Precipitation probability forecast method

Abstract

The present invention relates to the field of precipitation forecasting. In order to improve the accuracy of precipitation forecasting, a precipitation probability forecasting method is provided, including: 1. Selecting a numerical forecast result file corresponding to the time to be forecasted; 2. Extracting and calculating the precipitation in the forecast unit time; 3. Selecting a window of a certain size according to the deviation range of the precipitation falling area and the grid resolution of the numerical forecast product; 4. Calculating the weights of all grid points in the window respectively, and Gaussian weighting the precipitation in the unit time according to the weights; 5. Referring to the precipitation level standard or a custom precipitation threshold, calculating the precipitation probability according to the Gaussian weighted precipitation; 6. Calculating the weights of the forecast results of different starting times according to the time distance of the starting time relative to the forecast time, and weighting the precipitation probability to obtain the final precipitation probability forecast result; 7. Generating a precipitation probability forecast product of a specified threshold. The above method improves the accuracy of precipitation forecasting.

Description

Technical Field

[0001] The invention relates to the field of precipitation forecasting, and in particular to a precipitation probability forecasting method. Background Art

[0002] The Numerical Weather Prediction (NWP) model is the core of current weather forecasting. As long as the initial state and boundary conditions of the atmosphere are given, the future state of the atmosphere can be accurately calculated by solving the set of equations that describe the laws of atmospheric motion and change. However, due to the chaotic behavior of the numerical model system itself and people's limited understanding of the physical mechanism and external forcing of the weather and climate system, there are uncertainties and forecast errors in numerical weather forecasts. Traditional deterministic forecast products have relatively large forecast errors. Chaos theory shows that the evolution of the atmospheric state is very sensitive to small errors in the initial field. The extremely small errors in the initial field will gradually increase in the forecast. Practice has shown that the use of data assimilation technology can effectively improve the accuracy of numerical weather forecasts. Therefore, the current regional high-resolution rapid update cycle assimilation numerical forecast system has become the main direction of the development of numerical model technology. The current rapid update cycle assimilation system makes multiple (for example, 24) forecasts every day, with the longest forecast validity period being more than 24 hours. For a certain forecast moment in the future, there are multiple forecast results every day, and the forecast results will have certain differences. The development of probabilistic forecast products for the rapid update cycle assimilation numerical forecast system can effectively improve the forecast effect and continuously meet the public's growing demand for meteorological services. Summary of the invention

[0003] In order to improve the accuracy of precipitation forecast, the present invention provides a precipitation probability forecasting method.

[0004] The technical solution adopted by the present invention to solve the above problems is:

[0005] Precipitation probability forecasting methods include:

[0006] Step 1, select the numerical forecast result file corresponding to the time to be forecasted;

[0007] Step 2: extract and calculate the precipitation per unit time according to the numerical forecast result file;

[0008] Step 3: Select a window of a certain size according to the deviation range of the precipitation area and the grid resolution of the numerical forecast product, wherein the window has a central grid point;

[0009] Step 4: Calculate the weights of all grid points in the window respectively, and perform Gaussian weighting on the precipitation per unit time according to the weights;

[0010] Step 5: Calculate the precipitation probability according to the Gaussian weighted precipitation, referring to the precipitation level standard or the custom precipitation threshold;

[0011] Step 6: Calculate the weights of the forecast results at different onset times according to the time distance between the onset time and the forecast time, and weight the precipitation probability to obtain the final precipitation probability forecast result;

[0012] Step 7: Generate a precipitation probability forecast product with a specified threshold based on the precipitation probability forecast results.

[0013] Furthermore, the numerical forecast result file includes the start time, forecast time, cumulative precipitation RAINC generated by cumulus deep convection process, cumulative precipitation RAINNC generated by non-convective process and cumulative precipitation RAINSH generated by shallow convection process.

[0014] Furthermore, the step 2 is specifically as follows:

[0015] Step 21, extracting the cumulative precipitation RAINC generated by the cumulus deep convection process, the cumulative precipitation RAINNC generated by the non-convection process, and the cumulative precipitation RAINSH generated by the shallow convection process in the current numerical forecast result file, and summing them up to obtain the ground precipitation TP (time, lat, lon), where time represents time, lat represents latitude, and lon represents longitude;

[0016] Step 22, calculate the precipitation per unit time rain(time, lat, lon), rain(time, lat, lon) = TP(time, lat, lon) - TP(time-1, lat, lon).

[0017] Furthermore, the specific steps of calculating the weight in step 4 are:

[0018] Step 41: Calculate the initial weight W(x,y) of each grid in the window according to the Gaussian distribution function. x, y represent the horizontal and vertical distances of each grid point from the center of the window, respectively, and σ is the standard deviation of the normal distribution;

[0019] Step 42: Normalize the initial weight to obtain the weight w(x,y): w(x,y)=W(x,y) / sum(W(x,y)). Further, the weights of the forecast results at different starting times in step 6 are: min(t) represents the minimum value in the time distance series, α is a constant, t represents the time distance, and n represents the longest forecast time.

[0020] Furthermore, α is greater than 0 and less than half of the number of numerical prediction result files involved in the calculation.

[0021] Furthermore, the size of the window in step 3 is Where g is the window size, M is the maximum deviation distance of the precipitation area, and k is the horizontal resolution of the forecast product grid.

[0022] Compared with the prior art, the present invention has the beneficial effect that in the precipitation forecasting process, the forecast errors of numerical weather forecasting in time and space are fully taken into account, thereby improving the final precipitation forecast accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 This is a flow chart of the precipitation probability forecast method;

[0024] Figure 2 Schematic diagram of different grid weights;

[0025] Figure 3 This is a schematic diagram of the weights of forecast results at different onset times relative to different time distances to a certain moment in the future. DETAILED DESCRIPTION

[0026] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0027] like Figure 1 As shown in the figure, the precipitation probability forecasting method includes:

[0028] Step 1, select the numerical forecast result file corresponding to the forecast time; the numerical forecast result file includes information such as the start time, forecast time, cumulative precipitation RAINC generated by the cumulus deep convection process, cumulative precipitation RAINNC generated by the non-convective process, and cumulative precipitation RAINSH generated by the shallow convection process.

[0029] Step 2, extract and calculate the precipitation in the forecast unit time according to the numerical forecast result file; specifically: Step 21, extract the cumulative precipitation RAINC generated by the cumulus deep convection process, the cumulative precipitation RAINNC generated by the non-convective process, and the cumulative precipitation RAINSH generated by the shallow convection process in the current numerical forecast result file, and sum them to obtain the ground precipitation TP (time, lat, lon), where time represents time, lat represents latitude, and lon represents longitude;

[0030] Step 22, calculate the precipitation per unit time rain(time, lat, lon), rain(time, lat, lon) = TP(time, lat, lon) - TP(time-1, lat, lon).

[0031] Step 3: Select a window of a certain size according to the deviation range of the precipitation area and the grid resolution of the numerical forecast product, wherein the window has a central grid point; the size of the window is In the formula, g is the window size. In order to make the window have a central grid point, g should be an odd number, M is the maximum deviation distance of the precipitation area, and k is the horizontal resolution of the forecast product grid. The larger the window, the more forecast data of adjacent grid points participating in spatial weighting, and the smoother the forecast result.

[0032] Step 4: Calculate the weights of all grid points in the window respectively, and perform Gaussian weighting on the precipitation per unit time according to the weights; the specific steps for calculating the weights are as follows:

[0033] Step 41: Calculate the initial weight W(x,y) of each grid in the window according to the Gaussian distribution function. x, y represent the horizontal and vertical distances of each grid point from the center of the window, respectively, and σ is the standard deviation of the normal distribution;

[0034] Step 42: Normalize the initial weights to obtain w(x,y): w(x,y)=W(x,y) / sum(W(x,y)).

[0035] Step 5: Calculate the precipitation probability according to the Gaussian weighted precipitation, referring to the precipitation level standard or the custom precipitation threshold;

[0036] Step 6: Calculate the weights of the forecast results at different onset times according to the time distance between the onset time and the forecast time, and weight the precipitation probability to obtain the final precipitation probability forecast result; specifically, the weights of the forecast results at different onset times are min(t) represents the minimum value in the time distance sequence, t represents the time distance, n represents the longest forecast time, and α is a constant greater than 0 and less than half of the number of numerical forecast result files involved in the calculation.

[0037] Example

[0038] Step 1. Select the numerical forecast result file corresponding to the forecast time, for example: wrfout_d01_2021-12-20_00_00_00, wrfout identifies the result file of the weather forecast model (Weather Research and Forecast model, WRF), d01 identifies the first layer area of ​​the model, 2021-12-20_00_00_00 is the start time of the model, and the file is in the standard NetCDF (network Common Data Form) network common data format. Use the extract_times() function in Python's wrf-python to extract the forecast time. The forecast time is a time series with 1-hour intervals. The length of the sequence determines the maximum time of the forecast.

[0039] Assuming that the rapid update circulation assimilation model system reports 24 times a day, and the longest forecast time for each report is 30 hours, taking the forecast of hourly precipitation probability in the next 15 hours on 2021-12-20_00_00_00 as an example, if there is no missing data, the reporting time of the model forecast result file involved in the calculation of precipitation probability should be from 2021-12-18_19_00_00 to 2021-12-20_00_00_00, totaling 30 files.

[0040] Step 2. According to the output definition of the WRF model, the ground precipitation is the sum of the three variables: RAINC, the cumulative precipitation generated by the deep cumulus convection process, RAINNC, the cumulative precipitation generated by the non-convective process, and RAINSH, the cumulative precipitation generated by the shallow convection process. Use the getvar function in wrf-python to extract the values ​​of these three variables and sum them. The calculation result is a three-dimensional (time, latitude, longitude) array, recorded as TP(time,lat,lon). If the longest forecast time is 30 hours, the size of the time dimension time is 31. The physical meaning of the array in the time dimension is the cumulative precipitation relative to the start time. Subtract the previous time data from the current time data, rain(time,lat,lon)=TP(time,lat,lon)-TP(time-1,lat,lon), and you can get the precipitation at time in the past hour.

[0041] Step 3: Considering the certain error of precipitation in horizontal space, a window of a certain size is selected. The size of the window is generally related to the horizontal resolution of the model forecast product and the forecast performance of the model itself. The window has a central grid point and the size of the window is Where g is the window size, M is the maximum deviation distance of the precipitation area, and k is the horizontal resolution of the forecast product grid.

[0042] Step 4: According to the two-dimensional Gaussian distribution function Calculate the weights of all grid points in the window respectively, where x and y represent the horizontal distance and vertical distance from the center of the window respectively, and σ is the standard deviation of the normal distribution. In order to make the sum of the weights of all grid points equal to 1, use w(x,y)=W(x,y) / sum(W(x,y)) for normalization. w(x,y) is the weight of each grid point in the given window. The weight diagram of different grid points is shown in Figure 2 As shown. Use this weight coefficient to weight the hourly precipitation forecast results. g (time,lat,lon)=rain(time,lat,lon)*w(x,y), * is the cross-correlation operation.

[0043] Step 5: Select the precipitation threshold. Refer to the 1-hour precipitation level standard. The threshold can be 0.1mm, 1.6mm, 7mm, 15mm, 40mm, or 50mm. 0.1mm is used as the threshold for judging sunny or rainy days. If the precipitation is greater than or equal to 0.1mm, it is a rainy day. If it is less than 0.1mm, it is a sunny day. Taking 0.1mm as an example, judge each grid point of each hour of Gaussian weighted hourly precipitation. If it is greater than or equal to 0.1mm, it is recorded as 1, and if it is less than 0.1mm, it is recorded as 0. Convert the deterministic precipitation forecast result into a probabilistic forecast result. rain represents the precipitation, and threshold represents the selected threshold.

[0044] Step 6: Calculate the weights of forecast results at different starting times according to the time distance between the forecast product's starting time and the forecast moment, and weight the precipitation probability to obtain the final precipitation probability forecast result. For example, if 2021-12-20_00_00_00 is forecasted for the next hour, the time distances of the 30 starting times from 2021-12-18_19_00_00 to 2021-12-20_00_00_00 relative to the forecast moment are 30, 29, ..., 2, and 1, respectively, recorded as t 30 ,t 29 , ..., t2, t1, then the weight of the forecast product with time distance t is min(t) represents the minimum value in the time distance sequence, t represents the time distance, and n represents the longest forecast validity period; α is a constant. The larger its value is, the closer the weights of the forecast results at different onset times are. The smaller its value is, the greater the difference in the weights of the forecast results at different onset times is. Because the fast update cycle assimilation numerical system used performs 4 cold starts every day, the more cold starts, the closer the weights of the forecast results at different onset times are, the better. In this embodiment, there are 30 forecast result files involved in the calculation, 30÷2÷4≈5, so α is 5 in this embodiment. The weight results of different time distances are as follows: Figure 3shown.

[0045] The weight coefficient is used to weight the forecast results at different starting times, and the probability of precipitation forecast at the specified threshold at the future time is calculated as P1(t i ,lat,lon) is the precipitation probability calculated in step 5 after considering the spatial factors, (t i ,lat,lon) is the time distance t i any point in space.

[0046] Step seven: generate persistent data in the corresponding data format for the probability forecast results for a period of time in the future to form a probability forecast product.

[0047] The forecasting method of the present application takes into account the forecast errors of numerical forecasts in space and time, assigns different weights to forecast data at different spatial grid points and at different reporting times, gives full play to the advantages of the rapid update cycle assimilation system in short-term forecasts, and improves the performance of numerical precipitation forecasts. The probabilistic forecast product provided by the present invention is closer to actual precipitation.

Claims

1. A precipitation probability forecasting method, characterized in that: include: Step 1, select the numerical forecast result file corresponding to the time to be forecasted; Step 2: extract and calculate the precipitation per unit time according to the numerical forecast result file; Step 3: Select a window of a certain size according to the deviation range of the precipitation area and the grid resolution of the numerical forecast product, wherein the window has a central grid point; Step 4: Calculate the weights of all grid points in the window respectively, and perform Gaussian weighting on the precipitation per unit time according to the weights; Step 5: Calculate the precipitation probability according to the Gaussian weighted precipitation, referring to the precipitation level standard or the custom precipitation threshold; Step 6: Calculate the weights of the forecast results of different onset times according to the time distance between the onset time and the forecast time, and weight the precipitation probability to obtain the final precipitation probability forecast result; Step 7: Generate a precipitation probability forecast product with a specified threshold based on the precipitation probability forecast results.

2. The precipitation probability forecasting method according to claim 1, characterized in that: The numerical forecast result file includes the start time, forecast time, the cumulative precipitation RAINC generated by the cumulus deep convection process, the cumulative precipitation RAINNC generated by the non-convective process, and the cumulative precipitation RAINSH generated by the shallow convection process.

3. The precipitation probability forecasting method according to claim 2, characterized in that: The step 2 is specifically as follows: Step 21, extracting the cumulative precipitation RAINC generated by the cumulus deep convection process, the cumulative precipitation RAINNC generated by the non-convection process, and the cumulative precipitation RAINSH generated by the shallow convection process in the current numerical forecast result file, and summing them up to obtain the ground precipitation TP (time, lat, lon), where time represents time, lat represents latitude, and lon represents longitude; Step 22: Calculate the precipitation per unit time: rain(time, lat, lon), rain(time, lat, lon)=TP(time, lat, lon)-TP(time-1, lat, lon).

4. The precipitation probability forecasting method according to claim 1, characterized in that: The specific steps of calculating the weight in step 4 are: Step 41: Calculate the initial weight W(x,y) of each grid in the window according to the Gaussian distribution function. x, y represent the horizontal and vertical distances of each grid point from the center of the window, respectively, and σ is the standard deviation of the normal distribution; Step 42: Normalize the initial weight to obtain the weight w(x,y): w(x,y)=W(x,y) / sum(W(x,y)).

5. The precipitation probability forecasting method according to claim 1, characterized in that: The weights of the forecast results at different starting times in step 6 are: min(t) represents the minimum value in the time distance series, α is a constant, t represents the time distance, and n represents the longest forecast time.

6. The precipitation probability forecasting method according to claim 5, characterized in that: α is greater than 0 and less than half of the number of numerical prediction result files involved in the calculation.

7. The precipitation probability forecasting method according to any one of claims 1 to 6, characterized in that: The size of the window in step 3 is Where g is the window size, M is the maximum deviation distance of the precipitation area, and k is the horizontal resolution of the forecast product grid.

Images