Geoscience Reference
In-Depth Information
Fig. 2 Correlation between the reconstructions and the actual ECHO-G+MM5 model run for
winter season precipitation. a Analog method, b BARCAST. White stars indicate the location of
the pseudoproxies used to perform the reconstruction. The period used to calculate the correlations
is 1
1998. For BARCAST the reconstruction period was 1
1850, with a calibration interval of
1998. Values inside the area delimited by white contours are statistically signi cant with
p < 0.05. The signi cance level is estimated by Monte-Carlo simulations bootstrapping the
precipitation series
systematic comparison of different reconstruction methods. In a PPE, precipitation
sums simulated by a GCM or RCM are used as a basis to construct arti
cial proxy
data (pseudoproxies) at a few locations by distorting the time series with noise. The
kind of noise is based on the signal-to-noise ratio (SNR) between the proxy time
series and the observational variable under consideration. The typical SNR for real
world proxies is in the range between 0.25 and 0.5 (by standard deviation)
(Smerdon 2012 ). For our speci
c PPE, eleven locations are arbitrary selected as a
set of pseudoproxies. To better mimic real-world proxy data, white noise could be
added on the pseudoproxies at those locations, although in this case we started by
using perfect information (noise free case). Then, the two reconstruction methods
are applied to these pseudoproxy data over Europe. The resulting
fields are then
compared with the simulated
fields (cf. Fig. 2 ).
The analog method uses observations, i.e., instrumental or proxy data, at selected
locations during the target (reconstruction) interval. These data are then compared
to a pool of analog data
here the full ECHO-G+MM5 data. The analogs best
matching the proxy values best are then used as the reconstruction. The target
interval was the period 1500
1990, corresponding to the period reconstructed by
Pauling et al. ( 2006 ). The second method, BARCAST, uses a hierarchy of sto-
chastic models for the climate
field and the proxy response function. It uses
Bayesian inference to estimate the joint distribution of the model parameters and the
targeted climate
field variables (Tingley and Huybers 2010 , 2013 ). Its original
formulation uses an AR(1) process for the climate
field. While this is approximately
true for temperature data, precipitation data had to be transformed to a normal
Search WWH ::

Custom Search