Identifying alternatives to snow-based streamflow predictions to advance future drought predictability

For large populations across the western U.S., water supply prediction relies centrally on
knowledge of spring snow conditions, where anomalous snowpack can provide critical early
warning of drought. Yet, a warmer future portends for reduced snowpacks, presenting a major
challenge to the current paradigm of snow-based forecasting. The water management landscape
across the west is variable, with some smaller systems relying exclusively on snow information
to make local statistical forecasts, while others use operational forecast information from more
sophisticated statistical or dynamic forecasts that also leverage snow information.
Therefore, a comprehensive evaluation of the expected changes for snow-based prediction
techniques under historical and future climate conditions is essential. To overcome potential
shortcomings from current methods, there is a need for evaluation of possible alternatives to
snow-based predictions to better inform management and planning for key parts of the western
U.S., like the Intermountain West (IMW) and Pacific Northwest (PNW) Drought Early Warning
System (DEWS) regions.

We propose to develop and evaluate new techniques for drought prediction that will suit the
needs of western U.S. water management entities, considering regional characteristics and shifts
to a warmer, less snow-dominated future climate. The proposed analysis includes four elements:
1) An assessment of how changes in snowpack will impact drought predictability under future
climate. We focus on analyzing operational-type statistical forecasts, as well as conducting
new simulations with the state-of-the-art National Water Model (NWM), a possible
companion or successor model to the current River Forecast Center prediction models.
2) An evaluation into whether drought predictive skill can be recovered by including additional
non-snow predictors (e.g. soil moisture, geophysical, and extreme indicator data, etc.) with a
pipeline of machine learning algorithms informed by rigorous feature selection and tested
across alternative predictive model structures.
3) Obtaining direct input from water entities, via an online survey and in-person workshop, to
ensure that our experiments include current techniques and are relevant to decision makers.
Support letters have been received from regional DEWS partners (Colorado Climate Center
in the IMW and CIG in the PNW), and with other partners that work closely on issues of
water supply, the Colorado Basin RFC, the Western Utilities Climate Alliance (WUCA), and
Seattle Public Utilities to provide input on regionally dependent characteristics.
4) An integration of Elements 1-3 to identify the best drought indicators for each region,
considering current and future conditions, and constraints for adaptation by decision makers.

We address MAPP Priority Area C by evaluating drought predictability using current operational
techniques in the western U.S., focusing on the intervening process of snowpack evolution and
precursor mechanisms. We also address Priority Area B by using machine learning to develop
new/improved methods for drought early warning, applying probabilistic ensemble techniques
for two DEWS regional pilots, where a workshop will focus on pathways for future adoption.

Award Announcement: https://cires.colorado.edu/news/team-innovate-new-ways-predict-drought

Climate Risk Area: Water Resources