Statistical correlation analysis could clarify the relation between the two raster images. Good luck.

Dear Annelies,

First I would do some basic zonal statistics of each wetland categories (min, max, median, range, standard deviation, boxplot) and compare the categories with Tukey's HSD. Then building a simple classification model, e.g. Random Forest (called 'Random Trees' in ArcGIS), can answer your question related to predictability. Note however that one predictor variable is very-very few... some of the classification algorithms need a second dummy predictor (i.e. a constant, homogeneous raster layer) to work. And it is only a technical workaround, their real aim is working with several predictors.

I recommend using R for this task.

HTH,

Ákos

I would give a first (Prewitt filter) or second derivative (Laplacian filter) of the tpi a try as i would assume that slope gradients are correlated with the class changes. Then, i would compute a windowed correlation between all class (class borders) and slope gradients. After i would try to show, where it fits well and not. If you have any questions, just contact me.

Cheers, Chris

Christian Rogaß, many thanks for your answer! The TPI map is a result of some operations based on the digital elevation model (DEM). First, a low pass filter was applied on the DEM. Then the TPI was calculated on the resulting DEM (raw) and then the TPI was normalised. Lastly, a reclassification was applied to obtain values between 0 and 100.

So given the fact that we don't work with a 'raw' TPI, I assume that we cannot calculate the first or second derivative from this TPI (and use the slope gradient to compare the map with the wetland map). Actually, we want to validate our TPI map by comparing it with this wetland map. We want to figure out whether our method for TPI is a good metric/is a good indicator to map wetlands.

Dear Ákos Bede-Fazekas , thanks for answering my question! Unfortunately I cannot use Tukey's HSD as the data are not normally distributed...

If other members also have other opinions or solutions, I would really appreciate it!

Annelies Broeckx I understand. This is really an interesting task, but I wonder if those cut-outs really match. The TPI seems to be rotated. Can you provide georeferenced images such as geotiffs? I'd like to try some ideas.

Cheers, Chris

Christian Rogaß , excuse me, my fault. These images are not from the same area. The images in this attachment are. Do you think there is a way to compare the two maps?

Yes, i think so. Just provide both as geotiff (georeference must be included). The TPI should be gray-scaled.

Cheers, Chris

Christian Rogaß Here, in attachment. I hope I did it the right way. Thank you very much in advance!

Annelies Broeckx Thanks. I have downloaded the files and played a bit around with this. If you would assume, that any kind of gradients indicate different regions, then the following may work. I used mainly gradient operators & minkowski metric and and compared everything. Enclosed you will find some colored results which show potential overlaps as color code (whitish). I got 17% overlay for the gradients and if I dilate them 50% at max to respect influences of misclassification and misregistration. I hope it helps. If you are interested in this very basic approach, just write me an email.