Interpretations of systematic errors in the NCEP Climate Forecast System at lead times of 2, 4, 8, ..., 256 days

Abstract

The climatology of mean bias errors (relative to 1‐day forecasts) was examined in a 20‐year hindcast set from version 1 of the Climate Forecast System (CFS), for forecast lead times of 2, 4, 8, 16, … 256 days, verifying in different seasons. Results mostly confirm the simple expectation that atmospheric model biases should be evident at short lead (2‐4 days), while soil moisture errors develop over days‐weeks and ocean errors emerge over months. A further simplification is also evident: surface temperature bias patterns have nearly fixed geographical structure, growing with different time scales over land and ocean. The geographical pattern has mostly warm and dry biases over land and cool bias over the oceans, with two main exceptions: (1) deficient stratocumulus clouds cause warm biases in eastern subtropical oceans, and (2) high latitude land is too cold in boreal winter. Further study of the east Pacific cold tongue‐Intertropical Convergence Zone (ITCZ) complex shows a possible interaction between a rapidly‐expressed atmospheric model bias (poleward shift of deep convection beginning at day 2) and slow ocean dynamics (erroneously cold upwelling along the equator in leads > 1 month). Further study of the high latitude land cold bias shows that it is a thermal wind balance aspect of the deep polar vortex, not just a near‐surface temperature error under the wintertime inversion, suggesting that its development time scale of weeks to months may involve long timescale processes in the atmosphere, not necessarily in the land model. Winter zonal wind errors are small in magnitude, but a refractive index map shows that this can cause modest errors in Rossby wave ducting. Finally, as a counterpoint to our initial expectations about error growth, a case of non‐monotonic error growth is shown: velocity potential bias grows with lead on a time scale of weeks, then decays over months. It is hypothesized that compensations between land and ocean errors may cause this behavior.