An intra-diurnal precipitation simulation method, device, storage medium and program product based on a weather generator and precipitation decomposition

Abstract

This invention discloses a method, device, storage medium, and program product for intra-diurnal precipitation simulation based on a weather generator and precipitation decomposition. It establishes a complete precipitation simulation framework from diurnal to hourly scales, organically combining the weather generator and precipitation decomposition algorithm to overcome the shortcomings of traditional methods where each component is independent. An empirical rank rearrangement method is used for post-processing of the simulation sequence, effectively restoring the temporal correlation structure of precipitation and significantly improving the simulation accuracy of statistical characteristics such as the first-order autocorrelation coefficient. A joint stratified sampling strategy based on precipitation and temperature is employed for historical date matching, fully utilizing the covariant relationships between meteorological elements and improving the representativeness of hourly precipitation distribution patterns. An ensemble simulation framework is used, employing multiple realization and multiple sampling mechanisms to quantify simulation uncertainties, providing probabilistic forecasting capabilities for hydrological model applications and risk assessment.

Description

Technical Field

[0001] This invention relates to the field of hydro-meteorological simulation technology, and in particular to a method, equipment, storage medium, and program product for intra-diurnal precipitation simulation based on weather generators and precipitation decomposition. Background Technology

[0002] Precipitation is the most important input variable in the hydrological cycle. Due to observational limitations, most regions only have diurnal precipitation data, while intradiurnal (e.g., hourly) precipitation data is relatively scarce. Existing methods for obtaining intradiurnal precipitation mainly include: The first category is directly using measured precipitation data. This method has high data reliability, but is limited by the density of the observation network and the length of the observation history, making it difficult to meet the needs of long-sequence simulations. The second category is numerical model downscaling methods based on physical processes. This method outputs intradiurnal precipitation data through regional climate models or weather forecast models, but it involves large computational loads, complex parameterization schemes, and limited ability to simulate extreme precipitation events. The third category is stochastic simulation methods. This includes typical solution set models and precipitation simulation methods based on stochastic models. Although traditional typical solution set models and stochastic precipitation simulation models can simulate the statistical characteristics of intradiurnal precipitation, they are difficult to maintain the spatial correlation of the intradiurnal precipitation field simultaneously. Coupled with distributed hydrological models, they are prone to misestimation of runoff extrema.

[0003] In addition, existing methods generally have the following shortcomings: (1) the simulation of daily precipitation occurrence and precipitation amount are independent of each other and lack an effective integration mechanism; (2) the simulation sequence is difficult to reflect the spatial correlation of intra-diurnal precipitation field; (3) the intra-diurnal distribution information of historical precipitation is not fully utilized when daily precipitation is decomposed to intra-diurnal scale; (4) there is a lack of uncertainty quantification means, making it difficult to assess the reliability of simulation results.

[0004] Therefore, there is an urgent need to develop an intraday precipitation simulation method that can comprehensively utilize historical observation information, maintain the statistical characteristics of precipitation sequences at different time scales and the spatial correlation of precipitation fields, and provide uncertainty quantification. Summary of the Invention

[0005] Purpose of the invention: This invention provides a method, device, storage medium, and program product for intra-diurnal precipitation simulation based on weather generators and precipitation decomposition. It constructs a complete simulation framework from the generation of diurnal precipitation sequences to the decomposition of hourly precipitation, achieving high-precision reproduction of the statistical characteristics and spatial correlation structure of intra-diurnal precipitation processes. At the same time, it quantifies the simulation uncertainty through a multi-sampling ensemble simulation mechanism.

[0006] Technical solution: The intra-diurnal precipitation simulation method based on weather generators and precipitation decomposition described in this invention includes the following steps:

[0007] Step 1: Collect hourly precipitation data and daily average temperature data from meteorological stations within the study area, remove outliers and missing values, interpolate missing values ​​using adjacent stations to form continuous time series samples, calculate the regional daily average temperature, and form a training dataset.

[0008] Step 2: Based on the hourly precipitation data of each station, calculate the total daily precipitation to form a daily precipitation observation sequence; calculate the proportion of each hourly precipitation to the total daily precipitation to form an hourly precipitation proportion matrix for each station.

[0009] Step 3: Based on the Markov chain, calculate the dry-wet state transition probability matrix for each month, and simulate the daily precipitation sequence of each station by Monte Carlo random sampling.

[0010] Step 4: Based on the moving time window, the K-nearest neighbor resampling method (KNN) is used to randomly extract precipitation from the daily precipitation observation sequence to generate a simulated daily precipitation sequence for each station.

[0011] Step 5: Empirical rank rearrangement correction. The EC method is used to post-process the daily precipitation simulation series to restore the spatiotemporal correlation structure of the simulation series.

[0012] Step 6: Based on the joint stratified sampling method of regional average precipitation and regional average temperature, match historical observation dates for each precipitation day in the simulated sequence of regional average daily precipitation;

[0013] Step 7: Using the hourly precipitation ratio coefficients of each station corresponding to the matched historical observation dates, decompose the daily precipitation simulation sequence of each station into an hourly precipitation simulation sequence.

[0014] Furthermore, in step 2, for each precipitation day in the training dataset, the proportion of hourly precipitation to the total daily precipitation is calculated. :

[0015]

[0016] In the formula, The precipitation ratio coefficient for the h-th hour; The rainfall in the h-th hour is expressed in millimeters. This represents the total rainfall for the day, expressed in millimeters.

[0017] Furthermore, in step 3, based on the Markov chain, the dry-wet state transition probability matrix for each month is calculated, and the daily precipitation sequence for each station is simulated by Monte Carlo random sampling, specifically including the following steps:

[0018] Step 31: Define precipitation state variables When the daily rainfall is greater than or equal to a preset threshold (e.g., 0.1 mm) (Wet day), otherwise (Dry day);

[0019] Step 32: Count the number of state transitions between adjacent days by month and year, construct a transition count matrix, and count the frequency of the four types of transitions; normalize the counts in each row to obtain the transition probability matrix; the transition probability calculation formula is:

[0020]

[0021] In the formula, For the first Year Moon in state Transition to state The probability of; From state Transition to state The number of times; In a state The total number of days; when the sample size for a certain month or year is insufficient, i.e., the count of a certain row is zero, the corresponding row of the multi-year average transition probability matrix for that month is used as the substitute;

[0022] Step 33: For each simulated year, a year is randomly selected from historical years using the Monte Carlo method. The Markov chain is driven by the transition probability matrix of the corresponding month of that year. The state transitions of the Markov chain model follow the following probabilities:

[0023]

[0024] In the formula, Indicates the first Daily precipitation patterns; and These represent the precipitation status of the previous day and the current day, respectively, with values ​​of 0 (dry) or 1 (wet); From state Transition to state The conditional probability;

[0025] Step 34: Using the first day's dry / wet state in the corresponding segment of the daily precipitation observation sequence as the initial state, generate uniformly distributed random numbers for each day during the simulation period, compare them with the cumulative transition probability of the current state, determine the dry / wet state (0 or 1) for the next day, and recursively generate a complete precipitation occurrence sequence day by day.

[0026] Furthermore, in step 4, the random extraction of precipitation data from the daily precipitation observation sequence to generate the daily precipitation simulation sequence for each station specifically includes the following steps:

[0027] Step 41: For each wet day in the simulated daily precipitation sequence, determine its annual day number. ;

[0028] Step 42, with Centered on historical data, a time window is constructed. The window width is the preset number of days;

[0029] Step 43: Randomly select a precipitation value from all historical precipitation events within the time window as the simulated value;

[0030] Step 44: When the number of valid precipitation events within the window is insufficient, adaptively expand the window width until the minimum sample size requirement is met.

[0031] Furthermore, in step 5, the Empirical Copula (EC) method is used to post-process the daily precipitation simulation series to restore the spatiotemporal correlation structure of the simulation series. This specifically includes the following steps:

[0032] Step 51: Process the observed and simulated precipitation series separately by month to obtain the simulated series for each month. and observation sequence ;

[0033] Step 52: Calculate the rank and sort index of the observation sequence. , making Sort in ascending order;

[0034] Step 53: Sort the simulated sequences in ascending order to obtain... ;

[0035] Step 54: Reorder the sorted simulation sequences using the rank index of the observed sequences:

[0036]

[0037] In the formula, The corrected simulation sequence retains the marginal distribution characteristics of the simulation while having a spatiotemporal correlation structure similar to that of the observation.

[0038] Furthermore, in step 6, the joint stratified sampling method based on regional average precipitation and regional average temperature, specifically matching historical observation dates for each precipitation day in the simulated sequence of regional average daily precipitation, includes the following steps:

[0039] Step 61: For the daily precipitation observation sequence and the simulated daily precipitation sequence, use the arithmetic mean method to calculate the watershed average value, and form the time series of the regional average value of daily precipitation at the observation and simulation scales.

[0040]

[0041] In the formula, The average precipitation over the basin area on day t; Let N be the precipitation at the i-th station on day t; N is the total number of meteorological stations in the basin.

[0042] Step 62: Classify the observed time series of regional average daily precipitation and the time series of regional average daily temperature: precipitation is classified into intervals of 0-0.1 mm, 0.1-1 mm, and 5 mm increments up to the maximum value; daily average temperature is classified into intervals of 5℃ increments according to the actual range of the observed data.

[0043] Step 63: Using month, precipitation classification, and temperature classification as dimensions, record the dry and wet conditions of the day before and the day after each historical observation day to construct a historical observation date pool containing multiple dimensions.

[0044] Step 64: For each non-zero precipitation day in the simulated time series of regional average daily precipitation, determine the precipitation class to which it belongs. The temperature class is the class to which the regional average temperature observed on the simulated day belongs. At the same time, extract the wet and dry conditions of the day before and after the simulated day as constraints.

[0045] Step 65: Employ a multi-level backoff matching strategy, from strict to lenient, to filter candidate dates from the historical observation date pool according to the following priorities: Prioritize historical dates that simultaneously satisfy the four constraints of month, precipitation class, temperature class, and previous / next dry / wet state; if the candidate set is empty, relax the dry / wet state constraint and retain the three constraints of month, precipitation class, and temperature class; if it is still empty, further relax the temperature class constraint and retain only the month and precipitation class constraints; if it is still empty, search across months for dates that satisfy the precipitation class and temperature class constraints.

[0046] Step 66: Randomly select a historical date from the final candidate set as the matching result;

[0047] Step 67: To ensure sufficient candidate samples, perform multiple resampling operations (e.g., 50 times) to provide ensemble simulation capabilities.

[0048] Furthermore, in step 7, the daily precipitation simulation sequence for each station is decomposed into an hourly precipitation simulation sequence using the hourly precipitation proportion coefficients corresponding to the matched historical observation dates. This specifically includes the following steps:

[0049] Step 71: Based on the matched historical dates, first attempt to obtain the hourly precipitation ratio vector for the current station on that date. ;

[0050] Step 72: If the current site has no valid data, then randomly borrow the proportional coefficient from other sites on the same date;

[0051] Step 73: Multiply the daily precipitation by the proportionality coefficient for each hour to obtain the hourly precipitation:

[0052]

[0053] In the formula, For the simulation of the first Hourly precipitation; This is a simulated daily precipitation. For the first Historical proportion coefficient for hours.

[0054] Accordingly, an intra-diurnal precipitation simulation device based on weather generators and precipitation decomposition includes: one or more processors;

[0055] Storage device for storing one or more programs or user data;

[0056] When the one or more programs are executed by one or more processors, the one or more processors implement a method for intra-diurnal precipitation simulation based on weather generators and precipitation decomposition.

[0057] Accordingly, a storage medium for intra-diurnal precipitation simulation based on weather generator and precipitation decomposition is provided, on which a computer program is stored. When the program is executed by a processor, it implements an intra-diurnal precipitation simulation method based on weather generator and precipitation decomposition.

[0058] Accordingly, a product of an intra-diurnal precipitation simulation program based on weather generators and precipitation decomposition includes a computer program that, when executed by a processor, implements an intra-diurnal precipitation simulation method based on weather generators and precipitation decomposition.

[0059] Beneficial effects: Compared with the prior art, the present invention has the following significant advantages: (1) It establishes a complete precipitation simulation framework from daily to hourly scales, organically combining the weather generator with the precipitation decomposition algorithm, overcoming the defect that each link of the traditional method is independent; it adopts the empirical rank rearrangement method to post-process the simulation sequence, effectively restoring the time correlation structure of precipitation, and significantly improving the simulation accuracy of statistical features such as the first-order autocorrelation coefficient; it adopts a joint stratified sampling strategy based on precipitation and temperature to match historical dates, making full use of the covariance relationship between meteorological elements and improving the representativeness of the hourly precipitation distribution pattern; it adopts an ensemble simulation framework, quantifies the simulation uncertainty through multiple realization and multiple sampling mechanisms, and provides probabilistic forecasting capabilities for hydrological model application and risk assessment. Attached Figure Description

[0060] Figure 1 This is a schematic diagram of the method flow of the present invention.

[0061] Figure 2This is a scatter plot comparing the basic statistical characteristics of the daily precipitation sequence generated by this invention.

[0062] Figure 3 This is a scatter plot comparing the basic statistical characteristics of the hourly precipitation sequence generated by this invention.

[0063] Figure 4 This is a scatter plot comparing the correlations between stations in precipitation sequences at different time scales generated in this invention. Detailed Implementation

[0064] A method for simulating intra-diurnal precipitation based on weather generators and precipitation decomposition includes the following steps:

[0065] Step 1: Collect hourly precipitation data and daily average temperature data from meteorological stations within the study area, remove outliers and missing values, interpolate missing values ​​using adjacent stations to form continuous time series samples, calculate the regional daily average temperature, and form a training dataset.

[0066] Step 2: Based on the hourly precipitation data of each station, calculate the total daily precipitation to form a daily precipitation observation sequence; calculate the proportion of each hourly precipitation to the total daily precipitation to form an hourly precipitation proportion matrix for each station.

[0067] Step 3: Based on the Markov chain, calculate the dry-wet state transition probability matrix for each month, and simulate the daily precipitation sequence of each station by Monte Carlo random sampling.

[0068] Step 4: Based on the moving time window, the K-nearest neighbor resampling method (KNN) is used to randomly extract precipitation from the daily precipitation observation sequence to generate a simulated daily precipitation sequence for each station.

[0069] Step 5: Empirical rank rearrangement correction. The EC method is used to post-process the daily precipitation simulation series to restore the spatiotemporal correlation structure of the simulation series.

[0070] Step 6: Based on the joint stratified sampling method of regional average precipitation and regional average temperature, match historical observation dates for each precipitation day in the simulated sequence of regional average daily precipitation;

[0071] Step 7: Using the hourly precipitation ratio coefficients of each station corresponding to the matched historical observation dates, decompose the daily precipitation simulation sequence of each station into an hourly precipitation simulation sequence.

[0072] A method for simulating intra-diurnal precipitation based on weather generators and precipitation decomposition is presented. Taking the Fupingzi watershed of the Haihe-Daqinghe River Basin as an example, the following process is implemented on the R language platform, and the method steps are described in detail with reference to the attached figures.

[0073] like Figure 1 As shown, this intra-diurnal precipitation simulation method includes the following steps:

[0074] Step 1: Collect hourly precipitation observation data and daily average temperature data for the Fuping sub-basin. The hourly precipitation observation data comes from the hydrological yearbook compiled by the Hebei Provincial Hydrological Survey and Research Center, and the daily average temperature data comes from the CN05.1 gridded observation dataset.

[0075] Data from nine meteorological stations in the Fuping sub-basin from 1988 to June to October 2017 were collected and preprocessed to obtain hourly precipitation observation data; the daily average temperature in the CN05.1 dataset was processed into the Fuping basin surface average series.

[0076] Step 2: Based on hourly precipitation observation data, calculate the total daily precipitation to form a daily precipitation observation sequence; for each precipitation day, calculate the proportion of each hourly precipitation to the total daily precipitation. This forms an hourly precipitation ratio matrix:

[0077]

[0078] In the formula, The precipitation ratio coefficient for the h-th hour; The rainfall in the h-th hour is expressed in millimeters. This represents the total rainfall for the day, expressed in millimeters.

[0079] Step 3: Based on the Markov chain, calculate the dry-wet state transition probability matrix for each month, and simulate the daily precipitation sequence by Monte Carlo random sampling.

[0080] Step 31: Define precipitation state variables When the daily rainfall is greater than or equal to a preset threshold (e.g., 0.1 mm). (Wet day), otherwise (Dry Day). The daily precipitation data calculated in step 2 is selected as the input variable, and the dry and wet states are defined accordingly.

[0081] Step 32: Count the number of state transitions between adjacent days by month and year, construct a transition count matrix, and count the frequency of the four types of transitions; normalize the counts in each row to obtain the transition probability matrix; the transition probability calculation formula is:

[0082]

[0083] In the formula, For the first Year Moon in state Transition to state The probability of; From state Transition to state The number of times; In a state The total number of days. When the sample size for a certain month or year is insufficient (i.e., the count for a certain row is zero), the corresponding row of the multi-year average transition probability matrix for that month is used as the substitute;

[0084] Step 33: For each simulated year, a year is randomly selected from historical years using the Monte Carlo method. The Markov chain is driven by the transition probability matrix of the corresponding month of that year. The state transitions of the Markov chain model follow the following probabilities:

[0085]

[0086] In the formula, Indicates the first Daily precipitation patterns; and These represent the precipitation status of the previous day and the current day, respectively, with values ​​of 0 (dry) or 1 (wet); From state Transition to state The conditional probability.

[0087] Step 34: Using the first day's dry / wet state in the corresponding segment of the daily precipitation observation sequence as the initial state, generate uniformly distributed random numbers for each day during the simulation period, compare them with the cumulative transition probability of the current state, determine the dry / wet state (0 or 1) for the next day, and recursively generate a complete precipitation occurrence sequence day by day.

[0088] Step 4: Based on the moving time window, the K-nearest neighbor resampling method (KNN) is used to randomly extract precipitation from the daily precipitation observation sequence to generate a daily precipitation sequence;

[0089] Step 41: For each wet day in the simulated daily precipitation sequence, determine its annual day number. ;

[0090] Step 42, with Centered on historical daily precipitation data, a time window is constructed. The window width is the preset number of days;

[0091] Step 43: Randomly select a precipitation value from all historical precipitation events within the time window as the simulated value;

[0092] Step 44: When the number of valid precipitation events within the window is insufficient, adaptively expand the window width until the minimum sample size requirement is met.

[0093] Step 5: Empirical rank rearrangement correction. The simulated daily precipitation series is post-processed using the Empirical Copula (EC) method to restore the spatiotemporal correlation structure of the simulated series.

[0094] Step 51: Process the observed and simulated precipitation series separately by month to obtain the simulated series for each month. and observation sequence ;

[0095] Step 52: Calculate the rank and sort index of the observation sequence. , making Sort in ascending order;

[0096] Step 53: Sort the simulated sequences in ascending order to obtain... ;

[0097] Step 54: Reorder the sorted simulation sequences using the rank index of the observed sequences:

[0098]

[0099] In the formula, This is the corrected simulation sequence. The corrected simulation sequence retains the marginal distribution characteristics of the simulation while possessing a spatiotemporal correlation structure similar to that of the observations.

[0100] Step 6: Using a joint stratified sampling method based on precipitation and temperature, match historical observation dates for each simulated precipitation day;

[0101] Step 61: For the daily precipitation observation sequence and the simulated daily precipitation sequence, use the arithmetic mean method to calculate the watershed average value, and form the time series of the regional average value of daily precipitation at the observation and simulation scales.

[0102]

[0103] In the formula, The average precipitation over the basin area on day t; Let N be the precipitation at the i-th station on day t; N is the total number of meteorological stations in the basin.

[0104] Step 62: Classify the observed time series of regional average daily precipitation and the time series of regional average daily temperature: precipitation is classified into intervals of 0-0.1 mm, 0.1-1 mm, and 5 mm increments up to the maximum value; daily average temperature is classified into intervals of 5℃ increments according to the actual range of the observed data.

[0105] Step 63: Using month, precipitation classification, and temperature classification as dimensions, record the dry and wet conditions of the day before and the day after each historical observation day to construct a historical observation date pool containing multiple dimensions.

[0106] Step 64: For each non-zero precipitation day in the simulated time series of regional average daily precipitation, determine the precipitation class to which it belongs. The temperature class is the class to which the regional average temperature observed on the simulated day belongs. At the same time, extract the wet and dry conditions of the day before and after the simulated day as constraints.

[0107] Step 65: Employ a multi-level backoff matching strategy, from strict to lenient, to filter candidate dates from the historical observation date pool according to the following priorities: Prioritize historical dates that simultaneously satisfy the four constraints of month, precipitation class, temperature class, and previous / next dry / wet state; if the candidate set is empty, relax the dry / wet state constraint and retain the three constraints of month, precipitation class, and temperature class; if it is still empty, further relax the temperature class constraint and retain only the month and precipitation class constraints; if it is still empty, search across months for dates that satisfy the precipitation class and temperature class constraints.

[0108] Step 66: Randomly select a historical date from the final candidate set as the matching result;

[0109] Step 67: To ensure sufficient candidate samples, perform multiple resampling operations (e.g., 50 times) to provide ensemble simulation capabilities.

[0110] Step 7: Using the hourly precipitation ratio coefficient corresponding to the matched historical dates, decompose the simulated daily precipitation into hourly precipitation sequences.

[0111] Step 71: Based on the matched historical dates, first attempt to obtain the hourly precipitation ratio vector for the current station on that date. ;

[0112] Step 72: If the current site has no valid data, then randomly borrow the proportional coefficient from other sites on the same date;

[0113] Step 73: Multiply the daily precipitation by the proportionality coefficient for each hour to obtain the hourly precipitation:

[0114]

[0115] In the formula, For the simulation of the first Hourly precipitation; This is a simulated daily precipitation. For the first Historical proportion coefficient for hours.

[0116] The daily-scale precipitation sequence generated in step 5 is compared and analyzed with the observed sequence to evaluate the results. Four typical statistical measures are selected: mean, standard deviation, 95th quantile, and 99th quantile. Based on observed sequences from different stations and different months, and the simulated sequence generated by the method of this invention, the above indicators are calculated respectively, and a scatter plot comparing the simulated values ​​and observed values ​​is plotted, as shown below. Figure 2 As shown in the figure. The results demonstrate that the method of this invention can effectively simulate various statistical characteristics of diurnal precipitation sequences.

[0117] The generated hourly precipitation series was compared and analyzed with the observed series to evaluate the results. The generated hourly precipitation series and the observed series were aggregated to 3h, 6h, and 12h scales. Four typical statistics were selected: mean, variance, skewness, and wet / dry ratio. Based on the observed series from different stations and different months, and the simulated series generated by the method of this invention, the above indicators were calculated at different time scales, and scatter plots comparing the simulated values ​​and observed values ​​were drawn, as shown below. Figure 3 As shown in the figure. The results show that the method of the present invention can effectively simulate various statistical characteristics such as the mean and variance of intra-diurnal precipitation sequences in each month.

[0118] Furthermore, the inter-station correlation between the generated hourly precipitation series and the observed series was analyzed to assess the spatial correlation of the generated precipitation series. The generated hourly precipitation series and the observed series were aggregated to 3h, 6h, and 12h scales, and the inter-station correlation at different time scales was calculated. Scatter plots comparing simulated and observed values ​​were then generated. Figure 4 As shown in the figure, the results indicate that the intra-diurnal precipitation simulation method based on weather generators and precipitation decomposition can effectively simulate the spatial correlation of intra-diurnal precipitation sequences.

Claims

1. A method for simulating intra-diurnal precipitation based on weather generators and precipitation decomposition, characterized in that, Includes the following steps: Step 1: Collect hourly precipitation data and daily average temperature data from meteorological stations within the study area, remove outliers and missing values, interpolate missing values ​​using adjacent stations to form continuous time series samples, calculate the regional daily average temperature, and form a training dataset. Step 2: Based on the hourly precipitation data of each station, calculate the total daily precipitation to form a daily precipitation observation sequence; Calculate the proportion of each hourly precipitation to the total daily precipitation to form a matrix of hourly precipitation proportions for each station; Step 3: Based on the Markov chain, calculate the dry-wet state transition probability matrix for each month, and simulate the daily precipitation sequence of each station by Monte Carlo random sampling. Step 4: Based on the moving time window, the K-nearest neighbor resampling method (KNN) is used to randomly extract precipitation from the daily precipitation observation sequence to generate a simulated daily precipitation sequence for each station. Step 5: Empirical rank rearrangement correction. The EC method is used to post-process the daily precipitation simulation series to restore the spatiotemporal correlation structure of the simulation series. Step 6: Based on the joint stratified sampling method of regional average precipitation and regional average temperature, match historical observation dates for each precipitation day in the simulated sequence of regional average daily precipitation; Step 7: Using the hourly precipitation ratio coefficients of each station corresponding to the matched historical observation dates, decompose the daily precipitation simulation sequence of each station into an hourly precipitation simulation sequence.

2. The intra-diurnal precipitation simulation method based on weather generators and precipitation decomposition as described in claim 1, characterized in that, In step 2, for each precipitation day in the training dataset, the proportion of hourly precipitation to the total daily precipitation is calculated. : In the formula, The precipitation ratio coefficient for the h-th hour; The rainfall in the h-th hour is expressed in millimeters. This represents the total rainfall for the day, expressed in millimeters.

3. The intra-diurnal precipitation simulation method based on weather generators and precipitation decomposition as described in claim 1, characterized in that, In step 3, based on the moving time window, the K-nearest neighbor resampling method (KNN) is used to randomly extract precipitation data from the daily precipitation observation sequence to generate the simulated daily precipitation sequence for each station. The specific steps include the following: Step 31: Define precipitation state variables When the daily precipitation is greater than or equal to the preset threshold It is a wet day, otherwise For the day of the dry season; Step 32: Count the number of state transitions between adjacent days by month and year, construct a transition count matrix, and count the frequency of the four types of transitions; normalize the counts of each row to obtain the transition probability matrix; The formula for calculating the transition probability is: In the formula, For the first Year Moon in state Transition to state The probability of; From state Transition to state The number of times; In a state Total number of days; When the sample size for a certain month or year is insufficient, i.e., the count for a certain row is zero, the corresponding row of the multi-year average transition probability matrix for that month is used as a substitute. Step 33: For each simulated year, a year is randomly selected from historical years using the Monte Carlo method. The Markov chain is driven by the transition probability matrix of the corresponding month of that year. The state transitions of the Markov chain model follow the following probabilities: In the formula, Indicates the first Daily precipitation patterns; and These represent the precipitation status of the previous day and the current day, respectively, with values ​​of 0 or 1; From state Transition to state The conditional probability; Step 34: Using the first day's dry / wet state in the corresponding segment of the daily precipitation observation sequence as the initial state, generate uniformly distributed random numbers for each day during the simulation period, compare them with the cumulative transition probability of the current state, determine the dry / wet state of the next day as 0 or 1, and recursively generate a complete precipitation occurrence sequence day by day.

4. The intra-diurnal precipitation simulation method based on weather generators and precipitation decomposition as described in claim 1, characterized in that, Step 4, which involves randomly selecting precipitation data from the daily precipitation observation sequence to generate a simulated daily precipitation sequence for each station, includes the following steps: Step 41: For each wet day in the simulated daily precipitation sequence, determine its annual day number. ; Step 42, with Centered on historical data, a time window is constructed. The window width is the preset number of days; Step 43: Randomly select a precipitation value from all historical precipitation events within the time window as the simulated value; Step 44: When the number of valid precipitation events within the window is insufficient, adaptively expand the window width until the minimum sample size requirement is met.

5. The intra-diurnal precipitation simulation method based on weather generators and precipitation decomposition as described in claim 1, characterized in that, Step 5 involves post-processing the daily precipitation simulation series using the EC method to restore the spatiotemporal correlation structure of the simulation series. This includes the following steps: Step 51: Process the observed and simulated precipitation series separately by month to obtain the simulated series for each month. and observation sequence ; Step 52: Calculate the rank and sort index of the observation sequence. , making Sort in ascending order; Step 53: Sort the simulated sequences in ascending order to obtain... ; Step 54: Reorder the sorted simulation sequences using the rank index of the observed sequences: In the formula, The corrected simulation sequence retains the marginal distribution characteristics of the simulation while having a spatiotemporal correlation structure similar to that of the observation.

6. The intra-diurnal precipitation simulation method based on weather generators and precipitation decomposition as described in claim 1, characterized in that, In step 6, the joint stratified sampling method based on regional average precipitation and regional average temperature is used to match historical observation dates for each precipitation day in the simulated sequence of regional average daily precipitation. This specifically includes the following steps: Step 61: For the daily precipitation observation sequence and the simulated daily precipitation sequence, use the arithmetic mean method to calculate the watershed average value, and form the time series of the regional average value of daily precipitation at the observation and simulation scales. In the formula, The average precipitation over the basin area on day t; Let N be the precipitation at the i-th station on day t; N is the total number of meteorological stations in the basin. Step 62: Classify the observed time series of regional average daily precipitation and the time series of regional average daily temperature: precipitation is classified into intervals of 0-0.1 mm, 0.1-1 mm, and 5 mm increments up to the maximum value; daily average temperature is classified into intervals of 5℃ increments according to the actual range of the observed data. Step 63: Using month, precipitation classification, and temperature classification as dimensions, record the dry and wet conditions of the day before and the day after each historical observation day to construct a historical observation date pool containing multiple dimensions. Step 64: For each non-zero precipitation day in the simulated time series of regional average daily precipitation, determine the precipitation class to which it belongs. The temperature class is the class to which the regional average temperature observed on the simulated day belongs. At the same time, extract the wet and dry conditions of the day before and after the simulated day as constraints. Step 65: Employ a multi-level backoff matching strategy, from strict to lenient, to filter candidate dates from the historical observation date pool according to the following priorities: Prioritize historical dates that simultaneously satisfy the four constraints of month, precipitation class, temperature class, and previous / next dry / wet state; if the candidate set is empty, relax the dry / wet state constraint and retain the three constraints of month, precipitation class, and temperature class; if it is still empty, further relax the temperature class constraint and retain only the month and precipitation class constraints; if it is still empty, search across months for dates that satisfy the precipitation class and temperature class constraints. Step 66: Randomly select a historical date from the final candidate set as the matching result; Step 67: To ensure sufficient candidate samples, perform multiple resampling operations to provide ensemble simulation capabilities.

7. The intra-diurnal precipitation simulation method based on weather generators and precipitation decomposition as described in claim 1, characterized in that, Using the hourly precipitation ratio coefficients for each station corresponding to the matched historical observation dates, the daily precipitation simulation series for each station is decomposed into hourly precipitation simulation series, specifically including the following steps: Step 71: Based on the matched historical dates, first attempt to obtain the hourly precipitation ratio vector for the current station on that date. ; Step 72: If the current site has no valid data, then randomly borrow the proportional coefficient from other sites on the same date; Step 73: Multiply the daily precipitation by the proportionality coefficient for each hour to obtain the hourly precipitation: In the formula, For the simulation of the first Hourly precipitation; This is a simulated daily precipitation. For the first Historical proportion coefficient for hours.

8. A diurnal precipitation simulation device based on a weather generator and precipitation decomposition, comprising: One or more processors; Storage device for storing one or more programs or user data; When the one or more programs are executed by one or more processors, the one or more processors implement the intra-diurnal precipitation simulation method based on weather generators and precipitation decomposition as described in any one of claims 1 to 7.

9. A storage medium for intra-diurnal precipitation simulation based on a weather generator and precipitation decomposition, wherein a computer program is stored thereon, which, when executed by a processor, implements the intra-diurnal precipitation simulation method based on a weather generator and precipitation decomposition as described in any one of claims 1 to 7.

10. A product of an intra-scale precipitation simulation program based on a weather generator and precipitation decomposition, comprising a computer program that, when executed by a processor, implements the intra-scale precipitation simulation method based on a weather generator and precipitation decomposition as described in any one of claims 1 to 7.

Images