[GRASS-dev] Unexpected EVI range from "i.vi"

[Nikos Alexandris]

MM:
> >> >> I think I figured it out:  The EVI formula in i.vi is for MODIS.

NA:
> >> > That's precise, EVI is MODIS specific.  We should clearly describe this
> >> > in the manual (I will try to alter the respective text).

MM:
> >> From the literature, I gor the impression that EVI can be calculated
> >> from other sensors as well, as long as you get the coefficients right.

NA:
> > Yes, but this is not an easy job, is it?  This is (also) why, I think,
> > they (MODIS science team) developed the EVI2, which is cross-sensor.

MM:
> I am pretty sure that the EVI2 formula in i.vi is not cross-sensor,
> but also tailored to unscaled MODIS input bands: 
> EVI2 = G * ( nir - red)  / (nir + C1 * red + L)
 
> again, L needs to be adjusted to the actual input scale.

> AFAIU, both EVI and EVI2 can be applied to different sensors, given
> that the required bands are available and that the coefficients are
> adjusted if need be.

Right.  Yet, i.vi defines EVI and EVI2 (the latter suggested by me) with the 
MODIS-specific coefficients.  We do need to explain this in the manuals.

> >> >> The generic formula is
> >> >> G * ( nir - red)  / (nir + C1 * red - C2 * blue + L)
> >> >> where G is a gain factor, C1, C2 are coefficients to correct for
> >> >> aerosol influences in the red band using the blue band and L is the
> >> >> canopy background adjustment that addresses non-linear, differential
> >> >> NIR and red radiant transfer through a canopy.

I had a look back in papers I have read in the past. There is one by Hui qing 
Liu and Huete, A. [1], I only came to "discover" now.  According to it, the 
origins of EVI date back to 1995!

> >> Assuming that the input to i.vi should be properly preprocessed bands
> >> with a theoretically maximum range of [0, 1], you could set L to
> >> 0.0001 and would get reasonable EVI values, sensor-independent.

> > This reminds, if I am not wrong (didn't check) the scale factor for MOD09
> > surface reflectance products.  Makes sense.

> I suggested L = 0.0001 exactly because this is the MODIS scale factor.

> BTW, the satellite data you mentioned are ETM, not MODIS, thus
> applying the EVI formula developed for MODIS to ETM bands is a bit
> adventurous. In any case, EVI was developed for tropical rain forests
> because NDVI can saturate there. The Landsat scene you mentioned has
> only ocean and desert, no forest of any kind. NDVI should be just fine
> in this area.

True. Was just testing... not real work (yet).  However, from my last work 
using Landsat TM over Greece (fuel type mapping, where vegetation density is 
important), the experience (based on visual interpretation, no real 
assessment) was that using EVI2 helped more than the NDVI in visually 
discriminating forest vegetation types (of different degrees of density).  
Which, in turn, suggested (to me :-p) that using EVI2 could improve further 
the performance of i.cluster and subsequent classification attempts for 
example.  I never got to test/evaluate the idea though.

> I would suggest to test the EVI(2) formulas in i.vi with a MODIS NDVI
> product which also includes the required input bands. All bands in the
> MODIS NDVI product would need to be scaled according to the
> documentation prior to feeding them to i.vi, or r.mapcalc with
> adjusted formulas.

ToDo.

Just adding stuff from the literature:  EVI2 is supported to be a formula that 
can be used with other sensors as well. Taken from "Development of a two-band 
enhanced vegetation index without a blue band" (by Zhangyan Jianga, Alfredo R. 
Huetea, Kamel Didana, Tomoaki Miurab [2]):

    "EVI2 can be used for sensors without a blue band, such as the Advanced 
Very High Resolution Radiometer (AVHRR), and may reveal different vegetation 
dynamics in comparison with the current AVHRR NDVI dataset."


There is also a study "Using lidar and effective LAI data to evaluate IKONOS 
and Landsat7 ETM+ vegetation cover estimates in a ponderosa pine forest" [3] 
that derives EVI from IKONOS imagery by using the parameters

"suggested by Huete et al. (1997) of L=1, C1=6, C2=7.5, and G=2.5"

?!  -- didn't read carefully the whole of it though.

Nikos
---

[1] "A feedback based modification of the NDVI to minimize canopy background 
and atmospheric noise": <http://dx.doi.org/10.1109/36.377946>.

[2] <http://dx.doi.org/10.1016/j.rse.2008.06.006>

[3] <http://dx.doi.org/10.1016/j.rse.2003.11.003>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 230 bytes
Desc: This is a digitally signed message part.
URL: <http://lists.osgeo.org/pipermail/grass-dev/attachments/20130702/d31b1bb2/attachment.pgp>