Geoscience Reference
In-Depth Information
This synthesis shows whether the properties of areas that are monitored are represen-
tative of the unmonitored areas. This question is important, considering that extrapolation
from monitored to unmonitored areas will always be needed to some degree.
3 Results for GCM Projections Across Arctic Basins
Figure
3
illustrates GCM projections and Climatic Research Unit (CRU) observations
for the 14 major Arctic drainage basins (green and blue basins in Fig.
1
; adapted from
Bring and Destouni
2011
). Temperature projections are compatible with observed
deviations during the late twentieth and early twenty-first century, both for the TAR and
AR4, while precipitation projections indicate an increase that is hitherto not evident in
observations. The absolute GCM results compare well with the observations for tem-
perature, but the models generally overestimate precipitation. However, while the dif-
ference between temperature projections and observations has increased between the
TAR and AR4 model ensembles, precipitation projections have come closer to obser-
vations. Generally, GCM projections for temperature and precipitation have become
more precise in the AR4 ensemble, as the models in AR4 converge more closely on the
mean than do the TAR ensemble models.
Results from the error and performance analysis of the AR4 GCMs over the 14 largest
Arctic watersheds indicate that there is a large spread in error and relative performance
between the models used in the AR4 (Fig.
4
). Although the inter-model variation in MAE
is similar for temperature and precipitation, the difference between the systematic over-
estimation of precipitation, evident in the above-zero MBE for all models but one, and the
relatively smaller systematic underestimation of temperature, evident in MBEs closer to
zero, mean that the model performance index is less variable and, for most models, better
for temperature than for precipitation.
For some basins, and for the pan-Arctic in general, an above-average model perfor-
mance in temperature simulation does not correlate particularly strongly with performance
in simulating precipitation (negative rank correlations for MAE and d
r
; Table
2
). This
implies that choosing a ''best'' climate model for the Arctic or any of its major drainage
Fig. 3 Temperature (left) and precipitation (right) values for 1961-1990 and 1991-2002 for observations
and for 1961-1990 and 2010-2039 for GCM projections across 14 major Arctic basins. Error bars for GCM
projections indicate one standard deviation of different GCM results from the model ensemble mean. Error
bars for precipitation observations indicate upper and lower estimates, corresponding to uncorrected and
bias-corrected values, respectively
