[GRASS-user] NDVI analyses with Landsat 8

GRASS GRASS user
1

Hi there all! Greetings from a happy Grass-root GISser on an Archlinux box…

Is there already a way to analyze the new Landsat 8 images (they are already available!)?
We are in particularly interested in the Normalized Standardized Vegetation Index (NDVI). We noticed a shift in wavelength range that is best depicted here:

http://www.gisagmaps.com/landsat-8-about-refinements/
Doing the well-known raster map calculation with the bands 5 (NIR) and 4 (R) of L-8 doesn’t give a satisfying result. Very bluntly we subtracted 0.12 from the ‘ndvi’ raster derived from L-8. Without doubr there must be more elegant and correct solutions. Any ideas are very welcome

Cheers,

Huub Munstege
BPE 2836
Bamako, Rep. du Mali
Tel: +223 20226397
Port: +223 78370695

2

On Wednesday 12 of June 2013 02:05:48 Huub Munstege wrote:

Hi there all! Greetings from a happy Grass-root GISser on an Archlinux
box...

Hi!

Is there already a way to analyze the new Landsat 8 images (they are already
available!)?

In grass64 you would want to use i.landsat.toar first (with one of the
correction methods) which supports the new Landsat_8 OLI/TIRS sensor(s).

We are in particularly interested in the Normalized
Standardized Vegetation Index (NDVI).

Having (top-of-canopy) reflectances is, if I am not wrong, a *must* in order
to get the desired ration that reflects the status of the vegetation. You are,
I guess, probably interested in the "i.vi" module (with "viname=ndvi").

We noticed a shift in wavelength
range that is best depicted here:
http://www.gisagmaps.com/landsat-8-about-refinements/

Interesting, they "narrowed" the band. They probably know what they are doing.
New stuff to explore :smiley:

Doing the well-known raster map calculation with the bands 5 (NIR) and 4 (R)
of L-8 doesn't give a satisfying result. Very bluntly we subtracted 0.12
from the 'ndvi' raster derived from L-8.

Why subtract? Precisely, if I may ask, did you "threshold" it? Cut-off
anything below?

Without doubr there must be more
elegant and correct solutions. Any ideas are very welcome

Would you mind to elaborate a bit more?

Thanks, Nikos

3

Hello Nikos,

thx for the swift reply. I’ll check out the options you mentioned but at first sight I noticed that ‘i.vi’ tool is not available for me ( I’am on 6.4.3-rc, from the AUR package in Archlinux). I.vi is not a command in the ‘standard’ Grass trunk. Neither is it available as an add-on.
Some more explanation as you asked: previous Landsat 7 images we analyzed with a simple raster calculation with the following formula:

float(Band-4 - Band-3) / (Band-4 + Band-3).

But as you already noticed the bands and ranges have changed under Landsat 8. Various combinations of bands (4,5 and 8) give a result that is at best not so clear cut as the analysis done with L-7. The “bluntly subtraction with 0.12” is based on on a quick scan of the obtained result. Areas without vegetation in our project area should give values in the range between -0.05 - 0. The result of the combination of the 5 and 4 bands in the above formula gives values that are aprroximately 0.12 higher. Therefore the blunt and in-elegant subtraction which is definitely wrong. But it gives us for the time being a better deistinction between cultivated irrigation schemes and their surroundings.
Maybe I am simply too impatient and too eager to work with the new fresh data from L-8. Undoubtedly soon, Grass and the other software will incorporate specific modules that will be tailored to the L-8 sensors.

Cheers and keep up the good work,

Huub

Huub Munstege
BPE 2836
Bamako, Rep. du Mali
Tel: +223 20226397
Port: +223 78370695

De : Nikos Alexandris nik@nikosalexandris.net
À : grass-user@lists.osgeo.org; Huub Munstege hmunstege@yahoo.com
Envoyé le : Mercredi 12 juin 2013 11h23
Objet : Re: [GRASS-user] NDVI analyses with Landsat 8

On Wednesday 12 of June 2013 02:05:48 Huub Munstege wrote:

Hi there all! Greetings from a happy Grass-root GISser on an Archlinux
box…

Hi!

Is there already a way to analyze the new Landsat 8 images (they are already
available!)?

In grass64 you would want to use i.landsat.toar first (with one of the
correction methods) which supports the new Landsat_8 OLI/TIRS sensor(s).

We are in particularly interested in the Normalized
Standardized Vegetation Index (NDVI).

Having (top-of-canopy) reflectances is, if I am not wrong, a must in order
to get the desired ration that reflects the status of the vegetation. You are,
I guess, probably interested in the “i.vi” module (with “viname=ndvi”).

We noticed a shift in wavelength
range that is best depicted here:
http://www.gisagmaps.com/landsat-8-about-refinements/

Interesting, they “narrowed” the band. They probably know what they are doing.
New stuff to explore :smiley:

Doing the well-known raster map calculation with the bands 5 (NIR) and 4 (R)
of L-8 doesn’t give a satisfying result. Very bluntly we subtracted 0.12
from the ‘ndvi’ raster derived from L-8.

Why subtract? Precisely, if I may ask, did you “threshold” it? Cut-off
anything below?

Without doubr there must be more
elegant and correct solutions. Any ideas are very welcome

Would you mind to elaborate a bit more?

Thanks, Nikos

4

Huub Munstege wrote:

Hello Nikos,

Hello Huub!

thx for the swift reply. I'll check out the options you mentioned but at
first sight I noticed that 'i.vi' tool is not available for me ( I'am on
6.4.3-rc, from the AUR package in Archlinux). I.vi is not a command in the
'standard' Grass trunk. Neither is it available as an add-on.

Right! Apologies from my side. I have a setup giving access to all grass
modules outside of a grass session [see link to GRASS-Wiki below] which, if
not properly handled, as described in the wiki ("strip paths"), it even allows
access to all grass70 modules from inside a grass64 session.

Some more explanation as you asked: previous Landsat 7 images we analyzed
with a simple raster calculation with the following formula:
float(Band-4 - Band-3) / (Band-4 + Band-3).

I guess using grass64 means sticking to the formula above for NDVI.

But as you already noticed the bands and ranges have changed under Landsat
8. Various combinations of bands (4,5 and 8) give a result that is at best
not so clear cut as the analysis done with L-7.

I didn't test any L8 so far, though I have downloaded some scenes. Strange,
in a way -- I would rather think that their improvements would improve S-N-R
and, thus, derive "better" vegetation indices... :-?

The "bluntly subtraction with 0.12" is based on on a quick scan of the
obtained result. Areas without vegetation in our project area should give
values in the range between -0.05 - 0.

as per L7 data I guess...

The result of the combination of the 5 and 4 bands in the above formula
gives values that are aprroximately 0.12 higher.

I see.

Therefore the blunt and in-elegant subtraction which is definitely wrong.
But it gives us for the time being a better deistinction between cultivated
irrigation schemes and their surroundings. Maybe I am simply too impatient
and too eager to work with the new fresh data from L-8.

Sorry for asking again: did you correct the data in some way or did you
simply feed the formula with DNs?

Undoubtedly soon, Grass and the other software will incorporate specific
modules that will be tailored to the L-8 sensors.
Cheers and keep up the good work,

Thank you for all of the details regarding your work.

Nikos
---

<http://grasswiki.osgeo.org/wiki/Working_with_GRASS_without_starting_it_explicitly#Bash_examples_.28GNU.2FLinux.29&gt;

5

Dear all,

in the following link I’ve uploaded an image that gives a bit more context information:

http://www.flickr.com/photos/hmunstege/9036837366/

The problem boils down to the different wavelength range of the the NIR band of Landsat 8. It’s much narrower and applying the ndvi tools from Landsat 7 doesn’t give a clear image (see the image).
I am not a remote sensing expert, but we use frequently the ndvi index to monitor the exploitation of rice schemes along the Niger river in Mali. We would love to incorporate the Landsat 8 images in our analyses (without the ‘nodata scratches’ of Landsat 7!)

Thanks in advance for your attention.

Huub Munstege
BPE 2836
Bamako, Rep. du Mali
Tel: +223 20226397
Port: +223 78370695

De : Nikos Alexandris nik@nikosalexandris.net
À : Huub Munstege hmunstege@yahoo.com
Cc : “grass-user@lists.osgeo.org” grass-user@lists.osgeo.org
Envoyé le : Mercredi 12 juin 2013 21h58
Objet : Re: [GRASS-user] NDVI analyses with Landsat 8

Huub Munstege wrote:

Hello Nikos,

Hello Huub!

thx for the swift reply. I’ll check out the options you mentioned but at
first sight I noticed that ‘i.vi’ tool is not available for me ( I’am on
6.4.3-rc, from the AUR package in Archlinux). I.vi is not a command in the
‘standard’ Grass trunk. Neither is it available as an add-on.

Right! Apologies from my side. I have a setup giving access to all grass
modules outside of a grass session [see link to GRASS-Wiki below] which, if
not properly handled, as described in the wiki (“strip paths”), it even allows
access to all grass70 modules from inside a grass64 session.

Some more explanation as you asked: previous Landsat 7 images we analyzed
with a simple raster calculation with the following formula:
float(Band-4 - Band-3) / (Band-4 + Band-3).

I guess using grass64 means sticking to the formula above for NDVI.

But as you already noticed the bands and ranges have changed under Landsat
8. Various combinations of bands (4,5 and 8) give a result that is at best
not so clear cut as the analysis done with L-7.

I didn’t test any L8 so far, though I have downloaded some scenes. Strange,
in a way – I would rather think that their improvements would improve S-N-R
and, thus, derive “better” vegetation indices… :-?

The “bluntly subtraction with 0.12” is based on on a quick scan of the
obtained result. Areas without vegetation in our project area should give
values in the range between -0.05 - 0.

as per L7 data I guess…

The result of the combination of the 5 and 4 bands in the above formula
gives values that are aprroximately 0.12 higher.

I see.

Therefore the blunt and in-elegant subtraction which is definitely wrong.
But it gives us for the time being a better deistinction between cultivated
irrigation schemes and their surroundings. Maybe I am simply too impatient
and too eager to work with the new fresh data from L-8.

Sorry for asking again: did you correct the data in some way or did you
simply feed the formula with DNs?

Undoubtedly soon, Grass and the other software will incorporate specific
modules that will be tailored to the L-8 sensors.
Cheers and keep up the good work,

Thank you for all of the details regarding your work.

Nikos

<http://grasswiki.osgeo.org/wiki/Working_with_GRASS_without_starting_it_explicitly#Bash_examples_.28GNU.2FLinux.29>

6

Huub Munstege wrote:

[...]

But as you already noticed the bands and ranges have changed under Landsat
8. Various combinations of bands (4,5 and 8) give a result that is at best
not so clear cut as the analysis done with L-7. The "bluntly subtraction
with 0.12" is based on on a quick scan of the obtained result. Areas
without vegetation in our project area should give values in the range
between -0.05 - 0. The result of the combination of the 5 and 4 bands in
the above formula gives values that are aprroximately 0.12 higher.
Therefore the blunt and in-elegant subtraction which is definitely wrong.
But it gives us for the time being a better deistinction between cultivated
irrigation schemes and their surroundings. Maybe I am simply too impatient
and too eager to work with the new fresh data from L-8. Undoubtedly soon,
Grass and the other software will incorporate specific modules that will be
tailored to the L-8 sensors.

Dear Huub,

did you eventually progress regarding the VI from L8 images? Any findings you
could share?

Thanks, Nikos