.png)
Our team is always exploring ways to make our streamflow forecasts even better—including in snow-dominated basins where accuracy can hinge on what’s happening with snowpack and melt.
That’s why we’ve been experimenting with a new approach: Training our models to predict land surface conditions like Snow Water Equivalent (SWE) and soil moisture as auxiliary outputs, or provisional forecast targets. This strategy is intended to help our streamflow model build a richer internal understanding of land surface processes like snow accumulation, melt, and soil saturation.
In our recent experiments, we compared multiple model configurations to our seasonal forecasts. We compared streamflow-only models (also known as discharge, our current baseline) with models that produced additional outputs for SWE, soil moisture, and NDVI alongside discharge.
Here’s what we found: Discharge performance improved or stayed steady across most variants—the best model results were achieved with the combination of streamflow + SWE + soil moisture, where our model outperformed current baseline at 70% of sites.
Based on our results, these recent experiments have helped the model learn better internal hydrologic process representations—not just improving SWE and soil moisture predictions, but also making the streamflow model more physically grounded and explainable.
As Simon Topp, HydroForecast Machine Learning engineer, put it, “We're increasing the model's understanding of the interactions between complex hydrological processes, with the result being a better forecast and more information for end users who are managing critical water resources.”
We’ll continue building on these improvements to make our core seasonal streamflow forecasts even more accurate and reliable, especially in regions where snow dynamics are critical.
We’ll be releasing beta versions of SWE and soil moisture forecasts to select customers as bonus outputs from our seasonal streamflow model in 2026. These forecasts will be available via the dashboard and API, designed to improve the interpretability of streamflow forecasts and provide extra context for planning in snow-driven or drought-prone regions in particular.
Want early access when these bonus forecasts go live? Join the list and we’ll keep you posted when they’re available.
