I need to perform change detection between the years 2005, 2010 and 2015. I have tried classifying the 2015 Landsat 8 Imagery using SCP plug-in of QGIS. Some of the issues cropping up are dry riverbeds are getting classified either as bare soil or as built-up area. Therefore, I was thinking maybe I can mask the water features and then do the supervised classification.
If I need to do edge enhancement, then how do I extract the river features after that?
Thank you, Julian, for your input. I'll follow your suggestions and see if they help solve the issue. I do have waterway shapefile acquired from OSM but there are instances where the polyline does not fall in line with the river features in the Landsat image when overlayed. Hence I am reluctant to use this as a mask.
You can map first all water bodies using indexes (Hassani et al., 2015), then you can use the result as mask.
Data Hassani et al, 2015
You can apply normalized differential water index(NDWI) for year wise LANDSAT data. After deriving NDWI, use reclassify to clasiify the NDWI image in two parts such as Water bodies and land surface body in 1 and 0 respectively. Finally convert the reclassified image into vector format. By this you can delineate water bodies and can find area of individual water body. And also you can do change detection analysis by this method.
I would follow Julian's advice: try to find an image from a high streamflow period, then use the NDWI to identify water, then set those cells as Null before doing other classifications. If the high stream flow still doesn't fill the channels, perhaps you could do some sort of buffer around the water cells and then mask all of those cells out of hte classification. But, classifying bare river bed as 'bare soil' and built up seems to just be one of the challenges....that cause others to reject pixel-based classification in favor of other approaches.......
Thanks, Mohand. I've tried using the IWS to extract the water surfaces and the result was more or less similar to MNDWI. Oddly, I got an output similar to the images given in the paper only when applying the first half of the equation [(4*b5-b1)*2/b5].
Thank you for your suggestion, Biswajit. I've used MNDWI, however.
Dear Uday Pimple , the mask water dont'is problem , istintivly water in the word , but word of uman gender in a good voyage is very umanly !!
benedetta brasca
- Badges
- Science topic
- Similar topics
- Geoscience
- Remote Sensing