Hello all. my question is about difference between land use and land cover when we are trying to classify satellite image (Landsat in my work). when I done perprocessing, select training points and giver them to software (ENVI in my work) and run some algorithms (SVM, Maximum likelihood), the software give me a classified image. so my question is, in my work, for instance I want to measure city area so after classification I merge green areas (like parks, trees and so on) located in the city to city land use. or another instance is about fishing pool (aquaculture) there are many fishing pool which separated with tree or grass line edge (they are in one area side by side and only tree or grass line or road separate them). software separate reservoir and tree or grass edge but actually they are in one land use(fishing pool) so I need to merge them. now if I do accuracy assessment by validation points my classes accuracy come low, right?
should I done accuracy assessment in first classification image which software give me? and after accuracy assessment merge this classes?
forgive me for my terrible mistake in English writing, I am not good in English. and I wish I could convey my mean transparently.
thankfully.
Dear Ali,
You could calculate an accuracy (and Kappa Coefficient) before class merging but this is not the validation accuracy. What you can do is to define another set of samples (training areas, namely ROIs in ENVI, that are different from those you used for classification) based on your field investigation. You can use this set of ROIs to assess the accuracy of your land use and land cover map. To be useful, the overall accuracy (OA) of your map should not be below 85%.
By the way, I would not suggest you to merge the urban green area into city or city land use (green area is a specific urban land use) if it is clearly separated from other vegetated lands such as croplands, forests, etc. Otherwise, other vegetation cover will be also considered as urban or city land use, which is not true.
I have a paper that may be of interest to you. Please see the attached file.
Good try and all the best,
Weicheng
Hello!
Sugestion: you can classify each class separately, creating masks between each classification process (to avoid reclassifications of pixels). Then, you can do a bandmath to unify your set of images, apply the majority filter (to do some "cleaning") and do the final adjustments in GIS (vector format).
Best regards...
Firstly, the LAND COVER should be classified and evaluate this classification using statistic measures according to the general work flow of supervised and unsupervised classifications in remote sensing. Then, you might merge the land cover classes as land uses according to the established criteria (ad hoc). At this level, you do not evaluate again the accuracy of new merged classes (LAND USE). There are many examples where the same land covers are classified as different land uses. For example, a ind of tree cover could be natural forest, urban park, forest crops , etc. according to the scale, the context, and the criteria.
Good luck!
I am so appreciative for your consideration. your suggestions illustrated my workflow.
I wish bests for you in life. god bless you all.
Hope you expect this
do classification using ENVI or other software
do raster to vector conversion using ArcGIS
merge vector polygons wherever you want
calculate area by statistics tool in ArcGIS
- Badges
- Science topic
- Similar topics
- Geoscience
- Geoinformatics
- Geographic Information System