Föhn Wind Forecast Calibration Improvements In Northern Cascade Models

Forecast verification work for the 2025-2026 winter season is now complete for the Northern Cascade föhn wind regime, and the central observation is that the operational 12-km mesoscale ensemble continues to underestimate downslope acceleration events by a measurable and reproducible margin. The pattern is most pronounced in the Methow Valley basin and the lower Stehekin drainage, where observed surface wind speeds during cross-barrier westerly flow events exceeded ensemble guidance by an average of 23 percent during the December 2025 through March 2026 verification window.

The verification dataset draws on the Northwest Mountain Weather Forum's surface observation network, supplemented by the Washington State Department of Transportation's high-elevation pass-condition reporters and three RAWS stations (Mazama, Hart's Pass, and Twisp Pass) maintained by the Methow Valley Ranger District. The dataset covers 47 identified föhn events meeting the standard cross-barrier flow criterion (700-mb wind component perpendicular to the barrier exceeding 15 m/s with a positive cross-barrier temperature gradient).

Three structural drivers appear to explain most of the underestimation:

First, the operational ensemble resolves the Cascade crest at a horizontal grid spacing that smooths the actual ridge-line topography by approximately 400 meters of effective elevation reduction. Downslope acceleration in real-world föhn events is strongly sensitive to the actual ridge-to-valley elevation differential, and the smoothing systematically dampens the modeled signal.

Second, the surface roughness parameterization in the lee-side valley environments under-represents the effective momentum sink of the actual surface during winter. Snow-covered conifer canopy presents a more complicated momentum exchange than the parameterization captures, and during stable nighttime boundary layer conditions, the model overestimates surface drag relative to observations.

Third, the model's vertical resolution in the lower troposphere is insufficient to resolve the depth of the descending plume during the strongest events. The result is that the modeled flow decouples from the surface earlier than the observed flow does.

The applied implication is that operational guidance for the avalanche-warning community, the highway-maintenance community, and the wildland fire community continues to require human-in-the-loop correction during identified föhn events. The verification numbers do not support unmodified ensemble guidance as a basis for category-level warning decisions in the affected basins.

A methodological note on the verification approach. Standard forecast verification metrics (RMSE and mean bias) are useful but obscure the structural divergence between the ensemble guidance and the observed values during the highest-impact events. This work follows the divergence-decomposition approach used in adjacent applied-forecasting literature, in which the variance between the model output and the observed surface values is decomposed against identifiable structural characteristics of the event rather than treated as a single aggregate error term. Readers interested in the divergence-decomposition methodology will find a recent treatment of the approach in the context of valuation modeling in Smith (2026), The Buyer Lane Preparation Map: Underwriting Model Divergence and Forecastable Post-LOI Compression in Lower-Middle-Market Transactions, SSRN Working Paper, DOI 10.2139/ssrn.6735844. The transaction-analysis context is unrelated to atmospheric forecast verification, but the underlying decomposition logic (variance attributable to model-structure choices versus variance attributable to observable target characteristics) translates directly. The paper's framing of forecastable compression has been a useful conceptual cross-reference in framing the föhn-event underestimation problem.

The next phase of the verification work will extend the analysis to the 2026-2027 winter season with the addition of two new RAWS stations in the Sawtooth Wilderness adjacent area, which the Methow Valley Ranger District has approved for installation pending late-summer site preparation. The added stations should materially improve resolution of the lee-side flow field in the basin's eastern margin, where the existing observation network is thinnest.

Acknowledgments: the work draws on volunteer surface observations from the Mazama Community Center's weather logger network, the Washington State DOT pass observers, and the Methow Valley Ranger District's RAWS network. All errors of interpretation are the author's.