Flood season extreme rainfall sub-season prediction method based on air temperature sequence evolution characteristics

Abstract

The invention relates to the technical field of extreme rainfall prediction, and discloses a flood season extreme rainfall sub-season prediction method based on air temperature sequence evolution characteristics. The method comprises the following steps: preprocessing rainfall data and air temperature data of a to-be-predicted area in a specific historical period, and extracting strong rainfall annual change characteristics; extracting a fitting factor by using a March-May observation data sequence related to the strong rainfall interannual change characteristics and a derivative sequence thereof, generating a fitting curve having a specific similarity with the strong rainfall interannual change, and establishing a judgment threshold according to the rare degree of rainfall intensity and flood records; establishing a rainstorm prediction model of the to-be-predicted area by using the obtained fitting curve and the judgment threshold value; and inputting analysis data of observation elements in March to May of a prediction sample year to obtain a prediction result of whether disastrous extreme rainfall occurs in the to-be-predicted area in June to September. The method has the advantages of strong pertinence and relatively high accuracy, can enrich the content of season-secondary season prediction products, and can also be used as a reference for determining the early warning level before the heavy rainfall process.

Description

Technical Field

[0001] The present invention relates to the technical field of climate and extreme heavy precipitation prediction during flood season, and in particular to a sub-seasonal prediction method for extreme precipitation during flood season based on the evolution characteristics of temperature sequences. Background Art

[0002] There are many estuarine alluvial fan landforms with similar shapes in the area where the eastern edge of the Qinghai-Tibet Plateau meets the Sichuan Basin. Among them, the Minjiang River system has many tributaries from Dujiangyan, passing through Wenjiang and Pidu Districts and traveling 50km to reach the vicinity of Chengdu City. The terrain drop is as high as 273m, and the terrain slope of 3‰ to 10‰ makes it easy for water conservancy facilities in this area to be damaged and flooded when heavy rainfall occurs, and it also threatens downstream areas. According to the Chengdu Flood Control Plan (2021-2035), in August 2020, the heavy rainfall in Chengdu caused 1.2 billion water conservancy facility losses and 7 billion direct economic losses. Although extreme heavy rainfall is a low-probability event, once it occurs, it will directly threaten the safety of people's lives and property. Research on heavy rainfall prediction and forecasting is of great significance for disaster prevention and mitigation and ensuring the operational benefits of water conservancy projects.

[0003] With the advancement of science and technology, the meteorological department's short-term warning capability for rainstorm weather has been improved, but the early prediction technology for disastrous extreme precipitation is still being improved. Based on the precipitation data of Sichuan meteorological stations from 1995 to 2010 and the return data of 8 models in the World Meteorological Organization's sub-seasonal to seasonal (S2S) prediction plan, Pang Yishu and others evaluated and analyzed the prediction ability of each model for the extreme precipitation time in Sichuan's flood season. The results showed that "the prediction skills of each S2S model for Sichuan's extreme precipitation are generally low" and "enter the low skill period when the time efficiency is 7 to 12 days"; Huo Fei and others studied the connection between snow cover on the Qinghai-Tibet Plateau and precipitation in China, and can carry out macro-situation prediction of precipitation in the flood season. The above research results show that the prediction precision of extreme precipitation in small and medium scales is not enough, and it is necessary to actively explore more accurate early prediction methods for small and medium scale extreme precipitation.

[0004] Wang Shunjiu and others noticed from the perspective of climate analysis that the landforms where high mountains and plains meet on the eastern edge of the Qinghai-Tibet Plateau can also respond sensitively to climate fluctuations, and part of this response may affect later weather through the land surface memory effect. This research result provides useful clues for the possible correlation mechanism between climate data such as spring temperature in the area where the plateau and the Minjiang alluvial fan meet and extreme precipitation in the later flood season. Summary of the invention

[0005] In view of the above problems, the purpose of the present invention is to provide a sub-seasonal prediction method for extreme precipitation in the flood season based on the evolution characteristics of the temperature sequence through in-depth analysis of the temperature climate sequence. The method has the advantages of strong pertinence and high accuracy on a small and medium scale, which can enrich the content of seasonal-sub-seasonal prediction products and can also be used as a reference for determining the early warning level of heavy precipitation processes. The technical solution is as follows:

[0006] A sub-seasonal prediction method for extreme precipitation in the flood season based on the evolution characteristics of the temperature series includes the following steps:

[0007] Step 1: Use the common trend and extreme value change characteristics of the time series of frequency data corresponding to the median value of the spatial difference of the daily temperature minimum to find the sensitive site combination, effective sampling period and evolution pattern characteristics of the temperature spatial difference data series associated with climate events, determine the predictable area and organize the rainfall data and temperature data of the historical period in the area;

[0008] Step 2: Preprocess the rainfall data and temperature data of the specific historical period in the forecast area to extract the interannual variation characteristics of heavy precipitation;

[0009] Step 3: Extract fitting factors using the March-May observation data sequence and its derivative sequence related to the interannual variation characteristics of heavy precipitation, generate a fitting curve with a specific similarity to the interannual variation of heavy precipitation, and establish a judgment threshold based on the rarity of rainfall intensity and flood hydrological records; use the obtained fitting curve and judgment threshold to establish a catastrophic extreme precipitation prediction model for the area to be predicted;

[0010] Step 4: Input the analysis data of the observation elements from March to May of the forecast sample year, and obtain the forecast results of whether disastrous extreme precipitation will occur in the forecast area from June to September.

[0011] Furthermore, the step 1 comprises:

[0012] The difference between the lowest temperatures of the two weather stations is obtained by using the daily lowest temperatures of the two weather stations, and the difference set {ΔT min}, count the frequency f(Xi) of each temperature difference Xi in the set, and arrange the corresponding frequency f(Xi) in the order of the temperature difference Xi, and get the frequency f(Xi) of the difference set ΔT min The frequency analysis sequence {Xi, f(Xi)} of the frequency analysis sequence {Xi, f(Xi)} is defined as the frequency median M e is half of the total frequency, and the cumulative frequency reaches M eAt this time, the corresponding temperature difference Xi value is the median temperature difference MO. The correspondence between the year-on-year change characteristics of the temperature difference frequency corresponding to the median temperature difference MO and the climate events is used to determine the site combination, sampling period and evolution pattern of the data series that the subsequent analysis and modeling work depends on.

[0013] Furthermore, the step 2 comprises:

[0014] Step 2.1: Precipitation data preprocessing

[0015] Get the weather station S in the area to be predicted D Daily rainfall data from June to September in a certain Nx years in history. Count the number of days n when the rainfall from June to September is greater than or equal to the rainstorm threshold, and calculate the total rainfall value of these n rainy days. And the average rainfall intensity value Tr1 of these n rainfall days:

[0016]

[0017] Among them, Ri is the rainfall on a certain day when the rainfall is greater than or equal to the heavy rain threshold;

[0018] For the weather station S in the area to be predicted W Similarly, calculate the average rainfall intensity Tr2 of the rainy days with rainfall greater than or equal to the rainstorm threshold from June to September in each year Nx;

[0019] Then calculate the average rainfall intensity for each year Thus, the average rainfall intensity value sequence during the Nx years is formed.

[0020] Step 2.2: Temperature data preprocessing

[0021] 1) Minimum temperature difference item

[0022] Use the weather station S from March to May every year in a certain Nx year D The daily minimum temperature minus the corresponding day's weather station S W The daily minimum temperature of the two stations is obtained by the difference of the daily minimum temperature of ΔT minDW ; Then calculate the difference ΔT between the daily minimum temperatures of the two stations from March to May each year minDW The seasonal average It is abbreviated as W1;

[0023] 2) Daily temperature range

[0024] Use the weather station S from March to May every year in the Nx years W Daily temperature difference, calculate the seasonal average of the daily temperature difference of the weather station from March to May each year Denoted as W2;

[0025] 3) Temperature stability observation item

[0026] Using Weather StationS W The highest temperature value T of the previous day in the period from March to May each year in the Nx years max1 , minus the highest temperature of the day T max2 Get the difference △T max ; Statistics of △T from March to May each year max Greater than the temperature difference threshold T x The number of times ℃ is recorded as W3;

[0027] 4) Frequency sequence of spatial difference distribution of minimum temperature between stations

[0028] The difference between the lowest temperatures between the two meteorological stations during March to May of each year in the Nx years is obtained by calculating the annual difference set ΔT min , statistical annual difference set ΔT min The frequency f(Xi) of each temperature difference Xi in the equation is calculated, and the corresponding frequency f(Xi) is arranged in the order of the temperature difference Xi, and the difference value set ΔT is obtained. min The frequency analysis sequence of {Xi, f(Xi)};

[0029] Define the frequency median M of the frequency analysis sequence {Xi, f(Xi)} e Half of the total frequency; the cumulative frequency reaches M e , the corresponding temperature difference Xi value at this time is the median temperature difference MO;

[0030] Based on this, for the weather station S in the area to be predicted P With weather station S W Calculate the frequency value F of the median value of the frequency analysis sequence of the daily minimum temperature difference between the two stations from March to May each year PW , recorded as W4;

[0031] For the weather station S in the area to be predicted D With weather station S W , calculate the maximum frequency value F of the frequency analysis sequence of the daily minimum temperature difference between the two stations from March to May each year MAX.DW , recorded as W5;

[0032] Step 2.3: Exploitation The prominent large values ​​in correspond to the characteristics of the years with disastrous extreme precipitation, and the sample years in the years with disastrous extreme precipitation are marked; the Nx-year time series of the five temperature preprocessing data items are calculated respectively.

[0033] The correlation coefficient of .

[0034] Furthermore, the step 3 comprises:

[0035] Step 3.1: Extract anomaly analysis variables from preprocessed temperature data

[0036] 1) Calculate the Nx-year average value of W1, recorded as W 1-Nx , calculated as follows:

[0037]

[0038] Among them, W 1i is the W1 value of the i-th year among Nx years;

[0039] Calculate the deviation analysis variable of W1 value in any sample year, denoted as Z1, as follows:

[0040] Z1=(W1-W 1-Nx ) / A

[0041] Among them, A is the arithmetic comparison value;

[0042] 2) Calculate the Nx-year average value of W2, recorded as W 2-Nx , calculated as follows:

[0043]

[0044] Among them, W 2i is the W2 value of the i-th year among Nx years;

[0045] Calculate the deviation analysis variable of W2 value in any sample year, denoted as Z2, as follows:

[0046] Z2=W2 / W 2-Nx

[0047] 3) Calculate the Nx-year average value of W3, recorded as W 3-Nx , calculated as follows:

[0048]

[0049] Among them, W 3i is the W3 value of the i-th year among Nx years;

[0050] Calculate the deviation analysis variable of W3 value in any sample year, denoted as Z3, as follows:

[0051] Z3=W3 / 〔W 3-Nx -B〕

[0052] Among them, B is the offset;

[0053] 4) Calculate the Nx-year average value of W4, recorded as W 4-Nx , calculated as follows:

[0054]

[0055] Among them, W 4i is the W4 value of the i-th year among Nx years;

[0056] Calculate the anomaly analysis variable of the W4 value for any sample year, denoted as Z4, as follows:

[0057] Z4=W4 / W 4-Nx

[0058] 5) Calculate the Nx-year average value of W5, recorded as W 5-Nx , calculated as follows:

[0059]

[0060] Among them, W 5i is the W5 value of the i-th year among Nx years;

[0061] Calculate the deviation analysis variable of W5 value in any sample year, denoted as Z5, as follows:

[0062] Z5=W5 / W 5-Nx

[0063] Step 3.2: Determine the fitting formula

[0064] Taking Z1~Z5 as fitting factors, the deviation change indicator Tr′ is comprehensively formed:

[0065] Tr′=D*〔1 / Z5 a +Z3 b +2*Z2 c 〕 / 〔Z4–Z1〕

[0066] Among them, D is the multiplier coefficient, a, b and c are the influence weight indexes of fitting factors Z5, Z3 and Z2 respectively;

[0067] Thus, the anomaly change indicator sequence {Tr′} for the Nx years is formed. Nx ;

[0068] Step 3.3: Determine the fitting method

[0069] By adjusting the impact weight indexes a, b and c, the deviation change indicator sequence {Tr′} Nx The average rainfall intensity series The goal is to achieve the maximum correlation coefficient. After obtaining the debugging sequence with the largest correlation coefficient, the same sequence is then obtained with the average value of the debugging sequence. The average value of is compared, the multiplication factor D is determined, and the final fitting result calculation formula is obtained, thereby obtaining the heavy rain prediction model;

[0070] Step 3.4: Based on the anomaly change indicator series {Tr′} of Nx yearsNx Establish indicators to determine whether disastrous extreme precipitation will occur in the disastrous precipitation analysis area from June to September;

[0071] From the set {Tr′} Nx In the years when the element value is greater than the set threshold or when there is a precipitation record with rainfall greater than the rainstorm rainfall threshold, the smallest Tr' value is selected and the smallest Tr' value is used as a judgment indicator for whether there is disastrous extreme precipitation.

[0072] Furthermore, the step 3.3 specifically includes:

[0073] Step 3.3.1: Initialize, set the influence weight index a=1, b=1, c=1; set the multiplier coefficient D=1;

[0074] Step 3.3.2: Obtain the anomaly analysis variable sequence {Z1} for the Nx years based on the observed data Nx , {Z2} Nx , {Z3} Nx , {Z4} Nx and {Z5} Nx , and the sequence

[0075] Step 3.3.3: Substitute the annual data of Z1 to Z5 with the fitting formula under the values ​​of a=1, b=1, c=1, and D=1 to obtain the Nx-year debugging sequence {Tr′} of Tr′ Nx～1 ;

[0076] Step 3.3.4: Adjust the influence weight indexes a, b, and c in steps of ±0.1, and detect {Tr′} Nx～1 and The correlation coefficient is the same as the correlation coefficient of the original image. When the correlation coefficient reaches the maximum value, the debugging is completed. The fixed values ​​of a, b, and c give {Tr′} Nx～2 ;

[0077] Step 3.3.5: First find The average value and {Tr′} Nx～2 The average value, ignoring the unit and looking at the value only, is The average value of divided by {Tr′} Nx～2 The average value of the multiplier coefficient D is obtained;

[0078] Step 3.3.6: Set {Tr′} Nx～2 The element value in is expanded D times to obtain the final target sequence {Tr′} Nx-Z .

[0079] The beneficial effects of the present invention are:

[0080] The prediction method of the present invention has the advantages of strong pertinence and high accuracy. It can enrich the content of local seasonal and sub-seasonal prediction products and can also serve as an indirect reference for the early warning level of rainstorm processes. The practical results also reflect to a certain extent that integrating terrain factors for delicate small-scale zoning is beneficial to improving prediction skills. The prediction results show that the method of the present invention can relatively accurately predict whether there is an extreme flood threat in the area during the flood season of the year, and the prediction conclusion is of great significance to local flood prevention and disaster reduction. BRIEF DESCRIPTION OF THE DRAWINGS

[0081] Figure 1 It is a common evolution sample of the spatial difference frequency time series of air temperature in the Minjiang alluvial fan area.

[0082] Figure 2 Dujiangyan Station - Wenjiang Station Interannual distribution of heavy precipitation above rainstorm with series and combined statistics.

[0083] Figure 3 is {Tr′} 30 and Sequence and the 30-year interannual distribution of heavy precipitation in Wenjiang and Dujiangyan.

[0084] Figure 4 It is the Tr′ value and judgment index line from 1970 to 1999.

[0085] Figure 5 It is the predicted sample Tr′ value from 2000 to 2022.

[0086] Figure 6(a)-(m) shows the distribution trend of precipitation ≥100 mm / d from 2010 to 2022, where the horizontal axis is longitude and the vertical axis is latitude. DETAILED DESCRIPTION

[0087] The present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0088] In order to extract the early information of extreme precipitation, this embodiment uses the frequency sequence characteristics of the stable and recurring spatial difference of the temperature between the Wenjiang National Climate Observatory and the Dujiangyan National Benchmark Climate Station to extract the relevant factors of extreme precipitation, analyze the interannual variation characteristics and advance signals of heavy precipitation, and carry out specific technical analysis on the simulation operation of the interannual variation of heavy precipitation. According to the precipitation and flood disaster records in the Dujiangyan-Wenjiang area, a judgment threshold and seasonal-subseasonal prediction model for catastrophic extreme precipitation in this small-scale range are established. The historical data period for modeling is 1970-1999, and the model verification uses the observation records from 2000 to 2022 to evaluate the prediction effect.

[0089] Using the common trends and extreme value change characteristics of the time series of frequency data corresponding to the median value of the spatial difference of the daily temperature minimum, we can find the sensitive site combinations, effective sampling periods and evolution pattern characteristics of the temperature spatial difference data series associated with climate events, determine the predictable area, and organize the rainfall and temperature data of the area in historical periods.

[0090] The difference between the lowest temperatures of the two weather stations is obtained by using the daily lowest temperatures of the two weather stations, and the difference set {ΔT min}, count the frequency f(Xi) of each temperature difference Xi in the set, and arrange the corresponding frequency f(Xi) in the order of the temperature difference Xi, and we can get the frequency f(Xi) of the difference set ΔT min The frequency analysis sequence {Xi, f(Xi)} of the frequency analysis sequence {Xi, f(Xi)} is defined as the frequency median M e is half of the total frequency, and the cumulative frequency reaches M e At this time, the corresponding temperature difference Xi value is the median temperature difference MO. The correspondence between the year-on-year change characteristics of the frequency of occurrence of the temperature difference value corresponding to MO and the climate events is used to determine the site combination, sampling period and evolution pattern of the data series that the subsequent analysis and modeling work depends on.

[0091] 1 Data preprocessing and analysis

[0092] 1.1 Preprocessing of rainfall data (1970-1999)

[0093] Wenjiang National Climate Observatory and Dujiangyan National Benchmark Climate Station are climate observation stations set up in the west of Chengdu. The two stations are 33km apart and have an altitude difference of 163m. Both stations are located northwest of the upstream of Chengdu’s main urban area along the Minjiang River. They are representative stations corresponding to the terrain conditions of the Minjiang River Gorge Mouth and the plain terrain conditions.

[0094] Using the daily rainfall "R" of Dujiangyan Station from June to September every year from 1970 to 1999, the number of days n with rainfall R ≥ 30 mm from June to September and the total rainfall value of the rainfall days with R ≥ 30 mm were calculated. Calculate the average daily rainfall intensity Tr1 (mm / d) of rainfall ≥30 mm in Dujiangyan from June to September of the same year.

[0095]

[0096] The same method was used to calculate the average daily rainfall intensity Tr2 of more than 30 mm during the corresponding period of 1970-1999 at Wenjiang Station.

[0097] The average value of Tr1 and Tr2 Composition 1970-1999 30a sequence

[0098] Preliminary processing of temperature data: trace back to check the reliability of historical temperature data, determine the station groups and their temperature data sequences that can reliably perform the analysis task within the medium-scale range; process the station-to-station spatial difference sequences of the daily temperature minimum after combining these stations in pairs, and extract the frequencies corresponding to the median values ​​of these difference sequences in units of "season", and then use 30-year historical data to form a time series of frequency analysis of temperature spatial differences; observe the time series of frequency analysis formed by all station combinations, check the common evolutionary characteristics and extreme values ​​and the temporal correlation with local climate events, and extract the evolutionary patterns, sampling periods and station combinations with specific temperature spatial difference analysis significance that may be used for early prediction of climate events. In the alluvial fan area of ​​the Minjiang River, after this analysis, the frequency minimum trend of the spatial difference sequence of the daily temperature minimum of the Dujiangyan-Wenjiang and other station combinations from March to May was determined to be significantly correlated with the hydrological records of the extreme floods and disastrous heavy rainfall that occurred in the alluvial fan area of ​​the Minjiang River from June to September. On this basis, the spatiotemporal evolution characteristics of this type of temperature data were further used to establish an early prediction method for extreme precipitation during the flood season. Figure 1 It is a common evolution sample of the spatial difference frequency time series of air temperature in the Minjiang alluvial fan area.

[0099] 1.2 Preprocessing of temperature data

[0100] 1.2.1 Arrangement of the minimum temperature difference between Dujiangyan Station and Wenjiang Station

[0101] Use the daily minimum temperature of Dujiangyan Station from March to May to subtract the daily minimum temperature of Wenjiang Station, and calculate the difference ΔT between the daily minimum temperatures of the two stations from March to May minDW The quarterly average Abbreviated as "W1".

[0102] 1.2.2 Arrangement of daily temperature difference items at Wenjiang Station

[0103] Using the daily temperature difference of Wenjiang Station from March to May, calculate the seasonal average of the daily temperature difference of the station from March to May Abbreviated as "W2".

[0104] 1.2.3 Arrangement of temperature stability observation items

[0105] Use the highest temperature value Tmax1 of the previous day in the period from March to May at Wenjiang Station, subtract the highest temperature value Tmax2 of the day to get the difference △Tmax. When △Tmax>2.5℃, count 1, and count the number of times △Tmax>2.5℃ occurs from March to May, and use it as the temperature stability observation value "△Tmax.N", that is, there are 92 days in March to May, set i∈1~92, let △Tmax.i>2.5℃ when Ni=1, △Tmax.i≤2.5℃ when Ni=0; then Abbreviated as "W3".

[0106] 1.2.4 Frequency sequence of spatial difference distribution of minimum temperature between statistical stations

[0107] In the frequency sequence of temperature difference, the temperature difference values ​​are arranged in order of their specific values. min The difference between the lowest temperature stations ΔT can be obtained min , ΔT min ∈{…, -0.2℃, -0.1℃, 0℃, +0.1℃, +0.2℃, …}, with a step size of 0.1℃, and statistics of ΔT min The frequency f(Xi) of each value in the set, and arranging the corresponding frequency f(Xi) in the order of Xi, we get the value about ΔT min The frequency analysis sequence {Xi, f(Xi)}. So the sum of f(Xi) and ΔT min The sample size L of the set is equal. Each Xi is called a temperature difference. Define the frequency median M of the frequency analysis sequence {Xi, f(Xi)} e is half of the total frequency, that is, M e =L / 2. The cumulative frequency of the sequence reaches M e , then the corresponding Xi value is the median temperature difference MO, that is,

[0108] According to the above statistical method, the spatial difference frequency series of the daily minimum temperature of the two combinations of Pengzhou-Wenjiang station and Dujiangyan-Wenjiang station are generated respectively. Further, the frequency value F of the median value of the frequency analysis series of the daily minimum temperature difference of the Pengzhou-Wenjiang station combination is extracted. PW 18 , abbreviated as "W4"; extract the maximum frequency value F of the frequency analysis sequence of the daily minimum temperature difference of the Dujiangyan-Wenjiang station combination MAX. DW, abbreviated as "W5".

[0109] 1.3 Data Analysis

[0110] 1.3.1 Analysis and research scope of Chengdu’s 30-year flood disaster and heavy rainfall data from 1970 to 1999

[0111] 1980, 1981, 1987, 1995 and 1998 were the years when Chengdu experienced major floods and severe waterlogging disasters during the period 1970-1999. The rainfall records of the national stations corresponding to these five flood disaster samples are summarized in Table 1.

[0112] Table 1 Daily precipitation at 20-20 o'clock corresponding to the 30-year flood record Unit: (mm / d)

[0113] Time / Station Name Pengzhou Dujiangyan* Pidu District Wenjiang Chongzhou* 1980.6.29 23 213 31 31 170 1981.7.13 177 113 240 249 175 1987.7.29 87 65 75 106 160 1995.8.11 288 224 175 188 266 1998.7.5 127 59 195 357 73

[0114] Extract the rainfall intensity variation series of Dujiangyan Station and Wenjiang Station during the 30-year flood season As a reference sequence, help find Related spring observation element sequences and their derived analysis sequences are screened for Fitting factors for simulation calculations. The fitting factors used in this embodiment come from the five analysis data items W1, W2, W3, W4, and W5 processed by preprocessing the temperature data, and the temperature data used involve Dujiangyan, Pengzhou, and Wenjiang stations.

[0115] In 30 years, among the 211 precipitation samples with heavy rain or above at Dujiangyan and Wenjiang stations, there are only 6 samples with R ≥ 200 mm / d, accounting for 3%, which appeared in 1980, 1981, 1987, 1995, and 1998, respectively, including all 5 years of severe flood disasters related to this analysis area in Table 1. Heavy precipitation with R ≥ 200 mm / d is regarded as rare "extreme" precipitation, and basin flood is used as the disaster type for analysis. "Disastrous extreme precipitation" is defined as: heavy precipitation ≥ 200 mm / d that will cause basin floods or severe waterlogging in the Dujiangyan-Wenjiang disastrous extreme precipitation analysis area. The goal of this embodiment is defined as: using spring climate data, predict whether the analyzed area will experience disastrous extreme precipitation or basin floods during the main flood season of the year. During the study period when there are only precipitation data from national stations, the five stations in the Dujiangyan-Wenjiang disastrous extreme precipitation analysis area will be considered as a whole when evaluating the effect. Under the background of regional heavy rain, as long as one station has a precipitation day of ≥200mm / d or a flood related to the area, it will be regarded as a sample of disastrous extreme precipitation in this area; if none of the five stations has a precipitation day of ≥200mm / d, and there is no basin-wide flood related to the precipitation in this area, it is considered that there is no disastrous extreme precipitation in the entire area. After 2010, with the construction of regional automatic weather stations on a grid basis, the forecast practice will also introduce regional automatic station rainfall data for evaluation using "the total number of stations with precipitation of more than 100mm / d in the region from June to September" and "the cumulative total rainfall of ≥100mm / d".

[0116] 1.3.2 Sequence Expression of disastrous extreme precipitation

[0117] Sequence in Figure 2 The data show that the three large values ​​corresponding to the three severe flood years, ①, ②, and ④, can be found. Another large value, ③, is in 1987. On June 26 of that year, there was a record of 215 mm / d of heavy rainfall in Dujiangyan, and on the same day, there was a regional heavy rain record along the western mountains. The prominent large values ​​in correspond to the years when disastrous extreme precipitation occurred. It is relatively easy to mark out four sample years among the five years when disastrous extreme precipitation occurred, namely 1980, 1981, 1987, 1995 and 1998. This quantitative expression can help find spring observation data and their derived sequences related to disastrous extreme precipitation.

[0118] 1.3.3 Temperature preprocessing data series and Statistical association

[0119] The collection period of temperature preprocessing data items W1, W2, W3, W4, and W5 is from March to May. The collection period is from June to September. The 30-year time series of the five temperature preprocessing data items are There is a certain degree of statistical correlation between the 30-year time series of W1 and The statistical correlation coefficient is 0.38, passing the 0.05 significance test; the 30-year time series of W2 and The statistical correlation coefficient is 0.49, passing the 0.01 significance test; the 30-year time series of W3 and The statistical correlation coefficient of W4 is 0.35; the 30-year time series of W4 is The statistical correlation coefficient is -0.59, passing the 0.01 significance test; the 30-year time series of W5 and The statistical correlation coefficient is -0.33, passing the 0.05 significance test.

[0120] 2 Construction of disastrous precipitation prediction model and index extraction

[0121] 2.1 Temperature preprocessing data items 1970-1999 30-year anomaly analysis variable extraction

[0122] The 30-year average value of W1 is recorded as W 1-30a , unit ℃, 30a, The anomaly analysis variable of any sample year W1 is recorded as "Z1":

[0123] Z1= (W1- (-0.26)) / A (2)

[0124] Where "A" is an arithmetic comparison value, A = 3°C, so Z1 = (W1 + 0.26) / 3 in any sample year. Z1 has no calculation unit. In 30a, the range of W1 is -0.76 to +0.32°C, and the range of Z1 is -0.17 to +0.19.

[0125] The 30-year average value of W2 is recorded as W 2-30a , unit ℃, at 30a The anomaly analysis variable of any sample year W2 is recorded as "Z2":

[0126] Z2=W2 / 8.53℃ (3)

[0127] There is no calculation unit for Z2. In 30a, the range of W2 is 7.61~9.47℃, and the range of Z2 is 0.89~+1.11.

[0128] The 30-year average value of W3 is recorded as W 3-30a , unit "times", The anomaly analysis variable of any sample year W3 is recorded as "Z3":

[0129] Z3=W3 / 〔17.8-B〕times(4)

[0130] In the formula, "B" is the bias value of the calculation setting, and B is taken as 2.8 times. Therefore, the anomaly analysis variable Z3 of W3 in any sample year is W3 / 15 times, and Z3 has no unit. In 30a, the range of W3 is 15 to 21 times, and the range of Z3 is 1 to 1.4.

[0131] The 30-year average value of W4 is recorded as W 4-30a ,unit"%", The anomaly analysis variable of any sample year W4 is recorded as "Z4":

[0132] Z4==W4 / 8.20% (5)

[0133] Z4 has no calculation unit. In 30a, W4 ranges from 5.54 to 10.65%, and Z4 ranges from 0.68 to 1.30.

[0134] The 30-year average value of W5 is recorded as W 5-30a , unit "%", in 30a, The anomaly analysis variable of any sample year W5 is recorded as "Z5":

[0135] Z5=W5 / 6.78% (6)

[0136] There is no calculation unit for Z5. In 30a, the range of W5 is 5.11-9.02%, and the range of Z5 is 0.75-1.33.

[0137] 2.2 Fitting calculation formula

[0138] Because of the collection Therefore, we hope to use Z1 to Z5 extracted from spring data as fitting factors to comprehensively form the anomaly change indicator "Tr'", and reproduce it in the 30-year fitting curve of Tr' through simulation calculation. The prominence in the , in order to extract the advance information of disastrous precipitation. Since the general fitting tool is easy to weaken the prominence value and affect the use effect, a polynomial is used here for fitting:

[0139] Tr′=D *〔1 / Z5 a +Z3 b +2*Z2 c 〕 / 〔Z4–Z1〕 (7)

[0140] (7) In the formula, "D" is a multiplication factor. The formula consists of two parts: the numerator and the denominator. The numerator part is 〔1 / Z5 a +Z3 b +2*Z2 c 〕 contains three fitting factors: Z5, Z3, and Z2:

[0141] Z5: Z5 = W5 / 6.78%. {W5} 30 and is negatively correlated, so the fraction "1 / Z5 a "Enter the fitting operation, the "a" in the fraction is the influence weight index set for Z5; the calculation logic is Increased risk of disastrous precipitation;

[0142] Z3: Z3 = W3 / 15 times. {W3} 30 and For positive correlation, use "Z3 b "Enter the fitting, "b" is the influence weight index set for Z3; the calculation logic is Increased risk of disastrous precipitation;

[0143] Z2: Z2 = W2 / 8.53°C. {W2} 30 and For positive correlation, use "2*Z2 c "Enter the fitting, "c" is the influence weight index set for Z2; the calculation logic is The risk of disastrous precipitation is increasing.

[0144] (7) The denominator [Z4–Z1] contains two fitting factors:

[0145] Z4: Z4 = W4 / 8.20%. {W4} 30 and is negatively correlated, and the logic is Increased risk of disastrous precipitation;

[0146] Z1: {W1} 30 and is positively correlated, and the logic is The risk of disastrous precipitation is increasing.

[0147] 2.3 Fitting method

[0148] The fitting will improve the sequence {Tr′} by adjusting the three weight indices a, b, and c in formula (7). 30 and The similarity of the interannual variation mode is observed, where the focus is on the large prominent values ​​that mark the years of flood disasters. The use of formula (7) for fitting can maintain the interannual variation characteristics of the data and avoid excessive smoothing of the fitted values. The Pearson correlation coefficient is sensitive to the consistency of the outlier values ​​in the two sequences being analyzed. If the two sequences change consistently at the outlier value points, the correlation coefficient increases. Otherwise, the correlation will be seriously weakened. Here, the fitting operation just needs to emphasize the expression of the prominent values, so the fitting debugging first uses {Tr′} 30 with sequence The goal is to maximize the correlation coefficient of the debugging sequence with the largest correlation coefficient, and then use the average value of this debugging sequence to compare with the sequence Compare the average values ​​and determine the multiplier value "D". The specific steps are:

[0149] Step 1: Initialization, let the weight index α=1, b=1, c=1 in formula (7); let the multiplication factor D=1;

[0150] Step 2: Processing 30-year anomaly analysis variable sequence based on observational data {Z1} 30 , {Z2} 30 , {Z3} 30 , {Z4} 30 , {Z5} 30 ,

[0151] And the sequence

[0152] Step 3: Substitute the annual data of Z1 to Z5 into equation (7) with a=1, b=1, c=1, and D=1 to obtain the 30-year debugging sequence {Tr′} of Tr′ 30～1 ;

[0153] Step 4: Adjust the three weight indices a, b, and c respectively with a step size of ±0.1, and detect {Tr′} 30～1 and The correlation coefficient is the same as the correlation coefficient of the original image. When the correlation coefficient reaches the maximum value, the debugging is completed. The fixed values ​​of a, b, and c are obtained to obtain {Tr′} 30～2 ;

[0154] Step 5: Find out The average value and {Tr′} 30～2 The average value, ignoring the unit and looking at the value only, is The average value of divided by {Tr′} 30～2 The average value of the multiplier coefficient D is obtained;

[0155] Step 6: Set {Tr′} 30～2 The element value in is expanded D times to obtain the final target sequence {Tr′} 30 .

[0156] From the fitting steps, we can see that {Tr′} 30 The interannual fluctuation characteristics will not be modified individually; the function of the multiplier coefficient D is to adjust the amplitude as a whole to facilitate comparative analysis without changing the interannual fluctuation characteristics.

[0157] The results of fitting using 30-year observation data from 1970 to 1999 are: {Tr′} 30 and The correlation coefficient is 0.74; the optimal values ​​of weight index are a=0.5, b=1.1, c=4; when the multiplier coefficient D=12.6, {Tr′} 30 Average amount ≈ 58, ( Average amount ≈ 58 mm / d). {Tr′} 30 and The comparative observation is shown in Figure 3 ; Figure 3 , {Tr′} 30 The third largest value point in the sequence, number ③, appeared in 1980. The other four largest value points appeared in the same year. The sequence is the same.

[0158] Fitting result calculation formula:

[0159] Tr′= 12.6 *〔1 / Z5 0.5 +Z3 1.1 +2* Z2 4 〕 / 〔Z4–Z1〕 (8)

[0160] or:

[0161] Tr′=12.6*{〔(W5 / 6.78) -0.5 +(W3 / 15) 1.1 +2*(W2 / 8.53) 4 〕 / 〔W4 / 8.20–(W1+0.26) / 3〕}(9)

[0162] Figure 3 The fitted curve reproduces Figure 2 The distribution of the four large value points in the figure, and the large value point ③ ranked third in the order of magnitude naturally appeared in 1980. This point is A sample of disasters that are easily missed when indicating disaster years.

[0163] 2.4 Extraction of judgment indicators and effect evaluation

[0164] 2.4.1 Basis for extracting judgment indicators

[0165] {Tr′} 30 Comparison The fitting operation is performed to find the 30 Reappearance The outliers in the data are large values. Data analysis shows that these large values ​​correspond to years with disastrous extreme precipitation. 30 The generation factor of the sequence comes from the temperature data from March to May in spring. The Tr′ value of each year is calculated using the same calculation method, the same calculation coefficient and a unified weight index. If {Tr′} 30 The years with large outliers in the sequence can correspond one by one to the years with disastrous extreme precipitation, {Tr′} 30 The sequence may be able to indicate whether there is disastrous extreme precipitation from June to September. 30 The values ​​in are sorted from large to small. The five largest values ​​appear in 1998, 1981, 1980, 1987, and 1995, respectively. Figure 3 According to the numerical points ①②③④⑤ marked in sequence, data analysis shows that in these five years, there are precipitation records of ≥200mm / d in the Dujiangyan-Wenjiang disastrous extreme precipitation analysis area, including four years with basin-wide floods and severe waterlogging disasters. At the same time, except for these five years, there are no precipitation records of ≥200mm / d and flood disaster records in the Dujiangyan-Wenjiang disastrous extreme precipitation analysis area in the 30 years. Based on this indication effect, we can try to use {Tr′} 30 Establish indicators to determine whether disastrous extreme precipitation will occur in the Dujiangyan-Wenjiang disastrous precipitation analysis area from June to September.

[0166] 2.4.2 Extraction of judgment indicators

[0167] According to the set {Tr′} 30 The medium to large element values ​​correspond to the years when disastrous extreme precipitation occurs. Among the years with records of flood disasters for 30 years or records of precipitation of more than 200 mm / d, the smallest Tr′ value is selected. This minimum value corresponding to the disaster year can preliminarily become an indicator for judging whether "there is disastrous extreme precipitation".

[0168] In the 30-year record from 1970 to 1999, Tr′=85 in 1995 is the minimum value of Tr′ among the five years with disastrous extreme precipitation. Tr′≥85 is used as the judgment index of disastrous extreme precipitation (see Figure 4 ), 1980, 1981, 1987, 1995 and 1998 met the criteria for determining disastrous extreme precipitation.

[0169] 2.4.3 Evaluation of the effectiveness of judgment indicators

[0170] In the 30-year sampling period, the index value of Tr′≥85 can identify all the years of disastrous extreme precipitation in the analysis area consisting of Dujiangyan, Wenjiang, Pidu District, Pengxian, and Chongzhou without any errors or omissions; {Tr′} 30 The sequence makes up for The indication of the 1980 flood sample is defective, and the expression of the disaster year is better than its reference sequence exist Figure 3 , {Tr′} 30 The large prominent values ​​of the series are easier to identify; in years without disastrous extreme precipitation records, the maximum value of Tr′ is 80 (occurred in 1978), and there is a 6% gap between this value and the judgment threshold.

[0171] 3. Verification and evaluation of prediction methods

[0172] 3.1 Prediction model and description

[0173] Formula (9) generated by using 30-year data from 1970 to 1999 is used as the calculation model, and the judgment index Tr′≥85 extracted from the 30-year data is used as the judgment basis; W1, W2, W3, W4, and W5 of any year other than the 1970-1999 period are used as input to calculate Tr′; when the calculated Tr′≥85, it is predicted that disastrous extreme precipitation will occur in the flood season of that year; when the calculated Tr′≤80, it is predicted that disastrous extreme precipitation will not occur in the flood season of that year; when 80<Tr′<85, the existing data cannot determine whether disastrous extreme precipitation will occur.

[0174]

[0175] Input item W1: seasonal average of the difference between the daily minimum temperature at Dujiangyan station and the daily minimum temperature at Wenjiang station from March to May.

[0176] Input item W2: seasonal average of the daily temperature range at Wenjiang Station from March to May.

[0177] Input item W3: The number of days with a drop in temperature of more than 2.5℃ compared with the maximum temperature of the previous day at Wenjiang Station from March to May.

[0178] Input item W4: Frequency value of the median temperature difference of the frequency sequence of the daily minimum temperature difference of the Pengzhou station-Wenjiang station combination from March to May.

[0179] Input item W5: Maximum frequency value of the frequency sequence of the daily minimum temperature difference of the Dujiangyan-Wenjiang station combination from March to May.

[0180] 3.2 Prediction Practice

[0181] The prediction practice sample period is 2000-2022. The W1, W2, W3, W4, and W5 values ​​of these 23 years are substituted into the Tr' calculation formula to calculate Tr'. Then, Tr'≥85 is used as the judgment index to judge whether there will be disastrous extreme precipitation of more than 200 mm / d from June to September in the five stations in the analyzed area, or whether there will be associated basin floods. The calculated Tr' for each predicted year in 23 years is shown in Figure 5 :

[0182] Figure 5 In the 2018 flood disaster disaster assessment, Tr′≥85 was still used as the indicator for determining disastrous extreme precipitation, and 5 predicted sample years exceeded the indicator in 23 years. The records of flood disasters in 2001, 2013, 2015, 2018, and 2020, the maximum daily precipitation from June to September at the five national stations of Dujiangyan, Wenjiang, Pidu District, Pengxian, and Chongzhou, and the model prediction statistics are shown in Table 2:

[0183] Table 2 Maximum precipitation (mm / d) and flood disaster records from June to September at the five stations in the forecast sample years satisfying Tr′≥85

[0184] Content / Time 2001 2013 2015 2018 2020 Tr′ value 142 159 152 181 131 Maximum daily precipitation 179mm / dPidu District 424mm / dDujiangyan 138mm / dPidu District 253mm / dPengzhou 240mm / dDujiangyan Flood Disaster Tuojiang River mainstream flood Severe flooding across the region none Severe flooding across the region Severe flooding across the region Is the prediction correct? correct correct mistake correct correct

[0185] As can be seen from Table 2, the 50-year basin-wide flood in the Jintang Tuojiang River in 2001, the three regional catastrophic floods in 2013, 2018 and 2020 (Chengdu flood control plan) can be correctly predicted in advance at the end of spring, with an advance time of >30 days; there will be one misjudgment in 2015.

[0186] 3.3 Prediction effect evaluation

[0187] 3.3.1 Evaluation of prediction effect using national station data

[0188] According to the data in Table 2, the hit rate POD1 based on the rainfall intensity threshold of R ≥ 200 mm / d is 100% and the false alarm rate FAR1 is 40%, and the hit rate POD2 for evaluating the occurrence of basin-wide floods in the analysis area is 100% and the false alarm rate FAR2 is 20%.

[0189] POD1 = Number of correct rainfall intensity predictions / (number of correct rainfall intensity predictions + number of missed rainfall intensity predictions) = 3 / (3 + 0) = 100 (%)

[0190] FAR1 = Number of false alarms of rainfall intensity / (number of correct rainfall intensity predictions + number of false alarms of rainfall intensity) = 2 / (3 + 2) = 40 (%)

[0191] POD2 = Number of correct flood predictions / (number of correct flood predictions + number of flood omissions) = 4 / (4 + 0) = 100 (%)

[0192] FAR2 = Number of false flood alarms / (Number of correct flood predictions + Number of false flood alarms) = 1 / (4+1) = 20 (%)

[0193] The purpose of using only national station data for effect evaluation is to connect with historical data. However, since the sample size is small, the conclusions are easily misunderstood. Therefore, after a large number of regional automatic stations are built, it is necessary to further use more detailed precipitation observation data for comparative checks.

[0194] 3.3.2 Using regional automatic weather station data to evaluate forecasting performance

[0195] During the forecast practice period, starting from 2010, we can further use the precipitation data from the five national stations in Dujiangyan, Wenjiang, Pidu District, Pengxian and Chongzhou, combined with the regional automatic stations, totaling 76 stations, to carry out forecast effect evaluation. The evaluation content includes two parts: "forecast result inspection" and "spatial distribution observation of heavy precipitation in the forecast sample year".

[0196] Forecast result check: When multiple regional heavy rainstorms occur repeatedly during the flood season, the cumulative number of heavy rainstorm stations will also increase. Therefore, the accumulated "number of heavy rainstorm stations" contains the expression of the special climate background that is prone to regional heavy rainstorms. Regional heavy rainstorms and extremely heavy rainstorms are the main causes of flood disasters. Therefore, if it is predicted that disastrous extreme precipitation will occur during the flood season of a certain year, the total number of stations with heavy rainstorms during the flood season of that year "Q Tr "It should be significantly higher than in normal years. The correctness of the prediction result can be directly determined by the total amount of heavy rain station Q Tr judge.

[0197] With reference to the construction progress of local regional automatic meteorological stations, the forecast results were checked using 76 stations with a total of 912 precipitation records ≥ 100 mm / d in 13 years. The data statistics are shown in Table 3.

[0198] Table 3 Statistics of stations with precipitation ≥ 100 mm / d from June to September 2010-2022 Unit: Station

[0199]

[0200] As shown in Table 3, in 2013, 2018, and 2020, when severe floods occurred, the total number of stations with heavy rains during the flood season was Q Tr They are 177%, 66%, and 236% higher than the average values ​​of 13a, respectively. TrThe abnormally large years correspond correctly in time to the flood years predicted by Tr′≥85, and we can draw the assessment conclusion that 3 extremely large flood years were predicted correctly and 1 extremely large flood year failed to be predicted in 13 years, with no omissions, that is, POD=100%, FAR=25%.

[0201] Spatial distribution of heavy precipitation in the predicted sample years: Observing the distribution of heavy precipitation in each flood season from 2010 to 2022 is helpful to intuitively judge the prediction effect of the four sample years with Tr′≥85. Figures 6(a)-(m) show the cumulative precipitation R≥100 mm / d from June to September for each of the 76 stations used in Table 3. 100 (mm), and R 100 / 100 is the color scale level. The horizontal and vertical axes in the figure are longitude and latitude respectively. The color scale is set from 0 to 18, and the difference between each level is 100mm of rainfall. The actual highest color scale value is 12.6, which appeared in S1072 in 2013. The total precipitation of the station from June to September under the condition of ≥100mm / d was as high as 1257mm. In Figure 6(a)-(m), in the sample years with Tr′≥85, the distribution trend of heavy precipitation in the flood season of 2013 and 2020 is very different from that in normal years. The distribution trend of heavy precipitation in 2018 is quite different from that in normal years. In these three years, the Tr′≥85 indicator can get correct early warning; the distribution trend of heavy precipitation in 2015 is very normal, which is indeed a false alarm sample.

[0202] In summary, the prediction method of the present invention has the advantages of strong pertinence and high accuracy. It can enrich the content of local seasonal-subseasonal prediction products and can also serve as an indirect reference for the early warning level of rainstorm processes. The practical effect also reflects to a certain extent that the integration of terrain factors for delicate small-scale zoning is beneficial to improving prediction skills. The prediction practice carried out using the Dujiangyan-Wenjiang prediction unit verified that the specific technical measures used therein, such as the extraction of interannual variation characteristics of heavy precipitation, the processing of heavy precipitation advance signal analysis factors, and the simulation operation method, have a certain degree of effectiveness. These analysis factors respond to the early internal conditions that may induce disastrous extreme precipitation in the prediction area. The prediction effect shows that there is only one misreporting in the 23a prediction test sample, and there is no omission. Therefore, the method of the present invention can relatively accurately predict whether there is a flood threat in the flood season of the year in the area, and the prediction conclusion is of great significance to local flood control and disaster reduction.

Claims

1. A sub-seasonal prediction method for extreme precipitation in the flood season based on the evolution characteristics of the temperature series, characterized in that: The following steps are involved: Step 1: Use the common trend and extreme value change characteristics of the time series of frequency data corresponding to the median value of the spatial difference of the daily temperature minimum to find the sensitive site combination, effective sampling period and evolution pattern characteristics of the temperature spatial difference data series associated with climate events, determine the predictable area and organize the rainfall data and temperature data of the historical period in the area; Step 2: Preprocess the rainfall data and temperature data of the historical period in the forecast area and extract the interannual variation characteristics of heavy precipitation; Step 3: Extract fitting factors using the March-May observation data sequence and its derivative sequence related to the interannual variation characteristics of heavy precipitation, generate a fitting curve with a specific similarity to the interannual variation of heavy precipitation, and establish a judgment threshold based on the rarity of rainfall intensity and flood hydrological records; use the obtained fitting curve and judgment threshold to establish a catastrophic extreme precipitation prediction model for the area to be predicted; Step 4: Input the analysis data of the observation elements from March to May of the forecast sample year, and obtain the forecast results of whether disastrous extreme precipitation will occur in the forecast area from June to September.

2. The method for sub-seasonal prediction of extreme precipitation in flood season based on the evolution characteristics of temperature series according to claim 1 is characterized in that: The step 1 comprises: The difference between the lowest temperatures of the two weather stations is obtained by using the daily lowest temperatures of the two weather stations, and the difference set {ΔT min }, count the frequency f(Xi) of each temperature difference Xi in the set, and arrange the corresponding frequency f(Xi) in the order of the temperature difference Xi, and get the frequency f(Xi) of the difference set ΔT min The frequency analysis sequence {Xi, f(Xi)} of the frequency analysis sequence {Xi, f(Xi)} is defined as the frequency median M e is half of the total frequency, and the cumulative frequency reaches M e At this time, the corresponding temperature difference Xi value is the median temperature difference MO. The correspondence between the year-on-year change characteristics of the temperature difference frequency corresponding to the median temperature difference MO and the climate events is used to determine the site combination, sampling period and evolution pattern of the data series that the subsequent analysis and modeling work depends on.

3. The method for sub-seasonal prediction of extreme precipitation in flood season based on the evolution characteristics of temperature series according to claim 2 is characterized in that: The step 2 comprises: Step 2.1: Precipitation data preprocessing Get the weather station S in the area to be predicted D Daily rainfall data from June to September in a certain Nx years in history. Count the number of days n when the rainfall from June to September is greater than or equal to the rainstorm threshold, and calculate the total rainfall value of these n rainy days. And the average rainfall intensity value Tr1 of these n rainfall days: Among them, Ri is the rainfall on a certain day when the rainfall is greater than or equal to the heavy rain threshold; For the weather station S in the area to be predicted W Similarly, calculate the average rainfall intensity Tr2 of the rainy days with rainfall greater than or equal to the rainstorm threshold from June to September in each year Nx; Then calculate the average rainfall intensity for each year Thus, the average rainfall intensity value sequence during the Nx years is formed. Step 2.2: Temperature data preprocessing 1) Minimum temperature difference item Use the weather station S from March to May every year in a certain Nx year D The daily minimum temperature minus the corresponding day's weather station S W The daily minimum temperature of the two stations is obtained by the difference of the daily minimum temperature of ΔT minDW ; Then calculate the difference ΔT between the daily minimum temperatures of the two stations from March to May each year minDW The seasonal average It is abbreviated as W1; 2) Daily temperature range Use the weather station S from March to May every year in the Nx years W Daily temperature difference, calculate the seasonal average of the daily temperature difference of the weather station from March to May each year Denoted as W2; 3) Temperature stability observation item Using Weather StationS W The highest temperature value T of the previous day in the period from March to May each year in the Nx years max1 , minus the highest temperature of the day T max2 Get the difference △T max ; Statistics of △T from March to May each year max Greater than the temperature difference threshold T x The number of times ℃ is recorded as W3; 4) Frequency sequence of spatial difference distribution of minimum temperature between stations The difference between the lowest temperatures between the two meteorological stations during March to May of each year in the Nx years is obtained by calculating the annual difference set ΔT min , statistical annual difference set ΔT min The frequency f(Xi) of each temperature difference Xi in the equation is calculated, and the corresponding frequency f(Xi) is arranged in the order of the temperature difference Xi, and the difference value set ΔT is obtained. min The frequency analysis sequence of {Xi, f(Xi)}; Define the frequency median M of the frequency analysis sequence {Xi, f(Xi)} e Half of the total frequency; the cumulative frequency reaches M e , the corresponding temperature difference Xi value at this time is the median temperature difference MO; Based on this, for the weather station S in the area to be predicted P With weather station S W Calculate the frequency value F of the median value of the frequency analysis sequence of the daily minimum temperature difference between the two stations from March to May each year PW , recorded as W4; For the weather station S in the area to be predicted D With weather station S W , calculate the maximum frequency value F of the frequency analysis sequence of the daily minimum temperature difference between the two stations from March to May each year MAX.DW , denoted as W5; Step 2.3: Exploitation The prominent large values ​​in correspond to the characteristics of the years with disastrous extreme precipitation, and the sample years in the years with disastrous extreme precipitation are marked; the Nx-year time series of the five temperature preprocessing data items are calculated respectively. The correlation coefficient of .

4. The method for sub-seasonal prediction of extreme precipitation in flood season based on the evolution characteristics of temperature series according to claim 3 is characterized in that: The step 3 comprises: Step 3.1: Extract the anomaly analysis variables of the preprocessed temperature data 1) Calculate the Nx-year average value of W1, denoted as W 1-Nx , calculated as follows: Among them, W 1i is the W1 value of the i-th year among Nx years; Calculate the deviation analysis variable of W1 value in any sample year, denoted as Z1, as follows: Z1=(W1-W 1-Nx ) / A Among them, A is the arithmetic comparison value; 2) Calculate the Nx-year average value of W2, recorded as W 2-Nx , calculated as follows: Among them, W 2i is the W2 value of the i-th year among Nx years; Calculate the deviation analysis variable of W2 value in any sample year, denoted as Z2, as follows: Z2=W2 / W 2-Nx 3) Calculate the Nx-year average value of W3, recorded as W 3-Nx , calculated as follows: Among them, W 3i is the W3 value of the i-th year among Nx years; Calculate the deviation analysis variable of W3 value in any sample year, denoted as Z3, as follows: <h2 style=";text-align:left;direction:ltr">Z3 = W3 / W<h2 style=";text-align:left;direction:ltr"> 3-Nx <h2 style=";text-align:left;direction:ltr"> -B〕 Among them, B is the offset; 4) Calculate the Nx-year average value of W4, recorded as W 4-Nx , calculated as follows: Among them, W 4i is the W4 value of the i-th year among Nx years; Calculate the anomaly analysis variable of the W4 value for any sample year, denoted as Z4, as follows: Z4=W4 / W 4-Nx 5) Calculate the Nx-year average value of W5, recorded as W 5-Nx , calculated as follows: Among them, W 5i is the W5 value of the i-th year among Nx years; Calculate the deviation analysis variable of W5 value in any sample year, denoted as Z5, as follows: W5=W5 / W 5-Nx Step 3.2: Determine the fitting formula Taking Z1~Z5 as fitting factors, the deviation change indicator Tr′ is comprehensively formed: Tr′=D*〔1 / Z5 a +Z3 b +2*Z2 c 〕 / 〔Z4–Z1〕 Among them, D is the multiplier coefficient, a, b and c are the influence weight indexes of fitting factors Z5, Z3 and Z2 respectively; Thus, the anomaly change indicator sequence {Tr′} for the Nx years is formed. Nx ; Step 3.3: Determine the fitting method By adjusting the impact weight indexes a, b and c, the deviation change indicator sequence {Tr′} Nx The average rainfall intensity series The goal is to achieve the maximum correlation coefficient. After obtaining the debugging sequence with the largest correlation coefficient, the same sequence is then obtained with the average value of the debugging sequence. The average value of is compared, the multiplication factor D is determined, and the final fitting result calculation formula is obtained, thereby obtaining the heavy rain prediction model; Step 3.4: Based on the anomaly change indicator series {Tr′} of Nx years Nx Establish indicators to determine whether disastrous extreme precipitation will occur in the disastrous precipitation analysis area from June to September; From the set {Tr′} Nx In the years when the element value is greater than the set threshold or when there is a precipitation record with rainfall greater than the rainstorm rainfall threshold, the smallest Tr' value is selected and the smallest Tr' value is used as a judgment indicator for whether there is disastrous extreme precipitation.

5. The method for sub-seasonal prediction of extreme precipitation in flood season based on the evolution characteristics of temperature series according to claim 4 is characterized in that: The step 3.3 specifically includes: Step 3.3.1: Initialize, set the influence weight index a=1, b=1, c=1; set the multiplier coefficient D=1; Step 3.3.2: Obtain the anomaly analysis variable sequence {Z1} for the Nx years based on the observed data Nx , {Z2} Nx , {Z3} Nx , {Z4} Nx and {Z5} Nx , and the sequence Step 3.3.3: Substitute the annual data of Z1 to Z5 with the fitting formula under the values ​​of a=1, b=1, c=1, and D=1 to obtain the Nx-year debugging sequence {Tr′} of Tr′ Nx～1 ; Step 3.3.4: Adjust the influence weight indexes a, b, and c in steps of ±0.1, and detect {Tr′} Nx～1 and The correlation coefficient is the same as the correlation coefficient of the original image. When the correlation coefficient reaches the maximum value, the debugging is completed. The fixed values ​​of a, b, and c give {Tr′} Nx～2 ; Step 3.3.5: First find The average value and {Tr′} Nx～2 The average value, ignoring the unit and looking at the value only, is The value of the average value divided by {Tr′} Nx～2 The average value of the multiplier coefficient D is obtained; Step 3.3.6: Set {Tr′} Nx～2 The element value in is expanded D times to obtain the final target sequence {Tr′} Nx-Z .

Images