A method for adaptive correction of climatological state of numerical weather forecast products

Abstract

The present invention provides a method for adaptively correcting the climate state of a numerical weather forecast product, comprising the following steps: conducting a climate state time scale test on a set of historical observation data time series and a set of historical model data time series of weather variables to determine an optimal sliding average window length; using the optimal sliding average window length to extract the observed HMA climate state and the model HMA climate state, correcting the model data of the weather variable, and obtaining the forecast of the weather variable. The method for adaptively correcting the climate state of a numerical weather forecast product provided by the present invention can reduce the error of numerical model results, extend the effective forecast time, and improve the forecast accuracy. The present invention particularly improves the forecasting capabilities in the short to medium term - extended period, complex terrain areas, heavy precipitation weather systems, etc.

Description

Technical Field

[0001] The invention belongs to the technical field of numerical weather forecasting, and in particular relates to a method for adaptively correcting the climatic state of a numerical weather forecast product. Background Art

[0002] The development of numerical weather forecasting has achieved remarkable results, especially in the past decade or so. With the more precise model dynamic framework, more reasonable parameterization scheme, more intensive observation data, more perfect assimilation technology, as well as the advancement of computers, large-scale computing power and improved model resolution, the effective forecast time of numerical weather forecast has been continuously extended, and the forecast accuracy has continued to improve. Numerical forecasting has become the most important means of weather forecasting.

[0003] The development of numerical weather forecasting is to pursue the improvement of forecast accuracy, and the key issue is how to reduce forecast errors. Adopting more accurate and reliable observation instruments, more advanced weather radars and weather satellites to obtain observation data with a wider range and higher resolution, etc., to provide more reliable initial values, minimize the initial error of the model. Adopting more complete non-hydrostatic fully compressible equations and high-precision calculation methods to build the model dynamic framework, optimize the model physical process parameterization scheme, make the model's description of the dynamics and physical processes of atmospheric motion closer to the actual situation, and improve the model resolution, etc., can effectively reduce the error of the numerical model.

[0004] However, whether it is the description of the initial field or the model dynamic process or physical process, it is only an approximation of the infinitely complex process of the actual atmosphere. The improvement of the calculation scheme can only improve the degree of approximation to reality to a certain extent and reduce the model prediction error, but it cannot completely eliminate the error. Therefore, the study of effective correction and correction methods for numerical prediction errors has important scientific significance and practical application value.

[0005] The existing methods for correcting numerical forecast errors are mainly divided into four categories: empirical methods, variational correction methods, Kalman filtering and ensemble Kalman filtering methods, and dynamic-statistical correction methods. However, the existing methods for correcting various numerical forecast errors still have great limitations in reducing numerical forecast errors and improving numerical forecast accuracy, and cannot meet people's increasing demand for numerical forecast accuracy. Summary of the invention

[0006] In view of the defects of the prior art, the present invention provides a method for adaptively correcting the climatological state of numerical weather forecast products, which can effectively solve the above problems.

[0007] The technical solution adopted by the present invention is as follows:

[0008] The present invention provides a method for adaptively correcting the climatological state of a numerical weather forecast product, comprising the following steps:

[0009] Step 1: Identify the weather variables that need to be forecasted and corrected ;

[0010] Step 2: Get weather variables A collection of historical observation data time series and historical pattern data time series collection ; Wherein, the historical observation data is the weather variables The historical model data are the weather variables predicted by numerical weather forecast products. The pattern data;

[0011] Step 3: Weather variables A collection of historical observation data time series and historical pattern data time series collection Conduct climatological time scale experiments to determine the optimal sliding average window length ;

[0012] Step 4: Make corrections:

[0013] Step 4.1: Obtain the latest weather variables predicted by the numerical weather forecast product to be revised Pattern data ;

[0014] Step 4.2, in the weather variables A collection of historical observation data time series In the example, the sliding average optimal window length is used. As the sliding window length, the Hull moving average HMA climate state is extracted to obtain the observed HMA climate state. ;

[0015] In the historical mode data time series collection In the example, the sliding average optimal window length is used. As the sliding window length, the Hull moving average HMA climate state is extracted to obtain the model HMA climate state ;

[0016] Step 4.3, using formula (1), the corrected weather variables Pattern data Correction to obtain weather variables The forecast amount :

[0017] (1);

[0018] This completes the correction process of the model data.

[0019] Preferably, the weather variable , including geopotential height, temperature, wind vector and specific humidity.

[0020] Preferably, the historical observation data is specifically the atmospheric reanalysis data ERA5 of the Weather Forecast Center for weather variables observation data, and preprocessed historical observation data.

[0021] Preferably, the pretreatment specifically includes:

[0022] The atmospheric reanalysis data ERA5 is subjected to latitudinal storage transposition and data interpolation processing to obtain the processed atmospheric reanalysis data ERA5, so that the processed atmospheric reanalysis data ERA5 is consistent with the weather variables to be corrected. Pattern data , with the same resolution and grid distribution.

[0023] Preferably, step 3 is specifically:

[0024] Step 3.1, set the sliding average window length ;

[0025] Step 3.2: In the historical observation data time series collection In the example, the sliding average window length is used. , extract the Hull moving average HMA climate state, and obtain the observed HMA climate state ;

[0026] In the historical mode data time series collection In the example, the sliding average window length is used. , extract the Hull moving average HMA climate state, and obtain the model HMA climate state ;

[0027] Step 3.3, let the sliding average window length Take separately , and ,in, , and is a selectable value for the pre-set climatological averaging time scale;

[0028] Using the method in step 3.2, we can obtain the observed HMA climate states. , observe HMA climate state , observe HMA climate state 、HMA climate state model 、HMA climate state model and HMA climate state ;

[0029] Calculate the observed HMA climate state and HMA climate state The root mean square error between ;

[0030] Calculate the observed HMA climate state and HMA climate state The root mean square error between ;

[0031] Calculate the observed HMA climate state and HMA climate state The root mean square error between ;

[0032] Root mean square error , and Compare and determine the sliding average window length with the smallest root mean square error , which is the optimal sliding average window length .

[0033] Preferably, the historical observation data time series collection Method for extracting Hull Moving Average (HMA) climate state and historical model data time series collection The method for extracting the Hull Moving Average HMA climate state is the same.

[0034] Preferably, step 3.2 is specifically as follows:

[0035] Step 3.2.1: Collect the historical observation data time series and historical pattern data time series collection Unified representation: time series set ;in, Represents a collection of time series The amount of data in Represents the data of each year arranged in chronological order. If it is a set , then it is observation data; if it is a set , then it is the mode data;

[0036] Step 3.2.2, in the time series collection In the example, the sliding average window length is used. Perform mobile convolution calculation to obtain the convolution time series ;in, Represents the sliding average window length When , the number of convolutions obtained;

[0037] In the time series collection In the example, the sliding average window length is used. Perform mobile convolution calculation to obtain the convolution time series ;in, Represents the sliding average window length When , the number of convolutions obtained;

[0038] Step 3.2.3, obtain a new time series: ;

[0039] For new time series , using sliding average window length Perform mobile convolution calculation to obtain a time series set Hull Moving Average HMA Climate State , the formula is:

[0040] (2);

[0041] Therefore, in the sliding average window length When the historical observation data time series set The Hull moving average HMA climate state is expressed as , which is the observed HMA climate state ; For the historical model data time series collection The Hull moving average HMA climate state is expressed as , which is the climate state of the model HMA .

[0042] Preferably, in step 3.2.2, in the time series set In the example, the sliding average window length is used. Perform mobile convolution calculation to obtain the convolution time series , specifically:

[0043] Step 3.2.2.1, when the length is When the sliding average window moves to a certain position, let its corresponding time series be: Time series ;in: t represents the year;

[0044] Step 3.2.2.2, use formula (3) to get the time series in Filter function for year The expression of ;

[0045] (3);

[0046] Step 3.2.2.3, use formula (4) to get the time series In the window length Convolution The expression is:

[0047] (4);

[0048] Step 3.2.2.4: Convert the time series And substitute formula (3) into formula (4) to get the convolution The detailed expression of:

[0049] (5);

[0050] By using formula (5), we can calculate the time series In the window length Convolution .

[0051] The method for adaptively correcting the climatological state of a numerical weather forecast product provided by the present invention has the following advantages:

[0052] The present invention provides a method for adaptively correcting the climatological state of numerical weather forecast products, which can reduce the error of numerical model results, extend the effective forecast time, and improve the forecast accuracy. The present invention particularly improves the forecasting capabilities in the short to medium term, extended period, complex terrain areas, and heavy precipitation weather systems. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1 A schematic flow chart of a method for adaptively correcting climatic states of numerical weather forecast products provided by the present invention. DETAILED DESCRIPTION

[0054] In order to make the technical problems, technical solutions and beneficial effects solved by the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and 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.

[0055] See also Figure 1 The present invention provides a method for adaptively correcting the climatological state of numerical weather forecast products, comprising the following steps:

[0056] Step 1: Identify the weather variables that need to be forecasted and corrected ;

[0057] In the present invention, weather variables , including but not limited to geopotential height, temperature, wind vector, and specific humidity.

[0058] Step 2: Get weather variables A collection of historical observation data time series and historical pattern data time series collection ; Wherein, the historical observation data is the weather variables The historical model data are the weather variables predicted by numerical weather forecast products. The pattern data;

[0059] Historical observation data, specifically the atmospheric reanalysis data ERA5 of the Weather Forecast Center for weather variables The preprocessing is as follows: the atmospheric reanalysis data ERA5 is subjected to latitudinal storage transposition and data interpolation processing to obtain the processed atmospheric reanalysis data ERA5, and the processed atmospheric reanalysis data ERA5 is made consistent with the weather variables to be corrected. Pattern data , with the same resolution and grid distribution.

[0060] For example, weather variables A collection of historical observation data time series , which is the fifth generation of global atmospheric reanalysis data ERA5 of the European Centre for Medium-Range Weather Forecasts (ECMWF) in the past 30 years. ERA5 reanalysis data provides isobaric grid point data with a 1-hour interval and a horizontal resolution of 0.125° from January 1940 to the last 5 days. At the same time, because it assimilates a large number of multi-source observations such as ground observation data, sounding data, and satellite remote sensing data, it is considered to be the best quality global grid reanalysis data at present.

[0061] Weather variables A collection of historical pattern data time series , which are the model data predicted by the model numerical weather prediction products in the past 30 years;

[0062] In the present invention, the weather variables predicted by the numerical weather forecast product to be revised Pattern data , which is the model data that needs to be corrected.

[0063] For example, if the current year is 2024, get the collection of 2023 and previous years and collection , thereby correcting the current model data for 2024. Assuming that the current year is 2025, the collection of 2024 and previous years is obtained and collection , thereby revising the current model data for 2025. In other words, the set and collection , and is continuously updated over time to ensure the accuracy of the forecast.

[0064] Step 3: Weather variables A collection of historical observation data time series and historical pattern data time series collection Conduct climatological time scale experiments to determine the optimal sliding average window length ;

[0065] Specifically, WMO (World Meteorological Organization) defines climate as the 30-year average change trend, but the definition of climate state in numerical forecast correction is not necessarily the real "climate", but can be a basic state that can better serve the model error correction. For the statistical sliding average, the larger the calculation scale (window), the smoother the average value and the more serious the lag. The smaller the scale (window), the more high-frequency useful information the average value retains and the smaller the lag, but the complexity of the algorithm implementation increases significantly. Therefore, how to determine the sliding average window length that meets the accuracy requirements is a difficult problem that needs to be solved.

[0066] The present invention adopts the Hull Moving Average (HMA) to construct a new and more effective HMA climate state, filters out high-frequency disturbances in the time series, retains useful low-frequency trends, and minimizes the "lag" defect commonly found in moving averages, thereby achieving the effect of climate state adaptation and approaching the true climate variability, making the obtained observed climate state and model climate state closer to reality.

[0067] Step 3 is as follows:

[0068] Step 3.1, set the sliding average window length ;

[0069] Step 3.2: In the historical observation data time series collection In the example, the sliding average window length is used. , extract the Hull moving average HMA climate state, and obtain the observed HMA climate state ;

[0070] In the historical mode data time series collection In the example, the sliding average window length is used. , extract the Hull moving average HMA climate state, and obtain the model HMA climate state ;

[0071] In this step, the historical observation data time series set Method for extracting Hull Moving Average (HMA) climate state and historical model data time series collection The method for extracting the Hull Moving Average HMA climate state is the same; the specific extraction method is:

[0072] Step 3.2.1: Collect the historical observation data time series and historical pattern data time series collection Unified representation: time series set ;in, Represents a collection of time series The amount of data in Represents the data of each year arranged in chronological order. If it is a set , then it is observation data; if it is a set , then it is the mode data;

[0073] Step 3.2.2, in the time series collection In the example, the sliding average window length is used. Perform mobile convolution calculation to obtain the convolution time series ;in, Represents the sliding average window length When , the number of convolutions obtained;

[0074] In the time series collection In the example, the sliding average window length is used. Perform mobile convolution calculation to obtain the convolution time series ;in, Represents the sliding average window length When , the number of convolutions obtained; Represents rounding down;

[0075] In this step, in the time series set In the example, the sliding average window length is used. Perform mobile convolution calculation to obtain the convolution time series , specifically:

[0076] Step 3.2.2.1, when the length is When the sliding average window moves to a certain position, let its corresponding time series be: Time series ;in: t represents the year;

[0077] Step 3.2.2.2, use formula (3) to get the time series in Filter function for year The expression of ;

[0078] (3);

[0079] Step 3.2.2.3, use formula (4) to get the time series In the window length Convolution The expression is:

[0080] (4);

[0081] Step 3.2.2.4: Convert the time series And substitute formula (3) into formula (4) to get the convolution The detailed expression of:

[0082] (5);

[0083] By using formula (5), we can calculate the time series In the window length Convolution .

[0084] Step 3.2.3, obtain a new time series: ;

[0085] For new time series , using sliding average window length Perform mobile convolution calculation to obtain a time series set Hull Moving Average HMA Climate State , the formula is:

[0086] (2);

[0087] The Hull Moving Average HMA climate state obtained by the above method It can be very stable when climate fluctuations are not obvious, and can capture such changes in time when climate trends change, while ensuring sufficient sensitivity to climate fluctuations and minimizing response lag.

[0088] Therefore, in the sliding average window length When the historical observation data time series set The Hull moving average HMA climate state is expressed as , which is the observed HMA climate state ; For the historical model data time series collection The Hull moving average HMA climate state is expressed as , which is the climate state of the model HMA .

[0089] Step 3.3, let the sliding average window length Take separately , and ,in, , and is a selectable value for the pre-set climatological averaging time scale;

[0090] In practical applications, usually, =30 years, = 20 years, =15. Different values ​​of the climate state average time scale directly affect the accuracy of the climate state adaptive correction. Therefore, the present invention compares the accuracy of common climate state average time scales. In the current scenario, the climate state average time scale when the model HMA climate state is closest to the observed HMA climate state is obtained as the sliding average window length, thereby improving the accuracy of the climate state adaptive correction.

[0091] Using the method in step 3.2, we can obtain the observed HMA climate states. , observe HMA climate state , observe HMA climate state 、HMA climate state model 、HMA climate state model and HMA climate state ;

[0092] Calculate the observed HMA climate state and HMA climate state The root mean square error between ;

[0093] Calculate the observed HMA climate state and HMA climate state The root mean square error between ;

[0094] Calculate the observed HMA climate state and HMA climate state The root mean square error between ;

[0095] Root mean square error , and Compare and determine the sliding average window length with the smallest root mean square error , which is the optimal sliding average window length , thus achieving a balance between effectiveness and retaining small climate changes.

[0096] Step 4: Make corrections:

[0097] Step 4.1: Obtain the latest weather variables predicted by the numerical weather forecast product to be revised Pattern data ;

[0098] Step 4.2, in the weather variables A collection of historical observation data time series In the example, the optimal window length is adopted by sliding average As the sliding window length, the Hull moving average HMA climate state is extracted to obtain the observed HMA climate state. ;

[0099] In the historical mode data time series collection In the example, the sliding average optimal window length is used. As the sliding window length, the Hull moving average HMA climate state is extracted to obtain the model HMA climate state ;

[0100] Step 4.3, using formula (1), the corrected weather variables Pattern data Correction to obtain weather variables The forecast amount :

[0101] (1);

[0102] This completes the correction process of the model data.

[0103] Specifically, for weather forecasting, the climate field can be considered to be known and unchanged. Therefore, the present invention makes full use of historical observation data to obtain the true observed HMA climate state of the atmosphere. Moreover, due to the adoption of the Hull moving average method, a more accurate and realistic observed HMA climate state can be obtained. ;

[0104] Furthermore, in order to accurately extract the model error in the model data, the present invention analyzes the historical model data and uses the Hull moving average method to extract the model HMA climate state that can accurately reflect the model error. ; Therefore, in the weather variables to be corrected Pattern data In the first step, the HMA climate state model with response model errors is removed. , combined with observations of HMA climate states , that is, to obtain accurate forecast , specifically the forecast amount revised using ERA5 reanalysis data and historical model data.

[0105] The present invention provides a method for adaptively correcting the climatological state of numerical weather forecast products, which has the following characteristics:

[0106] Historical observation data is the most objective reflection of real atmospheric movement that can be obtained at present. At present, historical observation data of more than 30 years have been accumulated. Through the Hull moving average HMA climate state extraction method of the present invention, climate states that can objectively reflect the periodic changes of the atmosphere at various time scales can be extracted from a large amount of historical observation data. The present invention calculates the climate state of different time windows through HMA, compares the relative stability and time variability of climate states at different time scales, and obtains the optimal sliding average window length that can balance the algorithm complexity and retain small climate changes; the optimal sliding average window of this length is used to extract accurate observed HMA climate states. and HMA climate state , thereby ensuring that the weather variables obtained after the final correction The forecast amount accuracy.

[0107] The present invention provides a method for adaptively correcting the climate state of numerical weather forecast products. It selects different average time scales to perform HMA climate average state extraction and sliding average calculation window experiments, and selects the best sliding average window. The best sliding average window has the following characteristics: it can be relatively balanced in terms of retaining high-frequency key information of climate evolution and reducing hysteresis. Moreover, with this optimal time scale, the Hull moving average (HMA) is used to determine the climate state of historical observation data and historical model data, and the climate state that can accurately represent the relative stability of the climate and reflect the key evolution information of climate events is obtained through error analysis. The model results to be corrected are further corrected to obtain the weather variable The forecast amount .

[0108] The present invention provides a method for adaptively correcting the climatological state of numerical weather forecast products, which can reduce the error of numerical model results, extend the effective forecast time, and improve the forecast accuracy. The present invention particularly improves the forecasting capabilities in the short to medium term, extended period, complex terrain areas, and heavy precipitation weather systems.

[0109] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principle of the present invention. These improvements and modifications should also be considered as the scope of protection of the present invention.

Claims

1. A method for adaptively correcting the climatological state of numerical weather forecast products, characterized in that: The following steps are involved: Step 1: Identify the weather variables that need to be forecasted and corrected ; Step 2: Get weather variables A collection of historical observation data time series and historical pattern data time series collection ; Wherein, the historical observation data is the weather variables The historical model data are the weather variables predicted by numerical weather forecast products. The pattern data; Step 3: Weather variables A collection of historical observation data time series and historical pattern data time series collection Conduct climatological time scale experiments to determine the optimal sliding average window length ; Step 4: Make corrections: Step 4.1: Obtain the latest weather variables predicted by the numerical weather forecast product to be revised Pattern data ; Step 4.2, in the weather variables A collection of historical observation data time series In the example, the sliding average optimal window length is used. As the sliding window length, the Hull moving average HMA climate state is extracted to obtain the observed HMA climate state. ; In the historical mode data time series collection In the example, the sliding average optimal window length is used. As the sliding window length, the Hull moving average HMA climate state is extracted to obtain the model HMA climate state ; Step 4.3, using formula (1), the corrected weather variables Pattern data Correction to obtain weather variables The forecast amount : （1）； This completes the correction process of the model data.

2. A method for adaptively correcting the climatological state of numerical weather forecast products according to claim 1, characterized in that: The weather variables , including geopotential height, temperature, wind vector and specific humidity.

3. A method for adaptively correcting the climatological state of numerical weather forecast products according to claim 1, characterized in that: The historical observation data are specifically the atmospheric reanalysis data ERA5 of the Weather Forecast Center for weather variables. observation data, and preprocessed historical observation data.

4. A method for adaptively correcting the climatological state of numerical weather forecast products according to claim 3, characterized in that: The pre-processing is specifically as follows: The atmospheric reanalysis data ERA5 is subjected to latitudinal storage transposition and data interpolation processing to obtain the processed atmospheric reanalysis data ERA5, so that the processed atmospheric reanalysis data ERA5 is consistent with the weather variables to be corrected. Pattern data , with the same resolution and grid distribution.

5. A method for adaptively correcting the climatological state of numerical weather forecast products according to claim 1, characterized in that: Step 3 is as follows: Step 3.1, set the sliding average window length ; Step 3.2: In the historical observation data time series collection In the example, the sliding average window length is used. , extract the Hull moving average HMA climate state, and obtain the observed HMA climate state ; In the historical mode data time series collection In the example, the sliding average window length is used. , extract the Hull moving average HMA climate state, and obtain the model HMA climate state ; Step 3.3, let the sliding average window length Take separately , and ,in, , and is a selectable value for the pre-set climatological averaging time scale; Using the method in step 3.2, we can obtain the observed HMA climate states. , observe HMA climate state , observe HMA climate state 、HMA climate state model 、HMA climate state model and HMA climate state ; Calculate the observed HMA climate state and HMA climate state The root mean square error between ; Calculate the observed HMA climate state and HMA climate state The root mean square error between ; Calculate the observed HMA climate state and HMA climate state The root mean square error between ; Root mean square error , and Compare and determine the sliding average window length with the smallest root mean square error , which is the optimal sliding average window length .

6. A method for adaptively correcting the climatological state of numerical weather forecast products according to claim 5, characterized in that: For the historical observation data time series collection Method for extracting Hull Moving Average (HMA) climate state and historical model data time series collection The method for extracting the Hull Moving Average HMA climate state is the same.

7. A method for adaptively correcting the climatological state of numerical weather forecast products according to claim 6, characterized in that: Step 3.2 is as follows: Step 3.2.1: Collect the historical observation data time series and historical pattern data time series collection Unified representation: time series set ;in, Represents a collection of time series The amount of data in Represents the data of each year arranged in chronological order. If it is a set , then it is observation data; if it is a set , then it is the mode data; Step 3.2.2, in the time series collection In the example, the sliding average window length is used. Perform mobile convolution calculation to obtain the convolution time series ;in, Represents the sliding average window length When , the number of convolutions obtained; In the time series collection In the example, the sliding average window length is used. Perform mobile convolution calculation to obtain the convolution time series ;in, Represents the sliding average window length When , the number of convolutions obtained; Step 3.2.3, obtain a new time series: ; For new time series , using sliding average window length Perform mobile convolution calculation to obtain a time series set Hull Moving Average HMA Climate State , the formula is: （2）； Therefore, in the sliding average window length When the historical observation data time series set The Hull moving average HMA climate state is expressed as , which is the observed HMA climate state ; For the historical model data time series collection The Hull moving average HMA climate state is expressed as , which is the climate state of the model HMA .

8. A method for adaptively correcting the climatological state of numerical weather forecast products according to claim 7, characterized in that: In step 3.2.2, in the time series collection In the example, the sliding average window length is used. Perform mobile convolution calculation to obtain the convolution time series , specifically: Step 3.2.2.1, when the length is When the sliding average window moves to a certain position, let its corresponding time series be: Time series ;in: t represents the year; Step 3.2.2.2, use formula (3) to get the time series in Filter function for year The expression of ; （3）； Step 3.2.2.3, use formula (4) to get the time series In the window length Convolution The expression is: （4）； Step 3.2.2.4: Convert the time series And substitute formula (3) into formula (4) to get the convolution The detailed expression of: （5）； By using formula (5), we can calculate the time series In the window length Convolution .

Images