Skip to content

Latest commit

 

History

History
116 lines (88 loc) · 2.76 KB

D1_MarconiS.md

File metadata and controls

116 lines (88 loc) · 2.76 KB
layout title permalink
page
DayOne
/D1_MarconiS/

Tree based species identification and biodiversity estimation using NEON AOP data

In this module we will see how to use LiDAR and hiperspectral data to identify tree species out of a scenery

you will need:

  1. intermediate knowledge of R
  2. the lidR package

INDEX:

  • Visualize LiDAR data
  • Perform crown delineation
  • Use SVM to perform classification
  • Calculate biodiversity

Visualize LiDAR data

First of all, we will need a LiDAR file, download it from https://blablabla.io

Brief description of the site and file

library(lidR)
library(raster)
pt <-  "./inputs/LiDAR/"
f <- list.files(pt, pattern = ".laz")[1]
las = readLAS(paste(pt, f, sep=""))

normalization

lasnormalize(las, method = "knnidw", k = 10L)

Let's see what we have just imported in our environment To plot the point cloud data we will be serving of an external software: Xquartz

plot(las, color="Intensity", colorPalette = terrain.colors(50), trim=0.95)

pointCloud

You can also overimpose an rgb raster on top of it

image_rgb   = stack(paste(pt, list.files(pt, pattern = "image.tif")[1], sep="/"))
#plotRGB(image_rgb, r=2, g=1, b=3)
# Get each channel in a RasterLayer
ch1 = image_rgb@layers[[1]]
ch2 = image_rgb@layers[[2]]
ch3 = image_rgb@layers[[3]]

# Attribute the channels in fields R,G and B
lasclassify(las, ch1, "R")
lasclassify(las, ch2, "G")
lasclassify(las, ch3, "B")

# build the color palette give the 3 channels
lascolor(las)

pointRGB

#plot the scene
plot(las)

Create a raster of heights

    chm = grid_canopy(las, res = 0.5, subcircle = 0.2, na.fill = "knnidw", k = 4)
    chm = as.raster(chm)

Unfortunately, LiDAR point cloud data may be noisy. This means we would be willing to blur the image a little to avoid the crown delineation method to see artifact crowns

plot(chm)

chmNoSmooth

kernel = matrix(1,3,3)
chm = raster::focal(chm, w = kernel, fun = mean)
chm = raster::focal(chm, w = kernel, fun = mean)

plot(chm)

chmSmooth

Now, let's try to apply a crown delineation algorithm (Silva et al., 2016). First of all we will need to look for the highest points, which will represent the top of our trees. Let's look at what are the parameters involved:

?tree_detection
ttops = tree_detection(chm, 5, 2)

5 and 2 seem a reasonable way to go, for this case. At this point we can predict our crowns' shape!

crowns <-lastrees_silva(las, chm, ttops, max_cr_factor = 0.6, exclusion = 0.3, extra = T)
#find a way to polygonize which doesn't involve the use of python

plot(crowns)

crowns