radiometric, atmospheric and geometric preprocessing optical earth observed images

8/9/2019 Radiometric, Atmospheric and Geometric Preprocessing Optical Earth Observed Images

1/15

School ofGeographyUniversity of Leeds

2004/2005Level 2LouiseMackay

EARTH OBSERVATION AND GIS OF THE PHYSICAL ENVIRONMENT –

GEOG2750

Week 4

Lecture 4: Radiometric, Atmospheric and Geometric Pre-Processing of

Optical Earth Observed mages

Aims: This lecture will introduce you to the most common forms of processing that

need to be performed before Earth observation images can be used quantitatively to

characterise and map the Earth’s terrestrial environment. It covers the stages involved

in converting (inverting digital Earth observed multispectral images bac! to estimates

of at"surface reflectance. It also considers the most commonly employed approach for

the correction of geometric errors and registration of Earth observed images to digital

map data sets used in many #I$ applications.

%ecture & consists of the following topics:

'. hy pre"process remotely sensed digital images. ). *adiometric calibration.

+. Atmospheric correction approaches.

&. #eometric correction and image registration.

,. -onclusion /re"processing top"tips.

or! through each of the main topics in order. A reading list is supplied at the end of

the content or can be accessed at http:00lib,.leeds.ac.u!0rlists0geog0geog)1,2.htm.

!egin Lecture 4 here

"h# pre-process remotel# sensed digital images$

*arely are remotely"sensed images provided where they can be directly applied to an

environmental application. 3or e4ample5 emitted radiance from a surface travels

through the atmosphere6 this process can both attenuate and add to sensor radiance.

#eometric distortions also result from flightline orientation or orbit characteristics.An image may e4hibit all of these undesirable features which often depend on the

'

http://lib5.leeds.ac.uk/rlists/geog/geog2750.htmhttp://lib5.leeds.ac.uk/rlists/geog/geog2750.htmhttp://lib5.leeds.ac.uk/rlists/geog/geog2750.htm


2/15

supplier’s level of pre"processing. In certain cases they must be removed before

inferring or estimating properties about the Earth’s surface:

• $uch as when comparing quantitatively images5 estimated properties or cover

types acquired at different dates and0or

• hen wishing to derive properties that can only be estimated from the Earth’ssurface reflectance characteristics.

78TE: it is therefore necessary to apply Radiometric %alibration and Atmospheric

%orrection&

Radiometric calibration

Radiometric 'pectral-!and %alibration

A remotely sensed image comprises of a series of spectral bands5 the pi4els of which

each have a digital number ()*+. /i4el 97 is a linearly transformed representation

of at-sensor radiance for a discrete resolved area of the Earth’s surface (3igure '.

owever5 some studies and pre"processing operations need at"sensor radiance. $o the

radiance"to"97 procedure of image acquisition for each spectral wavelength must be

inverted (3igure ).

3igure ': 97"to"radiance 3igure ): *adiance"to"97

)


3/15

To summarise the relationship between

radiance (% and pi4el 97 loo! at 3igure

+ to the right. As pi4el 97 is a simple

linear transformation of radiance (3igures

' and )5 the slope and offset of this

linear transformation (which is specificfor each spectral band5 each sensor and

initial calibration can be used to

calculate radiance (% and inversely used

to calculate pi4el 97. 7ote the

dimensions of at"sensor radiance. ;ou

can use the reading list to chec! the unit

of measurement for the symbols you are

not familiar with.

/oints to 7ote:

(i The gain and offset values are unique

for each spectral band acquired by a

particular sensor.

(ii These are often provided with the

images and for common sensors can be

found in the main te4ts.

(iii owever5 these values change over

the life span of a sensor " so ma!e sure

you have the most recent.

(iv


4/15

At-'ensor 'pectral Reflectance

It is possible to convert at"

sensor radiance to apparent at"

sensor spectral reflectance "

required before atmosphericcorrection.

At sensor"reflectance involves

ta!ing into account temporal

changes in solar illumination

due to Earth"$un geometry. As

can be seen by 3igure & to the

right the Earth"$un geometry

changes with time of year.

3igure &: Earth"$un geometry seasonal changes.

Atmospheric correction approaches

At"sensor reflectance still has atmospheric scattering effects present.

In some studies atmospheric effects may have to be removed by deriving ratios of

different spectral"bands. This can be underta!en in two ways: by using reflectance in

physical"based environmental models6 or by using reflectance as the basis of change

detection over time.

The aim of atmospheric correction is to derive a good estimate of the true at"ground

upwelling radiance (reflectance. Approaches commonly employed in remote sensing

for atmospheric correction are:

'. -omple4 physical modelling this can be underta!en using specialised

software pac!ages.

). $emi"empirical modelling " this can be underta!en using specialised software

pac!ages.

+. Empirical estimation approaches " based on image data relationships.

Each of these approaches will now be considered in turn.

%omple Ph#sical odelling

/hysical modelling is based on a mathematical understanding of atmospheric

radiation transfer theory and atmospheric scattering. /hysical modelling requiresdetailed estimates of atmospheric state i.e. optical thic!ness and atmospheric aerosols.

&


5/15

The models are data intensive5 require field data5 validation and are computationally

intensive.

)ata collection:

This can beunderta!en using

an automatic sun

trac!ing

photometer (see

3igure , right

which measures

optical thic!ness

and atmospheric

aerosols.

.alidation:

This can be

underta!en by

using ground"

based collection

of reflectance

using a

spectrometer

(3igure = right.

'emi-Empirical odelling

This approach uses the same comple4 models but for a significantly reduced set of

input parameters. In the simplest case5 correction can be based on an estimate of

atmospheric visibility and standard atmospheric constants for latitude0longitude0date.

owever5 quite different results can be obtained for different estimates e.g.5

atmospheric visibility. 3or e4ample5 the following figures (3igures 1 and > show the

change of /i4el to $urface reflectance for different estimates of atmospheric visibility.

3igure 1: ,?m visibility 3igure >: )2!m visibility

In these two figures you may notice that pi4el reflectance is decreased with low

atmospheric visibility.

,


6/15

Empirical Estimation Approaches

Empirical estimation approaches use only the image data to remove atmospheric

effects. They only correct for atmospheric path radiance which is at"sensor radiance

contributed by atmospheric scattering. Additionally5 such approaches only account

for *ayleigh atmospheric scattering.

%orrection procedure:

(i The image must have pi4els of

e4pected @ero reflectance in the

visible and 7I* (3igure right.

(ii %inearly regress each visible

band against near infrared for these

pi4els.

(iii The B"a4is offset is considered

to be an e4pression of path

radiance.

(iv $ubtract the B"a4is offset from

the visible band.

7ote: ;ou will apply this approachin your atmospheric correction

practical.

3igure : Cero reflectance in the visible and

7I*.

Geometric correction and image registration

Geometric %orrection

Images are rarely provided in the correct proDection5 coordinate system and free of

geometric distortions. #eometric distortions occur due to Earth rotation during

acquisition and due to Earth curvature.

In order to combine images with other digital map data"sets they must be transformed

from the acquisition coordinate system (rows0cols to that of the digital map data sets.

3or well defined orbital geometry parameters this can be achieved using predefined

transformations (see ather p1="1> which model the aspect5 s!ew and rotational

distortions of a sensor. These terms are defined as follows:

'. Aspect ratio " unequal across"trac! and along"trac! pi4el scale.

). $!ew " image s!ew with respect to Earth’s north0south a4is.

=


7/15

+. *otation " column coincident pi4els do not have same longitude.

owever5 not all possible causes of geometric distortion can be modelled using these

predefined transformations " therefore a more general approach is required e.g.5 using

ground control data.

%orrection using Ground %ontrol

-orrecting for geometric distortions using ground control is a multi"step process that

involves the following procedures:

'. *ecognising in the image and map enough locations which are the same in

order to accurately e4press the distortion present.

). Fsing these locations to calculate the transformation between the image and

the map.

+. Actually performing the transformation to the new coordinate system for each pi4el to generate a new image.

&. -alculating the actual 97 that should be assigned to each pi4el in the new

image.

,. Estimating how well the procedure has actually performed.

e will now cover these important steps of geometric correction.


#round control correction is based on recognising points !nown in a map on the

image to model distortion and define an empirical transformation between the image

and the map as represented by 3igure '2 below. here the left"hand map represents

digital map data where ground points are recognised and used to model distortion

from reciprocal points on the image data on the right.

1


8/15

3igure '2: An illustration of the use of ground control points from verification data

(left to geo"correct image data (right.


There are a certain number of points to remember when recognising and choosing

ground control points (#-/s:

• They should be distinct features in both the map and image.

• They should have clear boundaries and regular geometry.

• #ood #-/s would be house corners5 road intersections5 field corners etc.

•


9/15

3igure '': %east squares estimation summarised in matri4 notation.

The previous derivation in 3igure '' is a first"order polynomial which corrects for

scaling5 rotation and shearing. igher"order polynomials (c and r raised to a power

can also be used to correct for warping (bending.

Assessing Geometric Accurac#

The co"ordinates resulting from a transformation are an estimate and need to bechec!ed for their degree of accuracy. Accuracy is normally e4pressed for each control

point and overall root mean square. 3or e4ample5 in Table ' below control points '

and ) have been evaluated for the degree of error they e4hibit in both the B and ;

direction5 producing an individual point error which then contributes to the overall

*$ error of the transformation. igh *$ values (H ' pi4el can indicate to the

user how inaccurate the transformation is and whether the chosen #-/s were

adequate5 the process can then be repeated with additional0different #-/s in order to

decrease the point and total error.

*ote: a high *$ error for a particular point can indicate to the user a #-/ that could

be removed or replaced.

Table ': *$ point and total error for a ) point e4ample

/oint Error B Error y /oint Error

' "'.,&)2)+ '.+&=,>> ).2&1))=

) "2.=2&' '.>+1)>2 ).21+'=

*$ '.2,>2 (pi4el

7ote: This e4ample in Table ' is only for !nown points6 it tells us nothing aboutun!nown locations.


10/15

Resampling

8nce the least squares coefficients have been derived and the (45y calculated for a

pi4el5 a 97 must be assigned to it. This is done by a reverse least squares estimation.

owever5 the resulting col0row of an B0; is li!ely not to be an integer as the corrected

pi4el lies across two or more pi4els in the raw image (as illustrated in 3igure ') below. $o we need to estimate the new 97 value by interpolation this is called

resampling.

3igure '): /i4el interpolation after geometric correction.

'2


11/15

Resampling

Three interpolation routines can be used for resampling the image.

(i 7earest neighbour:

• In this procedure a pi4el is selected that is nearest to the estimated col0row of the reverse least squares:

• The advantage of this approach is that it retains the original image histogram

(distribution.

• The disadvantage of this approach is that visually the resulting images can be

Jbloc!yJ.

(ii


12/15

3igure '+: After pre"processing the image may need clipping to a cover a specific

study area.

0he Result and the End

The following figure (3igure '& represents AT data that has been geometrically

corrected using a second"order least squares polynomial5 using +) ground control

points resulting in a *$ of '. pi4els5 nearest neighbour resampling and clipped to

match 8$ %andline (the left"hand figure. The final image after this pre"processing is

illustrated on the right.

3igure '&: An AT pre"processed image (right5 with digital land cover data (left.

%onclusion 2 Pre-processing top-tips

')


13/15

!efore starting a pro3ect ala#s:

• 9erive as much information from the vendor on what has been done to the

image supplied to you.

• -hec! carefully the header and all files supplied with the image.

• Kisually chec! the image for visual atmospheric effects e.g.5 visible ha@e can

occur in the blue spectral"band.

Evaluate #our needs:

• If not performing change detection or ratios you do not need to correct for

atmospheric effects.

• If producing an image for visual overlay with map data then select bicubic

resampling


14/15

+. ather5 /..5 '. -omputer processing of remotely"sensed images. -hapter

&5 pages >1", (E4cellent readable coverage5 including terrain correction

which is not covered in the lecture.

Lournal Articles:

'. -have@5 /.$.5 '>>. An improved dar!"obDect subtraction technique for

atmospheric scattering correction of multi"spectral data. *emote $ensing of

Environment5 )&5 &,"&1.

). -have@5 /.$.5 '=. Image based atmospheric corrections revisited and

improved. /hotogrammetric Engineering and *emote $ensing5 =)5 '2),"'2+=.

+. 9uggin5 .L.5 '>,. 3actors influencing the discrimination and quantification

of terrestrial features using remotely"sensed radiance. International Lournal of

remote $ensing5 =5 +")>.

&. ill5 L.5 and Aifadopoulou5 9.5 '2. -omparative analysis of %andsat", Tand $/8T *K"' data for use in multiple sensor approaches. *emote $ensing

of Environment5 +&5 ,,"12.

,. ar!ham5 .

1. /rice5 L.-.5 '&. An update on visible and near infrared calibration of satellite

instruments. *emote $ensing of Environment5 )&5 &'"&)).

>. Teillet5 /..5 and 3edoseDevs5 #.5 ',. 8n the dar! target approach to

atmospheric correction of remotely"sensed data. -anadian Lournal of *emote

$ensing5 )'5 +1&"+>1. (


15/15

'. ?ardoulas5 7.#.5

radiometric, atmospheric and geometric preprocessing optical earth observed images

Documents