A record-low snowpack, then a drought emergency: Tacoma Power runs its dams on satellite forecasts

Tacoma Power spent the 2025-26 water year defending its largest hydroelectric project against two opposite failures, a record inflow surge and a record-poor runoff season, and in both cases one of its inputs was a machine-learning river forecast built partly on NASA satellite data, according to an account published by NASA ^[1][2]. The project stores water behind the Mayfield and Mossyrock dams on the Cowlitz River and generates enough electricity to serve more than 151,000 homes a year ^[3], which makes a reservoir release decision a load decision too.

The product is Upstream Tech's HydroForecast, which combines weather forecasts and river measurements with NASA-produced satellite data on snow cover and vegetation conditions to predict river flow hours to days ahead, and refreshes every two hours ^[4]. That cadence is twelve model runs a day ^[5]. NASA turns observations from the VIIRS instrument on the Suomi-NPP satellite into snow-cover and vegetation-greenness products that cover entire watersheds, including areas where ground monitors are sparse ^[6]. Upstream Tech archives years of those products alongside weather forecast data and measured river flow from hundreds of watersheds, so the models learn common patterns in how water moves through a landscape and apply them at new locations ^[7]. Laura Read of Upstream Tech says tests across multiple basins found that adding snow and vegetation observations increased forecast skill ^[8].

The year gave the system something to work with. Unusual warmth over the winter of 2025-26 pushed a larger share of western precipitation to fall as rain rather than snow, with below-normal precipitation deepening deficits in some areas ^[9]. January, February and March each set the lowest western snow cover for that month in NASA's MODIS satellite record since 2001 ^[10], a record roughly 25 years long ^[11]. In December 2025 an atmospheric river produced one of the largest one-day inflow surges ever recorded at the Tacoma Power project ^[12], and the rain-heavy pattern moved water downstream quickly instead of banking it as snowpack, which held at 20% to 50% of normal ^[13]. On April 8 Washington state placed every watershed, the Cowlitz included, under a drought emergency ^[14]. From April through June, peak daily inflow to the project was among the lowest on record, according to Saul Villarreal, Tacoma Power's senior hydro operations manager ^[15]. Roughly four months separated the flood problem from the drought problem ^[16].

The utility uses HydroForecast alongside stream gauges, snow stations and operator judgment ^[17]. Villarreal says the short-term forecast during the December storm helped the utility anticipate how much water would reach the project and prepare for dynamic river conditions while meeting operating requirements ^[18]. As spring approached, Tacoma Power used the seasonal model to track the rising risk of weak runoff and began holding its reservoirs higher than usual to preserve water for summer, which left less space ^[19] - the standard flood-margin trade that a dry forecast forces.

Two things are worth watching. The first is pipeline dependence: Read says the short-term models run every two hours, so the inputs have to arrive on schedule, and while stopgaps exist, any interruption to the operational pipeline is, in her words, a huge deal ^[20]. A utility that has moved storage and release timing onto a two-hourly satellite-fed model has taken on the availability risk of that model's data supply. The second is whether the higher-than-normal spring storage was enough to carry summer demand ^[19][15], and whether the same playbook gets written down for other basins, since NASA's Erin Urquhart frames the point of the program as making freely available agency data useful to companies serving water managers ^[21].