Jökull - 01.01.2011, Side 2

P. Crochet and T. Jóhannesson
is important to estimate the spatial distribution of pre-
cipitation and temperature accurately in order to prop-
erly distinguish between rainfall and snowfall and ad-
equately estimate the snowpack evolution and the tim-
ing and magnitude of snow and glacier melt.
Accurate estimates of meteorological input in-
formation is crucial for robust calibration of hydro-
glaciological models and to avoid the introduction of
noise and bias related to this input information. This is
especially important in the context of climate change
impact studies where models calibrated for the present
climate are used to simulate future water resources
(Bergström et al., 2007; Jóhannesson et al., 2007;
Jónsdóttir, 2008; Einarsson and Jónsson, 2010).
While gridded precipitation fields with high
spatio-temporal resolution have recently been con-
structed for Iceland (Crochet et al., 2007; Jóhannes-
son et al., 2007), similar temperature data sets have
not been available.
A large number of methods of different complex-
ity have been proposed to interpolate climate data and
in particular temperature from sparse observations
(see for instance Bolstad et al., 1998; Gozzini et al.,
2000; Price et al., 2000; Hasenauer et al., 2003; Apay-
din et al., 2004; Chuanyan et al., 2005; Daly, 2006;
Björnsson et al., 2007). Several of these are based on
simple interpolation methods such as inverse-distance
weighting or truncated Gaussian weighting filters.
Others are based on more advanced methods such
as spline-surface fitting and various forms of kriging.
One of the advantages of kriging is the use of a spatial
covariance function or semi-variogram that describes
the spatial variability of the data, but in the context of
daily temperature mapping over several decades, esti-
mating such a function for each day is non-trivial al-
though automatic structural identification can be used.
Methods exist though to minimize the needed compu-
tational effort, based on the calculation of a so-called
climatological semi-variogram (Creutin and Obled,
1982; Lebel et al., 1987). As terrain features are
known to strongly influence temperature variations
(Daly, 2006), direct spatial interpolation in mountain-
ous terrain is problematic except for very high sta-
tion densities. Both kriging and spline-based meth-
ods can take other explanatory variables such as ele-
vation into account. Examples are co-kriging (Phillips
et al., 1992; Pardo-Iguzquiza, 1998), kriging with an
external drift (Hudson and Wackernagel, 1994; Pardo-
Iguzquiza, 1998) and trivariate thin-plate smoothing
splines (Sharples et al., 2005).
Another way to take the effect of elevation on tem-
perature into account is to use the so-called lapse-rate
method. The temperature at a given location is esti-
mated by adjusting measured temperature at a nearby
station given their respective elevation difference and
an appropriate temperature gradient (see for instance
Bolstad et al., 1998). However, factors other than el-
evation can influence spatial temperature variations,
especially in complex terrain. These spatial variations
may be due to orographic effects such as temperature
inversions resulting from cool air drained and trapped
into valley depressions, sharp temperature gradients
between air masses separated by topographic barri-
ers, local orographic effects such as different slope as-
pects leading to a different amounts of incoming solar
radiation, coastal effects leading to temperature con-
trasts between ocean and adjacent land masses, and
land use/landcover variations (Bolstad et al., 1998;
Chuanyan et al., 2005; Daly, 2006). For this rea-
son, multiple linear regression models that formulate
statistical relationships between temperature and local
or regional orographic, geographic and landscape fac-
tors have been proposed and often used for estimating
long-term averaged temperature in combination with
residual interpolation such as detrended kriging, to ac-
count for spatial variations not described by the re-
gression analysis. Such a method was used by Tveito
et al. (2000) and by Björnsson et al. (2007) to esti-
mate the 1961–1990 mean monthly seasonal and an-
nual temperature in Iceland. However, these relation-
ships may be cumbersome to derive for each day and
not necessarily valid or as accurate as for long-term
means. One possible solution for obtaining daily tem-
perature fields in complex terrain is to combine the
use of such method applied on long-term averages and
anomaly interpolation (see for instance the use of the
Aurelhy method in Gozzini et al., 2000).
This paper presents a gridded daily temperature
data set for Iceland with a 1 km resolution and eval-
uates its quality. The study is organized as follows.
2 JÖKULL No. 61, 2011