I'm trying to use NDVI data to define forested land in the southeastern United States. The data is made up of 1m pixels. My goal was to classify each pixel as vegetation or not, then use a cut-off size of contiguous vegetation pixels to define a forest. What are the arbitrary cut-off values in NDVI in order to define as vegetation (tree) vs non-vegetation/grass? What are sources to go along with this?
Also: is there a cut-off value to distinguish between deciduous and evergreen forests?
Hi Traylor,
I think you can use trial and error and fix your threshold based on the NDVI value in selected and randomly distributed number of samples in your study area. There is no universal and accepted NDVI cut-off value while dealing with vegetation classes. It depends on the vegetation structure and phenology (high canopy for dense forest, sparse vegetation for shrub maybe) in your study area which is quite different from one study area to another. It would also better to cluster your data into objects and analyze them using object-based methods since you are using high resolution data (1m Pixels)
Good luck
Theo
Actually there is no particular NDVI range for forest. Normally for vegetation it varies between 0.2 to 0.8. So, if the density of vegetation is higher then it goes towards higher range. But there is no exact cutoff value. This vegetation indices are very much site specific as well as vegetation type specific. So it varies according to study area. You check according to your study area and fixed the cutoff range. Even also it depends upon the spatial resolution of the data. For vegetation discrimination it will be better if you take different vegetation indices like SAVI or EVI.
Somnath Paramanik thank you. I used 0.3 initially and it seemed to work to differentiate forest/tree cover (blue) vs the lake (orange) and grass (yellow and light green). I will look into using other indices as well for further discrimination.
Theodomir Mugiraneza thanks for your input. Are there GIS commands to convert this kind of data to objects? I am not familiar with these methods. Thanks
Somnath Paramanik USDA/NAIP/DOQQ from Google earth engine (1m resolution)
Dear Traylor,
If your image is in a pre-processed JPG format I think the object-based analysis would be meaningless. I was thinking you are using multispectral image with different spectral bands.
Dense Vegetation > 0.7
Medium Vegetation 0.5-0.7
Sparse Vegetation 0.3-0.5
No Vegetation 0.1-0.3
Water Bodies < 0.1
- Badges
- Science topic
- Similar topics
- Geoscience
- Remote Sensing