In December 2025, a series of atmospheric river (AR) storms brought record flooding to the Pacific Northwest (PNW), testing the resilience of hydropower operators and regional water managers. After months of severe drought, hydrologists and dam operators were suddenly required to shift from managing critically low reservoirs to responding to intense inflows within a matter of days.
This rapid transition – from water scarcity to flood risk – captured a growing reality for water managers: extreme weather is no longer an outlier, but an operational constant.
During the events, HydroForecast provided extended lead-time predictions that helped operators prepare for rapidly changing hydrological conditions. At several customer sites, the company identified peak flows up to 10 days in advance, four days earlier than forecasts issued by the Northwest River Forecast Center (NWRFC).
For utilities including Tacoma Power, Seattle City Light and Evolugen, that additional lead time proved critical. By combining forecast signals with expert interpretation from HydroForecast’s hydrology team, operators were able to communicate release schedules to emergency response partners and flood control authorities such as the US Army Corps of Engineers (USACE), improving coordination at a moment when clarity was essential.
What set these forecasts apart was not just how early they arrived, but how they interpreted the broader environmental picture.
“HydroForecast began flagging elevated inflow risk up to ten days before the first peaks arrived, well before most public river forecasts were showing anything unusual,” explains Alden Keefe Sampson, Chief Technology Officer at Upstream Tech. “Our model was picking up on a combination of factors: the region had been in severe drought, so the landscape was primed to absorb a lot of water quickly, but once that capacity was exhausted, runoff would accelerate fast.”
He adds that the system also draws on a wider range of meteorological inputs than many traditional approaches. “Going into this event, it was able to use the relative agreement between forecasts to gain earlier confidence that a significant storm was coming.”
Balancing flood control and reservoir operations
The first wave of atmospheric rivers initially provided relief from drought conditions as reservoirs began to refill. But the speed and intensity of inflows quickly shifted the operational challenge from water conservation to flood risk management.
Using extended-range forecasts, operators were able to absorb the initial surge while maintaining capacity for subsequent storm systems. With up to 10 days of advance visibility, reservoir managers could strategically refill while preserving the flood control space needed for storms still approaching.
This balancing act required confidence in forecasts that extended well beyond the traditional short-term operational window, and beyond the limits of historical data.
“Traditional models are built on local history, so when a storm pushes beyond anything that’s happened before in a given basin, their accuracy can suffer,” says Sampson. “We train on data from around the world, which means the model has seen extreme flood scenarios that may never have occurred locally.”
During the December events, that global perspective proved critical. In several locations, peak flows exceeded historical records, yet the system was still able to generate reliable predictions by drawing on analogous patterns from other regions.
Real-time decision-making under pressure
As the storms intensified, forecasting became not just a planning tool, but a central component of day-to-day decision-making.
At one customer site, HydroForecast issued a warning nine days ahead of the first peak inflow – four days earlier than the NWRFC forecast – and predicted peak flow rates with 35% greater accuracy, approximately 815 ft³/sec closer to observed levels. Across the first two peaks, this represented a difference of around 15,000 acre-feet of water volume.
At another site, the system predicted peak flows 10 days in advance while improving peak-flow accuracy by approximately 75%.
Behind these improvements was an iterative operational process. Dam operators continuously updated reservoir scenarios, integrating HydroForecast outputs with other data sources and evaluating the implications for storage levels and release strategies.
“Going into and throughout an event like the December 2025 storms, dam operators are constantly pulling the latest data, adjusting plans, and coordinating with other water managers and stakeholders,” says Sampson. “HydroForecast is one of the few forward-looking data sources available to our dam operators.”
He emphasises that the system’s role extended beyond providing forecasts. “After each update, operators are revising reservoir scenarios and working out the impact on reservoir levels and dam releases. Our role was less about delivering a number and more about making sure operators had the data and context to make the best decisions under significant pressure.”
Throughout the storms, HydroForecast’s hydrology team remained in close contact with customers, helping interpret forecast updates and supporting communication with downstream agencies.
Managing multi-wave storm dynamics
The December atmospheric rivers arrived in multiple waves, compounding the complexity of reservoir operations.
Unlike a single storm event, multi-wave systems require operators to consider how successive pulses of precipitation will interact with already saturated ground and evolving reservoir conditions.
“Multi-wave events present unique operational challenges compared to single storms,” Sampson explains. “Operators not only have to consider immediate precipitation, but also how subsequent waves will interact with the imminent one.”
Continuous forecast updates provided the visibility needed to manage this complexity. Rather than reacting to each storm individually, operators could plan for the full sequence of events.
In practice, this enabled more measured responses. “One of our customers told us that having highly accurate forecasts gave them the confidence to make smaller, more gradual gate adjustments rather than large reactive releases,” Sampson notes. “That was better for the entire river system and downstream stakeholders.”
A shift toward predictive water management
The December 2025 storms underscored a broader shift in how water systems must be managed in an era of climate volatility.
Extreme swings between drought and flooding – often described as “weather whiplash” – are becoming more common, challenging traditional approaches that rely heavily on historical precedent.
“The December 2025 series of atmospheric rivers reinforced that you can’t make water management decisions today based solely on what’s happened in the past,” says Sampson. “The combination of a long drought followed immediately by record flooding is exactly the kind of pattern that’s becoming more common.”
For operators, this changing reality is accelerating the adoption of advanced forecasting tools.
“Providing operators advanced forecasting tools enables them to make informed decisions amidst extreme events that have little to no historical precedent,” he adds. “Operators that invested in modernising their forecasting tools prior to the December storms navigated the deluge with more confidence.”
Crucially, that confidence is built over time. “They had built enough familiarity with those tools that they could rely on them when it really mattered,” Sampson says. “That trust doesn’t happen overnight, which is why integrating these tools before a crisis is so important.”
From insight to infrastructure resilience
The December atmospheric rivers demonstrated that forecasting is no longer a supporting capability, it is becoming central to infrastructure resilience.
HydroForecast’s combination of machine learning, global data integration, and real-time expert support reflects a broader evolution toward predictive, data-driven operations in hydropower and water management.
As climate variability increases, the ability to anticipate – not just react to – extreme events will define operational success. For water managers across the Pacific Northwest and beyond, the lesson is clear: foresight is no longer optional. It is foundational.
Alden Keefe Sampson is Upstream Tech’s Chief Technical Officer and a co-founder of the company. After working at several software technology companies including Crashlytics, Twitter, and Drift, Alden co-founded Upstream Tech to apply advanced data analysis to our planet’s most pressing environmental challenges. Awarded Forbes 30 Under 30 in Energy, Alden focuses on large-scale data processing pipelines and machine learning to derive environmental measurements from satellite data and large geospatial datasets. Alden holds a BA in Computer Science from Tufts University.
About HydroForecast
HydroForecast, developed by Upstream Tech, is a data-driven hydrological forecasting platform designed to support operational decision-making in hydropower, water resources, and flood risk management. The system applies machine learning techniques trained on large, globally distributed hydrometeorological datasets to generate streamflow and inflow forecasts across diverse basin types, including those with limited or non-stationary historical records.
The platform integrates multiple data inputs, including numerical weather prediction (NWP) outputs, remote sensing observations, and in-situ measurements where available. These inputs are processed through ensemble-based modelling approaches to produce probabilistic forecasts, providing users with a range of possible hydrological outcomes rather than a single deterministic prediction.
HydroForecast delivers forecasts across multiple temporal scales, from short-term (hours to days) operational guidance to extended outlooks of up to approximately 10 days. Forecasts are updated continuously as new meteorological data becomes available, enabling near real-time situational awareness and iterative operational planning.
Outputs are typically provided at user-defined points of interest, such as reservoir inflows, river reaches, or ungauged catchments, and can be accessed via web-based dashboards or integrated into existing operational workflows via API. This flexibility allows utilities and water managers to incorporate forecast data directly into dispatch planning, reservoir optimisation, and flood mitigation strategies.
In addition to automated model outputs, HydroForecast includes support from a team of hydrologists who assist with forecast interpretation, uncertainty communication, and event-based analysis. This combination of machine learning–based modelling and expert oversight is intended to enhance forecast reliability and usability, particularly during extreme or rapidly evolving hydrological events.
HydroForecast is used by utilities, independent power producers, and public agencies to improve forecast lead times, quantify uncertainty, and support more informed, risk-aware water management decisions
