A method for correcting climate data errors

Abstract

This invention discloses a climate data error correction method, relating to the field of meteorological data processing technology. The method includes: identifying a target region; acquiring CMIP6 global climate model data and ERA5 reanalysis data for the target region; unifying the computational grids of the two datasets; correcting the data sequence in the time dimension using wavelet analysis, wavelet decomposition, quantile mapping, and wavelet reconstruction; correcting the spatial sequence in the longitude direction using wavelet analysis, quantile mapping, and wavelet reconstruction; and correcting the spatial sequence in the latitude direction using wavelet analysis, quantile mapping, and wavelet reconstruction; and saving the time, longitude, and latitude correction fields as standard data files. This invention can significantly reduce the systematic bias of CMIP6 data, improve extreme value performance and spatial resolution, and provide a reliable data foundation for high-precision downscaling simulations of future climate scenarios.

Description

Technical Field

[0001] This invention relates to the field of meteorological data processing technology, and in particular to a method for correcting climate data errors. Background Technology

[0002] Global climate models are core tools for understanding and predicting future climate change, such as the CMIP6 data from the Sixth Coupled Model Intercomparison Project. However, due to limitations in computational resources and simplifications in the parameterization of physical processes, global climate models typically operate at relatively low spatial resolutions, leading to biases in their simulations of regional-scale climate characteristics and making them difficult to directly use for regional-scale climate change impact assessments and adaptation planning.

[0003] Reanalysis data, which integrates global observations and numerical models, provides a high spatiotemporal resolution and physically consistent climate field, such as ECMWF's ERA5 data, which is often used as a reference benchmark for regional climate studies. To apply large-scale information from global climate models to regional scales, downscaling is necessary. Traditional downscaling methods include dynamic downscaling and statistical downscaling. Dynamic downscaling is computationally expensive, while statistical downscaling methods typically focus on corrections for mean-state conditions and have limited ability to correct for extreme events and climate variability across different time scales.

[0004] In existing technologies, quantile mapping is a commonly used bias correction method that can effectively correct the probability distribution of variables. However, it is usually performed on a single time scale and fails to fully consider the fluctuation characteristics of climate variables at different time scales. Wavelet analysis has excellent time-frequency localization capabilities and can reveal the evolution characteristics of signals at different time scales, but simple wavelet analysis lacks a direct correction mechanism for system bias.

[0005] In view of this, this invention is hereby proposed. Summary of the Invention

[0006] The purpose of this invention is to address the shortcomings of existing technologies by proposing a climate data error correction method that integrates wavelet analysis and quantile mapping, thereby achieving accurate matching between climate model output and reanalysis data in both time and space dimensions, and significantly improving the spatiotemporal accuracy of climate simulation results.

[0007] To achieve the above objectives, the present invention adopts the following technical solution:

[0008] A method for correcting errors in climate data includes the following steps:

[0009] S1: Determine the target area, acquire CMIP6 global climate model data and ERA5 reanalysis data for the target area, and unify the computational grids of the two types of data;

[0010] S2: In the time dimension, wavelet analysis is used to extract the primary and secondary fluctuation periods of each meteorological element in CMIP6 and ERA5 data respectively. The data is decomposed into different period sequences through wavelet decomposition. Error correction functions under each period are established using quantile mapping to correct each period sequence of CMIP6. Finally, wavelet reconstruction is performed to complete the time dimension correction.

[0011] S3: Based on the data corrected in the time dimension of S2, the longitude sequence is corrected in the longitude direction using wavelet analysis, wavelet decomposition, quantile mapping and wavelet reconstruction methods;

[0012] S4: Based on the data corrected in the longitude direction by S3, the latitude sequence is corrected in the latitude direction using wavelet analysis, wavelet decomposition, quantile mapping and wavelet reconstruction methods;

[0013] S5: Save the time correction field, longitude correction field, and latitude correction field as standard data files.

[0014] Preferably, step S1, which unifies the computational grids of the two types of data, includes the following steps:

[0015] S11: Unify the horizontal resolution of CMIP6 data and ERA5 data, that is, interpolate CMIP6 data to the horizontal grid of ERA5 data;

[0016] S12: Under the premise of conforming to the vertical physical relationship of the atmosphere, a three-dimensional interpolation function is used to interpolate the CMIP6 data to the ERA5 vertical grid;

[0017] S13: Consistentize time and units, ensure consistent time range for ERA5 and CMIP6 data, and convert temperature units from Kelvin (K) to Celsius (°C).

[0018] Preferably, the bilinear interpolation function is used to achieve uniformity in horizontal resolution as described in S11.

[0019] Preferably, step S2 includes the following steps:

[0020] S21: Using wavelet analysis, the main and secondary fluctuation periods of each meteorological element in the time dimension of CMIP6 and ERA5 data are analyzed respectively. The meteorological elements include: surface temperature, air pressure, precipitation, relative humidity, wind speed, and ground temperature.

[0021] S22: Using the wavelet decomposition method, the Daubechies wavelet basis (db5) is adopted, and the decomposition layer is 11 layers. The two types of data are decomposed into time series under each fluctuation period.

[0022] S23: For each fluctuation period sequence obtained from the decomposition in S22, a mapping function for each fluctuation period is established using the quantile mapping method:

[0023] S24: The time-dimension corrected meteorological element data are obtained by using wavelet reconstruction method for each periodic sequence after correction in S23.

[0024] Preferably, the mapping function in S23 is:

[0025] ;

[0026] in and These are the cumulative distribution functions for the pattern data and the reanalysis data, respectively.

[0027] Preferably, step S3 includes the following steps:

[0028] S31: Using the data obtained in S2 after time dimension correction, extract the one-dimensional data sequence in the longitude direction for each time step;

[0029] S32: For each set of longitude data sequences extracted in S31, wavelet analysis is performed to identify its main spatial fluctuation scale, and wavelet decomposition is performed to obtain spatial sequences at different scales.

[0030] S33: For the spatial sequences of different scales obtained by S32 decomposition, establish a mapping function between global climate model data and target reanalysis data, and make corrections;

[0031] S34: Perform wavelet reconstruction on the corrected spatial sequences at each scale to obtain the longitude correction field.

[0032] Preferably, step S4 includes the following steps:

[0033] S41: Using the data obtained in S3 after longitude correction, extract the one-dimensional data sequence in the latitude direction for each time step;

[0034] S42: For each set of latitude data sequences extracted in S41, wavelet analysis is performed to identify its main spatial fluctuation scale, and wavelet decomposition is performed to obtain spatial sequences at different scales.

[0035] S43: For the spatial sequences of different scales obtained by S42 decomposition, establish a mapping function between global climate model data and target reanalysis data, and make corrections;

[0036] S44: Perform wavelet reconstruction on the corrected spatial sequences at each scale to obtain the latitudinal correction field.

[0037] Preferably, the standard data file in S5 includes coordinate information, time index, and correction parameter metadata.

[0038] Compared with the prior art, the beneficial effects of this invention are as follows:

[0039] 1. Multi-level error correction: By combining wavelet analysis with quantile mapping, fine-grained error correction of climate data at different time scales is achieved, which not only improves the average state, but also enhances the simulation ability of climate fluctuations and extreme events at different scales such as interannual variability and intraseasonal oscillations.

[0040] 2. High spatiotemporal consistency: After completing the temporal dimension correction, similar scale separation and correction are further performed in the spatial dimension, which effectively eliminates the systematic bias in the spatial distribution of climate model data and ensures the consistency of the corrected data in the spatiotemporal dimensions.

[0041] 3. Wide applicability: This method can be widely applied to downscaling correction of various meteorological elements such as surface temperature, precipitation, air pressure, and wind speed, providing a high-quality data foundation for research on the impacts of climate change in fields such as hydrology, ecology, and agriculture.

[0042] 4. High degree of automation: The method and process are clear, and the programmable process can realize automated processing, which is suitable for post-processing of large amounts of climate model data and product generation. Attached Figure Description

[0043] Figure 1 This is a flowchart of the climate data error correction method of the present invention;

[0044] Figure 2 To verify the wavelet spectrum of the climate data in Example 1;

[0045] Figure 3 To verify the wavelet decomposition structure diagram of Example 1;

[0046] Figure 4 To verify the approximate and detail wave patterns after wavelet decomposition in Example 1;

[0047] Figure 5 To verify the quantile mapping diagram of Example 1;

[0048] Figure 6 To verify the representative point distribution map of Example 1;

[0049] Figure 7 To verify the time-series comparison of surface temperature at point A in Example 1;

[0050] Figure 8 To verify the time-series comparison of surface temperature at point B in Example 1;

[0051] Figure 9 To verify the time-series comparison of surface temperature at point C in Example 1;

[0052] Figure 10 To verify the spatial distribution of surface temperature at time T1 in Example 1;

[0053] Figure 11 To verify the spatial distribution of surface temperature at time T61 in Example 1;

[0054] Figure 12 To verify the spatial distribution of surface temperature at time T361 in Example 1. Detailed Implementation

[0055] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0056] Example 1:

[0057] like Figure 1 The diagram shows a flowchart of the climate data error correction method of the present invention; a climate data error correction method includes the following steps:

[0058] S1: Determine the target area, acquire CMIP6 global climate model data and ERA5 reanalysis data for the target area, and unify the computational grids of the two types of data;

[0059] In this preferred embodiment, unifying the computational grid for the two types of data includes the following steps:

[0060] S11: Unify the horizontal resolution of CMIP6 data and ERA5 data, that is, interpolate CMIP6 data to the horizontal grid of ERA5 data. In this embodiment, a bilinear interpolation function is used to achieve this; other methods can also be used in practical applications.

[0061] S12: Under the premise of conforming to the vertical physical relationship of the atmosphere, this embodiment uses a three-dimensional interpolation function to interpolate CMIP6 data to the ERA5 vertical grid; other methods can also be used in practical applications.

[0062] S13: Consistentize time and units, ensure consistent time range for ERA5 and CMIP6 data, and convert temperature units from Kelvin (K) to Celsius (°C).

[0063] S2: In the time dimension, wavelet analysis is used to extract the primary and secondary fluctuation periods of each meteorological element in the CMIP6 and ERA5 data respectively. Wavelet decomposition is used to decompose the data into different periodic sequences. Quantile mapping is used to establish error correction functions for each period, and the periodic sequences of CMIP6 are corrected. Finally, wavelet reconstruction is performed to complete the time dimension correction. This embodiment specifically includes the following steps:

[0064] S21: Using wavelet analysis, analyze the main and secondary fluctuation periods of each meteorological element in the time dimension of CMIP6 and ERA5 data respectively.

[0065] S22: Using the wavelet decomposition method, the Daubechies wavelet basis (db5) is adopted, and the decomposition layer is 11 layers. The two types of data are decomposed into time series under each fluctuation period.

[0066] S23: For each fluctuation period sequence obtained from the decomposition in S22, a mapping function for each fluctuation period is established using the quantile mapping method:

[0067] ;

[0068] in and These are the cumulative distribution functions for the pattern data and the reanalysis data, respectively;

[0069] S24: The time-dimension corrected meteorological element data are obtained by using wavelet reconstruction method for each periodic sequence after correction in S23.

[0070] S3: Based on the time-dimensional correction data from S2, the longitude sequence is corrected using wavelet analysis, wavelet decomposition, quantile mapping, and wavelet reconstruction methods. This embodiment specifically includes the following steps:

[0071] S31: Using the data obtained in S2 after time dimension correction, extract the one-dimensional data sequence in the longitude direction for each time step;

[0072] S32: For each set of longitude data sequences extracted in S31, wavelet analysis is performed to identify its main spatial fluctuation scale, and wavelet decomposition is performed to obtain spatial sequences at different scales.

[0073] S33: For the spatial sequences of different scales obtained by S32 decomposition, establish a mapping function between global climate model data and target reanalysis data, and make corrections;

[0074] S34: Perform wavelet reconstruction on the corrected spatial sequences at each scale to obtain the longitude correction field.

[0075] S4: Based on the longitude correction data from S3, wavelet analysis, wavelet decomposition, quantile mapping, and wavelet reconstruction are used to correct the latitude sequence. This embodiment specifically includes the following steps:

[0076] S41: Using the data obtained in S3 after longitude correction, extract the one-dimensional data sequence in the latitude direction for each time step;

[0077] S42: For each set of latitude data sequences extracted in S41, wavelet analysis is performed to identify its main spatial fluctuation scale, and wavelet decomposition is performed to obtain spatial sequences at different scales.

[0078] S43: For the spatial sequences of different scales obtained by S42 decomposition, establish a mapping function between global climate model data and target reanalysis data, and make corrections;

[0079] S44: Perform wavelet reconstruction on the corrected spatial sequences at each scale to obtain the latitudinal correction field.

[0080] S5: Save the time correction field, longitude correction field, and latitude correction field as standard data files. The standard data files described in this embodiment include coordinate information, time index, and correction parameter metadata.

[0081] The climate data error correction method in this embodiment addresses the systematic bias between global climate model data and high-precision reanalysis data in terms of spatiotemporal distribution. It proposes a comprehensive correction scheme that combines multi-scale decomposition and distribution mapping to achieve accurate matching between climate model output and reanalysis data in both time and space dimensions, thereby significantly improving the spatiotemporal accuracy of climate simulation results.

[0082] Based on this, the climate data error correction method in this embodiment can be widely applied to downscaling correction of various meteorological elements such as surface temperature, precipitation, air pressure, and wind speed, providing a high-quality data foundation for research on the impacts of climate change in fields such as hydrology, ecology, and agriculture.

[0083] This embodiment preferably uses the MATLAB scientific computing platform to read, crop, interpolate, matrix index, and extract sequences from ERA5 and CMIP6 NetCDF meteorological data. MATLAB has mature wavelet analysis, spatial interpolation, and statistical computing toolboxes, which can realize the steps required in this method, such as interpolation unification, time series decomposition, probability distribution mapping, and reconstruction. In practical applications, other methods can also be used.

[0084] Verification Example 1

[0085] To verify the effectiveness of the method in this embodiment, surface temperature is used as an example, and East Asia is selected as the study area, with a latitude range of 15°N-40°N and a longitude range of 115°E-145°E. This region has significant monsoon climate characteristics and complex topographic variations, making it a typical area for evaluating the error correction effect of climate models in temperature simulation.

[0086] In step S1, surface temperature data and corresponding latitude / longitude grids for the target area are first extracted from the ERA5 reanalysis data according to the set latitude and longitude range, forming a reference dataset. Simultaneously, surface temperature data for the same time period is extracted from the CMIP6 model output data. Due to differences in spatial resolution and grid structure between ERA5 and CMIP6 data, to ensure correspondence, this embodiment interpolates the CMIP6 model data onto the ERA5 spatial grid. Specifically, for horizontal resolution, bilinear interpolation is used to interpolate the CMIP6 data onto the ERA5 horizontal grid; for vertical resolution, a three-dimensional interpolation function is used to interpolate the CMIP6 data onto the ERA5 vertical grid. This ensures the interpolation results maintain physical continuity and reasonable terrain transitions.

[0087] Since ERA5 is hourly data and CMIP6 is daily data, to ensure consistency in the time dimension, the ERA5 data was converted to a daily scale using a moving average method and aligned with the CMIP6 data by time. Subsequently, all surface temperature data were uniformly converted to Celsius (Kelvin minus 273.15) using MATLAB and saved as an intermediate data file to provide input for subsequent temporal and spatial error correction.

[0088] In S2, the ERA5 and CMIP6 surface temperature time series for each grid point within the study area are analyzed. First, wavelet analysis is used to analyze the primary and secondary fluctuation periods of each meteorological element in the time dimension of the CMIP6 and ERA5 data (as shown in the appendix). Figure 2 The wavelet spectrum shown includes meteorological elements such as surface temperature, air pressure, precipitation, relative humidity, wind speed, and ground temperature. Next, wavelet decomposition is used to decompose the two datasets into time series under each fluctuation period. In this embodiment, the wavelet basis function is Daubechies (db5), and the decomposition level is eleven (see attached diagram). Figure 3 The wavelet decomposition structure diagram is shown below. Wavelet decomposition decomposes a time series into an approximate component (representing the long-term trend) and multiple detail components (representing the fluctuation characteristics of different periods) (as shown in the attached diagram). Figure 4The approximate and detail fluctuations after wavelet decomposition are shown in the diagram. Subsequently, this embodiment calculates the 0%, 25%, 50%, 75%, and 100% quantile values ​​for each wavelet scale component of ERA5 at each grid point, constructing a multi-scale quantile feature matrix to reflect the probability distribution characteristics of ERA5 data at different time scales. For each scale component of CMIP6 data, its quantile distribution is calculated, and a quantile mapping function is constructed using the corresponding scale quantile of ERA5 as the target distribution. The distribution of CMIP6 is mapped to the distribution space of ERA5, achieving statistical correction for each scale component (as shown in the attached diagram). Figure 5 (See the quantile mapping diagram shown). This mapping function employs smooth interpolation within each quantile interval of the model and reanalysis to avoid edge effects. Finally, the corrected scale components are re-superimposed, and wavelet reconstruction is performed to obtain the time-dimension corrected surface temperature series. This process preserves the multi-scale statistical characteristics of ERA5 and maintains the continuity and climate trend of the CMIP6 time series.

[0089] In S3, using the data corrected for the time dimension in S2, further refinement and correction are performed in the spatial dimension. First, wavelet decomposition and quantile mapping correction are performed along the longitude direction. For each time-corrected data field, MATLAB is used to extract a one-dimensional spatial sequence along each latitude. Wavelet decomposition is then used to divide it into multiple spatial scale components, and quantile mapping is performed on each scale component. The mapping target is the distribution characteristics of ERA5 along the same latitude direction. After mapping, inverse wavelet reconstruction is performed to obtain the longitude-corrected field.

[0090] In S4, using the longitude-corrected data from S3, for each time step, MATLAB is used to extract a one-dimensional sequence along the latitude direction, and the same wavelet decomposition, quantile mapping, and reconstruction steps are performed. Finally, the latitude-corrected surface temperature field is obtained.

[0091] Finally, the time correction field, longitude correction field, and latitude correction field are saved as standard data files. The result data after latitude correction is the final correction, which includes coordinate information, time index, and correction parameter metadata, facilitating subsequent reproduction and reanalysis.

[0092] To demonstrate the correction effect, representative points A, B, and C (as shown in the attached diagram) were selected within the study area in this embodiment. Figure 6 The distribution map of representative points is shown below. Point A is located in the central part of the land, point B is located in the coastal area, and point C is located over the ocean, exhibiting different topographical and climatic characteristics. The original ERA5 and CMIP6 values ​​and the corrected CMIP6 time series of the three points are extracted and plotted as shown in the attached diagram. Figure 7- Figure 9 shows a comparison of surface temperature time series. The results show that the original CMIP6 data significantly underestimated the low-temperature period. After correction, the three-point time series almost overlaps with ERA5, reducing the error by about 80%. The corrected curve can more accurately reflect daily and seasonal changes.

[0093] To demonstrate the spatial correction effect, this embodiment selects three typical time sections and plots the spatial distribution of surface temperatures of ERA5, the original CMIP6, and the corrected CMIP6, as shown in the attached diagram. Figure 10 - As shown in Figure 12. The comparison results show that the corrected data are significantly improved in terms of spatial extreme value distribution, temperature gradient structure, and temperature transition characteristics in near-topographic regions, with a significant improvement in spatial resolution, which is highly consistent with the ERA5 results.

Claims

1. A method for correcting errors in climate data, characterized in that, Includes the following steps: S1: Determine the target area, acquire CMIP6 global climate model data and ERA5 reanalysis data for the target area, and unify the computational grids of the two types of data; S2: In the time dimension, wavelet analysis is used to extract the primary and secondary fluctuation periods of each meteorological element in CMIP6 and ERA5 data respectively. The data is decomposed into different period sequences through wavelet decomposition. Error correction functions under each period are established using quantile mapping to correct each period sequence of CMIP6. Finally, wavelet reconstruction is performed to complete the time dimension correction. S3: Based on the data corrected in the time dimension of S2, the longitude sequence is corrected in the longitude direction using wavelet analysis, wavelet decomposition, quantile mapping and wavelet reconstruction methods; Includes the following steps: S31: Using the data obtained in S2 after time dimension correction, extract the one-dimensional data sequence in the longitude direction for each time step; S32: For each set of longitude data sequences extracted in S31, wavelet analysis is performed to identify its main spatial fluctuation scale, and wavelet decomposition is performed to obtain spatial sequences at different scales. S33: For the spatial sequences of different scales obtained by S32 decomposition, establish a mapping function between global climate model data and target reanalysis data, and make corrections; S34: Perform wavelet reconstruction on the corrected spatial sequences at each scale to obtain the longitude correction field; S4: Based on the longitude correction data from S3, wavelet analysis, wavelet decomposition, quantile mapping, and wavelet reconstruction are used to correct the latitude sequence; S5: Save the time correction field, longitude correction field, and latitude correction field as standard data files; Includes the following steps: S41: Using the data obtained in S3 after longitude correction, extract the one-dimensional data sequence in the latitude direction for each time step; S42: For each set of latitude data sequences extracted in S41, wavelet analysis is performed to identify its main spatial fluctuation scale, and wavelet decomposition is performed to obtain spatial sequences at different scales. S43: For the spatial sequences of different scales obtained by the 432 decomposition, establish a mapping function between global climate model data and target reanalysis data, and make corrections. S44: Perform wavelet reconstruction on the corrected spatial sequences at each scale to obtain the latitudinal correction field.

2. The climate data error correction method according to claim 1, characterized in that, The S1 step of unifying the computational grid for the two types of data includes the following steps: S11: Unify the horizontal resolution of CMIP6 data and ERA5 data, and interpolate CMIP6 data to the horizontal grid of ERA5 data. S12: Under the premise of conforming to the vertical physical relationship of the atmosphere, a three-dimensional interpolation function is used to interpolate the CMIP6 data to the ERA5 vertical grid; S13: Consistentize time and units, ensure consistent time range for ERA5 and CMIP6 data, and convert temperature units from Kelvin to Celsius.

3. The climate data error correction method according to claim 2, characterized in that, S11 describes the use of a bilinear interpolation function to achieve uniform horizontal resolution.

4. The climate data error correction method according to claim 1, characterized in that, S2 includes the following steps: S21: Using wavelet analysis, the main and secondary fluctuation periods of each meteorological element in the time dimension of CMIP6 and ERA5 data are analyzed respectively. The meteorological elements include: surface temperature, air pressure, precipitation, relative humidity, wind speed, and ground temperature. S22: Using the wavelet decomposition method, with the Daubechies wavelet basis and 11 decomposition layers, the two types of data are decomposed into time series under each fluctuation period. S23: For each fluctuation period sequence obtained from the decomposition in S22, a mapping function for each fluctuation period is established using the quantile mapping method: S24: The time-dimension corrected meteorological element data are obtained by using wavelet reconstruction method for each periodic sequence after correction in S23.

5. A climate data error correction method according to claim 4, characterized in that, The mapping function described in S23 is: ； in and These are the cumulative distribution functions for the pattern data and the reanalysis data, respectively.

6. The climate data error correction method according to claim 1, characterized in that, The standard data file described in S5 includes coordinate information, time index, and correction parameter metadata.

Images