Optimization method and system for improving CMIP6 water vapor precision based on GNSS

Abstract

The invention provides a GNSS-based CMIP6 water vapor precision improvement optimization method and system, and belongs to the field of water vapor precision prediction, and the method comprises the steps: obtaining GNSS observation data, CMIP6 water vapor data and ERA5 reanalysis water vapor data; processing the acquired data: interpolating the GNSS observation data to an observation station and a grid in space, and downsampling the GNSS observation data in time; taking the CMIP6 water vapor data, the latitude, the longitude, the altitude and the annual day of the located region as input parameters, taking the GNSS observation data as target parameters for output, training the data by utilizing a machine learning algorithm, and constructing a CMIP6 water vapor lifting model based on GNSS-PWV (Global Navigation Satellite System-Pulse Width Value); and inputting to-be-tested data into the constructed model to obtain improved CMIP6 water vapor data.

Description

Technical Field

[0001] The present invention belongs to the field of water vapor accuracy prediction, and in particular relates to an optimization method and system for improving CMIP6 water vapor accuracy based on GNSS. Background Art

[0002] The statements in this section merely provide background information related to the present invention and do not necessarily constitute prior art.

[0003] In recent years, numerous studies have extensively explored historical reconstructions and future projections of climate change based on simulations from the sixth phase of the international Coupled Model Intercomparison Project (CMIP6) model. While CMIP6 models play an important role in climate change simulation and projections, their simulations exhibit certain errors compared to observational references. For example, Tian's research indicates that simulated annual average precipitation is significantly overestimated in the southwestern river basins of China, while it is underestimated in the high-altitude regions of the northwest. A subset of CMIP6 models can recreate transient surface subsurface shortwave radiation well in some regions, but performs poorly in others. A series of studies have shown that CMIP6 models play a key role in global climate change simulations and projections, but their simulation accuracy still exhibits certain regional and systematic biases.

[0004] Current research indicates that most studies using CMIP6 to explore the evolution of meteorological and climate change in their study areas have difficulty minimizing errors. Furthermore, most studies have simply integrated and averaged multiple global climate models to minimize CMIP6 model errors, without incorporating external constraints to further reduce errors and improve accuracy. Driven by big data (parameterization schemes of models designed and organized by experts and institutions worldwide), there is an even greater need to improve and refine CMIP6 model accuracy, breaking down the uncertainty barriers of big data predictions and providing more reliable information for decision makers.

[0005] Global Navigation Satellite System (GNSS) meteorology is rapidly developing. GNSS offers the advantages of high precision and high temporal resolution, but is limited by station distribution and spatial discontinuity. Remote sensing, on the other hand, offers high spatial resolution and spatial continuity, but suffers from a poor revisit period. Based on these advantages and disadvantages, research both domestically and internationally has integrated GNSS and remote sensing, demonstrating numerous mutually beneficial improvements in water vapor accuracy.

[0006] In the era of big data, the spatial and temporal evolution of future water vapor and its high-precision simulation have become core challenges and key research areas in global climate change research. While traditional observational methods (such as GNSS and remote sensing) can provide highly accurate instantaneous water vapor information, they are insufficient for long-term climate prediction. Global climate models (such as CMIP6) can simulate water vapor changes under future climate scenarios, but their spatial and temporal resolution is low and subject to regional biases.

[0007] When using global climate models to perform numerical simulations, there are limitations such as regional simulation bias and insufficient observational constraints. In addition, there are also problems such as insufficient temporal and spatial resolution, making it difficult to accurately capture water vapor changes. Summary of the Invention To overcome the shortcomings of the above-mentioned existing technologies, the present invention provides a GNSS-based optimization method for improving the CMIP6 water vapor accuracy, constructing a high-temporal and spatial resolution, high-precision water vapor model to improve the accuracy and robustness of water vapor simulation at the regional scale, providing a more reliable data foundation for climate change research and disaster prediction.

[0008] To achieve the above objectives, one or more embodiments of the present invention provide the following technical solutions: First, a GNSS-based optimization method for improving CMIP6 water vapor accuracy is disclosed, including: Obtain GNSS observation data, CMIP6 water vapor data, and ERA5 reanalysis water vapor data; Process the acquired data: spatially, interpolate the target data onto the stations and grids; temporally, downsample the target data; The CMIP6 water vapor data, CMIP6 water vapor corresponding latitude, CMIP6 water vapor corresponding longitude, CMIP6 water vapor corresponding altitude, and CMIP6 water vapor corresponding annual cumulative day are used as input parameters, and the GNSS observation data are used as the target parameter output. The machine learning algorithm is used to train the data and construct the CMIP6 water vapor lifting model based on GNSS-PWV. The data to be measured are input into the constructed model to obtain improved CMIP6 water vapor data.

[0009] Secondly, a GNSS-based optimization system for improving CMIP6 water vapor accuracy is provided, including: The data acquisition module is configured to: acquire GNSS observation data, CMIP6 water vapor data, and ERA5 reanalysis water vapor data; The data processing module is configured to: process the acquired data: spatially, interpolate the target data onto the measuring stations and grids, and temporally, downsample the target data; The model building module is configured to take CMIP6 water vapor data, CMIP6 water vapor corresponding latitude, CMIP6 water vapor corresponding longitude, CMIP6 water vapor corresponding altitude, and CMIP6 water vapor corresponding annual cumulative day as input parameters, output GNSS observation data as target parameters, use machine learning algorithms to train data, and build a CMIP6 water vapor uplift model based on GNSS-PWV. The water vapor accuracy improvement module is configured to input the data to be measured into the constructed model to obtain improved CMIP6 water vapor data.

[0010] One or more of the above technical solutions have the following beneficial effects: The technical solution of this invention incorporates GNSS water vapor data into the CMIP6 climate model and utilizes machine learning algorithms to improve the accuracy of CMIP6 water vapor data. Global Navigation Satellite System (GNSS) observations, with their high precision and high temporal and spatial resolution, provide reliable data support for PWV inversion. By applying high-precision GNSS PWV data to calibrate the CMIP6 water vapor data for the global climate model, it effectively reduces systematic biases in the model across different regions and climate settings, further guiding bias optimization.

[0011] The technical solution of the present invention proposes a method for improving the water vapor accuracy of CMIP6 using GNSS observations based on a machine learning algorithm. This method can not only enhance the reconstruction capability of the model in the historical stage, but also help to improve the reliability of CMIP6 in future meteorological forecasts and climate change projections, and provide a more solid scientific basis for risk assessment and adaptation strategy formulation of extreme weather and climate events.

[0012] Advantages of additional aspects of the present invention will be given in part in the following description and in part will be obvious from the following description, or will be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] The accompanying drawings, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0014] Figure 1 This is a flowchart of a method according to an embodiment of the present invention; Figure 2The PWV scatter plots of the training set and test set from 2020 to 2023 on the GNSS station of the present invention are as follows: (a) training set of the CNN algorithm; (b) test set of the CNN algorithm; (c) training set of the XGBoost algorithm; (d) test set of the XGBoost algorithm; (e) training set of the LSTM algorithm; (f) test set of the LSTM algorithm; Figure 3 The RMSE performance of PWV improvement of CNN, XGBoost and LSTM models on training and test sets in different seasons; Figure 4 The RMSE performance of PWV improvement of CNN, XGBoost and LSTM models at different altitudes for training set, test set and unified station set; Figure 5 Scatter plot of Improved_CMIP6-PWV and GNSS-PWV for validation set 1; Figure 6 Scatter plot of Improved_CMIP6-PWV and ERA5-PWV for validation set 2; Figure 7 Technical roadmap of the exemplary technical solutions implemented in this invention. DETAILED DESCRIPTION

[0015] It should be noted that the following detailed descriptions are exemplary and intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the art to which the present invention belongs.

[0016] It should be noted that the terms used herein are for describing particular embodiments only and are not intended to limit the exemplary embodiments according to the present invention.

[0017] In the absence of conflict, the embodiments of the present invention and the features thereof may be combined with each other.

[0018] Explanation of terms: PWV (Precipitable Water Vapor): atmospheric precipitable water; CMIP6 (Coupled Model Intercomparison Project Phase 6): The sixth international coupled model intercomparison project; GNSS (Global Navigation Satellite System): Global Navigation Satellite System; GCM (Global Climate Model): Global Climate Model; CNN (Convolutional Neural Network): Convolutional Neural Network; XGBoost (eXtreme Gradient Boosting): extreme gradient boosting; LSTM (Long Short-Term Memory): Long Short-Term Memory Network; ERA5 (European Centre for Medium-Range Weather Forecasts ReAnalysisv5): The fifth generation of reanalysis datasets of the European Centre for Medium-Range Weather Forecasts; RMSE (Root Mean Square Error): root mean square error.

[0019] The increasing sophistication and robust development of the Global Navigation Satellite System (GNSS) have led to its widespread application in meteorology. One of its most prominent applications is the acquisition of atmospheric precipitable water vapor (PWV) information, a key meteorological parameter crucial for weather forecasting, climate monitoring, and hydrological modeling. GNSS-PWV retrieval relies primarily on the delay caused by water vapor in satellite signals as they pass through the atmosphere, particularly the zenith tropospheric delay (ZTD). This enables high-temporal-resolution, all-weather water vapor observations. Compared to traditional observational methods such as soundings and microwave radiometers, GNSS offers advantages in high precision, high temporal resolution, low cost, continuous operation, and global coverage, making it particularly suitable for regions where conventional observations are scarce. In recent years, GNSS meteorology has been widely used in the study of weather processes such as heavy rainfall, typhoons, monsoon systems, earthquakes, and atmospheric rivers, demonstrating its enormous potential in capturing rapid water vapor changes and revealing atmospheric water vapor transport mechanisms.

[0020] Water vapor plays a crucial role in numerical weather forecasting. Precipitable water vapor (PWV), a key indicator of water vapor, is closely related to actual precipitation. Therefore, improving the accuracy of PWV and its spatial and temporal resolution will greatly benefit meteorological observations, forecasts, and climate change analysis. Given that the newly released CMIP6 continues to simulate and predict large-scale extreme disasters, such as more frequent and intense extreme heat and precipitation events under continued global warming, and given that some uncertainties remain, improving the accuracy and precision of its simulations and projections of spatiotemporal variations in atmospheric water vapor is particularly important.

[0021] This paper combines machine learning algorithms to propose a GNSS-based optimization method for improving the water vapor accuracy of the CMIP6 global climate model. This method is validated and analyzed in Turkey. This approach falls within the realm of atmospheric water vapor research, which intersects GNSS and atmospheric science.

[0022] In this implementation, CMIP6 data can be considered big data, but its future prediction accuracy is lower than that of observational data. GNSS, on the other hand, is observational data and has high accuracy. Therefore, this implementation introduces high-precision GNSS water vapor data to improve the accuracy of the CMIP6 water vapor data. In this implementation example, it is proposed for the first time to use GNSS to improve CMIP6 water vapor, and experimental application verification is carried out.

[0023] Example 1 This invention falls within the scope of atmospheric water vapor research, which intersects GNSS and atmospheric science. This example utilizes water vapor data from the HadGEM3-GC31-HM_highres-future model, part of the High-Resolution Model Intercomparison Project, a joint initiative of the Nevada Geodetic Laboratory (NGL) GNSS PWV, Reindert J. Haarsma of the Royal Netherlands Meteorological Service, and Malcolm J. Roberts of the UK Met Office. Based on different machine learning algorithms, this paper proposes a research method for improving CMIP6 water vapor accuracy based on GNSS fusion for the evenly distributed Turkish region. The technical approach is described in the attached document. Figure 7 shown.

[0024] This embodiment specifically discloses a GNSS-based optimization method for improving CMIP6 water vapor accuracy, including: First, data was acquired and spatial and temporal matching of GNSS, PWV and CMIP6 water vapor data was performed; Secondly, we used the training data to construct CMIP6 water vapor enhancement models using different algorithms, used the test data to verify the accuracy at the spatial scale, and explored the effects of the models on CMIP6 water vapor enhancement at different seasons and altitudes. Then, conduct the validation of the model on a time scale; Finally, the new validation set is used to complete the model's accuracy verification of grid-scale CMIP6 water vapor.

[0025] The details and description are as follows: Figure 1 As shown, the present invention includes the following steps: Step 1: Data processing. The data required for obtaining PWV in this invention include GNSS observation data, CMIP6 water vapor data, and ERA5 reanalysis water vapor data.

[0026] It should be noted that GNSS technology has the advantages of high precision, all-weather capability, and low cost. This product data comes from the Nevada Geodetic Laboratory and is recorded as GNSS-PWV.

[0027] The CMIP6 water vapor data come from the HadGEM3-GC31-HM_highres-future model in the CMIP6 High-Resolution Model Intercomparison Project, denoted as CMIP6-PWV.

[0028] The ERA5 reanalysis water vapor data comes from the total column water vapor data of the European Centre for Medium-Range Weather Forecasts (ECMWF), denoted as ERA5-PWV.

[0029] Due to the different temporal and spatial resolutions of the input CMIP6 water vapor data (CMIP6-PWV) and the target water vapor parameters (GNSS-PWV) (i.e., GNSS observations), temporal and spatial matching was performed as follows. Spatially, bilinear interpolation was used to interpolate the target data onto the observation stations and grid. Specifically, CMIP6-PWV is grid data with a resolution of 50 km, and GNSS-PWV is point data based on ground stations. Bilinear interpolation is used to interpolate CMIP6 model data onto GNSS stations to obtain CMIP6-PWV (station_CMIP6-PWV) for all participating stations.

[0030] On the other hand, ERA5-PWV was used as the ground truth in the test set and validation set 1. ERA5-PWV is a 0.25° resolution gridded data. Bilinear interpolation was used to interpolate ERA5 data to the GNSS station. The ERA5-PWV at the station (station_ERA5-PWV) was obtained and used as the verification data for the test set and validation set 1.

[0031] In addition, in validation set 2, the ERA5 data were bilinearly interpolated onto the original CMIP6 grid points to obtain the ERA5-PWV data (Grid_ERA5-PWV) matching the CMIP6 grid points as the verification data for validation set 2. Temporally, the target data was downsampled to eight daily timestamps: 01:30, 04:30, 07:30, 10:30, 13:30, 16:30, 19:30, and 22:30. The CMIP6-PWV has a 3-hour resolution, corresponding to the eight daily timestamps: 01:30, 04:30, 07:30, 10:30, 13:30, 16:30, 19:30, and 22:30. The GNSS-PWV data was downsampled to retain only the eight daily timestamps: 01:30, 04:30, 07:30, 10:30, 13:30, 16:30, 19:30, and 22:30. The matching was performed using a conventional one-to-one location and one-to-one time matching method. The PWV data information and usage after step 1 of spatiotemporal matching are shown in Table 1.

[0032] Table 1 PWV data information and usage after time-space matching

[0033] The preprocessed data were used to train the model using the training set, test the model using the test set, and validate the model using validation sets 1 and 2. The model's accuracy was verified using Station_CMIP6-PWV data from the training set stations from 2020 to 2023 as the input water vapor parameter, and GNSS-PWV data from 2020 to 2023 as the output target water vapor. The model was trained using Station_CMIP6-PWV data from the test set stations from 2020 to 2023 as the input water vapor parameter, and GNSS-PWV and Station_ERA5-PWV data from the test set stations from 2020 to 2023 as the validation data for the test set. The Station_CMIP6-PWV data from the validation set 1 stations from 2024 were used as the input water vapor parameter, and GNSS-PWV and Station_ERA5-PWV data from the validation set 1 stations from 2024 were used as the validation data for validation set 1. The Grid_CMIP6-PWV data from the 2024 validation set 2 stations were used as the input water vapor parameters, and the Grid_ERA5-PWV data from the 2024 validation set 2 stations were used as the validation data for validation set 2. The trained model was spatially validated using the test set data, and temporally validated using the validation set data.

[0034] The advantages of this solution are: first, the accuracy of the trained model is verified from both spatial and temporal scales; second, and this is the greatest innovation of the present invention, it is the first to introduce the use of high-precision GNSS-PWV observation data to train CMIP6-PWV data, which greatly reduces the error of the CMIP6 climate model data in future predictions and compensates for the low accuracy of the climate model; third, at the spatial scale, verification is carried out at both the station and grid levels to ensure their comprehensive application at both the station and grid scales.

[0035] Step 2: Model Construction and Testing. This method combines multi-source input features with different machine learning algorithms: Convolutional Neural Network (CNN), eXtreme Gradient Boosting (XGBoost), and Long Short-Term Memory (LSTM), respectively, to capture the spatial structure, nonlinear relationships, and time series evolution patterns of the input features. The CNN model is designed as a lightweight convolutional architecture with 13 layers and a batch size of 128, optimized using the SDGM parameter. This architecture learns spatial distribution patterns and local neighborhood correlations, making it suitable for spatial error correction of large-scale grid data. The LSTM model is designed as a 5-layer LSTM network with a batch size of 64, optimized using the Adam parameter. This architecture models the daily-scale water vapor time series fluctuations, which are caused by the temporal fluctuations in PWV content, including seasonal cycles, diurnal variations, abrupt weather changes, and trend shifts. This effectively addresses the instability of CMIP6-PWV forecasts across different seasonal periods and examines the underlying dynamics of PWV evolution. XGBoost modeling utilizes a tree-based model structure to perform nonlinear multi-feature combinations, constructing a boosted tree model that is sensitive to high-dimensional inputs and exhibits strong generalization capabilities, particularly in low- and medium-altitude regions. In addition to traditional meteorological factors, it integrates spatial geographic features with temporal dynamic features. Spatial geographic features, such as longitude and latitude, and temporal dynamic features, such as the cumulative number of days per year, range from 1 to 365 and reflect the changing patterns of seasonal climate cycles. This creates a multidimensional input feature matrix, calculated as the number of samples × the number of input features. This significantly enhances the model's ability to discern regional topographic and temporal cyclical changes. In addition to water vapor data, geographic labels, including longitude, latitude, and elevation, are introduced as input features to enhance the model's adaptability to topographically sensitive regions. For topographically sensitive regions, such as plateaus and mountainous areas, temporal labels, such as the cumulative number of days per year, are also introduced to enhance the model's capture of temporal variations. The target output uses measured GNSS-PWV data as label values, representing the actual water vapor values.

[0036] In this embodiment, the time series fluctuation characteristics refer to the numerical fluctuation patterns of the daily water vapor PWV over time, including but not limited to: 1) Periodic fluctuations: A recurring pattern of changes within similar time intervals, influenced by diurnal and seasonal climate factors; 2) Short-term fluctuations: Rapid fluctuations in PWV values ​​within a short period of time caused by weather systems (such as fronts, typhoons, and severe convection). LSTM networks can capture these fluctuations and learn the inherent dynamics of PWV evolution over time, effectively reducing instability in daily forecasts across different seasons and weather conditions.

[0037] In this embodiment, a temporal dynamic feature refers to a data attribute that exhibits different values ​​over time and can reflect the periodicity, trend, and mutation of water vapor over time. In this invention, the annual cumulative day is used as the temporal dynamic feature. The annual cumulative day value ranges from 1–365 and is used to reflect the changing patterns of seasonal climate cycles. The input temporal dynamic feature is "annual cumulative day."

[0038] In this example, a feature matrix refers to a two-dimensional array of multiple feature variables organized in a unified format for input into the model for calculation. The multidimensional input feature matrix is ​​composed of n samples x 5 input parameters. CMIP6-PWV, the latitude of the corresponding CMIP6-PWV station / grid point, the longitude of the corresponding CMIP6-PWV station / grid point, the altitude of the corresponding CMIP6-PWV station / grid point, and the annual cumulative day of the corresponding CMIP6-PWV are used as input parameters. High-precision GNSS-PWV is output as the target parameter. Machine learning algorithms such as CNN, LSTM, and XGBoost are used to train data and construct a CMIP6 water vapor uplift model based on GNSS-PWV. The improvement effect is evaluated on a test set using GNSS-PWV as the ground truth. The improved CMIP6-PWV output by the model is denoted as Improved_CMIP6-PWV.

[0039] Step 3: Spatiotemporal Analysis: The model was divided into four seasons and five altitude zones. The water vapor uplift indicators (RMSE) and percentages were analyzed for each of these four seasons and five altitude zones. The performance of different models in improving CMIP6-PWV at different seasonal scales and geographic locations was further explored. The root mean square error (RMSE) and the percentage reduction in RMSE were used as indicators of model performance improvement. A smaller RMSE value indicates a better model correction of CMIP6-PWV; a smaller RMSE value indicates a moderate correction.

[0040]

[0041]

[0042] For the seasonal analysis, data were categorized by annualized days into the following groups: spring (March-May), summer (June-August), autumn (September-November), and winter (December-February). For each season, the original CMIP6-PWV data and the model-calibrated results were evaluated using the RMSE formula. The percentage reduction in RMSE, or improvement rate, was also calculated to reflect the degree of improvement in model performance across seasons, particularly between wet (e.g., summer) and dry (e.g., winter) seasons. For the analysis across geographic locations, GNSS stations in the study area were categorized by elevation, using the natural break classification method to divide the study area into five elevation zones. The water vapor RMSE and improvement rate were calculated for each of the five elevation zones before and after model calibration to analyze the model's generalization and stability under different terrain conditions. This approach, therefore, analyzes the model's performance across different seasons and geographic locations.

[0043] Step 4: Accuracy Verification. Using GNSS-PWV as the true value, the improvement effect for the observation period of 2024 (Validation Set 1), which was not involved in training, was verified: the accuracy of the 2024 station-level Improved_CMIP6-PWV output by the model was compared with the true GNSS-PWV. Furthermore, the original CMIP6 data is gridded, but the aforementioned accuracy improvement is achieved by interpolating it to GNSS stations, which does not reflect the contextual characteristics of the CMIP6 spatial grid. Therefore, this method performs grid-scale accuracy verification to preserve the inherent properties of the input product. Here, ERA5-PWV, which was not involved in modeling, is used as the validation data for comparison, including both station-level and grid-level results. A new validation (i.e., Validation Set 2) compares the accuracy of the improved 2024 grid-scale CMIP6-PWV data (Improved_Grid_CMIP6-PWV) output by the model with the true grid-scale ERA5 (Grid_ERA5-PWV). For comparability, the Improved_CMIP6-PWV results of the station were also compared with the ERA5-PWV data (station_ERA5-PWV) of the station that did not participate in the modeling for the accuracy of the test set and validation set 1.

[0044] In particular, when conducting a new validation (validation set 2), it was necessary to extract grid point data with non-negative elevations. This is because the input characteristics of the grid points are influenced by the equally spaced longitude and latitude positions of the original data. Since the grid within the study area contains locations below sea level, their elevations are necessarily negative. However, since the target parameters used in model construction are station GNSS-PWV observation data, and the GNSS station elevations are all above 0 m, necessary data quality checks are required to avoid relatively long air columns at negative elevation grid points, which can lead to overestimated PWV results.

[0045] The experimental results obtained by the present invention can be described as follows: Experimental Area: This study uses data from 2020 to 2024 provided by 162 GNSS stations evenly distributed across Turkey. Valid data was obtained from 98 stations in 2020, 16 stations in 2021, 141 stations in 2022, 139 stations in 2023, and 71 stations in 2024.

[0046] Data temporal and spatial matching: Spatially, bilinear interpolation was used to interpolate CMIP6 model data onto GNSS stations, resulting in CMIP6-PWV data for all participating stations. Temporally, CMIP6-PWV has a 3-hour resolution, with eight daily time points: 01:30, 04:30, 07:30, 10:30, 13:30, 16:30, 19:30, and 22:30. GNSS-PWV has a 300-second temporal resolution and requires downsampling to retain only the eight daily time points: 01:30, 04:30, 07:30, 10:30, 13:30, 16:30, 19:30, and 22:30.

[0047] On the other hand, the ERA5-PWV data were also temporally and spatially matched and interpolated onto the GNSS station and CMIP6 original grids using the bilinear interpolation method.

[0048] Finally, we obtained the 2020-2024 station CMIP6-PWV (station_CMIP6-PWV), GNSS-PWV, and station ERA5-PWV data (station_ERA5-PWV) with good temporal and spatial matching in Turkey, and the 2024 grid-scale CMIP6-PWV (Grid_CMIP6-PWV) and grid-scale ERA5-PWV data (Grid_ERA5-PWV) with good temporal and spatial matching.

[0049] Model construction and testing: Experimental results show that all models can significantly improve the accuracy of CMIP6-PWV, demonstrating the effectiveness of the introduction of GNSS, and the degree of improvement varies among models. We record the improved CMIP6 water vapor results output by the model as Improved_CMIP6-PWV. Using GNSS-PWV as the reference truth, we compare the accuracy of Improved_CMIP6-PWV with GNSS-PWV. Figure 2 As shown in the figure, the CNN algorithm achieved an RMSE of 4.25 mm for the training set and 4.57 mm for the test set, representing improvements of 38.49% and 36.17%, respectively. The Person correlation coefficients were 0.79 and 0.76, respectively, demonstrating effective spatial feature extraction. XGBoost significantly reduced learning costs, achieving RMSEs of 3.54 mm and 4.04 mm for the training and test sets, respectively, representing improvements of 48.77% and 43.57%, respectively. The Person correlation coefficients were 0.86 and 0.82. LSTM performed exceptionally well in time series modeling, achieving an RMSE of 4.85 mm for the training set and 4.93 mm for the test set, representing improvements of 29.81% and 31.14%, respectively. Overall, the XGBoost model clearly demonstrated its substantial optimization effect on CMIP6-PWV at the station.

[0050] Spatiotemporal Analysis: To further investigate the performance of different models in improving CMIP6-PWV at different seasonal scales, the data were divided into spring (MAM), summer (JJA), autumn (SON), and winter (DJF). A smaller RMSE indicates a better model correction of CMIP6-PWV; conversely, a lower RMSE indicates a moderate correction. Figure 3The RMSE results for different seasons in the training and test sets for models built using the CNN, XGBoost, and LSTM algorithms are shown. It is clear that the XGBoost model achieves the best improvement, with RMSEs of 3.12 mm, 4.95 mm, 4.08 mm, and 2.53 mm for the four seasons in the training set, and 3.60 mm, 5.62 mm, 4.58 mm, and 3.02 mm for the four seasons in the test set. Furthermore, summer has the largest RMSE of all seasons. This is because, although Turkey has a temperate continental climate with little rainfall year-round, summer is still a season of relatively active convection, with relatively high levels of water vapor content and variability, leading to large errors in the CMIP6 model's PWV simulations. Given this large baseline, the improved summer model also exhibits a higher RMSE than other seasons. The improvement rates for all models are above 25%, with XGBoost achieving the highest (over 40%). In winter, the water vapor content is low and stable, so the temperature difference is small.

[0051] The station elevations in Turkey are divided into five categories according to the natural break point classification method (as shown in Table 1). The improvement effects of the training set, test set and unified station set of different models on the CMIP6-PWV at different elevations are explored. The results are as follows: Figure 4 As shown in the figure, the XGBoost model achieved the best correction results, with RMSEs of 4.56, 3.28, 2.99, 2.89, and 2.42 mm across all five elevation categories, representing improvements of over 40%. The RMSE gradually decreases with increasing altitude. This is because the air column becomes shorter with increasing altitude, reducing the magnitude of the PWV content it contains, and thus the RMSE naturally decreases. In the training set, test set, and unified station set, the second category of low- and medium-altitude stations achieved the highest correction rates compared to the other four elevation categories, at 58.3%, 52.8%, and 56.5%, respectively (XGBoost). Furthermore, the number of stations in each elevation category was standardized to 16. XBGoost achieved the best RMSE for CMIP6 PWV improvements, with values ​​of 4.49 mm, 3.28 mm, 2.97 mm, 3.11 mm, and 2.11 mm, respectively. In terms of improvement rate, the Class II altitude zone achieved the greatest improvement, with rates of 45.0%, 56.5%, and 34.8% across the different models, followed by the Class III altitude zone, further reinforcing the strong evidence that the models provide better corrections for low- to medium-elevation regions. The RMSE of the CMIP6 water vapor data without model improvement decreased significantly linearly, but the RMSE of the CMIP6 models improved with the XBGoost and CNN algorithms remained unchanged for Classes II, III, and IV, indicating that the models learned topographical characteristics and achieved comparable corrections for the medium-elevation regions.

[0052] Table 2. Altitude classification of valid observation stations in Türkiye

[0053] Table 2 shows the elevation zones of valid stations in Turkey, using the GNSS-PWV validation system. After all model improvements, Improved_CMIP6-PWV at stations in eastern and central Turkey still exhibits smaller errors than at other stations. This suggests a trend in the distance from land and sea: the accuracy of CMIP6-PWV improves as the distance from the ocean increases. Ocean evaporation, high water vapor content, and dramatic humidity fluctuations make it difficult for CMIP6 to effectively simulate these small-scale processes under this parameterization scheme. This strongly suggests that CMIP6 water vapor simulations exhibit significant biases at the land-ocean interface. Applying CMIP6 to water vapor research, especially in coastal areas or at the land-ocean interface, requires additional calibration or integration with GNSS for error correction. Comparing different models, XGBoost continues to provide the best improvement, and this improvement is consistent across Turkey, with most exceeding 50%.

[0054] Accuracy verification: Time verification: The observation period of 2024, which was not involved in the model training, was selected as the verification set 1. The accuracy of the model output station Improved_CMIP6-PWV was compared with the true value GNSS-PWV, such as Figure 5 The CNN model performed best, with an RMSE of 5.31 mm (a 29.20% improvement) and a Person correlation coefficient of 0.73. The LSTM model performed second, with an RMSE of 5.61 mm (a 25.24% improvement) and a Person correlation coefficient of 0.70. The XGBoost model achieved an RMSE of 5.94 mm (a 20.76% improvement) and a Person correlation coefficient of 0.66.

[0055] The ERA5-PWV that was not involved in the modeling was used as the true observation value for comparison. The new validation set (i.e., validation set 2) is the 2024 Turkey CMIP6-PWV grid data (Grid_CMIP6-PWV). Since the original data is distributed in an equidistant latitude and longitude grid covering the entire region, and there are areas below sea level in the study area, the grid point elevations appear negative. However, the target parameter GNSS-PWV data used in this model construction are all from ground stations (elevation > 0 m). In order to ensure data consistency, strict quality control must be performed. Based on this consideration, the present invention eliminates grid point data with negative elevations, and the elimination ratio is 0.40% of the total data volume. The accuracy of the grid-scale Improved_CMIP6-PWV output by the model is compared with the true value ERA5-PWV, as shown in the following figure: Figure 6The results show that CNN improves CMIP6 water vapor data by 11.64% with an RMSE of 5.96 mm, XGBoost improves it by 17.71% with an RMSE of 5.55 mm, and LSTM improves it by 18.97% with an RMSE of 5.47 mm.

[0056] To ensure comparability of the evaluation results, we used ERA5-PWV data (station-scale), which were not used in modeling, as a unified reference benchmark to systematically validate the Improved_CMIP6-PWV results for each station. The RMSEs for the three algorithms for the test set were 5.40 mm, 5.66 mm, and 5.41 mm, respectively, while the RMSEs for validation set 1 were 5.36 mm, 6.03 mm, and 5.59 mm, respectively. Table 3 visually illustrates the evaluation results for all test and validation sets.

[0057] In summary, the evaluation results for all test and validation sets show significant accuracy improvements. Through GNSS data collaborative correction, the accuracy of CMIP6 water vapor products has been significantly improved at various spatial scales (including the original grid scale and GNSS station locations), with an average improvement of approximately 20%.

[0058] Table 3. Accuracy verification of Improved_CMIP6-PWV compared to ERA5-PWV at station and grid scales

[0059] The method of the present invention achieves optimal results using ARCGIS software and MATLAB. This method optimizes the CMIP6 water vapor accuracy based on high-precision GNSS and conducts a series of accuracy verifications. By integrating GNSS and employing machine learning algorithms to construct different models, the method improves the water vapor accuracy of the CMIP6 High-Resolution Model Intercomparison Project and verifies its accuracy at multiple scales.

[0060] In this paper, Türkiye is used as the study area, and the CMIP6 water vapor uplift model is constructed using GNSS PWV. Spatiotemporal testing, verification and analysis are carried out at both station scale and grid scale.

[0061] Example 2 The purpose of this embodiment is to provide a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the above method when executing the program.

[0062] Example 3 The purpose of this embodiment is to provide a computer-readable storage medium.

[0063] A computer-readable storage medium stores a computer program, which, when executed by a processor, performs the steps of the above method.

[0064] Example 4 The purpose of this embodiment is to provide a GNSS-based optimization system for improving CMIP6 water vapor accuracy, including: The data acquisition module is configured to: acquire GNSS observation data, CMIP6 water vapor data, and ERA5 reanalysis water vapor data; The data processing module is configured to: process the acquired data: spatially, interpolate the target data onto the measuring stations and grids, and temporally, downsample the target data; The model building module is configured to take CMIP6 water vapor data, CMIP6 water vapor corresponding latitude, CMIP6 water vapor corresponding longitude, CMIP6 water vapor corresponding altitude, and CMIP6 water vapor corresponding annual cumulative day as input parameters, output GNSS observation data as target parameters, use machine learning algorithms to train data, and build a CMIP6 water vapor uplift model based on GNSS-PWV. The water vapor accuracy improvement module is configured to input the data to be measured into the constructed model to obtain improved CMIP6 water vapor data.

[0065] Example 5 The purpose of this embodiment is to provide a computer program product containing instructions, which, when running on a computer, enables the computer to execute the methods and functions involved in any of the above embodiments. The steps involved in the apparatus of the above embodiment correspond to those of the method embodiment 1. For detailed implementation, please refer to the relevant description of embodiment 1. The term "computer-readable storage medium" should be understood to mean a single medium or multiple media containing one or more instruction sets; it should also be understood to include any medium capable of storing, encoding, or carrying an instruction set for execution by a processor and causing the processor to perform any method of the present invention.

[0066] Those skilled in the art will appreciate that the modules or steps of the present invention described above can be implemented using a general-purpose computer device. Alternatively, they can be implemented using program code executable by a computing device, which can then be stored in a storage device and executed by the computing device. Alternatively, they can be fabricated into separate integrated circuit modules, or multiple modules or steps can be fabricated into a single integrated circuit module for implementation. The present invention is not limited to any specific combination of hardware and software.

[0067] Although the above describes the specific embodiments of the present invention in conjunction with the accompanying drawings, it is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art on the basis of the technical solution of the present invention without any creative work are still within the scope of protection of the present invention.

Claims

1. A GNSS-based optimization method for improving CMIP6 water vapor accuracy, characterized by: include: Obtain GNSS observation data, CMIP6 water vapor data, and ERA5 reanalysis water vapor data; Process the acquired data: spatially, interpolate the target data onto the stations and grids; temporally, downsample the target data; The CMIP6 water vapor data, CMIP6 water vapor corresponding latitude, CMIP6 water vapor corresponding longitude, CMIP6 water vapor corresponding altitude, and CMIP6 water vapor corresponding annual cumulative day are used as input parameters, and the GNSS observation data are used as the target parameter output. The machine learning algorithm is used to train the data and construct the CMIP6 water vapor lifting model based on GNSS-PWV. The data to be measured are input into the constructed model to obtain improved CMIP6 water vapor data.

2. The optimization method for improving CMIP6 water vapor accuracy based on GNSS according to claim 1, characterized in that: In space, bilinear interpolation is used to interpolate the target data to the stations and grids.

3. The optimization method for improving CMIP6 water vapor accuracy based on GNSS according to claim 1, characterized in that: The GNSS observation data is based on point data of ground stations. Bilinear interpolation is used to interpolate the CMIP6 model data onto the GNSS stations to obtain data from all stations participating in the experiment.

4. The optimization method for improving CMIP6 water vapor accuracy based on GNSS according to claim 1, characterized in that: The CMIP6 water vapor data are grid data of a set resolution.

5. The optimization method for improving CMIP6 water vapor accuracy based on GNSS according to claim 1, characterized in that: In the process of constructing the CMIP6 water vapor uplift model based on GNSS-PWV, the ERA5 reanalysis water vapor data were used as the reference truth in the test and validation sets. The ERA5 reanalysis water vapor data are grid data with a set resolution. Bilinear interpolation was used to interpolate the ERA5 reanalysis water vapor data to the GNSS stations. The ERA5 reanalysis water vapor data at the stations were obtained as the verification data for the test and validation sets.

6. The optimization method for improving CMIP6 water vapor accuracy based on GNSS according to claim 5, characterized in that: In another validation set, the ERA5 reanalysis water vapor data were bilinearly interpolated onto the CMIP6 original grid points to obtain ERA5 reanalysis water vapor data that matched the CMIP6 grid points, which served as the validation data for another validation set.

7. A GNSS-based optimization system for improving CMIP6 water vapor accuracy, characterized by: include: The data acquisition module is configured to: acquire GNSS observation data, CMIP6 water vapor data, and ERA5 reanalysis water vapor data; The data processing module is configured to: process the acquired data: spatially, interpolate the target data onto the measuring stations and grids, and temporally, downsample the target data; The model building module is configured to take CMIP6 water vapor data, CMIP6 water vapor corresponding latitude, CMIP6 water vapor corresponding longitude, CMIP6 water vapor corresponding altitude, and CMIP6 water vapor corresponding annual cumulative day as input parameters, output GNSS observation data as target parameters, use machine learning algorithms to train data, and build a CMIP6 water vapor uplift model based on GNSS-PWV. The water vapor accuracy improvement module is configured to input the data to be measured into the constructed model to obtain improved CMIP6 water vapor data.

8. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, the method according to any one of claims 1 to 6 is implemented.

9. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, the steps of the method described in any one of claims 1 to 6 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps of the method described in any one of claims 1 to 6 are performed.

Images