Central Arctic weather forecasting: Confronting the ECMWF IFS with observations from the Arctic Ocean 2018 expedition

Forecasts with the European Centre for Medium-Range Weather Forecasts’
numerical weather prediction model are evaluated using an extensive set of
observations from the Arctic Ocean 2018 expedition on the Swedish icebreaker
Oden. The atmospheric model (Cy45r1) is similar to that used for the ERA5
reanalysis (Cy41r2). The evaluation covers 1 month,with the icebreaker moored
to drifting sea ice near the North Pole; a total of 125 forecasts issued four times
per day were used. Standard surface observations and 6-hourly soundings were
assimilated to ensure that the initial model error is small. Model errors can be
divided into two groups. First, variables related to dynamics feature errors that
grow with forecast length; error spread also grows with time. Initial errors are
small, facilitating a robust evaluation of the second group; thermodynamic variables.
These feature fast error growth for 6–12 hr, after which errors saturates;
error spread is roughly constant. Both surface and near-surface air temperatures
are too warm in the model. During the summer both are typically above zero
in spite of the ongoing melt; however, the warm bias increases as the surface
freezes. The warm bias is due to a too warm atmosphere; errors in surface sensible
heat flux transfer additional heat from the atmosphere to the surface. The
lower troposphere temperature error has a distinct vertical structure: a substantial
warm bias in the lowest few 100m and a large cold bias around 1 km; this
structure features a significant diurnal cycle and is tightly coupled to errors in
themodelled clouds. Clouds appear too often and in a too deep layer of the lower
atmosphere; the lowest clouds essentially never break up. The largest error in
cloud presence is aligned with the largest cold bias at around 1 km.