Discussion: Q for Remote Sensing Nerds

I know there are a lot us on here and having been away from the industry for a while need some input given progression over the past decade.

Consider time and money as no concern. Given the advances in remotely sensed data since "my day"... what would you do to best optically determine presence/absence of exposed bedrock in a boreal forest environment (northern ON/MB/SK) where exposures are patchy at best? There'd be no need for rock or mineral specificity, simply presence/absence of bedrock outcrop to guide future in situ geologic study.

Would you go consider airborne over satellite?

Would you go hyperspectral or multispectral?

The spatial target would be identifying outcrops as small as 2x2m

I have my thoughts, which would tend to lean towards airborn multispectral collection with focus on the contrast of NIR vs SWIR (reflection of vegetation vs absorption by exposed rock).

By 2015, this must be a common practice by geologist and those in the forestry industry. However, most of the literature I'm turning up is specific to rock or mineral identification mapping, which is overkill.

30sec back-of-the-napkin input would greatly appreciated.

If you collect thermal infrared an hour after sunset, wouldn't the rocks glow warmer? Sorry I don't know what state the detectors for that spectrum are in and whether they can resolve 2 m from ~3 km, not really my domain.

Francis Falardeau (contact at the link) has done the hot rocks trick. Can't remember if it was aerial or SPOT data. He's the guy who puts on all the Laurentian Rogaines in Quebec. Drop him a line, he'll be happy to share.

If I had to guess at the best wavelength for this I'd use Wein's Law. E.g. for peak emission at room temp (293K), wavelength = 2900/293 = ~10 microns.

I would also try looking at intensity of the ground returns from lidar. You get good contrast between grass and pavement, so if you could get enough ground returns from the dataset you may be able to see good contrast between the bare rock and dirt/plant forest floor.

Ah AP never fails. Thanks all so far. Hadn't thought of the thermal trick. I'll email Francois ...incidentally he and I raced the 2008(?) NARC together and this topic never came up ;)

Trying to stay optical. Lidar data is sadly not as continuous or freely available north of the border.

I would also add that if you have Lidar data you have the additional benefit of being able to see gaps in the tree cover by taking the difference between the first returns and the ground returns. Larger patches of rock will show up because trees don't generally grow on bare rock. Not the primary method you'd want to use but it could be useful as an extra check.

NDVI alone would probably do it. It requires visible and Near IR images, and is typically post-processed.

http://en.wikipedia.org/wiki/Normalized_Difference...

It is an index regarding the chlorophyll activity, and obviously (unless they're vegetation covered), rocks don't have that.

Near IR can be done with CCD-based point-and-shoot cameras with a special filter installed, so no expensive far-IR camera is required:

http://diydrones.com/profiles/blogs/modifying-a-po...

I found some quotes with google that implied roads are strongly contrasted in NDVI images.

For example, check out these:
http://diydrones.com/profiles/blogs/the-near-futur...

If you had strong canopy, you might need to pick a time when the leaves were off, but the forest floor had green (active chlorophyll) rather than pure winter dormancy.

Well, there could well be a 15 m × 15 m evergreen chlorophyllated canopy over your 2 m × 2 m bare rock outcrop...

First of all, I don't consider myself a remote-sensing nerd.

I didn't realize boreal means mostly coniferous trees.

Ed's idea of lidar intensity makes a lot of sense to me, just based on ground- or all-last-return- intensity images I've seen. Depending on the thickness of the canopy or how close together the trees are, you might have to throw away points away from nadir. (I'm assuming nadir is the best place for the laser to penetrate to the ground.)

The intensity image products I see tend to be grayscale images. One could probably determine the intensity returns of interest and apply filtering or false-colors to highlight them.

I see some search hits for things like "Multi-wavelength canopy LiDAR."
http://www.sciencedirect.com/science/article/pii/S...

A multispectral lidar that could do something like NVDI and let you only process points near the ground might work.

Saw this today. Thought it might be of interest:

Using UAVs and NIR photography in conservation land cover classification