CN103197305A - Sand-dust type aerosol inversion method based on support vector machine identification - Google Patents

Abstract

The invention discloses a sand-dust type aerosol inversion method based on support vector machine identification. The method comprises the steps of selecting a sand-dust aerosol layer with obvious characteristics, a thick cloud layer with obvious characteristics and a thin cloud layer with obvious characteristics as classification samples, training classifiers through various sample numbers and characteristic vectors, then confirming an optimal classifier, classifying data of a satellite borne laser radar sand-dust source region, obtaining a sand-dust aerosol identification result with high accuracy, and inverting level height and optical thickness of an aerosol. According to the sand-dust type aerosol inversion method based on the support vector machine identification, a support vector machine is used for confirming characteristics of a hyper-plane through support vectors, the requirements for the number of the samples can be reduced, interference on classification accuracy due to sample uncertainty is reduced, and 532nm polarization detection data and level height information of a satellite borne laser radar are effectively used for distinguishing non-spherical ice crystal cloud and sand-dust type aerosol particles. The sand-dust type aerosol inversion method based on the support vector machine identification is particularly suitable for processing satellite borne laser radar detection data of a sand-dust season in northwest region of our country.

Description

Sand and dust type gasoloid inversion method based on support vector machine identification

Technical field

The invention belongs to the satellite remote sensing technology field, design a kind of sand and dust type gasoloid inversion method based on support vector machine identification especially.

Background technology

China is attacked by a lot of sandstorms every year, and under the influence of this extreme weather, China's industrial and agricultural production, communications and transportation and human life are subjected to serious loss and harm safely.In recent years, the sand and dust coverage enlarges gradually, even has diffused to region of Southeast such as Foochow and Taiwan.Therefore, the fundamental research of carrying out sand and dust observation on a large scale and inversion method has important scientific meaning, will play a positive role in the major event of national economy such as sand and dust early warning, mitigation.

Laser radar is the best means of obtaining the aerosol vertical distribution profile, and at present unique is that the NASA of NASA is in the CALIPSO of emission in 2006 at the spaceborne atmospheric laser radar of rail.But owing to CALIPSO develop for cloud and gasoloid detect, detection is not optimized its international data processing method at sand and dust, can't differentiate dense sand and dust layer and cloud layer, the result that can lead to errors.In the NORTHWEST CHINA area, there is the phenomenon of cloud-Sha coexistence throughout the year, particularly at the dense sand and dust layer in sand and dust source region, the gasoloid recognition result (Version2) in the CALIPSO Takla Makan Desert area of NASA issue, error rate reaches 43%.Therefore, the research that utilizes CALIPSO to survey sand and dust early stage mainly concentrates on the sand and dust source region, the African northwestward (area, the Sahara) that long-term cloud amount is less, the mixability of cloud-Sha is lighter.CALIPSO is the best means of obtaining the vertical distribution profiles information of China's large tracts of land sand and dust, can survey sand and dust information on its hardware, but the defective of its data processing method makes it can not be applied to China's sand and dust research and forecast.Version2 method sand and dust classification error rate as NASA is higher, and Version3 method complexity, parameter are many, gasoloid mode region bad adaptability etc.

Summary of the invention

The purpose of this invention is to provide the sand and dust type gasoloid inversion method based on support vector machine identification, the aerosol suspension position in sand and dust source region when adopting this method to obtain the sandstorm generation, final inverting obtains the aerocolloidal height of sand and dust type, thickness and photoextinction.

Technical scheme of the present invention is a kind of sand and dust type gasoloid inversion method based on support vector machine identification, may further comprise the steps:

At first, training classifier comprises following 4 sub-steps,

Step 1.1, the sample of input sand and dust type gasoloid layer and cloud layer, the sample of described cloud layer comprises Bao Yun and spissatus sample; The spot, sample evidence sandstorm place of described sand and dust type gasoloid layer is obtained, and spot, sandstorm place is determined by passive satellite remote sensing image data that sensor obtains;

Step 1.2 is downloaded the active satellite sensor corresponding with passive satellite remote sensing image data that sensor obtains and is obtained laser radar profile data, obtains hoverheight information and the echoed signal value of sand and dust type gasoloid layer and cloud layer;

Step 1.3 according to hoverheight information and the echoed signal value of step 1.2 gained sand and dust type gasoloid layer and cloud layer, is tried to achieve the proper vector of sand and dust type gasoloid layer sample and cloud layer sample;

Step 1.4, according to the proper vector of step 1.3 gained sand and dust type gasoloid layer sample and cloud layer sample, training obtains sorter; Then, carry out the inverting of sand and dust type gasoloid according to sorter; Comprise following 3 sub-steps,

Step 2.1 is obtained time of origin and the zone of daily sandstorm according to meteorological measuring, when active satellite sensor during through this zone, downloads the laser radar profile data on corresponding date; According to the laser radar profile data of downloading, calculate and obtain the required proper vector of level to be sorted;

Step 2.2 is classified to the proper vector of step 2.1 gained level to be sorted based on step 1.4 gained sorter, and obtains the aerocolloidal laser radar ratio of sand and dust type;

Step 2.3, according to the aerocolloidal laser radar ratio of step 2.2 gained sand and dust type, take all factors into consideration the detection feature of active satellite sensor, adopt the near-end method of inversion, try to achieve the extinction coefficient of sand and dust type particulate, integration obtains the optical thickness value of the face of land overhead sand and dust type gasoloid layer.

And described active satellite sensor adopts the CALIOP sensor that carries on the CALIPSO satellite; The MODIS sensor that described passive satellite sensor adopts the Aqua satellite to carry.

The present invention proposes novel satellite-bone laser radar data processing method, namely replaces the probability density equation with support vector machine, improves the nicety of grading of sand and dust and cloud layer; The situation of the low accuracy data that can only passive use NASA provides can effectively be provided at present.Application of the present invention can solve bottleneck problems such as current China large tracts of land sand and dust Vertical Profile data disappearance, availability of data difference, satisfy the observation requirement in China and even global sand and dust gasoloid source region and diffusion transport zone thereof.Technical scheme provided by the invention has the following advantages and good effect:

1) reduces quantity demand to sample greatly, need not simulate the characteristic distributions of each proper vector;

2) improve the aerocolloidal nicety of grading of sand and dust type, avoid it is divided into cloud layer, and ignored its atmospheric action effect;

3) and then improve cognition to space distribution, diffusion and the scattering properties of sand and dust.

Description of drawings

Fig. 1 is the process flow diagram of the embodiment of the invention.

Embodiment

Sand and dust can not be ignored the influence of atmosphere radiation, and the sand and dust identification of mistake can't obtain the space distribution information of sand and dust, and the delustring of sand and dust and atmosphere radiation effect.Common mistake mainly is that the sand and dust layer erroneous judgement that sand and dust source region concentration is higher is cloud layer.This mainly be because: in the parameter that satellite-bone laser radar provides except the decay backscattering coefficient, dual wavelength is than being the dimensional properties of describing particle, depolarization ratio is the style characteristic of describing particle; And ice cloud crystal and macroparticle sand and dust gasoloid have large scale and aspheric characteristic simultaneously, therefore, have constituted the existence of this mis-classification.There is a large amount of mistakes in the probability density equation sorting technique that NASA official adopts in the version in early days; After the renewal, still have a few errors to exist, and sample quantity required big (need learn the distribution situation of each proper vector).The level identification of mistake will cause the wrong inverting of its optical characteristics.

The active satellite sensor of embodiment adopts the CALIOP sensor that carries on the CALIPSO satellite; The MODIS sensor that described passive satellite sensor adopts the Aqua satellite to carry; MODIS and CALIOP have collaborative observation effect, and what MODIS obtained is the satellite remote sensing influence of two dimensional surface, broad covered area; CALIOP is satellite-bone laser radar, the two-dimentional area coverage that arrives the face of land is little, but can obtain three-dimensional atmospheric descriptor, as: cloud and aerocolloidal space distribution situation, this spaceborne laser divides two wave band 532nm and 1064nm, and wherein the 532nm wave band has two channel of polarization of vertical and level.Can adopt computer software technology to realize automatic operational scheme during specific implementation.For overcoming above problem, the present invention is in conjunction with the MODIS detector that carries on the passive detection device A qua satellite.At first obtain the horizontal distribution information of sand and dust and cloud layer based on its bidimensional image, the characteristics of recycling master-passive detection equipment space-time consistency are obtained satellite-bone laser radar CALIOP sensor at the tangible sample of this provincial characteristics.Because the lineoid of support vector only needs a spot of support vector can determine to obtain, so the quantity required of sample also reduces the correctness of having guaranteed sample of maximum possible under the prerequisite short in the gentle sol layer life cycle of cloud layer, that validation difficulty is high greatly.Like this, classify reliably, identify and established solid prerequisite guarantee for the later stage.Classification results also can with corresponding being compared by dynamic image data, the result of checking shows that this method is than official's disposal route of NASA effect that has greatly improved.

As shown in Figure 1, embodiment carries out the inverting of sand and dust type gasoloid at the sand and dust source region in NORTHWEST CHINA area, and flow process is as follows:

At first, sorter is trained, to guarantee the correctness of follow-up classification results.Namely utilize the MODIS remote sensing image data, determine spot, sandstorm place and time of origin, choose sample, and sorter is trained, comprise following 4 steps,

Step 1.1, the sample of input sand and dust type gasoloid layer and cloud layer.Wherein the sample of cloud layer comprises Bao Yun and spissatus sample.Wherein the sample of sand and dust type gasoloid layer should be obtained with reference to remote sensing image data that passive satellite obtains.During concrete enforcement, can utilize the accurate synchronism of space-time of passive satellite sensor and active satellite sensor, with reference to remote sensing image data that passive satellite obtains, obtain the generation information of sandstorm, determine the aerocolloidal sample selection of sand and dust type zone, in the non-sandstorm outbreak period, determine that the sample of spissatus layer and thin layer is selected the zone in addition.

Because it is short that MODIS remote sensing image spectral information enriches, finishes once global all standing detection time, therefore, its execution earth observation and atmospheric correction can be seen sandstorm distribution and spread condition for many years comparatively intuitively.Correctness in order to ensure the MODIS image data, releasing news of the earth observatory page of embodiment by NASA official website, obtain the sandstorm image in NORTHWEST CHINA area, determine that the sandstorm sample in this sand and dust source region chooses scheme, simultaneously, during no sand and dust, determine that the sample of spissatus and thin cloud is chosen scheme.Aerocolloidal characteristics mainly are reflected in the weak reflected signal in low latitude; Spissatus characteristics mainly are the strong echoed signals in cloud top, and lower signal suddenly weakens; It is all stronger that the characteristics of Bao Yun are that the low echoed signal of cloud is arrived on the cloud top, can see subaerial weak signal simultaneously.Because this judgement mainly is based on human eye identification, thus the demand of sample size more less, feature is more obvious, just more can guarantee the precision of classifying.

Step 1.2 is downloaded the active satellite sensor corresponding with passive satellite remote sensing image data that sensor obtains and is obtained laser radar profile data, obtains hoverheight information and the echoed signal value of sand and dust type gasoloid layer and cloud layer.

After having determined the scheme of choosing of sample, download is corresponding to the CALIOP laser radar profile detection data of MODIS, the recycling satellite data fetch program, profile observation data in the corresponding longitude and latitude scope is read and preserves, mainly comprise: the decay backscattering coefficient of the rising of the atmospherium that has found, end height, 1064nm and 532nm passage, and the component value on the 532nm different polarization passage.Whole data handling system is finished based on IDL and matlab exploitation.

Step 1.3 according to hoverheight information and the echoed signal value of step 1.2 gained sand and dust type gasoloid layer and cloud layer, is tried to achieve the proper vector of sand and dust type gasoloid layer sample and cloud layer sample.

CALIOP is as satellite-bone laser radar, flying height height, speed are fast, and corresponding flow chart of data processing is meticulousr, need search earlier atmospherium (cloudlike and gasoloid) hoverheight at place, here embodiment will directly quote the level lookup result of NASA, and then the computation layer sub-eigenvector.The proper vector that can directly quote comprises the rising of level, end height, increased surface covering type etc., the support vector that need utilize the formula indirect calculation to obtain comprises decay backscattering coefficient (attenuated scattering coefficient), depolarization ratio (the volume depolarization ratio of level integration; VDR) and the dual wavelength signal than (total attenuated color ratio; ACR), it is emphasized that satellite-bone laser radar CALIOP belongs to pulsed Mie scattering laser light detection and ranging, valid data are in-2.0-40km atmosphere altitude range, and spatial resolution can change to some extent along with height, and high spatial resolution is 30m, and lowest spatial resolution is 300m.The level integration namely is that the signal in the respective heights scope is superposeed.The proper vector computing formula that embodiment adopts is as follows with reference to the formula that NASA provides:

${\mathrm{\γ}}_{\mathrm{layer}}^{\′}={\∫}_{\mathrm{top}}^{\mathrm{base}}{\mathrm{\β}}_p\left(r\right)\·T_p^2\left(r\right)\mathrm{dr}---\left(1\right)$

${\mathrm{\δ}}_{\mathrm{layer}}=(\underset{k=\mathrm{top}}{\overset{\mathrm{base}}{\mathrm{\Σ}}}{\mathrm{\β}}_{532,\⊥}^{\′}\left(z_k\right)/\underset{k=\mathrm{top}}{\overset{\mathrm{base}}{\mathrm{\Σ}}}{\mathrm{\β}}_{532,\left|\right|}^{\′}\left(z_k\right))---\left(2\right)$

$B_{\mathrm{\λ},k}=\frac{{\mathrm{\β}}_{\mathrm{\λ}}^{\′}\left(z_k\right)}{T_{m,\mathrm{\λ}}^2\left(z_k\right)\·T_{o_3,\mathrm{\λ}}^2\left(z_k\right)}---\left(3\right)$

${\mathrm{\χ}}_{\mathrm{layer}}^{\′}=(\underset{k=\mathrm{top}}{\overset{\mathrm{base}}{\mathrm{\Σ}}}{B}_{1064,k}/\underset{k=\mathrm{top}}{\overset{\mathrm{base}}{\mathrm{\Σ}}}{B}_{532,k})---\left(4\right)$

The decay backscattering coefficient γ ' of formula (1) representational level integration _Layer, r represents laser radar to the distance of target, dr'' is the range resolution of laser radar.β _p(r) expression is apart from the backscattering coefficient of r place particle,

Expression the two-way of particle sees through, and base and top represent rising and end height of atmospherium respectively.β ' expression decay backscattering coefficient, β _⊥' (z) and β _||' (z) represent the vertical and parallel component of channel of polarization respectively, the ratio behind both integrations namely is the depolarization ratio δ of level integration _Layer, (wherein, footmark m represents molecule, O by the conversion of (3) formula ₃Expression ozone, λ represents wavelength, z represents height, k is the hoverheight signal value of corresponding atmospherium), obtain the dual wavelength of level integration than χ ' _LayerParameter δ _Layer, χ ' _LayerThe shape and size feature of particle has been described respectively.

Specifically, z _kBe be suspended in level (cloud layer or sand and dust type gasoloid layer) in the atmosphere from the top height to k signaling point correspondence of bottom, β ' _{532, ⊥}(z _k) be that 532nm wave band vertical polarization passage is at height z _kThe decay backscattering coefficient at place, β ' _{532, ||}(z _k) be that 532nm wave band parallel polarization passage is at height z _kThe decay backscattering coefficient at place, β ' _λ(z _k) be that wave band λ is at height z _kThe decay backscattering coefficient,

Be that atmospheric molecule is at wave band λ, height z _kThe two-way transmitance at place, Be that ozone is at wave band λ, height z _kThe two-way transmitance at place.B _{1064, k}Expression expression 1064nm wave band hoverheight signal value k place intermediate conversion formula, B _{532, k}Expression 532nm wave band hoverheight signal value k place intermediate conversion formula.

Step 1.4, according to the proper vector of step 1.3 gained sand and dust type gasoloid layer sample and cloud layer sample, training obtains sorter.During concrete enforcement, by selecting different training sample quantity and feature space, carry out the comparative analysis of classification results, find optimum lineoid, obtain optimum sorter.

Because NASA official classification method adopts the probability density equation to classify, and the two kinds of methods in front and back adopt respectively and do not contain shake coefficient and contain the shake method of coefficient of depolarization and classify of depolarization, the depolarization shape description parameter of coefficient as particle of shaking, belong under the prerequisite of nonspherical particle at ice cloud crystal and sand and dust, how this parameter of reasonable use is classified, and is the meaning of this step.

Support vector machine mainly is to determine lineoid by training sample, and for the situation of two classification, when finding support vector, lineoid is also just determined thereupon.W is a vector vertical with the classification lineoid, and x represents an atmospherium to be sorted, and x=(x ₁..., x _n), x _iA kind of attribute of then representing it, the value of i are 1 ..., n, n are the attribute number of x, consider a some product space R ⁿ, super unilateral the writing:

{x∈R ⁿ|<w,x>+b＝0},w∈R ⁿ,b∈R （5）

Wherein, the R representation space, n represents dimension, R ⁿRepresent the some product space of a n dimension, b represents discrimination threshold.

Under the situation of linear separability, distinguish two kinds with (wx+b＞0, wx+b＜0).For inseparable situation, then need to add more proper vector, select suitable kernel function, and with its feature space mapping higher dimensional space, be converted into linear problem again, and then obtain classification results.

Repeatedly experimental result proves: the sand and dust layer in sand and dust source region is lower than cloud layer level height, therefore, the elevation information of level is aided with the depolarization coefficient that shakes can improve the nicety of grading of sand and dust source region sand and dust type gasoloid and cloud layer effectively, the employed proper vector of sorter both had been different from NASA Level1 and had not contained the depolarization coefficient that shakes in this patent, also was different from NASA Level2 and comprised depolarization shake coefficient and geography information simultaneously.Experiment shows, when sample size is 280, overall accuracy surpasses 98.4%, under the condition with respect to different sample sizes, 23456 combined feature spatial classification precision are the most stable, and therefore, the proper vector that the embodiment of the invention contains comprises: the elevation information z of level, dual wavelength ratio, depolarization ratio and decay backscattering coefficient.

Then, carry out the inverting of sand and dust type gasoloid according to sorter, mainly be time of origin and the zone that obtains daily sandstorm according to meteorological measuring, when this zone of active satellite sensor process, download the remote sensing image data on corresponding date, the sorter that trains before utilizing is classified, is identified atmospherium, and final inverting obtains the optical parameter value of each level.Specifically comprise following 3 steps:

Step 2.1, obtain the proper vector of level to be sorted: embodiment in conjunction with the detection data of China weather station, obtain generation, the development of the sandstorm of daily the Northwest, the active satellite sensor of downloading corresponding time and zone obtains laser radar profile data.According to the laser radar profile data of having downloaded, calculate and obtain the required proper vector of level to be sorted.

Receive auxiliary weather information, paid close attention to for the dust and sand weather in NORTHWEST CHINA area, determine that take place period in sandstorm, and download satellite-bone laser radar through the data in sand and dust source region, namely the official website at NASA downloads Level1 and Level2 data.Wherein, the Level1 data are the laser radar initial echo profile information (echoed signal value) that the area takes place through sandstorm, for finding the solution with the inverting of aerosol optical characteristics of subsequent characteristics vector prepared.If the Level2 data owner, is searched the relevant information (namely obtaining the hoverheight information of sand and dust layer, cloud layer) that obtains level according to signal in band, obtain the required support vector of classifying in conjunction with place height and echoed signal integration.The similar step 1.3 of computing method adopts formula (1), (2), (3), (4) to get final product.

Step 2.2 is classified to the proper vector of step 2.1 gained level to be sorted based on step 1.4 gained sorter, and obtains the aerocolloidal laser radar ratio of sand and dust type.

Carry out sort operation, sorting technique is based on existing sorter, is similar to step 1.4, and just the test sample book in the classifier optimization process being replaced is the atmospherium that will classify.Calculate cloud layer and sand and dust type gasoloid layer according to step 2.1 gained proper vector to be sorted, other classifications can directly be inherited the result of NASA, obtain final classification results figure, wherein can adopt different colours to represent clean atmosphere, cloud, gasoloid, the face of land, deep layer stratum and signal net loss district (that is, laser radar can't penetrate atmospherium) in this width of cloth scene respectively.By experiment this support vector machine SVM is compared with two versions of probability density distribution PDF, can see that it can not only correctly identify the sand and dust in low latitude, can also identify the thick sand and dust layer of high-latitude area (north latitude 40-42 degree).

After classification is finished, can comprise station observation and site observation with reference to China's existing sand and dust gasoloid data, the sand and dust aerated solids particle laser radar in selected sand and dust source region, northwest is than empirical value, and specific implementation is prior art.Laser radar is than being the extinction coefficient of particle and the ratio of backscattering coefficient, and the correct approximate hypothesis of this value can effectively improve the precision of gasoloid optical parametric inverting, and this step can be avoided the regional inadaptability that data produce outside simply the successor state.

Step 2.3, according to the aerocolloidal laser radar ratio of step 2.2 gained sand and dust type, take all factors into consideration the detection feature of active satellite sensor, adopt the near-end method of inversion, try to achieve the extinction coefficient of sand and dust type particulate, integration obtains the optical thickness value of the face of land overhead sand and dust type gasoloid layer.This method can effectively be avoided under cloud-Sha mixing condition, sand and dust are classified as cloud layer and have ignored its change in time and space characteristic and weather action effect, finally can correct inverting obtains the optical characteristics of sand and dust type particulate.

The laser radar equation expression formula can be write as following form:

$P\left(r\right)=c_0\&CenterDot;P_0\&CenterDot;\frac{\mathrm{<figure-callout\; id="2"\; label="A\; r"\; filenames="130321141310.png"\; state="\{\{state\}\}">A</figure-callout>}}{r^{}}\&CenterDot;\mathrm{\&Delta;}{r}^{\&prime;}\&CenterDot;\mathrm{\&beta;}(\mathrm{\&lambda;},r)\&CenterDot;\exp [-2{\&Integral;}_0^r\mathrm{\&alpha;}(\mathrm{\&lambda;},r)\mathrm{dr}]---\left(6\right)$

In the formula, P (r) receives from r to r+ Δ r' distance segment atmosphere echoed signal for laser radar;

P ₀Power for the emission laser beam;

c ₀Be the laser radar meter constant, relevant with the configuration of laser radar;

A is the area of receiving telescope;

Δ r' is the range resolution of laser radar system;

R is that laser radar is to the distance of target;

(λ is apart from the backscattering coefficient of laser radar r place target (being certain component in the atmosphere) in wavelength X r) to β;

(λ is apart from the extinction coefficient of laser radar r place target (being certain component in the atmosphere) at the wavelength X atmosphere r) to α.

At present, extensively the ground laser radar inversion method that adopts is the Fernald method of inversion of the prior art, and it is separately considered atmospheric molecule and aerocolloidal contribution, so laser radar equation can be expressed as:

$P\left(r\right)=\frac{C}{r^2}[{\mathrm{\&beta;}}_{\mathrm{mol}}\left(r\right)+{\mathrm{\&beta;}}_{\mathrm{par}}\left(r\right)]\exp \{-2{\&Integral;}_0^r[a_{\mathrm{mol}}\left(r^{\&prime;}\right)+a_{\mathrm{par}}\left(r^{\&prime;}\right)\left]{\mathrm{dr}}^{\&prime;}\right\}---\left(7\right)$

Subscript m ol represents atmospheric molecule; Subscript par represents gasoloid.C is the abbreviation of all parameters of laser radar system.

Specifically, β _Mol(r) the expression atmospheric molecule is at the backscattering coefficient at distance laser radar r place, β _Par(r) the expression particulate is at the backscattering coefficient at distance laser radar r place, α _Mol(r') represent the extinction coefficient that atmospheric molecule is located at distance laser radar r ', α _Par(r') represent the extinction coefficient that particulate is located at distance laser radar r ', r' represents integration variable, and dr' is the range resolution of laser radar.

Selected reference position r _c(being calibrated altitude) supposes known r _cThe corresponding backscattering coefficient in place is β (r _c) use

Describe the calculating of laser radar ratio, α (r) and β (r) represent extinction coefficient and the backscattering coefficient apart from laser radar r place target respectively.Wherein, the laser radar of particle is than need more needing the physical characteristics of zones of different particle to estimate, and the laser radar of molecule is than then adopting a constant value.

R then _cThe Aerosol Extinction of above height (forward direction integration) is:

${\mathrm{\&alpha;}}_{\mathrm{par}}\left(r\right)=-\frac{S_{\mathrm{par}}}{S_{\mathrm{mol}}}{\mathrm{\&alpha;}}_{\mathrm{mol}}\left(r\right)+\frac{P\left(r\right){\mathrm{<figure-callout\; id="2"\; label="r"\; filenames="130321141310.png"\; state="\{\{state\}\}">r</figure-callout>}}^2\&CenterDot;\exp [-2(\frac{S_{\mathrm{par}}}{S_{\mathrm{mol}}}-1){\&Integral;}_{r_c}^r{\mathrm{\&alpha;}}_{\mathrm{mol}}\left(r^{\&prime;}\right){\mathrm{dr}}^{\&prime;}]}{\frac{P\left(r_c\right){r_c}^2}{{\mathrm{\&alpha;}}_{\mathrm{par}}\left(r_c\right)+\frac{S_{\mathrm{par}}}{S_{\mathrm{mol}}}{\mathrm{\&alpha;}}_{\mathrm{mol}}\left(r_c\right)}-2{\&Integral;}_{r_c}^{r}P\left(r^{\&prime;}\right)r^{\&prime;2}\&CenterDot;\exp [-2(\frac{S_{\mathrm{par}}}{S_{\mathrm{mol}}}-1){\&Integral;}_{r_c}^r{\mathrm{\&alpha;}}_{\mathrm{mol}}\left(r^{\&prime;\&prime;}\right){\mathrm{dr}}^{\&prime;\&prime;}]{\mathrm{dr}}^{\&prime;}}---\left(8\right)$

Specifically, S _ParRepresent aerocolloidal laser radar ratio, S _MolThe laser radar ratio of expression atmospheric molecule, r'' represents integration variable, dr'' is the range resolution of laser radar.

And r _cThe Aerosol Extinction of following height (back is to integration) is:

${\mathrm{\&alpha;}}_{\mathrm{par}}\left(r\right)=-\frac{S_{\mathrm{par}}}{S_{\mathrm{mol}}}{\mathrm{\&alpha;}}_{\mathrm{mol}}\left(r\right)+\frac{P\left(r\right){\mathrm{<figure-callout\; id="2"\; label="r"\; filenames="130321141310.png"\; state="\{\{state\}\}">r</figure-callout>}}^2\&CenterDot;\exp \left[2(\frac{S_{\mathrm{par}}}{S_{\mathrm{mol}}}-1){\&Integral;}_r^{r_c}{\mathrm{\&alpha;}}_{\mathrm{mol}}\left(r^{\&prime;}\right){\mathrm{dr}}^{\&prime;}\right]}{\frac{P\left(r_c{r_c}^2\right)}{{\mathrm{\&alpha;}}_{\mathrm{par}}\left(r_c\right)+\frac{S_{\mathrm{par}}}{S_{\mathrm{mol}}}{\mathrm{\&alpha;}}_{\mathrm{mol}}\left(r_c\right)}+2{\&Integral;}_r^{r_c}P\left(r^{\&prime;}\right)r^{\&prime;2}\&CenterDot;\exp \left[2(\frac{S_{\mathrm{par}}}{S_{\mathrm{mol}}}-1){\&Integral;}_r^{r_c}{\mathrm{\&alpha;}}_{\mathrm{mol}}\left(r^{\&prime;}\right){\mathrm{dr}}^{\&prime;\&prime;}\right]{\mathrm{dr}}^{\&prime;}}---\left(9\right)$

The method that the forward direction integration is found the solution Aerosol Extinction is the near-end method of inversion, then is the far-end method of inversion to integration after (9) formula.So ground laser radar in the prior art is because the stable reference point r of remote signaling _cBe selected in far-end, namely adopt the far point method of inversion, and the steady-state signal of satellite-bone laser radar CALIOP belongs to high spacing wave, nearer from detector, therefore the present invention selects the near-end method of inversion opposite with the ground laser radar for use, after the inverting of employing formula (8) obtains extinction coefficient, with reference to level place height, the extinction coefficient profile is carried out integration, obtain the sand and dust aerosol optical depth in this sky, zone.

Below the sand and dust gasoloid identification of summary embodiment and inverting flow process are for implementing reference:

1. the development descriptor takes place in the NORTHWEST CHINA area sandstorm that provides with reference to the earth observation page on the NASA official website, and downloads corresponding MODIS remote sensing image and CALIPSO data.At the image data of MODIS, choose the laser radar profile data of corresponding sandstorm.Select the spissatus and thin cloud laser radar of evident characteristic profile data in this zone.

2. sorter is trained, it is less to obtain sample size, and the higher support vector of nicety of grading is chosen scheme.The optimum lineoid of final acquisition.

3. with reference to China's sandstorm weather data, download CALIPSO observation data Level1 and the Level2 of there and then correspondence, if the data owner of Level2 obtains the elevation information of level, comprise rising and end height of level.Utilizing this information that three parameters of Level are carried out integration, needed proper vector obtains classifying.

4. based on existing sorter the profile of Experimental Area is classified sign 0: cloud; 1: gasoloid.The 3-D display of atmosphere profile is finished in the face of land elevation information of recycling IDL software and CALIPSO Data Identification, clean atmosphere and laser beam net loss zone, observes hoverheight and the distribution situation of sand and dust type gasoloid and cloud layer intuitively.

Than empirical value, the laser radar echo signal is carried out integration on inverting and the corresponding hoverheight and then the aerocolloidal optical thickness value of the sand and dust of asking based on existing classification results and laser radar.For the research of follow-up its radiation characteristic is had laid a good foundation.

Specific embodiment described herein only is that the present invention's spirit is illustrated.Those skilled in the art can make various modifications or replenish or adopt similar mode to substitute described specific embodiment, but can't depart from spirit of the present invention or surmount the defined scope of appended claims.

Claims (2)

1. the sand and dust type gasoloid inversion method based on support vector machine identification is characterized in that, may further comprise the steps:

At first, training classifier comprises following 4 sub-steps,

Step 1.1, the sample of input sand and dust type gasoloid layer and cloud layer, the sample of described cloud layer comprises Bao Yun and spissatus sample; The spot, sample evidence sandstorm place of described sand and dust type gasoloid layer is obtained, and spot, sandstorm place is determined by passive satellite remote sensing image data that sensor obtains;

Step 1.2 is downloaded the active satellite sensor corresponding with passive satellite remote sensing image data that sensor obtains and is obtained laser radar profile data, obtains hoverheight information and the echoed signal value of sand and dust type gasoloid layer and cloud layer;

Step 1.3 according to hoverheight information and the echoed signal value of step 1.2 gained sand and dust type gasoloid layer and cloud layer, is tried to achieve the proper vector of sand and dust type gasoloid layer sample and cloud layer sample;

Step 1.4, according to the proper vector of step 1.3 gained sand and dust type gasoloid layer sample and cloud layer sample, training obtains sorter;

Then, carry out the inverting of sand and dust type gasoloid according to sorter; Comprise following 3 sub-steps,

Step 2.1 is obtained time of origin and the zone of daily sandstorm according to meteorological measuring, when active satellite sensor during through this zone, downloads the laser radar profile data on corresponding date; According to the laser radar profile data of downloading, calculate and obtain the required proper vector of level to be sorted;

Step 2.2 is classified to the proper vector of step 2.1 gained level to be sorted based on step 1.4 gained sorter, and obtains the aerocolloidal laser radar ratio of sand and dust type;

Step 2.3, according to the aerocolloidal laser radar ratio of step 2.2 gained sand and dust type, take all factors into consideration the detection feature of active satellite sensor, adopt the near-end method of inversion, try to achieve the extinction coefficient of sand and dust type particulate, integration obtains the optical thickness value of the face of land overhead sand and dust type gasoloid layer.

2. the sand and dust type gasoloid inversion method based on support vector machine identification according to claim 1 is characterized in that: the CALIOP sensor that carries on the described active satellite sensor employing CALIPSO satellite; The MODIS sensor that described passive satellite sensor adopts the Aqua satellite to carry.

Images