Detecting Willow - Need help combining elevation and spectral reflectance raster data

[StevenSullivan]

Hello! I'm trying to detect and quantify willow on a stream in NE Oregon. I have elevation data that gives me the height of plants. I have change in ndvi data that indicates plants with most photosynthetic activity. Essentially, grass (green up) and willow (leaf out) have the highest photosynthesis levels. I can separate willow from conifers based on photosynthesis. And with elevation, I can distinguish willow from grass. Yet, I'm looking for ways to combine these two data sets, so I'll be left with the high photosynthesis regions that are taller than a threshold for grass (say 1 meter) . Of course, I'd also like to be able to do some classification error clean up. I feel certain others have been in this situation before, but I'm not finding anything on point. Would any of you have a suggestion on what to try, and on what not to try, so I reduce spinning my wheels? Thank you!

[GISGeekpro · Accepted solution]

Hi Steven, what I would is use the raster calculator to extract a new layer according to a certain threshhold for the elevation, so you can end-up with a new layer with the areas with a certain elevation and above (or below if you like). Then , you can use this layer to mask your NDVI layer, where then you can use again the raster calculator to extract NDVI values - above or below a certain threshhold. As a result, you will end up with a layer which it will be the areas above or below a certain elevation, which will also have NDVI values above or below a certain value. You can also try to clip any of these layers within these masks, so you can actually keep your actual values within the areas of interest. 

Hope this helps! 

[BrianShepard · Accepted solution]

Tagging on to @GISGeekpro's reply - where you have a specific threshold, creating binary rasters using raster calculator can be helpful. We had a model that required us to isolate vegetation at least 20 ft. in height. We created a raster using raster calculator with cell values of 1 where height was >=20 ft. and 0 where it was < 20 ft. Multiply that raster by your NDVI and the cells with grass (lower than your height threshold) will be 0. If you're using a range of 1 to -1 for the NDVI values, 0 should be lower than the values for the willow. If you can identify a range of NDVI values that represent the willow (and grass), use raster calculator to set that range as 1 and anything out of range as 0. Multiplying that by the binary height raster will result in values of 1 where your NDVI is in the right range and height is above grass, and 0 everywhere else.

You could also look into doing a supervised classification using both datasets. If you're able to isolate the willow fairly well using NDVI and height, you may not gain much over using raster calculator. 

[StevenSullivan]

---

@BrianShepard wrote:

Tagging on to @GISGeekpro's reply - where you have a specific threshold, creating binary rasters using raster calculator can be helpful. We had a model that required us to isolate vegetation at least 20 ft. in height. We created a raster using raster calculator with cell values of 1 where height was >=20 ft. and 0 where it was < 20 ft. Multiply that raster by your NDVI and the cells with grass (lower than your height threshold) will be 0. If you're using a range of 1 to -1 for the NDVI values, 0 should be lower than the values for the willow. If you can identify a range of NDVI values that represent the willow (and grass), use raster calculator to set that range as 1 and anything out of range as 0. Multiplying that by the binary height raster will result in values of 1 where your NDVI is in the right range and height is above grass, and 0 everywhere else.

You could also look into doing a supervised classification using both datasets. If you're able to isolate the willow fairly well using NDVI and height, you may not gain much over using raster calculator. 

---

@GISGeekpro and @BrianShepard Thank you for your comments. Very helpful. I did use reclassification to get the desired NDVI and elevation layers. Yet specifically, I didn't know a best/good way to combine them and, generally, I didn't know if there might be a better way to go at it. I appreciate the comments about raster calculator and the like. And the comment about classification. I have found at times that sitting up schema and getting classification to work consistently can be problematic, though I do like some of the filter settings. Again, your responses are very much appreciated.