Agreed. You have to make sure you have valid training areas are for each image you want to classify. I've also found that dealing with clouds and their shadows is one of the biggest issues to consider when doing classifications.<div>
<br></div><div>- Nick J<br><br><div class="gmail_quote">On Sun, Dec 5, 2010 at 4:12 AM, Nikos Alexandris <span dir="ltr"><<a href="mailto:nikos.alexandris@uranus.uni-freiburg.de">nikos.alexandris@uranus.uni-freiburg.de</a>></span> wrote:<br>
<blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex;">Nick Jachowski:<br>
<div class="im"><br>
> I have been working a lot with SLC-Off imagery lately. Some people in my<br>
> department have used the gap filling programs floating around the net, but<br>
> I'm not familiar with them personally. I've settled on using r.patch as<br>
> well, although you have to be careful how you apply it. I found that even<br>
> if I used radiometrically corrected landsat images (using i.landsat.toar)<br>
> from the same season often the patched parts of the image did not fit<br>
> smoothly with the rest of the image (i.e. you could see striations where<br>
> the gaps had been).<br>
<br>
</div>Right. The same here. But I used the composites only for visual interpretation<br>
which was ok. It's always interesting how different tasks pose different<br>
challenges.<br>
<div class="im"><br>
> I'm using the imagery for land classification, so<br>
> I've found it works better if I do the classification on each landsat<br>
> image separately and then patch them. Using this method you can't tell<br>
> where the former gaps are, at least in my experience working with imagery<br>
> from the dry season in southeast asia.<br>
<br>
</div>Interesting. Yet, I guess, you had to use independent training areas (in case<br>
you performed supervised classifications), right?<br>
<br>
[...]<br>
<br>
Nikos A<br>
</blockquote></div><br></div>