The work aims to improve ecosystem health, fisheries management, infrastructure protection, and emergency response planning—helping decision-makers anticipate change rather than react to it.

Forecasting the unknown

At the heart of Johnson’s research is a deceptively simple question: How do you make reliable predictions when there’s very little data?

Traditional hydrologic models depend on long-term monitoring records—something Alaska often lacks due to its extreme remoteness and expansive terrain. Johnson’s team is tackling this challenge by combining machine learning (ML) and artificial intelligence (AI) with observational data from similar environments around the planet to bypass the limited regional data, a technique known as transfer learning.

“Transfer learning allows us to train our ML models to regions with similar hydrology and apply it to places where observations are limited, as in Southcentral and Southeast Alaska,” Johnson explained. “It’s a way to expand predictive capability without waiting decades to collect new data.”

The research will contribute to expanding the National Water Model (NWM) into Alaska—a much needed integration, yet one that presents major scientific and operational hurdles.

Why streamflow—and temperature—matter

The project focuses on two critical predictions:

• Streamflow, which informs flood forecasting, emergency response, and wildlife management
• Stream temperature, a key indicator of ecosystem health—particularly for salmon

Salmon are often considered a keystone species in Alaska’s river systems and an essential element of Alaskan life, culture, and tourism. Their life cycle connects oceans, rivers, forests, and wildlife, transporting nutrients inland and supporting everything from aquatic insects to birds and large mammals. Even small changes in water temperature or streamflow can disrupt this balance.

Water temperature thresholds can also trigger emergency regulations, such as temporary fishing closures designed to protect vulnerable salmon populations. With several salmon species already under stress, better temperature forecasts could help agencies act earlier and more precisely.

“Having more accurate streamflow and stream temperature predictions provides natural resource managers with critical information for making informed decisions that maintain healthier fisheries and create resilient communities,” Johnson said.

Use-Inspired model development

A defining feature of the project is its use-inspired model development—building scientific tools around the real needs of the people who rely on them.

Johnson’s team is partnering closely with local fisheries, nonprofits, regional stakeholders, and the Alaska Pacific River Forecasting Center (APRFC) to ensure the models are both scientifically robust and operationally useful.

“We’re able to work with end users from the beginning so the tools we build actually serve the people making decisions on the ground,” said Kaitlin Meyer and Liza McLatchy, two University of Utah civil & environmental engineering Ph.D. students involved in the project.

The collaboration also gives Johnson and his students the opportunity to collaborate with experts across institutions, including Martyn Clark of the University of Alberta, widely regarded as one of the world’s foremost hydrologic modelers.

Fieldwork at the last frontier

The research will take place during two key windows in Alaska’s hydrologic year:

• May, during peak snowmelt season
• August, when rivers experience low flows and glacier-driven contributions dominate

These field campaigns will give students firsthand experience in cold-region hydrology—working at the intersection of water, environment, data science, and real-world impact.

In Alaska, protecting salmon means protecting entire ecosystems. By improving how we forecast rivers today, this research helps ensure those ecosystems—and the communities that depend on them—can thrive tomorrow.