I applied supervised maximum likelihood classification to a yearly stack of Landsat images (resolution 30m). The land cover classes are related to coniferous forest, i.e., forest, clear-cut, fire, bark beetle. I had only few years of available reference aerial photos (resolution 0.5m). Aerial photos do not fully cover the extent of Landsat images (AOI). To assess the accuracy of the classification for single year scene, I extracted a overlapping part between the extent of Landsat and aerial photos:

enter image description here

I calculated the measures of accuracy for the extracted part of fully classified Landsat by set of stratified sampling points per mapped class.

I followed the methods published here:

  • Olofsson, P., Foody, G.M., Herold, M., Stehman, S. V., Woodcock, C.E., Wulder, M.A., 2014. Good practices for estimating area and assessing accuracy of land change. Remote Sens. Environ. 148, 42–57. doi:10.1016/j.rse.2014.02.015 and
  • Olofsson, P., Foody, G.M., Stehman, S. V., Woodcock, C.E., 2013. Making better use of accuracy data in land change studies: Estimating accuracy and area and quantifying uncertainty using stratified estimation. Remote Sens. Environ. 129, 122–131. doi:10.1016/j.rse.2012.10.031

to assess the accuracy of the classification.

Great R code summarizing this can be found here: https://github.com/openforis/accuracy-assessment/blob/master/error_matrix_analysis.R

The error matrix, by this method, is not based on the simple sample counts, but on the estimated proportions of the area. Including the proportion of the mapped classes in accuracy assessment, it is possible to quantify the uncertainty attributable to variability in the sampling. Simply, to adjust the mapped area to map classification error. The error-adjusted area estimates should be accompagnied by confidence interval.

My problem:

I classified 7 Landsat images. I calculated accuracy for extracted parts of them for two years, because of no availability of another reference data sets.

My questions:

  • How to calculate the adjusted mapped area for the whole classified Landsat scene? (i.e. the overlapped area, where I assessed accuracy, was 100 ha. The whole AOI was 500 ha. My adjusted area estimated for the overlapped area is 20 ha (100ha±20 ha)).
  • How to extent the margin of error number for the whole classified area? (500±100 ha?)
  • Just for clarification: You classified Landsat imagery (manually or supervised/unsupervised?) and you want to assess the accuracy of the classification in the Landsat images based on the aerial reference in the overlapping area, correct? May I ask what you classified? I assume that the spatial resolution is much higher in the aerial images. Is that correct?
    – C.Riedel
    Commented Sep 29, 2016 at 9:58
  • Yes, it is correct, I have updated my questions by those details.
    – maycca
    Commented Sep 29, 2016 at 12:22

1 Answer 1


If you want to assess the accuracy of a supervised classification, you compare your classification results to your ground truth. Thus, results and ground truth can only be compared within an area in which both data overlap. In your case, that would be area in which the aerial images intersect the Landsat data. If you want to verify your Landsat classification according to your ground truth in the aerial images, you can only give the accuracy of the classification based on the intersecting subset.

In general, there are two options:

Option one is to perform a supervised classification on all your Landsat images and assess the accuracy of the classification based on your ground truth from the overlapping areas. I don't know about your data and research area but in my opinion, this would be OK if the Landsat and aerial images were taken during the same season and if you took one set of training data from all the Landsat images (dense forest from scene 1-7, forest fire from scene 1-7 (if available), etc.). This means that same sensor & different period is compared to same sensor and different period.

Second option is to perform a supervised classification (as in the first option) but to have your ground truth taken from the Landsat images. This would actually be the better choice because you extracted your ground truth from a dataset with the same spatial resolution.

If you go for option one, be aware that you're expecting a lot from your supervised classification algorithm when you make an accuracy assessment between both image datasets. When you compare your classification results from a 30 m resolution image to your ground truth from a 0.5 m resolution image, you want your algorithm not only to give good results, you want your algorithm to detect more, than the spatial resolution has to offer. Although you may detect more by visual interpretation in the 0.5 m aerial images, transferring the information to a 30 m Landsat image is very vague.

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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