College of Engineering | Apply to CSU | Disclaimer | Equal Opportunity Statement | Privacy | Search CSU

Investigation Into a Displacement Bias in Numerical Weather Prediction Models' Forecasts of Mesoscale Convective Systems

You must be on the CSU network—either physically or using VPN—to watch this or any of the videos on this site.

October 26, 2012
Charles Yost
Hosted by Russ Schumacher (advisor), Sue van den Heever, Jorge Ramirez (Civil and Environmental Engineering)


Although often hard to correctly forecast, mesoscale convective systems (MCSs) are responsible for a majority of warm- season, localized extreme rain events. This study investigates displacement errors often observed by forecasters and researchers in the Global Forecast System (GFS) and the North American Mesoscale (NAM) models, in addition to the European Centre for Medium Range Weather Forecasts (ECMWF) and the 4-km convection allowing NSSL-WRF models. Using archived radar data and Stage IV precipitation data from April to August of 2009 to 2011, MCSs were recorded and sorted into unique six-hour intervals. The locations of these MCSs were compared to the associated predicted precipitation field in all models using the Method for Object-Based Diagnostic Evaluation (MODE) tool, produced by the Developmental Testbed Center and verified through manual analysis.

A northward bias exists in the location of the forecasts in all lead times of the GFS, NAM, and ECMWF models. The MODE tool found that 74%, 68%, and 65% of the forecasts were too far to the north of the observed rainfall in the GFS, NAM and ECMWF models respectively. The higher-resolution NSSL-WRF model produced a near neutral location forecast error with 52% of the cases too far to the south. The GFS model consistently moved the MCSs too quickly with 65% of the cases located to the east of the observed MCS. The forecast error from the GFS and NAM were on average 266 km and 249 km, respectively, while the ECMWF and NSSL-WRF produced a much lower average of 179 km and 158 km.

A case study of the Dubuque, IA MCS on 28 July 2011 was analyzed to identify the root cause of this bias. This MCS shattered several rainfall records and required over 50 people to be rescued from mobile home parks from around the area. This devastating MCS, which was a classic Training Line/Adjoining Stratiform archetype, had numerous northward- biased forecasts from all models, which are examined here. As common with this archetype, the MCS was triggered by the low-level jet impinging on a stationary front, with the heaviest precipitation totals in this case centered along the tri-state area of Iowa, Illinois, and Wisconsin. Low-level boundaries were objectively analyzed, using the gradient of equivalent potential temperature, for all forecasts and the NAM analysis. In the six forecasts that forecasted precipitation too far to the north, the predicted stationary front was located too far to the north, and therefore convection was predicted to initiate too far to the north. Subtracting the NAM analysis from these six northern forecasts at all vertical levels shows an area of enhanced potential temperature and stronger southerly winds directly to the south of the northerly forecasted precipitation maximum. This area of enhanced potential temperature and southerly winds is indicative of an incorrectly forecasted northern front. The neutral or good forecasts were devoid of this enhanced area of potential temperature and southerly winds.