If you have ground data for a few stands and the cedar stands look different than the pine stands in your input image when you visualize it in Erdas, then it should be possible to create a process that separates both types. If you cannot see a clear difference in pattern, texture, hue or tone, then it is unlikely you will be successful. What type of image will you be using? Are the stands pure (monospecific), are they even-aged, and are stands from different species away from each other or can they be adjacent?
It should be better if you have some sample plots in two different coniferous species. Using the supervised classification (Spectral signature and feature space) in ERDAS Imagine you can distinguish both easily. If you don't have direct field measurements you can take help with other ancillary information like geo tags Google photographs of that area, literature review etc.
Agree with Hammad, it would be best to use an ASD Spectrometer to collect ground truth measurements of the tree species. You can then down sample the spectra to the resolution of Landsat 8, perform atmospheric correction and many run your classification routine. If you don't have ground measurements, seek out those who may have collected that data already. This method will give you the best results. There are others, like using in-scene derived spectra of known plots of the different species to perform classification, but there is the risk of it being less accurate.
If you have ground data for a few stands and the cedar stands look different than the pine stands in your input image when you visualize it in Erdas, then it should be possible to create a process that separates both types. If you cannot see a clear difference in pattern, texture, hue or tone, then it is unlikely you will be successful. What type of image will you be using? Are the stands pure (monospecific), are they even-aged, and are stands from different species away from each other or can they be adjacent?
Use End member extraction technique and find many of the pure pixels, disaggregate your desired pure pixels from all and classify imagery using Spectral Angle Mapper and / or Unmixing (Linear, MTMF) techniques. If you have spectroradiometer spectra from ground locations, you can directly use them to classify the imagery.
You can also try object oriented image segmentation and classification.
thanks for your answers. I really appropriate your time and considerations.
with kind regards,
Dear Guillermo Castilla, I'm using LANDSAT image (downloadable), the stand is pure, a different age and are a stand from different species (adjacent: in altitude), in descending, there are cedar, green oak. in the other region atlas cedar then pine.
There are several studies dealing with distinguishing spectrally between conifer and broad leaf trees. one of them is change detection one image in the summer the other in the winter trusting the exfoliation to do the difference. if the broad leaf tree does not loose leaves on winter the color might do the job and there are studies dealing with that.
When looking for separation of two conifers, my suggestion is to look at them and if you see a difference - meaning different colors- hues it might be easier to distinguish spectrally. All of this is in case that each species is in a homogeneous forest, both forests have the same relative coverage of trees etc' . If there are differences between the two forests the classification might be based on that and not on spectral differences of the species (e,g,. one forest is on a slope facing north the other forest is on a slope facing south).