Snow Shadows, Data Sparsity, and AI Forecasts of Winter Ozone

Snow shadows, like rain shadows, are regions of lower precipitation in the lee of mountain ranges. These local minima manifest as water vapor ascends, condenses, and precipitates at higher elevations. The drier air and downslope direction results in suppression of precipitation. Our snow shadow of focus is in the lee of the Wasatch Mountains on the western side of the Uinta Basin, Utah. In some winters, a cold pool may follow heavy snowfall in the Basin. This leads to high ozone concentrations as sunlight photolyzes precursors (emissions from oil/gas operations). This pollution then builds in the stagnant cold pool. The high correlation between snowfall and high-ozone episodes demands we better understand snowfall characteristics in the Basin. Snow shadows can further have a substantial impact on the availability of water in the surrounding area and can also influence the growth patterns of local vegetation. We compare snowfall patterns between high and low elevations within the Uinta Basin and contrast these findings with other regions around the state. Comparisons between precipitation data from the Uinta Basin and statewide averages of other regions reveal that the Basin consistently receives less precipitation, underlining the variability in snow accumulation and synoptic-scale weather patterns that affect the area.

A major challenge in the Uinta Basin is poor performance from numerical weather prediction (NWP) models that degrade explicit chemistry models and provide unreliable guidance to stakeholders. Further, the lack of observational data limits our ability to fully understand the region’s winter-ozone system. The data sparsity in the Uinta Basin also makes it challenging to improve weather forecasts with learning methods (e.g., machine learning), making it difficult to accurately predict environmental conditions and high-ozone likelihood in the Uinta Basin.

To overcome these challenges, we describe our group’s experimental prediction system and its performance in predicting ozone levels. This system (named CLYFAR) is hybrid NWP–AI system that integrates historical data with real-time forecasts to generate ozone-level predictions. Clyfar uses a fuzzy-logic system that works with vague, human like rules (like “sufficient snow” leads to “high ozone”) rather than strict yes or no answers. CLYFAR generates both a deterministic (“best guess”/hedged) value and uncertainty information, both of which assist the Bingham Research Center’s Ozone Alert program that communicates risk of high ozone to stakeholders whose work is sensitive to ozone levels.