CN103197303A - Earth surface two-direction reflection characteristic retrieval method and earth surface two-direction reflection characteristic retrieval system based on multiple sensors - Google Patents

Abstract

The invention relates to the technical field of remote sensing image processing, in particular to a multi-sensor-based inversion method and system for surface bidirectional reflection characteristics. The method comprises the steps: S1. extracting multi-sensor observation data in the inversion period; S2. combining the observation data extracted in the step S1, using band conversion to construct a multi-sensor observation data set; S3. Angular multi-band model and multi-sensor observation data set for surface bidirectional reflectance inversion. The multi-sensor-based inversion method for surface bidirectional reflection characteristics provided by the present invention constructs a rich multi-angle observation data set by comprehensively utilizing the observation data of various sensors, and performs the inversion of surface bidirectional reflection characteristics, thereby solving the problem of The problem of insufficient inversion information in the two-way reflection characteristics is solved, and the accuracy and time resolution of the inversion two-way reflection characteristics are effectively improved.

Description

Based on the face of land two of multisensor to reflection characteristic inversion method and system

Technical field

The present invention relates to the remote sensing technology field, be specifically related to a kind of face of land two based on multisensor to reflection characteristic inversion method and system.

Background technology

Moderate Imaging Spectroradiomete (Moderate Resolution Imaging Spectroradio-meters, MODIS) two to the reflection distribution function (Bidirectional Reflectance Distri but-ion Function, BRDF) product is that the present most widely used face of land two is to the reflection characteristic product.Its theoretical method basis is the linear BRDF model that nuclear drives, and this model comes two tropism's reflectance signatures on the match face of land with the linear combination of nuclear.Briefly, the nuclear driving model can be represented by the formula:

Wherein,

Be two to reflection; K is each nucleoid (K _IsoFor isotropic nuclear, be expressed as 1; K _VolBe volume scattering nuclear; K _GeoBe geometrical optics nuclear), f is the shared ratio (f of corresponding each nuclear _IsoBe isotropic karyonide number; f _VolBe volume scattering karyonide number; f _GeoBe geometrical optics karyonide number); θ is solar zenith angle;

Be the observation zenith angle; φ is relative bearing; Λ is wide waveband.

In the prior art, MODIS two mainly comprises step to reflection characteristic inverting flow process:

S1 '. extract pixel multi-angle observation data from the MODIS satellite image;

For example, extract effectively observation data input of 16 days cloudless MODIS observation datas (MOD09GA and MYD09GA) conduct, observation data comprises 7 reflectivity wave bands, four angle wave bands (solar zenith angle, observation zenith angle, solar azimuth and observed azimuth).

S2 '. use the nuclear driving model to carry out two to the reflection characteristic inverting;

According to the number of effective observation data, two are divided into complete inverting and back-up algorithm inverting again to the reflection characteristic inverting.

S21 '. inverting fully; When effective observation data during more than or equal to 7 times, carry out complete inverting.Refutation process comprises:

S211 '. calculate nuclear expression;

(a) volume scattering nuclear:

The volume scattering nuclear that MODIS two selects for use to reflection model be the Roujean Luo Si thick-layer nuclear that equals to propose in 1992 (Ross-Thick kernel, RT) approximate with Ross radiation transfer theory (Ross, 1981), description be dense vegetation canopy.Set up three hypothesis in the modeling process of Roujean, the small scattering surface in background and the medium all is lambert's scattering; Scattering surface in the medium is towards being stochastic distribution; Only once scattering; And when zenith is observed (observation zenith angle and position angle are 0), the value of nuclear is normalized to 0.Luo Si thick-layer kernel form is as follows:

Wherein, ξ is the phasing degree,

(b) geometrical optics nuclear

The geometrical optics nuclear that MODIS two selects for use to reflection model is the sparse reciprocal kernel of Li Shi (the LiSparse Reciprocal kernel that Li Xiaowen etc. proposes, LSR), it describes the sparse canopy that distributes on lambert's scattering earth background based on the geometric optical model development of scape synthetic model.Because the data of multiple angles of moonscope usually has the very little problem of solar zenith angle variation range, what can cause that match obtains two occurs than mistake when being extrapolated to other solar zenith angles to reflection model.Model satisfies principle of reciprocity, i.e. the face of land two is constant to reflection during sun angle and observation angle reciprocity, to a certain extent departure.The concrete form of this nuclear is as follows:

Wherein:

${\mathrm{\θ}}^{\′}={\tan }^{-1}\left(\frac{b}{r}\tan \mathrm{\θ}\right)$

Wherein,

Be the ratio of having described tree crown spheroid height and the width, general value is 2.0.

S212 '. this two to reflection model by least square method, be finally inversed by the match observation data

Optimum f _k, i.e. known θ _l,

φ _lThe reflection observation data of angle By minimizing

Obtain

Wherein, d is degree of freedom, and namely the observation data sample number deducts karyonide and counts f _kNumber; w _l(Λ) be the weight of given corresponding l observation data.

S22 '. the back-up algorithm inverting; When effective observation data more than or equal to 3 times and during less than 7 times, carry out the back-up algorithm inverting.Back-up algorithm uses the reflectivity of historical observation data and the angle information of current observation data to carry out.Use the model of back-up algorithm to be rewritten as:

Wherein, l is the l time observation,

Be the l time conception of history geodetic table reflectivity,

It is historical karyonide number.Defining two parameters is:

${\mathrm{sum}}_1=\underset{l=1}{\overset{n}{\mathrm{\Σ}}}{\mathrm{\ρ}}_l^{\mathrm{obs}}{\mathrm{\ρ}}_l^{\mathrm{his}}{w}_l$

${\mathrm{<figure-callout\; id="2"\; label="sum"\; filenames="HDA00003020257300011.png"\; state="\{\{state\}\}">sum</figure-callout>}}_2=\underset{l=1}{\overset{n}{\mathrm{\&Sigma;}}}{\mathrm{\&rho;}}_l^{\mathrm{his}}{\mathrm{\&rho;}}_l^{\mathrm{his}}{w}_l$

Wherein, Be the earth surface reflection rate of the l time observation, w _lIt is the weight of current observation.So, new karyonide number is found the solution as follows:

$f_i^{\mathrm{new}}=\frac{{\mathrm{sum}}_1}{{\mathrm{sum}}_2}{f}_i^{\mathrm{his}}$

Be finally inversed by after the karyonide number by above dual mode, can obtain two under any solar incident angle and the view angle condition to reflection by the extrapolation of nuclear.This shows that nuclear drives the BRDF model based on the linear combination of nuclear, succinctly, at a high speed, the inverting analytic solution, the data fitting ability is strong, and this process of handling for data in enormous quantities is to have extremely huge advantage.Simultaneously, each nuclear has a certain physical meaning in the model, this make we in Extrapolating model when not having the direction of observation data, be hopeful to explain and to control the result of extrapolation.But of the prior art two still exist following defective to the reflection characteristic inversion method:

(1), production life cycle is long, temporal resolution is low

Moderate Imaging Spectroradiomete needs 16 days time to generate the BRDF product; And because time span is big, be difficult to react two the variations to reflection characteristic of the face of land when changing fast, especially in the quick season of growth of vegetation and snow melt season.

(2), fully retrieval products is few, filling product is many, influences the product precision

Because influence such as Yun Xue causes effective observation data to reduce, force and carry out the back-up algorithm inverting and historical data is filled more; The observation data that participates in inverting simultaneously reduces, and makes to have the quantity of information deficiency, and least square is found the solution instability, and then influences inversion accuracy.

(3), at single-sensor, two are subject to the sensor wave band to reflection model

Model mainly is at the single-sensor band setting, be subject to sensor during inverting, and regional the making with the gap of response of band setting can not be combined effectively from the data of different platform sensor between the different sensors, causes this model can't be applicable to that the collaborative inverting face of land two of multiple sensors is to reflection characteristic.

Summary of the invention

(1) technical matters that will solve

The object of the present invention is to provide a kind of face of land two based on multisensor to reflection characteristic inversion method and system, in order to overcome defectives such as single-sensor quantity of information deficiency in the prior art, temporal resolution be low.

(2) technical scheme

Technical solution of the present invention is as follows:

A kind of face of land two based on multisensor comprises step to the reflection characteristic inversion method:

S1. extract the observation data of multisensor in the inverting cycle;

S2. in conjunction with the observation data of extracting among the described step S1, use the wave band conversion to make up multisensor observation data collection;

S3. according to effective observation frequency, carry out the face of land two to the reflection characteristic inverting in conjunction with multi-angle multiband model and multisensor observation data collection.

Preferably, described multisensor comprises: Moderate Imaging Spectroradiomete, high resolution scanning radiometer and visible light infrared scanning radiometer, described multisensor observation data comprises: earth surface reflection rate, the sun and sensors observe angle, cloud mask and NDVI.

Preferably, described step S1 comprises:

S101. be benchmark with Moderate Imaging Spectroradiomete pixel position, the observation data of searching other sensor that is complementary according to the center longitude of pixel;

S102. according to the cloud mask item in each the sensors observe data that finds, if judge fine observation data, extract earth surface reflection rate, the sun and sensors observe angle and the NDVI data of this sensor.

Preferably, described step S2 comprises:

S201. use actual measurement ground-object spectrum data to return by least square method and obtain the wave band conversion coefficient that the different target sensor is expressed with the standard transducer unification;

S202. according to described wave band conversion coefficient, each sensor with unified standard transducer model tormulation, is made up multisensor observation data collection.

Preferably, described step S3 comprises:

S31. judge effective observation frequency: if be not less than 49, execution in step S32 then; If be not less than 21 and less than 49, then execution in step S33;

S32. in conjunction with multisensor observation data collection, carry out complete inverting according to multi-angle multiband model;

S33. in conjunction with multisensor observation data collection, introduce the inverting of priori according to multi-angle multiband model.

Preferably, described step S33 comprises:

S331. obtain the global face of land two to the reflection characteristic priori;

S332. utilize the global face of land two to the reflection characteristic priori, make up model priori observation data;

S333. the weighting of model priori observation data is introduced multisensor observation data collection, carried out inverting according to multi-angle multiband model.

Preferably, also comprise after the described step S33:

S4. according to net information content, inversion result is estimated and optimized described multisensor observation data collection.

Preferably, described step S4 comprises:

S41. according to inversion result among the described step S33, match obtains reflectivity, recursion step S33, and the wave band of the relative error maximum of at every turn selecting if remove greater than 50%, finally obtains the model karyonide number of preliminarily stabilised;

S42. use the model karyonide number of preliminarily stabilised, together with standard transducer observation data collection and newly-increased sensors observe data set, calculate quantity of information separately respectively; The difference of multi-sensor information amount and standard transducer quantity of information is net information content:

If just, then this model karyonide number is final mask karyonide number;

If negative, then remove this sensors observe data, recomputate model karyonide number according to described step S33.

The present invention also provides a kind of and has realized that the above-mentioned face of land two based on multisensor is to the system of reflection characteristic inversion method:

A kind of face of land two based on multisensor comprises to reflection characteristic inverting system: the multisensor observation data extraction module that sets gradually, multisensor observation data collection make up module, the face of land two to reflection characteristic inverting module and model result output module;

Described multisensor observation data collection makes up module and makes up multisensor observation data collection according to the observation data use wave band conversion that described multisensor observation data extraction module extracts; Described model result output module export the described face of land two to the face of land two that reflection characteristic inverting module is carried out in conjunction with described multisensor observation data collection to the reflection characteristic inversion result.

Preferably, comprise that also being arranged on the described face of land two optimizes the inverting module to the quantity of information between reflection characteristic inverting module and the model result output module;

Described quantity of information is optimized the inverting module according to net information content, and inversion result is estimated and optimized described multisensor observation data collection.

(3) beneficial effect

Observation data from the face of land two of multisensor to the reflection characteristic inversion method that pass through the comprehensive utilization multiple sensors based on provided by the present invention, make up the abundant multi-angle observation data set of structure, carry out the face of land two to the inverting of reflection characteristic, thereby solved the problem of the face of land two inverting quantity of information deficiency in the reflection characteristic, and effectively raised inverting two to temporal resolution and the precision of reflection characteristic.

Description of drawings

Fig. 1 is based on the face of land two of the multisensor schematic flow sheet to the reflection characteristic inversion method in the embodiment of the invention;

Fig. 2 is based on the face of land two of the multisensor module diagram to reflection characteristic inverting system in the embodiment of the invention.

Embodiment

Below in conjunction with drawings and Examples, the specific embodiment of the present invention is described further.Following examples only are used for explanation the present invention, but are not used for limiting the scope of the invention.

Process flow diagram a kind of face of land two based on multisensor as shown in fig. 1 mainly may further comprise the steps to the reflection characteristic inversion method:

S1. extract the multi-angle observation data of each sensor matching pixel in the inverting cycle by the Multi-sensor Satellite image; Multisensor in the present embodiment comprises: Moderate Imaging Spectroradiomete, high resolution scanning radiometer (Advanced Very High Resolution Radiometer, AVHRR) and the visible light infrared scanning radiometer (Visible and InfraRed Radiometer, VIRR); The observation data of each sensor mainly comprises wave band reflectivity, observation angle, cloud mask (being used for judging whether being fine day observation, i.e. valid data) and NDVI(Normalized Difference Vegetation Index, normalized differential vegetation index).Particularly, are reflectivity wave band 1-7 for the implication of MODIS sensors observe data, solar zenith angle, solar azimuth, observation zenith angle and observed azimuth, cloud mask and NDVI; Implication for AVHRR sensors observe data is reflectivity wave band 1,2,3A, solar zenith angle, observation zenith angle and relative bearing, cloud mask and NDVI; Implication for VIRR sensors observe data is reflectivity wave band 1,2,6,7,9, solar zenith angle, solar azimuth, observation zenith angle and observed azimuth, cloud mask and NDVI.The flow process that observation data is extracted specifically comprises:

S101. be benchmark with Moderate Imaging Spectroradiomete pixel position, the observation data of searching other sensor that is complementary according to the center longitude of pixel;

S102. according to the cloud mask item in each the sensors observe data that finds, select fine observation data (being valid data), extract earth surface reflection rate, the sun and sensors observe angle and the NDVI of this sensor.

S2. the observation data of extracting among the integrating step S1 uses the wave band conversion to make up multisensor observation data collection.The basis of multisensor observation data associating is that sensor with multiple source is with unified standard transducer model tormulation.The wave band reflectivity of supposing other sensor can be by the linear polynomial expression of standard transducer wave band reflectivity, selecting MODIS as standard transducer, is the linearity expression of seven wave band reflectivity of MODIS with other sensor (sensor of interest) wave band reflectivity conversion.So, unified standard transducer is expressed as:

ρ _T,i=a _i,1ρ _S,1+a _i,2ρ _S,2+...+a _i,7ρ _S,7 （1）

Wherein, a be standard transducer (S) to the wave band conversion coefficient of sensor of interest (T), i is i wave band.

In the present embodiment, obtaining of wave band conversion coefficient is to use 464 vegetation soil mixing wave spectrums and 79 ice and snow wave spectrums, wave band response function integration according to different sensors obtains the wave band reflectivity, wave band reflectivity substitution formula (1) can be obtained 7 wave band conversion coefficients of sensor of interest wave band by least square.Suppose to have n bar wave spectrum, the least squares equation group is:

$\left(\begin{array}{c}{\mathrm{\&rho;}}_1\\ {\mathrm{\&rho;}}_2\\ ...\\ ...\\ {\mathrm{\&rho;}}_n\end{array}\right)=\left(\begin{array}{ccccc}{\mathrm{\&rho;}}_{\mathrm{1,1}}& {\mathrm{\&rho;}}_{\mathrm{1,2}}& ...& ...& {\mathrm{\&rho;}}_{\mathrm{1,7}}\\ {\mathrm{\&rho;}}_{\mathrm{2,1}}& {\mathrm{\&rho;}}_{\mathrm{2,2}}& ...& ...& {\mathrm{\&rho;}}_{\mathrm{1,7}}\\ ...& ...& ...& ...& ...\\ ...& ...& ...& ...& ...\\ {\mathrm{\&rho;}}_{n,1}& {\mathrm{\&rho;}}_{n,2}& ...& ...& {\mathrm{\&rho;}}_{n,7}\end{array}\right)\&times;\left(\begin{array}{c}{a}_1\\ a_2\\ ...\\ ...\\ a_7\end{array}\right)---\left(2\right)$

7 wave band reflectivity conversion that table 1 and table 2 have provided MODIS respectively are the conversion coefficient of AVHRR and VIRR sensor wave band reflectivity.

Table 1MODIS is to the wave band conversion coefficient of AVHRR

Table 2MODIS is to the wave band conversion coefficient of VIRR

S3. according to effective observation frequency, carry out the face of land two to the reflection characteristic inverting in conjunction with multisensor observation data collection.This step is united the structure system of equations by multi-source data, utilizes the multi-angle multi-wavelength data to work in coordination with 21 karyonide numbers of inverting, finishes the BRDF inverting.Its core for the collaborative inverse model of multi-angle multiband (Multi-Angular and Multi-Band Model, MAB).

Model explanation is as follows:

Formula (1) is introduced following nuclear driving model expresses:

（3）

Wherein, θ is solar zenith angle,

Be the observation zenith angle, φ is relative bearing.

Sensor of interest wave band reflectivity can be expressed by formula (4):

（4）

Wherein, N is the wave band number of standard transducer, and as MODIS during as standard transducer, N is 7.By RS data, can make up the system of equations of all the sensors wave band general purpose core coefficient.

The expression of model center:

(a) volume scattering nuclear

The volume scattering of selecting for use in model nuclear be the Roujean Luo Si thick-layer that equals to propose in 1992 (Ross Thick kernel, RT) nuclear, description be dense vegetation canopy.Set up three hypothesis in the modeling process of Roujean, the small scattering surface in background and the medium all is lambert's scattering; Scattering surface in the medium is towards being stochastic distribution; Only once scattering.And when zenith is observed (observation zenith angle and position angle are 0), the value of nuclear is normalized to 0.Luo Si thick-layer nuclear expression is as follows:

Wherein, ξ is the phasing degree,

(b) geometrical optics nuclear

The geometrical optics that model is selected for use is Li Shi transition (Li Transmit, LT) nuclear that Li Xiaowen etc. proposes.In order to reflect the feature of original optical model more accurately, do not reduce the ability of fitting data again, this is checked and uses Li Shi in little zenith angle sparse (Li Sparse, LS) nuclear then use Li Shi dense (Li Dense, LD) nuclear when zenith angle is big.Embody as follows:

Wherein:

${\mathrm{\&theta;}}^{\&prime;}={\tan }^{-1}\left(\frac{b}{r}\tan \mathrm{\&theta;}\right)$

Wherein, Described the ratio of the vertical semiaxis of tree crown spheroid and horizontal semiaxis, general value is 1.0; Describe the overhead ratio of height and the vertical semiaxis of tree crown ellipsoid of tree crown spheroid center, general value is 2.0.

In the present embodiment, step S3 specifically comprises:

S31. judge effective observation frequency: if more than or equal to 49, execution in step S32 then; If more than or equal to 21 and less than 49, execution in step S33 then;

S32. in conjunction with multisensor observation data collection, directly use multi-angle multiband model to carry out complete inverting, because model is linear model, can find the solution by least square.Specifically describe as follows:

The associating multisensor makes up the multi-angle observation data, supposes to have in the inverting cycle earth surface reflection rate observation data n time.The least squares equation group of finding the solution can be expressed as:

Wherein, R represents the column vector of reflectivity observation.Use two to reflection model formula (4), system of equations can be rewritten into expression matrix:

R=K·F （10）

Wherein, K is nuclear matrix, is expressed as:

（11）

F is the column vector of 21 karyonide numbers of model parameter, is expressed as:

$F=\left(\begin{array}{c}{f}_{\mathrm{iso},1}\\ f_{\mathrm{vol},1}\\ f_{\mathrm{geo},1}\\ ...\\ ...\\ f_{\mathrm{iso},7}\\ f_{\mathrm{vol},7}\\ f_{\mathrm{geo},7}\end{array}\right)---\left(12\right)$

S33. in conjunction with multisensor observation data collection, introduce the inverting of priori according to multi-angle multiband model, namely during the observation information quantity not sufficient, introduce priori and carry out inverting.Refutation process mainly comprises:

S331. obtain the global face of land two to the reflection characteristic priori;

The face of land two, the whole world is to the reflection characteristic product based on MODIS two to the reflection characteristic priori, two of branch latitude zone (5 ° of intervals) statistics three kinds of surface vegetations, soil and ice and snow are average and the covariance matrix of many decades to reflective core, i.e. expectation and uncertain.

The judgment basis of face of land type is: judge that for ice and snow ruddiness or blue wave band reflectivity are greater than 0.3; Judge non-ice and snow and NDVI 〉=0.2 for vegetation; Judge non-ice and snow and NDVI＜0.2 for soil.

S332. make up model priori observation data;

Use the global face of land two to the reflection priori, can construct 21 priori observation.Its method is as follows:

Its cost function of linear inversion problem is:

Cost(f)=(Af-Y _obs) ^T(Af-Y _obs) （13）

The maximum likelihood solution of cost function is:

X=[A'A] ^-1A'Y _obs （14）

Because the restriction of observation geometric condition, make [A ' A] ^-1Might become nonsingular matrix, thereby make finding the solution of unknown vector X become difficult.Therefore, in the process of finding the solution unknown vector X, introduce priori.Based on Bayesian inference, the tentation data noise is Gaussian distribution, and the prior probability distribution of unknown parameter is Gaussian distribution, and then the maximum a posteriori probability solution is the extreme point of the cost function of following adding priori:

$\mathrm{Cost}\left(f\right)={(\mathrm{Kf}-{\mathrm{\&rho;}}_{\mathrm{obs}})}^T{\mathrm{\&Sigma;}}^{-1}(\mathrm{Kf}-{\mathrm{\&rho;}}_{\mathrm{obs}})$

$+{(f-f_{\mathrm{prior}})}^T{\mathrm{\&Delta;}}^{-1}(f-f_{\mathrm{prior}})$ （15）

Wherein, Σ is the covariance matrix of observation data noise and model uncertainty; Δ is the covariance matrix of priori.

Consider that Σ is difficult to obtain the introducing with priori, rewrites cost function:

$\mathrm{Cost}\left(f\right)={n(\mathrm{Kf}-{\mathrm{\&rho;}}_{\mathrm{obs}})}^T(\mathrm{Kf}-{\mathrm{\&rho;}}_{\mathrm{obs}})$

$+{(f-f_{\mathrm{prior}})}^T\left({K_{\mathrm{simu}}}^T{K}_{\mathrm{simu}}\right)(f-f_{\mathrm{prior}})$ （16）

Wherein, K _Simu' K _Simu=Δ ^-1N is weight.

N depends on the degree of confidence of observation data and priori, and namely n has a bowel movement and more levels off to the observation data solution, estimates away from priori.The method that adds priori can be by simulation BRDF data, and construction solution (expectation) is that the priori karyonide is counted average statistical f _PriorPriori nuclear matrix and reflectivity matrix add inverting:

K _simuf=ρ _simu （17）

The covariance matrix of priori carries out following decomposition:

Δ=E*∧*E' （18）

The priori reflectivity observation data parameter (nuclear matrix and reflectivity matrix) that can obtain constructing is:

K _simu=∧ ^-1/2*E' （19）

ρ _simu=∧ ^-1/2*E'*f _prior （20）

S333. the model priori observation data of weighting is introduced multisensor observation data collection, carry out inverting according to multi-angle multiband model, can increase quantity of information, make inverting more stable.The inverting matrix of introducing priori can be expressed as:

The least square method for solving that using is all-trans drills can solve 21 karyonide numbers of model parameter.

After step S33, the present invention has also introduced quantity of information and has optimized the multi-sensor data collection; Be step S4: the new sensors observe data of introducing are carried out the net information content evaluation and optimized multisensor observation data collection and carry out inverting.

The quantity of information source of carrying out inverting according to the collaborative inverse model of multi-angle multiband comprises two parts, namely increases deterministic observation angle and reduces deterministic observational error.Multi-sensor data two in the reflection characteristic inverting, introduce new sensing data source, increase the multi-angle observation data that participate in inverting and might not promote inversion accuracy.The new data source of introducing can be estimated by net information content (being also referred to as information index) for the determinacy influence of inverting, and the one-step optimization multisensor observation data of going forward side by side collection carries out inverting.

Given two probability density function f ₁(x) and f ₂(x), f ₁(x) with respect to f ₂(x) relative entropy (being also referred to as net information content or information index) may be defined as:

$I(f_1:f_2)={\&Integral;}_{\mathrm{\&Psi;}}{f}_1\left(x\right)\log \frac{f_1\left(x\right)}{f_2\left(x\right)}---\left(22\right)$

For the multi-sensor data inverting, namely to estimate by following formula and introduce the quantity of information that new data source (sensor or wave band) is introduced, information index is defined as follows:

I _net=I _multisensors-I _monosensor （23）

Wherein, quantity of information I _MultisensorsAnd I _MonosensorThe formula that is calculated as follows:

$I=\underset{1}{\overset{n}{\mathrm{\&Sigma;}}}\ln \left({\mathrm{\&lambda;}}_i\right)-\ln \left(\mathrm{MSE}\right)---\left(24\right)$

K'K=G'VG （25）

Wherein, K is nuclear matrix (formula 11), decomposes to obtain eigenmatrix V, and λ i is the eigenwert of eigenmatrix V.MSE is the mean square deviation sum that participates in each wave band reflectivity of inverting.

In the inverting of step S33, suppose that inversion result karyonide number brings the reflectivity that model formation (4) match obtains into, relative error was the error wave band greater than 50% o'clock; Recursion step S33 selects the wave band of relative error maximum at every turn, if remove greater than 50%, finally obtains the model karyonide number of preliminarily stabilised.Select the observation of MODIS sensor and the INTEGRATED SIGHT of multiple sensors to calculate its quantity of information respectively according to formula (24) then, further calculate poor (net information content) of its quantity of information.If net information content shows that for just the sensors observe data of introducing increase the inverting quantity of information, then this karyonide number is final mask karyonide number; If quantity of information then removes this sensors observe data for negative, recomputate model karyonide number according to step S33.

The present invention also provides a kind of and has realized that the above-mentioned face of land two based on multisensor is to the system of reflection characteristic inversion method, as shown in Figure 2, comprise that mainly the multisensor observation data extraction module, the multisensor observation data collection that set gradually make up module, the face of land two to reflection characteristic inverting module and model result output module; In the present embodiment two comprises also that to reflection characteristic inverting system being arranged on the face of land two optimizes the inverting module to the quantity of information between reflection characteristic inverting module and the model result output module.Below each module is illustrated respectively.

Multisensor observation data extraction module:

This module is used for extracting the wave band reflectivity of MODIS, AVHRR and VIRR sensing data, angle information, cloud mask and NDVI.

Multisensor observation data collection makes up module:

This module is used and is returned the different sensors wave band conversion coefficient that obtains, and the linearity that the multisensor observation watch is shown 7 wave bands of standard transducer (MODIS) is expressed, thereby makes up the multi-angle multiband observation data collection of multisensor.

Two to reflection characteristic inverting module:

This module is called the complete inverting module of submodule and is introduced priori inverting module according to effective observation data of participating in inverting.Complete inverting submodule is geometrical optics nuclear and the volume scattering nuclear matrix of computation model at first, makes up multi-angle multiband observation equation group.Use least square method to find the solution the multi-angle multiband observation equation group of multisensor, calculate two to reflection characteristic model karyonide number.Introduce priori inverting submodule and at first construct 21 group two to reflection priori analogue observation, the geometrical optics of computation model is examined and the volume scattering nuclear matrix then, make up the multi-angle multiband observation equation group of introducing priori, use least square method to calculate two to reflection characteristic model karyonide number.

Quantity of information is optimized the inverting module:

This module is for the bigger observation of rejecting error, and whether estimate the new sensor of introducing useful to inverting, thus the precision of the model karyonide number that the raising inverting obtains.

The model karyonide is counted output module:

This module is used for 21 model karyonide numbers will calculating, outputs to ENVI(The Environment for Visualizing Images, remote sensing image processing platform with 21 wave bands) in the standard format files.

Also experimental verification has been carried out to reflection characteristic inversion method and system in the face of land two based on multisensor that provides in the present embodiment.

Test is to have contrasted albedo and the main flow albedo product MCD43B3 that result of calculation of the present invention is produced, and calculates root-mean-square error with the true albedo of ground actual measurement respectively and estimate its precision separately.Temporal resolution that it should be noted that MCD43B3 albedo product is 8 days, and the temporal resolution of the MAB albedo product among the present invention has been brought up to 4 days.Test findings shows that the albedo that the karyonide number of use inverting of the present invention calculates can keep close precision with respect to existing main flow albedo product, even in some instances, also improve precision under the prerequisite that improves temporal resolution.

Experiment one:

Select the website Fort_Peck(48.308 ° N of ground measured data for use, 105.101 ° of W) the ice and snow coverage period estimates effect of the present invention.The input data are MODIS and the AVHRR sensing datas in the Fort_Peck website ice and snow cycle in 2003.Experimental result is as follows:

The experimental result of table 3 experiment one

Website	The time phase	RMSE MCD43B3	RMSE MAB
				Fort_Peck	Ice and snow	0.1648	0.1221

Experiment two:

Select the website Fort_Peck(48.308 ° N of ground measured data for use, 105.101 ° of W) non-ice and snow coverage period estimates effect of the present invention.The input data are MODIS and the AVHRR sensing datas in the Fort_Peck website non-ice and snow cycle in 2003.Experimental result is as follows:

The experimental result of table 4 experiment two

Website	The time phase	RMSE MCD43B3	RMSE MAB
				Fort_Peck	Non-ice and snow	0.0223	0.0172

Experiment three:

Select the website ARM_SGP_Main(36.605 ° N of ground measured data for use, 97.488 ° of W) the ice and snow coverage period estimates effect of the present invention.The input data are MODIS and the AVHRR sensing datas in the ARM_SGP_Main website ice and snow cycle in 2004.Experimental result is as follows:

The experimental result of table 5 experiment three

Website	The time phase	RMSE MCD43B3	RMSE MAB
				ARM_SGP_Main	Ice and snow	0.1978	0.1358

Experiment four:

Select the website ARM_SGP_Main(36.605 ° N of ground measured data for use, 97.488 ° of W) non-ice and snow coverage period estimates effect of the present invention.The input data are MODIS and the AVHRR sensing datas in the ARM_SGP_Main website non-ice and snow cycle in 2004.Experimental result is as follows:

The experimental result of table 6 experiment four

Website	The time phase	RMSE MCD43B3	RMSE MAB
				ARM_SGP_Main	Non-ice and snow	0.0242	0.0287

In sum, the face of land two based on multisensor provided by the present invention has solved the problem of the face of land two inverting quantity of information deficiency in the reflection characteristic to the reflection characteristic inversion method, and effectively raises resolution and the precision of the time of inverting.

Above embodiment only is used for explanation the present invention; and be not limitation of the present invention; the those of ordinary skill in relevant technologies field; under the situation that does not break away from the spirit and scope of the present invention; can also make a variety of changes and modification, so all technical schemes that are equal to also belong to protection category of the present invention.

Claims (10)

1. One kind based on the face of land two of multisensor to the reflection characteristic inversion method, it is characterized in that, comprise step:

   S1. extract the observation data of multisensor in the inverting cycle;

   S2. in conjunction with the observation data of extracting among the described step S1, use the wave band conversion to make up multisensor observation data collection;

   S3. according to effective observation frequency, carry out the face of land two to the reflection characteristic inverting in conjunction with multi-angle multiband model and multisensor observation data collection.
2. 2. the face of land two according to claim 1 is to the reflection characteristic inversion method, it is characterized in that, described multisensor comprises: Moderate Imaging Spectroradiomete, high resolution scanning radiometer and visible light infrared scanning radiometer, described multisensor observation data comprises: earth surface reflection rate, the sun and sensors observe angle, cloud mask and NDVI.
3. 3. the face of land two according to claim 1 and 2 is characterized in that to the reflection characteristic inversion method, and described step S1 comprises:

   S101. be benchmark with Moderate Imaging Spectroradiomete pixel position, the observation data of searching other sensor that is complementary according to the center longitude of pixel;

   S102. according to the cloud mask item in each the sensors observe data that finds, if judge fine observation data, extract earth surface reflection rate, the sun and sensors observe angle and the NDVI data of this sensor.
4. 4. the face of land two according to claim 3 is characterized in that to the reflection characteristic inversion method, and described step S2 comprises:

   S201. use actual measurement ground-object spectrum data to return by least square method and obtain the wave band conversion coefficient that the different target sensor is expressed with the standard transducer unification;

   S202. according to described wave band conversion coefficient, each sensor with unified standard transducer model tormulation, is made up multisensor observation data collection.
5. 5. the face of land two according to claim 4 is characterized in that to the reflection characteristic inversion method, and described step S3 comprises:

   S31. judge effective observation frequency: if be not less than 49, execution in step S32 then; If be not less than 21 and less than 49, then execution in step S33;

   S32. in conjunction with multisensor observation data collection, carry out complete inverting according to multi-angle multiband model;

   S33. in conjunction with multisensor observation data collection, introduce the inverting of priori according to multi-angle multiband model.
6. 6. the face of land two according to claim 5 is characterized in that to the reflection characteristic inversion method, and described step S33 comprises:

   S331. obtain the global face of land two to the reflection characteristic priori;

   S332. utilize the global face of land two to the reflection characteristic priori, make up model priori observation data;

   S333. the weighting of model priori observation data is introduced multisensor observation data collection, carried out inverting according to multi-angle multiband model.
7. 7. the face of land two according to claim 6 is characterized in that to the reflection characteristic inversion method, also comprises after the described step S33:

   S4. according to net information content, inversion result is estimated and optimized described multisensor observation data collection.
8. 8. the face of land two according to claim 7 is characterized in that to the reflection characteristic inversion method, and described step S4 comprises:

   S41. according to inversion result among the described step S33, match obtains reflectivity, recursion step S33, and the wave band of the relative error maximum of at every turn selecting if remove greater than 50%, finally obtains the model karyonide number of preliminarily stabilised;

   S42. use the model karyonide number of preliminarily stabilised, together with standard transducer observation data collection and newly-increased sensors observe data set, calculate quantity of information separately respectively; The difference of multi-sensor information amount and standard transducer quantity of information is net information content:

   If just, then this model karyonide number is final mask karyonide number;

   If negative, then remove this sensors observe data, recomputate model karyonide number according to described step S33.
9. 9. a realization is according to the system of any described face of land two based on multisensor of claim 1-8 to the reflection characteristic inversion method, it is characterized in that, comprising: the multisensor observation data extraction module that sets gradually, multisensor observation data collection make up module, the face of land two to reflection characteristic inverting module and model result output module;

   Described multisensor observation data collection makes up module and makes up multisensor observation data collection according to the observation data use wave band conversion that described multisensor observation data extraction module extracts; Described model result output module export the described face of land two to the face of land two that reflection characteristic inverting module is carried out in conjunction with described multisensor observation data collection to the reflection characteristic inversion result.
10. 10. the face of land two according to claim 9 is characterized in that to reflection characteristic inverting system, comprises that also being arranged on the described face of land two optimizes the inverting module to the quantity of information between reflection characteristic inverting module and the model result output module;

    Described quantity of information is optimized the inverting module according to net information content, and inversion result is estimated and optimized described multisensor observation data collection.

Images