DATE: 2018-09-07

AUTHOR: John L. Godlee

--------------------------------------------------------------------------------

layout: post title: "A guide about processing hemispherical photos" date: 2018-09-07

--------------------------------------------------------------------------------

I wrote a guide for some undergraduate students on a field course about hemispherical photography and calculating forest canopy traits. This is it. It's untested so far, so some parts may change depending on how well the field course goes. The guide may get updated, so the most up to date version can always be found here, on Github[1].

1: https://github.com/johngodlee/hemi_photo_guide

A list of tips for taking good hemispherical photos:

• Take photos under a uniformly overcast sky, ideally before the sun has risen too high in the sky, or just before sunset. I find in the morning the photos are generally better and at high latitudes you will have more time than in the tropics..
• Ensure that the camera is level and the lens is pointing straight up. Use the spirit level on the camera hotshoe to do this.
• Adjust the tripod so that the top of the camera lens is 1 m above the ground, or above any understorey vegetation, whichever is higher.
• Turn the camera so the top of the camera body is facing north, take a compass! This ensures that the top of the captured photo is also facing north, which is necessary for calculating LAI..
• Make use of the articulated display on the camera to get a good view of the photo before you take it.
• Set the camera:

* Manual shooting mode

* Manual focus

* Set the focus to infinity

* Exposure compensation = -0.7

* Capturing fine jpeg & RAW images at the same time

* The camera time and date is accurate (this is purely for ease of matching photos to sites)

* Set the Aperture to 5

* Adjust the ISO and shutter speed so the photo is neutrally exposed but the shutter speed is always over 1/60sec, otherwise you will introduce camera shake when you press the button

* Take all photos in landscape dimensions, never portait.

• Make sure you all duck down below the camera when the image is being taken!
• Make sure there is battery and you have the spare battery
• Make sure there is an SD card in the camera, and take a spare.
• Cover the lens with the lens cap between photos. PLEASE PLEASE PLEASE!!!

1. Open ImageJ

2. File -> Open, then select an image

3. Visually inspect the image to see that there isn't massive amounts of lens flare. If you have lots of lens flare, the photo should be thrown out! This is what lens flare looks like:

Lens flare example

4. Image -> Type -> 8-bit

5. Image -> Adjust -> Threshold, manually adjust the image so all the branches are red and the sky is white, or as near as you can get it.

6. Save the newly thresholded image as a jpeg in a folder called img.

7. Rinse and repeat for all images.

The above process can be automated with a macro, but this assumes that the images are all uniformly exposed.

This is the macro, saved as a .ijm file. This is untested so use at your own risk:

// Automatically create a thresholded image for use in further analysis. Change the values of setThreshold to achieve different results.

// Partially tested

// Save as a Jpeg in the Batch macro dialog in ImageJ

run("8-bit");
run("Threshold...");
setThreshold(0, 146);
setOption("BlackBackground", false);
run("Convert to Mask");

1.

Open RStudio.

2.

Open a new script (File -> New File -> R Script)

3.

Save the script in a folder above the images folder:

4.

Enter the following preamble into the R script:

# Set working directory to location of thresholded images
setwd("LOCATION_OF_ANALYSIS")

# Source the functions used to calculate stuff
source("hemiphot.R")

# Packages
library(jpeg)

5. Add white_image.jpg to the same folder where the thresholded images are found

6. Read in all the thresholded images and create an empty data frame which will later be filled with canopy trait statistics like LAI and canopy openness.

# List all images in the directory
all_images <- list.files("img/", pattern = ".JPG")

# How many images
img_length = length(all_images)

# Create empty dataframe, 6x7 and fill it with zeroes
all_data = data.frame(matrix(data = 0, nrow = img_length, ncol = 7))
names(all_data) = c("File", "CanOpen", "LAI", "DirectAbove", "DiffAbove", "DirectBelow", "DiffBelow")
# Fill first column with image names
all_data[,1] = all_images

7. Read in the reference image (white_img.jpg) as a matrix of pixel values:

white_img <- readJPEG("img/white_image.jpg", native = F)

8. Set some parameters for the location the photos are being taken. Approximate location (0.1 degrees latitude) is good enough for our purposes. Note that the values below are for somewhere in Africa and should be changed:

location.latitude   = -15
location.altitude   = 200
location.day        = 30
location.days       = seq(15,360,30)   # roughly each mid of the 12 months

9. Set some parameters for the images, cropping them to a circle and setting the threshold. These parameters are ones I have used on this camera, so don't need to be changed:

## Image parameters

### Drawing circles and identifying the image centre point
hemi_dim <- dim(white_img)
radius <- max(rowSums(white_img[,,1] > 0.4) / 2)

### determine using a single image and fill in here for batch processing
location.cx         = (hemi_dim[2] / 2)             # x coordinate of center of image
location.cy         = (hemi_dim[1] / 2)             # y coordinate of center image
location.cr         = radius             # radius of circle
location.threshold  = 0.42  # Must get this to match all images, or maybe could use a lookup table / dictionary?  Does R have dictionaries?

10. Set some atmospheric parameters. I've loosely estimated these for this location, but by no means is it scientific. I would not have much confidence in the statistics generated using these parameters, namely DirectAbove, DiffAbove, DirectBelow and DiffBelow.

# atmospheric parameters
## Atmospheric transmissivity - Normally set at 0.6, but can vary between 0.4-0.6 in the tropics
location.tau = 0.6

## Amount of direct light that is used as diffuse light in the Uniform Ovecast Sky (UOC)
location.uoc = 0.15

11. Run a big for loop to calculate the statistics for each photo

for(i in 1:img_length){
    ## read file
    image <- readJPEG(paste("test_img/", all_images[i], sep = ""), native = F)

    ## conver to Hemi image
    image <- Image2Hemiphot(image)

    ## set cirlce parameters
    image <- SetCircle(image, cx = location.cx, cy = location.cy, cr = location.cr)

    ## select blue channel
    image <- SelectRGB(image, "B")

    #threshold
    image <- ThresholdImage(im = image, th = location.threshold, draw.image = F)

    # canopy openness
    gap.fractions <- CalcGapFractions(image)
    all_data[i,2] = CalcOpenness(fractions = gap.fractions)

    ## calculate LAI according to Licor's LAI Analyzer
    all_data[i,3] = CalcLAI(fractions = gap.fractions)

    ## Photosynthetic Photon Flux Density (PPDF, umol m-1 s-1) P
    rad <- CalcPAR.Day(im = image,
    	lat = location.latitude, d = location.days,
    	tau = location.tau, uoc = location.uoc,
    									draw.tracks = F, full.day = F)
    all_data[i,4] = rad[1]
    all_data[i,5] = rad[2]
    all_data[i,6] = rad[3]
    all_data[i,7] = rad[4]
}

12. Finally, look at the output, which is stored in all_data

all_data

<hr>

The hemiphot.R source file comes from here[2].

2: https://github.com/naturalis/Hemiphot