GFS Water Vapor Forecast Error Evaluated Over The 2009-2010 West Coast Cool Season Using The Met/mode Object Analyses Package

Abstract

Research over the last decade and a half confirms that the vast majority of West Coast cool-season extreme precipitation events are due to the landfall of intense wind-driven streams of concentrated water vapor associated with extratropical cyclones called atmospheric rivers (ARs). Accurate prediction of the effects of ARs as they come ashore depends on accurate numeric modeling of integrated water vapor (IWV) over the Northeast Pacific (NEP). Quantifying the uncertainty in this forecast field is an important step toward understanding the causes of uncertainty in West Coast extreme event forecasts. To this end GFS (Global Forecast System) model output obtained in real time of the fields needed to calculate IWV were archived and analyzed. GFS was used because it is well known, it covers our area of interest, and the output is readily available to the community. To estimate forecast uncertainties we used an object-based method that allows quantitative comparisons of object location, size, shape, and intensity. In particular, we used MODE, the Method for Object-based Diagnostic Evaluation. MODE is an object-based verification tool from the MET (Model Evaluation Tools) package developed and supported by the Developmental Testbed Center (DTC). This package of verification tools is readily available and intended to provide the community with a common software package incorporating the latest advances in forecast verification. We describe results from two studies conducted as part of the Hydrometeorology Testbed (HMT)—DTC collaboration project. The studies are based upon Northeast Pacific (NEP) data collected during the 2009-2010 cool season. In the first study we focus on verifying GFS-analysis IWV against satellite-observed IWV throughout the NEP. Specifically, IWV GFS analysis objects are compared with 12-hour composite, satellite-derived Special Sensor Microwave/Imager (SSM/I) observational objects. Then we incorporate MODE object attributes related to object location, size, shape and intensity into metrics that quantify the degree of agreement between the analyses and the observations. The second study is carried out over a smaller domain partially covering the NEP but local to the United States West Coast to examine the way the uncertainty in forecast IWV object location, shape, and integrated water vapor intensity changes with forecast lead time. We use MODE analysis to compare the 24, 48, 72, and 96 hour GFS forecasts with the GFS analyses. As above, MODE object attributes are used to create metrics allowing estimates of the uncertainty. To date preliminary results with respect to location, based on analysis of the objects centroid (center of gravity), suggest small, if any, locational bias, but significant locational uncertainty. In particular, the interquartile range of the centroid displacement observed over the 2009-2010 cool season was about 70 km for the 24 hour forecasts, increasing to about 200 km for the 96 hour forecasts.