CN102662172A - Stormy cloud cluster extrapolation method based on Doppler radar reflectivity image - Google Patents

Abstract

The invention discloses a stormy cloud cluster extrapolation method based on a Doppler radar reflectivity image. Doppler radar reflectivity images at current time and at previous time are divided into seven layers respectively according to different reflectivity thresholds to form seven monochrome images; blocking processing is carried out on each of the monochrome images and thus the complex cloud cluster is decomposed into small block regions and large block regions; extrapolation results of all the regions at future time are respectively obtained; combination is carried out on the extrapolation results of all the regions in terms of layers so as to obtain a complete extrapolation imaged based on layers; by taking an extrapolation result of layer with a threshold of 25dBz as a base, extrapolation images of layers with higher thresholds are successively utilized to cover extrapolation images of layers with lower thresholds and at last, a region that has extremely high reflectivity and has an area less than 50 units is used for final coverage, so that an extrapolation image of the stormy cloud cluster is obtained. The experiments proves that the stormy cloud cluster that is extrapolated by the method is capable of embodying important information about the structure and the form, so that an extrapolation coincidence rate of a typical stormy cloud cluster is obviously higher than those in other existing methods.

Description

A kind of Extrapolation method of the storm cloud cluster based on the radar Doppler albedo image

Technical field

The present invention relates to meteorological field, particularly a kind of Extrapolation method of the storm cloud cluster based on radar Doppler.

Background technology

Human after getting into for 20th century, through obtaining the tracking of storm cloud cluster, the achievement in research of discerning and forecasting automatically based on the albedo image of Doppler radar.Become one type of popular and method easily based on the cloud cluster of extrapolation technique, wherein, monomer centroid method and cross-correlation technique are to use two kinds of methods the most widely.The track algorithm TREC that Rinehart proposes the earliest derives from from cross-correlation technique; Passed through the unremitting effort of some scholars such as Li, Lai and Wang Gaili etc. afterwards again; The TREC algorithm has obtained updating and is perfect; The basic way of TREC algorithm is the related coefficient of at first calculating between the radar return image zonule of adjacent moment cloud cluster, under the maximal correlation condition, obtains the mobile vector of each local cloud cluster echo, and then carries out outside forecast; Another kind of monomer centroid method is from being suggested so far; Same constantly perfect and development through numerous scholars; Present widely used TITAN and SCIT algorithm have been proposed; The basic way of these two kinds of algorithms is to identify on the storm monomer basis of (only containing a nuclear district), carries out the monomer coupling of adjacent moment, and through the adjacent matching result of carving is for a long time extrapolated.

The step that the TREC algorithm is followed the trail of the storm monomer is: go on foot the storm monomer that identifies through last one; Calculate the centroid position of two groups of cloud clusters of adjacent moment; Utilize maximal rate to control; Adopt maximum correlation coefficient to confirm the mobile vector of adjacent moment cloud cluster, predict the following centroid position constantly of corresponding monomer through the mobile vector that obtains.

The TITAN algorithm is followed the tracks of the barycenter of meteorological cloud cluster through the combinatorial optimization algorithm, describes the division and the merging of meteorological cloud cluster through geometric processing.The core concept of this algorithm is: similitude and distance based on meteorological cloud cluster current time are matched, and the similitude of cloud cluster is high more, and the distance of cloud cluster is more preferentially matched at a distance of near more.

The SCIT algorithm comprises three parts: cloud cluster identification, form coupling and position prediction.Will do matching treatment to it after having discerned cloud cluster: be the basis with the current time; Seek the corresponding with it cloud cluster of previous moment; Seek its track; On the basis of working in front, utilize the current motion vector of " historical position " match cloud cluster, this motion vector is used to do the following center of mass motion track constantly of cloud cluster.

The inventor is in realizing process of the present invention; Not enough below finding to exist at least in the prior art: the radar Doppler albedo image of strong convective weathers such as the hail in the meteorology, wind spout and heavy rain is the equal important information of carrier on inner structure and formalness, and these information can help the type of people's discrimination storm cloud cluster and storm cloud cluster.And existing extrapolation technique stresses is the extrapolation of cloud cluster position, in the knot fruit, the information of cloud cluster structure and cloud cluster form exactly can not be provided outside.

Summary of the invention

The invention provides a kind of storm cloud cluster Extrapolation method based on the radar Doppler albedo image; The present invention is on the basis of cloud cluster position extrapolation; The important form and the structural information of storm cloud cluster of extrapolation can be provided; Enriched the useful and important forecast information of extrapolation cloud cluster, filled up a blank can not releasing the storm cloud cluster that has structure and shape information outward, seen hereinafter for details and describe:

A kind of Extrapolation method of the storm cloud cluster based on the radar Doppler albedo image said method comprising the steps of:

(1) storm cloud cluster radar Doppler albedo image is carried out layering and decompose, obtain 7 monochromatic subimages of storm cloud cluster based on the different reflectivity threshold value;

(2) each Zhang Danse subimage of storm cloud cluster is carried out piecemeal and handle that to be decomposed into pocket regional with bulk; Wherein, with area in the storm cloud cluster less than 10 ⁴Regional area be defined as said pocket; With area more than or equal to 10 ⁴Regional area be defined as said bulk zone;

(3) adopt the linear extrapolation revised law to obtain the extrapolation result of said pocket and said bulk zone barycenter;

(4) according to the extrapolation result of said pocket barycenter, each said pocket is adopted the form extrapolation method based on scattering model, obtain the extrapolation result of said pocket local configuration;

(5) according to the extrapolation result of said bulk zone barycenter, the local fan-shaped edge to each said bulk zone adopts the form extrapolation method based on mathematical morphology, obtains the extrapolation result of said bulk Regional Bureau contouring;

(6) the extrapolation result to the extrapolation result of every layer said pocket local configuration and said bulk Regional Bureau contouring merges, and obtains the integral body extrapolation image of each monochromatic subimage;

(7) be that the integral body extrapolation result of the monochromatic subimage of 25dBz is substrate with the reflectivity threshold value; With the integral body extrapolation result of the monochromatic subimage of high reflectance threshold value the integral body extrapolation result of the monochromatic subimage of antiradar reflectivity threshold value is covered successively; Again high reflectivity regions is finally covered, obtain the extrapolation image of storm cloud cluster in the radar Doppler albedo image; Wherein, smaller or equal to 50 units, the reflectivity threshold value is defined as high reflectivity regions greater than the reflectivity regions of 55dBz with area;

(8) next one is regarded as current time constantly, repeats the step of (1) to (7), realize lasting extrapolation, obtain the extrapolation image of storm cloud cluster in the farther moment to the storm cloud cluster.

Saidly storm cloud cluster radar Doppler albedo image is carried out layering decompose and to be specially:

To current time t _iWith previous moment t _I-1The radar Doppler albedo image of storm cloud cluster is divided into 7 figure layers according to different reflectivity threshold values respectively, i figure layer T _RiGenerative process following, if storm cloud cluster zone in reflectance value f _{T (p)}>=R _iDBz then makes it at T _RiIn value f _{Tt (p)}=R _iDBz, otherwise make f _{Tt (p)}Be background colour, wherein R _i=25,30,35,40,45,50,55dBz.

Said each Zhang Danse subimage to the storm cloud cluster carries out piecemeal to be handled and to be decomposed into pocket and the bulk zone is specially:

(1) from the barycenter O of monochromatic subimage to border injection line;

(2) be that step-length rotates counterclockwise ray with 1 °, when ray and at least 3 intersection points of image outline formation, note first intersection point A of this ray and image outline;

(3) in the B point, AB is divided into two image to reverse extending AO intersection graph with line segment as profile, obtains two sub regions;

(4) for each subregion, repeat (1) to (3) process, until ray and profile intersection point situation no longer occurring greater than 1, with area less than 10 ⁴Subregion as said pocket; With area more than or equal to 10 ⁴Subregion as said bulk zone, piecemeal is finished dealing with.

The extrapolation result that said employing linear extrapolation revised law obtains said pocket and said bulk zone barycenter is specially:

If current time is t _i, through t _I-2Barycenter and t constantly _I-1Barycenter linear extrapolation constantly obtains t _iExtrapolation barycenter constantly calculates t _iExtrapolation barycenter constantly is a displacement vector with said offset tag, through t to the side-play amount of ti actual barycenter constantly _I-1The moment and t _iBarycenter linear extrapolation constantly obtains t _I+1Centroid position constantly, through said displacement vector to said t _I+1Centroid position is constantly revised.

Said extrapolation result according to said pocket barycenter adopts the form extrapolation method based on scattering model to each said pocket, and the extrapolation result who obtains said pocket local configuration is specially:

1) evenly scatters 24 rays from the pocket barycenter, form 24 intersection points with the pocket border;

2) move previous moment pocket F _I-1Up to barycenter and current time pocket F _iBarycenter overlap;

3) with said t _I+1Barycenter is as F constantly _I+1Barycenter, establish F _I-1And F _iWith ray l _i(i=1 ..., 4 (n _t-1) intersection point) is

With

For all intersection points of falling on the current monochromatic subimage profile, calculate

Intersection point for falling on the Region Segmentation line makes Δ _i=0 push away pocket at t _I+14 (n constantly _t-1) individual frontier point $p_{{\mathrm{Fl}}_i}\left(t_{i+1}\right)=p_{{\mathrm{Fl}}_i}\left(t_i\right)+{\mathrm{\Δ}}_i;$

4) with 4 (n _t-1) individual frontier point

Connect into polygon as angle point, and polygon is carried out pixel fill, obtain the extrapolation image F of pocket _I+1

5) move extrapolation image F _I+1Up to barycenter and said t _I+1Extrapolation barycenter constantly overlaps.

Said extrapolation result according to said bulk zone barycenter, the local fan-shaped edge to each said bulk zone adopts the form extrapolation method based on mathematical morphology, obtains the extrapolation result of said bulk Regional Bureau contouring;

1) moves previous moment bulk zone F _I-1Up to barycenter p _c(t _I-1) and current time bulk zone F _iBarycenter p _c(t _i) overlap;

2) stretching, extension or retraction zone are detected;

If the bulk cloud cluster of adjacent moment zone G _I-1And G _iFrontier point set be respectively P _B1={ p _B1And P (i) } _B2={ p _B2(i) }; Make that bulk regional background point value is 0, the object-point value is 1, by G _I-1And G _iComposograph G ' makes each pixel value of G ':

$f^{\′}\left(p_i\right)=f_{t_i}\left(p_i\right)-f_{t_{i-1}}\left(p_i\right)=\left\{\begin{array}{c}1\\ 0\\ -1\end{array}\right.$

Definition f ' (p _iThe point p of)=1 _iFor stretching point, f ' (p _iThe point p of)=-1 _iBe retraction point, f ' (p _iThe point p of)=0 _iBe invariant point, stretching, extension point and retraction Vertex Coloring with image obtain stretching area S _Ext, retraction district S _RetAnd constant region S _Inv

3) sector region is divided;

(1) search is stretched or the regional critical point set

that bounces back

To p _B1(i) ∈ P _B1And p _B2(i) ∈ P _B2, if p _B1(i) ∈ P _B2Set up, then p _B1(i) ∈ P _lIf p _B2(i) ∈ P _B1Set up, then p _B2(i) ∈ P _l

(2) from barycenter p _cObtain a plurality of sector region AS to each critical point line _i, i=1,2 ..., each sector region comprises a stretching area or a retraction district, and the sector region that will comprise the stretching area is defined as and stretches the fan section, and the sector region that will comprise the retraction district is defined as the retraction fan section;

4) corrosion expansion process;

Each is stretched fan section AS _iCarry out differential expansion and handle, to each retraction fan section AS _iCarrying out local corrosion handles;

(1) to said stretching, extension fan section AS _iTwo ray limits make to forbid marks for treatment;

(2) read expansion or corrosion information through dye marker;

(3) to said stretching area S _ExtAverage the calculating of span, to said retraction district S _RetAverage the calculating of amount of recovery;

If S _iFor detected certain stretching area or retraction district, if satisfy pixel (p ∈ S _i) ∩ (p ∈ P _B1) count and be N ₁, satisfy (p ∈ S _i) ∩ (p ∈ P _B2) count and be N ₂, satisfy p ∈ S _iCount and be N, then region S _iAverage span or amount of contraction D be:

$D=\frac{N}{0.5(N_1+N_2)}$

In the formula, denominator is a region S _iThe outer edge average length;

(4) use yardstick as the circular configuration element of d to retraction sector region corrosion n time, to stretching, extension sector region expansion n time, obtain a plurality of new sector region AS ' _i, wherein

$d=\left\{\begin{array}{cc}D,& D\≤2\\ 2,& D>2\end{array}\right.$ $n=\mathrm{int}(\frac{D}{2}+0.5)$

In the formula, int representes rounding operation;

5) zone is synthetic;

To a plurality of new sector region AS ' _iWith all constant region S _InvCarry out cup, splice by original relative position with the zone that remains unchanged in the zone that is about to handle, and obtains G _iExtrapolation image G _I+1

6) displacement;

Move image G _I+1Up to barycenter and said t _I+1Extrapolation barycenter constantly overlaps.

The beneficial effect of technical scheme provided by the invention is: this method is from the level formation characteristics of storm cloud cluster; Through being deep into the internal layer of cloud cluster; Utilization is based on the form Extrapolation method of scattering model with based on the form Extrapolation method of mathematical morphology; The preference policy and the subimage synthetic schemes that cooperate two kinds of methods; Realized keeping the extrapolation of detailed information such as the inner precipitation particles distribution of storm cloud cluster and each layer of cloud cluster border motif information; Filled up a blank can not releasing the storm cloud cluster that has structure and shape information outward, the method can adapt to objective differentiation phenomenons such as the division of storm cloud cluster, merging simultaneously, is of value to the forecast accuracy of strengthening the storm cloud cluster.

Description of drawings

Fig. 1 is the process flow diagram of a kind of storm cloud cluster Extrapolation method based on the radar Doppler albedo image provided by the invention;

Fig. 2 is the layering decomposing schematic representation of storm cloud cluster provided by the invention;

Fig. 3 is that the piecemeal of storm cloud cluster layered image provided by the invention is handled synoptic diagram;

Fig. 4-1, Fig. 4-2, Fig. 4-3 and Fig. 4-4 are the linear extrapolation revised law synoptic diagram of regional barycenter provided by the invention;

Fig. 5 is a scattering form Extrapolation method principle schematic provided by the invention;

Fig. 6 is the pocket extrapolation exemplary plot based on scattering model provided by the invention;

Fig. 7 is the form extrapolation method principle schematic of mathematical morphology provided by the invention;

Fig. 8 is the big zone extrapolation exemplary plot of the form extrapolation method based on mathematical morphology provided by the invention;

Fig. 9 is the synoptic diagram of 6 minutes test sample of extrapolation provided by the invention;

Figure 10 is the synoptic diagram of 6～18 minutes test sample of extrapolation provided by the invention.

Embodiment

For making the object of the invention, technical scheme and advantage clearer, embodiment of the present invention is done to describe in detail further below in conjunction with accompanying drawing.

The problem that people more are concerned about with regard to forecast is how on the radar Doppler image, accurately to predict moving and variation tendency of strong convection monomer cloud cluster; Problem comprises the cloud cluster processing of different appearance profiles, the centroid position in cloud cluster moment from the current time to the future and appearance profile evolutionary process etc.Small and Medium Sized diastrous weathers such as hail, hurricane line, wind spout and thunderstorm have take place to develop fast, move rapidly unusual and characteristics such as destructive power is huge; Wanting it is analyzed exactly and predicts is the difficult point problem that needs to be resolved hurrily in the weather forecast, also is simultaneously the problem that presses for solution in the diastrous weather prediction.In order to improve the accuracy of cloud cluster extrapolation, reduce rate of false alarm, the embodiment of the invention provides a kind of storm cloud cluster Extrapolation method based on radar Doppler, referring to Fig. 1, sees hereinafter for details and describes:

101: on the basis that utilizes the storm recognition technology, the albedo image of storm cloud cluster radar Doppler albedo image is carried out layering decompose, obtain 7 monochromatic subimages of storm cloud cluster, be i.e. 7 of the storm cloud cluster figure layers based on the different reflectivity threshold value;

Wherein, this step is specially:

Referring to Fig. 2,16 kinds of pseudo-colourss in the radar Doppler albedo image respectively with black background, 5dBZ, 10dBZ ..., the reflectivity intensity of 65dBZ is corresponding.And the outermost layer reflectivity intensity of storm cloud cluster is generally always greater than 25dBZ; And ecto-entad reflectivity intensity is elevated to more than the 55dBZ step by step; Area is corresponding to be reduced step by step; So this method is visualized as the two dimensional image of cloud cluster on the zone that the outer boundary by the 25dBz zone surrounds and constantly is superimposed with the more result of high reflectivity regions, forms 7 layers of branch layered scheme.To i figure layer, the reflectance value of subject area is " also " less than the zone of threshold value Ri of background area originally and reflectance value more than or equal to threshold value Ri, background area, and therefore, each figure layer all is a monochrome image.To any figure layer T _Ri, if value is R _iConnected domain area S≤50 of dBz are then with the direct result as extrapolation of this connected domain.Wherein, each figure layer T of storm cloud cluster _RiThe generation method following:

To current time t _iWith previous moment t _I-1The radar Doppler albedo image of storm cloud cluster is divided into 7 figure layers according to different reflectivity threshold values respectively, at i figure layer T _RiIn, if the reflectance value f in the storm cloud cluster zone _{T (p)}>=R _iDBz then makes it at T _RiIn value f _{Tt (p)}=R _iDBz, otherwise make f _{Tt (p)}Be background colour, wherein R _i=25,30,35,40,45,50,55dBz.

102: each Zhang Danse subimage of storm cloud cluster is carried out piecemeal handle that to be decomposed into pocket regional with bulk;

With area in the storm cloud cluster less than 10 ⁴Regional area be defined as pocket, area more than or equal to 10 ⁴Regional area be defined as the bulk zone.

Wherein, this step is specially:

(1) from the barycenter O of monochromatic subimage to border injection line;

(2) be that step-length rotates counterclockwise ray with 1 °, when ray and at least 3 intersection points of image outline formation, note first intersection point A of this ray and image outline;

(3) in the B point, AB is divided into two image to reverse extending AO intersection graph with line segment as profile, obtains two sub regions;

(4) for each subregion, repeat (1) to (3) process, until ray and profile intersection point situation no longer occurring greater than 1, with area less than 10 ⁴Subregion as pocket; With area more than or equal to 10 ⁴Subregion as bulk zone, piecemeal is finished dealing with.

Referring to Fig. 3, establishing fine line is the exact shape profile of cloud cluster, O ₁The barycenter of cloud cluster for this reason, ray O ₁A is that article one and profile intersection point become a plurality of rays by one, notes the A point, reverse extending AO ₁, meeting at B with profile, AB is divided into 2 to cloud cluster, wherein S ₁All barycenter scattered rays all only and S ₁Profile has an intersection point, need not divide again.Another piece then also will continue to be divided into 2 parts could satisfy all barycenter scattered rays all only have an intersection point with its profile condition, so this cloud cluster has been divided into S ₁, S ₂, S ₃Three zones.Every zone is all surrounded by monochromatic subimage profile and Region Segmentation line.

103: adopt the linear extrapolation revised law to obtain the extrapolation result of pocket and bulk zone barycenter;

Wherein, this step is specially:

If current time is t _i, through t _I-2Barycenter and t constantly _I-1Barycenter linear extrapolation constantly obtains t _iExtrapolation barycenter constantly calculates t _iExtrapolation barycenter constantly is to t _iThe side-play amount of actual barycenter constantly is a displacement vector with offset tag, through t _I-1The moment and t _iBarycenter linear extrapolation constantly obtains t _I+1Centroid position constantly, through displacement vector to t _I+1Centroid position is constantly revised.

Referring to Fig. 4, the solid-line curve among Fig. 4-1,4-2 and the 4-3 is represented the profile that the storm cloud cluster is actual, and some D, some A and some B are their barycenter,

Be at t _I-1Constantly push away moment t _iThe displacement of barycenter A, the C point is t _iThe physical location of moment cloud cluster barycenter,

Be the displacement vector of the reality of barycenter A, and

Be predicated error, will

As the amount of correcting of linear extrapolation to t _I+1Predicting the outcome constantly revised.Fig. 4-4 has provided two kinds of barycenter extrapolation signals, and some O is the linear extrapolation result of barycenter C, and some R is the extrapolation result through revising on the basis of linear extrapolation, and correction is exactly at t _iThe predicated error that obtains constantly, i.e. t _iCloud cluster barycenter C is to t constantly _I+1Displacement constantly $\stackrel{\→}{\mathrm{CR}}=\stackrel{\→}{\mathrm{CO}}+\stackrel{\→}{\mathrm{OR}},$ Wherein, $\stackrel{\→}{\mathrm{CO}}=\stackrel{\→}{\mathrm{AC}},\stackrel{\→}{\mathrm{OR}}=\stackrel{\→}{\mathrm{BC}}.$

104: according to the extrapolation result of pocket barycenter, each pocket is adopted the form extrapolation method based on scattering model, obtain the extrapolation result of pocket local configuration;

Wherein, this step is specially:

1) evenly scatters 24 rays from the pocket barycenter, form 24 intersection points with the pocket border;

2) move previous moment pocket F _I-1Up to barycenter and current time pocket F _iBarycenter overlap;

3) with t _I+1Barycenter is as F constantly _I+1Barycenter, establish F _I-1And F _iWith ray l _i(i=1 ..., 4 (n _t-1) intersection point) is

With For all intersection points of falling on the current monochromatic subimage profile, calculate

Intersection point for falling on the Region Segmentation line makes Δ _i=0 push away pocket at t _I+14 (n constantly _t-1) individual frontier point $p_{{\mathrm{Fl}}_i}\left(t_{i+1}\right)=p_{{\mathrm{Fl}}_i}\left(t_i\right)+{\mathrm{\Δ}}_i;$

4) with 4 (n _t-1) individual frontier point

Connect into polygon as angle point, and polygon is carried out pixel fill, obtain the extrapolation image F of pocket _I+1

5) move extrapolation image F _I+1Up to barycenter and t _I+1Extrapolation barycenter constantly overlaps.

Referring to Fig. 5, thicker solid line (t _I+1Constantly) be utilize other 2 than fine line (t _I-1The moment and t _iConstantly) according to the above-mentioned pocket profile that obtains based on the form extrapolation method of scattering model at (t constantly in future _I+1Extrapolation result constantly).

It is generally acknowledged that the area of storm cloud cluster nucleus on the radar Doppler reflectance map is not less than 50 pixels (pixel is apart from being 1km), therefore gets scattering model parameter n _t=7, number of rays then is 24.For the less cloud cluster of area; Through 24 rays are linked to each other as angle point with 24 intersection points of region contour successively; Can satisfy the appearance profile of sketching the contours of small size zone under the condition of certain precision, and for the bigger cloud cluster of area, can be very coarse through the extrapolation profile that connects the zone that 24 angle points obtaining through scattering model obtain successively; Tracing it to its cause is the increase along with region area; The distance of adjacent point-to-point transmission increases in 24 angle points, and for the large area region of band shape, each angle point has very big range difference to the distance of regional barycenter; These range differences are difficult to satisfy precision of prediction and couple together the point on an one magnitude not simply not on an one magnitude.Obviously, extrapolate and to become coarse along with the increase of cloud cluster area, so this method is suitable for the form extrapolation of pocket based on the cloud cluster form of scattering model.

Referring to Fig. 6, wherein, figure a, figure b, figure c are respectively certain cloudling group zone in t _I-2T constantly, _I-1The moment and t _iThis cloud cluster zone that real image constantly, figure d obtain for the form extrapolation method based on scattering model is at t _iExtrapolation image constantly; Can find out; Behind the enforcement scattering model of the less zone of area (figure Smalt zone); Region contour predict the outcome with actual conditions at form form basically identical, come the following form of prediction area so can utilize than the zonule based on the form extrapolation method of scattering model.

105: according to the extrapolation result of bulk zone barycenter, the local fan-shaped edge to each bulk zone adopts the form extrapolation method based on mathematical morphology, obtains the extrapolation result of bulk Regional Bureau contouring;

Wherein, this step is specially:

1) moves previous moment bulk zone F _I-1Up to barycenter p _c(t _I-1) and current time bulk zone F _iBarycenter p _c(t _i) overlap;

2) stretching, extension or retraction zone are detected;

If the bulk cloud cluster of adjacent moment zone G _I-1And G _iFrontier point set be respectively P _B1={ p _B1And P (i) } _B2={ p _B2(i) }; Make that bulk regional background point value is 0, the object-point value is 1, by G _I-1And G _iComposograph G ' makes each pixel value of G ':

$f^{\′}\left(p_i\right)=f_{t_i}\left(p_i\right)-f_{t_{i-1}}\left(p_i\right)=\left\{\begin{array}{c}1\\ 0\\ -1\end{array}\right.$

Definition f ' (p _iThe point p of)=1 _iFor stretching point, f ' (p _iThe point p of)=-1 _iBe retraction point, f ' (p _iThe point p of)=0 _iBe invariant point, stretching, extension point and retraction Vertex Coloring with image obtain stretching area S _Ext, retraction district S _RetAnd constant region S _Inv

3) sector region is divided;

(1) search is stretched or the regional critical point set

that bounces back

To p _B1(i) ∈ P _B1And p _B2(i) ∈ P _B2, if p _B1(i) ∈ P _B2Set up, then p _B1(i) ∈ P _lIf p _B2(i) ∈ P _B1Set up, then p _B2(i) ∈ P _l

(2) from barycenter p _cObtain a plurality of sector region AS to each critical point line _i, i=1,2 ..., each sector region comprises sector region that a stretching area or one retraction district will comprise the stretching area and is defined as and stretches the fan section, and the sector region that will comprise the retraction district is defined as the retraction fan section;

4) corrosion expansion process;

Each is stretched fan section AS _iCarry out differential expansion and handle, to each retraction fan section AS _iCarrying out local corrosion handles;

(1) to stretching fan section AS _iTwo ray limits make to forbid marks for treatment;

(2) read expansion or corrosion information through dye marker;

(3) to stretching area S _ExtAverage the calculating of span, to retraction district S _RetAverage the calculating of amount of recovery;

If S _iFor detected certain stretching area or retraction district, if satisfy pixel (p ∈ S _i) ∩ (p ∈ P _B1) count and be N ₁, satisfy (p ∈ S _i) ∩ (p ∈ P _B2) count and be N ₂, satisfy p ∈ S _iCount and be N, then region S _iAverage span or amount of contraction D be:

$D=\frac{N}{0.5(N_1+N_2)}$

In the formula, denominator is a region S _iThe outer edge average length;

(4) use yardstick as the circular configuration element of d to retraction sector region corrosion n time, to stretching, extension sector region expansion n time, obtain a plurality of new sector region AS ' _i, wherein

$d=\left\{\begin{array}{cc}D,& D\≤2\\ 2,& D>2\end{array}\right.$ $n=\mathrm{int}(\frac{D}{2}+0.5)$

In the formula, int representes rounding operation;

5) zone is synthetic;

To a plurality of new sector region AS ' _iWith all constant region S _InvCarry out cup, splice by original relative position with the zone that remains unchanged in the zone that is about to handle, and obtains G _iExtrapolation image G _I+1

6) displacement;

Move image G _I+1Up to barycenter and said t _I+1Extrapolation barycenter constantly overlaps.

Referring to Fig. 7, F _T1And F _T2Represent the real image of certain cloud cluster zone, move F in two adjacent moment _T1Their barycenter is overlapped, obtain containing the F of stretching area, retraction district, constant region _T12, wherein, the fan section that has the linear shadow district is " stretching, extension fan section ", and the fan section that has solid shaded is " retraction fan section ", and the sector region of losing the border is a constant region.

Referring to Fig. 8, wherein, figure a, figure b, figure c are respectively certain big cloud cluster zone in t _I-2T constantly, _I-1The moment and t _iThis cloud cluster zone that real image constantly, figure d obtain for the form extrapolation method based on mathematical morphology is at t _iExtrapolation image constantly can be found out, figure d is higher to the form degree of fitting of figure c.For area bigger bulk cloud cluster zone, particularly bulk belt-like zone, be suitable for the contour prediction in bulk cloud cluster zone based on the form extrapolation method of mathematical morphology.

Wherein, the embodiment of the invention does not limit the execution sequence of step 104 and step 105.

106: the extrapolation result of every layer pocket local configuration and the extrapolation result of bulk Regional Bureau contouring are merged, obtain the integral body extrapolation result of each monochromatic subimage;

107: with the reflectivity threshold value is that the integral body extrapolation result of the monochromatic subimage of 25dBz is substrate; With the integral body extrapolation result of the monochromatic subimage of high reflectance threshold value the integral body extrapolation result of the monochromatic subimage of antiradar reflectivity threshold value is covered successively; High reflectivity regions is finally covered, obtain the extrapolation image of storm cloud cluster in the radar Doppler albedo image;

Wherein, smaller or equal to 50 units, the reflectivity threshold value is defined as high reflectivity regions greater than the reflectivity regions of 55dBz with area.

108: the next one is regarded as current time constantly, and repeating step 101-step 107 realizes the lasting extrapolation to the storm cloud cluster, obtains the extrapolation image of storm cloud cluster in the farther moment.

Wherein, utilizing the time interval is two real image T of 6 minutes _I-1And T _iOrganize the scheme of laddering lasting extrapolation to be: to use T _I-1And T _iRelease is apart from T _iImage T after 6 minutes _I+1, use T _iAnd T _I+1Release is apart from T _I+1Image T after 6 minutes _I+2, by that analogy.

The feasibility of a kind of storm cloud cluster Extrapolation method based on radar Doppler of verifying that the embodiment of the invention provides with a concrete test below, see hereinafter for details and describe:

With the radar base data is data source, gets 100 albedo images that are multinuclear band shape, multinuclear bulk and each two storm process of monokaryon bulk and makees sample, this method is tested, and make similarity assessment, and evaluation index designs as follows.

If t ₃The true picture of cloud cluster is F (t constantly ₃), the extrapolation image is F ¹²(t ₃), their nuclear district area is respectively S _h(t ₃) and S _h ¹²(t ₃), the region area when descending successively from examining district's beginning intensity is respectively S ₅₀(t ₃), S ₄₅(t ₃) ..., S ₂₅(t ₃) and S ₅₀ ¹²(t ₃), S ₄₅ ¹²(t ₃) ..., S ₂₅ ¹²(t ₃), note by abridging and be S ₆, S ₅..., S ₁And S ₆ ¹², S ₅ ¹²..., S ₁ ¹²

(1) nuclear district area similarity

${\mathrm{\ρ}}_1=\frac{\min \{S_h\left(t_3\right),{S_h}^{12}\left(t_3\right)\}}{\max \{S_h\left(t_3\right),{S_h}^{12}\left(t_3\right)\}}$

(2) cloud cluster area similarity

${\mathrm{\ρ}}_2=\frac{1}{n}\underset{i=1}{\overset{n}{\mathrm{\Σ}}}\frac{\min \{S_i,S_i^{12}\}}{\max \{S_i,S_i^{12}\}},n\≤6$

Be positioned at F simultaneously if satisfy the some p of strength condition " R (p)>=i " ¹²With the quantity among the F be n _i, only be positioned at F ¹²Or the quantity among the F is n ' _i, then

(3) high intensity region shape similarity

${\mathrm{\ρ}}_3=\underset{i=\mathrm{40,45,50},\≥55\mathrm{dbz}}{\mathrm{\Σ}}\frac{n_i}{n_i+n_i^{\′}}$

(4) hypo-intense regions shape similarity

${\mathrm{\ρ}}_4=\underset{i=\mathrm{25,30,35}\mathrm{dbz}}{\mathrm{\Σ}}\frac{n_i}{n_i+n_i^{\′}}$

(5) comprehensive similarity

With ρ _i, i=1 ..., 4 carry out weighted sum, the comprehensive similarity ρ of the cloud cluster that obtains extrapolating:

$\mathrm{\ρ}=\underset{i=1}{\overset{4}{\mathrm{\Σ}}}{k}_i{\mathrm{\ρ}}_i,$ $\underset{i=1}{\overset{4}{\mathrm{\Σ}}}{k}_i=1$

Consider that the availability of the high reflectivity region domain information of storm cloud cluster always is better than the information availability of low reflectivity regions, gets k={k especially ₁, k ₂, k ₃, k ₄}={ 0.4,0.2,0.3,0.1}.

Referring to Fig. 9; It has provided the test sample of extrapolating 6 minutes, wherein, and with regard to the zone of 35dBz; The sub-yellow area of figure (a) upside develops into figure (c) and constantly with the below yellow area merging has taken place; With regard to the zone of 50dBz, figure red subregion in the middle of (c) again from figure (a) constantly the below red area split off, the division in above-mentioned this cloud cluster zone be incorporated in 6 minutes and all obtained reflection more accurately among the extrapolation figure.

Referring to Figure 10, this is the test sample that another one was extrapolated 6～18 minutes preferably.Table 1 has provided the comprehensive assessment result based on extrapolation in 6 minutes of test sample book, and table 2 is based on extrapolations in 6～18 minutes of test sample book and assesses the table of comparisons and average operating time thereof.

The evaluation index of table 1 pair extrapolation in 6 minutes

The evaluation index of table 2 pair extrapolation in 6～18 minutes

The extrapolation time	6min	12min	18min
				Average similarity	0.94	0.90	0.81
Average operating time	870ms	1.51s	2.43s

With regard to test sample book, the average similarity of its 6 minutes extrapolate image and true pictures surpasses 92%, and along with the increasing of extrapolation time, average similarity decreases, but 18 minutes average similarity has been verified the feasibility of this method still greater than 80%.

In sum; The embodiment of the invention provides a kind of Extrapolation method of the storm cloud cluster based on radar Doppler; This method is on the basis of it being carried out the hierarchical block decomposition; Proposed to cooperate the preference policy and the subimage synthetic schemes of two kinds of methods, realized keeping the extrapolation of detailed information such as the inner precipitation particles distribution of storm cloud cluster and each layer of cloud cluster border motif information based on the form Extrapolation method of scattering model with based on the form Extrapolation method of mathematical morphology; Filled up the blank that outer release has the storm cloud cluster of structure and shape information, the method can adapt to objective differentiation phenomenons such as the division of storm cloud cluster, merging simultaneously; And has higher similarity through the extrapolation cloud cluster that this method obtains through experimental verification.

It will be appreciated by those skilled in the art that accompanying drawing is the synoptic diagram of a preferred embodiment, the invention described above embodiment sequence number is not represented the quality of embodiment just to description.

The above is merely preferred embodiment of the present invention, and is in order to restriction the present invention, not all within spirit of the present invention and principle, any modification of being done, is equal to replacement, improvement etc., all should be included within protection scope of the present invention.

Claims (6)

1. the Extrapolation method based on the storm cloud cluster of radar Doppler albedo image is characterized in that, said method comprising the steps of:

(1) storm cloud cluster radar Doppler albedo image is carried out layering and decompose, obtain 7 monochromatic subimages of storm cloud cluster based on the different reflectivity threshold value;

(2) each Zhang Danse subimage of storm cloud cluster is carried out piecemeal and handle that to be decomposed into pocket regional with bulk; Wherein, with area in the storm cloud cluster less than 10 ⁴Regional area be defined as said pocket; With area more than or equal to 10 ⁴Regional area be defined as said bulk zone;

(3) adopt the linear extrapolation revised law to obtain the extrapolation result of said pocket and said bulk zone barycenter;

(4) according to the extrapolation result of said pocket barycenter, each said pocket is adopted the form extrapolation method based on scattering model, obtain the extrapolation result of said pocket local configuration;

(5) according to the extrapolation result of said bulk zone barycenter, the local fan-shaped edge to each said bulk zone adopts the form extrapolation method based on mathematical morphology, obtains the extrapolation result of said bulk Regional Bureau contouring;

(6) the extrapolation result to the extrapolation result of every layer said pocket local configuration and said bulk Regional Bureau contouring merges, and obtains the integral body extrapolation image of each monochromatic subimage;

(7) be that the integral body extrapolation result of the monochromatic subimage of 25dBz is substrate with the reflectivity threshold value; With the integral body extrapolation result of the monochromatic subimage of high reflectance threshold value the integral body extrapolation result of the monochromatic subimage of antiradar reflectivity threshold value is covered successively; Again high reflectivity regions is finally covered, obtain the extrapolation image of storm cloud cluster in the radar Doppler albedo image; Wherein, smaller or equal to 50 units, the reflectivity threshold value is defined as high reflectivity regions greater than the reflectivity regions of 55dBz with area;

(8) next one is regarded as current time constantly, repeats the step of (1) to (7), realize lasting extrapolation, obtain the extrapolation image of storm cloud cluster in the farther moment to the storm cloud cluster.

2. the Extrapolation method of a kind of storm cloud cluster based on the radar Doppler albedo image according to claim 1 is characterized in that, saidly storm cloud cluster radar Doppler albedo image is carried out layering decomposes and is specially:

To current time t _iWith previous moment t _I-1The radar Doppler albedo image of storm cloud cluster is divided into 7 figure layers according to different reflectivity threshold values respectively, i figure layer T _RiGenerative process following, if storm cloud cluster zone in reflectance value f _{T (p)}>=R _iDBz then makes it at T _RiIn value f _{Tt (p)}=R _iDBz, otherwise make f _{Tt (p)}Be background colour, wherein R _i=25,30,35,40,45,50,55dBz.

3. the Extrapolation method of a kind of storm cloud cluster based on the radar Doppler albedo image according to claim 1 is characterized in that, said each Zhang Danse subimage to the storm cloud cluster carries out piecemeal to be handled and be decomposed into pocket and the bulk zone is specially:

(1) from the barycenter O of monochromatic subimage to border injection line;

(2) be that step-length rotates counterclockwise ray with 1 °, when ray and at least 3 intersection points of image outline formation, note first intersection point A of this ray and image outline;

(3) in the B point, AB is divided into two image to reverse extending AO intersection graph with line segment as profile, obtains two sub regions;

(4) for each subregion, repeat (1) to (3) process, until ray and profile intersection point situation no longer occurring greater than 1, with area less than 10 ⁴Subregion as said pocket; With area more than or equal to 10 ⁴Subregion as said bulk zone, piecemeal is finished dealing with.

4. the Extrapolation method of a kind of storm cloud cluster based on the radar Doppler albedo image according to claim 1 is characterized in that, the extrapolation result that said employing linear extrapolation revised law obtains said pocket and said bulk zone barycenter is specially:

If current time is t _i, through t _I-2Barycenter and t constantly _I-1Barycenter linear extrapolation constantly obtains t _iExtrapolation barycenter constantly calculates t _iExtrapolation barycenter constantly is to t _iThe side-play amount of actual barycenter constantly is a displacement vector with said offset tag, through t _I-1The moment and t _iBarycenter linear extrapolation constantly obtains t _I+1Centroid position constantly, through said displacement vector to said t _I+1Centroid position is constantly revised.

5. the Extrapolation method of a kind of storm cloud cluster based on the radar Doppler albedo image according to claim 1; It is characterized in that; Said extrapolation result according to said pocket barycenter; Each said pocket is adopted the form extrapolation method based on scattering model, and the extrapolation result who obtains said pocket local configuration is specially:

1) evenly scatters 24 rays from the pocket barycenter, form 24 intersection points with the pocket border;

2) move previous moment pocket F _I-1Up to barycenter and current time pocket F _iBarycenter overlap;

3) with said t _I+1Barycenter is as F constantly _I+1Barycenter, establish F _I-1And F _iWith ray l _i(i=1 ..., 4 (n _t-1) intersection point) is

With For all intersection points of falling on the current monochromatic subimage profile, calculate Intersection point for falling on the Region Segmentation line makes Δ _i=0 push away pocket at t _I+14 (n constantly _t-1) individual frontier point $p_{{\mathrm{Fl}}_i}\left(t_{i+1}\right)=p_{{\mathrm{Fl}}_i}\left(t_i\right)+{\mathrm{\Δ}}_i;$

4) with 4 (n _t-1) individual frontier point

Connect into polygon as angle point, and polygon is carried out pixel fill, obtain the extrapolation image F of pocket _I+1

5) move extrapolation image F _I+1Up to barycenter and said t _I+1Extrapolation barycenter constantly overlaps.

6. the Extrapolation method of a kind of storm cloud cluster based on the radar Doppler albedo image according to claim 1; It is characterized in that; Said extrapolation result according to said bulk zone barycenter; Local fan-shaped edge to each said bulk zone adopts the form extrapolation method based on mathematical morphology, obtains the extrapolation result of said bulk Regional Bureau contouring;

1) moves previous moment bulk zone F _I-1Up to barycenter p _c(t _I-1) and current time bulk zone F _iBarycenter p _c(t _i) overlap;

2) stretching, extension or retraction zone are detected;

If the bulk cloud cluster of adjacent moment zone G _I-1And G _iFrontier point set be respectively P _B1={ p _B1And P (i) } _B2={ p _B2(i) }; Make that bulk regional background point value is 0, the object-point value is 1, by G _I-1And G _iComposograph G ' makes each pixel value of G ':

$f^{\′}\left(p_i\right)=f_{t_i}\left(p_i\right)-f_{t_{i-1}}\left(p_i\right)=\left\{\begin{array}{c}1\\ 0\\ -1\end{array}\right.$

Definition f ' (p _iThe point p of)=1 _iFor stretching point, f ' (p _iThe point p of)=-1 _iBe retraction point, f ' (p _iThe point p of)=0 _iBe invariant point, stretching, extension point and retraction Vertex Coloring with image obtain stretching area S _Ext, retraction district S _RetAnd constant region S _Inv

3) sector region is divided;

(1) search is stretched or the regional critical point set

that bounces back

To p _B1(i) ∈ P _B1And p _B2(i) ∈ P _B2, if p _B1(i) ∈ P _B2Set up, then p _B1(i) ∈ P _lIf p _B2(i) ∈ P _B1Set up, then p _B2(i) ∈ P _l

(2) from barycenter p _cObtain a plurality of sector region AS to each critical point line _i, i=1,2 ..., each sector region comprises a stretching area or a retraction district, and the sector region that will comprise the stretching area is defined as and stretches the fan section, and the sector region that will comprise the retraction district is defined as the retraction fan section;

4) corrosion expansion process;

Each is stretched fan section AS _iCarry out differential expansion and handle, to each retraction fan section AS _iCarrying out local corrosion handles;

(1) to said stretching, extension fan section AS _iTwo ray limits make to forbid marks for treatment;

(2) read expansion or corrosion information through dye marker;

(3) to said stretching area S _ExtAverage the calculating of span, to said retraction district S _RetAverage the calculating of amount of recovery;

If S _iFor detected certain stretching area or retraction district, if satisfy pixel (p ∈ S _i) ∩ (p ∈ P _B1) count and be N ₁, satisfy (p ∈ S _i) ∩ (p ∈ P _B2) count and be N ₂, satisfy p ∈ S _iCount and be N, then region S _iAverage span or amount of contraction D be:

$D=\frac{N}{0.5(N_1+N_2)}$

In the formula, denominator is a region S _iThe outer edge average length;

(4) use yardstick as the circular configuration element of d to retraction sector region corrosion n time, to stretching, extension sector region expansion n time, obtain a plurality of new sector region AS ' _i, wherein

$d=\left\{\begin{array}{cc}D,& D\≤2\\ 2,& D>2\end{array}\right.$ $n=\mathrm{int}(\frac{D}{2}+0.5)$

In the formula, int representes rounding operation;

5) zone is synthetic;

To a plurality of new sector region AS ' _iWith all constant region S _InvCarry out cup, splice by original relative position with the zone that remains unchanged in the zone that is about to handle, and obtains G _iExtrapolation image G _I+1

6) displacement;

Move image G _I+1Up to barycenter and said t _I+1Extrapolation barycenter constantly overlaps.

Images