Combined temperature forecast correction method and system
By decomposing temperature forecast data into mean, trend, and residual terms for independent correction and combination, the problem of existing methods being unable to take into account multiple sources of error is solved, thereby improving the accuracy and flexibility of temperature forecasts.
Patent Information
- Application Number
- CN202610114964.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-01-28
- Publication Date
- 2026-05-15
AI Technical Summary
Existing temperature forecast correction methods cannot simultaneously take into account climatological shifts, long-term trend evolution, and residual structure, resulting in decreased forecast reliability and insufficient practicality.
Temperature forecast data is decomposed into mean, trend, and residual terms for independent correction and then combined to generate forecast results under various correction scenarios. The trend is fitted by univariate linear regression, the residuals are corrected by conditional Gaussian, the rank order is maintained, and a combined temperature forecast result is generated.
It improves the accuracy and reliability of temperature forecasts, provides flexible correction methods applicable to different application scenarios, and enhances the application value and credibility of climate forecast products.
Smart Images
Figure CN122045765A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the technical field of weather forecasting, and more specifically, to a combined temperature forecast correction method and system. Background Technology
[0002] Climate model forecasts are crucial tools for seasonal climate prediction and interannual climate change projection, and they hold significant application value in areas such as meteorological disaster prevention and mitigation decision support, agricultural production layout optimization, refined water resource management, and energy dispatch planning. However, influenced by factors such as model physical process parameterization schemes, initial field uncertainties, and boundary condition errors, climate model forecasts generally suffer from systematic biases. These systematic biases severely restrict the direct operational application value and scientific reliability of forecast products, significantly reducing their practicality and credibility in meteorological disaster prevention and mitigation decision support and risk assessment. Correcting these systematic biases is therefore a key step in enhancing the application value and reliability of climate forecast products.
[0003] With the intensification of global climate change, temperature series exhibit significant non-stationarity. The main sources of error in temperature forecasts are climatological shifts, long-term trend evolution, and residual structure. Traditional temperature forecast correction methods struggle to simultaneously address these three characteristics. Traditional methods primarily include mean bias correction and quantile mapping. Mean bias correction simplifies calculation by adjusting the forecast mean to align with historical observation means, but it neglects the variance structure and long-term trends of the forecast and observations, leading to decreased forecast reliability. Quantile mapping, through cumulative distribution function transformation, matches the forecast distribution with the observed distribution, effectively improving probabilistic forecasting skills. However, it fails to capture the non-stationar characteristics under the background of climate change and exhibits instability with limited sample sizes. Summary of the Invention
[0004] To address the problem that existing temperature forecast correction methods struggle to accommodate multiple error sources, resulting in poor correction performance, this invention proposes a combined temperature forecast correction method and system. This method independently corrects and combines the mean, trend, and residual terms of the temperature forecast data to generate forecast results under various correction scenarios. It offers advantages such as good correction performance, flexible application, and high computational efficiency.
[0005] To achieve the above-mentioned technical effects, the technical solution of the present invention is as follows: A combined temperature forecast correction method includes the following steps: Acquire the observed temperature data and raw forecast temperature data during the training period, as well as the raw forecast temperature data during the testing period. The training period and the testing period are the periods corresponding to the training and testing of the preset temperature forecast correction model, respectively. Considering the sources of error in temperature forecasts, the observed temperature data during the training period and the original forecast temperature data are both decomposed into three independent components: mean, trend, and residual. Calculate the statistical parameter characteristics of the residual terms after decomposition, and based on the statistical parameter characteristics of the residual terms after decomposition, perform residual correction on the original forecast temperature data during the test period to obtain the residual correction results. Obtain the rank order of the set containing the raw forecast temperature data for the test period, where the set is the set of raw forecast temperature data for the test period; sort the raw forecast temperature data after residual correction in ascending order to obtain ascending-order raw forecast temperature data; rearrange the ascending-order raw forecast temperature data according to the rank order. Based on the requirements, perform mean correction, trend correction, or residual correction of the original forecast temperature data, and obtain the mean correction results, trend correction results, and residual correction results of the original forecast temperature data. The mean correction result, trend correction result, and residual correction result of the original forecast temperature data are combined to output a combined temperature correction result.
[0006] Preferably, the observed temperature data during the training period is denoted as... The raw forecast temperature data during the training period is denoted as ,in, For the time in the observed temperature data Temperature observations below, For the time in the original forecast temperature data The next Temperature forecast values for a set.
[0007] Preferably, the time center point for averaging the observed temperature data during the training period and the original forecast temperature data is expressed as follows:
[0008] in, This represents the number of data samples during the training period. For the first Time values for each data sample; The long-term trend of observed temperature data and original forecast temperature data is fitted by univariate linear regression, and the process satisfies the expression:
[0009]
[0010] in, This represents the ensemble mean of the original forecast temperature data during the training period. and These represent the slopes of the observed temperature data and the predicted temperature data obtained by fitting a linear trend line, respectively. and For the trend term of the decomposition, and These represent the observed temperature data and the predicted temperature data at the time mean point, respectively. The mean term, and These represent the residual terms of the observed temperature data and the predicted temperature data, respectively.
[0011] Preferably, the statistical parameter characteristics of the residual terms after decomposition are calculated, and the process satisfies the expression:
[0012]
[0013]
[0014] in, and These are the standard deviations of the residuals from the observed temperature data and the forecast temperature data, respectively. This is the correlation coefficient between the residual terms of the observed temperature data and the residual terms of the predicted temperature data.
[0015] Preferably, based on the statistical parameter characteristics of the residual terms after decomposition, a conditional Gaussian correction method is used to correct the residuals of the original forecast temperature data during the test period. The process satisfies the expression:
[0016]
[0017]
[0018] in, The ensemble mean of the original forecast temperature data. The standard normal distribution quantiles generated for the Weibull plotting position. It is the inverse function of the standard normal distribution. The set mean of the corrected original forecast temperature data. This is the revised forecast time.
[0019] Preferably, the rank order of the set containing the original forecast temperature data for the testing period is obtained, and the process satisfies the expression:
[0020] in, This function returns the index of each element in the original array after sorting the array in ascending order. The original forecast temperature data after residual correction are sorted in ascending order to obtain the ascending-order original forecast temperature data. The process satisfies the expression:
[0021] in, This function sorts the elements in an array in ascending order and returns the sorted array. The original forecast temperature data are rearranged in ascending order according to rank, and the process satisfies the expression:
[0022] This indicates rearranging the original forecast temperature data. The revised forecast temperature data has the same set structure as the original forecast temperature data.
[0023] Preferably, the raw forecast data during the test period Satisfies the expression: ; When only the mean of the original forecast temperature data is corrected, the expression is satisfied: ; When only trend correction is performed on the raw forecast temperature data, the expression is satisfied: ; When only residual correction is performed on the original forecast temperature data, the expression is satisfied: ; in, , , This represents the mean of the original forecast temperature data. This represents the mean of the original forecast temperature data after mean correction. This represents the trend term of the original forecast temperature data. This represents the trend term after trend correction of the original forecast temperature data. This represents the residual term of the original forecast temperature data. This represents the residual term after residual correction of the original forecast temperature data.
[0024] Preferably, the mean correction result, trend correction result, and residual correction result of the original forecast temperature data are combined, including: Combining mean correction and trend correction satisfies the expression: ; Combining mean correction and residual correction, the expression is satisfied: ; Combining trend correction and residual correction satisfies the expression: ; Combining mean correction, trend correction, and residual correction, the expression is satisfied: .
[0025] Preferably, the method further includes: storing the combined temperature correction results in NetCDF format and evaluating the correction effect using a visualization analysis tool.
[0026] This invention also proposes a combined temperature forecast correction system to implement the above-mentioned combined temperature forecast correction method, the system comprising: The data processing module is used to acquire the observed temperature data and raw forecast temperature data during the training period, as well as the raw forecast temperature data during the testing period. The training period and the testing period are the periods corresponding to the training and testing of the preset temperature forecast correction model, respectively. Considering the sources of error in temperature forecast, the observed temperature data and raw forecast temperature data during the training period are decomposed into three independent components: mean, trend, and residual. The parameter calculation module is used to calculate the statistical parameter characteristics of the residual terms after decomposition. An independent correction module is used to correct the mean and trend terms separately, obtaining mean correction results and trend correction results. Based on the statistical parameter characteristics of the decomposed residual terms, residual correction is performed on the original forecast temperature data to obtain residual correction results. The rank order of the set containing the original forecast temperature data of the test period is obtained, where the set is the set of original forecast temperature data of the test period. The original forecast temperature data after residual correction is sorted in ascending order to obtain ascending-order original forecast temperature data. The ascending-order original forecast temperature data is then rearranged according to the rank order. The multi-dimensional correction module is used to combine the mean correction results, trend correction results, and residual correction results of the original forecast temperature data to output a combined temperature forecast correction result.
[0027] Compared with the prior art, the beneficial effects of the technical solution of the present invention are: This invention proposes a combined temperature forecast correction method and system. It acquires observed temperature data and raw forecast temperature data during the training period, as well as raw forecast temperature data during the test period, providing a data foundation for subsequent operations. Considering the sources of temperature forecast error, both the observed temperature data during the training period and the raw forecast temperature data are decomposed into three independent components: mean, trend, and residual, making the correction more targeted. The statistical parameter characteristics of the decomposed residuals are calculated, and based on these characteristics, the raw forecast temperature data during the test period is further corrected. The method involves error correction to obtain residual correction results, effectively improving forecast performance. The members of the corrected original forecast temperature data set are rearranged according to rank order to maintain the rank correlation structure of the corrected original forecast temperature data set, ensuring reasonable spatial consistency and probabilistic characteristics in the corrected ensemble forecast. Based on requirements, three error components are independently corrected. The mean correction result, trend correction result, and residual correction result of the original forecast temperature data are combined to output a combined temperature forecast correction result. This combined correction method allows users to flexibly select the optimal combination according to specific application scenarios and can be used to diagnose the independent contribution of each correction component to the final forecast skill. This invention provides a temperature forecast correction method with clear physical meaning and good correction effect. Attached Figure Description
[0028] Figure 1 A flowchart illustrating a combined temperature forecast correction method proposed in an embodiment of the present invention; Figure 2 This diagram illustrates the structure of a combined temperature forecast correction system proposed in this embodiment of the invention. Figure 3 This represents the fitting and statistical characteristic graph of the temperature forecast correction parameters proposed in this embodiment of the invention; Figure 4 This diagram shows the comparison between the predicted and observed values of the eight combined temperature forecast correction methods proposed in this embodiment of the invention. Figure 5 This diagram shows a comparison of CRPS scores and a diagnostic chart of Shapley value error sources for the eight combined temperature forecast correction methods proposed in this invention. Detailed Implementation
[0029] The accompanying drawings are for illustrative purposes only and should not be construed as limiting the scope of this patent. To better illustrate this embodiment, some parts of the accompanying drawings may be omitted, enlarged, or reduced, and do not represent the actual dimensions; It is understandable to those skilled in the art that some well-known details may be omitted from the accompanying drawings.
[0030] The technical solution of the present invention will be further described below with reference to the accompanying drawings and embodiments.
[0031] The positional relationships depicted in the accompanying drawings are for illustrative purposes only and should not be construed as limiting this patent. Example 1 This embodiment proposes a combined temperature forecast correction method, see [link to relevant documentation]. Figure 1 This includes the following steps: S1: Obtain the observed temperature data and raw forecast temperature data during the training period, as well as the raw forecast temperature data during the testing period. The training period and the testing period are the periods associated with training and testing the preset temperature forecast correction model. S2: Considering the sources of error in temperature forecasts, both the observed temperature data during the training period and the original forecast temperature data are decomposed into three independent components: mean, trend, and residual. In this embodiment, the decomposition of the mean, trend, and residual is as follows: Observed temperature data and original forecast temperature data from multiple years of training are collected. These data are then divided into monthly units. For each monthly subsequence, a univariate linear regression is performed using the least squares method. Linear regression is then performed on the time series data of the observed and original forecast temperatures to obtain a linear trend line and the residual. The value of the linear trend line at the time mean is used as the mean, and the value on the linear trend line minus the mean is used as the trend.
[0032] S3: Calculate the statistical parameter characteristics of the residual terms after decomposition. Based on the statistical parameter characteristics of the residual terms after decomposition, perform residual correction on the original forecast temperature data during the test period to obtain the residual correction results. S4: Obtain the rank order of the set containing the raw forecast temperature data for the test period. The set is composed of the raw forecast temperature data for the test period. Sort the raw forecast temperature data after residual correction in ascending order to obtain the ascending-order raw forecast temperature data. Rearrange the ascending-order raw forecast temperature data according to the rank order. S5: As needed, perform mean correction, trend correction, or residual correction of the original forecast temperature data, and obtain the mean correction result, trend correction result, and residual correction result of the original forecast temperature data. S6: Combine the mean correction results, trend correction results, and residual correction results of the original forecast temperature data to output a combined temperature forecast correction result.
[0033] In this embodiment, the mean, trend, and residual of the data are decomposed into three parts that can be independently added or subtracted. By separating and independently processing the mean term error, trend term error, and residual term error, three mutually independent correction processes are constructed. Finally, the independent correction results are combined to form a complete set of correction schemes, which covers all states from the original forecast to complete conditional Gaussian distribution correction. This method is used for temperature forecast correction and has the advantages of good correction effect, flexible use, and high computational efficiency.
[0034] Example 2 This embodiment proposes a combined temperature forecast correction method based on Embodiment 1. The observed temperature data during the training period are denoted as follows: The raw forecast temperature data during the training period is denoted as ,in, For the time in the observed temperature data Temperature observations below, For the time in the original forecast temperature data The next The system uses a set of temperature forecast values. In this embodiment, the original forecast time series data and observed time series data are used as input training data. Commonly used formats for input and output data include CSV, NetCDF, etc. First, the training data is read using Python, serving as the training basis for the temperature forecast correction model.
[0035] The expression for calculating the time center point of the average of the observed temperature data and the original forecast temperature data during the training period is as follows:
[0036] in, This represents the number of data samples during the training period. For the first Time values for each data sample; The long-term trend of observed temperature data and original forecast temperature data is fitted by univariate linear regression, and the process satisfies the expression:
[0037]
[0038] in, This represents the ensemble mean of the original forecast temperature data during the training period. and These represent the slopes of the observed temperature data and the predicted temperature data obtained by fitting a linear trend line, respectively. and For the trend term of the decomposition, and These represent the observed temperature data and the predicted temperature data at the time mean point, respectively. The mean term, and These represent the residual terms of the observed temperature data and the predicted temperature data, respectively.
[0039] In this embodiment, the average time center point is used to avoid calculation errors caused by excessively large time values; the trend term is used to correct the long-term trend rate deviation of the forecast, the mean term is used to correct the systematic deviation of the forecast, and the residual term is used to support the subsequent conditional Gaussian residual correction process. By calculating the ensemble mean of the original forecast temperature data, the high-dimensional ensemble forecast is reduced to a single-value sequence, matching the dimension of the observed data and ensuring the feasibility of subsequent regression analysis.
[0040] In an optional embodiment, the statistical parameter characteristics of the residual terms after decomposition are calculated, and the process satisfies the expression:
[0041]
[0042]
[0043] in, and These are the standard deviations of the residuals from the observed temperature data and the forecast temperature data, respectively. This is the correlation coefficient between the residual terms of the observed temperature data and the residual terms of the predicted temperature data.
[0044] Based on the statistical parameter characteristics of the residuals after decomposition, a conditional Gaussian correction method is used to correct the residuals of the original forecast temperature data during the test period. The process satisfies the expression:
[0045]
[0046]
[0047] in, The ensemble mean of the original forecast temperature data. The standard normal distribution quantiles generated for the Weibull plotting position. It is the inverse function of the standard normal distribution. The set mean of the corrected original forecast temperature data. This is the revised forecast time.
[0048] In this embodiment, by calculating the statistical characteristics of the residuals, the conditional Gaussian correction method is used to accurately correct the residual terms. By combining the conditional mean and standard deviation, the corrected residual set members are generated, which not only corrects the systematic bias but also preserves the randomness and diversity of the ensemble forecast, thereby improving the accuracy and probabilistic reliability of temperature forecasts.
[0049] In an optional embodiment, the rank order of the set containing the raw forecast temperature data for the testing period is obtained, and the process satisfies the expression:
[0050] in, This function returns the index of each element in the original array after sorting the array in ascending order. The original forecast temperature data after residual correction are sorted in ascending order to obtain the ascending-order original forecast temperature data. The process satisfies the expression:
[0051] in, This function sorts the elements in an array in ascending order and returns the sorted array. The original forecast temperature data are rearranged in ascending order according to rank, and the process satisfies the expression:
[0052] This indicates rearranging the original forecast temperature data. The revised forecast temperature data has the same set structure as the original forecast temperature data.
[0053] In this embodiment, the relative order of the original set members may reflect the uncertainty of the physical process. The Schaake Shuffle method is used to ensure that this structure is not destroyed after correction. By preserving the rank order, the corrected set can still accurately reflect the range of forecast uncertainty.
[0054] In an optional embodiment, the raw forecast data during the test period Satisfies the expression: ; When only the mean of the original forecast temperature data is corrected, the expression is satisfied: ; When only trend correction is performed on the raw forecast temperature data, the expression is satisfied: ; When only residual correction is performed on the original forecast temperature data, the expression is satisfied: ; in, , , This represents the mean of the original forecast temperature data. This represents the mean of the original forecast temperature data after mean correction. This represents the trend term of the original forecast temperature data. This represents the trend term after trend correction of the original forecast temperature data. This represents the residual term of the original forecast temperature data. This represents the residual term after residual correction of the original forecast temperature data.
[0055] In this embodiment, the three individual corrections correspond to the three independent components of error decomposition. The correction target can be selected according to actual needs to avoid overcorrection or ineffective correction.
[0056] In an optional embodiment, the mean correction result, trend correction result, and residual correction result of the original forecast temperature data are combined, including: Combining mean correction and trend correction satisfies the expression: ; Combining mean correction and residual correction, the expression is satisfied: ; Combining trend correction and residual correction satisfies the expression: ; Combining mean correction, trend correction, and residual correction, the expression is satisfied: .
[0057] The combined temperature correction results are stored in NetCDF format, and the correction effect is evaluated using visualization analysis tools.
[0058] In this embodiment, the mean correction, trend correction, and residual correction results of the original forecast temperature data are combined. This approach integrates the advantages of each correction method, more comprehensively eliminating forecast errors and improving the accuracy of temperature forecasts. Utilizing visualization analysis tools to evaluate the correction effect allows for a clear view of the changes in temperature forecasts before and after correction, providing a basis for further improving forecast models and correction methods, thereby enhancing the quality of temperature forecast corrections.
[0059] Example 3 This embodiment proposes a combined temperature forecast correction system, see [link to documentation]. Figure 2It includes a data processing module, a parameter calculation module, an independent correction module, and a multi-dimensional correction module. The data processing module is used to acquire the observed temperature data and raw forecast temperature data during the training period, as well as the raw forecast temperature data during the testing period. The training period and the testing period are the periods corresponding to the pre-set temperature forecast correction model during training and testing, respectively. Considering the sources of error in temperature forecast, the observed temperature data and raw forecast temperature data during the training period are decomposed into three independent components: mean, trend, and residual.
[0060] The parameter calculation module is used to calculate the statistical parameter characteristics of the residual terms after decomposition.
[0061] The independent correction module is used to correct the mean and trend terms separately, obtaining mean correction results and trend correction results. Based on the statistical parameter characteristics of the decomposed residual terms, it performs residual correction on the original forecast temperature data, obtaining residual correction results. It obtains the rank order of the set containing the original forecast temperature data of the test period, which is the set composed of the original forecast temperature data of the test period. It sorts the original forecast temperature data after residual correction in ascending order, obtaining ascending-order original forecast temperature data. It rearranges the ascending-order original forecast temperature data according to the rank order.
[0062] The multi-dimensional correction module combines the mean correction results, trend correction results, and residual correction results of the original forecast temperature data to output a combined temperature forecast correction result.
[0063] In this embodiment, the data processing module serves as the system's entry point, providing decomposed independent components for subsequent modules. The parameter calculation module calculates the statistical parameters of the residual terms, determining the accuracy of the residual correction. The independent correction module independently corrects the mean, trend, and residual terms, significantly improving the correction effect for the main sources of error in temperature forecasts while maintaining the ensemble rank structure, ensuring that the corrected ensemble forecast retains reasonable spatial consistency and probabilistic characteristics. The multi-dimensional correction module combines the component correction results to generate diverse correction schemes, supporting on-demand correction. It allows for flexible selection of the optimal combination based on specific application scenarios and is used to diagnose the independent contribution of each correction component to the final forecast skill. This system constructs an automated data analysis system that generates corrected forecasts based on the input raw forecast data.
[0064] Example 4 This embodiment is based on the method proposed in Embodiment 1 and is used for statistical post-processing of climate model temperature forecast data. This embodiment uses NCEP Climate Forecast System Version 2 (CFSv2) monthly forecast products and Extended Reconstructed Sea Surface Temperature Version 6 (ERSSTv6) observational data as inputs to establish a global ocean surface temperature correction model. The specific steps are as follows: S1: Use the open_dataset function in the Xarray library to read the preprocessed CFSv2 data. The file contains sea surface temperature data for 24 ensemble members. After reading, the data is stored in the cfs_data dictionary. Use the open_dataset function in the Xarray library to read the ERSSTv6 observation data. Select the time range from 1993 to 2022 and load it into memory and store it in the da_obs variable. Create an output directory structure to store the correction results and correction parameters.
[0065] S2: Determine the training window for the forecast start month (June) and target year (2023). Taking June 2023 as the target month as an example, when the forecast period is 0 months, extract the data from June 1993 to 2022 for this grid as training data, and use the original forecast for June 2023 as test data. This embodiment uses a 30-year sliding training window, which can capture long-term climate change trends while adapting to recent climate conditions, and has good adaptability to non-stationary climate sequences.
[0066] For each spatial grid point, extract the forecast set data from the training period. With observation data and the corresponding time series ; The `train_model` function is used for parameter estimation. This function performs the following operations: a) Calculate the average value of the forecast ensemble; b) Construct an effective data mask to exclude samples containing NaN; c) Fit the linear trend of observed data to forecast data using np.polyfit. ); d) Calculate the time center point t_m and its corresponding mean ( ); e) Calculate the residual sequence after detrending and mean removal; f) Estimate the standard deviation of the residuals ( ) and correlation coefficient .
[0067] Returns a dictionary containing 7 key parameters.
[0068] S3: For the initial month of the target year 2023, June, extract the test forecast data f_test_np and the test period time. The `predict` function is used for multi-dimensional correction. This function performs the following operations: a) Gaussian condition correction: Calculate the conditional mean Conditional standard deviation Generate quantile sequences based on weibullplotting positions. ; b) Schaake Shuffle: (for...) The results were then rearranged using the schaake shuffle function, preserving the original set's rank structure, resulting in the rearranged set. ; c) Calculate the mean of the original forecasts for the test period. ,trend residual Mean of the revised forecast after the test period ,trend residual And the correction results were generated when eight combinations were generated.
[0069] A multi-process parallel processing framework is employed to process data from different starting months simultaneously, balancing computational efficiency and system load. This embodiment is developed using Python and utilizes libraries such as multiprocessing to achieve efficient parallel computing, demonstrating significant performance advantages when processing high-resolution global data.
[0070] S4: Organize the correction results into an xarray Dataset according to the method, and save the xarray Dataset results in NetCDF format. Save the parameter dataset in the same way, containing 7 key parameters. Finally, use the Python visualization libraries Matplotlib and Cartopy to draw comparison graphs before and after correction.
[0071] Figure 3 This demonstrates the differences between the original forecast and observed values in terms of mean, trend, and residuals. Figure 3 It can be seen that there are significant differences between the original temperature forecast and the observations in terms of mean, trend, and residual distribution. The original forecast systematically underestimates the mean compared to the observed values. The original forecast overestimates the temperature's upward trend over time compared to the observed values, and the residual distribution of the original forecast is more concentrated than that of the observations.
[0072] Figure 4The figure shows the comparison between forecast and observation values for eight combined temperature forecast correction methods. As can be seen from the figure, the probability forecast score (CRPS) decreases as the three different types of errors are gradually corrected.
[0073] Figure 5 This paper presents a comparison of CRPS scores for eight combined temperature forecast correction methods and diagnoses the sources of error in Shapley values. Figure 5 As can be seen, during the training phase, Shapley values show that all three error correction processes contribute to the reduction of CRPS.
[0074] The embodiments described are merely examples to clearly illustrate the present invention and are not intended to limit the implementation of the invention. Those skilled in the art will recognize that other variations or modifications can be made based on the above description. It is neither necessary nor possible to exhaustively describe all possible implementations. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the claims of the present invention.
Claims
1. A combined temperature forecast correction method, characterized in that, Includes the following steps: Acquire the observed temperature data and raw forecast temperature data during the training period, as well as the raw forecast temperature data during the testing period. The training period and the testing period are the periods corresponding to the training and testing of the preset temperature forecast correction model, respectively. Considering the sources of error in temperature forecasts, the observed temperature data during the training period and the original forecast temperature data are both decomposed into three independent components: mean, trend, and residual. Calculate the statistical parameter characteristics of the residual terms after decomposition, and based on the statistical parameter characteristics of the residual terms after decomposition, perform residual correction on the original forecast temperature data during the test period; Obtain the rank order of the set containing the raw forecast temperature data for the test period, where the set is the set of raw forecast temperature data for the test period; sort the raw forecast temperature data after residual correction in ascending order to obtain the ascending-order raw forecast temperature data. Reorder the original forecast temperature data in ascending order according to rank; Based on the requirements, perform mean correction, trend correction, or residual correction of the original forecast temperature data, and obtain the mean correction results, trend correction results, and residual correction results of the original forecast temperature data. The mean correction result, trend correction result, and residual correction result of the original forecast temperature data are combined to output a combined temperature correction result.
2. The combined temperature forecast correction method according to claim 1, characterized in that, The observed temperature data during the training period are denoted as The raw forecast temperature data during the training period is denoted as ,in, For the time in the observed temperature data Temperature observations below, For the time in the original forecast temperature data The next Temperature forecast values for a set.
3. The combined temperature forecast correction method according to claim 2, characterized in that, The expression for calculating the time center point of the average of the observed temperature data and the original forecast temperature data during the training period is as follows: in, This represents the number of data samples during the training period. For the first Time values for each data sample; The long-term trend of observed temperature data and original forecast temperature data is fitted by univariate linear regression, and the process satisfies the expression: in, This represents the ensemble mean of the original forecast temperature data during the training period. and These represent the slopes of the observed temperature data and the predicted temperature data obtained by fitting a linear trend line, respectively. and For the trend term of the decomposition, and These represent the observed temperature data and the predicted temperature data at the time mean point, respectively. The mean term, and These represent the residual terms of the observed temperature data and the predicted temperature data, respectively.
4. The combined temperature forecast correction method according to claim 3, characterized in that, The statistical parameter characteristics of the residual terms after decomposition are calculated, and the process satisfies the expression: in, and These are the standard deviations of the residuals from the observed temperature data and the forecast temperature data, respectively. This is the correlation coefficient between the residual terms of the observed temperature data and the residual terms of the predicted temperature data.
5. The combined temperature forecast correction method according to claim 4, characterized in that, Based on the statistical parameter characteristics of the residuals after decomposition, a conditional Gaussian correction method is used to correct the residuals of the original forecast temperature data during the test period. The process satisfies the expression: in, The ensemble mean of the original forecast temperature data. The standard normal distribution quantiles generated for the Weibull plotting position. It is the inverse function of the standard normal distribution. The set mean of the corrected original forecast temperature data. This is the revised forecast time.
6. The combined temperature forecast correction method according to claim 1, characterized in that, Obtain the rank order of the set containing the raw forecast temperature data for the test period, satisfying the expression: in, This function returns the index of each element in the original array after sorting the array in ascending order. The original forecast temperature data after residual correction are sorted in ascending order to obtain the ascending-order original forecast temperature data. The process satisfies the expression: in, This function sorts the elements in an array in ascending order and returns the sorted array. The original forecast temperature data are rearranged in ascending order according to rank, and the process satisfies the expression: This indicates rearranging the original forecast temperature data. The revised forecast temperature data has the same set structure as the original forecast temperature data.
7. The combined temperature forecast correction method according to claim 1, characterized in that, Let the raw forecast data for the test period be Satisfies the expression: ; When only the mean of the original forecast temperature data is corrected, the expression is satisfied: ; When only trend correction is performed on the raw forecast temperature data, the expression is satisfied: ; When only residual correction is performed on the original forecast temperature data, the expression is satisfied: ; in, , , This represents the mean of the original forecast temperature data. This represents the mean of the original forecast temperature data after mean correction. This represents the trend term of the original forecast temperature data. This represents the trend term after trend correction of the original forecast temperature data. This represents the residual term of the original forecast temperature data. This represents the residual term after residual correction of the original forecast temperature data.
8. The combined temperature forecast correction method according to claim 7, characterized in that, The mean correction results, trend correction results, and residual correction results of the original forecast temperature data are combined, including: Combining mean correction and trend correction satisfies the expression: ; Combining mean correction and residual correction, the expression is satisfied: ; Combining trend correction and residual correction satisfies the expression: ; Combining mean correction, trend correction, and residual correction, the expression is satisfied: .
9. The combined temperature forecast correction method according to claim 1, characterized in that, Also includes: The combined temperature correction results are stored in NetCDF format, and the correction effect is evaluated using visualization analysis tools.
10. A combined temperature forecast correction system, used to implement the combined temperature forecast correction method according to any one of claims 1 to 9, characterized in that, include: The data processing module is used to acquire the observed temperature data and raw forecast temperature data during the training period, as well as the raw forecast temperature data during the testing period. The training period and the testing period are the periods corresponding to the training and testing of the preset temperature forecast correction model, respectively. Considering the sources of error in temperature forecast, the observed temperature data and raw forecast temperature data during the training period are decomposed into three independent components: mean, trend, and residual. The parameter calculation module is used to calculate the statistical parameter characteristics of the residual terms after decomposition. An independent correction module is used to correct the mean and trend terms separately, obtaining mean correction results and trend correction results. Based on the statistical parameter characteristics of the decomposed residual terms, residual correction is performed on the original forecast temperature data to obtain residual correction results. The rank order of the set containing the original forecast temperature data of the test period is obtained, where the set is the set of original forecast temperature data of the test period. The original forecast temperature data after residual correction is sorted in ascending order to obtain ascending-order original forecast temperature data. The ascending-order original forecast temperature data is then rearranged according to the rank order. The multi-dimensional correction module is used to combine the mean correction results, trend correction results, and residual correction results of the original forecast temperature data to output a combined temperature correction result.