A high-precision retrieval method and system for cloud base height of a domestic passive meteorological satellite

Abstract

The application provides a domestic passive meteorological satellite cloud bottom height high-precision inversion method and system, and belongs to the field of atmospheric cloud parameter remote sensing inversion and quantitative remote sensing mechanism. The application fuses satellite multi-channel atmospheric top reflectivity observation, brightness temperature observation and atmospheric temperature profile reanalysis data, develops a complex nonlinear feature extraction and modeling method, improves the nonlinear mapping capability among the atmospheric top reflectivity, the brightness temperature and the cloud bottom height, combines the whole period infrared channel characteristics and the stratification information, enhances the night and all-time inversion capability, improves the generalization adaptability of the model through the multi-source sample expansion and the stacking learning strategy, designs the sample expansion and the auxiliary training mechanism, effectively alleviates the problem of insufficient high-quality training data, and finally realizes the high-precision inversion of the all-time cloud bottom height suitable for the domestic stationary or polar orbit meteorological satellite.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of atmospheric cloud parameter remote sensing inversion and quantitative remote sensing mechanism, and particularly relates to a high-precision retrieval method and system for cloud base height of a domestic passive meteorological satellite. BACKGROUND

[0002] Cloud Base Height (CBH) is an important physical parameter for describing the vertical structure of cloud, which is of great significance for weather monitoring, climate change assessment, radiation budget calculation and aviation safety.

[0003] At present, there are many methods for retrieving cloud base height using passive satellite data internationally. The commonly used schemes are mainly divided into three categories: (1) indirect scheme of cloud thickness; (2) scheme based on the ratio of cloud water path to liquid water path (ice water path to liquid water path); (3) machine learning scheme. Among the three main schemes for retrieving cloud base height, because the cloud effective path, cloud effective emissivity, cloud optical thickness, cloud water path and cloud liquid water content parameters cannot be obtained for free on the official website of the National Meteorological Satellite Center, the schemes based on cloud thickness and cloud water path cannot be used for the retrieval of cloud base height of the domestic meteorological satellite developed independently by China. Therefore, the machine learning scheme is expected to become a general scheme for the retrieval of all-weather cloud base height of the domestic meteorological satellite.

[0004] In the prior art, the cloud base height retrieval method based on meteorological satellite has been widely used in meteorological monitoring and climate research field. Among them, the cloud base height retrieval technology based on machine learning method has gradually become a research hotspot in recent years due to its strong nonlinear fitting ability and multi-source data fusion advantage, but there are still many deficiencies in the practical application of this technology system: 1. The multi-channel brightness temperature information provided by satellite passive remote sensing mainly reflects the radiation characteristics of the cloud top and the upper layer, while the cloud base height is controlled by the complex factors such as cloud internal microphysical process, atmospheric stratification structure and radiation transmission, and there is no direct linear or unique mapping relationship between brightness temperature information and cloud base height. There is a high degree of non-linear coupling between them. This makes it difficult for traditional machine learning models to fully characterize the complex non-linear relationship between brightness temperature and cloud base height during the training process, and is prone to fitting deviation or model instability problems; 2. Most of the existing machine learning models rely on daytime visible light and short-wave infrared channel information, and the relevant channels are missing at night, which leads to a significant reduction in the input feature dimension of the model under night or all-day conditions, affecting the accuracy and applicability of cloud base height retrieval, and cannot meet the all-weather observation demand; 3. The existing machine learning models generally have strong dependence on training sample area and poor climate adaptability. Most of the models are concentrated in specific regions, specific seasons or specific cloud types, lack of comprehensive sample support covering a wide range, multiple climate zones, multiple time periods and multiple cloud types, and the generalization ability of the model under cross-regional, cross-seasonal or complex weather system conditions is limited, and the stability is insufficient; 4. Although active remote sensing satellites can provide high-precision cloud vertical profile data to provide reference data sources for machine learning model training, due to the limited spatial and temporal coverage, it is difficult to form a long-term continuous, large sample and balanced distribution of training data set, especially in complex cloud system conditions such as extreme weather and multi-layer cloud overlap, the training sample is seriously insufficient, which restricts the training sufficiency and application expansion ability of the machine learning model. SUMMARY

[0005] To solve the above problems in the prior art, the present application provides a high-precision retrieval method and system for cloud base height of domestic passive meteorological satellite. The present application develops a complex nonlinear feature extraction and modeling method by fusing satellite multi-channel atmospheric top reflectivity observation, brightness temperature observation and atmospheric temperature profile reanalysis data, which improves the nonlinear mapping ability between atmospheric top reflectivity, brightness temperature and cloud base height; combined with the characteristics of the infrared channel and the stratification information in all time periods, the night and all-time retrieval ability is enhanced; through the multi-source sample expansion and stacking learning strategy, the generalization adaptability of the model is improved; and the sample expansion and auxiliary training mechanism is designed, which effectively alleviates the problem of insufficient high-quality training data, and finally realizes the high-precision retrieval of all-time cloud base height suitable for domestic geostationary or polar orbit meteorological satellite. In order to achieve the above purpose, the technical scheme is as follows:

[0006] In one aspect, the application provides a high-precision retrieval method for cloud base height of a domestic passive meteorological satellite, which comprises the following steps:

[0007] S1, according to the existing first stationary meteorological satellite data set and the active remote sensing satellite cloud profile data set, a multi-source database set is obtained through time matching and space matching;

[0008] S2, according to the multi-source database set, an atmospheric top brightness temperature data set is obtained by comparing with the second stationary meteorological satellite data set and performing multi-channel spectral function response, the atmospheric top brightness temperature data set includes the atmospheric top brightness temperature data set of the first stationary meteorological satellite and the atmospheric top brightness temperature data set of the second stationary meteorological satellite;

[0009] S3, according to the atmospheric top brightness temperature data set, the second stationary meteorological satellite data set is corrected to obtain the retrieval data set of the first stationary meteorological satellite;

[0010] S4, the multi-source database set and the retrieval data set of the first stationary meteorological satellite are merged to obtain an all-time retrieval data set;

[0011] S5, according to the all-time retrieval data set, the all-time cloud base height is obtained by a stacked machine learning method.

[0012] Optionally, the multi-source database set includes L1 level multi-channel data and solar zenith angle data, L2 level cloud amount data and cloud mask data, cloud base height, temperature profile and relative humidity profile data, atmospheric total water vapor content and elevation data.

[0013] Optionally, in S2, the atmospheric top brightness temperature data set is obtained by comparing the multi-channel spectral response function of the multi-source database set with the second stationary meteorological satellite data set, which includes:

[0014] S21, according to the multi-source database set and the second stationary meteorological satellite data set, the brightness temperature channel with large spectral response function difference is obtained by comparing the spectral response function;

[0015] S22, according to the brightness temperature channel with large spectral response function difference, the atmospheric top brightness temperature data set is obtained by MODTRAN radiation transfer simulation.

[0016] Optionally, in S3, according to the atmospheric top brightness temperature data set, the second stationary meteorological satellite data set is corrected to obtain the retrieval data set of the first stationary meteorological satellite, which includes:

[0017] S31, according to the atmospheric top brightness temperature data set, a linear relationship model between the brightness temperature channels of the first stationary meteorological satellite and the second stationary meteorological satellite is established;

[0018] S32, according to the linear relationship model between the brightness temperature channels of the first geostationary meteorological satellite and the second geostationary meteorological satellite, the second geostationary meteorological satellite dataset is corrected to obtain the first geostationary meteorological satellite inversion dataset.

[0019] Optionally, according to the all-day inversion dataset, the all-day cloud base height is obtained by a stacked machine learning method in S5, including:

[0020] S51, according to the all-day inversion dataset, the daytime inversion data and the nighttime inversion data are screened to obtain a daytime inversion dataset and a nighttime inversion dataset;

[0021] S52, according to the daytime inversion dataset, a plurality of daytime preliminary cloud base heights are obtained by different machine learning models;

[0022] S53, according to the nighttime inversion dataset, a plurality of nighttime preliminary cloud base heights are obtained by different machine learning models;

[0023] S54, according to the plurality of daytime preliminary cloud base heights, a daytime cloud base height is obtained by a stacking model;

[0024] S55, according to the plurality of nighttime preliminary cloud base heights, a nighttime cloud base height is obtained by a stacking model;

[0025] S56, according to the daytime cloud base height and the nighttime cloud base height, an all-day cloud base height is obtained.

[0026] Optionally, the machine learning model includes: an RF machine learning model, an XGBOOST machine learning model, and a LightGBM machine learning model.

[0027] Optionally, the method of screening daytime inversion data and nighttime inversion data includes:

[0028] If the solar zenith angle in the all-day inversion dataset is less than 90, it is put into the daytime inversion dataset; if the solar zenith angle in the all-day inversion dataset is not less than 90, it is put into the nighttime inversion dataset.

[0029] On the other hand, the present application provides a domestic passive meteorological satellite cloud base height high-precision inversion system, which is applied to a domestic passive meteorological satellite cloud base height high-precision inversion method, and the system includes:

[0030] A database set construction module is used to obtain a multi-source database set by time matching and space matching according to an existing first geostationary meteorological satellite dataset and an active remote sensing satellite cloud profile dataset;

[0031] a data set comparison module configured to obtain an atmospheric top brightness temperature data set by comparing the second geostationary meteorological satellite data set and performing a multi-channel spectral function response according to the multi-source database set, the atmospheric top brightness temperature data set comprising the atmospheric top brightness temperature data set of the first geostationary meteorological satellite and the atmospheric top brightness temperature data set of the second geostationary meteorological satellite;

[0032] a data set correction module configured to correct the second geostationary meteorological satellite data set according to the atmospheric top brightness temperature data set to obtain an inversion data set of the first geostationary meteorological satellite;

[0033] an inversion data set construction module configured to merge the multi-source database set and the inversion data set of the first geostationary meteorological satellite to obtain an all-time inversion data set;

[0034] a cloud bottom height inversion module configured to obtain an all-time cloud bottom height by a stacked machine learning method according to the all-time inversion data set.

[0035] Compared with the prior art, the technical scheme of the present application has at least the following beneficial effects:

[0036] The above-mentioned scheme has the following effects: on the one hand, the complex nonlinear feature extraction and modeling method is developed by fusing satellite multi-channel atmospheric top reflectivity observation, brightness temperature observation and atmospheric temperature profile reanalysis data, thereby improving the nonlinear mapping capability among the atmospheric top reflectivity, brightness temperature and cloud bottom height; on the other hand, the night and all-time inversion capability is enhanced by combining the all-time infrared channel features and layer information; and on the third hand, the generalization adaptability of the model is improved by the multi-source sample expansion and stacked learning strategy, and the problem of insufficient high-quality training data is effectively alleviated by designing the sample expansion and auxiliary training mechanism, so as to finally realize the high-precision inversion of the all-time cloud bottom height suitable for the domestic geostationary or polar orbit meteorological satellite. BRIEF DESCRIPTION OF DRAWINGS

[0037] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed in the embodiment description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0038] Figure 1 is a flowchart of the high-precision cloud bottom height inversion method for domestic passive meteorological satellites according to the present application;

[0039] Figure 2 is a flowchart of obtaining the atmospheric top brightness temperature data set in the high-precision cloud bottom height inversion method for domestic passive meteorological satellites according to the present application;

[0040] Figure 3is a flow chart of obtaining the inversion data set of the first geostationary meteorological satellite in the high-precision inversion method of the domestic passive meteorological satellite cloud bottom height embodiment of the present application;

[0041] Figure 4 is a flow chart of obtaining the all-time cloud bottom height in the high-precision inversion method of the domestic passive meteorological satellite cloud bottom height embodiment of the present application;

[0042] Figure 5 is a daytime cloud bottom height training result chart in the high-precision inversion method of the domestic passive meteorological satellite cloud bottom height embodiment of the present application, wherein a) is a daytime preliminary cloud bottom height training result chart obtained by an RF machine learning model, b) is a daytime preliminary cloud bottom height training result chart obtained by an XGBOOST machine learning model, c) is a daytime preliminary cloud bottom height training result chart obtained by a LightGBM machine learning model, and d) is a daytime cloud bottom height training result chart obtained by a stacking model;

[0043] Figure 6 is a daytime cloud bottom height verification result chart in the high-precision inversion method of the domestic passive meteorological satellite cloud bottom height embodiment of the present application, wherein a) is a daytime preliminary cloud bottom height verification result chart obtained by an RF machine learning model, b) is a daytime preliminary cloud bottom height verification result chart obtained by an XGBOOST machine learning model, c) is a daytime preliminary cloud bottom height verification result chart obtained by a LightGBM machine learning model, and d) is a daytime cloud bottom height verification result chart obtained by a stacking model;

[0044] Figure 7 is a nighttime cloud bottom height training result chart in the high-precision inversion method of the domestic passive meteorological satellite cloud bottom height embodiment of the present application, wherein a) is a nighttime preliminary cloud bottom height training result chart obtained by an RF machine learning model, b) is a nighttime preliminary cloud bottom height training result chart obtained by an XGBOOST machine learning model, c) is a nighttime preliminary cloud bottom height training result chart obtained by a LightGBM machine learning model, and d) is a nighttime cloud bottom height training result chart obtained by a stacking model;

[0045] Figure 8 is a nighttime cloud bottom height verification result chart in the high-precision inversion method of the domestic passive meteorological satellite cloud bottom height embodiment of the present application, wherein a) is a nighttime preliminary cloud bottom height verification result chart obtained by an RF machine learning model, b) is a nighttime preliminary cloud bottom height verification result chart obtained by an XGBOOST machine learning model, c) is a nighttime preliminary cloud bottom height verification result chart obtained by a LightGBM machine learning model, and d) is a nighttime cloud bottom height verification result chart obtained by a stacking model;

[0046] Figure 9This is a training result diagram of cloud base height under all weather conditions in an embodiment of the high-precision inversion method for cloud base height of domestically produced passive meteorological satellites of the present invention;

[0047] Figure 10 This is a diagram showing the all-day cloud base height verification results in an embodiment of the domestically developed passive meteorological satellite cloud base height high-precision inversion method of the present invention;

[0048] Figure 11 This is a system block diagram of an embodiment of the domestically developed passive meteorological satellite cloud base height high-precision inversion system of the present invention. Detailed Implementation

[0049] The technical solution of the present invention will now be described with reference to the accompanying drawings.

[0050] In embodiments of the present invention, words such as "exemplarily," "for example," etc., are used to indicate that something is an example, illustration, or description. Any embodiment or design described as "exemplary" in the present invention should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of the word "exemplary" is intended to present the concept in a concrete manner. Furthermore, in embodiments of the present invention, the meaning expressed by "and / or" can be both, or either one.

[0051] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.

[0052] like Figure 1 The flowchart shown is an embodiment of the high-precision cloud base height retrieval method for domestically produced passive meteorological satellites according to the present invention. The present invention provides a high-precision cloud base height retrieval method for domestically produced passive meteorological satellites, which is implemented by a high-precision cloud base height retrieval system for domestically produced passive meteorological satellites. The method includes:

[0053] S1. Based on the existing first geostationary meteorological satellite dataset and active remote sensing satellite cloud profile dataset, a multi-source database set is obtained through time matching and spatial matching;

[0054] Specifically, this multi-source database set includes: L1 level multi-channel data and solar zenith angle data, L2 level cloud cover data and cloud mask data, cloud base height, temperature profile and relative humidity profile data, total atmospheric water vapor content and elevation data.

[0055] Further, the first stationary meteorological satellite FY-4B dataset is used in the embodiment, and the CALIOP cloud profile data is available at the same time, which is from June 1, 2022 to June 30, 2023. The cloud mask data of FY-4B is used to identify cloud pixels. Through the time and space matching of the FY-4B dataset and CALIOP, and the spatial matching of DEM, a multi-source database set is constructed, which includes FY-4B L1 multi-channel data, L2 cloud amount, cloud type and cloud mask, CALOP cloud base height, ERA5 temperature profile and relative humidity profile, and elevation.

[0056] S2, according to the multi-source database set, the atmospheric layer top brightness temperature dataset is obtained by comparing with the second stationary meteorological satellite dataset and performing multi-channel spectral function response, the atmospheric layer top brightness temperature dataset includes the atmospheric layer top brightness temperature dataset of the first stationary meteorological satellite and the atmospheric layer top brightness temperature dataset of the second stationary meteorological satellite;

[0057] Specifically, as shown in the flow chart of obtaining the atmospheric layer top brightness temperature dataset in the domestic passive meteorological satellite cloud base height high-precision inversion method embodiment of the application, Figure 2 the S2 in the embodiment includes:

[0058] S21, according to the multi-source database set and the second stationary meteorological satellite dataset, the brightness temperature channel with large spectral response function difference is obtained by spectral response function comparison;

[0059] Further, the difference between the second stationary meteorological satellite FY-4A and FY-4B is compared. The results show that the spectral response function of FY-4A and FY-4B has small difference in 1-6 reflectivity channel, and the spectral response function of FY-4A 7~14 and FY-4B 7~15 brightness temperature channel has large difference.

[0060] S22, according to the brightness temperature channel with large spectral response function difference, the atmospheric layer top brightness temperature dataset is obtained by MODTRAN radiation transfer simulation.

[0061] Further, the 2284 clear sky profiles in the atmospheric profile are screened in the embodiment. First, the atmospheric layer top brightness temperature dataset of FY-4A 7~14 channel and FY-4B 7~15 channel is constructed by MODTRAN radiation transfer simulation, that is, the atmospheric layer top brightness temperature dataset.

[0062] S3, according to the atmospheric layer top brightness temperature dataset, the second stationary meteorological satellite dataset is corrected to obtain the inversion dataset of the first stationary meteorological satellite;

[0063] Specifically, such as Figure 3 The flowchart shown in this embodiment of the high-precision inversion method for cloud base height of domestically produced passive meteorological satellites of the present invention is used to obtain the inversion dataset of the first geostationary meteorological satellite. In step S3, the second geostationary meteorological satellite dataset is corrected based on the atmospheric top brightness temperature dataset to obtain the inversion dataset of the first geostationary meteorological satellite, including:

[0064] S31. Based on the brightness temperature dataset of the top of the atmosphere, establish a linear relationship model between the brightness temperature channels of the first geostationary meteorological satellite and the second geostationary meteorological satellite;

[0065] S32. Based on the linear relationship model between the brightness and temperature channels of the first geostationary meteorological satellite and the second geostationary meteorological satellite, the dataset of the second geostationary meteorological satellite is corrected to obtain the inversion dataset of the first geostationary meteorological satellite.

[0066] Furthermore, a linear relationship model between the brightness temperature channels of FY-4A and FY-4B was established, and the brightness temperature channels 7-14 of FY-4A were corrected to channels 7-15 of FY-4B. This enabled the previously constructed 2019-2020 FY-4A cloud base height inversion database to be converted into the dataset required for FY-4B cloud base height inversion, namely the inversion dataset of the first geostationary meteorological satellite.

[0067] S4. Merge the multi-source database set with the inversion dataset of the first geostationary meteorological satellite to obtain the all-day inversion dataset;

[0068] S5. Based on the all-day inversion dataset, obtain the all-day cloud base height using a stacked machine learning method.

[0069] Specifically, such as Figure 4 The flowchart shown in this embodiment of the high-precision cloud base height retrieval method for domestically produced passive meteorological satellites of the present invention illustrates the process of obtaining the all-day cloud base height. In step S5, based on the all-day inversion dataset, the all-day cloud base height is obtained through a stacked machine learning method, including:

[0070] S51. Based on the all-day inversion dataset, filter the daytime inversion data and the nighttime inversion data to obtain the daytime inversion dataset and the nighttime inversion dataset;

[0071] Furthermore, the method for filtering daytime and nighttime inversion data includes:

[0072] If the solar zenith angle in the all-day inversion dataset is less than 90 degrees, it is placed in the daytime inversion dataset; if the solar zenith angle in the all-day inversion dataset is not less than 90 degrees, it is placed in the nighttime inversion dataset.

[0073] S52, obtaining multiple daytime preliminary cloud base heights through different machine learning models according to the daytime inversion data set;

[0074] Further, the machine learning model comprises an RF machine learning model, an XGBOOST machine learning model and a LightGBM machine learning model, first, the RF machine learning model, the XGBOOST machine learning model and the LightGBM machine learning model are used to estimate daytime FY-4B cloud base height respectively. The 29 atmospheric temperature layers and relative humidity layers of ERA5 in the database, the surface features (ERA5 LST and CWV, DEM, FY-4B cloud cover, longitude and latitude), and the FY-4B satellite observation angle, 1-6 channel reflectivity and 7-15 brightness temperature are all used as feature parameters, and CALIOP CBH is used as a label (true value) to train the above-mentioned three machine learning models to obtain the estimated daytime FY-4B preliminary cloud base height; then, the daytime FY-4B preliminary cloud base height predicted by the three models is used to train a LightGBM stacked model to obtain the daytime cloud base height, and the training result and the verification result are as shown in Figure 5 The daytime cloud base height training result diagram in the domestic passive meteorological satellite cloud base height high-precision inversion method embodiment of the application is shown in the daytime cloud base height training result diagram obtained by the RF machine learning model, the daytime cloud base height training result diagram obtained by the XGBOOST machine learning model, the daytime cloud base height training result diagram obtained by the LightGBM machine learning model, and the daytime cloud base height training result diagram obtained by the stacked model, and the daytime cloud base height verification result diagram is as shown in Figure 6 The daytime cloud base height verification result diagram in the domestic passive meteorological satellite cloud base height high-precision inversion method embodiment of the application is shown in the daytime cloud base height verification result diagram obtained by the RF machine learning model, the daytime cloud base height verification result diagram obtained by the XGBOOST machine learning model, the daytime cloud base height verification result diagram obtained by the LightGBM machine learning model, and the daytime cloud base height verification result diagram obtained by the stacked model.

[0075] S53, obtaining multiple nighttime preliminary cloud base heights through different machine learning models according to the nighttime inversion data set;

[0076] Further, the machine learning model includes: an RF machine learning model, an XGBOOST machine learning model, and a LightGBM machine learning model. In the embodiment, the model input data is basically consistent with the daytime, except for the FY-4B 1~6 reflectivity bands. First, the RF machine learning model, the XGBOOST machine learning model, and the LightGBM machine learning model are used to estimate the preliminary cloud base height at night, and then the LightGBM training stacking model is used to estimate the final cloud base height at night. The training results and the verification results are as shown in FIGS. 6 and 7. Figure 7 FIG. 6 shows the training results of the cloud base height at night in the embodiment of the high-precision retrieval method for the cloud base height of the domestic passive meteorological satellite of the application, wherein a) is the training results of the preliminary cloud base height at night obtained by the RF machine learning model, b) is the training results of the preliminary cloud base height at night obtained by the XGBOOST machine learning model, c) is the training results of the preliminary cloud base height at night obtained by the LightGBM machine learning model, and d) is the training results of the cloud base height at night obtained by the stacking model, and FIG. 7 shows the verification results of the cloud base height at night in the embodiment of the high-precision retrieval method for the cloud base height of the domestic passive meteorological satellite of the application, wherein a) is the verification results of the preliminary cloud base height at night obtained by the RF machine learning model, b) is the verification results of the preliminary cloud base height at night obtained by the XGBOOST machine learning model, c) is the verification results of the preliminary cloud base height at night obtained by the LightGBM machine learning model, and d) is the verification results of the cloud base height at night obtained by the stacking model. Figure 8 FIG. 6 shows the training results of the cloud base height at night in the embodiment of the high-precision retrieval method for the cloud base height of the domestic passive meteorological satellite of the application, wherein a) is the training results of the preliminary cloud base height at night obtained by the RF machine learning model, b) is the training results of the preliminary cloud base height at night obtained by the XGBOOST machine learning model, c) is the training results of the preliminary cloud base height at night obtained by the LightGBM machine learning model, and d) is the training results of the cloud base height at night obtained by the stacking model, and FIG. 7 shows the verification results of the cloud base height at night in the embodiment of the high-precision retrieval method for the cloud base height of the domestic passive meteorological satellite of the application, wherein a) is the verification results of the preliminary cloud base height at night obtained by the RF machine learning model, b) is the verification results of the preliminary cloud base height at night obtained by the XGBOOST machine learning model, c) is the verification results of the preliminary cloud base height at night obtained by the LightGBM machine learning model, and d) is the verification results of the cloud base height at night obtained by the stacking model.

[0077] S54, obtaining the daytime cloud base height by the stacking model according to the plurality of preliminary cloud base heights at daytime.

[0078] S55, obtaining the cloud base height at night by the stacking model according to the plurality of preliminary cloud base heights at night.

[0079] S56, obtaining the all-day cloud base height according to the daytime cloud base height and the cloud base height at night.

[0080] Further, the training results and the verification results of the all-day cloud base height are as shown in FIGS. 8 and 9. Figure 9 FIG. 8 shows the training results of the all-day cloud base height in the embodiment of the high-precision retrieval method for the cloud base height of the domestic passive meteorological satellite of the application, and FIG. 9 shows the verification results of the all-day cloud base height in the embodiment of the high-precision retrieval method for the cloud base height of the domestic passive meteorological satellite of the application. Figure 10 FIG. 8 shows the training results of the all-day cloud base height in the embodiment of the high-precision retrieval method for the cloud base height of the domestic passive meteorological satellite of the application, and FIG. 9 shows the verification results of the all-day cloud base height in the embodiment of the high-precision retrieval method for the cloud base height of the domestic passive meteorological satellite of the application.

[0081] FIG. 8 shows the training results of the all-day cloud base height in the embodiment of the high-precision retrieval method for the cloud base height of the domestic passive meteorological satellite of the application, and FIG. 9 shows the verification results of the all-day cloud base height in the embodiment of the high-precision retrieval method for the cloud base height of the domestic passive meteorological satellite of the application. Figure 11The system block diagram of the shown embodiment of the domestic passive meteorological satellite cloud bottom height high-precision inversion system of the application, the application provides a domestic passive meteorological satellite cloud bottom height high-precision inversion system, which is applied to a domestic passive meteorological satellite cloud bottom height high-precision inversion method, and the system comprises a database set construction module, a data set comparison module, a data set correction module, an inversion data set construction module and a cloud bottom height inversion module, in particular,

[0082] The database set construction module is used for obtaining a multi-source database set through time matching and space matching according to an existing first stationary meteorological satellite data set and an active remote sensing satellite cloud profile data set;

[0083] The data set comparison module is used for obtaining an atmospheric top brightness temperature data set through comparison with a second stationary meteorological satellite data set and multi-channel spectral function response according to the multi-source database set;

[0084] The data set correction module is used for correcting the second stationary meteorological satellite data set according to the atmospheric top brightness temperature data set to obtain an inversion data set of the second stationary meteorological satellite;

[0085] The inversion data set construction module is used for merging the multi-source database set and the inversion data set of the second stationary meteorological satellite to obtain an all-time inversion data set;

[0086] The cloud bottom height inversion module is used for obtaining an all-time cloud bottom height through a stacked machine learning method according to the all-time inversion data set.

[0087] The application provides a domestic passive meteorological satellite cloud bottom height high-precision inversion method and system, the application fuses satellite multi-channel atmospheric top reflectivity observation, brightness temperature observation and atmospheric temperature profile reanalysis data, develops a complex nonlinear feature extraction and modeling method, improves the nonlinear mapping ability among the atmospheric top reflectivity, the brightness temperature and the cloud bottom height, combines the infrared channel features and the stratification information in the whole period, enhances the night and all-time inversion ability, improves the generalization adaptability of the model through the multi-source sample expansion and the stacked learning strategy, designs the sample expansion and the auxiliary training mechanism, effectively alleviates the problem of insufficient high-quality training data, and finally realizes the high-precision inversion of the all-time cloud bottom height suitable for the domestic stationary or polar orbit meteorological satellite.

[0088] It is to be understood that the present application is described by way of example only, and that modifications of detail can be made without departing from the scope of the application. Various modifications and changes can be made thereto by those skilled in the art which freely substitute various features and embodiments thereof, without departing from the spirit of the application. Further, it is to be understood that the application is not limited in its application to the details set forth in the description contained herein or to the Examples, and that the application is capable of other embodiments and of being practiced or being carried out in various ways. Accordingly, the application is to be realized in a manner that does not result in a practical application of the specific embodiments disclosed herein, and all such modifications of detail are intended to be included within the scope of the application.

Claims

1. A high-precision retrieval method for cloud base height of a homemade passive meteorological satellite, characterized in that, The method comprises: S1, obtaining a multi-source database set through time matching and space matching according to an existing first stationary meteorological satellite data set and an active remote sensing satellite cloud profile data set; S2, obtaining an atmospheric top brightness temperature data set through comparison with a second stationary meteorological satellite data set and multi-channel spectral function response according to the multi-source database set, wherein the atmospheric top brightness temperature data set comprises an atmospheric top brightness temperature data set of the first stationary meteorological satellite and an atmospheric top brightness temperature data set of the second stationary meteorological satellite; S3, correcting the second stationary meteorological satellite data set according to the atmospheric top brightness temperature data set to obtain an inversion data set of the first stationary meteorological satellite; S4, merging the multi-source database set and the inversion data set of the first stationary meteorological satellite to obtain an all-day inversion data set; S5, obtaining all-day cloud base height through a stacked machine learning method according to the all-day inversion data set. The S2 comprises: S21, obtaining a brightness temperature channel with large spectral response function difference through spectral response function comparison according to comparison of the multi-source database set and the second stationary meteorological satellite data set; S22, obtaining an atmospheric top brightness temperature data set through MODTRAN radiation transfer simulation according to the brightness temperature channel with large spectral response function difference; The S3 comprises: S31, establishing a linear relationship model between brightness temperature channels of the first stationary meteorological satellite and the second stationary meteorological satellite according to the atmospheric top brightness temperature data set; S32, correcting the second stationary meteorological satellite data set according to the linear relationship model between the brightness temperature channels of the first stationary meteorological satellite and the second stationary meteorological satellite to obtain an inversion data set of the first stationary meteorological satellite; The S5 comprises: S51, screening daytime inversion data and nighttime inversion data according to the all-day inversion data set to obtain a daytime inversion data set and a nighttime inversion data set; S52, obtaining multiple daytime preliminary cloud base heights through different machine learning models according to the daytime inversion data set; S53, obtaining multiple nighttime preliminary cloud base heights through different machine learning models according to the nighttime inversion data set; S54, obtaining daytime cloud base height through a stacking model according to the multiple daytime preliminary cloud base heights; S55, obtaining nighttime cloud base height through a stacking model according to the multiple nighttime preliminary cloud base heights; S56, obtaining all-day cloud base height according to the daytime cloud base height and the nighttime cloud base height.

2. The method according to claim 1, wherein, The multi-source database set comprises L1-level multi-channel data and solar zenith angle data, L2-level cloud amount data and cloud mask data, cloud base height, temperature profile and relative humidity profile data, atmospheric total water vapor content and elevation data.

3. The method according to claim 1, wherein, The machine learning model comprises an RF machine learning model, an XGBOOST machine learning model and a LightGBM machine learning model.

4. The method according to claim 1, wherein, The method for screening daytime inversion data and nighttime inversion data comprises: If the solar zenith angle in the all-day inversion data set is less than 90, the all-day inversion data set is put into the daytime inversion data set; if the solar zenith angle in the all-day inversion data set is not less than 90, the all-day inversion data set is put into the nighttime inversion data set.

5. A high-precision retrieval system for cloud base height of a domestic passive meteorological satellite, for implementing the high-precision retrieval method for cloud base height of a domestic passive meteorological satellite according to any one of claims 1-4, characterized in that, The system comprises: a database set construction module for obtaining a multi-source database set through time matching and space matching according to an existing first stationary meteorological satellite data set and an active remote sensing satellite cloud profile data set; a data set comparison module for obtaining an atmospheric top brightness temperature data set through comparison with a second stationary meteorological satellite data set and multi-channel spectral function response according to the multi-source database set, the atmospheric top brightness temperature data set comprising an atmospheric top brightness temperature data set of the first stationary meteorological satellite and an atmospheric top brightness temperature data set of the second stationary meteorological satellite; a data set correction module for correcting the second stationary meteorological satellite data set according to the atmospheric top brightness temperature data set to obtain an inversion data set of the first stationary meteorological satellite; an inversion data set construction module for merging the multi-source database set and the inversion data set of the first stationary meteorological satellite to obtain an all-day inversion data set; a cloud base height inversion module for obtaining an all-day cloud base height through a stacked machine learning method according to the all-day inversion data set.

Images