Interpretations of Coupled Forecast System systematic errors at leads of 2d, 4d, 8d,...256d

We made a comprehensive examination of the climatology of the NOAA EMC Coupled Forecast System hindcast set, expressed as differences from the day-1 forecast ("errors"). Errors at lead times from 2 days to weeks to months and seasons can be interpreted as involving more and more components: just the atmosphere model initially, then the land surface model, and increasingly the ocean model. This interpretation of time scale as error source works well for a few phenomena: for example, tropical cold tongue-ITCZ complex errors seem to stem from rapidly established atmosphere errors, whose wind signature engages ocean coupling with longer lead times. Land model errors show up as hot surface and low latent heat flux biases in the weeks to a month range of lead times. Oceanic mean bias drifts for months to seasons. Other error growth patterns are harder to interpret, like the northern winter jet stream bias, which grows along with cold Siberian biases in the thermal wind sense, but continues to increase with lead time even out to very long leads. Does this long growth time scale mean it is driven by the ocean's long-growing SST errors? Or might there be long inherent timescales to the winter polar vortex itself, perhaps related to stationary waves driven by the Siberian cold bias, as argued by some authors in relation to autumn snow anomalies in Asia?