A nonlinear correction method and system for geostationary satellite atmospheric precipitable water
Patent Information
- Application Number
- CN202611427178.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-09-15
- Publication Date
- 2026-10-23
AI Technical Summary
[0003]针对上述情况，为克服现有技术的缺陷，本发明提供了一种静止轨道卫星大气可降水量非线性校正方法及系统，针对一般静止轨道卫星大气可降水量非线性校正方法存在未充分利用空间邻近区域和连续观测时刻之间的误差相关性，容易导致相邻像元或者相邻时刻得到的校正结果存在不必要的差异，降低校正PWV场的时空连续性，进而影响校正结果准确性的问题，本方案利用邻域卫星PWV误差一致性对时空传播条件进行筛选，仅在当前卫星PWV观测样本与时空邻域卫星PWV观测样本具有较高误差一致性时进行邻域特征传播，并通过时空邻域传播系数控制邻域特征与当前卫星PWV观测样本特征的融合程度，从而在利用静止轨道卫星时空连续观测信息提高PWV误差特征连续表达能力的同时，避免异常误差跨区域传播以及真实局地水汽突变特征被过度平滑，进而提高静止轨道卫星大气可降水量非线性校正结果的准确性；针对一般静止轨道卫星大气可降水量非线性校正方法存在容易在局部区域产生过大的校正量，造成校正PWV空间突变，无法根据不同卫星PWV卫星PWV观测样本的原始误差程度调整校正幅度的问题，本方案通过水汽场连续性约束限制校正PWV与真实水汽场之间的空间变化差异；根据卫星PWV观测值与GNSS参考PWV之间的差异构建校正幅度调节系数，对校正增量进行自适应约束，使原始误差较小的卫星PWV保持较小的校正幅度，而原始误差较大的卫星PWV获得更大的校正自由度，从而减少局部异常校正和过度校正，提高静止轨道卫星大气可降水量非线性校正结果的稳定性
[0044]（1）针对一般静止轨道卫星大气可降水量非线性校正方法存在未充分利用空间邻近区域和连续观测时刻之间的误差相关性，容易导致相邻像元或者相邻时刻得到的校正结果存在不必要的差异，降低校正PWV场的时空连续性，进而影响校正结果准确性的问题，本方案利用邻域卫星PWV误差一致性对时空传播条件进行筛选，仅在当前卫星PWV观测样本与时空邻域卫星PWV观测样本具有较高误差一致性时进行邻域特征传播，并通过时空邻域传播系数控制邻域特征与当前卫星PWV观测样本特征的融合程度，从而在利用静止轨道卫星时空连续观测信息提高PWV误差特征连续表达能力的同时，避免异常误差跨区域传播以及真实局地水汽突变特征被过度平滑，进而提高静止轨道卫星大气可降水量非线性校正结果的准确性。
Smart Images
Figure CN122964433A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of data processing technology, specifically to a method and system for nonlinear correction of atmospheric precipitable water content of geostationary satellites. Background Technology
[0002] The nonlinear correction method for atmospheric precipitable water vapor (PWV) of geostationary satellites utilizes the error between PWV retrieved from geostationary satellites and high-precision PWV reference values from ground-based GNSS (Global Navigation Satellite System). It establishes a nonlinear mapping relationship by incorporating temporal, spatial, and meteorological factors to compensate for the error in the satellite PWV retrieval results, thereby obtaining a PWV product that more closely approximates the true atmospheric water vapor state. However, general nonlinear correction methods for atmospheric precipitable water vapor of geostationary satellites often fail to fully utilize the error correlation between spatially adjacent regions and consecutive observation times. This can easily lead to unnecessary differences in correction results obtained from adjacent pixels or adjacent times, reducing the spatiotemporal continuity of the corrected PWV field and thus affecting the accuracy of the correction results. Furthermore, general nonlinear correction methods for atmospheric precipitable water vapor of geostationary satellites can easily generate excessive correction amounts in local regions, causing spatial abrupt changes in the corrected PWV, and cannot adjust the correction amplitude according to the original error level of different satellite PWV observation samples. Summary of the Invention
[0003] To address the above issues and overcome the shortcomings of existing technologies, this invention provides a nonlinear correction method and system for atmospheric precipitable water (PWV) of geostationary satellites. Addressing the problem that general nonlinear correction methods for PWV of geostationary satellites fail to fully utilize the error correlation between spatially adjacent regions and continuous observation times, easily leading to unnecessary differences in correction results obtained from adjacent pixels or adjacent times, reducing the spatiotemporal continuity of the corrected PWV field, and thus affecting the accuracy of the correction results, this scheme utilizes the consistency of PWV errors from neighboring satellites to screen spatiotemporal propagation conditions. Neighborhood feature propagation is only performed when the current satellite PWV observation sample and the spatiotemporally adjacent satellite PWV observation samples have a high degree of error consistency. Furthermore, the spatiotemporal neighborhood propagation coefficient controls the degree of fusion between neighborhood features and the features of the current satellite PWV observation sample. This improves the continuous representation capability of PWV error features while utilizing continuous spatiotemporal observation information from geostationary satellites, thus avoiding abnormal errors. The propagation differences across regions and the abrupt changes in local water vapor characteristics are excessively smoothed, thereby improving the accuracy of the nonlinear correction results for atmospheric precipitable water (PWV) of geostationary satellites. Addressing the common problem of excessively large correction values in local areas leading to spatial abrupt changes in the corrected PWV and the inability to adjust the correction amplitude based on the original error levels of different satellite PWV observation samples, this scheme limits the spatial variation difference between the corrected PWV and the actual water vapor field through water vapor field continuity constraints. A correction amplitude adjustment coefficient is constructed based on the difference between the satellite PWV observations and the GNSS reference PWV, adaptively constraining the correction increment. This ensures that satellite PWVs with smaller original errors maintain smaller correction amplitudes, while satellite PWVs with larger original errors gain greater correction freedom, thus reducing local anomaly corrections and overcorrections, and improving the stability of the nonlinear correction results for atmospheric precipitable water (PWV) of geostationary satellites.
[0004] The technical solution adopted by this invention is as follows: This invention provides a nonlinear correction method for atmospheric precipitable water content of geostationary satellites, which includes the following steps:
[0005] Step S1: Constructing satellite PWV observation samples;
[0006] Step S2: Satellite error characterization;
[0007] Step S3: Enhanced spatiotemporal propagation;
[0008] Step S4: Design of the error feature mapping layer;
[0009] Step S5: Design of the error contribution fusion layer;
[0010] Step S6: Output of the nonlinear correction network.
[0011] Furthermore, in step S1, the satellite PWV observation sample construction is to establish satellite PWV observation samples using high-precision ground-based observations, using geostationary orbit satellite PWV observation values as satellite PWV values to be corrected, and GNSS inverted PWV reference values as training supervision values.
[0012] Match satellite PWV observation samples with consistent time and spatial location, calculate the deviation of satellite PWV values from GNSS reference PWV values, and use this as the satellite PWV error.
[0013] Further, in step S2, the satellite error characterization involves characterizing the influencing factors of the satellite PWV error; including:
[0014] The satellite PWV error is divided into three dimensions: spatial, temporal, and meteorological, and corresponding feature vectors are constructed for each dimension.
[0015] Z-score standardization is applied to the error impact characteristics of each dimension;
[0016] Joint encoding combines the three-dimensional features to form the satellite PWV error impact features.
[0017] Further, in step S3, the spatiotemporal propagation enhancement involves performing spatiotemporal propagation screening and, based on the screening results, utilizing the error correlation between geostationary satellite PWV observations in spatially adjacent regions and continuous observation times to perform spatiotemporal neighborhood enhancement on the satellite PWV error impact characteristics; including:
[0018] For the current satellite PWV observation sample, select neighboring satellite PWV observation samples according to spatial distance and time interval to construct a spatiotemporal neighborhood set;
[0019] For current satellite PWV observation samples and neighboring satellite PWV observation samples, spatiotemporal propagation weights are constructed based on the normalized spatial distance and time interval between the two.
[0020] For the selection of PWV observation samples from the spatiotemporal propagation satellite, for the i-th satellite PWV observation sample, the neighborhood error consistency is calculated within the spatiotemporal neighborhood. Set the propagation determination threshold τ c ,when The current satellite PWV observation sample is identified as a spatiotemporal propagation enhanced sample; otherwise, the current satellite PWV observation sample is identified as a non-propagation sample.
[0021] For spatiotemporal propagation enhancement, the satellite PWV error impact characteristics of the spatiotemporal propagation enhanced samples are adaptively fused with the spatiotemporal neighborhood characteristics to obtain the satellite PWV error impact characteristics after spatiotemporal propagation enhancement.
[0022] Further, in step S4, the error feature mapping layer design is based on the satellite PWV error impact characteristics, establishing a nonlinear mapping relationship between influencing factors and satellite PWV error; including:
[0023] Error hidden features under the combined influence of different environmental factors are extracted using a nonlinear feature mapping network;
[0024] A multi-branch error contribution output is constructed, and the fused hidden features are mapped to normalized spatial error contribution, normalized time period error contribution, and normalized meteorological disturbance error contribution, respectively.
[0025] Furthermore, in step S5, the error contribution fusion layer design learns the importance of each error source under different environmental conditions through a weight generation network to obtain the error contribution weight;
[0026] For the normalized error contribution, inverse normalization is performed based on the statistical scale of satellite PWV observation sample error to obtain the actual PWV error of each error contribution.
[0027] The actual PWV error is fused using the error contribution weight to obtain the final satellite PWV compensation error;
[0028] Corrected atmospheric precipitation data for geostationary satellites.
[0029] Further, in step S6, the output of the nonlinear correction network uses the satellite PWV error influence characteristics enhanced by spatiotemporal propagation as the network input to construct the nonlinear correction network, while introducing water vapor field continuity constraints and correction increment constraints; including:
[0030] Using the satellite PWV error impact characteristics after spatiotemporal propagation enhancement of satellite PWV observation samples as network input, a PWV error correction network is constructed. The network structure includes an input layer, an error feature mapping layer, and an error contribution fusion layer.
[0031] The error feature mapping layer uses the GELU activation function to extract multi-factor coupled error features; it outputs the spatial, temporal and meteorological error contributions respectively.
[0032] The error contribution fusion layer uses error contribution weights to fuse the error contribution amounts to obtain the final satellite PWV compensation error; and obtains the corrected geostationary orbit satellite PWV;
[0033] A correction amplitude adjustment coefficient is constructed based on the difference between the satellite PWV observation value and the GNSS reference PWV; the PWV correction increment is determined based on the difference between the corrected geostationary satellite PWV and the original satellite PWV; and a correction increment constraint is constructed.
[0034] The training loss function includes data error loss, water vapor field continuity constraint, and correction increment constraint;
[0035] Correction result output: The trained model is applied to the new geostationary satellite PWV product to achieve automatic pixel-by-pixel correction.
[0036] This invention provides a nonlinear correction system for atmospheric precipitable water on geostationary satellites, comprising a satellite PWV observation sample construction module, a satellite error characterization module, a spatiotemporal propagation enhancement module, an error feature mapping layer design module, an error contribution fusion layer design module, and a nonlinear correction network output module;
[0037] The satellite PWV observation sample construction module acquires PWV observation values from geostationary orbit satellites and GNSS inverted PWV reference values, constructs satellite PWV observation samples, and calculates satellite PWV errors.
[0038] The satellite error characterization module is based on satellite PWV error, extracts features from three dimensions: space, time and meteorology, standardizes them and performs joint encoding to construct satellite PWV error impact features.
[0039] The spatiotemporal propagation enhancement module utilizes the error correlation between PWV observations from geostationary satellites in spatial neighborhoods and between consecutive observation times to enhance the spatiotemporal neighborhood characteristics of satellite PWV error impact.
[0040] The error feature mapping layer design module takes the satellite PWV error impact features enhanced by spatiotemporal propagation as input, extracts error hidden features, and maps them to generate normalized error contribution amounts respectively.
[0041] The error contribution fusion layer design module takes the normalized error contribution as input, obtains the error contribution weight through the weight generation network, and then performs weighted fusion after inversely normalizing various normalized error contribution quantities into the actual PWV error to output the satellite PWV compensation error.
[0042] The nonlinear correction network output module uses the satellite PWV error impact characteristics enhanced by spatiotemporal propagation as the network input, and constructs a nonlinear correction network by combining the error feature mapping layer and the error contribution fusion layer. After the network training is completed, real-time nonlinear correction of atmospheric precipitable water of geostationary satellites is performed.
[0043] The beneficial effects achieved by adopting the above solution are as follows:
[0044] (1) In view of the problem that the general geostationary satellite atmospheric precipitable water nonlinear correction method does not make full use of the error correlation between the spatial neighboring region and the continuous observation time, which can easily lead to unnecessary differences in the correction results obtained by adjacent pixels or adjacent time, reduce the spatiotemporal continuity of the corrected PWV field, and thus affect the accuracy of the correction results, this scheme uses the neighboring satellite PWV error consistency to screen the spatiotemporal propagation conditions. Neighborhood feature propagation is only carried out when the current satellite PWV observation sample and the spatiotemporal neighboring satellite PWV observation sample have a high error consistency. The degree of fusion between the neighborhood features and the current satellite PWV observation sample features is controlled by the spatiotemporal neighborhood propagation coefficient. In this way, while using the spatiotemporal continuous observation information of geostationary satellites to improve the continuous expression ability of PWV error features, it avoids the cross-regional propagation of abnormal errors and the excessive smoothing of the real local water vapor change features, thereby improving the accuracy of the geostationary satellite atmospheric precipitable water nonlinear correction results.
[0045] (2) To address the problem that general geostationary satellite atmospheric precipitable water nonlinear correction methods are prone to generating excessive correction amounts in local regions, causing spatial abrupt changes in the corrected PWV and making it impossible to adjust the correction amplitude according to the original error level of different satellite PWV observation samples, this scheme limits the spatial variation difference between the corrected PWV and the real water vapor field by constraining the continuity of the water vapor field; constructs a correction amplitude adjustment coefficient based on the difference between the satellite PWV observation value and the GNSS reference PWV, and adaptively constrains the correction increment, so that the satellite PWV with smaller original error maintains a smaller correction amplitude, while the satellite PWV with larger original error obtains a larger degree of correction freedom, thereby reducing local abnormal correction and overcorrection, and improving the stability of the geostationary satellite atmospheric precipitable water nonlinear correction results. Attached Figure Description
[0046] Figure 1 A flowchart illustrating a nonlinear correction method for atmospheric precipitable water on a geostationary satellite provided by the present invention;
[0047] Figure 2 This is a schematic diagram of a nonlinear correction system for atmospheric precipitable water on a geostationary satellite, provided by the present invention.
[0048] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used together with the embodiments of the invention to explain the invention and do not constitute a limitation thereof. Detailed Implementation
[0049] 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. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.
[0050] In the description of this invention, it should be understood that the terms "upper", "lower", "front", "rear", "left", "right", "top", "bottom", "inner", "outer", etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the system or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.
[0051] Example 1, see Figure 1 The present invention provides a nonlinear correction method for atmospheric precipitable water content of geostationary satellites, the method comprising the following steps:
[0052] Step S1: Satellite PWV observation sample construction, obtain geostationary orbit satellite PWV observation values and GNSS inverted PWV reference values, construct satellite PWV observation samples and calculate satellite PWV error;
[0053] Step S2: Satellite error characterization. Based on satellite PWV error, features are extracted from three dimensions: space, time, and meteorology. After standardization, these features are jointly encoded to construct the satellite PWV error impact features.
[0054] Step S3: Spatiotemporal propagation enhancement. By utilizing the error correlation between PWV observations from geostationary satellites in the spatial neighborhood and between consecutive observation times, the spatiotemporal neighborhood characteristics of satellite PWV error impact are enhanced.
[0055] Step S4: Error feature mapping layer design. Taking the satellite PWV error impact features enhanced by spatiotemporal propagation as input, extract the hidden error features and map them to generate normalized error contribution amounts respectively.
[0056] Step S5: Error contribution fusion layer design. The normalized error contribution is used as input. The error contribution weight is obtained through the weight generation network. The various normalized error contribution quantities are denormalized to the actual PWV error and then weighted and fused to output the satellite PWV compensation error.
[0057] Step S6: Output of the nonlinear correction network. The satellite PWV error impact characteristics after spatiotemporal propagation enhancement are used as the network input. The nonlinear correction network is constructed by combining the error feature mapping layer and the error contribution fusion layer. After the network training is completed, real-time nonlinear correction of atmospheric precipitable water of geostationary satellites is performed.
[0058] Example 2, see Figure 1 This embodiment is based on the above embodiment. In step S1, the satellite PWV observation sample construction is to establish satellite PWV observation samples using high-precision ground-based observations, using geostationary orbit satellite PWV observation values as satellite PWV values to be corrected, and GNSS inverted PWV reference values as training supervision values.
[0059] Match satellite PWV observation samples with consistent time and spatial location, calculate the deviation of the satellite PWV value from the GNSS reference PWV value, and use this as the satellite PWV error, expressed as:
[0060] ; This is the satellite PWV error, in mm; This is the GNSS inversion PWV reference value, in mm; These are PWV observations from geostationary satellites, in mm.
[0061] Example 3, see Figure 1 This embodiment is based on the above embodiment. In step S2, the satellite error characterization is to characterize the influencing factors of the satellite PWV error; including:
[0062] The satellite PWV error is divided into three dimensions: spatial, temporal, and meteorological, and corresponding feature vectors are constructed for each dimension.
[0063] The time dimension is used to describe the periodic errors caused by the solar radiation cycle and the seasonal variation of water vapor;
[0064] The meteorological dimension is used to describe the dynamic impact of changes in temperature, pressure, and water vapor content on the satellite remote sensing inversion process;
[0065] Spatial error affects eigenvectors ; It is the longitude of the satellite observation position, in degrees; H is the latitude of the satellite observation location, in degrees; H is the Earth's surface elevation, in meters.
[0066] Time error affects feature vector By employing sine and cosine periodic mapping, the time variable is transformed from discrete dates into a continuous periodic space: ; This is the accumulated days of the year;
[0067] Meteorological error affects the feature vector T is the ground temperature, in K; e is the ground vapor pressure, in hPa; P is the ground air pressure, in hPa.
[0068] Z-score standardization is applied to the error impact features of each dimension. The standardized spatial, temporal, and meteorological error impact feature vectors are represented as follows: , and ;
[0069] Joint encoding combines the three-dimensional features to form the satellite PWV error impact features. , .
[0070] Example 4, see Figure 1 This embodiment is based on the above embodiment. In step S3, the spatiotemporal propagation enhancement involves performing spatiotemporal propagation screening and, based on the screening results, utilizing the error correlation between geostationary satellite PWV observations in the spatial neighborhood and between consecutive observation times to perform spatiotemporal neighborhood enhancement on the satellite PWV error impact characteristics; including:
[0071] For the current PWV observation sample of the i-th satellite, select neighboring satellite PWV observation samples based on spatial distance and time interval to construct a spatiotemporal neighborhood set N. i The normalized spatial distance between the neighboring satellite PWV observation sample in the spatiotemporal neighborhood set and the current satellite PWV observation sample does not exceed the preset spatial neighborhood range (0.4~0.8), and the normalized time interval does not exceed the preset time neighborhood range (0.4~0.8).
[0072] For the current satellite PWV observation sample i and the neighboring satellite PWV observation sample j, construct the spatiotemporal propagation weights based on the normalized spatial distance and time interval between them: ; and These are the normalized spatial distances between the current satellite PWV observation sample i and the neighboring satellite PWV observation samples j and k, respectively. and These are the normalized time intervals between the current satellite PWV observation sample i and the neighboring satellite PWV observation samples j and k, respectively. It is a spatial propagation scale parameter, with a value ranging from 0.2 to 0.8; It is a time propagation scale parameter, with a value ranging from 0.2 to 0.8; It is the spatiotemporal propagation weight of neighboring satellite PWV observation sample j to current satellite PWV observation sample i;
[0073] For the selection of PWV observation samples from the spatiotemporal propagation satellite, for the i-th satellite PWV observation sample, the neighborhood error consistency is calculated within the spatiotemporal neighborhood: ; It is the neighborhood error consistency of the PWV observation sample of the i-th satellite; and These are the normalized satellite PWV errors of the current satellite PWV observation samples and the neighboring satellite PWV observation samples, respectively; a propagation decision threshold τ is set. c ,when The more consistent the error changes of the current satellite PWV observation sample with those of satellite PWV observation samples in its spatiotemporal neighborhood, the stronger the spatiotemporal error continuity, and the current satellite PWV observation sample is identified as a sample with enhanced spatiotemporal propagation; conversely, it indicates that there are local water vapor abrupt changes and rapid weather changes in the current region, and the current satellite PWV observation sample is identified as a non-propagation sample to avoid neighborhood errors from propagating abnormal errors to the current satellite PWV observation sample; Values range from 0.05 to 0.30;
[0074] For spatiotemporal propagation enhancement, the satellite PWV error impact characteristics of the spatiotemporal propagation enhanced samples are adaptively fused with spatiotemporal neighborhood characteristics to obtain the satellite PWV error impact characteristics after spatiotemporal propagation enhancement. ; It is the satellite PWV error impact characteristic of the spatiotemporal propagation enhancement sample; It is the satellite PWV error influence characteristic of neighboring satellite PWV observation samples; It is the spatiotemporal neighborhood propagation coefficient, used to control the degree of propagation of neighborhood error influence characteristics to the current satellite PWV observation sample, with a value of 0~1; It is the characteristic of satellite PWV error impact after the spatiotemporal propagation enhancement of the current satellite PWV observation sample; it improves the spatiotemporal continuity expression capability of satellite PWV error characteristics.
[0075] By performing the above operations, this scheme addresses the problem that conventional geostationary satellite atmospheric precipitable water nonlinear correction methods fail to fully utilize the error correlation between spatially adjacent regions and continuous observation times, easily leading to unnecessary differences in correction results obtained from adjacent pixels or adjacent times, reducing the spatiotemporal continuity of the corrected PWV field, and thus affecting the accuracy of the correction results. This scheme utilizes the consistency of neighboring satellite PWV errors to screen spatiotemporal propagation conditions. Neighborhood feature propagation is only performed when the current satellite PWV observation sample and the spatiotemporally adjacent satellite PWV observation samples have high error consistency. Furthermore, the spatiotemporal neighborhood propagation coefficient controls the degree of fusion between neighborhood features and the features of the current satellite PWV observation sample. This approach enhances the continuous expression capability of PWV error features by utilizing continuous spatiotemporal observation information from geostationary satellites while avoiding the cross-regional propagation of abnormal errors and the over-smoothing of true local water vapor abrupt changes, thereby improving the accuracy of geostationary satellite atmospheric precipitable water nonlinear correction results.
[0076] Example 5, see Figure 1 This embodiment is based on the above embodiment. In step S4, the error feature mapping layer design is based on the satellite PWV error influence characteristics, establishing a nonlinear mapping relationship between influencing factors and satellite PWV error; including:
[0077] Error hidden features under the combined influence of different environmental factors are extracted using a nonlinear feature mapping network, and are represented as follows: ; It is an error-hiding feature; It is the error feature mapping weight matrix; It is a bias term; It is a non-linear activation function, using the GELU function; These are the satellite PWV error impact characteristics after spatiotemporal propagation enhancement; both the satellite PWV error impact characteristics of non-propagation samples and the satellite PWV error impact characteristics after spatiotemporal propagation enhancement are represented by... express;
[0078] To improve the interpretability of error sources, a multi-branch error contribution output is constructed, mapping the fused hidden features to normalized spatial error contribution, normalized temporal periodic error contribution, and normalized meteorological disturbance error contribution, respectively: Normalized Spatial Error Contribution Normalized time error contribution Normalized meteorological error contribution ; , and These are three types of normalized error contribution output layer weight matrices; , and These are the three types of normalized error contribution bias terms.
[0079] Example 6, see Figure 1 This embodiment is based on the above embodiment. In step S5, the error contribution fusion layer is designed so that the influence of spatial factors, time cycle factors, and meteorological disturbance factors on the PWV inversion error of geostationary satellites varies under different geographical regions, seasonal conditions, and atmospheric water vapor conditions. Therefore, the weight generation network learns the importance of each error source under different environmental conditions to obtain the error contribution weight A, which is expressed as: Expand ; , and These are the spatial error contribution weights, the temporal period error contribution weights, and the meteorological disturbance error contribution weights; satisfying... ; and These are the learnable weight matrix and bias vector of the weight generation network, respectively.
[0080] For the normalized error contribution, inverse normalization is performed based on the statistical scale of the error in the PWV observation samples of the training satellite to restore the actual PWV error dimensions, thus obtaining the actual PWV error for each error contribution, which are expressed as follows: , and ;
[0081] The actual PWV error is fused using error contribution weights to obtain the final satellite PWV compensation error. ;
[0082] The corrected atmospheric precipitable water content of the geostationary satellite is obtained (satellite PWV compensation error + satellite PWV observation value).
[0083] Example 7, see Figure 1 This embodiment is based on the above embodiment. In step S6, the nonlinear correction network output uses the satellite PWV error influence characteristics after spatiotemporal propagation enhancement as the network input to construct the nonlinear correction network. Simultaneously, it introduces water vapor field continuity constraints and correction increment constraints to avoid abnormal abrupt changes in the correction results that do not conform to actual atmospheric change patterns, and to avoid over-correction of the original satellite PWV with relatively small errors; including:
[0084] Using the satellite PWV error impact characteristics after spatiotemporal propagation enhancement of satellite PWV observation samples (including the original satellite PWV error impact characteristics of non-propagation samples) as network input, a PWV error correction network is constructed. The network structure includes an input layer, an error feature mapping layer, and an error contribution fusion layer.
[0085] The input layer is 8-dimensional, including 3-dimensional spatial features, 2-dimensional temporal periodic features, and 3-dimensional meteorological features;
[0086] The error feature mapping layer uses the GELU activation function to extract multi-factor coupled error features; it outputs the spatial, temporal and meteorological error contributions respectively.
[0087] The error contribution fusion layer uses error contribution weights to fuse the error contribution amounts to obtain the final satellite PWV compensation error; and obtains the corrected geostationary orbit satellite PWV;
[0088] To avoid excessive correction by the nonlinear correction network in regions where the original satellite PWV error is small, a correction amplitude adjustment coefficient is constructed based on the difference between the satellite PWV observations and the GNSS reference PWV. , is represented as: ; This is the correction amplitude adjustment parameter, with a value ranging from 0.1 to 0.5; It is the normalized PWV observation value of the i-th geostationary satellite; It is the normalized GNSS reference PWV; the PWV correction increment is determined based on the difference between the corrected geostationary satellite PWV and the original satellite PWV. , is represented as: Constructing correction increment constraints N is the number of satellite PWV observation samples;
[0089] Training loss functions include: data error loss Continuity constraints of the water vapor field Total loss ;in, It is the corrected PWV value of the geostationary satellite corresponding to the normalized PWV observation sample of the i-th satellite; It is the normalized spatial gradient of the PWV corrected by the PWV observation sample of the i-th satellite, which represents the rate at which the PWV changes with spatial location; It is the normalized spatial gradient of the GNSS reference PWV of the PWV observation sample of the i-th satellite, representing the spatial variation trend of the real water vapor field;
[0090] Model training process: During the training phase, GNSS-inverted PWV is used as a reference value, and network parameters are optimized through loss backpropagation.
[0091] Initialize network parameters: W∼N(0,0.01); bias: b=0; input satellite PWV observation samples; calculate the error contribution of each of the three types; calculate the error contribution weight; obtain the corrected PWV; calculate the loss based on the prediction results and GNSS reference values; use Adam to optimize and update parameters;
[0092] Correction result output: Apply the trained model to the new geostationary satellite PWV product to achieve automatic pixel-by-pixel correction: ; This is the corrected PWV value for geostationary satellites; These are new PWV observations from geostationary satellites.
[0093] By performing the above operations, this scheme addresses the problem that conventional geostationary satellite atmospheric precipitable water volume (PWV) nonlinear correction methods tend to generate excessive correction values in local regions, causing spatial abrupt changes in the corrected PWV and making it impossible to adjust the correction amplitude based on the original error level of different satellite PWV observation samples. This scheme limits the spatial variation difference between the corrected PWV and the actual water vapor field by constraining the continuity of the water vapor field. It constructs a correction amplitude adjustment coefficient based on the difference between the satellite PWV observations and the GNSS reference PWV, adaptively constraining the correction increment. This ensures that satellite PWVs with smaller original errors maintain a smaller correction amplitude, while satellite PWVs with larger original errors gain greater correction freedom, thereby reducing local abnormal corrections and overcorrections, and improving the stability of the geostationary satellite atmospheric precipitable water volume nonlinear correction results.
[0094] Example 8, see Figure 2 Based on the above embodiments, this embodiment provides a nonlinear correction system for atmospheric precipitable water on geostationary satellites, including a satellite PWV observation sample construction module, a satellite error characterization module, a spatiotemporal propagation enhancement module, an error feature mapping layer design module, an error contribution fusion layer design module, and a nonlinear correction network output module.
[0095] The satellite PWV observation sample construction module acquires PWV observation values from geostationary orbit satellites and GNSS inverted PWV reference values, constructs satellite PWV observation samples, and calculates satellite PWV errors.
[0096] The satellite error characterization module is based on satellite PWV error, extracts features from three dimensions: space, time and meteorology, standardizes them and performs joint encoding to construct satellite PWV error impact features.
[0097] The spatiotemporal propagation enhancement module utilizes the error correlation between PWV observations from geostationary satellites in spatial neighborhoods and between consecutive observation times to enhance the spatiotemporal neighborhood characteristics of satellite PWV error impact.
[0098] The error feature mapping layer design module takes the satellite PWV error impact features enhanced by spatiotemporal propagation as input, extracts error hidden features, and maps them to generate normalized error contribution amounts respectively.
[0099] The error contribution fusion layer design module takes the normalized error contribution as input, obtains the error contribution weight through the weight generation network, and then performs weighted fusion after inversely normalizing various normalized error contribution quantities into the actual PWV error to output the satellite PWV compensation error.
[0100] The nonlinear correction network output module uses the satellite PWV error impact characteristics enhanced by spatiotemporal propagation as the network input, and constructs a nonlinear correction network by combining the error feature mapping layer and the error contribution fusion layer. After the network training is completed, real-time nonlinear correction of atmospheric precipitable water of geostationary satellites is performed.
[0101] It should be noted that, in this document, the terms “comprising,” “including,” or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.
[0102] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention.
[0103] The present invention and its embodiments have been described above. This description is not restrictive, and the accompanying drawings are only one embodiment of the present invention; the actual structure is not limited thereto. In conclusion, if those skilled in the art are inspired by this description and design similar structures and embodiments without departing from the spirit of the invention, such designs should fall within the protection scope of the present invention.
Claims
1. A nonlinear correction method for atmospheric precipitable water content of geostationary satellites, characterized in that: The method includes the following steps: Step S1: Satellite PWV observation sample construction, obtain geostationary orbit satellite PWV observation values and GNSS inverted PWV reference values, construct satellite PWV observation samples and calculate satellite PWV error; Step S2: Satellite error characterization. Based on satellite PWV error, features are extracted from three dimensions: space, time, and meteorology. After standardization, these features are jointly coded to construct the satellite PWV error impact features. Step S3: Spatiotemporal propagation enhancement. By utilizing the error correlation between PWV observations from geostationary satellites in the spatial neighborhood and between consecutive observation times, the spatiotemporal neighborhood characteristics of satellite PWV error impact are enhanced. Step S4: Error feature mapping layer design. Taking the satellite PWV error impact features enhanced by spatiotemporal propagation as input, extract the hidden error features and map them to generate normalized error contribution amounts respectively. Step S5: Error contribution fusion layer design. The normalized error contribution is used as input. The error contribution weight is obtained through the weight generation network. The various normalized error contribution quantities are denormalized to the actual PWV error and then weighted and fused to output the satellite PWV compensation error. Step S6: Output of the nonlinear correction network. The satellite PWV error impact characteristics after spatiotemporal propagation enhancement are used as the network input. The nonlinear correction network is constructed by combining the error feature mapping layer and the error contribution fusion layer. After the network training is completed, real-time nonlinear correction of atmospheric precipitable water of geostationary satellites is performed.
2. The nonlinear correction method for atmospheric precipitable water content of a geostationary satellite according to claim 1, characterized in that: In step S3, the spatiotemporal propagation enhancement involves performing spatiotemporal propagation screening and, based on the screening results, utilizing the error correlation between geostationary satellite PWV observations in the spatial neighborhood and between consecutive observation times to perform spatiotemporal neighborhood enhancement on the satellite PWV error impact characteristics; including: For the current satellite PWV observation sample, select neighboring satellite PWV observation samples according to spatial distance and time interval to construct a spatiotemporal neighborhood set; For current satellite PWV observation samples and neighboring satellite PWV observation samples, spatiotemporal propagation weights are constructed based on the normalized spatial distance and time interval between the two. For the selection of PWV observation samples from the spatiotemporal propagation satellite, for the i-th satellite PWV observation sample, the neighborhood error consistency is calculated within the spatiotemporal neighborhood. Set the propagation determination threshold τ c ,when The current satellite PWV observation sample is identified as a spatiotemporal propagation enhanced sample; otherwise, the current satellite PWV observation sample is identified as a non-propagation sample. For spatiotemporal propagation enhancement, the satellite PWV error impact characteristics of the spatiotemporal propagation enhanced samples are adaptively fused with the spatiotemporal neighborhood characteristics to obtain the satellite PWV error impact characteristics after spatiotemporal propagation enhancement.
3. The nonlinear correction method for atmospheric precipitable water content of a geostationary satellite according to claim 1, characterized in that: In step S6, the output of the nonlinear correction network uses the satellite PWV error influence characteristics enhanced by spatiotemporal propagation as the network input to construct the nonlinear correction network, while introducing water vapor field continuity constraints and correction increment constraints; including: Using the satellite PWV error impact characteristics after spatiotemporal propagation enhancement of satellite PWV observation samples as network input, a PWV error correction network is constructed. The network structure includes an input layer, an error feature mapping layer, and an error contribution fusion layer. The error feature mapping layer uses the GELU activation function to extract multi-factor coupled error features; it outputs the spatial, temporal and meteorological error contributions respectively. The error contribution fusion layer uses error contribution weights to fuse the error contribution amounts to obtain the final satellite PWV compensation error; and obtains the corrected geostationary orbit satellite PWV; A correction amplitude adjustment coefficient is constructed based on the difference between the satellite PWV observation value and the GNSS reference PWV; the PWV correction increment is determined based on the difference between the corrected geostationary satellite PWV and the original satellite PWV; and a correction increment constraint is constructed. The training loss function includes data error loss, water vapor field continuity constraint, and correction increment constraint; Correction result output: The trained model is applied to the new geostationary satellite PWV product to achieve automatic pixel-by-pixel correction.
4. The nonlinear correction method for atmospheric precipitable water content of a geostationary satellite according to claim 1, characterized in that: In step S2, the satellite error characterization involves characterizing the influencing factors of the satellite PWV error; including: The satellite PWV error is divided into three dimensions: spatial, temporal, and meteorological, and corresponding feature vectors are constructed for each dimension. Z-score standardization is applied to the error impact characteristics of each dimension; Joint encoding combines the three-dimensional features to form the satellite PWV error impact features.
5. The nonlinear correction method for atmospheric precipitable water content of a geostationary satellite according to claim 1, characterized in that: In step S4, the error feature mapping layer design is based on the satellite PWV error impact characteristics, establishing a nonlinear mapping relationship between influencing factors and satellite PWV error; including: Error hidden features under the combined influence of different environmental factors are extracted using a nonlinear feature mapping network; A multi-branch error contribution output is constructed, and the fused hidden features are mapped to normalized spatial error contribution, normalized time period error contribution, and normalized meteorological disturbance error contribution, respectively.
6. The nonlinear correction method for atmospheric precipitable water content of a geostationary satellite according to claim 1, characterized in that: In step S5, the error contribution fusion layer design learns the importance of each error source under different environmental conditions through a weight generation network to obtain the error contribution weights; For the normalized error contribution, inverse normalization is performed based on the statistical scale of satellite PWV observation sample error to obtain the actual PWV error of each error contribution. The actual PWV error is fused using the error contribution weight to obtain the final satellite PWV compensation error; Corrected atmospheric precipitation data for geostationary satellites.
7. The nonlinear correction method for atmospheric precipitable water content of a geostationary satellite according to claim 1, characterized in that: In step S1, the satellite PWV observation sample construction is to establish satellite PWV observation samples using high-precision ground-based observations, using geostationary orbit satellite PWV observation values as satellite PWV values to be corrected, and GNSS inverted PWV reference values as training supervision values. Match satellite PWV observation samples with consistent time and spatial location, calculate the deviation of satellite PWV values from GNSS reference PWV values, and use this as the satellite PWV error.
8. A nonlinear correction system for atmospheric precipitable water content of a geostationary satellite, used to implement the nonlinear correction method for atmospheric precipitable water content of a geostationary satellite as described in any one of claims 1-7, characterized in that: It includes a satellite PWV observation sample construction module, a satellite error characterization module, a spatiotemporal propagation enhancement module, an error feature mapping layer design module, an error contribution fusion layer design module, and a nonlinear correction network output module; The satellite PWV observation sample construction module acquires PWV observation values from geostationary orbit satellites and GNSS inverted PWV reference values, constructs satellite PWV observation samples, and calculates satellite PWV errors. The satellite error characterization module is based on satellite PWV error, extracts features from three dimensions: space, time and meteorology, standardizes them and performs joint encoding to construct satellite PWV error impact features. The spatiotemporal propagation enhancement module utilizes the error correlation between PWV observations from geostationary satellites in spatial neighborhoods and between consecutive observation times to enhance the spatiotemporal neighborhood characteristics of satellite PWV error impact. The error feature mapping layer design module takes the satellite PWV error impact features enhanced by spatiotemporal propagation as input, extracts error hidden features, and maps them to generate normalized error contribution amounts respectively. The error contribution fusion layer design module takes the normalized error contribution as input, obtains the error contribution weight through the weight generation network, and then performs weighted fusion after inversely normalizing various normalized error contribution quantities into the actual PWV error to output the satellite PWV compensation error. The nonlinear correction network output module uses the satellite PWV error impact characteristics enhanced by spatiotemporal propagation as the network input, and constructs a nonlinear correction network by combining the error feature mapping layer and the error contribution fusion layer. After the network training is completed, real-time nonlinear correction of atmospheric precipitable water of geostationary satellites is performed.