Remote Sensing
Remotely sensed
imagery, including multispectral, hyperspectral, thermal infrared, and
panchromatic, is one of the most powerful mapping tools that we currently have.
Remotely sensed data can greatly enhance GIS databases.
If you did not
take Intro GIS or even if you did, you may want to review http://www5.egi.utah.edu/GIS__CVEEN/Remote_Sensing/remote_sensing.html
before starting this assignment.
Some Terms
1. Spatial
resolution: The amount of the Earth’s surface represented in a single
pixel, i.e. an image with 30 m spatial resolution has pixels that each
represent 30 m on the ground.
2. Sampling
interval: the band-widths produced by a remote sensing instrument.
Smaller sampling intervals produce better the spectral resolutions. Spectral
resolution is roughly the full-width half-maximum (FWHM) of the sampling
interval.
3. BV =
brightness value (also called DN for digital number)
3. Visible:
400 – 700 nm
4. Near
infrared: 700 – 1000 nm
5. Shortwave
infrared (SWIR): 1000 nm – 3000 nm
6. Long wave
infrared or thermal infrared (TIR): 8 μm – 14 μm
7. Microwave
or radar: 1000 μm - ~ 10 m
Available Data Types
Remotely sensed
data is usually broken-down into several categories including:
1. Multispectral:
Multispectral sensors produce relatively broad bands (medium to low spectral
resolution) ranging from low to high spectral resolution. This data can be
either airborne or satellite derived. Some examples include LandSat, ASTER, and
SPOT. Multispectral data can cover visible, near infrared, shortwave infrared,
and thermal infrared spectral regions. Multispectral data is generally
relatively inexpensive.
2. Hyperspectral:
Hyperspectral sensors produce relatively narrow bands (high spectral
resolution) and are generally airborne platforms. Some examples are AVIRIS and
HyVista HyMap. Hyperspectral data generally covers visible, near infrared, and
shortwave infrared spectral regions. Hyperspectral data can be expensive.
3. Panchromatic:
Panchromatic data has a single broad-band. This data is similar to a black and
white photograph. The spatial resolution runs from medium to very high. Some
examples are LandSat 7 ETM (15 m) and QuickBird (0.6 m). This data is produced
from both satellite and airborne platforms. The cost of pan data ranges from
free to moderately expensive.
4. Radar: Radar
data sets generally cover one of several subsets of the microwave region. For
example X-band (~3.0 cm), C-band (~5.5 cm), and L-band (~27cm). Radar datasets
have widely varying spatial resolution. Some examples are RadarSat and JPL’s
Imaging Radar program. Radar data is priced from low to moderate cost.
This bit of info
plus the review at http://www5.egi.utah.edu/GIS__CVEEN/Remote_Sensing/remote_sensing.html
gives you enough information to get started on your introductory remote sensing
assignment, but this introduction is not intended to educate you on effectively
using remote sensing tools. If you have an interest in remote sensing and would
like to develop skills in this area, several classes are offered by the
Geography Department.
Assignment
This assignment
will use the Raster Calculator in ArcGIS 8.3 Spatial Analyst. The objectives
will be to preprocess a multispectral image to prepare it for merging with a
high spatial resolution panchromatic image to enhance the spatial resolution of
the multispectral image. Additionally, we will look at using band ratios to
enhance surface features based upon their spectral absorption. The
assignment will be in three parts.
NOTE: Keep data (workspace) paths
and grid names simple – no spaces and short.
Part 1: Simple Atmospheric Correction and
Spatial Enhancement
First, we will
recalculate our spectral values in the multispectral data set to help
ameliorate deleterious atmospheric conditions. The method we will be using is
called Dark Object Subtraction (DOS). This method is very simple and is not as
effective as more complex methods, such as MODTRAN (MODerate
spectral resolution atmospheric TRANSsmittance algorithm). However, DOS
often noticeably improves multispectral imagery.
DOS is based upon
the assumption that an object exists in an image having a brightness value (BV)
greater than one, but which should actually have a BV of 0. Our data set is
Advanced Spaceborne Thermal Emission and Reflection Radiometer (ASTER)
registered radiance at the sensor with an 8 bit quantization level. Therefore,
our pixel brightness values (radiance) are represented with a range from 0 –
255. These data consist of 9 bands covering the Visible and near infrared (VNIR
– 3 bands) and SWIR (6 bands). The VNIR data are 15 m spatial resolution as are
the SWIR data. However, the SWIR data were originally 30 m spatial resolution
and were resampled to 15 m. See http://asterweb.jpl.nasa.gov/characteristics.asp
for ASTER details (this is a must).
As you learned in
the review or in the intro GIS course, there are atmospheric windows which
allow reflected electromagnetic radiation to be detected by sensors in space.
Sensors are generally designed to accommodate these windows. However,
atmospheric scattering can still affect the data. Rayleigh scattering
affects short wavelengths (primarily blue and UV) though interaction with small
particles -- smaller than the affected wavelengths -- such as atmospheric
nitrogen molecules. This is the type of scattering that makes the sky appear
blue.
Mie
scattering, which affects
longer wavelengths, is caused by larger particles, such as dust and
water molecules.
DOS works best on images acquired on clear days with low atmospheric
water. And although the ASTER data do not contain a blue visible band, the
technique can be applied to the VNEAR bands. To accomplish this using ArcGIS
download the following data:
Store the folders
in this data set on a path with folder names no longer than 8 characters and no
spaces. For example: F:\remsen. You must then set this to the working folder in
Spatial Analyst > Options.
The data set contains 10 grids -- 9 discrete bands (Gisc1 - Gisc9) and a 9-band
image (GIS). The discrete bands represent, in order,VNEAR and SWIR spectral
regions.
Dark Object Subtraction (haze reduction)
To apply the DOS
method add GISC1, GISC2, GISC3 grids to the table of contents using ArcCatalog.
These represent ASTER VISNEAR bands 1, 2, & 3.
You now must
determine the BV where the histogram for each band begins to rise rapidly (this
will be the DOS value). You can view the histogram by clicking the Symbology
tab in Properties. Record this value. The DOS value be subtracted from
the green, red, and NIR bands using Raster Calculator.
Open Raster
Calculator and enter an expression for each band. For example, let’s say the
DOS value for GISC1 (green band) is 14:
con([gisc1] - 14
GE 1, [gisc1] - 14, 1)
This can be
thought of as: If band1 minus 14 is greater then or equal to one then band 1
minus 14 else one.
Stack the con
statements for the green, red and NIR bands and run the calculation. The
outputs will all be called “Calculation” in the table of contents – rename
(hint: the first raster calculator input will be the bottom output in the table
of contents). After renaming you can right-click each new band and Make
Permanent if wanted. If not, save your map document.
We will now
stretch the data to cover a full 8-bit range:
Conceptually:
((band - global minimum)/(global maximum – global minimum)) * 255
For the new DOS
corrected green band you would use the raster calculator as follows:
(float([BandName]
- 1) / (87 - 1)) * 255 - where BandName is your new DOS green band
Data Fusion
You can now work
on data merging. First add the Pan data to the table of contents. To
merge the data, we first need to resample the new DOS data to 1 m. Go into the Spatial
Analyst > Options and change the cell size to that of the Pan
data. Now, using Raster Calculator, enter the three calculated bands with no
operators. This will create three new bands with 1 m pixels.
We will use the
Brovey Method to merge the data sets. The formula is as follows:
Rnew = (R/R+G+B) * Pan
GNew = (G/ R+G+B) * Pan
Bnew = (B/ R+G+B) * Pan
Where R = the band to be assigned to the red color gun, G = the band to be
assigned to the green color gun, and B = the band to be assigned to the blue
color gun. For this assignment this will equal:
R = GISC3
G = GISC2
B = GISC1
The calculation to use in
Raster Calculator looks like this:
(Float([b3])/(Float([b1]) + Float([b2]) + Float([b3]))) * Float([pan])
(Float([b2])/(Float([b1]) + Float([b2]) + Float([b3]))) * Float([pan])
(Float([b1])/(Float([b1]) + Float([b2]) + Float([b3]))) * Float([pan])
Where b1, b2, b3
= your DOS processed green, red, and near infrared input bands.
Use Raster Calculator to create
3 new merged bands.
After the merge
you can combine the new bands into one three-band image using the MAKESTACK
command. The syntax is as follows:
MAKESTACK Stack1
LIST [Band1] [Band2] [Band3] … etc. – make sure to leave a space between input
bands.
Where Stack1 is the path and name of the
output file, e.g. f:\gis\remsen\bndstack – the path must be the same as that of
the input layers and all must reside in a single folder (workspace).
Now try changing
the band combinations in Symbology. Also, view the original 9 band image
(GIS – add with ArcCatalog). Try different band combinations with the GIS
image. Determine the best combination and create a merged dataset using those
bands. Export to a .jpg and e-mail to gnash@egi.utah.edu.
Part 2 - Other Enhancements
There are many methods of
enhancing spectral data. Under Symbology you can stretch the data
using standard deviations. Try this using one, two, and three standard
deviations. This clips the data at the ends of the histogram and stretches
the remaining data between 0 and 255.
Aside from general enhancements such as the above stretch, specific features
can be enhanced. High spectral resolution data works best for this and can be
unmixed to reveal specific types of ground cover. However, even multispectral
data can be used for select ground cover enhancement. For example, vegetation
can be enhanced using a vegetation index, the most simple of which is derived
by dividing the near infrared maximum by the visible red minimum. For the ASTER
data you are using this would be band 3/band 2. Try this with Raster Calculator
and compare the results with ASTER Band 3 (maximum vegetation brightness band).
You will need to assign the data as Float. To view vegetation it
is customary to assign near infrared, visible red, and visible green or blue to
R, G, B color guns respectively. As ASTER does not have a visible blue band you
will use visible green. Use MAKESTACK to create a new image with ASTER bands 1
and 2 and your new vegetation enhancement band. Compare this to the raw ASTER
imagery assigned to Bands 3, 2, 1 = R, G, B. Do you see more vegetation in red
in the new image? Another, superior vegetation index (normalized vegetation
index - NDVI) is: IR -Red/IR + Red or in this case float([Band3] -
[Band2]) / float([Band3] + [Band2]) – replace the band names with your DOS
processed bands. Give it a try and save the results.
There is little vegetation in this image, so enhancement is necessary to even
see it.
Band ratios can be useful for other materials as well. For instance clay
minerals have a diagnostic absorption around ~2200 nm. This is found in ASTER
band 6 in our data set (ASTER SWIR band 3). The reflectance maximum for clays
is at around 1600 - 1700 nm -- band 4 in our data set (ASTER SWIR band 1). Try
a Band4/Band6 ratio (remembering raster calculator syntax) and determine if any
clay minerals are abundant in the image. Save the results.
I will review your results in class when you are finished.