Deployment of Enhanced Snowpack Assimilation and Physics on the Cloud to Drive Snowmelt-Driven Seasonal Water Supply Forecasts.

For multiple snow seasons, airborne lidar retrieved snowpack estimates were assimilated into the WRF-Hydro modeling system to improve snowpack representation in high elevation areas for California, Colorado, and Utah. For the past 2024 season, operational assimilation and seasonal runoff predictions were migrated to the cloud for robust forecast delivery to key stakeholders in these regions. In preparation for the 2025 operations season, multiple upgrades were made to the scientific components of the operations pipeline. These upgrades include an updated assimilation component, improvements to land cover physics and parameterizations, and an updated model calibration.

The first set of upgrades were done to the assimilation method to minimize loss of mass during the ablation period. This was done to ensure a continuity of snowpack mass, which helps minimize errors during assimilation increments. Additionally, the assimilation pipeline was upgraded to allow for the assimilation of snow albedo. This is crucial during the ablation period as significant differences between modeled and observed albedos can exist, which lead to errors in modeled ablation.

In addition to upgrades to the assimilation pipeline, improvements were made to the land cover physics. More specifically, upgrades were made to better represent the impact of wildfire burn areas on land cover physics. Both representation spatially in the parameters and improvements to the model physics are needed to accurately represent the localized impacts of burn areas, which includes a modified decay in albedo. Work was done to implement both in a retrospective study and results are presented here.

Finally, a set of updated model calibrations were performed for key forecast locations in California. This is needed as any changes to model physics and parameters dictates an updated set of calibrations be performed. Further, sources of additional observations in a changing climate regime provide an opportunity to improve overall model performance for key forecast regions.

While scientific improvements are critical for model performance from a validation perspective, work was also done to optimize model execution on the cloud. This work is critical moving forward as efficient execution is required to perform model calibration and operational forecasting in an scalable manner. This work provides a unique opportunity to allow for both cutting edge scientific performance, but also a scalable enterprise that can provide operational forecasting that goes beyond legacy forecast systems that are limited to a set of forecast locations or a limited geographic area.