Urban rainfall area IDF curve establishment constraint method and system

Abstract

The invention provides an urban rainfall area IDF curve establishment constraint method and system. The method comprises the following steps: collecting actually measured rainfall data and GCM simulated rainfall data; establishing a spatial downscaling model, and performing spatial downscaling on the grid monthly rainfall estimated by the GCM in the future period; calculating the condition transition probability of rainfall occurrence in each calendar month in the future period; generating a parameter set for driving a random rainfall generator, and inputting the parameter set into the random rainfall generator to obtain a daily rainfall sequence in a future period; performing time discretization from the daily scale to the hour scale on the daily rainfall sequence in the future time period; extracting an annual maximum value sequence from the hour drop sequence in the future time period to carry out regional frequency analysis, and constructing an urban rainfall region IDF curve; constructing a random forest model, and deducing alternative factors of IDF curve migration change; and an emergence constraint model is constructed, and the uncertainty of urban rainfall risk prediction under future climate change is reduced. The accuracy of rainfall frequency analysis is improved.

Description

Technical Field

[0001] The invention belongs to the technical field of water conservancy projects, and in particular relates to a method and system for establishing constraints on an IDF curve in an urban rainfall area. Background Art

[0002] When designing and assessing the risk of urban water conservancy project infrastructure, it is necessary to estimate the extreme rainfall intensity of different durations and return periods. The rainfall intensity-duration-frequency (IDF) curve can meet this requirement and is one of the important design bases for urban basic drainage facilities.

[0003] The traditional method of establishing an IDF curve is to extract the annual maximum rainfall values ​​of different durations from the hourly measured rainfall data of a single station, and use a suitable probability distribution function to perform rainfall frequency analysis. The IDF curve derived from historical measured rainfall data is based on the consistent extreme value hypothesis, which assumes that the exceedance probability of extreme rainfall events will not change significantly over time during the service life of urban infrastructure drainage facilities. However, in the context of global climate change, the intensity and frequency of extreme rainfall events in most parts of the world have increased, and the extreme flood events caused by this will increase the flood control pressure of urban infrastructure drainage facilities. The IDF curve based on the consistency hypothesis may not be able to cope with the design requirements of urban infrastructure drainage facilities under future climate change. On the other hand, the IDF curve established based on rainfall data from a single station is difficult to guarantee the accuracy of rainfall frequency analysis due to problems such as data shortage or lack of representativeness.

[0004] At present, how to consider future climate change to establish IDF curves has become a new research hotspot. To address this issue, domestic and foreign scholars mainly use two types of methods: 1) constructing non-consistent IDF curves based on covariates; 2) constructing IDF curves based on future rainfall data simulated by global climate models (GCMs). The former models the measured extreme rainfall trends and uses covariates to incorporate the trends in the simulated measured rainfall data into the consistent IDF curve to construct a non-consistent curve. However, the process of constructing a non-consistent IDF curve based on covariates is relatively complicated and its accuracy depends on the selection of the optimal covariates, which is difficult to select, so it is less used; GCM provides rainfall sequences under different future climate change scenarios, which can provide useful future climate change information for the construction of IDF curves.

[0005] However, the spatial and temporal resolution of rainfall data simulated by GCM is low, that is, the spatial resolution of GCM output data is generally greater than 1°×1°, which makes it difficult to accurately reflect the spatial heterogeneity of urban rainfall and flood events; in terms of time, it often only provides data on a daily or monthly scale. Even if some GCMs can provide data on a 3-hour or 6-hour scale, the accuracy is low; at the same time, the rainfall data estimated by GCM often has large systematic deviations, which is particularly obvious under complex urban convection conditions. Therefore, it is impossible to directly use the rainfall data output by GCM to construct the IDF curve. Before using the future rainfall data estimated by GCM to construct the IDF curve, it is necessary to perform spatiotemporal downscaling to obtain high-resolution and high-precision rainfall data series. Statistical downscaling methods have been widely used due to their advantages such as simple use, small amount of calculation and high accuracy. However, the existing statistical downscaling methods mostly focus on spatial downscaling. At present, there is a lack of methods that use statistical methods to directly downscale the coarse-grid GCM monthly rainfall data to station hourly data, which cannot meet the actual requirements of IDF curve construction in urban areas. Summary of the invention

[0006] The purpose of the present invention is to address the shortcomings of the prior art and provide a method for establishing constraints on the IDF curve of an urban rainfall area. The spatiotemporal cascade statistical downscaling method is used to compensate for the low spatiotemporal resolution of rainfall data simulated by GCM, and the rainfall information of multiple stations in the consistent area is comprehensively considered to improve the accuracy of rainfall frequency analysis.

[0007] In order to solve the above technical problems, the present invention adopts the following technical solutions:

[0008] A method for establishing a constraint on an IDF curve in an urban rainfall area comprises the following steps:

[0009] Step 1: Collect rainfall data in the study area and the gridded monthly rainfall in the future period estimated by GCM;

[0010] Step 2: establish the relationship between the measured monthly rainfall at the site and the grid monthly rainfall in the historical period simulated by GCM based on the data collected in step 1, and obtain the site-scale monthly rainfall forecast data series for the future period based on the relationship and the grid monthly rainfall data for the future period forecasted by GCM;

[0011] Step 3, obtaining the relationship between the rainfall conditional transfer probability and the monthly average rainfall based on the measured daily rainfall data, and then calculating the monthly rainfall data for the future period obtained in step 2 to obtain the conditional transfer probability of rainfall occurring in each calendar month in the future period;

[0012] Step 4, based on the monthly rainfall in the future period obtained in step 2 and the rainfall transfer probability obtained in step 3, a parameter set for driving the random rainfall generator is generated, and the parameter set is input into the random rainfall generator to obtain a data sequence of daily rainfall forecast in the future period;

[0013] Step 5, based on the daily rainfall forecast data sequence for the future period, discretize the daily rainfall forecast data for the future period from the daily scale to the hourly scale;

[0014] Step 6: Use a sliding time window to extract the maximum rainfall annual value series of different durations from the hourly rainfall forecast data for future periods obtained in step 5, perform regional frequency analysis on them, and calculate the design rainfall values ​​of different return periods to construct the urban rainfall regional IDF curve considering future climate change;

[0015] Step 7: Based on the IDF curve established in step 6, the changes in urban rainfall risk under each climate model are deduced, and a random forest model is constructed with the daily meteorological factors output by the climate model to deduce the candidate factors for the migration change of the IDF curve;

[0016] Step 8: Based on the candidate factors selected in step 7, an emergence constraint model is constructed to reduce the uncertainty of urban rainfall risk prediction under future climate change.

[0017] Furthermore, step 2 includes the following sub-steps:

[0018] Step 2.1 For different calendar months, rank the measured monthly rainfall at the site and the GCM simulated monthly rainfall at the historical grid, and use a polynomial to fit the nonlinear relationship between the two. obs,his,m =f1(R GCM,his,m ), as a spatial downscaling model, where R GCM,his,m is the grid monthly rainfall data for the historical period, R obs,his,m It is the monthly rainfall data of the measured sites;

[0019] Step 2.2 Substitute the grid monthly rainfall data for the future period predicted by GCM into the spatial downscaling model established in step 2.1 to obtain the station monthly rainfall forecast data R for the future period. fut,m =f1(R GCM,fut,m ), R GCM,fut,m Gridded monthly rainfall data for future periods predicted by GCM.

[0020] Furthermore, step 3 includes the following sub-steps:

[0021] Step 3.1: Divide the measured daily rainfall data into multiple rainfall subsets of equal length. For different rainfall subsets, calculate the conditional transition probability of rainfall in each calendar month: P 11 = P{It rains on the tth day|It rains on the t-1th day}, P01 = P{rainfall on day t|no rainfall on day t-1};

[0022] Step 3.2 For different rainfall subsets, calculate the monthly average rainfall of the measured rainfall data for each calendar month, denoted as μ m , and then fit the conditional transition probability and the average monthly rainfall μ m The linear relationship between 11 =f2(μ m ), P 01 =f3(μ m );

[0023] Step 3.3 Based on the linear relationship fitted in step 3.2 and combined with the monthly rainfall data of the future period that has been spatially downscaled in step 2, calculate the conditional transition probability P of rainfall in each calendar month in the future period 11,fut =f2{e[f1(R GCM,fut,m )]},P 01,fut =f3{E[f1(R GCM,fut,m )]}, where E[f1(R gCM,fut,m )] is the mean monthly rainfall in the future period after spatial downscaling in step 2, R GCM,fut,m Gridded monthly rainfall data for future periods predicted by GCM.

[0024] Furthermore, step 4 includes the following sub-steps:

[0025] Step 4.1 Calculate the mean and variance of daily rainfall in the future period based on the monthly rainfall in the future period obtained in step 2, where the mean of daily rainfall in the future period is:

[0026] In the formula, E[f1(R gCM,fut,m )] is the mean monthly rainfall in the future period after spatial downscaling, N d is the number of days in the calendar month, is the probability of rainfall in each calendar month in the future period, is the average number of rainy days in that calendar month;

[0027] Variance of daily rainfall in future periods:

[0028]

[0029] In the formula, D[f1(R GCM,fut,m )] is the variance of the monthly rainfall in the future period after spatial downscaling for the calendar month, P 01,fut -P 11,fut Autocorrelation coefficient of unconditional probability of rainfall;

[0030] Step 4.2: The conditional transition probability of rainfall in each calendar month in the future period calculated in step 3 and the variance of daily rainfall in the future period calculated in step 4.1 are organized into a rainfall parameter set par{P 11,fut ,P 01,fut ,μ d,fut ,σ 2 d,fut}, and input the parameter set into the random rainfall generator to obtain the daily rainfall data sequence for the future period, denoted as R fut,d .

[0031] Furthermore, step 5 includes the following sub-steps:

[0032] Step 5.1 Use the K-nearest neighbor resampling algorithm to select the K nearest neighbors of the target day’s rainfall, calculate the corresponding weights, and select the parent individuals required by the genetic algorithm from them, denoted as x p , and the parent individual, denoted as x o* ;

[0033] Step 5.2: x o and x p* Perform crossover and mutation to obtain offspring individuals, denoted as Based on the K nearest neighbors obtained in step 5.1, select a nearest neighbor x δ,h right Each chromosome of is mutated to obtain the mutated offspring individuals. Finally, based on the obtained offspring individuals, the hourly rainfall distribution ratio α is calculated. t,h , and as a time discrete model on a daily to hourly scale;

[0034] Step 5.3 Set the target day rainfall R fut,d(t) Substituting into the time discrete model established in step 5.2, we get the rainfall from 1 to 24 hours on day t: R t,h＝ α t,h R fut,d(t) , where t is the tth day, and h is 1-24 hours;

[0035] Step 5.4 Repeat the above steps n times, where n is the daily rainfall sequence R fut,d until the hourly rainfall data sequence R for the future period after time and space downscaling is obtained fut,h .

[0036] Furthermore, step 6 includes the following sub-steps:

[0037] Step 6.1: Use a sliding time window to extract the annual maximum rainfall series of different durations from the hourly-scale rainfall forecast data for future periods obtained in step 5;

[0038] Step 6.2 uses the partitioned linear moment method to perform regional frequency analysis on the annual maximum rainfall series extracted in step 6.1;

[0039] Step 6.3: Based on the design rainfall values ​​of different return periods of each station obtained by regional frequency analysis, construct the urban rainfall regional IDF curve considering climate change information.

[0040] Furthermore, the specific implementation of step 6.2 is as follows:

[0041] 1) The initial hydrological consistency area is divided according to the annual maximum rainfall sequence, and the inharmony test and uniformity test are performed based on the linear moment coefficient of each station in the initial hydrological consistency area, and the outliers are eliminated to obtain the final hydrological consistency area;

[0042] 2) Perform dimensionless processing on the sample sequences of each station in the determined hydrological consistent area to generate dimensionless sample sequences reflecting the commonality of regional rainfall and scale factors reflecting the rainfall characteristics of this station;

[0043] 3) Calculate the regional linear moment coefficient based on the dimensionless sample sequence of each station;

[0044] 4) Calculate the parameters of the alternative probability distribution line type based on the regional linear moment coefficient, and use the Monte Carlo simulation test method to perform a goodness of fit test to determine the optimal distribution line type and obtain the regional growth curve;

[0045] 5) Based on the scale factor of each station and the regional growth curve, the index flood method is used to calculate the design rainfall value of each station under different return periods.

[0046] Furthermore, step 7 includes the following sub-steps:

[0047] Step 7.1 collects multiple observed meteorological factor data and interpolates them to each rainfall station;

[0048] Step 7.2: After calibration with multiple climate models in the historical period, the annual maximum rainfall intensity value obtained from the IDF curve and the candidate meteorological factor sequence of the same period of maximum rainfall intensity are used to establish a random forest model, and the prediction effect of the random forest model is verified;

[0049] Step 7.3: Based on the prediction results of the verified random forest model, the out-of-bag error index is used to obtain the degree of explanation of each meteorological factor on rainfall intensity, and the meteorological factors with the greatest importance in predicting rainfall intensity series are selected as candidate factors for constructing the emergence constraint model.

[0050] Further, step 8 includes the following sub-steps:

[0051] Step 8.1 Collect the measured data of the candidate factors inferred in step 7, denoted as X O, the rainfall intensity in the future period is recorded as Y, and the correlation coefficient ρ between the historical candidate factor simulated by GCM and the rainfall intensity in the future period estimated by GCM is calculated, where the historical candidate factor simulated by GCM is recorded as X GCM , the rainfall intensity in the future period predicted by GCM is recorded as Y GCM ;

[0052] Step 8.2 Standard deviation of historical candidate factors simulated by GCM and the standard deviation σ of the observation data n Calculate the observation signal-to-noise ratio:

[0053]

[0054] Among them, σ represents the standard deviation of each variable, and the observation error is recorded as n;

[0055] Step 8.3: Based on the data obtained in steps 8.1 and 8.2, the emergence constraint relationship is established using the Bayesian hierarchical emergence constraint method:

[0056]

[0057] in, is the mean of the precipitation intensity in the future period after the emergence constraint forecast, is the regression coefficient;

[0058] The selected main controlling meteorological factors are used to constrain the extreme rainfall intensity at different durations in the future to reduce the uncertainty in estimating urban rainstorm risks.

[0059] Another object of the present invention is to provide a system for implementing the above-mentioned urban rainfall area IDF curve establishment constraint method, comprising:

[0060] The data acquisition module collects rainfall data in the study area and the gridded monthly rainfall in the future period estimated by GCM;

[0061] The monthly rainfall prediction module is used to establish the relationship between the measured monthly rainfall at the site and the grid monthly rainfall in the historical period simulated by GCM based on the collected data, and obtain the site-scale monthly rainfall prediction data series for the future period based on the established relationship and the grid monthly rainfall data for the future period predicted by GCM;

[0062] The conditional transfer probability calculation module is used to obtain the relationship between the rainfall conditional transfer probability and the monthly average rainfall according to the measured daily rainfall data, and then calculate the conditional transfer probability of rainfall in each calendar month in the future period using the obtained monthly rainfall data in the future period;

[0063] The daily rainfall prediction module is used to generate a parameter set for driving the random rainfall generator according to the obtained monthly rainfall in the future period and the obtained rainfall transfer probability, and input the parameter set into the random rainfall generator to obtain a data sequence of daily rainfall prediction in the future period;

[0064] The rainfall discretization module is used to discretize the daily rainfall forecast data in the future period from the daily scale to the hourly scale according to the daily rainfall forecast data sequence in the future period;

[0065] The urban rainfall regional IDF curve construction module is used to extract the annual maximum rainfall series of different durations from the hourly rainfall forecast data obtained in the future period using a sliding time window, and conduct regional frequency analysis on them, and calculate the design rainfall values ​​of different return periods to construct the urban rainfall regional IDF curve considering future climate change;

[0066] The module for inferring alternative factors is used to infer the changes in urban rainfall risks under various climate models based on the established IDF curve, and to build a random forest model with the daily meteorological factors output by the climate model to infer alternative factors for the migration changes of the IDF curve;

[0067] The urban rainfall risk prediction module is used to construct an emergent constraint model based on the preferred alternative factors to reduce the uncertainty of urban rainfall risk prediction under future climate change.

[0068] Compared with the prior art, the beneficial effects of the present invention are as follows: the present invention first obtains a sequence of estimated rainfall extremes with high spatiotemporal resolution for future periods. Subsequently, the partitioned linear moment method is further used to divide the study area into several hydrologically consistent areas, and the rainfall extreme sequence of each station in the consistent area is subjected to regional frequency analysis to obtain the rainfall design value under different return periods, and the urban rainfall regional IDF curve considering future climate change information is constructed based on this, and finally the uncertainty of the extreme rainfall risk prediction results is reduced by the emergence constraint method. The present invention compensates for the low spatiotemporal resolution and large deviation of the rainfall data simulated by GCM through the proposed spatiotemporal cascade statistical downscaling method, and adopts a regional frequency analysis method based on the partitioned linear moment method to comprehensively consider the rainfall information of multiple stations in the consistent area, thereby improving the accuracy of rainfall frequency analysis, and reducing the uncertainty of the prediction results by adopting the emergence constraint method, thereby improving the reliability of the prediction results. Therefore, the present invention can scientifically and rationally establish an urban rainfall regional IDF curve considering future climate change information. BRIEF DESCRIPTION OF THE DRAWINGS

[0069] Figure 1 A specific flow chart of the method for establishing constraints on the IDF curve of the urban rainfall area according to the embodiment of the invention;

[0070] Figure 2A schematic diagram of a QQ map of monthly rainfall before and after spatial downscaling using the conversion function method of an embodiment of the invention;

[0071] Figure 3 A schematic diagram of the relationship between the average monthly rainfall in each calendar month and the conditional transition probability of rainfall occurrence in an embodiment of the invention;

[0072] Figure 4 A schematic diagram showing a comparison between the multi-year average hourly rainfall values ​​of the historical period measured and simulated by the time discrete model of the embodiment of the invention;

[0073] Figure 5 It is a schematic diagram of regional IDF curves under different recurrence periods in the future period of an embodiment of the invention. DETAILED DESCRIPTION

[0074] The technical solutions in the embodiments of the present invention will be described clearly and completely below in combination with the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0075] It should be noted that, in the absence of conflict, the embodiments of the present invention and the features in the embodiments may be combined with each other.

[0076] The present invention will be further described below in conjunction with specific embodiments, but the present invention is not limited thereto.

[0077] The method for establishing a constraint on the IDF curve of an urban rainfall area in an embodiment of the present invention is based on a constructed spatiotemporal cascade statistical downscaling model. First, a conversion function method is used to spatially downscale the grid-scale monthly rainfall data predicted by GCM to obtain station-scale monthly rainfall prediction data. Subsequently, a random rainfall generator is used to disaggregate the spatially downscaled monthly rainfall prediction data into a daily rainfall sequence. In the process of temporal downscaling, the rainfall occurrence state is downscaled on the one hand, and the rainfall amount is downscaled on the other hand. Finally, the daily rainfall prediction data generated by the random rainfall generator is input into a time discrete model coupled with K-nearest neighbor resampling and genetic algorithm to obtain hourly-scale rainfall prediction data and extract rainfall extreme value sequences of different durations. Further, a partitioned linear moment method is used to divide the study area into a number of hydrologically consistent areas, and a regional frequency analysis is performed on the rainfall extreme value sequences of each station in the consistent area to obtain the design rainfall values ​​under different recurrence periods. Finally, an IDF curve of an urban rainfall area is established according to the design rainfall value. The specific process is described in detail in Figure 1 :

[0078] Step 1: Collect rainfall data in the study area and GCM-estimated grid monthly rainfall in the future period; this step includes:

[0079] Step 1.1 Collect the hourly rainfall data series measured at the stations in the study area (denoted as R his,h ), and accumulate to obtain the corresponding daily rainfall data series (denoted as R his,d ) and the monthly rainfall data series (denoted as R his,m ); The station measured hourly rainfall data sequence collected in this embodiment is from January 1, 1981 to December 31, 2010, a total of 30 years of station hourly rainfall data;

[0080] Step 1.2 Collect the historical grid monthly rainfall data (denoted as R) simulated by GCM in the study area. GCM,his,m ), and the gridded monthly rainfall data for the future period predicted by GCM (denoted as R GCM,fut,m ). The simulated rainfall data collected in this embodiment include the gridded monthly rainfall data of the historical period from January 1, 1981 to December 31, 2010, a total of 30 years, and the gridded monthly rainfall data of the future period from January 1, 2011 to December 31, 2040, a total of 30 years, simulated by GCM.

[0081] Step 2: establish the relationship between the measured monthly rainfall at the site and the grid monthly rainfall in the historical period simulated by GCM based on the data collected in step 1, and obtain the site-scale monthly rainfall forecast data series for the future period based on the relationship and the grid monthly rainfall data for the future period forecasted by GCM; this step includes the following sub-steps:

[0082] Step 2.1 For different calendar months, arrange the measured monthly rainfall at the site and the monthly rainfall at the grid in the historical period simulated by GCM in ascending order, and use a polynomial to fit the nonlinear relationship between the two: R obs,his,m =f1(R GCM,his,m ) and use it as a spatial downscaling model;

[0083] Step 2.2 Substitute the grid monthly rainfall data of the future period predicted by GCM into the spatial downscaling model established in step 2.1 to obtain the estimated monthly rainfall data of the station in the future period:

[0084] R fut,m =f1(R GCM,fut,m );

[0085] Figure 2 The QQ plots of GCM simulated monthly rainfall and measured monthly rainfall before and after spatial downscaling using the transfer function method are shown, which can reflect the performance of the transfer function method in downscaling grid data to station data.

[0086] Step 3, based on the measured daily rainfall data, the relationship between the rainfall conditional transfer probability and the average monthly rainfall is obtained, and then the monthly rainfall data for the future period obtained in step 2 is used to calculate the conditional transfer probability of rainfall occurring in each calendar month in the future period; specifically:

[0087] Step 3.1: Divide the measured daily rainfall data into four rainfall subsets of equal length, and calculate the conditional transition probability of rainfall occurrence in different subsets;

[0088] First, the measured daily rainfall data are divided into four rainfall subsets of equal length, including two rainfall subsets divided equally by year (e.g., 50 years of data are divided equally into two 25-year periods), and two rainfall subsets divided by monthly rainfall, the driest group and the wettest group (e.g., 50 years of data are divided into 25 years with more rainfall and 25 years with less rainfall); then, for different rainfall subsets, the conditional transition probability of rainfall occurring in each calendar month is calculated separately:

[0089] P 11 = P{It rains on the tth day|It rains on the t-1th day}

[0090] P 01 = P{rainfall on day t|no rainfall on day t-1}

[0091] Among them, P 11 is the conditional transition probability that there is rainfall on the t-th day of the calendar month and there is rainfall on the t-1th day, P 01 is the conditional transition probability that there is rainfall on the t-th day of the calendar month and no rainfall on the t-1th day;

[0092] In this embodiment, the rainfall threshold is 0.1 mm; if the rainfall on that day is greater than 0.1 mm, it is a rainy day, otherwise it is a non-rainy day. Based on the above steps, for each calendar month, a set of conditional transition probabilities of rainfall occurrence containing four subsets will be derived, namely, P 11 Set and P 01 A collection of data to characterize the occurrence of rainfall under different climatic conditions;

[0093] Step 3.2 For different rainfall subsets, calculate the monthly average rainfall (denoted as μ m ), and respectively fit P 11 , P 01 and μ m The linear relationship between:

[0094] P 11 =f2(μ m )

[0095] P 01 =f3(μ m )

[0096] Figure 3 The linear relationship between the average monthly rainfall and the conditional transition probability of rainfall occurrence in each calendar month is shown;

[0097] Step 3.3: Combine the linear relationship fitted in step 3.2 with the monthly rainfall data of the future period that has been spatially downscaled in step 2 to calculate the conditional transition probability of rainfall in each calendar month in the future period:

[0098] P 11,fut =f2{E[f1(R GCM,fut,m )]}

[0099] P 01,fut =f3{E[f1(R GCM,fut,m )]}

[0100] Among them, E[f1(R GCM,fut,m )] is the mean monthly rainfall in the future period after spatial downscaling in step 2. Through the above two formulas, the monthly rainfall after spatial downscaling and the conditional transfer probability of rainfall in temporal downscaling are effectively cascaded.

[0101] Step 4, based on the monthly rainfall in the future period obtained in step 2 and the rainfall transition probability obtained in step 3, a parameter set for driving the random rainfall generator is generated, and the parameter set is input into the random rainfall generator to obtain a data sequence of daily rainfall prediction in the future period; specifically, step 4 includes:

[0102] Step 4.1 Calculate the mean and variance of daily rainfall in the future period based on the monthly rainfall in the future period obtained in step 2; first, for each calendar month, construct an analytical formula to calculate the mean of daily rainfall in the future period:

[0103]

[0104] In the formula, E[f1(R GCM,fut,m )] is the mean monthly rainfall in the future period after spatial downscaling, N d is the number of days in the calendar month, is the probability of rainfall in each calendar month in the future period, is the average number of rainy days in that calendar month;

[0105] Then construct an analytical formula to calculate the variance of daily rainfall in the future period:

[0106]

[0107] In the formula, D[f1(R GCM,fut,m )] is the variance of the monthly rainfall in the future period after spatial downscaling for the calendar month, P 01,fut -P 11,fut Autocorrelation coefficient of unconditional probability of rainfall;

[0108] Through the above two formulas, an effective cascade of spatial downscaling and daily temporal downscaling is achieved;

[0109] Step 4.2 For each calendar month, a set of rainfall parameter sets containing four parameters is generated based on the mean and variance of daily rainfall in the future period obtained in step 4.2 and the rainfall conditional transition probability obtained in step 3. 11,fut ,P 01,fut ,μ d,fut ,σ 2 d,fut}, this parameter set is the input parameters required by the common random rainfall generator to simulate the daily rainfall sequence in the future period;

[0110] This embodiment assumes that the skewness coefficient of daily rainfall in the future period does not change, and calculates the skewness coefficient g of daily rainfall based on the measured daily rainfall data.

[0111] Step 4.3 Input the parameter set generated in step 4.2 into the random rainfall generator to obtain the daily rainfall data sequence for the future period (denoted as R fut,d ). This embodiment uses the rainfall module in the random rainfall generator CLIGEN to simulate the daily rainfall data sequence in the future period. This step is a conventional technique in the technical field. For ease of understanding, the principle of the random rainfall generator CLIGEN involved in this step is briefly introduced below.

[0112] When simulating daily rainfall, CLIGEN uses a first-order two-state Markov chain to simulate the conditional transition probability P of rainfall occurrence. 11 and P 01 The basic idea is that the probability of rainfall on a certain day is only related to the dry and wet conditions of the previous day. When simulating the rainfall state of a certain day, firstly, the rainfall state of the previous day (rainy day or no rainfall day) is given arbitrarily, and then a random number in the interval [0,1] is generated and compared with the conditional transition probability P 11 or P 01 For comparison, if the generated random number is less than the conditional transition probability, then the day is a rainy day. For a rainy day, this embodiment uses a skewed normal distribution function to simulate daily rainfall, and the calculation formula is:

[0113]

[0114] Where P is the daily rainfall, μ, σ and g are the mean, standard deviation and skewness coefficient of the daily rainfall in the calendar month, respectively, and x is the standard normal deviation of the daily rainfall in the calendar month, which is generated by two uniformly distributed random numbers in the range [0,1];

[0115] On the contrary, the day is a rainless day, and the rainfall is 0. Then, the dry and wet state of this day is used as the initial value to determine the rainfall state of the next day, and the rainfall of the next day is calculated. By analogy, the daily rainfall data sequence of the required period can be obtained.

[0116] Step 5, based on the daily rainfall forecast data sequence for the future period, discretize the daily rainfall forecast data for the future period from the daily scale to the hourly scale;

[0117] Based on the daily rainfall forecast data in the future period, the K-nearest neighbor algorithm is used to select the K nearest neighbors with the highest similarity from the measured hourly rainfall data sequence and calculate the corresponding weights. The father and mother individuals are selected from the nearest neighbors according to the calculated weights. Then, the genetic algorithm is used to perform crossover and mutation on the two to obtain the offspring individuals. A time discrete model is established based on the hourly rainfall distribution ratio within the day to discretize the daily rainfall forecast data in the future period from the daily scale to the hourly scale. The specific steps include the following:

[0118] Step 5.1 Use the K-nearest neighbor resampling algorithm to select the K nearest neighbors of the target day’s rainfall, calculate the corresponding weights, and select the parent individuals required by the genetic algorithm (denoted as x p ) and the parent individual (denoted as x p* ); For each calendar month, firstly take the daily rainfall on the tth day of the future period daily rainfall sequence obtained in step 4 as the target daily rainfall (denoted as R fut,d(t) ), and calculate its difference with the daily rainfall R on the ith day of the measured daily rainfall data series obs,his,d(i) The Euclidean distance is:

[0119] (R fut,d(t) -R obs,his,d(i) ) 2

[0120] The smaller the Euclidean distance is, the higher the similarity between the two.

[0121] Then, D i Arrange in ascending order, according to the time index i of the first K distances, select the corresponding K days of hourly rainfall in the measured hourly rainfall data sequence as the K nearest neighbors, and calculate the corresponding weights:

[0122]

[0123] The number of nearest neighbors set in this embodiment is K=20;

[0124] Then, based on the calculated weights, two are selected from the K nearest neighbors as the parent individuals x and x required to run the genetic algorithm. p =[x p,h ] h∈{1,24} and the parent individual x p* =[xp*,h ] h∈{1,24} , the rainfalls from 1 to 24 hours corresponding to the two correspond to the 24 chromosomes of the genetic algorithm;

[0125] Step 5.2 for x p and x p* Perform crossover and mutation to obtain offspring individuals (denoted as ); For each calendar month, first, the x obtained in step 5.1 p and x p* Each chromosome of is crossovered to obtain the offspring individuals after crossover:

[0126]

[0127] Among them, ε is a random number uniformly distributed between the interval [0,1], P c is the crossover probability;

[0128] The crossover probability P used in this embodiment is c =0.1;

[0129] Then, based on the K nearest neighbors obtained in step 5.1, select another nearest neighbor x δ,h right Each chromosome of is mutated to obtain the mutated offspring individuals:

[0130]

[0131] Among them, ε is a random number uniformly distributed between the interval [0,1], P m is the crossover probability;

[0132] The mutation probability P used in this embodiment is m =0.01.

[0133] Finally, based on the obtained offspring individuals, calculate their hourly rainfall distribution ratio within the day:

[0134]

[0135] α h As a time discrete model on a daily to hourly scale;

[0136] Through steps 5.1 and 5.2, the coupling of the nearest neighbor resampling algorithm and the genetic algorithm is achieved.

[0137] Step 5.3 Set the target day rainfall R fut,d(t) Substituting into the time discrete model established in step 5.2, we get the hourly rainfall:

[0138] R t,h =α t,hR fut,d(t)

[0139] In the formula, t is the tth day, and h is 1-24 hours;

[0140] Step 5.4 Repeat the above steps n times (n is the daily rainfall sequence R fut,d until the hourly rainfall data sequence R for the future period after time and space downscaling is obtained. fut,h .

[0141] The above steps achieve an effective cascade of downscaling from daily scale to hourly scale. The time discrete model used in this embodiment is a random model. To eliminate the uncertainty of random simulation, this embodiment generates 300 30-year random rainfall sequences each time to capture the overall trend of hourly rainfall.

[0142] Figure 4 This is a comparison chart between the multi-year average of the hourly rainfall data series in the historical period simulated by the time discrete model and the multi-year average of the measured hourly rainfall data series, where the simulated values ​​are displayed through box plots of 300 random rainfall sequences.

[0143] Step 6, using a sliding time window to extract the annual maximum rainfall series of different durations from the hourly rainfall forecast data for future periods obtained in step 5, and perform regional frequency analysis on them, calculate the design rainfall values ​​of different return periods to construct the urban rainfall regional IDF curve considering future climate change; this step includes the following sub-steps:

[0144] Step 6.1: Use a sliding time window to extract the annual maximum rainfall series of different durations from the hourly rainfall forecast data for future periods obtained in step 5;

[0145] When this embodiment uses a sliding time window to extract the annual maximum rainfall sequence, the window lengths are set to 1 hour, 3 hours, 6 hours, 12 hours and 24 hours (corresponding to different rainfall durations), and the step length of each sliding is 1 hour.

[0146] Step 6.2 uses the partitioned linear moment method to perform regional frequency analysis on the annual maximum rainfall sequence extracted in step 6.1 to obtain the design rainfall values ​​of different return periods at each station. The partitioned linear moment method is a conventional technology in this field and has been widely used. The following is a brief introduction, which includes the following 5 steps:

[0147] 1) The initial hydrological consistency area is divided according to the annual maximum rainfall sequence, and the inharmony test and uniformity test are performed based on the linear moment coefficient of each station in the initial hydrological consistency area, and the outliers are eliminated to obtain the final hydrological consistency area;

[0148] 2) Perform dimensionless processing on the sample sequences of each station in the determined hydrological consistent area to generate dimensionless sample sequences reflecting the commonality of regional rainfall and scale factors reflecting the rainfall characteristics of this station;

[0149] 3) Calculate the regional linear moment coefficient based on the dimensionless sample sequence of each station;

[0150] 4) Based on the regional linear moment coefficient, the parameters of the alternative probability distribution line type are calculated, and the Monte Carlo simulation test method is used to perform a goodness of fit test to determine the optimal distribution line type and obtain the regional growth curve.

[0151] 5) Based on the scale factor of each station and the regional growth curve, the index flood method is used to calculate the design rainfall value of each station under different return periods;

[0152] Step 6.3: Connect the design rainfall values ​​of different return periods of each station obtained by regional frequency analysis to obtain the IDF curve of urban rainfall area considering future climate change. Figure 5 Shown is a schematic diagram of regional IDF curves under different return periods.

[0153] Step 7: Based on the IDF curve established in step 6, the change of urban rainfall risk under each climate model is deduced, and a random forest model is constructed with the daily meteorological factors output by the climate model to deduce the candidate factors for the migration change of the IDF curve; step 7 further includes the following sub-steps:

[0154] Step 7.1 collects multiple observed meteorological factor data and interpolates them to each rainfall station using the inverse distance weighted interpolation method. The meteorological factors considered in this embodiment include daily average temperature, relative humidity, shortwave radiation, leaf area index (LAI), specific humidity, soil moisture, and annual temperature variability, relative humidity variability, shortwave radiation variability, LAI variability, specific humidity variability, and soil moisture variability;

[0155] Step 7.2: After calibration with multiple climate models of historical periods lasting d (d = 1h, 2h, 4h, ...), the annual maximum rainfall intensity value obtained from the IDF curve and the candidate meteorological factor sequence of the same period of maximum rainfall intensity are used to establish a random forest model, and the root mean square error of the rainfall intensity prediction result and the original result is calculated. And the correlation coefficient Verify the prediction effect of the random forest model, y pre,i Represents the rainfall intensity result predicted by the random forest model, y ori,i represents the original result of rainfall intensity, and m represents the total number of prediction samples;

[0156] Step 7.3 Based on the prediction results of the verified random forest model, the out-of-bag error index is used to measure the importance of each meteorological factor to the rainfall intensity of duration d.d , screen out the meteorological factors with the greatest importance in predicting the rainfall intensity series of duration d as candidate factors for constructing the emergence constraint model; the method of selecting important factors based on the out-of-bag error index is a conventional technology in this field.

[0157] Step 8: Based on the candidate factors obtained in step 7, an emergent constraint model is constructed to reduce the uncertainty of urban rainfall risk prediction under future climate change;

[0158] The meteorological factors with the greatest importance in rainfall intensity prediction are used as candidate factors for constructing emergent constraints in this step. Taking into account the uncertainty of observation data, the Bayesian hierarchical emergent framework is used to further constrain rainfall intensity under different scenarios in the future. This step specifically includes the following sub-steps:

[0159] Step 8.1 Collect the measured data of the candidate factors inferred in step 7 (denoted as X O ) and the rainfall intensity in the future period (denoted as Y), and calculate the historical alternative factor (denoted as X) simulated by GCM GCM ) and the rainfall intensity in the future period predicted by GCM (denoted as Y GCM )’s correlation coefficient ρ;

[0160] Step 8.2 Due to the uncertainty in observational data, use the standard deviation of the historical candidate factor simulated by GCM and the standard deviation σ of the observation data n , calculate the signal-to-noise ratio of the observed data:

[0161]

[0162] Step 8.3 Calculate the regression coefficient using the model historical candidate factors (predictors) and the model future rainfall intensity (predicted values):

[0163]

[0164] Among them, ρ is the correlation coefficient between the current prediction factor and the future prediction value, and The standard deviation of each;

[0165] On this basis, the Bayesian hierarchical emergence constraint method is used to construct the emergence constraint model. Specifically, the Bayesian hierarchical emergence constraint relationship with observation uncertainty added is:

[0166]

[0167] That is, through the above formula, the selected alternative meteorological factors are used to constrain the rainfall intensity of different durations in the future to reduce the uncertainty of estimating urban rainfall risks.

[0168] The embodiment of the present invention further provides a system for implementing the above-mentioned method for establishing constraints on the IDF curve of the urban rainfall area, comprising:

[0169] The data acquisition module collects rainfall data in the study area and the gridded monthly rainfall in the future period estimated by GCM;

[0170] The monthly rainfall prediction module is used to establish the relationship between the measured monthly rainfall at the site and the grid monthly rainfall in the historical period simulated by GCM based on the collected data, and obtain the site-scale monthly rainfall prediction data series for the future period based on the established relationship and the grid monthly rainfall data for the future period predicted by GCM;

[0171] The conditional transfer probability calculation module is used to obtain the relationship between the rainfall conditional transfer probability and the monthly average rainfall according to the measured daily rainfall data, and then calculate the conditional transfer probability of rainfall in each calendar month in the future period using the obtained monthly rainfall data in the future period;

[0172] The daily rainfall prediction module is used to generate a parameter set for driving the random rainfall generator according to the obtained monthly rainfall in the future period and the obtained rainfall transfer probability, and input the parameter set into the random rainfall generator to obtain a data sequence of daily rainfall prediction in the future period;

[0173] The rainfall discretization module is used to discretize the daily rainfall forecast data in the future period from the daily scale to the hourly scale according to the daily rainfall forecast data sequence in the future period;

[0174] The urban rainfall regional IDF curve construction module is used to extract the annual maximum rainfall series of different durations from the hourly rainfall forecast data obtained in the future period using a sliding time window, and conduct regional frequency analysis on them, and calculate the design rainfall values ​​of different return periods to construct the urban rainfall regional IDF curve considering future climate change;

[0175] The module for inferring alternative factors is used to infer the changes in urban rainfall risks under various climate models based on the established IDF curve, and to build a random forest model with the daily meteorological factors output by the climate model to infer alternative factors for the migration changes of the IDF curve;

[0176] The urban rainfall risk prediction module is used to construct an emergent constraint model based on the obtained alternative factors to reduce the uncertainty of urban rainfall risk prediction under future climate change.

[0177] The above are only preferred embodiments of the present invention, and are not intended to limit the implementation methods and protection scope of the present invention. Those skilled in the art should be aware that all solutions obtained by equivalent substitutions and obvious changes made using the contents of the specification of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for establishing constraints on IDF curves in urban rainfall areas, characterized in that: The following steps are involved: Step 1: Collect rainfall data in the study area and the gridded monthly rainfall in the future period estimated by GCM; Step 2: establish the relationship between the measured monthly rainfall at the site and the grid monthly rainfall in the historical period simulated by GCM based on the data collected in step 1, and obtain the site-scale monthly rainfall forecast data series for the future period based on the relationship and the grid monthly rainfall data for the future period forecasted by GCM; Step 3, obtaining the relationship between the rainfall conditional transfer probability and the monthly average rainfall based on the measured daily rainfall data, and then calculating the monthly rainfall data for the future period obtained in step 2 to obtain the conditional transfer probability of rainfall occurring in each calendar month in the future period; Step 4, based on the monthly rainfall in the future period obtained in step 2 and the rainfall transfer probability obtained in step 3, a parameter set for driving the random rainfall generator is generated, and the parameter set is input into the random rainfall generator to obtain a data sequence of daily rainfall forecast in the future period; Step 5, based on the daily rainfall forecast data sequence for the future period, discretize the daily rainfall forecast data for the future period from the daily scale to the hourly scale; Step 6: Use a sliding time window to extract the maximum rainfall annual value series of different durations from the hourly rainfall forecast data for future periods obtained in step 5, perform regional frequency analysis on them, and calculate the design rainfall values ​​of different return periods to construct the urban rainfall regional IDF curve considering future climate change; Step 7: Based on the IDF curve established in step 6, the changes in urban rainfall risk under each climate model are deduced, and a random forest model is constructed with the daily meteorological factors output by the climate model to deduce the candidate factors for the migration change of the IDF curve; Step 8: Based on the candidate factors obtained in step 7, an emergence constraint model is constructed to reduce the uncertainty of urban rainfall risk prediction under future climate change.

2. The method for establishing constraints on the IDF curve of urban rainfall areas according to claim 1 is characterized in that: Step 2 includes the following sub-steps: Step 2.1 For different calendar months, rank the measured monthly rainfall at the site and the GCM simulated monthly rainfall at the historical grid, and use a polynomial to fit the nonlinear relationship between the two, R obs，his，m =f1(R GCM，his，m ), as a spatial downscaling model, where R GCM,his，m is the grid monthly rainfall data for the historical period, R obs，his，m It is the monthly rainfall data of the measured sites; Step 2.2 Substitute the grid monthly rainfall data for the future period predicted by GCM into the spatial downscaling model established in step 2.1 to obtain the station monthly rainfall forecast data R for the future period. fut，m =f1(R GCM,fut，m ), R GCM,fut，m Gridded monthly rainfall data for future periods predicted by GCM.

3. The method for establishing constraints on the IDF curve of urban rainfall areas according to claim 1 is characterized in that: Step 3 includes the following sub-steps: Step 3.1: Divide the measured daily rainfall data into multiple rainfall subsets of equal length. For different rainfall subsets, calculate the conditional transition probability of rainfall in each calendar month: P 11 = P{It rains on the tth day|It rains on the t-1th day}, P 01 = P{rainfall on day t|no rainfall on day t-1}; Step 3.2 For different rainfall subsets, calculate the monthly average rainfall of the measured rainfall data for each calendar month, denoted as μ m , and then fit the conditional transition probability and the average monthly rainfall μ m The linear relationship between 11 =f2(μ m ), P 01 =f3(μ m ); Step 3.3 Based on the linear relationship fitted in step 3.2 and combined with the monthly rainfall data of the future period that has been spatially downscaled in step 2, calculate the conditional transition probability P of rainfall in each calendar month in the future period 11，fut =f2{E[f1(R GCM,fut，m )]},P 01，fut =f3{E[f1(R GCM,fut，m )]}, where E[f1(R GCM,fut，m )] is the mean monthly rainfall in the future period after spatial downscaling in step 2, R GCM,fut，m Gridded monthly rainfall data for future periods predicted by GCM.

4. The method for establishing constraints on the IDF curve of urban rainfall areas according to claim 3 is characterized in that: Step 4 includes the following sub-steps: Step 4.1 Calculate the mean and variance of daily rainfall in the future period based on the monthly rainfall in the future period obtained in step 2, where the mean of daily rainfall in the future period is: In the formula, E[f1(R GCM,fut，m )] is the mean monthly rainfall in the future period after spatial downscaling, N d is the number of days in the calendar month, is the probability of rainfall in each calendar month in the future period, is the average number of rainy days in that calendar month; Variance of daily rainfall in future time periods: In the formula, D[f1(R GCM,fut，m )] is the variance of the monthly rainfall in the future period after spatial downscaling for the calendar month, P 01，fut-P11，fut Autocorrelation coefficient of unconditional probability of rainfall; Step 4.2: The conditional transition probability of rainfall in each calendar month in the future period calculated in step 3 and the variance of daily rainfall in the future period calculated in step 4.1 are organized into a rainfall parameter set par{P 11,fut , P 01,fut， μ d,fut ,σ 2 d,fut }, and input the parameter set into the random rainfall generator to obtain the daily rainfall data sequence for the future period, denoted as R fut，d .

5. The method for establishing constraints on the IDF curve of urban rainfall areas according to claim 3 is characterized in that: Step 5 includes the following sub-steps: Step 5.1 Use the K-nearest neighbor resampling algorithm to select the K nearest neighbors of the target day’s rainfall, calculate the corresponding weights, and select the parent individuals required by the genetic algorithm from them, denoted as x p , and the parent individual, denoted as x p* ; Step 5.2: x p and x p* Perform crossover and mutation to obtain offspring individuals, denoted as Based on the K nearest neighbors obtained in step 5.1, select a nearest neighbor x δ,h right Each chromosome of is mutated to obtain the mutated offspring individuals. Finally, based on the obtained offspring individuals, the hourly rainfall distribution ratio α is calculated. t，h , and as a time discrete model on a daily to hourly scale; Step 5.3 Set the target day rainfall R fut，d(t) Substituting into the time discrete model established in step 5.2, we get the rainfall from 1 to 24 hours on day t: R t,h =α t，h R fut，d (t), where t is the i-th day and h is 1-24 hours; Step 5.4 Repeat the above steps n times, where n is the daily rainfall sequence R fut，d until the hourly rainfall data sequence R for the future period after time and space downscaling is obtained fut,h .

6. The method for establishing constraints on the IDF curve of urban rainfall areas according to claim 1 is characterized in that: Step 6 includes the following sub-steps: Step 6.1: Use a sliding time window to extract the annual maximum rainfall series of different durations from the hourly-scale rainfall forecast data for future periods obtained in step 5; Step 6.2 uses the partitioned linear moment method to perform regional frequency analysis on the annual maximum rainfall series extracted in step 6.1; Step 6.3: Based on the design rainfall values ​​of different return periods of each station obtained by regional frequency analysis, construct the urban rainfall regional IDF curve considering climate change information.

7. The method for establishing constraints on the IDF curve of urban rainfall areas according to claim 6 is characterized in that: The specific implementation method of step 6.2 is: 1) The initial hydrological consistency area is divided according to the annual maximum rainfall sequence, and the inharmony test and uniformity test are performed based on the linear moment coefficient of each station in the initial hydrological consistency area, and the outliers are eliminated to obtain the final hydrological consistency area; 2) Perform dimensionless processing on the sample sequences of each station in the determined hydrological consistent area to generate dimensionless sample sequences reflecting the commonality of regional rainfall and scale factors reflecting the rainfall characteristics of this station; 3) Calculate the regional linear moment coefficient based on the dimensionless sample sequence of each station; 4) Calculate the parameters of the alternative probability distribution line type based on the regional linear moment coefficient, and use the Monte Carlo simulation test method to perform a goodness of fit test to determine the optimal distribution line type and obtain the regional growth curve; 5) Based on the scale factor of each station and the regional growth curve, the index flood method is used to calculate the design rainfall value of each station under different return periods.

8. The method for establishing constraints on the IDF curve of urban rainfall areas according to claim 1 is characterized in that: Step 7 includes the following sub-steps: Step 7.1 collects multiple observed meteorological factor data and interpolates them to each rainfall station; Step 7.2: After calibration with multiple climate models in the historical period, the annual maximum rainfall intensity value obtained from the IDF curve and the candidate meteorological factor sequence of the same period of maximum rainfall intensity are used to establish a random forest model, and the prediction effect of the random forest model is verified; Step 7.3: Based on the prediction results of the verified random forest model, the out-of-bag error index is used to obtain the degree of explanation of each meteorological factor on rainfall intensity, and the meteorological factors with the greatest importance in predicting rainfall intensity series are selected as candidate factors for constructing the emergence constraint model.

9. The method for establishing constraints on the IDF curve of urban rainfall areas according to claim 1, characterized in that: Step 8 includes the following sub-steps: Step 8.1 Collect the measured data of the candidate factors inferred in step 7, denoted as X O , the rainfall intensity in the future period is recorded as Y, and the correlation coefficient ρ between the historical candidate factor simulated by GCM and the rainfall intensity in the future period estimated by GCM is calculated, where the historical candidate factor simulated by GCM is recorded as X GCM , the rainfall intensity in the future period predicted by GCM is recorded as Y GCM ; Step 8.2 Standard deviation of historical candidate factors simulated by GCM and the standard deviation σ of the observation data n Calculate the observation signal-to-noise ratio: Among them, σ represents the standard deviation of each variable, and the observation error is recorded as n; Step 8.3: Based on the data obtained in steps 8.1 and 8.2, the emergence constraint relationship is established using the Bayesian hierarchical emergence constraint method: in, is the mean of the precipitation intensity in the future period after the emergence constraint forecast, is the regression coefficient; The selected main controlling meteorological factors are used to constrain the extreme rainfall intensity at different durations in the future to reduce the uncertainty in estimating urban rainstorm risks.

10. A system for implementing the method for establishing constraints on the IDF curve of an urban rainfall area according to any one of claims 1 to 9, characterized in that: include: The data acquisition module collects rainfall data in the study area and the gridded monthly rainfall in the future period estimated by GCM; The monthly rainfall prediction module is used to establish the relationship between the measured monthly rainfall at the site and the grid monthly rainfall in the historical period simulated by GCM based on the collected data, and obtain the site-scale monthly rainfall prediction data series for the future period based on the established relationship and the grid monthly rainfall data for the future period predicted by GCM; The conditional transfer probability calculation module is used to obtain the relationship between the rainfall conditional transfer probability and the monthly average rainfall according to the measured daily rainfall data, and then calculate the conditional transfer probability of rainfall in each calendar month in the future period using the obtained monthly rainfall data in the future period; The daily rainfall prediction module is used to generate a parameter set for driving the random rainfall generator according to the obtained monthly rainfall in the future period and the obtained rainfall transfer probability, and input the parameter set into the random rainfall generator to obtain a data sequence of daily rainfall prediction in the future period; The rainfall discretization module is used to discretize the daily rainfall forecast data in the future period from the daily scale to the hourly scale according to the daily rainfall forecast data sequence in the future period; The urban rainfall regional IDF curve construction module is used to extract the annual maximum rainfall series of different durations from the hourly rainfall forecast data obtained in the future period using a sliding time window, and conduct regional frequency analysis on them, and calculate the design rainfall values ​​of different return periods to construct the urban rainfall regional IDF curve considering future climate change; The module for inferring alternative factors is used to infer the changes in urban rainfall risks under various climate models based on the established IDF curve, and to build a random forest model with the daily meteorological factors output by the climate model to infer alternative factors for the migration changes of the IDF curve; The urban rainfall risk prediction module is used to construct an emergent constraint model based on the obtained alternative factors to reduce the uncertainty of urban rainfall risk prediction under future climate change.

Images