Readme for Tree Training Dataset

Collin Bode, Feb, 2016

The ESRI shapefile, "TreeID_by_LiDARvegheight_poly.shp", is a training dataset for hyperspectral imaging analysis.  Projection is UTM, zone 10, datum NAD83.

It contains vector polygons defining stands of similar vegetation types.  It has two degrees of classification: VegClass and TreeName.  These are found in the fields of the shapefile.  

Vegetation Classification is the most complete of the two characterizations.  It defines areas into the following broad categories:

VegClassID	Vegetation Classification	Explaination
1		Meadow		Grass fields with some leafy plants under 1 m.
2		Chaparral	Scrub (1-4 m height). Manzanita is dominant, but large mix of species.
3		Broadleaf	Drought tolerant Californian hardwoods. Oak and Madrone co-dominant, with Bay, Alder, and several other species intermixed. Heights vary greatly from 2 - 40 m.
4		Conifer		Douglas fir dominant, with Redwoods found near water and sheltered areas. Heights 20 - 80 m.
5		Riparian	River and stream channel. Tend not to be drought stressed. High variety of species within this zone. Probably not worth trying to do spp identification here.  
9		Building	Isolated buildings within study area.

Within these general classifications there are some individual species that have been identified.  

TreeID, Common Name, General Classification
1. Grasses	Meadow
2. Manzanita	Chaparral
3. Madrone	Broadleaf
4. Bay		Broadleaf
5. Oak		Broadleaf	(live oak, black oak, tan oak, etc)
6. Douglas Fir	Conifer
7. Redwood	Conifer


Identification Methods

This training set was created by hand by Collin Bode.  I am not a biologist. My identification is by a combination of several factors.  I have worked in the Angelo Reserve for over 10 years.  The polygons are only in areas I have personally hiked, so I know the area on the ground.  

Polygon creation was done in ArcMap 10.2, using the following:
Vegetation canopy map: derived from the 2014 Gemini LiDAR dataset.  Each grid cell was assigned the highest elevation found within the point cloud.  Hillshade layer was used.
Vegetation heigh map: canopy map minus bare-earth elevation. Please see "Hyperspectral_Training.pdf" for classification.
Angelo Reserve Infrastructure:  several shapefiles showing legal boundaries, watersheds, streams, and buildings.

Factors used:
Conifers are very tall.  They are obvious in a vegetation height map.  Douglas firs tend to have narrow conical tops.  Redwoods have broad umbrella shaped tops.  Redwoods at certain times of the year, are indeed redder than the blue-green of Doug firs. 

Aspect: all north facing slopes tend to be conifer dominant, south facing slopes tend to be broadleaf and chaparral dominant.

Broadleaf canopy tends to be relatively uniform in height and relatively smooth, compared to conifers.  As such, it is difficult to pick out the oak species from madrone or bay, unless you know an area if one species or another from the ground.  Madrones are shade-intolerant, so have a more open canopy than the oaks. 

Chaparral is incredibly dense and even more uniform than broadleaves.  Older LiDAR flights could not distinguish it from ground, as no ground hits were recorded in the chaparral.  Only occurs on south slopes and ridges.

Meadows mostly occur next to bends in the river channel in this region.  They are actually strathe terraces - bedrock flat areas caused by the river cutting into the hillside, then abandoned.  Very little soil, so the only thing that grows there is grass.

Riparian polygons were not hand drawn.  These were imported from a prior calculation using hydraulic geometry (equation using slope and drainage area).  

No assumptions made for areas without polygons.  This project was done within 2 days and I covered as many places as I was reasonably confident about the identification.  

Note most conifer classifications are identified as redwood or douglas fir.  Only some broadleaf are identified, as it is much harder to ID them remotely with the tools at hand. Chaparral probably should remain a single class instead of being broken into individual species. Likewise grasses are not possible to distinguish.  No Bay were individually identified in this excersize. 






