Improvements in Forecast Skill of the NCEP Production Suite

The Environmental Modeling Center (EMC) is devoted to implementing, maintaining and improving the numerical forecast systems that make up the NCEP production suite. This paper examines how the forecast skills of many of these prediction systems have improved over the years. It will assess global and regional, deterministic and ensemble forecast systems that predict weather, waves and tropical storms. It will include assessments against the systems' own analyses and against observations. Problems in assessing forecast skill will be considered, such as uncertainties in tropical analyses and in scoring precipitation forecasts. Where possible the paper will seek to establish what type of changes-data assimilation, resolution, numerical methods or model physics—led to specific improvements in skill. Many implementations involve more than one type of change. In addition, model improvements are often needed to exploit fully improvements in data assimilation. If the model physics is incompatible with the initial fields, rapid adjustment in the forecasts to the model's preferred state will occur. Over the last 30 years remarkable progress has been made in predicting mid-latitude circulations. For example current 8 day forecasts of anomalies in Northern Hemisphere 500 hPa height by NCEP's global forecast system are more skillful than 5 day forecasts were 30 years ago. The improvement in model forecasts of 500 hPa height forecasts has been fairly steady over much of that period. In 2010 an EMC study tracked changes in the GFS and improvements in GFS forecast skill over 15 years. They found that model resolution increases benefited the Northern Hemisphere more than the Southern Hemisphere, model changes improved both hemispheres and changes in data assimilation and observations improved both hemispheres but reduced the number of bad forecasts more in the Southern Hemisphere than in the Northern. This study will seek to update that effort. Changes in skill in other fields such as model forecasts of precipitation over the continental United States will also be examined. Object-based verification methods for precipitation forecasts will be used in addition to traditional verification measures; in the future object-based verification may also be applied to locations of jets, lows, troughs, and storms on both global and regional scales.