Supervised Classification Problems on Landsat 8 scene

[PedroPinto]

I’m really having a problem in the supervised classification of one Landsat 8 - OLI scene were I can’t get any results. I’m using for starters de Maximum Likelihood and I’m getting this kind of results:

I did obtained results with the Unsupervised Classification (K-Means):

Even with minimum distance classifier (although very poor results).

Can anyone please, help me solve this problem?

Thank you so much

[ABraun]

Did you use pins as training vectors?

[PedroPinto]

Hi Abraun,
No. Only vectors.
Here’s an example:

[ABraun]

sorry, I just wanted to go sure because there are pins in the screenshot.

The overlay of vectors and rasters is a bit faulty for some UTM zones. Have you tried projecting the data in WGS84 (import the vectors again afterwards) as described here?

Rndom forest classification steps

I reproject the data into Geographic Lat/Lon. After that, the values do not change. I try some times with some different data and I find out that in the classification step, it works with the Lat/Lon project.

Sentinel-2 Classification Problem

Dear Braun When I changed sentinel-2 coordinate system to wgs84, classification problem has solved completely. Thanks for your excellent support. Sincerely,

[PedroPinto]

I will recap my steps:
I imported a LS product: LC082040312017110601T1-SC20200511135125.tar.gz (using optical sensors, Landsat, 30m resolution function). Then a subset for the region of interest. After a re-projection to WGS:

Then I did a vector import of ESRI shapefiles (created in the original subset image without reprojection):

Then ML classifier and no results again

[PedroPinto]

It’s very strange, because to the previous pre-processing actions, the RF Classifier did work (partially I have to say, because is not classifying the water pixel reflectances, at least most of it, only the water breaking zones), take a look:

For the same input the only thing I get is a blank result, in the ML case:

Could this be related to low spectral values of L8 images for the water ? I really don’t understand, because Unsupervised Class did it’s job for the same values(!?). It’s confusing.

[ABraun]

strange indeed, but I was able to repdoduce this error.

@marpet Do you have an idea what is wrong?

[PedroPinto]

ABraun,
I edit my previous reply with some additional comment about RF class, that maybe has also something to do with what could be wrong. Please take a look.

[ABraun]

I have seen it but do not have an explanation why it worked for RF but not ML.

[PedroPinto]

Hi guys!
Have already anyone found a solution for this error?

[YannMa]

Hi @PedroPinto,
i had the same problem with landsat 8 and landsat 7 data. with the RF the classification worked well (also with the Minimum Distance), but with the ML it was the same as your case. i didn´t found a solution yet.