WC146 JanFeb 2026 - Magazine - Page 24
STORMWATER
Forecasting the Unpredictable
How machine learning is 昀椀lling gaps in stream昀氀ow forecasting Canada-wide
OR CANADA, a nation defined by its vast
freshwater resources, reliable streamflow forecasting is paramount for public safety and
infrastructure management. From annual
spring melts in major river basins to sudden, intense summer thunderstorms, the ability to
accurately forecast water movement is essential for
flood warning, dam operations and drinking water
security. Today, the sector faces compounding pressures from climate change, which introduces more
predictable, volatile weather patterns. This demand
for faster, more scalable flood forecasting capabilities
is an opportunity for modern technology, specifically Artificial Intelligence (AI) and Machine Learning
(ML), a welcome complement to established physics-based models of streamflow forecasting.
At the forefront of this shift in Canada is HydroSphereAI (HSAI), the recipient of the 2025 Water
Canada New Tech award. HSAI, developed by
Waterloo-based water science and technology firm
Aquanty, part of Rocscience group, was inspired
by the realization some five or six years ago that
AI would play a critical role in the future of water
resources management.
“As a water technology company who has been
a leader in physics-based hydrologic modelling, we
realized we had to integrate AI into the suite of tools
that we provide our customers,” says Steve Frey,
director of research services at Aquanty.
F
Lauren Belayneh is
Associate Editor of
Water Canada.
24
WATER C AN ADA • JANUARY/ FEBRUARY 2026
Filling gaps in streamflow forecasting
The physics-based model has long been the gold
standard in hydrology. While these models provide
accurate data that conservation authorities and
municipalities rely upon, they take lots of time and
funding to set up—and there are still many ungauged
watersheds. Aquanty realized there was an opportunity to create a tool that fills the gaps and can be
deployed and scaled widely.
“Because the models are trained on many different
basins, they learn the major physics of runoff generation like snowmelt. Because of that, they can also be
transferred between basins and to ungauged basins,
which is a very powerful feature,” explains Andre
Erler, senior climate scientist at Aquanty. “This means
they are still helpful in a changing climate.”
The result, HSAI, is designed to provide a highly
skillful solution for forecasting, especially in locations
where physical models have not yet been implemented.
Just how skillful is HSAI? According to internal
testing across Canada, the results are impressive.
“After doing a meta analysis of streamflow forecasts
across hundreds of stations across Canada, we’re
seeing exceptional skill of over 86 per cent—or a
skill score of 0.86, for the more technically minded.
A general industry standard of 75 per cent would
be considered very good,” says Brayden McNeill,
Aquanty’s sales and marketing lead.
How AI predicts streamflow
HSAI’s secret to achieving this high skill level lies in
both the data it analyses and its architecture—specificially, a type of machine learning called Long ShortTerm Memory (LSTM) systems. If you think of a
WAT E R C A N A D A . N E T
B.C. Ministry of Transportation and Transit
BY LAUREN BELAYNEH
