[GRASS5] i.tm.dehaze problem

[Markus Neteler]

On Tue, Aug 13, 2002 at 01:12:53AM +0100, Gualter Barbas Baptista wrote:
> Markus Neteler wrote:
> |On Mon, Aug 12, 2002 at 11:03:06PM +0100, Gualter Barbas Baptista wrote:
> |[...]
> |>|>>GRASS:~ > i.tm.dehaze inverno.tasscap.4 b1 b2 b3 b4 b5 b7
> |>|>>r.stats:  100%
> |>|>>1 variables
> |>|>>0 observations
> |>|>>ERROR: SVDCMP: You must augment A with extra zero rows
> |
> |>|With my TM data it runs well.
> |>But only with a smaller region or with the whole image?
> |
> |It should be sufficient to zoom a, say 500x500 region and run it
> |again (since GRASS raster is region sensitive).
> 
> It works well with a smaller region. But now I have the problem of not
> "dehazing" the whole scene. Is there any simple way of dehazing the
> whole scene or will I have to do something like cut, dehaze and put
> together again?

The problem of locally dehazing and then merging the results is that the
linear regression will slightly differ. That means that the areas
do not match 100%, some will be better, some less good dehazed.
The only solution is to dehaze all-in-one.

> I will probably not need to dehaze the whole scene for the work I'm
> doing now, but for the future this might prove necessary.
>
> Is this a bug of r.linear.regression? Would this bug fix be very
> complicated?

It is not a bug, but a limitation. r.linear.regression reads from
ASCII file and then does the regression. An idea may be to replace
this regression tool with a better software allowing for large
data sets. The script i.tm.dehaze itself is very simple.

Maybe some other developer knows about a (linear) regression tool
which accepts large data sets and operates in the style of
r.linear.regression?

Markus