How we model on seasonal and extended horizons

This article describes our process for creating the HydroForecast analog similarity model for an individual basin.  This model has a native 10-day time step and has been implemented to provide forecasts out 12 months into the future.

Using our neural network approach, we train a “base” model across ~470 sites in North America. The goal of this model is to learn general hydrological principles at a variety of basins. This model is driven by observed historic weather (called “reanalysis”), satellite observations of snow and vegetation, drainage characteristics, and snow products such as NSIDC’s SNODAS.
This base reanalysis model is the starting point for models designed for a specific basin. Using a machine learning technique called transfer learning, we adjust the base model parameters by “tuning” the model to a specific site. This tuned model generally shows improved accuracy across a broad range of metrics that are tracked internally. By this point, the model is very good at understanding the current hydrological conditions of the basin (such as snowpack level) and it knows how future weather translates into future flows.
In the next step we use this tuned reanalysis model to create an analog model. At each forecast issue time, the analog model combines its understanding of the current conditions of the basin with possible future weather patterns which come from 40 years (1982-2021) of historical weather patterns and operational weather forecasts. The combination of these two pieces of information creates 40 forecast traces that simulate realistic future flow scenarios, one for each historical year.

Finally, we select a subset of these flow traces that represent the most likely of the 40 years to materialize over the forecast time range and use these to create the analog similarity model. The mean of these traces is the HydroForecast mean issued in the dashboard and API, and represents our best forecast. 

At this time, similarity is based on NOAA's Global Ensemble Forecasting System (GEFS) forecasts, and additional research is underway to expand on the variables utilized for the similarity selection, e.g. as other climate indices, and sub-seasonal weather forecasts.

Email us at for any questions, thoughts or comments!