Channel dynamic cloud detection method and system based on meteorological satellite hyperspectral infrared detector

Abstract

The present invention discloses a channel dynamic cloud detection method and system based on a meteorological satellite hyperspectral infrared detector. First, the method proposes the optimal selection of nonlinear channels based on the probability approximation of the Gaussian mixture model to quantize the relative entropy, which can reduce the error distribution of temperature and humidity. Secondly, based on the layer-by-layer cloud simulation and clear sky brightness temperature simulation, a cloud profile inversion method of the weighted L1 norm constraint inverse problem of the signal degree of freedom optimization channel weight of Gaussian perturbation + numerical approximation is proposed, and cloud amount information of multiple levels from bottom to top is obtained to meet the actual situation of the vertical distribution of actual clouds. Finally, an adaptive dynamic cloud detection method is used to customize the cloud detection results for each channel at each field of view point, which is helpful to reasonably use the brightness temperature data of the hyperspectral infrared channel and realize the application of partial cloud area data, so as to increase the data volume of the brightness temperature of the variational assimilation channel, and thus improve the accuracy of numerical weather forecasting.

Description

Technical Field

[0001] The invention relates to a channel dynamic cloud detection method and system based on a meteorological satellite hyperspectral infrared detector, belonging to the technical field of numerical weather forecasting. Background Art

[0002] Meteorological satellite observations have become indispensable in the current numerical weather prediction (NWP) system. In meteorological satellite observations, hyperspectral infrared detectors are an important factor in improving forecast accuracy. However, this contribution comes largely from the assimilation of clear sky brightness temperature using hyperspectral infrared detectors. Due to the difficulty in correctly simulating infrared radiation under cloudy conditions, the use of meteorological satellite hyperspectral infrared detectors for data assimilation under cloudy conditions is very limited in terms of the amount of data used and applications. Considering that about 60% of the earth is covered by clouds at a given moment, the additional use of hyperspectral detector data in "cloudy areas" in data assimilation has the potential to improve the accuracy of numerical weather forecasts. At present, the application of hyperspectral infrared detector data is to assimilate the brightness temperature data of the "clear sky field point" channel or the brightness temperature data of the "clear sky channel" that is not affected by clouds. Therefore, it is necessary to study cloud detection methods to mine more available information from hyperspectral data.

[0003] Meteorological satellite hyperspectral infrared detectors are divided into polar-orbiting satellite hyperspectral infrared detectors and geostationary satellite hyperspectral infrared detectors. At present, the international meteorological polar-orbiting satellite hyperspectral infrared detectors mainly include: Atmospheric Infrared Sounder (AIRS), Infrared Atmospheric Sounding Interferometer (IASI), Cross-track Infrared Sounder (CrIS), and Fengyun-3D satellite hyperspectral infrared atmospheric sounder (HIRAS). AIRS, IASI, and CrIS have 2378, 8461, 1305, and 2287 channels, respectively. At present, the geostationary satellite hyperspectral infrared detectors in the world are only owned by or only in China, namely, the interferometric atmospheric vertical sounder (Geostationary Interferometric Infrared Sounder, GIIRS) carried by FengYun-4A (FY-4A) and FengYun-4B.

[0004] Cloud detection based on satellite hyperspectral infrared detector channel brightness temperature data has always been a hot and difficult issue in international research. At present, satellite hyperspectral infrared brightness temperature cloud detection is mainly divided into variants based on "clear sky view point" and "clear sky channel" and related methods. The principle of cloud detection based on "clear sky view point" is that if it is judged that this view point (also called observation point or pixel point) has clouds, then all channel brightness temperature data of this view point will be eliminated. The consequence of this "one size fits all" is that a lot of information available for the channel is lost. "Clear sky channel" cloud detection is based on the relationship between the brightness temperature deviation of different channels and the channels. The "channel sorting method" is used to determine whether the peak layer of the "equivalent weight function" of a channel at a certain view point is above the cloud top. If it is "above the cloud top", it is marked as a "clear sky channel" not affected by clouds, otherwise it is marked as "affected by clouds". The "clear sky channel" cloud detection method can retain some of the information lost by the "clear sky view point" cloud detection method.

[0005] Regarding "clear sky channel" cloud detection, the current mainstream method internationally is the "channel sorting method", but this method can only determine whether a certain channel is affected by clouds at a certain point in the field of view, but cannot obtain cloud cover information.

[0006] At present, the main international methods for “clear sky view point” cloud detection are: “minimum residual method” and “multivariate minimum residual method”;

[0007] (1) The “minimum residual method” can obtain the single-layer cloud parameter information of the field of view point, including cloud amount and cloud top pressure, under the premise of judging the existence of clouds at the field of view point. However, the “minimum residual method” cannot determine whether different channels are “really” affected by clouds at this field of view point, because the peak layer of the weight function of some channels is relatively high and is not affected by low-level clouds at this time, but the “one-size-fits-all” minimum residual method also eliminates the information of this channel at this field of view point. At the same time, the “minimum residual method” can only obtain single-layer cloud parameter information, while actual clouds are often multi-layered, and there are often low clouds under high clouds. This method does not conform to the actual vertical distribution of actual clouds.

[0008] (2) The “multivariate minimum residual method” can obtain the cloud amount profile (multi-layer cloud information) of the view point under the premise that there are clouds at the view point. However, the “multivariate minimum residual method” is essentially based on “least squares fitting”, and the results obtained have cloud amount values ​​for almost every vertical height layer or model pressure layer. However, actual clouds are often multi-layered, and there are often low clouds under high clouds, but it is impossible for clouds to exist at every height. This method does not quite conform to the actual vertical distribution of actual clouds.

[0009] As can be seen from the above, the "minimum residual method" obtains single-layer cloud information (cloud amount and cloud top pressure), while the "multivariate minimum residual method" obtains multi-layer cloud amount information; the "minimum residual method" obtains "two" information of cloud amount and cloud top pressure, while the "multivariate minimum residual method" obtains "one" information of cloud amount. They cannot meet the actual situation of the vertical distribution of actual clouds. Summary of the invention

[0010] The purpose of the present invention is to solve the problem that the cloud detection method in the prior art cannot meet the actual situation of the vertical distribution of actual clouds, and to provide a channel dynamic cloud detection method and system based on a meteorological satellite hyperspectral infrared detector.

[0011] In order to achieve the above object, the technical solution proposed by the present invention is as follows:

[0012] A channel dynamic cloud detection method based on a meteorological satellite hyperspectral infrared detector comprises the following steps:

[0013] Step 1, obtaining the observed brightness temperature of the meteorological satellite hyperspectral infrared detector and the background field data of the global data assimilation system that are matched in time and space, and preprocessing the observed brightness temperature, interpolating the background field data to the field of view point based on the latitude and longitude information of the field of view point of the observed brightness temperature, and obtaining the processed background field data;

[0014] Step 2: Based on the processed background field data, the brightness temperature of each channel and each field point of the hyperspectral infrared detector is simulated to obtain the simulated brightness temperature of all channels;

[0015] Step 3, using a nonlinear channel optimal selection method to select the channel of the meteorological satellite hyperspectral infrared detector to obtain the optimal channel;

[0016] Step 4, select the simulated brightness temperature and the observed brightness temperature corresponding to the preferred channel to perform field point cloud profile inversion to obtain cloud profile information that conforms to the actual characteristics of cloud cover;

[0017] Step 5: Based on the cloud profile information, the channel adaptive cloud detection index method is used to dynamically determine whether each preferred channel of the hyperspectral infrared detector is affected by clouds at a certain field of view point.

[0018] Furthermore, in step 1, the observed brightness temperature is preprocessed by apodization, and the background field data is interpolated to the field of view point using a bilinear interpolation method.

[0019] Furthermore, in step 3, a nonlinear channel optimal selection method is used to select the channel of the meteorological satellite hyperspectral infrared detector, and the preferred channel is specifically:

[0020] Step 3.1, calculate the relative entropy of the hyperspectral infrared detector channel:

[0021]

[0022] Where p(x) represents the probability distribution of meteorological variables of the background field in the field of view; p(x|y) represents the probability distribution of meteorological variables of the analysis field formed by correcting the meteorological variables of the background field after obtaining the channel brightness temperature of the field of view;

[0023] Step 3.2, calculate the mutual information of the channel;

[0024]

[0025] Where p(y) represents the probability distribution of brightness temperature observed by the channel;

[0026] Step 3.3, select the channel with the largest mutual information as the selected channel;

[0027] Step 3.4, updating the meteorological variables of the analysis field and the error covariance matrix of the meteorological variables of the analysis field according to the channel brightness temperature of the selected channel;

[0028] The expression of the error covariance matrix of the meteorological variables in the analysis field after the n+1th iteration is as follows:

[0029]

[0030] Where A represents the error covariance matrix of meteorological variables in the analysis field; A n+1 represents the error covariance matrix of meteorological variables in the analysis field after the n+1th iteration; B represents the error covariance matrix of meteorological variables in the background field; R represents the error covariance matrix of the selected channel observations; H represents the sensitivity of the selected satellite channel observations or simulated brightness temperature to the inverted atmospheric temperature or humidity, which is called the Jacobi matrix; H n represents the Jacobi matrix after the nth iteration;

[0031] Step 3.5, repeat steps 3.1 to 3.4 for iterative optimization until the required number of selected channels are selected as the preferred channels.

[0032] Furthermore, in step 3, p(x), p(y) and p(x|y) are obtained by a method of numerically approximating probability distribution based on nonlinear resampling of a Gaussian mixture model.

[0033] Furthermore, the fast radiation transfer mode is used at each field of view point to simulate the brightness temperature of each channel of the hyperspectral infrared detector. Specifically, the simulated radiation value is first calculated, and then the radiation value is converted into brightness temperature through the Planck function. The expression of the simulated radiation value is:

[0034]

[0035] In the formula, R vrepresents the channel simulation radiation value with frequency v; N is the number of radiation transmission mode layers; c k is the cloud cover of the kth layer of the model; represents the cloud radiation simulation when there are clouds in the kth layer of the model; c 0 is the percentage of clear sky radiation; Represents clear-sky radiation.

[0036] Furthermore, the brightness temperature of each field of view point in each channel of the hyperspectral infrared detector is simulated by the field of view point-by-field of view channel simulation method. The brightness temperature simulation is divided into clear sky simulation and layer-by-layer cloud simulation: the clear sky simulation is to simulate the clear sky brightness temperature of the channel when the field of view point is assumed to be a clear sky field of view point; the layer-by-layer cloud simulation is to simulate the cloud brightness temperature of the channel when it is assumed that there is a cloud at the field of view point. The cloud simulation here is the cloud simulation in different mode layers.

[0037] Furthermore, in step 4, the target minimization function of the field of view point cloud volume profile c is:

[0038]

[0039] Where W is the weight of the channel brightness temperature to the minimization function; (represents the contribution rate of different channel brightness temperatures to the target minimization function); λ is the regularization parameter; BT sim is the simulated brightness temperature of the preferred channel; BT obs is the observed brightness temperature of the preferred channel; BT sim The simulated radiation value in step 4 is calculated by the Planck function to obtain:

[0040]

[0041] Where n c and N represent the number of satellite channels and the number of mode layers, respectively; Indicates the nth c The clear sky simulated brightness temperature of each channel; Indicates the nth c The observed brightness temperature of each channel; Indicates the nth c The cloud-simulated brightness temperature of the channel when there are clouds in the Nth layer of the model (cloud-simulated brightness temperature);

[0042] The constraint condition of the target minimization function of the field of view point cloud volume profile c is:

[0043]

[0044] In the formula, c i represents the cloud cover of the i-th model layer; N represents the number of model layers.

[0045] Furthermore, the weight W of the channel brightness temperature to the minimization function is calculated by numerically approximating the signal degree of freedom DFS. The specific process is as follows:

[0046] ① Solve the following minimization function and obtain two optimal solutions c and c respectively * ;

[0047]

[0048] In the formula, For BT obs The Gaussian perturbation of ξ 0 represents a random Gaussian variable;

[0049] ② Calculate W by numerically approximating the signal degree of freedom DFS;

[0050]

[0051] In the formula, R error represents the error covariance matrix.

[0052] Furthermore, in step 5, a channel-adaptive cloud detection index method is used to dynamically determine whether each preferred channel of the hyperspectral infrared detector is affected by clouds at a certain field of view point, specifically:

[0053] If ∑(c i *BT sim,i,j )-BT sim,clr,i,j |<0.01*BT sim,clr,i,j

[0054] Then, it means that channel j is not affected by the cloud at the field of view point i, marked as label i,j =0, the observation data of this channel at this field of view point can be used;

[0055] If ∑(c i *BT sim,i,j )-BT sim,clr,i,j |≥0.01*BT sim,clr,i,j

[0056] Then, it means that channel j is affected by the cloud at the field of view point i, marked as label i,j =1, the observation data of this channel at this field of view point will be eliminated;

[0057] c i =(c i,1 ,c i,2 ,...,c i,N ,c i,0 ) T Indicates the cloud profile at the viewing point i; BT sim,i,j =[BTsim,i,j,k ,BT sim,clr,i,j ] k＝1,...N , [BT sim,i,j,k ] k＝1,...N Indicates the cloud-simulated brightness temperature of the jth channel at the i-th field of view for the 1st to Nth layers of the model; N represents the number of radiation transfer model layers; BT sim,clr,i,j Represents the clear sky simulated brightness temperature of the jth channel at the i-th field of view point.

[0058] A channel dynamic cloud detection system based on a meteorological satellite hyperspectral infrared detector includes the following parts:

[0059] The data acquisition and preprocessing module acquires the observed brightness temperature of the meteorological satellite hyperspectral infrared detector and the background field data of the global data assimilation system with time and space matching, and preprocesses the observed brightness temperature. Based on the latitude and longitude information of the field of view point of the observed brightness temperature, the background field data is interpolated to the field of view point to obtain the processed background field data;

[0060] The simulation data construction module simulates the brightness temperature of each channel and each field point of the hyperspectral infrared detector based on the processed background field data to obtain the simulated brightness temperature of all channels;

[0061] The nonlinear channel optimal selection module uses the nonlinear channel optimal selection method to select the channel of the meteorological satellite hyperspectral infrared detector to obtain the optimal channel;

[0062] Inversion of the field point cloud profile: Select the simulated brightness temperature and observed brightness temperature corresponding to the preferred channel to invert the field point cloud profile, and obtain the cloud profile information that conforms to the actual characteristics of the cloud;

[0063] The channel-adaptive dynamic cloud detection module uses the channel-adaptive cloud detection index method based on cloud profile information to dynamically determine whether each preferred channel of the hyperspectral infrared detector is affected by clouds at a certain field of view point.

[0064] Compared with the prior art, the present invention has the following beneficial effects:

[0065] The present invention proposes a nonlinear channel optimal selection method based on the background field and the iteratively obtained analysis field prior information, which can reduce the inversion instability caused by the redundant information observed by the hyperspectral detector to solve the channel "dimensionality disaster" problem. Compared with the internationally used "entropy reduction method-Shannon entropy", the channel optimal selection method of the present invention can reduce the error distribution of temperature and humidity, especially in the middle pressure layer of the model.

[0066] When the present invention adopts the "relative entropy" nonlinear channel optimal selection method based on the background field and the iteratively obtained analysis field prior information, due to the nonlinear mapping between the mode state space and the observation space, when the nonlinear observation operator is considered, it is impossible to give an analytical expression of the mutual information (Mutual information, referred to as MI). To address this problem, the present invention proposes a method for approximating probability distribution based on nonlinear resampling of a Gaussian Mixed Model (Gaussian Mixed Model, referred to as GMM), and gives an analytical expression of the mutual information.

[0067] The cloud detection method of the present invention is different from the minimum residual method. It provides multiple levels of cloud information from bottom to top through layer-by-layer cloud simulation, which can form three-dimensional cloud information and meet the actual situation that the cloud is multi-layered, there are low clouds under the high clouds, but not at every height. Different from the multivariate minimum residual method, the present invention combines the actual distribution of clouds with the method of cloud profile inversion, introduces the L1 norm, and proposes a cloud profile inversion method for the weighted L1 norm constrained inverse problem. This method is closer to the actual "vertical distribution of clouds". The L1 norm is a relatively effective method for solving sparse solutions. It can retain the maximum amount of original information with a small number of atoms, so the "features" are more in line with the actual "cloud". The method proposed by the present invention solves the multi-layer cloud amount distribution of clouds, while avoiding the problem of cloud amount values ​​at each height layer, and can build a model that conforms to the actual cloud distribution characteristics.

[0068] In the inversion of field point cloud profile based on weighted L1 norm constraint, the present invention proposes a method of optimizing the weights of the minimization function of the cloud profile inversion model of different channels in the "optimal channel combination" based on the degree of freedom for signal (DFS), and adopts the method of "Gaussian perturbation" + "numerical approximation" to quantify the signal degree of freedom value of the brightness temperature of each channel as the weight.

[0069] The adaptive dynamic cloud detection method proposed in the present invention can customize the cloud detection results for each channel at each field of view point, which is helpful to rationally use the brightness temperature data of the hyperspectral infrared channel and realize the application similar to that of the partial cloud area data, so as to increase the amount of data information of the brightness temperature of the numerical forecast assimilation channel, thereby improving the accuracy of numerical weather forecast and contributing to the meteorological disaster prevention and mitigation cause.

[0070] The cloud amount information obtained by the present invention can be used not only for numerical weather forecasting, but also for fields such as solar energy new energy power generation and forest fire monitoring. BRIEF DESCRIPTION OF THE DRAWINGS

[0071] Figure 1 This is a schematic diagram of a process of Embodiment 1 of the present invention;

[0072] Figure 2 This is a principle block diagram of a dynamic cloud detection method according to Embodiment 2 of the present invention;

[0073] Figure 3 The cloud map, brightness temperature distribution map and cloud amount profile distribution map detected in the area affected by Typhoon Maria; (a) is the cloud map of the multi-channel scanning imaging radiometer of Fengyun-4A; (b) is the brightness temperature distribution map of different channels of the interferometric atmospheric vertical sounder of Fengyun-4A in clear sky and cloudy sky; (c) and (d) are the cloud amount profile distribution of the clear sky field of view point FOV-1 and the cloudy field of view point FOV-2 respectively obtained by the method of the present invention;

[0074] Figure 4 The figures are effect diagrams of cloud detection in the high-frequency observation area of ​​typhoon Maria; (a), (b), (c) and (d) are respectively the dynamic cloud detection diagrams of the channels based on the present invention for different channels; (e) is the effect diagram of the cloud amount product of the same type. DETAILED DESCRIPTION

[0075] The present invention is described in detail below with reference to the accompanying drawings and specific embodiments.

[0076] Embodiment 1:

[0077] like Figure 1 As shown, the channel dynamic cloud detection method based on the meteorological satellite hyperspectral infrared detector of the present invention comprises the following steps:

[0078] Step 1, obtaining the observed brightness temperature of the meteorological satellite hyperspectral infrared detector and the background field data of the global data assimilation system that are matched in time and space, and preprocessing the observed brightness temperature, interpolating the background field data to the field of view point based on the latitude and longitude information of the field of view point of the observed brightness temperature, and obtaining the processed background field data;

[0079] Step 2: Based on the processed background field data, the brightness temperature of each channel and each field point of the hyperspectral infrared detector is simulated to obtain the simulated brightness temperature of all channels;

[0080] Step 3, using a nonlinear channel optimal selection method to select the channel of the meteorological satellite hyperspectral infrared detector to obtain the optimal channel;

[0081] Step 4, select the simulated brightness temperature and the observed brightness temperature corresponding to the preferred channel to perform field point cloud profile inversion to obtain cloud profile information that conforms to the actual characteristics of cloud cover;

[0082] Step 5: Based on the cloud profile information, the channel adaptive cloud detection index method is used to dynamically determine whether each preferred channel of the hyperspectral infrared detector is affected by clouds at a certain field of view point.

[0083] This embodiment may also include:

[0084] Step 6, accuracy assessment and verification, compare the cloud detection results of the present invention with the cloud image products of FY-4A Advanced Geosynchronous Radiation Imager (AGRI) and the cloud film products of FY-4A / AGRI cloud detection. Furthermore, it can also be compared with the cloud cover in ERA5.

[0085] The present invention discloses a channel dynamic cloud detection method and system based on a meteorological satellite hyperspectral infrared detector. First, the method proposes the optimal selection of nonlinear channels based on the probability approximation of the Gaussian mixture model to quantize the relative entropy, which can reduce the error distribution of temperature and humidity. Secondly, based on the layer-by-layer cloud simulation and clear sky brightness temperature simulation, a cloud profile inversion method of the weighted L1 norm constraint inverse problem of the signal degree of freedom optimization channel weight of Gaussian perturbation + numerical approximation is proposed, and cloud amount information of multiple levels from bottom to top is obtained to meet the actual situation of the vertical distribution of actual clouds. Finally, an adaptive dynamic cloud detection method is used to customize the cloud detection results for each channel at each field of view point, which is helpful to reasonably use the brightness temperature data of the hyperspectral infrared channel and realize the application of partial cloud area data, so as to increase the data volume of the brightness temperature of the variational assimilation channel, and thus improve the accuracy of numerical weather forecasting.

[0086] Embodiment 2:

[0087] like Figure 2 As shown, the channel dynamic cloud detection method based on satellite hyperspectral infrared detector of this embodiment includes 6 steps, namely:

[0088] Step 1: Data acquisition and preprocessing.

[0089] ① Data acquisition. Obtain the observed brightness temperature (data / data) of the satellite hyperspectral infrared detector that matches time and space and the background field data of the global data assimilation system; in this embodiment, the observation data of the satellite hyperspectral infrared detector adopts the FY-4A / GIIRS hyperspectral channel to observe the brightness temperature, and the FY-4A / GIIRS data comes from the official website of the National Satellite Meteorological Center and can be obtained by online registration. The background field data uses the Final Global Data Assimilation System (FNL) data of the National Centers for Environmental Prediction (NCEP) of the United States, and the data is obtained by registration on the relevant official website.

[0090] ② Data preprocessing. Before using the FY-4A / GIIRS observation brightness temperature data, the FY-4A / GIIRS observation brightness temperature data was preprocessed by using the hamming apodization function of the three-point filter (0.23, 0.54 and 0.23) of the FY-4A / GIIRS connected channel. And based on the longitude and latitude information of the field of view of FY-4A / GIIRS, the NCEP / FNL data was interpolated to the FY-4A / GIIRS field of view point using the bilinear interpolation method to obtain the processed background field data.

[0091] Step 2, simulation data construction. In the cloud profile inversion method constructed by the present invention, data needs to be input, and the FY-4A / GIIRS infrared brightness temperature data obtained by simulation based on the fast radiative transfer mode is used as one of the data sources of the cloud profile inversion method of the present invention. The simulated brightness temperature involves background field data. The background field data of this embodiment adopts the processed background field NCEP / FNL data obtained in step 1. The fast radiative transfer mode of this embodiment adopts the radiative transfer mode of Tiros operational vertical sounder (Radiative Transfer for Tirosn Operational Vertical Sounder, referred to as RTTOV). When performing channel brightness temperature simulation, in order to simplify the optimization algorithm for field of view point cloud profile inversion, this embodiment proposes a "layer-by-layer cloud simulation method", that is, assuming that each model layer has "clouds" respectively, the brightness temperature of each channel of each field of view point under the "cloud" condition of each model layer is calculated (simulated).

[0092] Step 3, optimal selection of nonlinear channels. In addition to the simulated brightness temperature, the data required in the present invention also requires channel observation brightness temperature data, and channel selection is performed to construct the optimal channel brightness temperature combination. The significance of hyperspectral channel selection is that the optimal channel selection, as a key technology for the inversion of hyperspectral atmospheric vertical detectors, can reduce the inversion malfunction caused by the redundant information observed by hyperspectral detectors to solve the channel "dimensionality curse" problem. Different from the international mainstream "entropy subtraction-Shannon entropy", this embodiment proposes a nonlinear channel optimal selection method based on Gaussian mixture resampling of the "relative entropy" of the background field and the iteratively obtained analysis field prior information. When specifically implementing "relative entropy", the present invention proposes a probabilistic nonlinear resampling method based on the Gaussian Mixed Model (GMM). A set of optimal "channel subsets" is obtained through channel optimal selection. The channel observation brightness temperature selected in the "channel subset" is used as one of the data sources of the cloud profile inversion method in step 4 of this embodiment.

[0093] Step 4, inversion of the cloud volume profile of the field of view point. This embodiment obtains the cloud volume profile of the field of view point by constructing a weighted L1 norm constraint based on the FY-4A / GIIRS observed brightness temperature and simulated brightness temperature (clear sky simulation and layer-by-layer cloud simulation). The advantage of the L1 norm is that it has been proven to be a relatively effective method for solving sparse solutions, and a small number of atoms can be used to retain the maximum amount of original information, so the "feature" is more in line with the actual vertical distribution of the "cloud". Different from the "least squares fitting" (L2 norm) of the "multivariate minimum residual method", this embodiment proposes to invert the cloud volume profile of the field of view point based on the weighted L1 norm constraint. With regard to the weight W of the cloud volume profile inversion model of different channels in the "optimal channel combination" of FY-4A / GIIRS, the present invention proposes to optimize W based on a method of numerical approximation of the degrees of freedom for signal (DFS for short).

[0094] Step 5, channel adaptive dynamic cloud detection. Different from the minimum residual method and other methods for judging whether there are clouds at a viewing point, this embodiment judges whether a channel is truly "affected by clouds" at a viewing point. Based on step 4, the present invention proposes a channel adaptive cloud detection index method to dynamically judge whether a channel is truly "affected by clouds" at a viewing point. If a channel is not affected by clouds at a viewing point, the observation data of this channel at this viewing point can be used, and the information of this channel at this viewing point is retained to form a subset of "effective information" of the viewing point for later data assimilation and numerical weather forecasting. If a channel is affected by clouds at a viewing point, the observation data of this channel at this viewing point is discarded.

[0095] Step 6: Accuracy assessment and verification. This example compares the cloud detection results with the FY-4A Advanced Geosynchronous Radiation Imager (AGRI) cloud image product and the FY-4A / AGRI cloud detection cloud mask (CLM) product. Furthermore, it is compared with the cloud cover in ERA5.

[0096] Embodiment three:

[0097] The further design of this embodiment is that step 3 adopts a nonlinear channel optimal selection method to select the channel of the meteorological satellite hyperspectral infrared detector, and the specific process is as follows:

[0098] 3.1) When performing optimal channel selection, a variational inversion and prior information update method is first constructed:

[0099] The variational inversion objective functional is defined as follows:

[0100]

[0101] In the formula, x represents the control variable; b represents the background field (including temperature, humidity, wind field, surface air temperature and other meteorological variables); B and R represent the background and observation error covariance matrices respectively; the superscripts -1 and T represent the matrix inversion and transposition marks respectively; y o Represents the observation value, which here represents the observed brightness temperature of the satellite hyperspectral infrared detector; H represents the observation operator, which is also called the (fast) radiation transfer mode for satellite data, and realizes the projection mapping of the mode space variables (such as different meteorological variables such as temperature) to the observation space (such as the brightness temperature observed by the satellite hyperspectral infrared detector).

[0102] That is, in the above formula, y o represents the channel observed brightness temperature; H(x) represents the channel simulated brightness temperature; x b represents the background field; minimize the iterative J(x) objective function, and the final solution x is the analytical field x a .

[0103] Perform iterative calculation of J(x). The estimated value of the error covariance matrix of the meteorological variables in the analysis field after the n+1th iteration in the iterative process is as follows:

[0104]

[0105] Where A represents the error covariance matrix of meteorological variables in the analysis field. n+1 represents the error covariance matrix of meteorological variables in the analysis field after the n+1th iteration; B represents the error covariance matrix of meteorological variables in the background field; R represents the error covariance matrix of the selected channel observations; H represents the sensitivity of the selected satellite channel observations or simulated brightness temperature to the inverted atmospheric temperature or humidity, which is called the Jacobi matrix; H n represents the Jacobi matrix after the nth iteration;

[0106] 3.2) The optimal selection method of nonlinear channels of meteorological satellite hyperspectral infrared detectors, the specific steps are as follows:

[0107] ① Establish a channel blacklist. Remove the channels in the blacklist and keep the remaining channels;

[0108] ② Calculate the relative entropy of the remaining channels based on the Gaussian mixture model nonlinear resampling numerical approximation probability distribution method;

[0109] Relative entropy (RE) is defined as:

[0110]

[0111] When calculating the relative entropy estimates of the probability distribution of p(x) and p(x|y), due to the nonlinear mapping between the mode state space and the observation space, when considering the nonlinear observation operator, it is impossible to give an analytical expression for the mutual information (MI). This embodiment proposes to numerically approximate the probability distribution of p(x) and p(x|y) based on nonlinear resampling of the Gaussian Mixed Model (GMM).

[0112] p(x) represents the probability distribution of different meteorological variables (e.g., temperature) in the background field at the view point in the studied area.

[0113] p(x|y) represents the probability distribution of different meteorological variables in the analysis field after obtaining the channel observation data of the field of view points in the study area (such as the channel brightness temperature of the satellite), that is, the probability distribution of the meteorological variables of the original background field after obtaining the channel observation data.

[0114] p(y) represents the probability distribution of brightness temperature observed in the channel.

[0115] p(y|x) represents the probability distribution of the channel simulated brightness temperature after obtaining the background field data of the field of view points in the studied area.

[0116] For p(x), the Gaussian mixture model is given by:

[0117]

[0118] Where G represents the number of Gaussian components. k Indicates the proportion of each Gaussian component. k Represents the mean of the Gaussian components. ∑ k Represents the covariance of the Gaussian component. Such parameters are optimized by the expectation-maximization (EM) method. After completing the probabilistic nonlinear resampling, it is necessary to iterate the EM estimation parameters.

[0119] Assume that p(y|x) and p(x) have means μ y and μ x , where the variances are Gaussian distributions of R and B respectively.

[0120] Specifically, when the nonlinear estimation method is used to perform the maximum mutual information (MI) iteration of the resampled distribution to select channels, the prior distribution does not need to be assumed to be Gaussian after the first iteration, so the distribution has "weak Gaussianity".

[0121] To represent p(y) and p(x|y), first generate M samples of p(y|x) and N samples of p(x), which are defined as follows:

[0122] x i :N(μ x ,B) i=1,2,...,N (5)

[0123] y j :N(μ y ,R) j=1,2,...,M (6)

[0124] Where B and R represent the background and channel observation error covariance matrices, respectively. Here N(·,·) represents the form of probability distribution.

[0125] Furthermore, the prior distribution can be expressed as the sum of delta functions, then:

[0126]

[0127] According to Bayesian theory, the jth sample of p(y|x) can be expressed as a weighted sum of delta functions, which means:

[0128]

[0129] w i,j The definition is as follows:

[0130] w i,j =p(y j |x i ) / Np(y j ) (9)

[0131] p(y j )=∫p(x,y j )dx (10)

[0132] According to Bayes' theorem,

[0133]

[0134] After derivation,

[0135] The relative entropy of the jth sample of p(y|x) is defined as follows:

[0136]

[0137] ③ Calculate the mutual information of relative entropy based on the background field and the iteratively obtained analysis field prior information;

[0138] Mutual information is a measure of the entropy of the observed data. It is given in the form of prior and posterior distributions:

[0139]

[0140] Mutual information MI can be estimated by "linear" and "nonlinear" methods. If the observed data has a large impact on the analysis results, the posterior distribution will change significantly compared to the prior distribution. In this case, the mutual information MI can be interpreted as the relative entropy and the weighted sum of all possible observation probabilities. The relative entropy can be expressed as

[0141] ④ Select the channel with the largest mutual information as the selected channel for this "channel subset";

[0142] ⑤ Update the above-mentioned prior information according to the brightness temperature information of the selected channel, that is, formula (2);

[0143] ⑥ Repeat steps ②-⑤ until the required number of channels are selected.

[0144] Embodiment 4:

[0145] A further design of this embodiment is that, when performing channel brightness temperature simulation, this embodiment adopts a field of view point-by-view channel-by-channel simulation method.

[0146] The brightness temperature simulation in this example is divided into two cases: ① Clear sky simulation. Assuming that the field of view point is a clear sky field of view point, the FY-4A / GIIRS clear sky brightness temperature is simulated; ② Cloud simulation. Assuming that there are clouds at the field of view point, the FY-4A / GIIRS cloud brightness temperature is simulated. Here, the simulation is of clouds in different model layers.

[0147] Calculate the brightness temperature simulation under "clear sky conditions" for each channel at each field of view point. When simulating the "brightness temperature with clouds" of the channel, in order to simplify the optimization algorithm for inverting the cloud volume profile of the field of view point, the present invention proposes a "layer-by-layer cloud simulation method", that is, assuming that each model layer has "clouds", calculate the brightness temperature simulation of clouds for each channel at each field of view point under the "cloudy" condition of each model layer.

[0148] In this embodiment, each field of view point is regarded as a multi-layer cloud profile. According to the radiation transfer theory, the satellite channel simulated radiation value R of frequency v is calculated by the radiation transfer model. v , the channel radiation value can be converted into channel brightness temperature through the Planck function. In a finite-dimensional space, the satellite channel simulated radiation value of frequency v can be expressed as the following relationship:

[0149]

[0150] In the formula, R v It represents the simulated radiation value of the channel with frequency v; N is the number of radiation transmission mode layers (i.e., the number of RTTOV layers). k is the cloud cover of the kth layer of the model. It represents the cloud brightness temperature simulation when there are clouds in the kth layer of the model.0 is the percentage of clear sky radiation. Represents clear-sky radiation.

[0151] c 0 It represents the percentage of clear sky radiation and is defined as follows:

[0152]

[0153] It should be noted that it is assumed here that the sum of the total cloud amounts of the cloud profiles is 1. This is used as one of the constraints of the field of view point cloud profile inversion method in step 4.

[0154] In step 4, the field of view point cloud profile is inverted based on the relationship between the hyperspectral infrared detector observation and the simulated brightness temperature.

[0155] Formula (14) Noted in matrix and vector form, it is represented as follows:

[0156] R v =A·R (16)

[0157] Where A represents the proportion of cloud cover and clear sky radiation in different layers of the model, and R represents the cloudy radiation value and clear sky radiation value simulated in different model layers.

[0158] The above calculated radiation values ​​are converted into brightness temperature values ​​by Planck function. Assume that the simulated brightness temperature (Simulatedbrightnesstemperature, marked as BT) of the optimal combination subset of FY-4A / GIIRS channels (obtained by the nonlinear channel optimal selection method) at each field of view point is sim ) and observed brightness temperature (marked as BT obs ) are recorded as:

[0159]

[0160] Where n c and N represent the number of satellite channels (the preferred channel is obtained by optimal selection of nonlinear channels) and the number of model layers, respectively. Indicates the nth c The clear sky simulated brightness temperature of each channel; Indicates the nth c The observed brightness temperature of each channel; Indicates the nth c The cloud-simulated brightness temperature of each channel when there are clouds in the Nth layer of the model layer (cloud-simulated brightness temperature).

[0161] Assume that the cloud amount profile is c = (c 1 ,c 2 ,...,cN ,c 0 ) T , e represents the brightness temperature simulation error, then:

[0162] BT obs =c·BT sim +e (17)

[0163] It can be seen that the cloud profile c can be solved by the relationship between the brightness temperature observed by the hyperspectral infrared detector and the simulated brightness temperature.

[0164] This embodiment proposes to invert the field of view point cloud volume profile c based on the weighted L1 norm constraint, and the target minimization function and constraint conditions are defined as follows:

[0165]

[0166] Constraints:

[0167]

[0168] Wherein, W represents the weight of the channel brightness temperature, that is, the contribution rate of different channel brightness temperatures to the target minimization function. λ is the regularization parameter, which is given as λ=0.1 in the present invention. The regularization parameter can be solved by curve optimization in the actual operation process. The minimization function is solved by constrained quadratic programming.

[0169] In this embodiment, the cloud cover is limited to between [0, 1] by adding the L1 norm, and the accumulated value of the cloud cover profile is 1 as the constraint condition of the objective function, instead of calculating the cloud cover of each layer and then dividing it by the total value and then returning the sum to 1. Therefore, the method of this embodiment is more reasonable.

[0170] Since the L1 norm is not differentiable at the origin, equation (18) is a non-smooth convex problem. The advantage of using the L1 norm is that it has been proven to be a more effective method for solving sparse solutions. It can retain the maximum amount of original information with a small number of atoms. Therefore, this "feature" is more in line with the actual vertical distribution of the "cloud".

[0171] In this embodiment, the degree of freedom for signal (DFS) is used to calculate the weight W of the brightness temperature of different channels on the cloud profile inversion objective function. The calculation formula of DFS is:

[0172] DFS=tr(KH) (20)

[0173] Where tr(·) represents the trace of the matrix. K represents the gain matrix in variational assimilation. H is the Jacobi matrix, which represents the sensitivity of the selected satellite channel observations or simulated brightness temperatures to the inverted atmospheric temperature and humidity, etc. The trace of the matrix KH can be regarded as the sensitivity of the analysis field to the observation data.

[0174] When calculating DFS, this embodiment proposes to use the method of "Gaussian perturbation" plus "numerical approximation". More specifically, consider two variables BT obs and The latter variable is a Gaussian perturbation of the former variable, satisfying ξ 0 represents a random Gaussian variable.

[0175] Based on these two variables, W in the present invention is reduced to solving the minimization problem of equations (21) and (22).

[0176]

[0177] We obtain two optimal solutions c and c respectively. * The DFS (i.e., W value) in the present invention is obtained by the following numerical approximation:

[0178]

[0179] In the formula, R error represents the error covariance matrix, which is obtained by statistics of a large number of samples in formula (21).

[0180] In step 5 of this embodiment, based on the inversion of the field of view point cloud profile, it is determined whether the meteorological satellite hyperspectral infrared detector channel j is affected by clouds at the field of view point i.

[0181] Assume that the cloud profile at the viewing point i is c i =(c i,1 ,c i,2 ,...,c i,N ,c i,0 ) T , the simulated brightness temperature of the jth channel at the i-th field of view is expressed as BT sim,i,j =[BT sim,i,j,k ,BT sim,clr,i,j ] k＝1,...N . [BT sim,i,j,k ] k＝1,...N Indicates the cloud-simulated brightness temperature of the jth channel at the i-th field of view for the 1st to Nth layers of the model; N represents the number of radiation transfer model layers; BT sim,clr,i,j Represents the clear sky simulated brightness temperature of the jth channel at the i-th field of view point.

[0182] To determine whether a meteorological satellite hyperspectral infrared detector channel j is affected by clouds at a field of view point i, the channel adaptive cloud detection index method of this embodiment is defined as the following threshold determination method.

[0183] ①If

[0184] |∑(c i *BTsim,i,j )-BT sim,clr,i,j |<0.01*BT sim,clr,i,j (twenty four)

[0185] Then, it means that channel j is not affected by the cloud at the field of view point i, marked as label i,j =0, the observation data of this channel at this field of view point can be used.

[0186] ②If

[0187] |Σ(c i *BT sim,i,j )-BT sim,clr,i,j |≥0.01*BT sim,clr,i,j (25)

[0188] Then, it means that channel j is affected by the cloud at the field of view point i, marked as label i,j =1, the observation data of this channel at this field of view point will be eliminated.

[0189] Embodiment five:

[0190] This example provides a channel-adaptive dynamic cloud detection system based on a meteorological satellite hyperspectral detector. The system includes the following parts:

[0191] The data acquisition and preprocessing module acquires the observed brightness temperature of the meteorological satellite hyperspectral infrared detector and the background field data of the global data assimilation system with time and space matching, and preprocesses the observed brightness temperature. Based on the latitude and longitude information of the field of view point of the observed brightness temperature, the background field data is interpolated to the field of view point to obtain the processed background field data;

[0192] The simulation data construction module simulates the brightness temperature of each channel and each field point of the hyperspectral infrared detector based on the processed background field data to obtain the simulated brightness temperature of all channels;

[0193] The nonlinear channel optimal selection module uses the nonlinear channel optimal selection method to select the channel of the meteorological satellite hyperspectral infrared detector to obtain the optimal channel;

[0194] Inversion of the field point cloud profile: Select the simulated brightness temperature and observed brightness temperature corresponding to the preferred channel to invert the field point cloud profile, and obtain the cloud profile information that conforms to the actual characteristics of the cloud;

[0195] The channel-adaptive dynamic cloud detection module uses the channel-adaptive cloud detection index method based on cloud profile information to dynamically determine whether each preferred channel of the hyperspectral infrared detector is affected by clouds at a certain field of view point.

[0196] This example may also include:

[0197] The accuracy assessment and verification module is used to compare the obtained cloud detection results with the existing FY-4A / AGRI cloud detection products, the cloud cover in the ERA5 data or the cloud cover information of other products to evaluate and verify the accuracy.

[0198] Application examples:

[0199] This example takes the 2018 typhoon "Maria (1808)" as an example, and uses the FY-4A / GIIRS long-wave channel data to conduct experiments based on the channel dynamic cloud detection method based on the geostationary meteorological satellite hyperspectral infrared detector of the present invention.

[0200] The (sparse) cloud profile of the viewing point can be obtained by using the weighted L1 norm constraint method. Figure 3 The distribution of window channel cloud images of the FengYun-4A (FY-4A) multi-channel scanning imaging radiation in the area affected by Typhoon Maria is given, as follows: Figure 3 (a). Two viewpoints, “clear sky” (FOV-1) and “cloudy” (FOV-2), are selected on the FY-4A / AGRI cloud map to interpret the correlation method. The brightness temperature distribution of the FY-4A / GIIRS channel at the viewpoints FOV-1 and FOV-2 is further given, as shown in Figure 3 (b). The weighted L1 norm constraint method is used to obtain the cloud profile distribution of the field of view points FOV-1 and FOV-2 as shown in Figure 3 (c) and Figure 3 As shown in (d).

[0201] Depend on Figure 3 It can be seen that at the 45th layer of the "clear sky" field of view (FOV-1) mode, the cloud amount is 0.001 (this value can be ignored), and at the 97th layer of the mode, the cloud amount is 0.999. The mathematical solution of cloud amount 0.999 indicates that the clouds are located near the ground, but this does not conform to the actual situation. The physical solution combined with meteorological background knowledge shows that this is a clear sky field of view point.

[0202] At the 45th layer of the "cloudy" field of view (FOV-2) mode, the cloud amount is 0.372, at the 53rd layer of the mode, the cloud amount is 0.286, and at the 97th layer of the field of view mode, the cloud amount is 0.342. This highlights the information of multiple layers of clouds (different from the minimum residual method), and not every layer of the model has a cloud amount value (different from the multivariate minimum residual method).

[0203] This example analyzes the channel adaptive dynamic cloud detection effect of the present invention as follows:

[0204] Considering that high-impact weather is often accompanied by the formation of clouds, and considering that the numerical weather forecast systems of most international operational weather forecast units can only assimilate satellite infrared detector data at clear sky view points. However, if a cloud detection method for "clear sky view points" is used, some useful information will be lost. Based on this, the channel-adaptive dynamic cloud detection method of the present invention starts from identifying the "clear sky channel" that is not affected by clouds, and can dynamically and adaptively mine the brightness temperature information of each selected hyperspectral channel.

[0205] Figure 4 The effect diagram of channel adaptive dynamic cloud detection in the high-frequency observation area of ​​typhoon Maria at 05:30 on July 10, 2018 is given. Considering the space limitation, only the FY-4A / GIIRS long-wave channel 9 (the peak layer of the weight function is located at 300hPa, Figure 4 Middle (a)), channel 47 (the peak layer of the weight function is located at 555.167hPa, Figure 4 Middle (b)), channel 88 (the peak layer of the weight function is located at 729.886hPa, Figure 4 (c)) and channel 112 (the weight function peak layer is located at 1042.2 hPa, Figure 4 The cloud detection effect in (d) is shown in Figure 1. The color code 0 is marked as “clear sky” and the color code 1 is marked as “cloudy”. The cloud distribution map of the FY-4A / AGRI cloud product matching the FY-4A / GIIRS field of view is further given ( Figure 4 In other words, similar traditional cloud products such as Figure 4 (e) and the method of the present invention Figure 4 Middle (a), Figure 4 Middle (b), Figure 4 Middle (c), Figure 4 Comparison of cloud cover products obtained in (d).

[0206] Depend on Figure 4 It can be seen that for channels with a peak layer of the channel weight function at a high level (for example, channel 9), the number of clear sky field points (areas marked with 0) obtained is relatively large, while for channels at a low level (for example, channel 112), the number of clear sky field points obtained is relatively small compared to the high level. The method of the present invention can identify channels affected by clouds, rather than the "one-size-fits-all" method of "minimum residual method" field point cloud detection. The cloud detection results of the present invention retain more data from different channels and realize adaptive cloud detection of channels. This lays the foundation for the variational assimilation of hyperspectral infrared detector data in the later stage, and is expected to improve the accuracy of numerical weather forecasting.

[0207] The technical solutions of the present invention are not limited to the above-mentioned embodiments, and any technical solutions obtained by equivalent replacement methods fall within the scope of protection required by the present invention.

Claims

1. A channel dynamic cloud detection method based on a meteorological satellite hyperspectral infrared detector, characterized in that: The steps include: Step 1, obtaining the observed brightness temperature of the meteorological satellite hyperspectral infrared detector and the background field data of the global data assimilation system that are matched in time and space, and preprocessing the observed brightness temperature, interpolating the background field data to the field of view point based on the latitude and longitude information of the field of view point of the observed brightness temperature, and obtaining the processed background field data; Step 2: Based on the processed background field data, the brightness temperature of each channel and each field point of the hyperspectral infrared detector is simulated to obtain the simulated brightness temperature of all channels; Step 3, using a nonlinear channel optimal selection method to select the channel of the meteorological satellite hyperspectral infrared detector to obtain the optimal channel; In the step 3, a nonlinear channel optimal selection method is used to select the channel of the meteorological satellite hyperspectral infrared detector, and the preferred channel is specifically: Step 3.1, calculate the relative entropy of the hyperspectral infrared detector channel: Where p(x) represents the probability distribution of meteorological variables of the background field in the field of view; p(x|y) represents the probability distribution of meteorological variables of the analysis field formed by correcting the meteorological variables of the background field after obtaining the channel brightness temperature of the field of view; Step 3.2, calculate the mutual information of the channel; Where p(y) represents the probability distribution of brightness temperature observed by the channel; Step 3.3, select the channel with the largest mutual information as the selected channel; Step 3.4, updating the meteorological variables of the analysis field and the error covariance matrix of the meteorological variables of the analysis field according to the channel brightness temperature of the selected channel; The expression of the error covariance matrix of the meteorological variables in the analysis field after the n+1th iteration is as follows: Where A represents the error covariance matrix of meteorological variables in the analysis field; A n+1 represents the error covariance matrix of meteorological variables in the analysis field after the n+1th iteration; B represents the error covariance matrix of meteorological variables in the background field; R represents the error covariance matrix of observations in the selected channels; H n represents the Jacobi matrix after the nth iteration; Step 3.5, repeating steps 3.1 to 3.4 for iterative optimization until the required number of selected channels are selected as the preferred channels; Step 4, select the simulated brightness temperature and the observed brightness temperature corresponding to the preferred channel to perform field point cloud profile inversion to obtain cloud profile information that conforms to the actual characteristics of cloud cover; Step 5: Based on the cloud profile information, the channel adaptive cloud detection index method is used to dynamically determine whether each preferred channel of the hyperspectral infrared detector is affected by clouds at a certain field of view point.

2. The channel dynamic cloud detection method based on a meteorological satellite hyperspectral infrared detector according to claim 1 is characterized in that: In the step 1, the observed brightness temperature is pre-processed by apodization, and the background field data is interpolated to the field of view point by using bilinear interpolation.

3. The channel dynamic cloud detection method based on a meteorological satellite hyperspectral infrared detector according to claim 2 is characterized in that: In step 3, p(x), p(y) and p(x|y) are obtained by using a method of numerically approximating the probability distribution based on a Gaussian mixture model nonlinear resampling.

4. The channel dynamic cloud detection method based on a meteorological satellite hyperspectral infrared detector according to claim 3 is characterized in that: At each field of view point, the fast radiation transfer mode is used to simulate the brightness temperature of each channel of the hyperspectral infrared detector. Specifically, the simulated radiation value is first calculated, and then the radiation value is converted into brightness temperature through the Planck function. The expression of the simulated radiation value is: In the formula, R v represents the channel simulation radiation value with frequency v; N is the number of radiation transmission mode layers; c k is the cloud cover of the kth layer of the model; It represents the cloud radiation simulation when there are clouds in the kth layer of the model; c0 is the clear sky radiation ratio; Represents clear-sky radiation.

5. The channel dynamic cloud detection method based on a meteorological satellite hyperspectral infrared detector according to claim 4 is characterized in that: The brightness temperature of each field of view point in each channel of the hyperspectral infrared detector is simulated by the field of view point-by-field of view channel simulation method. The brightness temperature simulation is divided into clear sky simulation and layer-by-layer cloud simulation: the clear sky simulation is to simulate the clear sky brightness temperature of the channel when the field of view point is a clear sky field of view point; the layer-by-layer cloud simulation is to simulate the cloud brightness temperature of the channel when it is assumed that there is a cloud at the field of view point. The cloud simulation here is the cloud simulation in different mode layers.

6. The channel dynamic cloud detection method based on a meteorological satellite hyperspectral infrared detector according to claim 5 is characterized in that: In step 4, the target minimization function of the field of view point cloud volume profile c is: Where W is the weight of the channel brightness temperature on the minimization function; λ is the regularization parameter; BT sim is the simulated brightness temperature of the preferred channel; BT obs is the observed brightness temperature of the preferred channel; BT sim The simulated radiation value in step 4 is calculated by the Planck function to obtain: and BT obs = (BT obs，1 , BT obs，2 ,..., BT obs,n ) T Where n c and N represent the number of satellite channels and the number of mode layers, respectively; Indicates the nth c The clear sky simulated brightness temperature of each channel; Indicates the nth c The observed brightness temperature of each channel; Indicates the nth c The cloud-simulated brightness temperature of the channel when there are clouds in the Nth layer of the model; The constraint condition of the target minimization function of the field of view point cloud volume profile c is: In the formula, c i represents the cloud cover of the i-th model layer; N represents the number of model layers.

7. The channel dynamic cloud detection method based on a meteorological satellite hyperspectral infrared detector according to claim 6 is characterized in that: The weight W of the channel brightness temperature to the minimization function is calculated by numerically approximating the signal degree of freedom DFS. The specific process is as follows: ① Solve the following minimization function and obtain two optimal solutions c and c respectively * ; and In the formula, For BT obs The Gaussian perturbation of ξ0 represents a random Gaussian variable; ② Calculate W by numerically approximating the signal degree of freedom DFS; In the formula, R error represents the error covariance matrix.

8. The channel dynamic cloud detection method based on a meteorological satellite hyperspectral infrared detector according to claim 5 is characterized in that: In step 5, a channel-adaptive cloud detection index method is used to dynamically determine whether each preferred channel of the hyperspectral infrared detector is affected by clouds at a certain field of view point, specifically: If |∑(c i *BT sim,i,j )-BT sim,clr,i,j |<0.01*BT sim,clr,i,j Then, it means that channel j is not affected by the cloud at the field of view point i, marked as label i,j =0, this channel is used to observe data at this field of view point; If |∑(c i *BT sim,i,j )-BT sim,clr,i,j |≥0.01*BT sim,clr,i,j Then, it means that channel j is affected by the cloud at the field of view point i, marked as label i,j =1, the observation data of this channel at this field of view point will be eliminated; c i =(c i,1 ,c i,2 ,...,c i,N ,c i,0 ) T represents the cloud profile at the viewing point i; BT sim,i,j =[BT sim,i,j,k ,BT sim,clr,i,j ] k＝1,...N , [BT sim,i,j,k ] k＝1,...N Indicates the cloud-simulated brightness temperature of the jth channel at the i-th field of view for the 1st to Nth layers of the model; N represents the number of radiation transfer model layers; BT sim,clr,i,j Represents the clear sky simulated brightness temperature of the jth channel at the i-th field of view point.

9. A channel dynamic cloud detection system based on a meteorological satellite hyperspectral infrared detector, used to implement the channel dynamic cloud detection method based on a meteorological satellite hyperspectral infrared detector according to any one of claims 1 to 8, characterized in that: It includes the following parts: The data acquisition and preprocessing module acquires the observed brightness temperature of the meteorological satellite hyperspectral infrared detector and the background field data of the global data assimilation system with time and space matching, and preprocesses the observed brightness temperature. Based on the latitude and longitude information of the field of view point of the observed brightness temperature, the background field data is interpolated to the field of view point to obtain the processed background field data; The simulation data construction module simulates the brightness temperature of each channel and each field point of the hyperspectral infrared detector based on the processed background field data to obtain the simulated brightness temperature of all channels; The nonlinear channel optimal selection module uses the nonlinear channel optimal selection method to select the channel of the meteorological satellite hyperspectral infrared detector to obtain the optimal channel; Inversion of the field point cloud profile: Select the simulated brightness temperature and observed brightness temperature corresponding to the preferred channel to invert the field point cloud profile, and obtain the cloud profile information that conforms to the actual characteristics of the cloud; The channel-adaptive dynamic cloud detection module uses the channel-adaptive cloud detection index method based on cloud profile information to dynamically determine whether each preferred channel of the hyperspectral infrared detector is affected by clouds at a certain field of view point.

Images