WC146 JanFeb 2026 - Magazine - Page 26
STORMWATER
HSAI was inspired by the realization
some 昀椀ve or six years ago that AI
would play a critical role in the future
of water resources management.
HSAI represents a signi昀椀cant development in hydrological forecasting—expanding the accessibility and
reach of detailed predictions.
26
WATER C AN ADA • JANUARY/ FEBRUARY 2026
CATCHMENT ATTRIBUTES: A database of geographic attributes for
thousands of catchments (up to 7,000 in Canada), detailing
everything from elevation and slope to land use and soil type.
“You take those three ingredients, which are not just based
on a single watershed, but span the entire content. And the
machine learning algorithm learns how to predict streamflow
in any catchment under any climate or weather forcing data,”
says McNeill.
By learning from this wide range of data, the model can
generalize its knowledge, meaning it can forecast streamflow in
areas that don’t have physical gauges.
Challenges across streamflow forecasting models
While the model is generalized, Aquanty says it performs especially well in snow-dominated areas—think alpine areas in the
West or Northern Ontario and Quebec—often boasting scores
over 90 per cent.
The lower-scoring areas present challenges for all hydrological
WAT E R C A N A D A . N E T
Aquanty, OPG
typical statistical model as having short-term memory—it only
looks at the input right now—the LSTM has a crucial upgrade.
“The ML models have a memory cell,” explains Erler. “They
can actually track the state of the watershed. In this sense, they’re
quite similar to physical models.”
This memory is vital in hydrology because it lets the model
account for slow-moving, long-term conditions like snowpack
and groundwater.
To teach this model how water behaves across the entire continent, Aquanty didn’t train it on just individual watersheds; they
used a massive, diverse dataset to make it truly generalized. The
model learns from three big sets of data:
STREAMFLOW HISTORY: Up to 50 years of observed streamflow data
from the Water Survey of Canada hydrometric network (about 1,500
gauges in Canada) and gauges from the U.S. Geological Survey.
WEATHER HISTORY: Up to 50 years of observed weather data paired
with the streamflow records, compiled from global and Canadian
datasets.
