Dear Vikram,
It is not good to calculate NDVI from DN. The purpose of DN (digital
number) is only for packing as much information as possible into a given
?BIT-size? (in Landsat 8-bit/pixel/band).
DN is practically the radiance (I think TOA?) modified with an offset
and a gain in order to press or expand the range between 0 and 255.
This means that the relation between bands is changed, and this is a
crucial need for NDVI (and generally for indices).
Probably the minimum pre-processing you absolutely need to do is the
conversion from DN to RADIANCE (TOA) using the gain and bias information
in the MTL file (metadata). Better and more precise is to calculate the
ground reflectance. I did this with the RemoteSensing package on
R-Forge. But the current version doesn't work because of incomplete
modifications in the package code.
Non-vegetated areas can have positive values. I think the soil has
often a NDVI of 0.1. Clouds sometimes have even 0.2 as...anyway you
can't trust a NDVI from DN and you can't sharply differentiate
veg/nonVeg by positive or negative ndvi values. Water normally has a
clear negative NDVI (~-0.2).
Also implemented in the RemoteSensing package are functions to compute
other vegetation indices, I did not test them but maybe you can find
something useful there (type ??ndvi? after installing and loading the
package).
Cheers Matteo
Vikram Ranga <babuawara at gmail.com> 1/18/2012 10:49 am >>>
Hello list,
I am working with landsat images and would like to create a
phenological
cycle with NDVI.
I am using R for that but NDVI values changes if i use on sensor
reflectance instead of DN values i.e. giving different result in every
case.
I tried to compare histograms seems to have changed and I am not
getting
negative values where there is no vegetation but if i use DN values it
gives negative values to such locations. Do anybody have any idea what
went wrong??
Thanks in advance and would be much appreciated.
Vikram