Multi-mode dynamic weight integrated prediction method based on climate state
Through the multi-mode dynamic weight integration prediction method, combined with multiple climate prediction products and inspection indicators, the weight coefficient is dynamically determined, which solves the problems of uncertainty and accuracy differences in climate numerical model prediction in Qinghai area, and achieves a more stable and accurate climate predict.
Patent Information
- Application Number
- CN202411461774.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-10-18
- Publication Date
- 2025-01-03
AI Technical Summary
There are problems of uncertainty and uneven prediction accuracy in existing climate numerical models in short-term climate prediction, especially in Qinghai, where the single model prediction effect is insufficient, resulting in large differences in forecast performance.
Using a multi-mode dynamic weight integration prediction method based on climate states, multiple climate prediction products are selected to integrate unequal weights, combined with sliding average temperature and precipitation distance percentage data in the past five years, multiple test indicators are used to evaluate the various modes. Predict the accuracy rate, dynamically determine the weight coefficient, form the optimal forecast field, and improve the stability and accuracy of climate prediction.
It improves the stability and accuracy of climate numerical model forecasting, enhances the reliability of temperature and precipitation prediction in Qinghai, and is suitable for climate prediction in other periods and regions.
Smart Images
Figure CN119247518A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of climate prediction, and in particular to a multi-mode dynamic weight integrated prediction method based on climate state. Background Art
[0002] Before the popularization of numerical forecasting technology, due to the lack of objective quantitative forecasting data, short-term climate forecasting mainly relied on empirical statistical methods, and the forecasting effect was usually not stable enough. In recent years, with the rapid development and application of computer technology in the fields of weather and climate, many grid-based refined precipitation and temperature data covering the world have been developed, such as station-based grid precipitation products, satellite precipitation data, and reanalysis data. In order to further enhance the defense capabilities against natural disasters, the meteorological department has developed climate forecast products based on global climate models based on global refined real-time data. Therefore, numerical forecasting based on climate models has become the mainstream development direction of short-term climate forecasting and is also the most widely used method in climate forecasting business.
[0003] Numerical model prediction is to solve the basic equations of atmospheric motion under given initial and boundary conditions. Since the atmosphere is a highly nonlinear chaotic system, the initial field of the numerical model is uncertain, and the parameterization scheme of the physical process is not perfect. These factors lead to large uncertainties in a single numerical model. In addition, the prediction accuracy of different climate numerical models for different regions and months is also uneven. Therefore, further understanding the model prediction performance, how to fully mine the model forecast information for the deviation between the model forecast and the actual situation, and how to use historical data to post-correct the multi-model output results to further improve the model forecast accuracy are scientific research and business issues that need to be solved urgently.
[0004] The climate numerical model has a certain prediction performance in Qinghai climate forecast, but the prediction performance of different numerical models varies greatly, and the prediction effect of a single model is not stable enough. When the prediction ability of a single model is limited, ensemble technology is used to improve the model's prediction ability for temperature and precipitation. Ensemble technology has gradually developed from a single model multi-initial field ensemble to a comprehensive prediction using multiple samples and multiple models to reduce the uncertainty of model prediction. The multi-initial value ensemble can reduce the error caused by the uncertainty of the initial value conditions of the dynamic model, and the multi-model ensemble can reduce certain model errors and random errors by combining model results from different sources, and usually has higher prediction skills than a single model. As early as the 1960s and 1970s, some scholars found that by making a specific linear combination of two independent prediction results, the root mean square error of the prediction can be less than the root mean square error of a single prediction. By combining the effective forecast information of multiple models, multi-model integration improves the reliability of forecasts, and the forecast skills are better than those of single models. Commonly used multi-model integration schemes can be divided into two categories: equal weight and unequal weight. Summary of the invention
[0005] The purpose of the present invention is to overcome the shortcomings of the prior art and provide a multi-model dynamic weight integrated prediction method including climate background information to improve the stability of climate numerical model forecasts.
[0006] The objective of the present invention is achieved through the following technical solutions: A multi-model dynamic weighted integrated prediction method based on climate state, comprising the following steps: S1 model data selection: select at least three climate prediction products from the sub-seasonal S2S model product category and the MODESv2 model product category as candidate members of the integrated prediction, obtain the sliding mean temperature and precipitation anomaly percentage data of a certain plateau area in the past five years as a candidate member of the integrated prediction, and preprocess all the candidate members of the integrated prediction; S2 candidate member determination: multiple test indicators are used to evaluate the prediction accuracy of each integrated prediction candidate member for the monthly temperature anomaly and precipitation anomaly percentage in a certain plateau area; According to the monthly mode, the scores of each candidate member of the ensemble forecast are ranked, and the best 1 member, 2 members, 3 members...k members are selected as the average forecast fields of multiple unequal weighted ensembles, where k is the number of candidate members of the ensemble forecast; S3 optimal integration scheme: the first scheme of unequal weight integration of temperature anomaly / precipitation anomaly percentage based on model climate state and the second scheme of unequal weight integration of average temperature / precipitation based on actual standard climate state are used to conduct monthly temperature anomaly / precipitation anomaly percentage prediction tests respectively; The climate forecast data of the most recent year was used as an independent sample to conduct a back-calculation test, and the forecast scores were used to evaluate the forecast results of the first and second schemes under different numbers of candidate members of the ensemble forecast. The model data updated in real time every month is substituted into the calculation, and 2-3 forecast candidate members from the first and second schemes are used for ensemble forecasting to comprehensively form a temperature anomaly / precipitation anomaly percentage forecast product.
[0007] Furthermore, the S1 mode data selection specifically includes: In the sub-seasonal S2S model product category, the Chinese model product CMA-CPS, the American model product S2S-NCEP and the European central model product S2S-EC participating in the international S2S program are selected as candidate members for integrated forecasting; The model prediction data of the National Climate Center of China (NCC), the National Center for Environmental Prediction (NCEP) of the United States, and the European Center (ECMWF) based on the multi-model dataset MODES are used as candidate members of the integrated prediction; Obtain the sliding mean temperature and precipitation anomaly percentage data of a certain plateau area in the past five years as a candidate member of an integrated forecast; Spatial interpolation processing was performed on the site data of the seven selected ensemble prediction candidate members.
[0008] Furthermore, the inspection indicators include the anomaly sign consistency rate AS, the anomaly correlation coefficient ACC, the mean variance skill score MSSS, the grading score Pg and the trend anomaly synthesis Ps.
[0009] Furthermore, the scores of the candidate members of the integrated forecast are ranked according to the monthly mode, and the optimal 1 member, 2 members, 3 members, ... k members are selected as the average forecast fields of multiple unequal weighted sets, where k is the number of candidate members of the integrated forecast, specifically: According to the AS of the symbol agreement rate of each ensemble forecast candidate member, the first member, the first two members, the first three members... and the first k members are selected to form a total of k ensemble forecast fields; According to the symbol consistency rate AS, unequal weights are assigned to the integrated forecast candidate members in the ensemble forecast field. The optimal weight coefficients of the integrated forecast candidate members are as follows: in, AS i For the i The symbol consistency rate of item members, n is the total number of members in the set.
[0010] Furthermore, the first scheme for integrating the temperature anomaly / precipitation anomaly percentages with unequal weights based on the model climate state is specifically as follows: Calculate the monthly average temperature predicted by each candidate member, and then calculate the temperature anomaly based on the climate state of each candidate member; then perform unequal weight integration on the monthly temperature anomalies predicted by the selected candidate members to obtain the monthly temperature anomaly prediction result after unequal weight integration; Calculate the predicted monthly precipitation of each candidate member, and then calculate the precipitation anomaly percentage according to the climate state of each candidate member; then perform unequal weight integration on the predicted monthly precipitation anomaly percentage of the selected ensemble members to obtain the prediction result of the monthly precipitation anomaly percentage after unequal weight integration.
[0011] Furthermore, the second solution of unequal weight integration of average temperature / precipitation based on the actual standard climate state is specifically as follows: The monthly temperature forecast results of each candidate member are integrated to obtain the monthly temperature forecast results after unequal weight collection, and then the temperature anomaly is calculated according to the actual climate state to obtain the integrated monthly temperature anomaly forecast results; The monthly precipitation forecast results of each candidate member are integrated to obtain the monthly precipitation forecast results after unequal weight collection. Then, the precipitation anomaly percentage is calculated according to the actual climate state to obtain the integrated monthly precipitation anomaly percentage forecast results.
[0012] Furthermore, the forecast score includes six indicators, namely, the anomaly sign consistency rate AS, the anomaly correlation coefficient ACC, the average variance skill score and its skill improvement value relative to the National Climate Center's guidance forecast.
[0013] The beneficial effects of the present invention are: 1) Refined evaluation of model performance. The plateau has complex terrain and there are obvious differences in model prediction performance. The prediction performance of each model was comprehensively evaluated by region and month, and the local climate background information of the past five years was incorporated to improve the reliability and stability of each model in localized integrated applications.
[0014] 2) Dynamic weighted ensemble prediction: According to the model prediction performance evaluation results, the ensemble members and dynamic weight coefficients are determined, and the optimal model is selected for unequal weighted ensemble, which improves the accuracy of monthly climate prediction in a targeted manner.
[0015] 3) Strong scalability. This method has good applicability in climate forecasting and is suitable for application to other time periods and regions. BRIEF DESCRIPTION OF THE DRAWINGS
[0016] Figure 1 A technical flow chart of a multi-mode dynamic weighted integrated prediction method based on climate state provided in an embodiment of the present invention. DETAILED DESCRIPTION
[0017] The technical solution of the present invention will be clearly and completely described below in conjunction with the embodiments. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present invention.
[0018] See also Figure 1 , the present invention provides a technical solution: A multi-model dynamic weighted integrated prediction method based on climate state, comprising the following steps: S1 model data selection: select at least three climate prediction products from the sub-seasonal S2S model product category and the MODESv2 model product category as candidate members of the integrated prediction, obtain the sliding mean temperature and precipitation anomaly percentage data of a certain plateau area in the past five years as a candidate member of the integrated prediction, and preprocess all the candidate members of the integrated prediction; In a specific embodiment, the S1 mode data selection specifically includes: In the sub-seasonal S2S model product category, the Chinese model product CMA-CPS, the American model product S2S-NCEP and the European central model product S2S-EC participating in the international S2S program are selected as candidate members for integrated forecasting; The model prediction data of the National Climate Center of China (NCC), the National Center for Environmental Prediction (NCEP) of the United States, and the European Center (ECMWF) based on the multi-model dataset MODES are used as candidate members of the integrated prediction; Obtain the sliding mean temperature and precipitation anomaly percentage data of a certain plateau area in the past five years as a candidate member of an integrated forecast; Spatial interpolation processing was performed on the site data of the seven selected ensemble prediction candidate members.
[0019] This invention introduces the sliding mean temperature and precipitation anomaly data of the Qinghai Plateau in the past five years, and combines the model products of China (CMA), the United States (NCEP), and the European Center (EC) participating in the international S2S program and the model prediction data of China (NCC_CSM), the United States (NCEP), and the European Center (ECMWF) in the multi-model data set (MODES) of the National Climate Center, comprehensively utilizes the effective information of multiple models, and gives full play to the role of different models in the prediction of temperature and precipitation trends in Qinghai.
[0020] S2 candidate member determination: multiple test indicators are used to evaluate the prediction accuracy of each integrated prediction candidate member for the monthly temperature anomaly and precipitation anomaly percentage in a certain plateau area; According to the monthly mode, the scores of each candidate member of the ensemble forecast are ranked, and the best 1 member, 2 members, 3 members...k members are selected as the average forecast fields of multiple unequal weighted ensembles, where k is the number of candidate members of the ensemble forecast; In a specific embodiment, the inspection indicators include the anomaly sign consistency rate AS, the anomaly correlation coefficient ACC, the mean variance skill score MSSS, the grading score Pg and the trend anomaly synthesis Ps.
[0021] Among them, the sign consistency rate of anomalies (anomaly percentage) (AS) is referred to as the same sign rate, which expresses the similarity between the forecast and the actual situation in the form of a percentage. It refers to the percentage of the number of meteorological stations with the same sign of anomalies as the actual value, or with an anomaly of 0, to the total number of meteorological stations (grid points) participating in the scoring: Where N is the total number of meteorological stations (grid points) participating in the scoring, N t The number of meteorological stations (grid points) for which the predicted and actual anomalies (anomaly percentage) have the same sign or one of the anomalies is zero.
[0022] The anomaly correlation coefficient (ACC) mainly reflects the spatial similarity between the forecast value and the actual value, and can also be called the spatial similarity coefficient.
[0023] The mean square skill score (MSSS) is mainly used for unclassified deterministic forecast testing and evaluation.
[0024] The grading score Pg is the scoring method in the "Measures for the Grading and Verification of Short-term Climate Forecast Quality" (Weather Forecast Letter
[2009] No. 141) promulgated by the Forecast and Network Department of the China Meteorological Administration in 2009. The calculation formula for the total score of the multi-station climate trend forecast verification is: Where: PS is the multi-station climate trend prediction score, P i is the score of a single station, and N is the total number of stations participating in the scoring in the province (region, city).
[0025] The trend anomaly comprehensive test (PS) scoring method mainly considers the trend item, anomaly item and underreporting item of the forecast element respectively, and is a comprehensive indicator for the evaluation of element prediction. Among them, the trend is based on whether the anomaly signs of the forecast and the actual situation are consistent, and the judgment is made station by station. When the signs of the forecast and the actual anomaly are consistent, the prediction of the station is considered correct. The anomaly is to examine the forecast ability of the forecast for the first level anomaly (2 ℃>X≥1℃, -1℃≥X>-2℃) and the second level anomaly (≥2 ℃, ≤-2℃) of the forecast object X. The judgment is made station by station and level by level. The calculation formula of the trend anomaly comprehensive test (PS) is as follows: Ps=100*(A*N0+B*N1+C*N2) / (N-N0+A*N0+B*N1+C*N2+M) N0 is the total number of stations with correct trend predictions, N1 is the total number of stations with correct level 1 anomaly predictions, N2 is the total number of stations with correct level 2 anomaly predictions, M is the number of missed stations, N is the total number of stations actually participating in the assessment, a, b and c are the weight coefficients of the climate trend item, level 1 anomaly item and level 2 anomaly item, respectively, where a=2, b=2, c=4.
[0026] The scores of the candidate members of the integrated forecast are ranked according to the monthly mode, and the average forecast fields of multiple unequally weighted sets of the best 1 member, 2 members, 3 members...k members are selected, where k is the number of candidate members of the integrated forecast, specifically: According to the AS of the symbol agreement rate of each ensemble forecast candidate member, the first member, the first two members, the first three members... and the first k members are selected to form a total of k ensemble forecast fields; According to the symbol consistency rate AS, unequal weights are assigned to the integrated forecast candidate members in the ensemble forecast field. The optimal weight coefficients of the integrated forecast candidate members are as follows: in, AS i For the i The symbol consistency rate of item members, n is the total number of members in the set.
[0027] S3 optimal integration scheme: the first scheme of unequal weight integration of temperature anomaly / precipitation anomaly percentage based on model climate state and the second scheme of unequal weight integration of average temperature / precipitation based on actual standard climate state are used to conduct monthly temperature anomaly / precipitation anomaly percentage prediction tests respectively; The climate forecast data of the most recent year was used as an independent sample to conduct a back-calculation test, and the forecast scores were used to evaluate the forecast results of the first and second schemes under different numbers of candidate members of the ensemble forecast. The model data updated in real time every month is substituted into the calculation, and 2-3 forecast candidate members from the first and second schemes are used for ensemble forecasting to comprehensively form a temperature anomaly / precipitation anomaly percentage forecast product.
[0028] The first scheme of integrating the temperature anomaly / precipitation anomaly percentages with different weights based on the model climate state is specifically as follows: Calculate the monthly average temperature predicted by each candidate member, and then calculate the temperature anomaly based on the climate state of each candidate member; then perform unequal weight integration on the monthly temperature anomalies predicted by the selected candidate members to obtain the monthly temperature anomaly prediction result after unequal weight integration; Calculate the predicted monthly precipitation of each candidate member, and then calculate the precipitation anomaly percentage according to the climate state of each candidate member; then perform unequal weight integration on the predicted monthly precipitation anomaly percentage of the selected ensemble members to obtain the prediction result of the monthly precipitation anomaly percentage after unequal weight integration.
[0029] The second solution of the unequal weight integration of average temperature / precipitation based on the actual standard climate state is specifically: The monthly temperature forecast results of each candidate member are integrated to obtain the monthly temperature forecast results after unequal weight collection, and then the temperature anomaly is calculated according to the actual climate state to obtain the integrated monthly temperature anomaly forecast results; The monthly precipitation forecast results of each candidate member are integrated to obtain the monthly precipitation forecast results after unequal weight collection. Then, the precipitation anomaly percentage is calculated according to the actual climate state to obtain the integrated monthly precipitation anomaly percentage forecast results.
[0030] Furthermore, the forecast score includes six indicators, namely, the anomaly sign consistency rate AS, the anomaly correlation coefficient ACC, the average variance skill score and its skill improvement value relative to the National Climate Center's guidance forecast.
[0031] The present invention comprehensively utilizes the effective information of multiple models, uses the model credibility to assign dynamic weight coefficients in the objective integration process, and integrates the climate background characteristics into it, thereby obtaining an unequal weighted ensemble average containing climate characteristic information, thereby improving the forecast stability. A dynamic weighted integrated prediction method with unequal weights is established month by month, and the prediction accuracy of the multi-model dynamic weighted integrated prediction method based on the climate state is tested and evaluated, realizing business application.
[0032] The above is only a preferred embodiment of the present invention. It should be understood that the present invention is not limited to the form disclosed herein, and should not be regarded as excluding other embodiments, but can be used in various other combinations, modifications and environments, and can be modified within the scope of the concept described herein through the above teachings or the technology or knowledge of the relevant field. The changes and modifications made by those skilled in the art shall not deviate from the spirit and scope of the present invention, and shall be within the scope of protection of the claims attached to the present invention.
Claims
1. A multi-mode dynamic weighted integrated prediction method based on climate state, characterized by: The following steps are involved: S1 model data selection: select at least three climate prediction products from the sub-seasonal S2S model product category and the MODESv2 model product category as candidate members of the integrated prediction, obtain the sliding mean temperature and precipitation anomaly percentage data of a certain plateau area in the past five years as a candidate member of the integrated prediction, and preprocess all the candidate members of the integrated prediction; S2 candidate member determination: multiple test indicators are used to evaluate the prediction accuracy of each integrated prediction candidate member for the monthly temperature anomaly and precipitation anomaly percentage in a certain plateau area; According to the monthly mode, the scores of each candidate member of the ensemble forecast are ranked, and the best 1 member, 2 members, 3 members...k members are selected as the average forecast fields of multiple unequal weighted ensembles, where k is the number of candidate members of the ensemble forecast; S3 optimal integration scheme: the first scheme of unequal weight integration of temperature anomaly / precipitation anomaly percentage based on model climate state and the second scheme of unequal weight integration of average temperature / precipitation based on actual standard climate state are used to conduct monthly temperature anomaly / precipitation anomaly percentage prediction tests respectively; The climate forecast data of the most recent year was used as an independent sample to conduct a back-calculation test, and the forecast scores were used to evaluate the forecast results of the first and second schemes under different numbers of candidate members of the integrated forecast. The model data updated in real time every month is substituted into the calculation, and 2-3 forecast candidate members from the first and second schemes are used for ensemble forecasting to comprehensively form a temperature anomaly / precipitation anomaly percentage forecast product.
2. The multi-mode dynamic weighted integrated prediction method based on climate state according to claim 1 is characterized by: The S1 mode data selection specifically includes: In the sub-seasonal S2S model product category, the Chinese model product CMA-CPS, the American model product S2S-NCEP and the European central model product S2S-EC participating in the international S2S program are selected as candidate members for integrated forecasting; The model prediction data of the National Climate Center of China (NCC), the National Center for Environmental Prediction (NCEP) of the United States, and the European Center (ECMWF) based on the multi-model dataset MODES are used as candidate members of the integrated prediction; Obtain the sliding mean temperature and precipitation anomaly percentage data of a certain plateau area in the past five years as a candidate member of an integrated forecast; Spatial interpolation processing was performed on the site data of the seven selected ensemble prediction candidate members.
3. The multi-mode dynamic weight integrated prediction method based on climate state according to claim 1 is characterized by: The inspection indicators include the anomaly sign consistency rate AS, the anomaly correlation coefficient ACC, the mean variance skill score MSSS, the grading score Pg and the trend anomaly comprehensive score Ps.
4. The multi-mode dynamic weight integrated prediction method based on climate state according to claim 3 is characterized by: The scores of the candidate members of the integrated forecast are ranked according to the monthly mode, and the average forecast fields of multiple unequally weighted sets of the best 1 member, 2 members, 3 members...k members are selected, where k is the number of candidate members of the integrated forecast, specifically: According to the AS score of the symbol agreement rate of each ensemble forecast candidate member, the first member, the first two members, the first three members... and the first k members are selected to form a total of k ensemble forecast fields; According to the symbol consistency rate AS, unequal weights are assigned to the integrated forecast candidate members in the ensemble forecast field. The optimal weight coefficients of the integrated forecast candidate members are as follows: in, AS i For the i The symbol consistency rate of item members, n is the total number of members in the set.
5. The multi-mode dynamic weight integrated prediction method based on climate state according to claim 1 is characterized by: The first scheme of integrating the temperature anomaly / precipitation anomaly percentages with different weights based on the model climate state is specifically as follows: Calculate the monthly average temperature predicted by each candidate member, and then calculate the temperature anomaly based on the climate state of each candidate member; then perform unequal weight integration on the monthly temperature anomalies predicted by the selected candidate members to obtain the monthly temperature anomaly prediction result after unequal weight integration; Calculate the predicted monthly precipitation of each candidate member, and then calculate the precipitation anomaly percentage according to the climate state of each candidate member; then perform unequal weight integration on the predicted monthly precipitation anomaly percentage of the selected ensemble members to obtain the prediction result of the monthly precipitation anomaly percentage after unequal weight integration.
6. The multi-mode dynamic weight integrated prediction method based on climate state according to claim 1 is characterized by: The second solution of the unequal weight integration of average temperature / precipitation based on the actual standard climate state is specifically: The monthly temperature forecast results of each candidate member are integrated to obtain the monthly temperature forecast results after unequal weight collection, and then the temperature anomaly is calculated according to the actual climate state to obtain the integrated monthly temperature anomaly forecast results; The monthly precipitation forecast results of each candidate member are integrated to obtain the monthly precipitation forecast results after unequal weight collection. Then, the precipitation anomaly percentage is calculated according to the actual climate state to obtain the integrated monthly precipitation anomaly percentage forecast results.
7. The multi-mode dynamic weight integrated prediction method based on climate state according to claim 1 is characterized by: The forecast score includes six indicators, namely, the anomaly sign consistency rate AS, the anomaly correlation coefficient ACC, the average variance skill score and its skill improvement value relative to the National Climate Center's guidance forecast.