(transfered from https://stackoverflow.com/questions/39348977/maximum-likelihood-classification-landsat-why-to-exclude-thermal-band)

I search for an argument (which I could cite, ideally) to support my decision to exclude Thermal band 6 from Maximum likelihood classification (MLC) of Landsat (5-7) imagery. To my knowledge, the thermal band 6 is suggested to exclude from MLC because of its coarser spatial resolution (~ 120 m), comparing to another bands (30 m).

Can you suggest me some papers resulting this?

The only one I've found is: http://www.isprs.org/proceedings/XXXIV/part1/paper/00077.pdf ... a classification utilizing all of the TM bands (excluding the thermal band due to its coarse spatial resolution) ...

Do you have suggestions for another paper/handbook?

  • 1
    What are you attempting to classify? General land cover? Forest? The TIR band is important for some classifications and less useful for others, but the usefulness is application specific. – Radar Sep 6 '16 at 16:24
  • I was trying to classify forest and disturbances as fire, clear-cuts, tree damage by bark beetles... would it be useful for that? – maycca Sep 12 '16 at 8:47

In my opinion, the coarser resolution of the thermal band is not necessarily the reason why it gets excluded in many applications, after all, you can always resample the thermal band to unify the resolutions. As @Radar has suggested in the comment, whether the thermal band needs to be included or not needs to be determined based on the specific relationship between the classes of interest and their thermal properties. If the thermal information is irrelevant, there is no benefit in including them.

  • I was trying to classify forest and disturbances as fire, clear-cuts, tree damage by bark beetles... thus I thing the termal properties may enhance the characteristics. However, the forest damage causes by beetles is so small, that the resolution of 60 ~ 120 m of thermal band could be crucial to mask the differencies in reflectance... What do you think? – maycca Sep 12 '16 at 14:21
  • @maycca Understanding forest disturbances (mostly fires and logging) using remote sensing happens to be my major focus too. There has been many applications using Landsat images to classify fires, clearcuts/logging and insect outbreaks. In my opinion, the thermal band is not very helpful here. The spectral characters of the post-fire and post-logging forest stands are significantly different, at 30-m resolution you can use the optical reflectance bands of the Landsat images to differentiate them quite well (as long as you have good before- and after- image pairs). – TonyC Sep 12 '16 at 18:17
  • @maycca The insect outbreaks are tricky. Like you said they are sometimes quite small. In addition, they don’t exhibit significant spectral differences as the other disturbances do. That is the same with the thermal band, especially considering the much coarser resolution of the thermal band. Having said that, I’m not trying to convince you not to use the thermal band in your classification. It’s always good to give it a test before dumping it completely. – TonyC Sep 12 '16 at 18:17
  • @maycca In addition to that, I’d also suggest you experiment with some spectral indices to see how they work. I haven’t dealt with insect outbreak classification myself before but I’ve seen some people who got some really good results using them. – TonyC Sep 12 '16 at 18:17
  • @maycca Another piece of suggestion, if your classification result is not to your satisfaction, is to consider using certain machine learning classification algorithms such as decision tree and random forest instead of the maximum likelihood classification method. In my personal experience, these methods may give you better results, especially when the spectral properties of the forest disturbances are complex, which is quite often the case. – TonyC Sep 12 '16 at 18:18

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.