Dr. Jeff Boisvert Professor jbb@ualberta.ca
A list of my publications can be found here.
My research focus is spatial numerical modeling, uncertainty quantification, data analytics, and machine learning applied to mining, petroleum, environmental, and wildland fire applications to make important decisions such as: should a mine be built; should exploration wells be drilled; how much land is contaminated by a spill; where will a wildfire spread in the next 24 hours? The usual goal is uncertainty management which aims to improve decision making in the presence of limited/uncertain data, often in a way that maximizes efficiency, generates value, minimizes environmental impact, and/or reduces risk. Research results contribute to improving spatial data collection, assessing data error, using data in spatial modeling, evaluating model response to relevant processes (e.g., mine designs, well designs, groundwater flow, fire spread), and improving engineering decisions based on the range of model behavior. Some examples of recent areas of focus follow:
1. Model nonstationary domains: Geology often displays complex non-linear features such as veins, channels, or faults, resulting in complex spatial relationships that manifest as nonstationary features that are important to model; incorporating these locally varying features improves model accuracy when geology is complex. This includes automatic domaining and model parameterization using machine learning algorithms and provides an alternative workflow for modeling non-stationary domains.
2. Stochastic modeling of geological features: I contribute to improving stochastic techniques for modeling geological features, rock types may be stochastically distributed and not follow geological structures; the goal is to generate models with increased geological realism, when geology is better reproduced stochastic models are more likely to be accurate and decisions made are improved. Quantifying and demonstrating this improvement is also important.
3. Mineral resource/reserve modeling: Common methodologies for calculating resources often rely on geometric measures such as distance to samples or sample spacing. Moreover, mine planning rarely considers uncertainty in mineral grades; design is based on a single model due to algorithm and computational limits. My research in this area advance mine planning, stope design, and drill pattern optimization considering uncertainty; designs that incorporate uncertainty in grade models minimize economic risk and lead to sustainable mining practices.
4. Numerical spatial modeling utilizing remote data: This work uses drone and satellite data to improve spatial modeling of wildland fires. Convolutional neural networks identify and classify fuel (i.e., trees) from RGB, thermal, multispectral, and LiDAR data. Variables of interest relate to fuel and weather; spatial models of these variables help predict fire behavior, make hazard/risk assessments, remotely triage values at risk around wildland fires, and help support other operational decisions at active fires.
All my research is conducted through the Centre for Computational Geostatistics (CCG), a UoA industry consortium of 30+ mining, petroleum, and software companies that I co-direct; my long-term objective is to establish the CCG as the industry leader in geostatistical research and training. Specifically, long-term objectives of my research are to develop spatial modeling solutions to improve decisions in the presence of incomplete data, make implementation practical to encourage widespread industry use, and train exceptional MSc and PhD students in modeling and decision making in the presence of uncertainty. This overarching long-term objective is broken into four subobjectives (1) encourage and improve the quantification of uncertainty in resource modeling for risk reduction and design planning, with tools that are widely accessible by diverse users (2) automate spatial modeling processes that can be automated, while demonstrating the dangers of full automation (3) promote, integrate, and use remote data autonomously and in near real-time to build flexible models that incorporate all available data at all scales and adapt to new information in real-time (4) transition from data-fusion to decision-fusion, where all relevant sources of data are used to build models (data-fusion) but workflows exist to aid in making risk quantified decisions based on all available information (decision-fusion).
My wildland fire research is an extension of geostatistical modeling and uses similar workflows. The long term goal is to expand remote data collection, improve spatial modeling of relevant variables, and create workflows for using spatial models in fire management decision-making; we aim to demonstrate to firefighters, fire prevention managers, and governments that collecting this data has value, improves fire mitigation and response strategies, enables hazard assessments, and increases prescribed burn safety. This research resulted from my experience as a firefighter and noticing that spatial data is often used ineffectively in fire management, there is a clear research gap. The long-term plan is to collect data at active fires, show how it could have improved decision-making, and simplify use of this data in decision making. This research has clear benefits in reducing health impacts of fire, limiting damage to values-at-risk, lowering costs of fire management, and increasing effectiveness of proactive approaches. I continue to work as a paid-on-call firefighter with Parkland County.