Multi-Model Evaluation of Atmospheric River Forecast Skill and Uncertainty over the Himalayas

Atmospheric Rivers (ARs) are long (>2000 km) and narrow (<1000 km) corridors of enhanced moisture transport, with typical water vapour fluxes of 400–500 kg m⁻¹ s⁻¹. When these moisture-laden systems encounter the steep Himalayan terrain, strong orographic uplift produces intense precipitation, supplying much of the seasonal snowpack that sustains regional rivers and water resources. Approximately 56–73% of extreme precipitation events and floods in the Himalayas occur during the presence of ARs, underscoring their critical role in hydrological extremes and downstream water availability for millions of people.

Numerical Weather Prediction (NWP) models play a crucial role in forecasting extreme weather systems, and evaluating their performance over the complex terrain of the Himalayas is a vital first step toward improving regional predictability. In this study, we assessed the capability of multiple NWP models, including ECMWF, IMD, NCEP, and NCMRWF, to detect and forecast ARs at various lead times. ARs were identified using the tARget algorithm based on Integrated Vapour Transport (IVT) thresholds. Our analysis shows that the Hit Rate varies between 0.3 and 0.6 across models and lead times, while the False Alarm Rate ranges from 0.03 to 0.09, indicating considerable uncertainty in AR prediction. The ECMWF generally performs better at short lead times, capturing a larger fraction of observed AR events, whereas the NCEP model exhibits comparatively better skill at longer lead times, extending beyond 10 days. For all models, forecast skill consistently decreases with increasing lead time, reflecting the growing uncertainty associated with longer-range predictions. The relatively low hit rate of the IMD model can be largely attributed to its tendency to overestimate IVT over the Indian subcontinent. This positive bias leads to an exaggerated frequency of AR detections, thereby inflating false alarms and reducing the overall reliability of the forecasts.

Beyond event detection, substantial discrepancies are also found in AR characteristics, including their intensity, spatial extent, geographical position, and orientation. These differences highlight limitations in how current NWP models represent moisture transport and orographic interactions over the Himalayas. Consequently, further improvements in physical processes, parameterizations, and model resolution are required to achieve more accurate and reliable AR forecasts for this highly complex and hydrologically sensitive region.