Lead-time independence of gravity-wave forecast skill in operational analysis and forecasts

Atmospheric gravity waves (GWs) are a key driver of vertical energy and momentum transport in the atmosphere, with important implications for large-scale dynamics and chemistry. However, they remain difficult to predict in operational weather and climate models due to their small spatial scales relative to model resolution, and are typically not assimilated into numerical weather prediction (NWP) systems because of the large departures they introduce from model initial conditions.Here we use stratospheric temperature measurements from the Atmospheric Infrared Sounder (AIRS) and the Cross-track Infrared Sounder (CrIS) to evaluate how well archived operational analyses and forecasts from ECMWF’s Integrated Forecast System reproduce observed GW activity over Greenland, a major Northern Hemisphere source region for orographic GWs. The combined AIRS–CrIS sampling at high latitudes provides an unusually high measurement cadence, enabling assessment of forecast performance and time variability at relatively fine temporal resolution.Operational analyses and forecasts with lead times of up to 240 h are sampled at the AIRS and CrIS measurement footprints and regridded to a common resolution to allow consistent spectral analysis. A 2D+1 Stockwell Transform is applied to both synthetic and real observations to characterise GW amplitudes and spatial structure, producing directly comparable GW fields across forecast lead times.Using a Structure–Amplitude–Location (SAL) framework adapted from precipitation forecast verification, we quantify the evolution of GW forecast skill with lead time. We find that model performance exhibits only weak dependence on forecast range: across all lead times, the model systematically produces GWs with smaller horizontal scales and reduced amplitudes relative to observations, while errors in wave location increase only modestly with lead time. This behaviour is unexpected, as shorter lead times are associated with more accurate resolved winds, and would therefore be expected to yield more accurate GW generation. The results suggest that errors in simulated GW characteristics in operational forecasts are dominated by structural and representational limitations rather than by forecast wind errors alone.