The NOAA Climate Forecast System (CFS) model data is used in these plots. The CFS Reanalysis (CFSR) ran from 1979 to 2010. In 2011 the model resolution was increased and was renamed as the CFSv2 - as far as I am aware, the model physics are the same. Analysis (and intermediate forecast) fields of the CFSv2 are archived as a continuation of the CFSR. The CFSR is the only reanalysis to include a sea ice model (the GFDL Sea Ice Simulator). See Saha et al., (2010) and Saha et al., (2014) for further details.
The data used here is taken from archives stored on a 0.5 degree Latitude/Longitude grid and are therefore area weighted so as to give a true mean.
Arctic Ocean 925hPa Air Temperature (925hPa Tair)
The value shown in the Figures is the area weighted mean of all non-land points within an arbitrarily chosen region of the Arctic basin that is relevant to sea ice. The map below shows the region over which the mean 925hPa Tair is computed. The 00z, 06z 12z and 18z analysis fields are used for making the daily average temperature. Under conditions of a Standard Atmosphere, 925hPa is at an elevation of about 760m (~2500ft), though in the Arctic 925hPa is closer to ~710m (~2330ft).
925hPa temperature was chosen as it is less influenced by the melting surface in the summer months than 2m temperature (and should be more influenced by data assimilation). Thus 925hPa Tair is more illustrative of heat in the Arctic that might impact sea ice; for example, compare the years 1996 and 2007 using the both 925hPa Tair and the +80N 2m Temperature. (Note that radiation is still the biggest player with respect to ice melt forced by the atmosphere.)
+80N 2m Air Temperature
The 2m Temperature (T2M) is a diagnosed variable (rather than a prognositc variable). T2M is calculated as
a function of the surface radiating temperature, the lowest model atmosphere layer temperature, wind and
roughness length. Over a melting surface T2M will be influenced by the latent heat sink. The +80N Figures show that
in the CFS model, T2m over sea ice areas rarely goes above 2C and there is very little interannual variability
from mid-June to mid-August due to the melt.
In the NOAA model T2M is a purely forecast variable, while in ECMWF it is an analyzed field (meaning that
observations are assimilated into the field). However, over the Arctic Ocean (particularly regions above
80N), observations are sparse (and not all buoy data is used). The T2M analysis at ECMWF is an optimal interpolation with an approximate
e-folding distance of 420km. The ECMWF background and observation errors are set to 1.5K and 2K respectively, meaning
that adjustment is not dramatic and the model diagnosed background field is actually given more weight than
the observations (remembering that the model represents a grid-box area while observations are at points).
In short, for T2M, the ECMWF data will be somewhat (rather marginally) influenced by observed data, for example from the Nth Pole Observatory.
See the ECMWF data assimilation documentation for more details.
The +80N T2M images on this site are similar to those shown by the
Danish Meteorological Institute (DMI)
except that I use the NOAA model (vs. ECMWF) and
area weighting is applied. All points at 80N and higher are used in the average (including land points, as with DMI).
The NOAA model T2M is heavily influenced by the latent heat sink during the melt. NOAA models have previously had noted issues over snow/ice.
Because the data used here is from a Latitude/longitude grid, data should be area weighted so as to give a true mean. A 0.5 x 0.5 degree grid box located at 85N is considerably smaller than a 0.5 x 0.5 grid box at the equator, so this needs to accounted for (as a function of the cosine of the latitude) The data shown on this site are all area weighted.
A simple example when using a 0.5 degree grid: Of the area above 80N, the first five latitude bands (80N-83N) contribute 43% of the area, while the last five latitude bands (87N-90N) contribute less than 5% of the area (see the below Figure). Without area weighting these two regions are considered equivalent within the averaging process.
