Safety: This solution could lead to significant cost savings as well as safety benefits by reducing reliance on helicopter surveys.
Cost: These surveys currently cost Albertans $2.6 million annually. A 50% reduction would save over $1 million annually.
Forest Health and Adaptation strives to lead in science-based, innovative management of forest health. Part of their mandate includes monitoring outbreaks of biological and environmental disorders that adversely affect the health of Alberta's forests.
Forest Health works closely with forestry companies to help monitor and control mountain pine beetle infestations to enable companies to adjust logging plans in order to recuperate as much economic value out of the forest while helping to minimize the spread of infestation.
This case study focuses on the central-west area of Alberta, close to Hinton, because it has a large population of pine trees and is an area of interest for Forest Health due to the number of Forest Management Agreements (FMA). Through an FMA, a company is given certain rights, including the right to establish, grow, harvest and remove Crown timber for economic value, in exchange for various responsibilities such as forest management planning. Forestry is a significant economic driver in this area, accounting for thousands of jobs; thus, it is imperative that Forest Health identify mountain pine beetle disturbances early and accurately to support FMAs in minimizing spread of infestation while obtaining economic value from wood products.
Mountain Pine Beetles (MPB) are attacking the province's pine trees resulting in significant forest damage. Left unmanaged, MPB will devastate Alberta's pine forests and spread eastward across Canada's boreal region.
The mountain pine beetle (MPB) infestation continues to represent a threat to timber supply and healthy forest ecosystems in Alberta. Over 100,000 infested trees were controlled under the MPB management program in 2019-2020, with funding of up to $30M allocated to helping manage infestations in Alberta. Surveys make up an important part of this cost, with helicopter surveys accounting for $2.6 million in annual costs. In addition, they are time-consuming and have a limited breadth of areas surveyed. Given that there is a short time window between when an infestation occurs to when Forest Health can prevent a spread, Forest Health was looking for a solution using satellite imaging to help with timely detection across the province so that it could effectively and expediently respond to infestations and provide communication to stakeholders such as the Wildfire Management Branch to assess disturbances to the forest that could increase the risk of wildfires.
Thus, the project’s goal was to equip Forest Health Officers with timely tools that could help them better manage MPB's across the province while reducing their reliance on helicopter surveys.
We developed a machine learning solution that complements Forest Health and Adaptation's process of detecting red-attack Mountain Pine Beetle stand infestations. It takes satellite imagery as the input, and it maps in a file format that can be used by Forest Health’s planning software.
The model: Single class object detection was applied to high resolution (1.5m) satellite imagery to predict and locate infected trees. The model was trained on manually-labeled images, and testing found that the model has over 61% precision and 85% recall, meaning that 85% of the infected pine trees were detected by the satellite image-based ML model. The results will be incorporated into Forest Health Officers’ workflow through a simple web-based user interface. This software tool augments the capabilities of forest health officers, giving them the ability to take proactive action with a solution that is accurate and cost-effective.
Through early ML model results of the single tree detection model, we are seeing strong potential to detect red trees at the single tree level, leading to the potential freeing up costs associated with heli-GPS surveys. Costs could be alleviated by better prioritizing infested areas in advance of costly aerial surveying techniques.
Proactive: Potential for ground identification of infected pine locations prior to aerial surveys.
Augmenting: Train the eyes of forest health officers for efficient detection of infected pine during fixed wing flights.
Accurate: Increase in accuracy of detection of infected pine locations for better control planning.
Cost-Effective: Up to 20% of freed up costs related to reducing the amount of aerial survey areas.
AltaML is working with the Forest Health and Adaptation team to build additional features to be deployed for the 2023 season. In addition, we are exploring the possibility of bringing these benefits to other regions, and for use with other pests. If you’d like to connect us with potential customers, discuss opportunities, or just talk about Machine Learning please give us a shout at firstname.lastname@example.org.