Beyond the Point: Closing Idaho’s Snow-Information Gaps through Station Diagnostics and Spatial Downscaling
Mountain snowpack is the dominant source of Idaho’s surface water, yet the snow-information layer that supports water-supply forecasting rests on two structural limitations. SNOTEL site selection was historically driven by practical considerations such as accessibility and long-term protection, and how well individual stations reflect the surrounding water-producing terrain, and in turn basin-scale snow conditions, remains an open, station-specific question. Even where that link is strong, an observation network of fixed points is inherently silent in the space between stations, where much of the water-producing terrain lies. The warm, dry conditions of Water Year 2026 brought both challenges into focus.
This talk presents two complementary strands of ongoing research at Boise State University. The first develops a statewide SNOTEL representativeness and hydrological utility framework. It combines a physiographic index based on the terrain and vegetation surrounding each station, calibrated with airborne LiDAR at Mores Creek Summit, with a complementary measure of how much predictive skill each station contributes to seasonal water-supply forecasts, using the regression framework that underlies operational forecasting.
The second strand develops a physics-informed U-Net to downscale 32-km NARR atmospheric reanalysis and static topographic information to daily 1-km snow water equivalent across Idaho, using SNODAS as the training target. Three physically motivated design choices aim to improve the representation of infrequent snow-drought extremes: signed departures from a leave-one-year-out climatology, an extreme-weighted loss function designed to reduce regression toward average conditions, and a compact predictor set representing multiscale radiative fluxes, mid-tropospheric thermodynamics, and cumulative above-freezing days. Cross-validation across contrasting climate conditions—including Water Year 2026—evaluates the model’s ability to reproduce years excluded from training.
Together, these research strands seek to move Idaho’s snow-information system toward a more diagnostic station network and a spatially complete snow product for water-supply forecasting and climate-change applications.
Irene Cionni is a Senior Research Scholar in the Department of Geosciences at Boise State University. She holds a Ph.D. in Physics from the University of L’Aquila, Italy, and has a background in atmospheric physics and climate science. She has contributed to European research programs focused on Earth-system model evaluation and climate services. At Boise State, she is the principal investigator of a NASA EPSCoR project on deep-learning downscaling of snow water equivalent and an NSF I-CREWS seed grant on compound energy droughts in the Pacific Northwest. Her current work also examines SNOTEL station representativeness and hydrological utility, which she will present in this seminar.