Sub-seasonal mode rainfall correction method and system based on climate partition and sea temperature background constraint

Abstract

The invention relates to the field of meteorological services, and provides a sub-seasonal mode rainfall correction method and system based on climate partitioning and sea temperature background constraint, and the method comprises the steps: climate partitioning, sea temperature background constraint addition, reanalysis data modeling, mode historical return data modeling, real-time prediction model matching, and prediction result correction. The system comprises a data preprocessing unit, a climate partitioning unit, a sea temperature background constraint unit, a reanalysis data modeling unit, a mode historical return data modeling unit and a real-time model matching and prediction result correction unit. According to the method, the importance of rainfall regional differences and sea temperature background constraints is considered, and the problem that the correction method is not applicable due to the fact that the daily rainfall regional differences in summer in China are obvious, overfitting is likely to occur to single-point modeling and physical constraints are lacked is solved; a predictable source with physical significance is effectively utilized to transform the cumulative probability density correction method of percentile mapping, and the rainfall prediction capacity of China in the summer sub-seasons is improved.

Description

Technical Field

[0001] The present invention relates to the field of meteorological services, and particularly to a sub-seasonal model precipitation correction method and system based on climate zoning and sea surface temperature background constraints. By separately considering the climate zoning factor and the sea surface temperature background factor, the precipitation prediction results of the sub-seasonal model are corrected, significantly improving the accuracy of the prediction results. Background Art

[0002] Under the background of global warming, extreme catastrophic climate events occur frequently. The sub-seasonal climate prediction on the time scale of two weeks to two months in the future can provide early warnings for extreme events and strong scientific and technological support for disaster prevention and mitigation. However, there is still an obvious gap between the current sub-seasonal prediction level and short-term weather forecasting and medium- and long-term climate prediction. In the field of sub-seasonal precipitation prediction, dynamic prediction and statistical prediction are mainly used at home and abroad, but the prediction capabilities of both methods have limitations. Specifically, for the mainstream sub-seasonal to seasonal (S2S) dynamic models, such as the model of the European Centre for Medium-Range Weather Forecasts (ECMWF), the prediction skill for precipitation usually only lasts for about two weeks (Domeisen et al., 2022), and as the prediction lead time extends, the prediction skill drops rapidly, especially the prediction ability for heavy precipitation processes is insufficient. In addition, although statistical models (such as the spatio-temporal projection STPM, Zhu and Li, 2017) or dynamic-statistical combined prediction models (Wu et al., 2022) have a certain prediction ability for the trend of sub-seasonal precipitation, there are obvious biases in the prediction of the central position and intensity of precipitation processes. Generally speaking, the current sub-seasonal climate prediction technology is not yet sufficient to meet the major service needs of meteorological disaster prevention and mitigation.

[0003] Model error correction is a key technical means to combine historical long-term observations and model reforecast data, correct the direct output of the model through the statistical characteristics of historical observations, and improve prediction skills. In particular, the percentile mapping correction method (QM) based on the cumulative distribution function (CDF) mainly corrects the cumulative distribution function curve of the element to be corrected, and has an obvious correction effect on the mean value and extreme values. For example, Zhang Daquan and Chen Lijuan (2016) corrected the monthly average temperature of the DERF2.0 model, effectively reducing the root mean square error of the model forecast, and also improving the temperature anomaly distribution forecast to varying degrees. However, the existing CDF-QM method is not applicable to correct the sub-seasonal model prediction results of daily precipitation in China. On the one hand, the precipitation probability distribution shows a skewed distribution, which is quite different from the normal distribution characteristics of temperature. At the same time, precipitation is significantly affected by topography, and its spatial distribution characteristics are relatively complex. The sample size for correction at each grid point is too small, and overfitting is likely to occur. On the other hand, in the absence of physical background constraints, the CDF-QM method can only correct systematic biases and cannot reflect the modulation effect of external forcing signals such as sea surface temperature on the precipitation process.

[0004] To solve the above technical problems, it is necessary to appropriately increase the sample size and consider reasonable physical background constraints. When increasing the sample size required for modeling, a certain quantity needs to be reached to avoid overfitting, and at the same time, the consistency between samples should be taken into account to prevent the established model from being inapplicable to the grid points to be corrected. When considering physical background constraints, it is required that the physical signal has characteristics such as stability, slow variation, and significant influence. Therefore, how to solve the existing technical problems and improve the prediction ability of the sub-seasonal model for summer precipitation in China requires sufficient meteorological knowledge reserves and certain technical difficulties. By solving this technical problem, the existing sub-seasonal dynamic model information can be effectively mined, and the prediction ability of sub-seasonal precipitation in summer in China can be significantly improved, providing important scientific and technological support for disaster prevention, mitigation, and water resource management. Summary of the Invention

[0005] The purpose of the present invention is to overcome the deficiencies of the prior art. Aiming at the problem that the CDF-QM correction method is not applicable due to the obvious regional differences in daily summer precipitation in China, the tendency of overfitting in single-point modeling, and the lack of physical constraints, a sub-seasonal model precipitation correction method and system based on climate zoning and sea surface temperature background constraints are provided to correct the error of the sub-seasonal model prediction results of daily summer precipitation in China and improve the prediction ability of sub-seasonal precipitation in summer in China.

[0006] Explanation of terms in this article:

[0007] Reanalysis precipitation data: A precipitation data set that is spatio-temporally continuous and generated by fusing various observational data and numerical model simulation results.

[0008] Model historical return data: The starting reporting date is past time point 1, and the precipitation data at past time point 2 is predicted using the sub-seasonal model.

[0009] Precipitation prediction result: The precipitation data at a future time point is predicted using the sub-seasonal model.

[0010] Prediction for a certain starting reporting date and a certain forecasting date: Taking a certain starting reporting date x as the starting point, the precipitation at a certain forecasting date y is predicted.

[0011] The present invention adopts the following technical solutions:

[0012] On the one hand, the present invention provides a sub-seasonal model precipitation correction method based on climate zoning and sea surface temperature background constraint, including:

[0013] S1. Climate zoning: The area to be corrected is gridded at a certain resolution; based on the daily characteristics of summer precipitation, using the reanalysis precipitation data, the area to be corrected is clustered and divided into several climate sub-regions.

[0014] S2. Adding sea surface temperature background constraint: Using the sea surface temperature background model, adding sea surface temperature background constraints to all the climate sub-regions in step S1.

[0015] S3. Reanalysis data modeling: For each climate sub-region with sea surface temperature background constraint, according to the reanalysis precipitation data of a certain date within the historical period, sliding window CDF modeling is carried out to obtain the first CDF curve of this climate sub-region on this date and under this sea surface temperature background constraint.

[0016] S4. Model historical return data modeling: For each climate sub-region with sea surface temperature background constraint, based on the sub-seasonal model, sliding window CDF modeling is carried out on all the model historical return data of a certain starting reporting date and a certain forecasting date to obtain the second CDF curve of this climate sub-region on this starting reporting date, this forecasting date, and under this sea surface temperature background constraint.

[0017] S5. Real-time prediction model matching: Judging the sea surface temperature background constraint to which the current year belongs, based on the sub-seasonal model, projecting the precipitation prediction results of each grid point of a certain climate sub-region, a certain starting reporting date, and a certain forecasting date onto the second CDF curve corresponding to the same climate sub-region, the same sea surface temperature background constraint, the same starting reporting date, and the same forecasting date in step S4 to obtain the probability density percentile position corresponding to the precipitation prediction result.

[0018] S6. Prediction result correction: Map the percentile position of the probability density obtained in step S5 to the first CDF curve of the same climate sub-region, the same sea surface temperature background constraint, and the same date as the forecast date obtained in step S3, to obtain the reanalysis precipitation value corresponding to the percentile position of the probability density. This reanalysis precipitation value is used as the corrected prediction result for a certain starting date and a certain forecast date.

[0019] In any of the possible implementation manners described above, a further implementation manner is provided. The method further includes:

[0020] S7. Repeat steps S5 and S6 to perform real-time prediction model matching and prediction result correction for each prediction date of a certain starting date, and obtain the corrected prediction results for all forecast dates of a certain starting date.

[0021] In any of the possible implementation manners described above, a further implementation manner is provided. In step S1, the K-means clustering method is used to classify the area to be corrected; all grid points are divided into K groups, and K grid points are randomly selected as the initial clustering centers. Then, calculate the distance (mean square error) between each grid point and the precipitation time series of other grid points except these K grid points, and assign each grid point to the clustering center closest to it; the clustering centers and the grid points assigned to them represent a cluster; each time a grid point is assigned, the clustering centers are recalculated according to all the grid points existing in the cluster; this process is repeated continuously until no grid point is reassigned to a different cluster, no clustering center changes, and the sum of squared errors is minimized within the same cluster, that is, the classification of the area to be corrected is completed; each cluster forms a climate sub-region.

[0022] In any of the possible implementation manners described above, a further implementation manner is provided. In step S2, the area to be corrected is the entire region of China, and the sea surface temperature background model uses the Indian Ocean Basin Mode (IOBM). It is divided into normal years, positive anomaly years, and negative anomaly years according to the IOBM index; years with an IOBM index greater than 1.5 times the standard deviation are positive anomaly years, years with an IOBM index less than -1.5 times the standard deviation are negative anomaly years, and years with an IOBM index within ±1.5 times the standard deviation are normal years.

[0023] In any of the possible implementation manners described above, a further implementation manner is provided. The IOBM index is defined as the anomaly value of the area-averaged sea surface temperature in the region of 20°S - 20°N, 40°E - 110°E after removing the linear trend;

[0024] The calculation method for standardizing the IOBM index is:

[0025] Here

[0026] where xi Denote the IOBM index as y i Denote the standardized IOBM index is the mean of the IOBM index, and n is the number of IOBM indices.

[0027] For any of the possible implementation manners as described above, a further implementation manner is provided. In step S3, the method for obtaining the first CDF curve is as follows:

[0028] S31. Set the precipitation group interval, and the precipitation group interval is set to change gradually, with the small precipitation group interval being smaller than the large precipitation group interval;

[0029] S32. For the daily precipitation reanalysis values within the sliding window in the positive anomaly years, negative anomaly years, and normal years of the IOBM index for all grid points within each climate sub-region across the country from May 1st to September 30th according to the clustering results in step S1, calculate the cumulative probability density values of the precipitation amounts within each precipitation interval according to the set gradually changing group interval, and obtain the cumulative probability density curve within the entire 0 - 100 mm range, which is the first CDF curve.

[0030] For any of the possible implementation manners as described above, a further implementation manner is provided. In step S4, the method for obtaining the second CDF curve is as follows:

[0031] S41. Set the precipitation group interval, and the precipitation group interval is the same as that in step S31;

[0032] S42. For the daily precipitation amounts of the model historical returns within the sliding window in the positive anomaly years, negative anomaly years, and normal years of the IOBM index for all grid points within each climate sub-region across the country from May 1st to September 30th for the future 0 - 60 days (taking CMA - CPSv3 as an example, the forecast lead times of different models are different) according to the clustering results in step S1 within the entire historical reforecast period (taking CMA - CPSv3 as an example, its historical reforecast period is the past 15 years with the starting year as a reference), calculate the cumulative probability density values of the model historical return precipitation amounts within each precipitation interval according to the set gradually changing group interval, and obtain the cumulative probability density curve within the 0 - 100 mm range, which is the second CDF curve.

[0033] For any of the possible implementation manners as described above, a further implementation manner is provided. The precipitation group interval is set as follows: within the range of 0 - 1 mm, the interval is 0.1 mm; within the range of 1 - 10 mm, the interval is 0.2 mm; within the range of 10 - 25 mm, the interval is 0.5 mm; within the range of 25 - 50 mm, the interval is 1.0 mm; within the range of 50 - 100 mm, the interval is 2.0 mm.

[0034] For any of the possible implementation manners described above, a further implementation manner is provided. In step S5, for all the forecast days of each starting report date (taking CMA-CPSv3 as an example, the result of forecasting the next 0 - 60 days in one starting report), all grid points are judged and calculated one by one.

[0035] On the other hand, the present invention also provides a sub-seasonal model precipitation correction system based on climate zoning and sea surface temperature background constraint, including:

[0036] A data preprocessing unit, which is used for the standardization processing of precipitation data and the grid division of the area to be corrected; the specific standardization processing of the precipitation data includes: processing the cumulative precipitation amount reported by the sub-seasonal model and the forecast cumulative precipitation amount into daily average precipitation amounts; interpolating the reanalysis precipitation data, the daily average precipitation amounts reported by the sub-seasonal model, and the forecast daily average precipitation amounts to the grid points with the same spatial resolution;

[0037] A climate zoning unit, based on the daily variation characteristics of summer precipitation, uses the reanalysis precipitation data processed by the data preprocessing unit to cluster and divide the area to be corrected into several climate sub-regions;

[0038] A sea surface temperature background constraint unit, which uses a sea surface temperature background model to add sea surface temperature background constraints to all the climate sub-regions;

[0039] A reanalysis data modeling unit, for each sub-region with sea surface temperature background constraint, performs sliding window CDF modeling on the reanalysis precipitation data within the historical period on a certain date, and obtains the first CDF curve of the climate sub-region on this date and under this sea surface temperature background constraint;

[0040] A model historical return data modeling unit, for each climate sub-region with sea surface temperature background constraint, based on the sub-seasonal model, performs sliding window CDF modeling on all the model historical return data of a certain forecast day of a certain starting report date, and obtains the second CDF curve of the climate sub-region on this starting report date, this forecast day, and under this sea surface temperature background constraint;

[0041] A real-time model matching and prediction result correction unit, judges the sea surface temperature background constraint to which the current year belongs, projects the precipitation prediction result of each grid point of a certain climate sub-region, a certain starting report date, and a certain forecast day onto the second CDF curve corresponding to the same climate sub-region, the same sea surface temperature background constraint, the same starting report date, and the same forecast day in step S4, and obtains the probability density percentile position corresponding to the precipitation prediction result; maps the probability density percentile position onto the first CDF curve corresponding to the same climate sub-region, the same sea surface temperature background constraint, and the date same as the forecast day, and obtains the reanalysis precipitation amount value corresponding to the probability density percentile position, and this reanalysis precipitation amount value is used as the corrected prediction result of a certain starting report date and a certain forecast day.

[0042] The beneficial effects of the present invention are as follows:

[0043] 1. The present invention effectively utilizes the scientific support that the summer precipitation in China shows obvious regional differences due to the northward movement of the rain belt from south to north, and the main feature that the summer precipitation in China is significantly affected by the Indian Ocean Basin Mode (IOBM) in spring. The technical method is designed and good correction effects are achieved, which reflects the importance of considering regional differences in precipitation and sea surface temperature background constraints in the model error correction for predicting daily summer precipitation in China at the sub-seasonal scale, as well as the effectiveness of scientific methods in guiding technical improvement.

[0044] 2. It solves the problem that the CDF-QM correction method is not applicable due to the obvious regional differences in daily summer precipitation in China, the easy occurrence of overfitting in single-point modeling, and the lack of physical constraints.

[0045] 3. Use the grid point data in the area where the temporal variation of precipitation is relatively consistent to jointly establish a CDF model to avoid overfitting problems. Further consider different sea surface temperature backgrounds for physical constraints, and correct the prediction results of all grid points in this area through this model. At the same time, an unequal interval group interval setting of first dense and then sparse is adopted in the establishment of the CDF model, so as to effectively utilize the physically meaningful source of predictability to transform the cumulative probability density correction method of percentile mapping, and then correct the error of the sub-seasonal model prediction results of daily summer precipitation in China, and improve the sub-seasonal precipitation prediction ability in summer in China.

[0046] 4. The method and system of the present invention can be extended to other regions outside China and have broader practicality. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 The figure shows a schematic flow chart of a sub-seasonal model precipitation correction method based on climate zoning and sea surface temperature background constraints according to an embodiment of the present invention.

[0048] Figure 2 The figure shows a schematic diagram of the results of climate zoning for China in the embodiment.

[0049] Figure 3The figure shows the verification score chart of the error correction effect of the daily precipitation for 15 - 30 days directly output by the CMA - CPSv3 sub - seasonal model in the embodiment; (a) PC, (b) ETS, and (c) HSS are the forecast accuracy and fairness skill scores, and the larger the value, the higher the forecast skill; (d) BIAS is the error score, and the smaller the value, the higher the forecast skill; where Origin is the original result directly output by the sub - seasonal model, CM1 is the correction result of directly applying the cumulative probability density method without considering the climate division and sea - surface temperature background constraints technology, and CM2 is the correction result of the cumulative probability density method considering the climate division and sea - surface temperature background constraints technology.

[0050] Figure 4 The figure shows the error correction effect diagram of the precipitation in China on June 9, 2024 by the CMA - CPSv3 sub - seasonal model 15 days in advance and 30 days in advance in Embodiment 2; (a) shows the distribution of observed precipitation in the re - analysis data on June 9, 2024; (b) shows the prediction result of the CMA - CPSv3 sub - seasonal model for the precipitation in China on June 9, 2024, 15 days in advance.

[0051] (c) shows the result after correcting the precipitation in (b) using the cumulative probability density correction method considering the climate division and sea - surface temperature background constraints technology; (d) is the same as (b) but for the forecast result 30 days in advance; (e) is the same as (c) but for the result after correcting (d).

[0052] Figure 5 The figure shows the schematic diagram of the real - time prediction model matching and prediction result correction process in steps S5 and S6 of the embodiment. Detailed implementation manners

[0053] The specific embodiments of the present invention will be described in detail below with reference to the specific drawings. It should be noted that the technical features described in the following embodiments or the combination of technical features should not be considered in isolation, and they can be combined with each other to achieve better technical effects.

[0054] As Figure 1 shown, an embodiment of the present invention provides a sub - seasonal model precipitation correction method based on climate division and sea - surface temperature background constraints, including:

[0055] S1. Climate division: The area to be corrected is gridded at a certain resolution; based on the daily characteristics of summer precipitation, using the re - analysis precipitation data, the area to be corrected is clustered and divided into several climate sub - regions;

[0056] S2. Adding sea - surface temperature background constraints: Using the sea - surface temperature background model, add sea - surface temperature background constraints to all the climate sub - regions in step S1;

[0057] S3. Re - analysis data modeling: For each of the climate sub - regions with SST background constraints, based on the re - analysis precipitation data on a certain date within the historical period, perform sliding - window CDF modeling to obtain the first CDF curve of this climate sub - region on this date under this SST background constraint.

[0058] S4. Model historical hindcast data modeling: For each of the climate sub - regions with SST background constraints, based on the sub - seasonal model, perform sliding - window CDF modeling on all the model historical hindcast data for a certain forecast date of a certain initialization date to obtain the second CDF curve of this climate sub - region on this initialization date, this forecast date, and under this SST background constraint.

[0059] S5. Real - time prediction model matching: Determine the SST background constraint to which the current year belongs. Based on the sub - seasonal model, project the precipitation prediction results of each grid point of a certain climate sub - region, a certain initialization date, and a certain forecast date onto the second CDF curve corresponding to the same climate sub - region, the same SST background constraint, the same initialization date, and the same forecast date in step S4 to obtain the probability density percentile position corresponding to the precipitation prediction result.

[0060] S6. Prediction result correction: Map the probability density percentile position obtained in step S5 to the first CDF curve of the same climate sub - region, the same SST background constraint, and the same date as the forecast date obtained in step S3 to obtain the re - analysis precipitation value corresponding to the probability density percentile position. This re - analysis precipitation value is used as the corrected prediction result for a certain initialization date and a certain forecast date.

[0061] In a specific embodiment, the method further includes: S7. Repeat steps S5 and S6 to perform real - time prediction model matching and prediction result correction for each forecast date of a certain initialization date to obtain the corrected prediction results for all forecast dates of a certain initialization date.

[0062] In a specific embodiment, the area to be corrected is the Chinese region. When performing climate zoning in S1, based on the daily variation characteristics of precipitation in summer (May - September) in China, use the K - means clustering method to divide the grid precipitation re - analysis data with a resolution of 25 km across the country (10 - 20 types of partitions across the country are acceptable).

[0063] In a specific embodiment, in S2, the SST background model adopts the Indian Ocean Basin - wide Mode (IOBM). For the positive anomaly years (years with IOBM index greater than 1.5 times the standard deviation), negative anomaly years (years with IOBM index less than - 1.5 times the standard deviation), and normal years (years with IOBM index within ±1.5 times the standard deviation) of the IOBM index (the average value of March and April in spring), distinguish the re - analysis precipitation and sub - seasonal model historical hindcast precipitation within each climate sub - region.

[0064] In a specific embodiment, in S3, for all grid points in each climate sub-region on a certain day (e.g., May 1st, May 2nd, ……, September 29th, September 30th) within the historical period (currently taking 1991 to 2020), according to the positive anomaly years, negative anomaly years, and normal years of the IOBM index, the precipitation reanalysis data in each climate sub-region is respectively subjected to CDF modeling with a 15-day sliding window (taking May 1st as the target day, the sliding window includes April 24th to April 30th and May 1st to May 8th), to obtain the first CDF curve of each climate sub-region under this certain day and this sea surface temperature background constraint (any combination of the three factors of climate sub-region, date, and sea surface temperature background constraint corresponds to a first CDF curve).

[0065] In a specific embodiment, in S4, based on the sub-seasonal model, starting from a certain day (e.g., May 1st, May 2nd, ……, September 29th, September 30th), the historical return results of all models for the xth day in the future (for example, the prediction time limit of the CMA-CPSv3 sub-seasonal model is 0 - 60 days, so the value range of x is 0 - 60) are used. According to the positive anomaly years, negative anomaly years, and normal years of the IOBM index, for all grid points in each climate sub-region, CDF modeling is respectively carried out with a 15-day sliding window (the target day of the sliding window refers to the forecast day, that is, the aforementioned xth day in the future. For example, when the starting day is May 1st and the target day is May 2nd, the sliding window includes the precipitation values predicted from April 24th to April 25th, predicted from April 25th to April 26th, ……, predicted from May 7th to May 8th, and predicted from May 8th to May 9th), to obtain the second CDF curve of each grid point in this climate sub-region under this starting day, this forecast day, and this sea surface temperature background constraint. For each starting day, any combination of the three factors of climate sub-region, forecast day, and sea surface temperature background constraint corresponds to a second CDF curve.

[0066] In a specific embodiment, in S5, the sea surface temperature background to which the current year belongs is judged, and the precipitation prediction results of each grid point for the xth day in the future (for example, the prediction time limit of the CMA-CPSv3 sub-seasonal model is 0 - 60 days, so the value range of x is 0 - 60) starting from a certain starting day (e.g., May 1st, May 2nd, ……, September 29th, September 30th) of the sub-seasonal model are projected into the second CDF curve of the CDF model with the same sea surface temperature background constraint, the same starting day, the same forecast day, and the same climate sub-region established in step S4, to determine the percentile position of the predicted precipitation in the second CDF curve.

[0067] In a specific embodiment, in S6, among the first CDF curves of the CDF models within the same climate sub-region under the SST background constraint where the grid points to be corrected on the forecast date in step S5 are found in step S3, determine the reanalysis precipitation value corresponding to the percentile position obtained in step S5 in this first CDF curve, and use this reanalysis precipitation value as the correction result to replace the sub-seasonal model prediction result to be corrected in step S5.

[0068] Figure 5 The real-time prediction model matching and prediction result correction processes of steps S5 and S6 in a specific embodiment are given. The left image is the second CDF curve, and the right image is the first CDF curve. The predicted precipitation value is X f (d) Find the corresponding percentile ratio P on the second CDF curve c (X f (d)), and according to this P c (X f (d)) find the corresponding reanalysis precipitation value X fcorr (d) on the first CDF curve. This reanalysis precipitation value X fcorr (d) is used as the final corrected prediction result.

[0069] Repeat steps S3 and S4 to achieve CDF modeling of the reanalysis (observed) precipitation for all dates and all sub-regions within the period from May 1st to September 30th under different IOBM SST background constraints, and CDF modeling of the predicted precipitation for all starting dates, all forecast dates, and all climate sub-regions within the period from May 1st to September 30th.

[0070] Repeat steps S5 to S6 to correct the prediction results for all forecast dates of one starting date and all grid points one by one.

[0071] In a specific implementation, in step S1, the K-means clustering method is used to classify the area to be corrected; all grid points are divided into K groups, and K grid points are randomly selected as the initial clustering centers. Then calculate the distance (mean square error) between each grid point and the precipitation time series of other grid points except these K grid points, and assign each grid point to the clustering center closest to it; the clustering centers and the grid points assigned to them represent a cluster; each time a grid point is assigned, the clustering centers are recalculated based on all the grid points existing in the cluster; this process is continuously repeated until no grid point is reassigned to a different cluster, no clustering center changes, and the sum of squared errors is minimized within the same cluster, that is, the classification of the area to be corrected is completed; each cluster forms a climate sub-region.

[0072] Figure 2 The clustering results in a certain embodiment are given, which are divided into 16 categories, that is, divided into 16 climate sub-regions.

[0073] In a specific implementation, the IOBM index is defined as the anomaly value of the regional average sea surface temperature in the area of 20°S - 20°N, 40°E - 110°E after removing the linear trend;

[0074] The calculation method for standardizing the IOBM index is as follows:

[0075] Here

[0076] where x i represents the IOBM index, and y i represents the standardized IOBM index, is the mean value of the IOBM index, and n is the number of IOBM indices.

[0077] In a specific embodiment, in step S3, the method for obtaining the first CDF curve is as follows:

[0078] S31. Set the precipitation group interval, and the precipitation group interval is set to change gradually, with the small precipitation group interval being smaller than the large precipitation group interval;

[0079] S32. For the daily precipitation reanalysis values of all grid points within each climate sub-region in the positive anomaly years, negative anomaly years, and normal years of the IOBM index within the sliding window across the country from May 1st to September 30th, calculate the cumulative probability density values of the precipitation within each precipitation interval according to the set gradually changing group interval, and obtain the cumulative probability density curve within the entire range of 0 - 100 mm, which is the first CDF curve.

[0080] In a specific embodiment, in step S4, the method for obtaining the second CDF curve is as follows:

[0081] S41. Set the precipitation group interval, and the precipitation group interval is the same as that in step S31;

[0082] S42. For the daily precipitation of the model historical reforecast for the next 0 - 60 days (taking CMA - CPSv3 as an example, the forecast lead time is different for different models) starting from May 1st to September 30th across the country, according to the clustering results in step S1, for all grid points within each climate sub-region within the entire historical reforecast period of the model (taking CMA - CPSv3 as an example, its historical reforecast period is the past 15 years with the starting year as a reference), calculate the cumulative probability density values of the precipitation within each precipitation interval according to the set gradually changing group interval for the positive anomaly years, negative anomaly years, and normal years of the IOBM index, and obtain the cumulative probability density curve within the range of 0 - 100 mm, which is the second CDF curve.

[0083] In a specific embodiment, the inter-group spacing of the precipitation is set as follows: the spacing is 0.1 mm within the range of 0 - 1 mm, 0.2 mm within the range of 1 - 10 mm, 0.5 mm within the range of 10 - 25 mm, 1.0 mm within the range of 25 - 50 mm, and 2.0 mm within the range of 50 - 100 mm.

[0084] An embodiment of the present invention provides a sub-seasonal model precipitation correction system based on climate zoning and sea surface temperature background constraints, comprising:

[0085] A data preprocessing unit for standardizing precipitation data and gridifying the area to be corrected; the standardization processing of the precipitation data specifically includes: processing the cumulative precipitation reported by the sub-seasonal model and the predicted cumulative precipitation into daily average precipitation; interpolating the reanalysis precipitation data, the daily average precipitation reported by the sub-seasonal model, and the predicted daily average precipitation to grid points with the same spatial resolution;

[0086] A climate zoning unit, based on the daily characteristics of summer precipitation, uses the reanalysis precipitation data processed by the data preprocessing unit to cluster and divide the area to be corrected into several climate sub-regions;

[0087] A sea surface temperature background constraint unit that adds sea surface temperature background constraints to all the climate sub-regions using a sea surface temperature background model;

[0088] A reanalysis data modeling unit that performs sliding window CDF modeling on the reanalysis precipitation data within the historical period on a certain date for each sub-region with sea surface temperature background constraints, to obtain a first CDF curve of the climate sub-region on this date and under this sea surface temperature background constraint;

[0089] A model historical return data modeling unit that performs sliding window CDF modeling on all the model historical return data on a certain forecast date for a certain initialization date for each climate sub-region with sea surface temperature background constraints, to obtain a second CDF curve of the climate sub-region on this initialization date, this forecast date, and under this sea surface temperature background constraint;

[0090] The real-time model matching and prediction result correction unit determines the SST background constraint to which the current year belongs, projects the precipitation prediction results of each grid point in a certain climate sub-region, a certain starting date, and a certain prediction date onto the second CDF curve corresponding to the same climate sub-region, the same SST background constraint, the same starting date, and the same prediction date in step S4, and obtains the probability density percentile position corresponding to the precipitation prediction result; the probability density percentile position is mapped onto the first CDF curve corresponding to the same climate sub-region, the same SST background constraint, and the same date as the prediction date, and the reanalysis precipitation value corresponding to the probability density percentile position is obtained. This reanalysis precipitation value serves as the corrected prediction result for a certain starting date and a certain prediction date.

[0091] Embodiment 1

[0092] The error correction effect of the CMA-CPSv3 sub-seasonal model's precipitation prediction results for multiple times with a 15- to 30-day lead in summer from 2021 to 2023.

[0093] ① Data standardization processing and grid point partitioning

[0094] The cumulative precipitation of the CMA-CPSv3's historical return for the summer from 2008 to 2020 and its real-time prediction for the summer from 2021 to 2023 are both processed into daily average precipitation; the reanalysis precipitation, the precipitation of the model's historical return, and the predicted precipitation are all interpolated onto grid points with a resolution of 25 km; according to the interannual variation characteristics of the reanalysis precipitation in summer from 1991 to 2020, the precipitation grid points across the country are divided into 16 climate sub-regions using the Kmeans method.

[0095] Meanwhile, calculate the IOBM index for spring (average of March and April) from 1991 to 2023, and based on this index, divide the climate period of the reanalysis data (1991 to 2020) and the historical return period of the CMA-CPSv3 (2008 - 2020) into IOBM index positive anomaly years, IOBM index negative anomaly years, and IOBM index normal years according to the threshold of 1.5 times the standard deviation, and determine that the years to be corrected (2021 to 2023) all belong to the IOBM normal background years.

[0096] ② CDF modeling of reanalysis precipitation

[0097] For the summer reanalysis precipitation of the IOBM index positive anomaly years, negative anomaly years, and normal years on the 25 km grid points in different regions obtained in step ①, establish CDF models with a 15-day sliding window for each region on a daily basis.

[0098] ③ CDF modeling of the sub-seasonal model's return precipitation

[0099] For the daily precipitation in different regions obtained in step ① at the 25-km grid points of CMA-CPSv3 for the positive anomaly years, negative anomaly years, and normal years of the IOBM index in summer with different starting forecast times, a CDF model with a 15-day sliding window is established for each region respectively.

[0100] ④ Correct the errors of the daily precipitation in summer in China from 2021 to 2023 predicted by CMA-CPSv3 15 - 30 days in advance

[0101] Project the daily forecast results of the CMA-CPSv3 sub-seasonal model 15 to 30 days in advance onto the corresponding CDF model established in step ③ for each grid point to obtain the CDF percentile corresponding to the corrected value; then project this percentile onto the corresponding CDF model established in step ② to obtain the corrected value of the prediction result.

[0102] The skill scores of the prediction results before and after correction are as Figure 3 shown. (a) PC, (b) ETS, and (c) HSS are the skill scores for forecast accuracy and fairness, and the larger the value, the higher the forecast skill; (d) BIAS is the error score, and the smaller the value, the higher the forecast skill; the test results of multiple indicators all show that the method of the present invention can effectively correct the prediction skill of the sub-seasonal model for summer precipitation in China.

[0103] Example 2

[0104] Correct the prediction results of the precipitation in China on June 9, 2024 by the CMA-CPSv3 sub-seasonal model 15 days in advance and 30 days in advance respectively (2024 belongs to the positive anomaly background year of IOBM). The correction results are as Figure 4 shown. (a) shows the distribution of the observed precipitation in the reanalysis data on June 9, 2024; (b) is the prediction result of the precipitation in China on June 9, 2024 by the CMA-CPSv3 sub-seasonal model 15 days in advance; (c) is the result after correcting the precipitation in (b) using the cumulative probability density correction method considering climate zoning and sea surface temperature background constraints; (d) is the same as (b) but for the forecast result 30 days in advance; (e) is the same as (c) but for the result after correcting (d). It can be seen that the method of the present invention has an obvious correction effect on the position and magnitude of the large precipitation value areas predicted by CMA-CPSv3 for precipitation in China.

[0105] The present invention effectively utilizes the scientific support that the summer precipitation in China shows obvious regional differences due to the northward movement of the rain belt, and the main feature that the summer precipitation in China is significantly affected by the Indian Ocean Basin Mode (IOBM) in spring, designs technical methods and achieves good correction effects.

[0106] Although several embodiments of the present invention have been given in this text, those skilled in the art should understand that the embodiments in this text can be changed without departing from the spirit of the present invention. The above embodiments are only exemplary and should not be used as a limitation of the scope of the rights of the present invention with the embodiments in this text.

Claims

1. A sub-seasonal precipitation correction method based on climate zoning and sea temperature background constraints, characterized in that: The method comprises: S1. Climate zoning: The area to be corrected is gridded at a certain resolution; based on the daily characteristics of summer precipitation, the area to be corrected is clustered into several climate sub-regions using reanalysis precipitation data; S2, adding sea temperature background constraints: using the sea temperature background model, adding sea temperature background constraints to all the climate sub-areas in step S1; S3, reanalysis data modeling: for each of the climate sub-regions with sea temperature background constraints, a sliding window CDF modeling is performed based on the reanalysis precipitation data of a certain date in a historical period to obtain the first CDF curve of the climate sub-region under the constraint of the sea temperature background on the date; S4, modeling of historical model return data: for each climate sub-region with sea temperature background constraints, based on the sub-seasonal model, all model historical return data of a certain forecast day of a certain start date are modeled by sliding window CDF, and a second CDF curve of the climate sub-region under the constraints of the start date, the forecast day and the sea temperature background is obtained; S5, real-time prediction model matching: determine the sea temperature background constraint of the current year, and based on the sub-seasonal mode, project the grid-by-grid precipitation prediction results of a certain climate sub-region, a certain start date, and a certain forecast day onto the second CDF curve of the same climate sub-region, the same sea temperature background constraint, the same start date, and the same forecast day corresponding to step S4, and obtain the probability density percentile position corresponding to the precipitation prediction result; S6. Correction of forecast results: The probability density percentile position obtained in step S5 is matched to the first CDF curve of the same climate sub-region, the same sea temperature background constraint, and the same date as the forecast day obtained in step S3, and the reanalysis precipitation value corresponding to the probability density percentile position is obtained. The reanalysis precipitation value is used as the corrected forecast result for the forecast day of the certain reporting start date.

2. The sub-seasonal precipitation correction method based on climate zoning and sea temperature background constraints according to claim 1, characterized in that: The method further comprises: S7. Repeat steps S5 and S6 to perform real-time prediction model matching and prediction result correction for each forecast day of a certain start date, and obtain the corrected prediction results of all forecast days of a certain start date.

3. The sub-seasonal precipitation correction method based on climate zoning and sea temperature background constraints according to claim 1, characterized in that: In step S1, the K-means clustering method is used to classify the area to be corrected; all grid points are divided into K groups, K grid points are randomly selected as initial cluster centers, and then the distance between each grid point and the precipitation time series of other grid points except these K grid points is calculated, and each grid point is assigned to the cluster center closest to it; the cluster center and the grid points assigned to them represent a cluster; each time a grid point is assigned, the cluster center is recalculated based on all existing grid points in the cluster; this process is repeated until no grid points are reassigned to different clusters, no cluster center changes, and the sum of squared errors is minimized within the same cluster, that is, the classification of the area to be corrected is completed; each cluster forms a climate sub-region.

4. The sub-seasonal precipitation correction method based on climate zoning and sea temperature background constraints according to claim 1, characterized in that: In step S2, the area to be corrected is the entire China, the sea temperature background model adopts the Indian Ocean Basin Consistent Mode (IOBM), and is divided into normal years, positive abnormal years, and negative abnormal years according to the IOBM index; years with an IOBM index greater than 1.5 times the standard deviation are positive abnormal years, years with an IOBM index less than -1.5 times the standard deviation are negative abnormal years, and years with an IOBM index within ±1.5 times the standard deviation are normal years.

5. The sub-seasonal precipitation correction method based on climate zoning and sea temperature background constraints as claimed in claim 4, characterized in that: The IOBM index is defined as the anomaly of the average sea surface temperature in the region of 20°S-20°N, 40°E-110°E after removing the linear trend; The calculation method of IOBM index standardization is: here Among them, x i represents the IOBM index, y i represents the standardized IOBM index, is the mean of the IOBM index, and n is the number of IOBM indices.

6. The sub-seasonal precipitation correction method based on climate zoning and sea temperature background constraints according to claim 4, characterized in that: In step S3, the method for obtaining the first CDF curve is: S31, setting the spacing between precipitation groups, wherein the spacing between precipitation groups is gradually set, and the spacing between small precipitation groups is smaller than the spacing between large precipitation groups; S32. For the precipitation reanalysis values ​​within the sliding window of the IOBM index positive anomaly years, negative anomaly years, and normal years for all grid points in the climate sub-regions nationwide from May 1 to September 30, according to the clustering results of step S1, the cumulative probability density values ​​of precipitation in each precipitation interval are calculated according to the set gradient group spacing, and the cumulative probability density curve in the entire range of 0-100 mm is obtained, that is, the first CDF curve.

7. The method for correcting subseasonal precipitation based on climate zoning and sea temperature background constraints according to claim 6, characterized in that: In step S4, the method for obtaining the second CDF curve is: S41, setting the interval between precipitation groups, the interval between precipitation groups is the same as step S31; S42. For the daily reports from May 1 to September 30, the next 0-60 days are forecasted nationwide on a daily basis. According to the clustering results in step S1, for all grid points in the climate sub-region during the entire historical return period, the model historical return precipitation in the sliding window is calculated according to the set gradient group spacing for positive anomaly years, negative anomaly years, and normal years. The cumulative probability density value of the model historical return precipitation in each precipitation interval is calculated to obtain the cumulative probability density curve in the range of 0-100 mm, that is, the second CDF curve.

8. The sub-seasonal precipitation correction method based on climate zoning and sea temperature background constraints according to claim 7, characterized in that: The spacing between the precipitation groups is set as follows: 0.1 mm in the range of 0-1 mm, 0.2 mm in the range of 1-10 mm, 0.5 mm in the range of 10-25 mm, 1.0 mm in the range of 25-50 mm, and 2.0 mm in the range of 50-100 mm.

9. The sub-seasonal precipitation correction method based on climate zoning and sea temperature background constraints according to claim 1, characterized in that: In step S5, judgment and calculation are performed one by one for all forecast days and all grid points of each forecast start date.

10. A sub-seasonal precipitation correction system based on climate zoning and sea temperature background constraints, characterized in that: The system comprises: A data preprocessing unit is used for standardization of precipitation data and gridding of the area to be corrected; the standardization of precipitation data specifically includes: processing the cumulative precipitation reported by the sub-seasonal model and the predicted cumulative precipitation into daily average precipitation; interpolating the reanalysis precipitation data, the daily average precipitation reported by the sub-seasonal model, and the predicted daily average precipitation to grids with the same spatial resolution; A climate partitioning unit, based on the daily characteristics of summer precipitation, uses the reanalyzed precipitation data processed by the data preprocessing unit to cluster and divide the area to be corrected into a plurality of climate sub-areas; A sea temperature background constraint unit, using a sea temperature background model to add sea temperature background constraints to all the climate sub-regions; The reanalysis data modeling unit performs sliding window CDF modeling on the reanalysis precipitation data of each sub-region with the sea temperature background constraint on a certain date within a historical period, and obtains the first CDF curve of the climate sub-region under the sea temperature background constraint on the date; The model historical report data modeling unit performs sliding window CDF modeling on all model historical report data of a certain forecast day of a certain report start date for each climate sub-region with sea temperature background constraint based on the sub-seasonal model, and obtains a second CDF curve of the climate sub-region under the constraints of the report start date, the forecast date and the sea temperature background; The real-time model matching and prediction result correction unit determines the sea temperature background constraint to which the current year belongs, and projects the grid-point precipitation prediction results of a certain climate sub-region, a certain start date, and a certain forecast date onto the second CDF curve of the same climate sub-region, the same sea temperature background constraint, the same start date, and the same forecast date corresponding to step S4, to obtain the probability density percentile position corresponding to the precipitation prediction result; the probability density percentile position is matched to the first CDF curve of the same climate sub-region, the same sea temperature background constraint, and the same date as the forecast date, to obtain the reanalysis precipitation value corresponding to the probability density percentile position, and the reanalysis precipitation value is used as the corrected prediction result of the certain start date and a certain forecast date.

Images