Vegetation management is often a numbers game. It involves constantly evaluating the odds of an incident, and making calls based on often limited information. But it’s also about massive scales: huge distances, broad expanses, and hundreds of thousands, even millions of trees.
Tree species is an essential aspect of vegetation management, whether you’re looking after entire forests, or stretches of woodland along a powerline or railway ROW. Being able to identify tree species rapidly, accurately, and at scale offers major technological, economical and ecological benefits.
The reason is simple: species is strongly correlated to a number of important tree characteristics, from median height and span to foliage density and seasonality, wood hardness, growth rate, disease risk, combustibility, biomass, carbon fixation and sequestration potential, economic value, and more.
Any improvement in the accuracy on these characteristics, leads in turn to significantly improved outcomes for key vegetation management functions: predicting infrastructure hazards, forest management and valuation including biosphere protection and wildfire mitigation, as well as carbon and GHG management.
In an ideal world, a vegetation manager would want to know the species of every tree under his or her care. But the reality is very different. Because of the problem of scale, this is virtually impossible using traditional means.
Traditional tree species identification is often done from the ground–on foot, horse or vehicle–but this is so labor intensive that it is limited to key areas. Since the 1940s, aerial photography has been used too. typically relying on pictures taken from airplane or helicopter overflights. This approach is limited in a number of ways however, perhaps most particularly by the need to collate and reassemble all the pictures in order to increasingly map of the results. Drone photography has helped overcome some of these limitations, but not completely.
In the past few years, the spatial and spectral resolution and availability of satellite imagery have greatly increased, and the price of satellite data has reduced dramatically. What’s more, a single satellite image can cover 10,000 hectares of forest. No need for calibrating, aligning and stitching the images, as you need to do with images from lower-altitude sources. And resolutions are now sufficient for tree species identification, even down to the level of individual trees.
When combined with ML, the result is powerful, accurate vegetation identification and inventorying on a massive scale. For the 10,000 hectares mentioned above, we could be talking upwards of 10,000,000 trees - the kind of scale that only AI can reasonably manage.
Better still, the insights provided by the species identification algorithms can be fed into other AIs designed for broader purposes such as those described above: growth prediction, risk identification, biotope monitoring…
While many take a “recognise the note” approach, tuning their AIs to recognise the species of a tree from a single snapshot, only LiveEO takes a “learn the melody” approach: observing changes over time and identifying the patterns in those changes, looking back across months, seasons, and years. The patterns identified this way are far more distinctive, and can work on images at lower resolutions than simple point-in-time identification. The result is improved accuracy and lower cost on satellite data requirements.
Would you like to know more about Species Identification at LiveEO? Get in touch!
