Jökull - 01.01.2011, Blaðsíða 6

P. Crochet and T. Jóhannesson
sea-level temperature at the location of each station
with a vertical lapse rate dT/dz. The estimated sea-
level temperature is then gridded onto a 1-km mesh
using a tension-spline interpolation which is a gener-
alization of the minimum-curvature method in which
large oscillations and extraneous inflection points are
eliminated by adding tension to the elastic-plate flex-
ure equation (Smith and Wessel, 1990). This method
offers a relatively high level of sophistication with-
out requiring the selection and estimation of a covari-
ance function or semi-variogram for each day which
would have required much more computational effort
without necessarily obtaining substantially better re-
sults, especially for periods when the network den-
sity is sparse. Finally, the gridded sea-level tempera-
ture is adjusted to the terrain elevation with the same
vertical lapse rate using a DEM with the same 1-km
grid mesh. No attempt was made here to capture other
types of spatial variations, for example related to dis-
tance to the coast or surface characteristics other than
elevation.
Four different sets of gridded temperature fields
were constructed. Two sets were made using manned
stations (group-1) only by applying i) a spatially and
temporally constant vertical lapse rate and ii) a spa-
tially constant but temporally variable lapse rate. Two
additional sets were calculated by adding the data
from the group-2 automatic stations installed after
1995 using the same two sets of lapse rates, respec-
tively.
The temporally constant lapse rate was some-
what arbitrarily chosen as 6.5 ◦C/km. This lapse
rate is often considered to represent the average ver-
tical temperature gradient in the troposphere (Stone
and Carlson, 1979; Engen-Skaugen, 2007; Li and
Williams, 2008) and it has often been used in glacio-
logical and hydrological studies to adjust temperature
measurements over catchment areas (see for instance
Michlmayr et al., 2008; Hebeler and Purves, 2008).
The monthly variable lapse rate (Table 3) was taken
from Tveito et al. (2000), where this lapse rate was
calculated as one of the components of a multiple lin-
ear regression (MLR) analysis between temperature
and various topographic and geographic factors. This
lapse rate estimate could, however, be influenced by
the presence of other variables in the MLR relation-
ship. The lapse rate of Tveito et al. (2000) displays
a seasonal pattern. It is close to dry-adiabatic in the
winter and spring seasons and close to the standard
value adopted in this study, in summer.
The gridded temperature fields obtained as de-
scribed above are based on a regular 1x1 km grid of
the topography of Iceland. This is a rather high spatial
resolution for many applications but errors due to er-
rors in the assumed topography may arise in some ap-
plications if temperatures for a particular point loca-
tion are interpolated directly from the gridded temper-
atures. In general, it is recommended that temperature
estimates for point locations with a known altitude are
corrected for the difference between the known alti-
tude of the point in question and the interpolated alti-
tude from the 1x1 km grid of Iceland, which for this
reason is included as an integral part of the data set
(using a lapse rate of 6.5 ◦C/km). Estimated station
temperatures used in the validation described below
are calculated in this manner.
VALIDATION
The gridded temperature fields were evaluated over
the period 1995–2010 against independent stations
whose measurements were not used in the spatial in-
terpolation. The first two data sets made with the
manned network only (group-1) were verified against
group-2 stations and group-3 stations. The other two
Table 3. Variable lapse rates in ◦C/km derived from Tveito et al. (2000). – Breytilegt hitafall með hæð í ◦C/km
skv. Tveito o.fl. (2000).
Period Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec
lapse rate 9.7 9.6 8.9 9.7 9.3 7.9 7.1 7.0 7.5 8.0 9.4 9.4
6 JÖKULL No. 61, 2011