About Rivermatics
River conditions, made legible — for the people whose work depends on water.
Mission
Rivermatics makes river conditions legible for the people whose work depends on water — fishing guides, growers, and the agencies that manage western watersheds. It pulls scattered public data — USGS flows, National Weather Service thresholds, SNOTEL snowpack, weather forecasts — into one clear, honest picture. Rivermatics gives estimates, not guarantees: it exists to sharpen your judgment and point you to official sources, never to replace them.
What it does today
Rivermatics tracks live and historical flows for thousands of USGS stream gauges across the western US — the Missouri headwaters, the Rio Grande, the Colorado, the Great Basin, the Pacific Northwest, and California. Every gauge shows current discharge and stage, a multi-year hydrograph with percentile bands placing today against normal, and flood or low-flow status from NWS thresholds. Rivermatics also runs its own forecasts — for both water levels and water temperature. A deep-learning model — a temporal-fusion transformer trained on each basin's observed flows, snowpack, and weather — issues daily predictions up to two weeks ahead for most of the fleet, drawn on the hydrograph as a median line inside a shaded confidence band. The band is the honest part, showing the range the flow will likely stay within, not a single number to bet on. At gauges with a temperature sensor, a companion model predicts each day's water-temperature high and low, drawn as a curve that follows the river's natural daily rhythm — afternoon peak, pre-dawn trough — with its own confidence band. Weather and snowpack layer onto an interactive map. Three lenses tailor the view — a guide sees floatability windows, agriculture sees irrigation and curtailment signals, an agency sees raw forecast and verification detail — and watchlists keep your key gauges close.
Accuracy, in public
A forecast you can't check is marketing. Rivermatics grades its own models in public: every gauge's forecast is replayed over days the model never trained on and compared against the baselines any forecast must beat — yesterday's reading carried forward, and the calendar's normal for the date. The result is published on the gauge screen itself as a plain-language track record: how much smaller a typical miss is, whether the timing of rises and falls runs true, whether the level and the swing do too. Water temperature is the model's strongest suit — at most scored gauges its daily highs and lows track the river closely enough to make the seasonal normal look blunt. Flow is the harder problem, and the honest answer varies by river — which is exactly why the score is published per gauge instead of averaged into a headline. Each gauge also explains itself: the model reports which signals — snowpack, recent rain, soil moisture, the river's own behavior — drove the line it drew, and that rationale is printed beside the chart.
On the roadmap
Forecasts reach further next: the model extends from two weeks toward a 30-day range as longer weather forcing proves out, and long-range climatology takes over beyond — where point forecasts stop being honest, percentile bands take their place. Forecast coverage deepens within the fleet as younger gauges accrue the observation record the models need. Curtailment alerts will warn water-right holders when diversions are legally restricted, and agencies gain deeper verification and export tools. The throughline holds — clear, sourced, current information you can act on.