Forecast Sources
RiverSat offers multiple river level forecast sources, each with different methodologies, strengths, and coverage. This guide explains what each forecast type is, how it works, and when to use it.
RiverSat
RiverSat is a machine learning-based river forecast developed by EarthDaily Analytics. It is the default forecast source for built-in RiverSat stations.
How It Works
RiverSat trains site-specific ML models on historical weather reanalysis data (ECMWF ERA5-Land) and observed river discharge measurements. The models learn each basin's hydrologic response to weather variables including precipitation, temperature, snow cover and depth, soil moisture, and surface heat flux.
For real-time forecasting, the trained models ingest weather forecast data from two NOAA ensemble systems:
GEFS (Global Ensemble Forecast System) -- 31 ensemble members providing forecasts out to 16 days.
CFS (Climate Forecast System) -- 4 ensemble members extending the forecast horizon beyond 16 days.
The displayed uncertainty bands represent the range of outcomes derived from all combinations of GEFS and CFS ensemble members, giving a probabilistic view of future river levels.
Key Details
Forecast horizon: ~16 days (GEFS), extended further with CFS
Output: River level (stage height) with uncertainty bands
Update frequency: Daily
Coverage: Built-in RiverSat stations with sufficient historical data
When To Use
RiverSat is the recommended default for stations where it is available. Its ML approach captures basin-specific behavior that general-purpose hydrologic models may miss.
NOAA Deterministic
The NOAA deterministic forecast is a single-value river stage prediction produced by the National Weather Service (NWS) River Forecast Centers (RFCs). It is sourced from the National Water Prediction Service (NWPS) stageflow API.
How It Works
NWS hydrologists at 13 regional River Forecast Centers run calibrated hydrologic models for their areas of responsibility. These models ingest observed precipitation and streamflow, radar-based precipitation estimates (MRMS, Stage IV), and numerical weather prediction (NWP) output from models such as GFS, HRRR, RAP, and NAM.
The result is a single "best estimate" forecast for river stage at each gauge location. Forecasters may manually adjust model output based on local knowledge, upstream dam operations, and other factors not captured by automated models.
Key Details
Forecast horizon: Typically 3-7 days, depending on the basin and RFC
Output: A single forecast trace of river stage over time
Update frequency: Multiple times per day as new observations and weather forecasts become available
Coverage: All NOAA river gauge locations with a valid LID (Location Identifier)
When To Use
Use the deterministic forecast when you want a single, official NWS best-estimate of future river levels. This is the forecast most commonly referenced for operational decision-making and flood watches/warnings.
NOAA Ensemble
The NOAA Ensemble forecast comes from the Hydrologic Ensemble Forecast Service (HEFS), a probabilistic forecasting system developed by the NWS. In RiverSat, this view displays the forecast as percentile bands (quantiles).
How It Works
HEFS is a multi-component system that propagates weather uncertainty through hydrologic models. The main processing components are:
The quantile view aggregates all ensemble members into percentile bands. For example, the 10th percentile band means only 10% of ensemble members predict a lower value, while the 90th percentile means 90% predict a lower value. The bands widen over time as forecast uncertainty increases.
Key Details
Forecast horizon: Up to several weeks, depending on the location
Output: Percentile bands (e.g. 10th, 25th, 50th, 75th, 90th) of river stage
Update frequency: Updated with each new HEFS forecast cycle
Coverage: NOAA gauge locations with HEFS data available (subset of all gauges)
When To Use
Use the ensemble quantile view when you want to understand the range of possible outcomes and the probability distribution of future river levels. The width of the bands communicates forecast confidence -- narrow bands indicate higher certainty, wide bands indicate greater uncertainty.
NOAA Ensemble Members
This view shows the individual ensemble traces from HEFS, rather than the aggregated percentile bands. Each line on the chart represents a single forecast scenario.
How It Works
The HEFS system generates approximately 30-40 individual forecast traces. Each trace results from running the hydrologic model with a different plausible weather scenario. In the short range (days 1-14), these scenarios blend current weather model predictions with historical weather patterns. Beyond day 14, forecasts rely primarily on climatological patterns (historical weather sequences applied to current watershed conditions).
Each trace is a complete, self-consistent forecast of river stage over time. Early in the forecast period, traces tend to cluster together (higher agreement). Further out, they diverge as uncertainty grows, visually showing the "cone of uncertainty."
Key Details
Forecast horizon: Same as NOAA Ensemble (up to several weeks)
Output: Individual forecast traces (typically 30-40 lines)
Update frequency: Updated with each new HEFS forecast cycle
Coverage: Same as NOAA Ensemble
When To Use
Use the ensemble members view when you want to see the individual scenarios rather than a statistical summary. This is useful for understanding the diversity of possible outcomes, identifying clusters of similar scenarios, or examining specific worst-case/best-case traces. The quantile view (NOAA Ensemble) is generally easier to interpret for quick assessments.