CN102129566A - Method for identifying rainstorm cloud cluster based on stationary meteorological satellite - Google Patents

Abstract

The invention relates to a method for identifying rainstorm cloud clusters based on stationary meteorological satellites, and belongs to the technical field of atmospheric monitoring. In order to improve the accuracy of rainstorm cloud identification, the method includes: segmenting the cloud image to obtain the category of each cloud at the current moment; synthesizing multiple cloud images within a set period of time to obtain a short-term basic brightness temperature map; calculating the cloud image at the current moment and The gray value difference image of the short-term basic brightness temperature map; the gray value difference image is segmented to identify the candidate cloud clusters in rainstorm weather; for the obtained various cloud clusters, combined with the candidate cloud clusters, use historical sample data to identify Get the final storm cloud cluster. The technical scheme of the present invention compares with the current rainstorm identification method that only considers the static intensity characteristics and texture characteristics of cloud clusters, considers the evolution process of cloud clusters such as generation, development, splitting, and merging, and proposes the concept and calculation method of short-term basic brightness temperature maps, It can identify the position of cloud clusters that change drastically in a short period of time, which is conducive to the early detection of targets in the formation of storm clouds, with high accuracy.

Description

A Method for Identifying Rainstorm Clouds Based on Geostationary Meteorological Satellites

technical field

The invention relates to the technical field of atmospheric monitoring, in particular to a method for identifying rainstorm clouds based on stationary meteorological satellites.

Background technique

Heavy rainfall is a key factor triggering floods, landslides, and mudslides. Accurate identification of the location and range of convective cloud clusters that cause heavy rainfall, as well as tracking the movement path of cloud clusters, can provide correct trigger conditions and corresponding parameters for the activation of downstream disaster sub-chains such as floods, landslides, and debris flows.

Geostationary meteorological satellites can continuously observe about one-third of the earth's surface for 24 hours, and generate a set of remote sensing data every half an hour. The observation range is wide and the observation frequency is high, and weather phenomena that change rapidly over time can be captured. It is especially suitable for early warning of mesoscale strong convective weather. These advantages are not possessed by polar-orbiting meteorological satellites and ground observation means. Therefore, it is of great practical significance to use geostationary satellite remote sensing data to identify and track strong convective clouds.

Contents of the invention

(1) Technical problems to be solved

The technical problem to be solved by the present invention is how to improve the accuracy of the rainstorm cloud recognition method.

(2) Technical solution

In order to solve the problems of the technologies described above, the present invention provides a method for identifying storm clouds based on stationary meteorological satellites, said method comprising the steps of:

Step S1: Segment the cloud image obtained by the stationary meteorological satellite to obtain the category of each cloud group at the current moment in the observation area;

Step S2: According to the cloud images acquired by stationary meteorological satellites, synthesize multiple cloud images in the set time period before the current moment to obtain the basic brightness temperature map of each cloud group in the set time period, and obtain the basic brightness temperature map obtained after the synthesis. The brightness temperature map is defined as the short-term basic brightness temperature map;

Step S3: Calculate the gray value difference image between the cloud image at the current moment and the short-term basic brightness temperature image;

Step S4: Segment the gray value difference image to identify candidate cloud clusters that may have rainstorm weather;

Step S5: For the various types of cloud clusters obtained by segmentation in step S1, combined with the candidate cloud clusters identified in step S4, use the historical sample data of the observation area obtained by stationary meteorological satellites to identify the final rainstorm cloud cluster .

In the step S1, the following steps are specifically included:

Step S101: Read the cloud image at the current time t and the time t-1 of the previous hour obtained by the stationary meteorological satellite, use the gray value threshold method to segment the cloud image, and divide the grid points in the cloud image into There are two categories greater than or equal to the threshold and less than the threshold, and the parts greater than or equal to the threshold are respectively recorded as point set S(t) and point set S(t-1); where point set S(t) is the cloud at the current time t Image point set, point set S(t-1) is the cloud cluster image point set at time t-1 in the last hour;

Step S102: Mark the connected regions in the point set S(t) and the point set S(t-1), and record the parameters related to the cloud precipitation intensity, which specifically include: the maximum gray value of the cloud , the minimum gray value of the cloud, the area of the cloud and the change of the position of the cloud;

Step S103: For each cloud in the point set S(t), by judging whether there is a corresponding source cloud in the point set S(t-1), the source cloud in the point set S(t-1) The cloud clusters in the point set S(t) are divided into ten categories based on the number of cloud clusters, whether their average gray value increases or decreases when compared with the source cloud clusters, and the change of the cloud cluster area.

The ten categories in the step S103 belong to four major categories, and the four major categories specifically include: newly added cloud clusters, growth and change cloud clusters, split cloud clusters, and merged cloud clusters;

If a certain cloud group in the point set S(t) does not intersect with any cloud group in the point set S(t-1), it is considered as a newly added cloud group;

If a certain cloud in the point set S(t) only intersects with a certain cloud in the point set S(t-1), it is considered that it is caused by the change of the cloud in the point set S(t-1). The growth and change class cloud clusters that come; further, according to the change situation of the cloud cluster area, the growth change class cloud clusters are also specifically divided into translational change class cloud clusters, expansion change class cloud clusters and contraction change class cloud clusters; The areas of the cloud cluster at time t-1 and time t are A _t-1 and A _t respectively, then when m ₁ *A _t-1 ≤A _t ≤n ₁ *A _t-1 , the growth and change type cloud cluster is considered It is a translational change cloud cluster; when A _t >n ₁ *A _t-1 , the growth change cloud cluster is considered to be an expansion change cloud cluster; when A _t <m ₁ *A _t-1 , the growth change cloud cluster is determined to be The changing cloud clusters are shrinking and changing cloud clusters; among them, the parameter values m ₁ and n ₁ are preset values according to the actual situation, and n ₁ >m ₁ ≥1;

If multiple cloud clusters in the point set S(t) intersect with a certain cloud cluster C _j in the point set S(t-1), then these cloud clusters can be identified as all formed by the point set S(t-1 ) in the cloud group C _j in the split cloud group; further, according to the point set S(t) in the multiple cloud clusters and the area relationship of the cloud group C _j , the split cloud group is also specifically divided into Growth-split cloud clusters, common split-type cloud clusters, and independent-split cloud clusters; if the area of one of these cloud clusters intersecting C _j in the point set S(t) is A _t , the area of _Cj is A _Cj , then when A _t >n ₂ *A _Cj , the split-type cloud cluster is considered to be a growing split-type cloud cluster; when m ₂ *A _Cj ≤A _t ≤n ₂ *A _Cj , the split-type cloud cluster is considered The cluster is an ordinary split cloud cluster; when A _t <m ₂ *A _Cj , the split cloud cluster is considered to be an independent split cloud cluster; among them, the parameter values m ₂ and n ₂ are pre-set according to the actual situation value, and n ₂ >m ₂ >0;

If a certain cloud in the point set S(t) of the current cloud image intersects with multiple clouds in the point set S(t-1), then the cloud can be identified as the point set S(t-1). ) in a plurality of cloud clusters merged; further, according to whether the area of the current cloud cluster is greater than the area summation of a plurality of cloud clusters in S(t-1), the merged cloud cluster is also specific They are divided into growth-merging cloud clusters, common merger-type cloud clusters, and possible false-merger cloud clusters; if the area of the current cloud cluster is A _t , the number of n cloud clusters in the intersecting point set S(t-1) The areas are A _i , i=1, 2,...n; then when A _t > sum(A _i ), the merged cloud cluster is considered to be a growing merged cloud cluster; when max(A _i )≤A When _t ≤ sum(A _i ), the merged cloud cluster is considered to be a common merged cloud cluster; when A _t < max(A _i ), the merged cloud cluster is considered to be a possible false merged cloud cluster.

In the step S2, it specifically includes:

Step S201: Aligning multiple cloud images within a set time period before the current moment;

Step S202: For the gray value at a certain grid point in the short-term basic brightness temperature map, select the smallest gray value among the multiple gray values at the same grid point in multiple cloud images;

Step S203: For the gray values at all grid points in the short-term basic brightness temperature map, use the method in step S202 to obtain gray value values, so as to synthesize and obtain the short-term basic brightness temperature map .

In the step S3, it specifically includes:

Step S301: aligning the cloud image at the current moment with the short-term basic brightness temperature image;

Step S302: For the gray value at a certain grid point in the gray value difference image, subtract the short-term basic brightness temperature from the gray value at the same grid point in the cloud image at the current moment The gray value at the same grid point in the figure, the obtained difference is selected as the gray value at the grid point in the gray value difference image; if the difference is less than zero, the gray value The gray value at the grid point in the difference image is 0;

Step S303: For the gray values at all the grid points in the gray value difference image, the method in step S302 is used to obtain the gray value, so as to obtain the gray value difference image.

In the step S4, the gray value difference image is segmented by a threshold method.

In the step S5, for the newly-added cloud group obtained by segmentation in the step S1, specifically include the following steps:

Step S501a: judging whether it is the candidate cloud group identified in step S4, if yes, proceed to S502a, otherwise, identify it as a non-rainstorm cloud group;

Step S502a: Whether the maximum gray value of the grid points it contains is greater than the preset threshold T ₁ , if yes, identify it as a rainstorm cloud cluster; otherwise, identify it as a non-rainstorm cloud cluster; where, the threshold T ₁ Based on the one-hour minimum precipitation of the rainstorm defined by meteorology, it is determined by the historical sample data in the past set months according to the minimum misjudgment probability criterion. The historical sample data includes: the maximum gray value of the cloud cluster, The minimum gray value of the cloud, the area of the cloud, and the change of the position of the cloud.

In the step S5, for the growth and change class cloud clusters obtained by segmentation in the step S1, specifically include the following steps:

Step S501b: judging whether it is the candidate cloud group identified in step S4, if yes, proceed to S502b or S503b, otherwise, identify it as a non-storm cloud group;

Step S502b: For the cloud clusters of the translational change type and the expansion change type, if the maximum gray value of the grid points contained therein is greater than the threshold T ₂ , identify it as a rainstorm cloud cluster; otherwise, identify it as a non-rainstorm cloud cluster;

Step S503b: For the shrinking and changing cloud cluster, if its area is greater than the threshold value V ₁ , identify it as a rainstorm cloud cluster; otherwise, identify it as a non-rainstorm cloud cluster; wherein, the thresholds T ₂ and V ₁ are both meteorologically defined The one-hour minimum precipitation of the rainstorm is based on the historical sample data in the past set month according to the minimum misjudgment probability criterion. The historical sample data includes: the maximum gray value of the cloud cluster, the minimum gray value of the cloud cluster value, cloud area, and cloud position changes.

In the step S5, for the split-type cloud cluster obtained by segmentation in the step S1, specifically include the following steps:

Step S501c: judging whether it is the candidate cloud group identified in step S4, if yes, proceed to S502c or S503c, otherwise, identify it as a non-storm cloud group;

Step S502c: For the cloud clusters of growth splitting type and common splitting type, identify them as torrential rain cloud clusters;

Step S503c: For the independent split cloud cluster, identify it as a non-storm cloud cluster.

In the step S5, for the merging class cloud group obtained by segmentation in the step S1, specifically include the following steps:

Step S501d: judging whether it is the candidate cloud group identified in step S4, if yes, proceed to S502d or S503d, otherwise, identify it as a non-rainstorm cloud group;

Step S502d: For the cloud clusters of growth-merging type and common-merging type, identify them as torrential rain cloud clusters.

Step S503d: For the possible false merged cloud cluster, if its area is smaller than the threshold value V ₂ , identify it as a non-rainstorm cloud cluster; otherwise, identify it as a rainstorm cloud cluster; wherein, the threshold V ₂ is defined by meteorology on rainstorm Based on the one-hour minimum precipitation of , it is determined by the historical sample data in the past set month according to the minimum misjudgment probability criterion. The historical sample data includes: the maximum gray value of the cloud cluster, the minimum gray value of the cloud cluster, Cloud area and cloud position changes.

(3) Beneficial effects

The present invention compares with prior art, and its beneficial effect is:

(1) Compared with the existing rainstorm identification method that only considers the static intensity characteristics and texture features of cloud clusters, the technical solution of the present invention considers the evolution process of cloud clusters such as generation, development, splitting, and merging, and improves the accuracy of rainstorm cloud cluster identification sex.

(2) The technical scheme of the present invention proposes the concept and calculation method of the short-term basic brightness temperature map, which can be used to identify the cloud position that changes drastically in a short time, and is conducive to the early detection of targets in the formation of heavy rain clouds. Therefore, compared with the prior art , with higher accuracy.

Description of drawings

Fig. 1 is the flow chart of the method for identifying rainstorm cloud group involved in the specific embodiment of the present invention;

Fig. 2 is a short-term basic brightness temperature diagram related to a specific embodiment of the present invention;

FIG. 3 is a gray value difference image related to a specific embodiment of the present invention;

Fig. 4 is the identification result figure of the rainstorm cloud group involved in the specific embodiment of the present invention;

Fig. 5 is a comparison chart of the recognition result involved in the specific embodiment of the present invention and the actual rainfall situation in the future.

Detailed ways

In order to make the purpose, content and advantages of the present invention clearer, the specific implementation manners of the present invention will be further described in detail below in conjunction with the accompanying drawings and embodiments.

Through the analysis of a large number of historical sample data, combined with the basic principles of atmospheric radiation, the radiation characteristics, texture characteristics and movement characteristics of the rainstorm cloud on each channel of the FY2 satellite are studied. On this basis, the rainstorm cloud is proposed. Detection method.

After analysis, it is found that the characteristics of the rainstorm cloud cluster are the most obvious on the image of the infrared split window channel 1 (hereinafter referred to as the IR1 channel), which can be summarized as follows:

(1) The average gray value of strong convective cloud clusters is higher, that is, the temperature is lower. Heavy precipitation cloud clusters with a gray value above 200 account for more than half of all heavy precipitation processes. However, due to latitude, altitude, season and other factors, the average gray value of some non-precipitating cloud clusters is also above 200, so it is impossible to judge whether there is heavy precipitation only by the gray value;

(2) The possibility of heavy precipitation appearing on the windward side of the cloud cluster is greater than the possibility of appearing in other locations;

(3) Clouds that change drastically in a short period of time are more likely to have heavy precipitation;

(4) After several newly formed small cloud clusters are merged into one large cloud cluster, there is a greater possibility of heavy precipitation;

(5) Smaller cloud clusters independent from the original larger cloud clusters generally do not experience heavy precipitation.

According to the above analysis, we use the IR1 channel as the data source. The cloud images of several consecutive times before the current moment are regarded as a whole, and the cloud cluster is used as the unit for research. At the same time, the generation and evolution process of the cloud cluster is investigated, and various indicators are integrated for identification. The important indicators related to the precipitation intensity of the cloud cluster include the maximum and minimum gray value of the cloud cluster, the area of the cloud cluster, the position change of the cloud cluster, and the evolution of the cloud cluster (new generation, splitting, merging, etc.).

So far, in order to improve the accuracy of the rainstorm cloud identification method, the method for identifying rainstorm clouds based on stationary meteorological satellites provided by the present invention, as shown in Figure 1, specifically includes:

1. Identification of strong convective cloud clusters

Step S1: Segment the cloud image obtained by the stationary meteorological satellite to obtain the category of each cloud group at the current moment in the observation area;

In the step S1, the following steps are specifically included:

Step S101: Read the cloud image at the current time t and the time t-1 of the previous hour obtained by the stationary meteorological satellite, use the gray value threshold method to segment the cloud image, and divide the grid points in the cloud image into There are two categories greater than or equal to the threshold and less than the threshold, and the parts greater than or equal to the threshold are respectively recorded as point set S(t) and point set S(t-1); where point set S(t) is the cloud at the current time t Image point set, point set S(t-1) is the cloud cluster image point set at time t-1 in the last hour;

Step S102: Mark the connected regions in the point set S(t) and the point set S(t-1), and record the parameters related to the cloud precipitation intensity, which specifically include: the maximum gray value of the cloud , the minimum gray value of the cloud, the area of the cloud, and the position change of the cloud; then do an open operation to remove isolated points and smooth the boundary of the region;

Step S103: For each cloud in the point set S(t), by judging whether there is a corresponding source cloud in the point set S(t-1), the source cloud in the point set S(t-1) The cloud clusters in the point set S(t) are divided into ten categories based on the number of cloud clusters, whether their average gray value increases or decreases when compared with the source cloud clusters, and the change of the cloud cluster area.

The ten categories in the step S103 belong to four major categories, and the four major categories specifically include: newly added cloud clusters, growth and change cloud clusters, split cloud clusters, and merged cloud clusters;

If a certain cloud group in the point set S(t) does not intersect with any cloud group in the point set S(t-1), it is considered as a newly added cloud group;

If a certain cloud in the point set S(t) only intersects with a certain cloud in the point set S(t-1), it is considered that it is caused by the change of the cloud in the point set S(t-1). The growth and change class cloud clusters that come; further, according to the change situation of the cloud cluster area, the growth change class cloud clusters are also specifically divided into translational change class cloud clusters, expansion change class cloud clusters and contraction change class cloud clusters; The areas of the cloud cluster at time t-1 and time t are A _t-1 and A _t respectively, then when m ₁ *A _t-1 ≤A _t ≤n ₁ *A _t-1 , the growth and change type cloud cluster is considered It is a translational change cloud cluster; when A _t >n ₁ *A _t-1 , the growth change cloud cluster is considered to be an expansion change cloud cluster; when A _t <m ₁ *A _t-1 , the growth change cloud cluster is determined to be The changing cloud clusters are shrinking and changing cloud clusters; among them, the parameter values m ₁ and n ₁ are preset values according to the actual situation, and n ₁ >m ₁ ≥1; for example, m ₁ =1 can be set, n ₁ =2;

If multiple cloud clusters in the point set S(t) intersect with a certain cloud cluster C _j in the point set S(t-1), then these cloud clusters can be identified as all formed by the point set S(t-1 ) in the cloud group C _j in the split cloud group; further, according to the area relationship between the cloud group and the cloud group Cj in the point set S(t), the split class cloud group is also specifically divided into growth Split-type cloud clusters, ordinary split-type cloud clusters, and independent split-type cloud clusters; if the area of one of these cloud clusters intersecting C _j in the point set S(t) is A _t , the area of C _j is A _Cj , then when A _t >n ₂ *A _Cj , the split-type cloud cluster is considered to be a growing split-type cloud cluster; when m ₂ *A _Cj ≤A _t ≤n ₂ *A _Cj , the split-type cloud cluster is considered to be The cluster is an ordinary split cloud cluster; when A _t <m ₂ *A _Cj , the split cloud cluster is considered to be an independent split cloud cluster; among them, the parameter values m ₂ and n ₂ are pre-set according to the actual situation value, and n ₂ >m ₂ >0; for example, m ₂ =0.5, n ₂ =1 can be set;

If a certain cloud in the point set S(t) of the current cloud image intersects with multiple clouds in the point set S(t-1), then the cloud can be identified as the point set S(t-1). ) in a plurality of cloud clusters merged; further, according to whether the area of the current cloud cluster is greater than the area summation of a plurality of cloud clusters in S(t-1), the merged cloud cluster is also specific They are divided into growth-merging cloud clusters, common merger-type cloud clusters, and possible false-merger cloud clusters; if the area of the current cloud cluster is A _t , the number of n cloud clusters in the intersecting point set S(t-1) The areas are A _i , i=1, 2,...n; then when A _t > sum(A _i ), the merged cloud cluster is considered to be a growing merged cloud cluster; when max(A _i )≤A When _t ≤ sum(A _i ), the merged cloud cluster is considered to be a common merged cloud cluster; when A _t < max(A _i ), the merged cloud cluster is considered to be a possible false merged cloud cluster.

2. Image preprocessing

Step S2: According to the cloud images acquired by stationary meteorological satellites, synthesize the N cloud images within the set time period before the current moment, calculate the minimum value of the gray value at each grid point corresponding to each of the N cloud images, and obtain the set The basic brightness temperature map of each cloud group in the time period, the basic brightness temperature map obtained after the synthesis is defined as the short-term basic brightness temperature map;

In the step S2, it specifically includes:

Step S201: Aligning multiple cloud images within a set time period before the current moment;

Step S202: For the gray value at a certain grid point in the short-term basic brightness temperature map, select the smallest gray value among the multiple gray values at the same grid point in multiple cloud images;

Step S203: For the gray values at all grid points in the short-term basic brightness temperature map, use the method in step S202 to obtain gray value values, so as to synthesize and obtain the short-term basic brightness temperature map ;

Calculated with one satellite image every half hour, if N takes 3, that is, use the three images in the past 2 hours to calculate the highest brightness temperature, indicating the basic brightness temperature of each cloud cluster in the past 2 hours. As shown in Figure 2, Figure 2 is a short-term maximum brightness temperature map synthesized from cloud images at three moments before 0:00 on August 1 (that is, 23:30, 23:00, and 22:30 on July 31).

Step S3: Calculate the gray value difference image between the cloud image at the current moment and the short-term basic brightness temperature image;

In the step S3, it specifically includes:

Step S301: aligning the cloud image at the current moment with the short-term basic brightness temperature image;

Step S302: For the gray value at a certain grid point in the gray value difference image, subtract the short-term basic brightness temperature from the gray value at the same grid point in the cloud image at the current moment The gray value at the same grid point in the figure, the obtained difference is selected as the gray value at the grid point in the gray value difference image; if the difference is less than zero, the gray value The gray value at the grid point in the difference image is 0;

Step S303: For the gray values at all the grid points in the gray value difference image, the method in step S302 is used to obtain the gray value, so as to obtain the gray value difference image.

The short-term maximum brightness temperature map involves fewer pictures and the time span is small, so it is not an approximate "eye-space brightness temperature" map, but the basic brightness temperature of each cloud cluster in a short period of time. After making a difference with the current cloud map, the cloud group that has been significantly strengthened in the near future can be marked out. As shown in Figure 3, Figure 3 is the image obtained by doing the difference between the 0-point cloud image on August 1 and Figure 2. In Figure 3, the larger the gray value, the greater the increase in cloud cover in the area in the past 2 hours. There is a greater possibility of heavy precipitation in these areas. Use an image segmentation algorithm, such as the threshold method, to segment the cloud cluster in Figure 3, and identify the candidate areas of the rainstorm cloud cluster.

Step S4: Using the threshold method to segment the gray value difference image to identify candidate cloud clusters that may have rainstorm weather;

Wherein, the order of step S1 and step S2-step S4 can be reversed, that is, the identification process of the strong convective cloud layer and the image preprocessing process can be reversed.

3. Identifying rainstorm clouds

Step S5: For the various types of cloud clusters obtained by segmentation in step S1, combined with the candidate cloud clusters identified in step S4, use the historical sample data of the observation area obtained by stationary meteorological satellites to identify the final rainstorm cloud cluster .

In the step S5, for the newly-added cloud group obtained by segmentation in the step S1, specifically include the following steps:

Step S501a: judging whether it is the candidate cloud group identified in step S4, if yes, proceed to S502a, otherwise, identify it as a non-rainstorm cloud group;

Step S502a: Whether the maximum gray value of the grid points it contains is greater than the preset threshold T ₁ , if yes, identify it as a rainstorm cloud cluster; otherwise, identify it as a non-rainstorm cloud cluster; where, the threshold T ₁ Based on the one-hour minimum precipitation of the rainstorm defined by meteorology, it is determined by the historical sample data in the past set months according to the minimum misjudgment probability criterion. The historical sample data includes: the maximum gray value of the cloud cluster, The minimum gray value of the cloud, the area of the cloud, and the change of the position of the cloud.

In the step S5, for the growth and change class cloud clusters obtained by segmentation in the step S1, specifically include the following steps:

Step S501b: judging whether it is the candidate cloud group identified in step S4, if yes, proceed to S502b or S503b, otherwise, identify it as a non-storm cloud group;

Step S502b: For the cloud clusters of the translational change type and the expansion change type, if the maximum gray value of the grid points contained therein is greater than the threshold T ₂ , identify it as a rainstorm cloud cluster; otherwise, identify it as a non-rainstorm cloud cluster;

Step S503b: For the shrinking and changing cloud cluster, if its area is greater than the threshold value V ₁ , identify it as a rainstorm cloud cluster; otherwise, identify it as a non-rainstorm cloud cluster; wherein, the thresholds T ₂ and V ₁ are both meteorologically defined The one-hour minimum precipitation of the rainstorm is based on the historical sample data in the past set month according to the minimum misjudgment probability criterion. The historical sample data includes: the maximum gray value of the cloud cluster, the minimum gray value of the cloud cluster value, cloud area, and cloud position changes.

In the step S5, for the split-type cloud cluster obtained by segmentation in the step S1, specifically include the following steps:

Step S501c: judging whether it is the candidate cloud group identified in step S4, if yes, proceed to S502c or S503c, otherwise, identify it as a non-storm cloud group;

Step S502c: For the cloud clusters of growth splitting type and common splitting type, identify them as torrential rain cloud clusters;

Step S503c: For the independent split cloud cluster, identify it as a non-storm cloud cluster.

In the step S5, for the merging class cloud group obtained by segmentation in the step S1, specifically include the following steps:

Step S501d: judging whether it is the candidate cloud group identified in step S4, if yes, proceed to S502d or S503d, otherwise, identify it as a non-rainstorm cloud group;

Step S502d: For the cloud clusters of growth-merging type and common-merging type, identify them as torrential rain cloud clusters.

Step S503d: For the possible false merged cloud cluster, if its area is smaller than the threshold value V ₂ , identify it as a non-rainstorm cloud cluster; otherwise, identify it as a rainstorm cloud cluster; wherein, the threshold V ₂ is defined by meteorology on rainstorm Based on the one-hour minimum precipitation of , it is determined by the historical sample data in the past set month according to the minimum misjudgment probability criterion. The historical sample data includes: the maximum gray value of the cloud cluster, the minimum gray value of the cloud cluster, Cloud area and cloud position changes.

Among them, the selection of the above-mentioned historical sample data is preferably the historical sample data of the past 30 days, but if the current month is May, since the rainstorms are mainly concentrated in June, July, and August, the historical sample data is selected from last year’s June and July. , August data.

Finally, as shown in Figure 4, it is the recognition result of the 0-point cloud map on August 1, and the position of the rainstorm cloud cluster has been marked by the boundary line.

4. Inspection method

Next, a method for detecting whether the result of the above-mentioned identified rainstorm cloud cluster is true or not is described.

Since the weather system cannot be reproduced, and there is currently no technical means that can continuously and completely detect all weather phenomena, no matter what kind of inspection method, there are certain limitations.

Here we use the one-hour precipitation record of the automated rainfall station as a reference standard to evaluate the algorithm. At present, there are more than 20,000 automatic rainfall stations in my country, which are densely distributed in most of the eastern regions, and the spatial resolution is close to the resolution of the FY2 satellite infrared channel. In addition, due to the small number and sparse distribution of rainfall stations in the western region, it is impossible to carry out the test. Therefore, when testing, only the accuracy of the algorithm in the eastern region is considered.

The inspection standard is that for the rainstorm cloud identified by the above algorithm, if the precipitation in one hour measured by a certain rainfall station in the area where the cloud is located exceeds 8mm at the current moment and within the next 2 hours, the identification is considered correct, otherwise the identification is considered wrong.

The specific inspection steps are as follows:

(1) Assuming the current time t, for each rainstorm cloud cluster identified by the above algorithm, read the one-hour precipitation information of all self-chemical rainfall stations in the area where the cloud cluster is located at the current time and in the next two hours, that is, if the current is 8 point, read the three one-hour precipitation information of each station from 7:00 to 8:00, from 8:00 to 9:00, and from 9:00 to 10:00;

(2) If the one-hour precipitation of a station exceeds 8mm, it is considered that the cloud cluster is indeed a rainstorm cloud cluster, and the identification is correct;

(3) If the three one-hour precipitation at all stations does not exceed 8mm, the cloud cluster is considered not to be a rainstorm cloud cluster, and the identification is wrong;

(4) Record all the correct and incorrect times of recognition, and calculate the correct rate of recognition.

As shown in Figure 5, it is a comparison chart of the recognition result at 0:00 on August 1 and the precipitation in the next 2 hours and 1 hour. Among them, the one-hour precipitation less than 5mm is divided by the boundary line of the box, the area greater than 5mm and less than 8mm is distributed at the fine points in the dot matrix area at the upper right of the box boundary, and the area greater than 8mm is distributed at the lower left of the box boundary At the large particle points in the lattice area.

5. Result inspection

A total of 738 cloud images (6 cloud images are missing) were detected from 00:00, July 1, 2010 to 23:00, July 31, 2010. A total of 1,565 rainstorm cloud clusters were detected in Sichuan, and 1,335 were correct, with a correct rate of 85.30%. A total of 6217 rainstorm cloud clusters were detected within the range (excluding the western region), and 4746 of them were correct, with a correct rate of 76.34%.

The algorithm detects a total of 238 cloud images (2 cloud images are missing) from 0:00 on August 1, 2010 to 23:00 on August 10, 2010. A total of 329 rainstorm cloud clusters were detected in Sichuan, 272 of which were correct, and the correct rate was 82.67%. A total of 2021 rainstorm cloud clusters were detected across the country (except the western region), and 1476 of them were correct, with a correct rate of 73.03%.

The above is only a preferred embodiment of the present invention, it should be pointed out that for those of ordinary skill in the art, without departing from the technical principle of the present invention, some improvements and modifications can also be made. It should also be regarded as the protection scope of the present invention.

Claims (10)

1. A method for identifying storm clouds based on stationary meteorological satellites, characterized in that, the method may further comprise the steps: Step S1: Segment the cloud image obtained by the stationary meteorological satellite to obtain the category of each cloud group at the current moment in the observation area; Step S2: According to the cloud images acquired by stationary meteorological satellites, synthesize multiple cloud images in the set time period before the current moment to obtain the basic brightness temperature map of each cloud group in the set time period, and obtain the basic brightness temperature map obtained after the synthesis. The brightness temperature map is defined as the short-term basic brightness temperature map; Step S3: Calculate the gray value difference image between the cloud image at the current moment and the short-term basic brightness temperature image; Step S4: Segment the gray value difference image to identify candidate cloud clusters that may have rainstorm weather; Step S5: For the various types of cloud clusters obtained by segmentation in step S1, combined with the candidate cloud clusters identified in step S4, use the historical sample data of the observation area obtained by stationary meteorological satellites to identify the final rainstorm cloud cluster . 2. the method for identifying storm clouds based on stationary meteorological satellites as claimed in claim 1, is characterized in that, in described step S1, specifically comprises the following steps: Step S101: Read the cloud image at the current time t and the time t-1 of the previous hour obtained by the stationary meteorological satellite, use the gray value threshold method to segment the cloud image, and divide the grid points in the cloud image into There are two categories greater than or equal to the threshold and less than the threshold, and the parts greater than or equal to the threshold are respectively recorded as point set S(t) and point set S(t-1); where point set S(t) is the cloud at the current time t Image point set, point set S(t-1) is the cloud cluster image point set at time t-1 in the last hour; Step S102: Mark the connected regions in the point set S(t) and the point set S(t-1), and record the parameters related to the cloud precipitation intensity, which specifically include: the maximum gray value of the cloud , the minimum gray value of the cloud, the area of the cloud and the change of the position of the cloud; Step S103: For each cloud in the point set S(t), by judging whether there is a corresponding source cloud in the point set S(t-1), the source cloud in the point set S(t-1) The cloud clusters in the point set S(t) are divided into ten categories based on the number of cloud clusters, whether their average gray value increases or decreases when compared with the source cloud clusters, and the change of the cloud cluster area. 3. the method for identifying storm clouds based on stationary meteorological satellites as claimed in claim 2, is characterized in that ten categories in the step S103 belong to four major categories, and the four major categories specifically include: newly added Cloud-like clusters, growth-changing cloud clusters, split-type cloud clusters, and merge-like cloud clusters; If a certain cloud group in the point set S(t) does not intersect with any cloud group in the point set S(t-1), it is considered as a newly added cloud group; If a certain cloud in the point set S(t) only intersects with a certain cloud in the point set S(t-1), it is considered that it is caused by the change of the cloud in the point set S(t-1). The growth and change class cloud clusters that come; further, according to the change situation of the cloud cluster area, the growth change class cloud clusters are also specifically divided into translational change class cloud clusters, expansion change class cloud clusters and contraction change class cloud clusters; The areas of the cloud cluster at time t-1 and time t are A _t-1 and A _t respectively, then when m ₁ *A _t-1 ≤A _t ≤n ₁ *A _t-1 , the growth and change type cloud cluster is considered It is a translational change cloud cluster; when A _t >n ₁ *A _t-1 , the growth change cloud cluster is considered to be an expansion change cloud cluster; when A _t <m ₁ *A _t-1 , the growth change cloud cluster is determined to be The changing cloud clusters are shrinking and changing cloud clusters; among them, the parameter values m ₁ and n ₁ are preset values according to the actual situation, and n ₁ >m ₁ ≥1; If multiple cloud clusters in the point set S(t) intersect with a certain cloud cluster C _j in the point set S(t-1), then these cloud clusters can be identified as all formed by the point set S(t-1 ) in the cloud group C _j in the split cloud group; further, according to the point set S(t) in the multiple cloud clusters and the area relationship of the cloud group C _j , the split cloud group is also specifically divided into growth splitting cloud clusters, common splitting cloud clusters and independent splitting cloud clusters; if the area of one of these cloud clusters intersecting C _j in the point set S(t) is A _t , the area of C _j is A _Cj , then when A _t >n ₂ *A _Cj , the split cloud cluster is considered to be a growing split cloud cluster; when m ₂ *A _Cj ≤A _t ≤n ₂ *A _Cj , the split cloud cluster is identified as The cloud group is an ordinary split cloud group; when A _t <m ₂ *A _Cj , the split cloud group is considered to be an independent split cloud group; among them, the parameter values m ₂ and n ₂ are preset according to the actual situation and n ₂ >m ₂ >0; If a certain cloud in the point set S(t) of the current cloud image intersects with multiple clouds in the point set S(t-1), then the cloud can be identified as the point set S(t-1). ) in a plurality of cloud clusters merged; further, according to whether the area of the current cloud cluster is greater than the area summation of a plurality of cloud clusters in S(t-1), the merged cloud cluster is also specific They are divided into growth-merging cloud clusters, common merger-type cloud clusters, and possible false-merger cloud clusters; if the area of the current cloud cluster is A _t , the number of n cloud clusters in the intersecting point set S(t-1) The areas are A _i , i=1, 2,...n; then when A _t > sum(A _i ), the merged cloud cluster is considered to be a growing merged cloud cluster; when max(A _i )≤A When _t ≤ sum(A _i ), the merged cloud cluster is considered as an ordinary merged cloud cluster; when A _t < max(A _i ), the merged cloud cluster is considered as a possible false merged cloud cluster. 4. the method for identifying storm clouds based on stationary meteorological satellites as claimed in claim 1, is characterized in that, in described step S2, specifically comprises: Step S201: Aligning multiple cloud images within a set time period before the current moment; Step S202: For the gray value at a certain grid point in the short-term basic brightness temperature map, select the smallest gray value among the multiple gray values at the same grid point in multiple cloud images; Step S203: For the gray values at all grid points in the short-term basic brightness temperature map, use the method in step S202 to obtain gray value values, so as to synthesize and obtain the short-term basic brightness temperature map . 5. the method for identifying storm clouds based on stationary weather satellites as claimed in claim 1, is characterized in that, in described step S3, specifically comprises: Step S301: aligning the cloud image at the current moment with the short-term basic brightness temperature image; Step S302: For the gray value at a certain grid point in the gray value difference image, subtract the short-term basic brightness temperature from the gray value at the same grid point in the cloud image at the current moment The gray value at the same grid point in the figure, the obtained difference is selected as the gray value at the grid point in the gray value difference image; if the difference is less than zero, the gray value The gray value at the grid point in the difference image is 0; Step S303: For the gray values at all the grid points in the gray value difference image, the method in step S302 is used to obtain the gray value, so as to obtain the gray value difference image. 6. The method for identifying rainstorm clouds based on stationary meteorological satellites according to claim 1, characterized in that, in the step S4, a threshold method is used to segment the gray value difference image. 7. the method for identifying storm clouds based on stationary meteorological satellites as claimed in claim 3, is characterized in that, in described step S5, for the newly-added class cloud that segmentation obtains in step S1, specifically comprises the following steps: Step S501a: judging whether it is the candidate cloud group identified in step S4, if yes, proceed to S502a, otherwise, identify it as a non-rainstorm cloud group; Step S502a: Whether the maximum gray value of the grid points it contains is greater than the preset threshold T ₁ , if yes, identify it as a rainstorm cloud cluster; otherwise, identify it as a non-rainstorm cloud cluster; where, the threshold T ₁ Based on the one-hour minimum precipitation of the rainstorm defined by meteorology, it is determined by the historical sample data in the past set months according to the minimum misjudgment probability criterion. The historical sample data includes: the maximum gray value of the cloud cluster, The minimum gray value of the cloud, the area of the cloud, and the change of the position of the cloud. 8. the method for identifying storm clouds based on stationary meteorological satellites as claimed in claim 3, is characterized in that, in described step S5, for the growth change class cloud that segmentation obtains in step S1, specifically comprises the following steps: Step S501b: judging whether it is the candidate cloud group identified in step S4, if yes, proceed to S502b or S503b, otherwise, identify it as a non-storm cloud group; Step S502b: For the cloud clusters of the translational change type and the expansion change type, if the maximum gray value of the grid points contained therein is greater than the threshold T ₂ , identify it as a rainstorm cloud cluster; otherwise, identify it as a non-rainstorm cloud cluster; Step S503b: For the shrinking and changing cloud cluster, if its area is greater than the threshold value V ₁ , identify it as a rainstorm cloud cluster; otherwise, identify it as a non-rainstorm cloud cluster; wherein, the thresholds T ₂ and V ₁ are both meteorologically defined The one-hour minimum precipitation of the rainstorm is based on the historical sample data in the past set month according to the minimum misjudgment probability criterion. The historical sample data includes: the maximum gray value of the cloud cluster, the minimum gray value of the cloud cluster value, cloud area, and cloud position changes. 9. the method for identifying storm clouds based on stationary meteorological satellites as claimed in claim 3, is characterized in that, in described step S5, for the split class cloud that segmentation obtains in step S1, specifically comprises the following steps: Step S501c: judging whether it is the candidate cloud group identified in step S4, if yes, proceed to S502c or S503c, otherwise, identify it as a non-storm cloud group; Step S502c: For the cloud clusters of growth splitting type and common splitting type, identify them as torrential rain cloud clusters; Step S503c: For the independent split cloud cluster, identify it as a non-storm cloud cluster. 10. the method for identifying storm clouds based on stationary meteorological satellites as claimed in claim 3, is characterized in that, in described step S5, for the merging class cloud that segmentation obtains in step S1, specifically comprises the steps: Step S501d: judging whether it is the candidate cloud group identified in step S4, if yes, proceed to S502d or S503d, otherwise, identify it as a non-rainstorm cloud group; Step S502d: For the cloud clusters of growth-merging type and common-merging type, identify them as torrential rain cloud clusters. Step S503d: For the possible false merged cloud cluster, if its area is smaller than the threshold value V ₂ , identify it as a non-rainstorm cloud cluster; otherwise, identify it as a rainstorm cloud cluster; wherein, the threshold V ₂ is defined by meteorology on rainstorm Based on the one-hour minimum precipitation of , it is determined by the historical sample data in the past set month according to the minimum misjudgment probability criterion. The historical sample data includes: the maximum gray value of the cloud cluster, the minimum gray value of the cloud cluster, Cloud area and cloud position changes.

Images