We're excited to share that HydroForecast is now running operationally inside Delft-FEWS at Ontario Power Generation (OPG), building on the multi-year deployment we announced last year. It's a live integration built to be reproducible and to scale. We recently presented this work at CWRA FEWS User Days alongside Ming Han, Water Resources Engineer at OPG.
Enhance your forecast stack with HydroForecast
HydroForecast gives you a cutting-edge AI streamflow model built on years of hydrologic modeling expertise. It's designed to deliver more accurate forecasts than traditional approaches, especially in basins where other models struggle — ungauged basins, fast-changing conditions, or locations with limited historical data.
Many operators already run their forecasting workflows in FEWS, which handles orchestration, scenario testing, and data visualization. Rather than asking you to stand up a separate system, we built HydroForecast to slot directly into that existing environment. You get our forecast where you already work, without disrupting the workflows your team relies on.
The practical challenge has historically been the integration work itself: managing installations across machines, dependency conflicts, training internal experts to debug model issues, ongoing maintenance overhead. That friction is real, and it's the main reason new models are slow to make it into operational FEWS environments. We removed it.
Set up once, run anywhere
Getting a new model's software stack installed and running correctly — without conflicting with everything else on the machine — is often one of the biggest barriers to adoption. If you've ever heard the phrase "well, it worked on my computer," you know the struggle. Containerization solves this by bundling everything a model needs into a single, self-contained package.
To run a HydroForecast model on your machine, it's now: one tool to install, one file to download, one command to run. It's the difference between being handed a box of parts and being handed something that's already put together and ready to go — helpful when you're trying to run an operational system, not a science project. For Ming Han at OPG, getting HydroForecast running for the first time took less than five minutes, and worked on the first try.
Because every model runs through the same container with different parameters, this also makes debugging far more reliable — we can replicate your exact environment when something goes wrong, not a close approximation, so we can support you from anywhere.
All of our models — different model types, any basin, any input data — run with the same single execution command. If you can run one, you can run any of them.
A templated FEWS configuration
Once you have a container that runs anywhere, the FEWS integration itself gets simple. Partnering with Dave Casson, Water Resources Engineer and FEWS expert, we built around a few principles:
- One General Adapter handles all import/export activity — every model call goes through the same container, so it all generalizes into a single configuration.
- Templatized workflows with a small set of global settings that cascade everywhere they're needed. Updating to a new model version is a one-line change.
- Reusable statistics and visualization starting points, so new models plug into the same evaluation infrastructure already built for existing ones.
The result: a FEWS environment that's easier to maintain and easier to extend. Trying a new setup or model configuration means branching the template, setting a few variables, and looking at results — not rebuilding the integration from scratch.
Better forecasts, without the integration headache
If you're running FEWS and want to add a cutting-edge AI streamflow model that's proven out in production — not a research prototype — the setup is simple: no ground-up integration effort, no dedicated debugging expertise required. This is what's running at OPG today, and as we keep investing in containerization and templatization, adding HydroForecast to a new FEWS environment keeps getting simpler.
Ready to bring HydroForecast into your FEWS environment, without the integration lift?
