He reckoned there was only fragmented data on pockets of tree species across New Zealand — “but nothing comprehensive”.
“This is why we are trying to capture an inventory of exotic planted forests and trees in New Zealand,” he said.
“Forest Insights aims to answer: What trees do we have? Where are they? And eventually, what can we do with them?”
The idea is the new interactive tool — powered by machine learning and deep learning models — will provide forest owners, managers and wood processors with an overview of the changing availability and growth of a range of tree species over time.
“So you’ll be able to go to a website and, for example, type in a particular species of tree and be presented with not only a satellite map of where they are located across Aotearoa but also how much volume there is estimated for each species,” Evans said.
“The aim of this is that processing plants, sawmills — all of the industry in fact — can start saying: ‘There is starting to be a significant volume of X here and there is a good-looking supply so maybe it’s worth looking at investing in setting up a second or specialised processing plant’.”
It outlines the boundaries for stands of trees above 4ha and provides essential details such as age class, area in production, and the number of stems per hectare.
“To do this, Scion’s data and geospatial intelligence team have developed a deep learning-based model that can detect and map planted forests using only aerial imagery.”
The AI model targets planted exotic forests and can map stands as early as two or three years after planting, once a minimum canopy size is reached.
Although the model can identify multiple exotic species, the results for radiata pine are shown because this species makes up most of the exotic forests planted in New Zealand.
Scion’s data and geospatial team collected and labelled more than 400sq km of aerial imagery to build the model, making it one of the largest data sets for high-resolution landcover mapping.
Grant said this model’s advantage was that it produced very accurate forest boundaries using only RGB aerial imagery.
“The model is also trained to work across imagery with different visual characteristics and resolutions.
“This allows us to take advantage of the aerial imagery datasets that are regularly captured by regional entities and served by Linz as a national aerial imagery base map.”
Cyclone damage
The Forest Insights AI model had its genesis in trying to map the impacts of Cyclone Gabrielle, but Scion has plans to provide the same data for a wider range of trees for other regions.
Significant ground truthing has taken place to help reinforce the model’s accuracy.
“The first areas mapped using the Forest Insights’ AI model included Gisborne and Hawke’s Bay — areas that were severely affected by Cyclone Gabrielle in February 2023,” Evans said.
“These layers can serve as a snapshot of the forest before the cyclone and the team has repurposed the model to quantify the forest loss and windthrow damage in these areas using a range of imagery sources.”
The next steps for the team will be to combine the high-resolution forest boundaries with LiDAR for the remainder of New Zealand to estimate attributes such as age class, stand density, height, timber volume and industry infrastructure like ports, processing plants and sawmills.
A digital “twin”
By partnering with Linz to access its LiDAR data warehouse, the team aims to extend this approach to all regions of New Zealand.
“In the longer term, we aspire to turn this into a national-scale ‘digital twin’ of the productive forest estate in New Zealand,” Evans said.
“This would allow mapping and monitoring of our forests using remote sensing.
“To create the twin we need regularly updated data flowing into the model; the team is currently exploring this.”
Evans said because forests were a critical tool in the fight against climate change, it was important to understand how they will adapt under present and future conditions.
“Sensed data inside our digital forest can help us simulate the future and improve decision-making in the forestry sector.”
Find out more here.
-SunLive.
