial prediction factor data component data based on a plurality of target time-lagged high-resolution prediction factor data using the time series decomposition algorithm; Data integration is performed on the multiple initial prediction factor data component data and the multiple monthly land water storage residual components to generate multiple second component data.

5. The method for generating high-resolution long-sequence terrestrial water storage data according to claim 1, characterized in that: The preset machine learning statistical algorithm includes an initial multivariate linear regression model and an initial multilayer perceptron neural network; the preset machine learning statistical algorithm is used to generate multiple target high-resolution long-sequence terrestrial water storage data according to multiple first component data, multiple second component data and the terrestrial water storage abnormality data set, including: Based on multiple preset specific time periods, using multiple first component data to respectively perform model training on an initial multiple linear regression model and an initial multilayer perceptron neural network, and determining a target multiple linear regression model and a target multilayer perceptron neural network corresponding to each of the component data sets in the specific time period; constructing a second component data set according to the plurality of the second component data; Using the second component data set as input of each of the target multiple linear regression models and each of the target multi-layer perceptron neural networks, and outputting a plurality of first land water reserve prediction values ​​corresponding to each of the target multiple linear regression models and a second land water reserve prediction value corresponding to each of the target multi-layer perceptron neural networks; Respectively perform mean processing on each of the first land water reserve prediction values ​​and each of the second land water reserve prediction values, and output the first land water reserve mean value and the second land water reserve mean value; Calculating a first error value according to the first terrestrial water storage mean value and the terrestrial water storage abnormal data set; Calculating a second error value according to the second terrestrial water storage mean value and the terrestrial water storage anomaly data set; comparing the first error value and the second error value; Multiple terrestrial water storage prediction values ​​corresponding to the smallest error value are selected as multiple target high-resolution long-sequence terrestrial water storage data.

6. The method for generating high-resolution long-sequence terrestrial water storage data according to claim 1, characterized in that: The method of using the first component data to respectively perform model training on the initial multiple linear regression model and the initial multilayer perceptron neural network based on the plurality of preset specific time periods, and determining the target multiple linear regression model and the target multilayer perceptron neural network corresponding to the component data sets of each specific time period, comprises: constructing a first component data set according to the plurality of the first component data; Based on a plurality of preset specific time periods, the first component data set is divided to generate specific time period component data sets corresponding to each of the preset specific time periods; The initial multiple linear regression model and the initial multilayer perceptron neural network are trained respectively using the component data sets of each specific time period, and the target multiple linear regression model and the target multilayer perceptron neural network corresponding to the component data sets of each specific time period are determined.

7. A high-resolution long-sequence terrestrial water storage data generation device, characterized in that: include: An acquisition module is used to acquire hydrological and meteorological data sets and terrestrial water storage anomaly data sets; A first generating module is used to generate a plurality of target high-resolution terrestrial water reserve prediction data based on the hydrological and meteorological data set and the terrestrial water reserve anomaly data set based on a bilinear interpolation method; A first determination module is used to determine a plurality of monthly land water storage residual components based on a plurality of preset exponential decay parameters and a time series decomposition algorithm according to a plurality of target high-resolution land water storage prediction data and a plurality of daily precipitation data in the hydrological and meteorological data set; A second determination module is used to determine a plurality of first component data and a plurality of second component data based on a plurality of preset lag times and using the time series decomposition algorithm according to the hydrological and meteorological data set, a plurality of the target high-resolution land water storage prediction data and a plurality of the monthly land water storage residual components; The second generation module is used to generate multiple target high-resolution long-sequence terrestrial water reserve data based on multiple first component data, multiple second component data and the terrestrial water reserve anomaly data set by adopting a preset machine learning statistical algorithm.

8. A computer device, characterized in that: It comprises a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of the method for generating high-resolution long-sequence terrestrial water storage data as described in any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed, the method for generating high-resolution long-sequence terrestrial water storage data as described in any one of claims 1 to 6 is implemented.

10. A computer program product, characterized in that The computer program product includes a computer program stored on a non-transitory computer-readable storage medium, wherein the computer program includes program instructions, wherein when the program instructions are executed by a computer, the computer executes the high-resolution long-sequence terrestrial water storage data generation method as described in any one of claims 1-6.

Images