Diagnosis of systematic errors in atmospheric river forecasts using satellite observations of integrated water vapor

Atmospheric rivers are long, narrow, filamentary structures of water vapor flux in the atmosphere responsible for 90% of the meridional poleward water vapor transport in less than 10% of the earth's circumference. Studies have shown that these atmospheric rivers were present and an important contributor to recent major winter flooding events along the US west coast. Previous work at the NOAA Earth System Research Laboratory developed objective characteristics for the identification of atmospheric river (AR) events in integrated water vapor (IWV) retrievals from the Special Sensor Microwave Imager (SSM/I). These techniques have been extended in the development of an automated AR detection procedure in which potential AR are first identified through thresholding and location of strong gradients in the IWV data, and then further distinguished through the image processing technique of skeletonization and determination of feature width and length. This tool, providing identification of the AR axis, its width, and an estimate of strength based on the IWV magnitude, is applicable both to satellite-derived and numerical weather prediction (NWP) fields of IWV.

Given the important hydrological impact of AR events, understanding whether or not these phenomena are well forecast is of significant interest. We have applied the automated AR detection tool to multiple seasons of observations from the SSM/I and corresponding forecast fields from several of the operational NWP models included in the THORPEX Interactive Grand Global Ensemble (TIGGE) to evaluate and compare the ability of the models to accurately reproduce the frequency, size and intensity of AR events. Results are presented as a function of forecast lead time in terms of quantities including probability of detection and false alarm rate. Overall, the frequency and timing of events is generally well forecast, though the occurrence of landfall tends to be overestimated, particularly at longer forecast lead times.