The Skill of Precipitation and Surface Temperature Forecasts by the NMC Global Model during DERF II

Abstract

This study assesses the skill of forecasts of precipitation and surface temperature by the National Meteorological Center's (NMC) global model in the 108 consecutive 30-day forecasts [known as Dynamical Extended Range Forecast II (DERF II)] that were made from initial conditions 24 h apart between 14 December 1986 and 31 March 1987. Since precipitation forecasts are notoriously difficult to verify, model precipitation accumulated during the first 24 h of each 30-day forecast was used as verification. Anomalies were calculated by averaging the precipitation for a given forecast length over all 108 forecasts and subtracting the resulting mean from the precipitation for that forecast length in each individual forecast. A similar procedure was used for surface skin temperature. The skill of the model's forecasts of precipitation and surface temperature anomalies was then assessed, using anomalies from day-1 forecasts as verification. Precipitation forecasts for all regions of the globe exhibit substantially more skill than persistence. Precipitation forecasts for the Northern Hemisphere (NH) extratropics show substantial skill 1.5 days further into the forecasts than forecasts for the tropics and on average exceed the mean skill of 1-day persistence forecasts until forecast day 7. Even the worst individual forecast for the NH extratropics exceeds the mean skill of persistence through day 5. Time-mean precipitation forecasts for the NH extratropics display an anomaly correlation of 0.69 for forecast days 2–5 and 0.53 for forecast days 2–10 when verified against day-1 precipitation anomalies. Surface skin-temperature anomalies are more persistent than precipitation anomalies; forecasts of surface temperature anomalies have considerably higher skill than forecasts of precipitation anomalies. However, forecasts of time-mean surface temperature anomalies for the NH extratropics for forecast days 2–30 and 11–30 exhibit similar levels of skill to forecasts of time-mean precipitation anomalies. This implies that forecast skill for such long forecast periods primarily reflects skill in predicting planetary-scale variations.