REIA / Research and development
Four research programs sit underneath every REIA deployment. Some of this is solved engineering. Some of it is genuinely unsolved, and we would rather say which is which than let a demo imply otherwise.
Programs
Each one starts from data that is already public and already free. Hardware and sensors come later, and only where nothing exists, because a research program that cannot start until the funding lands is a research program that never starts.
An alarm tells you a problem arrived. A forecast tells you it is coming. We train baseline models on decades of federal hydrometric and climate records for a specific watershed, then project condition days to weeks ahead from precipitation, snowpack and evaporation inputs. The output is guidance for a Nation's own decisions: when to sample, when to hold a drawdown, when to plant.
The safety rule is absolute and it shapes the architecture. The system never declares water safe. A false negative on drinking water is the deadliest failure mode a tool like this has, so the absence of a warning is never an all clear, and the interface is built so it cannot be read that way.
Optical imagery handles vegetation condition, surface water extent and algal blooms, and it is the obvious first tool. It is also useless under cloud, under smoke, and through a polar winter.
Synthetic aperture radar does not care about any of those. It sees through cloud and darkness, which matters enormously the further north you go, and interferometry between passes detects ground displacement at millimetre scale. That makes it the right instrument for coastal erosion, permafrost subsidence and slow ground movement, which are exactly the changes a person standing on the ground cannot perceive.
Cumulative effects analysis is bottlenecked on one thing: somebody has to delineate every footprint. Cutlines, wellpads, roads, cutblocks, borrow pits, going back decades. Doing that manually across a whole territory is why cumulative assessments are rare and why they are usually out of date when they land.
Segmentation models can extract those footprints from public imagery at territory scale. The open problems are the boring ones that decide whether it is usable: false positives on natural features, recovering older disturbance that has partially regrown, and knowing when the model is wrong without a human checking every polygon.
Canadian environmental assessment evaluates projects individually. The land experiences all of them at once. That gap is the single most consequential structural problem in how development on Indigenous territory gets approved, and it is a data problem before it is a legal one.
The work is aggregation across time and across proponents, producing a defensible figure for how much of a territory is disturbed, how that has changed across decades, and what it means for the things a Nation actually cares about: caribou range, water, medicine grounds, the ability to harvest.
Honest limits
Every one of these is a live problem. If one of them is your field, this is the part of the page we most want you to read.
How well does a water forecaster trained on one watershed transfer to a different one, and how do we know when it has stopped being valid rather than merely being wrong this week?
How do we surface model uncertainty to a Guardian in the field in a way that changes what they do, rather than being a number nobody reads?
How much historical disturbance is recoverable from archive imagery before the resolution simply is not there, and how do we express that limit honestly in a total?
What does meaningful validation look like when ground truth is expensive, seasonal, and in some places only reachable four months a year?
How do we build systems that degrade usefully on a satellite connection at thirty degrees below, rather than ones that assume an office?
Where is the line between a model that supports a Nation's decision and one that quietly makes it for them? We think it matters more than the accuracy figure.
Rooting depth against a soil profile. Some of what matters most is not visible from orbit at all, which is why field measurement is a research input and not an afterthought.
How we work
Evidence, not decisions. The system reports what it measures. Chief and Council decide what it means and what to do. We do not build tools that file, submit or escalate on a Nation's behalf.
Traditional knowledge is never training data. It stays its own knowledge system, on its own terms. Where it informs the work, it does so by guiding what we look at and what counts as health, not by being absorbed into a model.
Nothing sacred is mapped. Ceremonial sites, burials, harvesting and medicine grounds are excluded from collection entirely. Not blurred, not permissioned. A blurred marker still tells an outsider something is there and roughly where.
Publish the failures. Method and uncertainty go out with every result, including the plantings that died and the models that did not beat a naive baseline.