An email about hemispherical photography


This is a transcript of an email conversation with a colleague that I had about hemispherical photography and the various merits of different methods. I thought the content was useful and wanted to record it so I’ve put it up here. I’ve cleaned up the email a tiny bit, just to make things clearer.

Colleague’s email

Dear John,

My supervisor suggested that I get in contact regarding your knowledge of a new hemispherical camera that she thinks you’re using in your research. I’m not sure if it’s the tool for the job, but I’m trying to map the canopy cover of a very small area of woodland in the Scottish Borders.

One part of my PhD research is looking at the influence of this strip of mixed woodland on downslope soil moisture and groundwater dynamics, so I’m mainly gathering sub-surface data on a transect across the woodland. However, I also need to estimate canopy cover in summer/winter as well as the extent of shading on the land either side of the woodland.

I’m a geologist, not and ecologist, and I’m sure there are standard methods for doing this, so if you have any thoughts either using the hemispherical camera or other methods, I’d be interested in any quick thoughts that you have.

Best wishes,

My reply


Yep I got some money from the School last semester to buy some hemispherical photography gear for the equipment store, which is free to use. The available equipment consists of:

I’ve used this camera to estimate canopy cover of savannas in southern Africa and I can only imagine that it would work fine for your patch of woodland as well.

When you say “map” do you mean get a spatially explicit estimate of the canopy cover throughout the site? I’ve only ever used hemispherical photography to get a single plot level estimate of the canopy cover. Basically the mean and variance of the percentage canopy cover as estimated from many photographs taken at points on a regular grid laid out in the woodland area. Each photo is essentially a point estimate of the canopy cover. You could probably do a map, but you would have to increase the density of the grid quite a bit to truly capture the variation over space. To give you a rough idea, on a 100x100 m (1 Ha) plot, I normally take 100 photos to get a plot level estimate. Taking the photos doesn’t take very long at all, but setting up the grid can be a faff if the woodland is thick.

Processing the hemispherical images can be a pain but is fairly automated once you have a workflow set up. In the past I’ve used imageJ ([]), and if you only want to estimate percentage canopy cover then I see no problem with using it. I have some imageJ macros to batch analyse images if you want. If you want to estimate more advanced things like Leaf Area Index (the unit leaf area per unit ground area) or available Photosynthetically Active Radiation below the canopy, you will need to use something more advanced. I’ve recently discovered a set of R scripts collectively known as HemiPhot ([]) which can estimate these parameters.

The main thing to remember when taking hemispherical photos of the canopy is that you have to do them early in the morning or late in the evening, before the sun is overhead and too bright but with some ambient light, otherwise you will find that you get a sun flare on the lens, which makes the image basically unusable for analysis.

I’ve attached a few papers which you can read if you want to on the subject of how hemispherical photography (and other methods) is used in forest/woodland/plantation contexts to estimate tree canopy structure. By no means should you read them all, but they might be useful further down the line.

There are other methods for estimating canopy cover, but having experimented with most of them, I think hemispherical photography gives the most accurate result. Other options are to use a convex mirror densiometer ([]) or to use a periscope densitometer ([]). The periscope densitometer might be an option for making a high point density map of your site, as the measurements are quite quick so you can do more of them. The periscope densitometer method just requires you to talk along the grid and at each point take a yes/no reading of whether there is canopy touching the crosshairs of the periscope mirror. You wouldn’t be able to make a map of percentage canopy cover with the periscope densitometer, only a plot level estimate as it uses the binomial nature of the measurement (yes or no) to statistically estimate percentage cover, the value of each point on its own isn’t useful. I wouldn’t EVER recommend the convex mirror densiometer as they suffer from pretty serious researcher bias.

Measuring the shade on the land either side of the woodland would require a different method I think, though I’ve never done it myself. I get the impression that in a closed canopy woodland at this high a latitude, you could assume that when the path from the Sun to the open ground adjacent to the woodland is blocked by the woodland, all the direct sunlight is blocked. Considering this, you could just measure the maximum tree height at increments along the edge of the woodland using a clinometer or a laser range finder, measure the orientation of the woodland edge, then use that to model how long the shadow is at different times of the year and how many hours during the day a given distance from the woodland is shaded as the angle of the Sun changes. This has some assumptions/caveats though, 1) the woodland is thick enough to block all direct sunlight, and 2) the boundary of the woodland edge is a straight line. If the woodland edge isn’t a straight line it gets marginally more difficult as you would have to include more measurements of the distance of the woodland edge into your calculations of shade at different points.

These are the papers I attached: