By Ian Longley
When it comes to tracking the emission and dispersion of smoke there’s a lot you could learn just by watching. Are cameras a viable smoke monitoring tool?
When I’ve been asked about the monitoring of smoke, I automatically think of instruments that measure particulate matter, ideally continuously. This is what happened last year when I was asked if we could do a study to understand outdoor burning in a semi-rural South Island basin. Of course, I thought. We can scatter some monitors across the basin and observe the build up (or not) of smoke.
However, doubts started to creep in. Doesn’t smoke from outdoor fires rise due to buoyancy? And does it spread out or does it stay localised? We didn’t even really know where or when the fires were occurring (that was part of the point of the study). Were we looking at a serious risk of the smoke evading all of our monitors, no matter how many we deployed? We’d end up spending a lot of money but potentially learning nothing.
But monitors are not the only form of monitoring!
One of the reasons for the study, I suspected, was that smoke from outdoor burning can be highly visible over a wide area, often (unlike home heating sources) occurring when people are outdoors and more likely to notice it. Are public perceptions of outdoor burning smoke out of proportion to its measurable ground-level impact?
It is that very visibility that we chose to leverage by trialling something we’d never done before – use cameras.
Okay, that’s not entirely true. At NIWA we occasionally used cameras to monitor smoke for a few Regional Councils. This was over 10 (nearly 20) years ago. The cameras were quite epensive and the images were grainy and pixelated and difficult to analyse. As anyone with a smartphone knows, digital cameras have come a long way since then.
So we opted to purchase four different webcam models – all manufactured by Reolink. These are the type mainly sold as security devices. They were all solar powered, three with a wireless link to the cloud via the 4G mobile network, and a fourth via wifi. All retail around the $500 mark.
We installed these at elevated vantage points on private property around the basin, either on large tripods staked into the ground, or strapped to pre-existing poles. An elevation of ~60 m above the basin floor gave us a comprehensive wide view over the treetops at the expense of greater distance to the target compared to a lower vantage point. The cameras were set to capture regular still images that were bundled and sent to the cloud every 24 hours. We then ran a script to stitch the images into ~2 minute time-lapse videos (there are cameras available that will do this automatically).
And the results?
For years I have spent hours staring at the thousands of numbers recorded by monitors trying to conjure up an image in my mind of how smoke is rising and moving and shifting to form those numbers. But now I could see it for real.
Each morning watching the next video felt like opening the next chapter of an illustrated textbook of atmospheric dispersion. All the features I’d studied as a student were there in their real-world complexity – the turbulent interaction of colliding air masses, the formation and collapse of inversions, the lofting or fumigation of plumes as they interact with inversions, the exquisite “dance” of multiple plumes shimmying in unison as wind directions fluctuate. We could see haze forming as morning fog clears leading to a particle “blip” of an hour or so.
By encircling the basin with cameras we were able to use triangulation to ascertain where the fires were burning. We were able to determine their start and end times and duration.
Perhaps most importantly for this study we were able to see exactly how the impact of outdoor burning was exquisitely sensitive to the presence or absence of inversions at the time of the burn. Being able to use camera images to inform and back up explanations of the data collected by ground-level monitoring instruments gave us much more confidence in these explanations, which otherwise have been merely hypotheses.
The cameras were easy to install (<1 hour each), and completely reliable. The biggest cost for us was the time spent watching them.
However, if anything has developed faster than camera technology it might be computer vision technology. We are confident that far more information could be extracted from our image dataset, also greatly reducing the cost of analysis. Live detection and pinpointing of a fire should be relatively easy. Combine that with an understanding of current meteorological conditions and the impact of any burn could be both predicted and evaluated in real-time.
We think this method has lots of untapped potential and are keen to explore further.
