Battarbee R.W., Binney H.A. (Eds.) Natural Climate Variability and Global Warming: A Holocene Perspective

Подождите немного. Документ загружается.

170

Hugues Goosse et al.

associated with internal variability, the maximum temperatures occurred at differ-

ent locations at different times.

Even if perfect models and perfect forcing time-series were available, each model

experiment would simulate a possible evolution of the internal variability but not

necessarily the one followed in the real world. Very precise initial conditions would

only allow having an agreement between model and observations during a few

weeks at most, a classic problem in weather forecasting (Lorenz 1963). In complex

models (such as AOGCMs and some EMICs) that include a representation of the

internal variability, it is possible to compare the statistics of one simulation with

the observations or to try to extract the response to a forcing in the simulation and

in available reconstructions. A precise comparison between temperature changes

simulated in a model with a reconstruction in a particular region, for a particular

year (or decade), is usually useless because it is generally impossible to state if any

difference is due to uncertainties in the reconstruction, model, or forcing deﬁcien-

cies, or simply to a different realization associated with the internal variability of

the system.

In this framework, ensembles of simulations are particularly useful. To per-

form such an ensemble, slightly different initial conditions are selected. After a

few simulated days, the internal variability evolves in a totally different way in

the various members. If the ensemble is large enough, a reasonable sampling of the

model variability could be obtained. In order to have an agreement between

model and a reconstruction, the reconstruction must correspond to a reasonable

member of the ensemble (Figure 7.4). As the members of the ensemble provide

independent samples of the internal variability, averaging over all the members of

the ensemble tends to ﬁlter out the internal variability, leaving only the forced

response of the system. This provides a clear advantage compared with individual

simulations of complex models that always provide a mixture between the forced

response and the internal variability.

1100

1700

1400 150013001200 1600 19001800

Time (years)

1.8

–1.2

–0.9

–0.6

0.0

1.2

0.6

–0.3

0.9

1.5

0.3

Temperature (K)

Figure 7.4 Anomaly of

summer mean temperature in

Fennoscandia averaged over

25 simulations that differ only

in their initial conditions

(black). The mean over the

ensemble of 25 simulations

plus and minus two standard

deviations of the ensemble are

in grey. The reconstruction of

Briffa et al. (1992) is in green.

As the reconstruction is

within the range of the

various simulations

performed with the model,

we can consider that there is

no disagreement between

model results and the

reconstruction. The reference

period is 1500–1850. The

time series are grouped in

10-year averages.

9781405159050_4_007.qxd 6/3/08 4:44 PM Page 170

Climate of the past millennium

171

Using model-derived pseudo-proxies to test

reconstruction methods

As discussed above there are primarily two distinct methods that have been used

in proxy-based climate reconstruction: the CPS method, and the CFR method.

Experiments using synthetic proxies or “pseudo-proxies” derived from climate

model simulations have been used to test the performance of both the CPS (Mann

et al. 2005b) and CFR methods (Mann and Rutherford 2002; Rutherford et al.

2003; Zorita et al. 2003; von Storch et al. 2004; Mann et al. 2005b; Wahl et al. 2006).

Since our primary interest here is on regional and spatial patterns of change, we

will focus on the CFR approach, which provides spatial patterns as well as hemi-

spheric mean reconstructions.

Most of the studies described above have found that CFR-based reconstruc-

tions for past centuries are likely to be skilful, given the spatial distributions

and estimated signal-to-noise characteristics of available proxy records. Von

Storch et al. (2004) argue that proxy-based CFR reconstructions are likely to

underestimate low-frequency variability, based on experiments using pseudo-

proxy networks derived from a millennial simulation of the GKSS ECHO-G

coupled model. Wahl et al. (2006) have challenged the von Storch et al. (2004)

calculations, arguing that they suffer from an artefact of an originally undis-

closed procedure in which data were detrended by von Storch et al. (2004) prior

to calibration. Wahl et al. (2006) note that such a procedure a priori removes

the primary pattern of low-frequency variability from the surface temperature

data.

Von Storch et al. (2004) considered only the more primitive EOF-based CFR

method of Mann et al. (1998), and not the RegEM CFR approach used in recent

work by Mann and collaborators (e.g. Mann and Rutherford 2002; Rutherford

et al. 2003, 2005). Mann et al. (2005b) tested the “RegEM” CFR method using a

simulation of the climate of the past millennium using the NCAR CSM 1.4 coupled

ocean–atmosphere model driven by estimated long-term natural and anthro-

pogenic radiative forcing histories. The simulated Northern Hemisphere mean

surface temperature history indicates modestly greater variability than most

other simulations (cf. Figure 7.3), providing a challenging test for CFR methods.

“Pseudo-proxy” data were constructed from the model surface temperature ﬁeld

to have signal-to-noise attributes similar to actual proxy data networks used in

reconstructing past climates (e.g. as in Mann et al. 1998). Application of the

RegEM method to the calibration of the surface temperature ﬁeld against the

pseudo-proxy networks produced skilful reconstructions of the actual surface

temperature history, underscoring that CFR methods are likely to yield reliable

reconstructions of past variations given realistic assumptions regarding the statis-

tical attributes of actual proxy data networks.

Zorita et al. (2007) have, in turn, argued that the skilful results demon-

strated by Mann et al. (2005b) may represent an artefact of the use of a long

9781405159050_4_007.qxd 6/3/08 4:44 PM Page 171

172

Hugues Goosse et al.

combined 19th–20th century (1856–1980) rather than a short, 20th century only

(1900–1980) calibration interval. They also suggest that the skilful results demon-

strated by Mann et al. (2005b) might be peculiar to the speciﬁc (NCAR CSM 1.4)

model simulation used and not possible in other simulations such as the GKSS

ECHO-G simulation used in their analyses. Finally, Zorita et al. (2007) and von

Storch et al. (2006) argue that the results presented by Mann et al. (2005b) may

represent an artefact of an inappropriate model for the proxy noise, and that

poorer results would be achieved if proxy noise was assumed to be spectrally “red”

rather than, as assumed in previous studies such as Mann et al. 2005b and von

Storch et al. (2004), “white”.

We investigate these issues based on more recent experiments by Mann et al.

(2007a,b) wherein the RegEM approach was applied to surface temperature

reconstructions using pseudo-proxy networks diagnosed from two different

model simulations: the NCAR CSM 1.4 coupled simulation used by Mann et al.

2005b, and the GKSS ECHO-G “ERIK” simulation used by von Storch et al.

(2004) and Zorita et al. (2007). The pseudo-proxy networks, as in Mann et al.

2005b, von Storch et al. (2004), and Zorita et al. (2007), have the spatial dis-

tribution of the full Mann et al. (1998) proxy network. Reconstructions were

performed based on the “short” (1900–1980) calibration interval, and a lower

signal-to-noise ratio (“SNR”) than used by Zorita et al. (2007; the proxy signal-

to-noise amplitude ratio was 0.4 whereas Zorita et al. (2007) used SNR = 0.5). The

proxy noise was speciﬁed to be spectrally “red” as in Zorita et al. (2007), using the

average noise autocorrelation coefﬁcient ρ=0.32 estimated from the actual Mann

et al. (1998) network (see Mann et al. (2007a) for details). The resulting RegEM

reconstructions are observed to track closely the actual model temperature his-

tories for both the NCAR and GKSS simulations, and the Northern Hemisphere

mean reconstructions lie entirely within the self-consistently estimated uncertain-

ties of the true respective model histories (Figure 7.5). The most pronounced

feature in the simulations – the cold temperatures of the 15th–19th centuries asso-

ciated with anomalous negative radiative forcing by a combination of solar irradi-

ance reduction and active explosive volcanic aerosol forcing – is well captured in

both simulations, even with the short calibration interval and “red” proxy noise

employed.

As discussed earlier the anomalous initial warmth and much of the subsequent

long-term cooling trend in the GKSS “ERIK” simulation is an artefact of the model

initialization procedures used. Mann et al. 2005b speculated that the unphysical

drift resulting from this initialization might degrade CFR performance in tests

using the “ERIK” simulation, since the drift pattern might not be captured over

the modern training intervals used for calibration. The above test using the ERIK

simulation (i.e. Figure 7.5b), however, demonstrates that this drift does not in fact

pose an obstacle for the RegEM CFR method. In conclusion, the RegEM CFR

approach should provide reliable reconstructions of past surface temperature

histories over the past millennium within estimated uncertainties, given available

proxy data networks.

9781405159050_4_007.qxd 6/3/08 4:44 PM Page 172

Climate of the past millennium

173

Using proxy reconstructions to constrain model

parameters and detect the role of the various forcings

The time evolution during the past millennium of the annual mean tempera-

ture averaged over the Northern Hemisphere generally closely follows variations in

the external forcing. This is probably the main reason for the success of relatively

–1.2

–1.0

–0.8

–0.6

–0.4

–0.2

0.0

0.2

0.4

0.6

Year

Actual NH mean

Reconstructed (1900–1980 calibration,

SNR = 0.4, ρ = 0.32) with uncertainties

Actual NH mean

Reconstructed (1900–1980 calibration,

SNR = 0.4, ρ = 0.32) with uncertainties

20001100 1200 1300 1400 1500 17001600 1800 1900900 1000

Year

20001100 1200 1300 1400 1500 17001600 1800 19001000

1.5

–2.0

–1.5

–1.0

–0.5

0.0

0.5

1.0

Temperature anomaly (°C)

(a)

(b)

Figure 7.5 Reconstruction of

Northern Hemisphere (NH)

mean temperature based on

RegEM CFR reconstructions

using “pseudo-proxy”

networks taken from

(a) NCAR CSM 1.4 and

(b) GKSS ECHO-G “ERIK”

simulations. In both cases, the

pseudo-proxy network

locations correspond to the

104 unique locations used by

Mann et al. (1998), a proxy

signal-to-noise ratio SNR =

0.4, red proxy noise with noise

autocorrelation ρ=0.32, and

a 1900–1980 calibration

interval is used. Self-

consistent uncertainties in the

reconstructions are estimated

from the unresolved residual

variance during an

1856–1899 “validation”

interval, and are indicated by

shading (95% uncertainty

region). Actual model NH

series is shown for

comparison (black). All series

are decadally smoothed.

(From Mann et al. 2007a,

©American Meteorological

Society.)

9781405159050_4_007.qxd 6/3/08 4:44 PM Page 173

174

Hugues Goosse et al.

simple models in simulating this variable. At this spatial scale, the response of the

climate system to the total forcing is, in good approximation, equal to the sum of

the responses to all the individual forcings. It is thus possible to disentangle the role

of the various forcings by performing simulations in which only one forcing is

applied at a time (e.g. Crowley 2000; Bertrand et al. 2002; Bauer et al. 2003; see also

Figure 7.6).

Such simulations have allowed that, as expected, a higher than average solar

irradiance is associated with higher temperature and vice versa. As a consequence,

the solar forcing alone, as reconstructed in Figure 7.2, induces similarly high tem-

peratures during the 12th, 13th, and 20th centuries, whereas lower temperatures

are simulated during the 16th and the period covering the late 18th to early 19th

centuries. The large volcanic forcing is responsible for the simulated cold periods

after large volcanic eruptions, in particular those of 1258, 1452, 1600, 1641, and

1815. The land–use changes are associated with a long-term cooling trend and

the simulated cooling due to aerosols is mainly restricted to the 20th century.

Greenhouse gas forcing mainly induces a large warming during the past 150 years,

and if the forcing is excluded it is not possible to simulate the large temperature

increase observed during the 20th century.

The temperature response averaged over the Southern Hemisphere is more

complex than in the Northern Hemisphere. Between 50°S and 70°S, the Earth’s

surface is nearly exclusively covered by oceans, implying that the effective heat

capacity at these latitudes is much larger than in the Northern Hemisphere, induc-

ing a damped and delayed response to the forcing and thus a less clear imprint of

the volcanic signal than in the Northern Hemisphere (Figure 7.7). Furthermore,

the Southern Ocean experiences large-scale upwelling of relatively old water

masses that acquire their characteristics decades to centuries before they reach the

surface in the Southern Ocean. As a consequence, the surface temperature at one

particular time in the Southern Ocean is not only inﬂuenced by the radiative forc-

ing at that time but also by the previous history of forcing changes, which has left

its imprint on the characteristics of the older water masses that upwell there. This

Time (years)

20001100 1200 1300 1400 1500 17001600 1800 1900

0.6

–0.1

0.0

0.1

0.2

0.3

0.4

0.5

Temperature (K)

–0.3

–0.2

Greenhouse gas

Aerosols

Volcanos

Solar irradiance

Land-use changes

Figure 7.6 Annual mean

temperature anomaly

averaged over the Northern

Hemisphere in simulations

performed with

ECBILT–CLIO–VECODE

using only one forcing at a

time: greenhouse gas forcing

(red), aerosol forcing (green),

land-use changes (dark blue),

volcanic forcing (orange), and

solar forcing (light blue). The

curves shown are the average

over an ensemble of 10

simulations performed with

the model using one of the

forcings described in Figure

7.2. A 25-year running mean

has been applied to all the

time series. The reference

period is 1500–1850.

9781405159050_4_007.qxd 6/3/08 4:44 PM Page 174

Climate of the past millennium

175

results in a much more complex link between the forcing and the response than in

the Northern Hemisphere. For instance, this has led to the hypothesis that the

Medieval Warm Period was delayed in the Southern Ocean by 50 to 200 years com-

pared with the Northern Hemisphere (Goosse et al. 2004).

The previous discussion is mainly based on model results. The data are only

used to show that the models are providing results in agreement with the recon-

structions if all the forcings are included and that the conclusions deduced from

model simulations appear reasonable. The results could thus be model-dependent

and strongly inﬂuenced by uncertainties in the reconstruction of one particular

forcing. It is possible, however, to combine directly the model results and observa-

tions in order to detect the inﬂuence of various forcings on Northern Hemispheric

temperature during the past millennium (Hegerl et al. 2003), using techniques

similar to the ones applied for the 20th century, and to distinguish the anthro-

pogenic and natural forcings (e.g. Tett et al. 1999; Stott et al. 2001). The ﬁrst step is

to obtain, from model simulations, the signal associated with a forcing, i.e. the time

evolution of the temperature change when only one forcing is applied. This is

usually called the “ﬁngerprint” of the forcing. A multiple regression is then used

to obtain the linear combination of those ﬁngerprints that provide the best agree-

ment with the reconstructions. In other words, the model provides the shape of the

response to a forcing whereas it is the ﬁtting of the linear combination of the

ﬁngerprints with observations that gives the amplitude of the response to this forc-

ing. The conclusions of such analyses are relatively insensitive to the magnitude of

the response of the model to a particular forcing or to uncertainties in the ampli-

tude of the forcing. Indeed, an error in the amplitude of the forcing of a factor of

two would simply induce a change by a factor of two in the regression coefﬁcient, if

the response of the system could be considered as linear (e.g. Tett et al. 1999).

Using an EBM to obtain the ﬁngerprints of the externally forced signals and

various reconstructions of Northern Hemisphere mean temperature, Hegerl et al.

(2003) were able to detect clearly the inﬂuence of volcanic forcing on temperature

changes during the past millennium. Furthermore, in their model, the time-scale

1100 1200 1300 1400 1500 17001600 1800 1900

Time (years)

5.0

1.0

1.5

2.0

2.5

3.0

3.5

4.5

Normalized temperature

0.0

0.5

–2.0

–1.5

–1.0

–0.5

4.0

Figure 7.7 Annual mean

temperature anomaly

averaged over the Northern

Hemisphere (red) and over

the latitude band 50–70°S in

simulations performed with

ECBILT–CLIO–VECODE

driven by natural and

anthropogenic forcings. The

time series are grouped by 25-

year average and divided by

their standard deviation to

facilitate the comparison. The

reference period is

1500–1850.

9781405159050_4_007.qxd 6/3/08 4:44 PM Page 175

176

Hugues Goosse et al.

of the response is similar to the one deduced from the reconstructions. The

inﬂuence of anthropogenic forcings during the late 20th century is also detected,

in agreement with studies focusing on the 20th century. Solar forcing has been

marginally detected only over some periods and in some records, thus they con-

clude that the role of this forcing has probably been weak for the multi-decadal

variability of the temperature averaged over the Northern Hemisphere.

It is possible to combine model results and data to constrain the magnitude of

the changes during the past millennium as well as the value of some important

characteristics of the climate system such as the climate sensitivity. Using a coupled

physical–biogeochemical climate model, driven by different reconstructions of

past changes in solar irradiance and volcanic forcing, Gerber et al. (2003) com-

pared the simulated CO

and temperature evolution with available observations.

In their analysis, for large changes in solar irradiance and hemispheric mean tem-

perature, they ﬁnd large discrepancies between observed CO

concentration and

that simulated by the model. They argue that, between 1100 and 1700, changes in

multi-decadal Northern Hemispheric mean surface temperature larger than 1°C

are inconsistent with CO

records and thus not realistic. Following a similar objec-

tive to Gerber et al. (2003), Hegerl et al. (2006) recently performed a large ensemble

of simulations with an EBM over the past millennium varying two important

model parameters, namely the equilibrium climate sensitivity to a doubling of CO

and the effective ocean diffusivity, and using different scaling of the reconstruc-

tions of past radiative forcing. By comparing the results of this ensemble with

several reconstructions of surface temperature in the Northern Hemisphere at

(nearly) hemispheric scale, using a technique based on a Bayesian approach, they

derive a probability density function (PDF) for climate sensitivity. Such methods,

however, could be strongly dependent on prior assumptions made in the process

of the PDF estimation (e.g. Frame et al. 2005) and additional work is still required

before being able to constrain efﬁciently the model parameters using all the avail-

able information from the data.

Analyzing regional changes using observations

and models

At regional scales, the temperature and precipitation anomalies are strongly

inﬂuenced by the dominant mode of variability of the climate system, such as the

North Atlantic Oscillation (NAO) or El Niño–Southern Oscillation (ENSO). Any

change in these modes of variability, for instance, a shift from a positive to a nega-

tive phase, could be purely related to internal dynamics or caused by changes in the

forcing. The analysis of proxy data could allow determination of the magnitude of

the changes and the potential correlation with the reconstructions of past forcings.

Alternatively, if a model is able to simulate reasonably well the observed changes,

this could provide a reasonable hypothesis for the mechanism responsible for the

change. Despite the complementary information from model and data, determining

9781405159050_4_007.qxd 6/3/08 4:44 PM Page 176

Climate of the past millennium

177

the cause of a change in one mode of variability or in the probability to have one

phase of a mode is nearly impossible, currently, as internal variability alone could

explain the observed changes in those modes.

The observed substantial cooling in large parts of Europe during the late 17th

and early 18th centuries appears to be related to long-term variations in the NAO

pattern of climate variability (Figure 7.8), which are associated with variations over

time in the pattern of the Northern Hemisphere jet stream. In model simulations,

these changes are associated with a large-scale dynamical response of the climate

system to natural radiative forcing by explosive volcanic activity (Robock 2000;

Schmidt et al. 2004; Shindell et al. 2004) and solar output (Shindell et al. 2001;

Schmidt et al. 2004) that interacts with the NAO pattern of atmospheric circula-

tion. This has led to the suggestion that the moderate apparent lowering of solar

irradiance during the 17th century leads to only moderate decreases in hemi-

spheric mean temperature, but a tendency for strong annual mean cooling in some

regions, such as Europe.

Recent work suggests that the El Niño–Southern Oscillation may also be an

important component of the response of the climate to forcing over the past

1000 years. Model simulations using simple models of the coupled tropical

ocean–atmosphere to study the response of ENSO to solar and volcanic forcing

(Mann et al. 2005a) indicate a counterintuitive tendency towards El Niño (warm

Proxy data reconstruction, 1660–80 vs 1770–90, ANN surface temperature

GCM simulation, 1680 vs 1780 solar + volcano, ANN surface temperature

–1.5 0.6 0.80.40.20.0–0.2–0.4–0.6–0.8 1.0

Figure 7.8 Reconstructed

(top) and simulated (bottom:

GCM, general circulation

model) annual (ANN)

average surface temperature

differences between 1660–80

and 1770–90. The

reconstructed surface

temperatures are based on a

multi-proxy estimate using

tree rings, ice cores, corals,

and historical data. Model

results are based on the sum

of the response in two

simulations: one

incorporating reconstructed

solar irradiance changes

during this period and one

using volcanic forcing scaled

to changes over this time.

(From Schmidt et al. 2004.

9781405159050_4_007.qxd 6/3/08 4:44 PM Page 177

178

Hugues Goosse et al.

eastern and central tropical Paciﬁc) conditions in response to negative radiative

forcing (past explosive tropical volcanic eruptions or decreases in solar irradiance)

and a tendency for La Niña-like conditions in response to positive radiative forcing

(i.e., increases in solar irradiance). This prediction matches available evidence

from tropical Paciﬁc coral records (Cobb et al. 2003). The response of ENSO to the

combined effects of solar and volcanic forcing in past centuries provides a poten-

tial explanation for why the tropical Paciﬁc appears to have been in a cold La Niña-

like state during the so-called “Medieval Warm Period” and a warm El Niño-like

state during the “Little Ice Age” (Adams et al. 2003).

Changes in oceanic circulation could also have an impact on regional changes

during the past millennium. Unfortunately, there are not enough high-frequency

observations in the ocean to derive a clear and comprehensive view of oceanic

changes during this period. The available information is mainly derived from some

local observations and model results. A thorough model–data comparison has not

been performed yet but that could certainly be very useful and would help to ﬁll

some of the gaps in our knowledge. In particular, when driven by a reduction of the

solar irradiance (or more generally in case of a moderate cooling), models tend to

simulate an intensiﬁcation of the meridional overturning circulation in the

Atlantic (Cubasch et al. 1997; Goosse and Renssen 2004; Weber et al. 2004). This is

due to an increase in the density of the surface water in the North Atlantic, caused

by cooling and by changes in the freshwater budget, inducing more vigorous deep

water formation there. The intensiﬁcation of the oceanic circulation implies a

larger heat transport towards the high latitudes and is thus associated with a nega-

tive feedback in those regions of the initial cooling. For a moderate warming, the

opposite mechanism is noticed, presenting some similarities with what is expected

for the 21st century (e.g. Gregory et al. 2005). A reduction of the solar forcing

might also modify the probability of having years with signiﬁcantly reduced inﬂow

of warm Atlantic waters at high latitudes (Goosse and Renssen 2004). This leads to

a higher probability of having very cold conditions in the Nordic Seas and over

Scandinavia, resulting overall in a very strong positive feedback in the northern

North Atlantic and surrounding regions during those cold periods. There is also

evidence for the potentially important response of the Atlantic meridional over-

turning to low-frequency changes in the NAO (Delworth and Dixon 2000), which

have in turn been simulated as a possible response to long-term solar and volcanic

forcing in past centuries (Shindell et al. 2001, 2004; Schmidt et al. 2004).

Assimilation of observations

Classically, the information deduced from the analysis of observations and from

model simulations is obtained from two different, clearly separated streams that

are then combined to provide additional insight, in order to give more strength to

the conclusions, or to propose a new interpretation. In the sections above, we have

shown examples where a stronger coupling between modeling and data analysis

9781405159050_4_007.qxd 6/3/08 4:44 PM Page 178

Climate of the past millennium

179

could be very instructive. This is precisely the goal of the assimilation of observa-

tions or data assimilation, which is deﬁned by Talagrand (1997) “as the process

through which all the information is used to estimate as accurately as possible

the state of the atmospheric or oceanic ﬂow. The available information essentially

consists of the observations proper, and of the physical laws which govern the

evolution of the ﬂow. The latter are available in practice under the form of a

numerical model.”

Data assimilation techniques have been used successfully in various domains of

meteorology and oceanography during the past decades (for a review, see for

instance Talagrand 1997; Kalnay 2003, Brasseur 2006). One of the initial and most

important applications of such techniques is in weather forecasting, to determine

the initial conditions at a time t

from which subsequent evolution of the system at

time t > t

will be deduced. This step, which is called the analysis, combines some

background information (also denoted as ﬁrst guess or prior information) with

available observations. Following the notation and formulation of Kalnay (2003),

observed variables are noted y

, while x

is the background ﬁeld. It is generally

obtained from a previous forecast using a model simulation starting from an ana-

lysis at time t − 1 (t − 1<t

) until time t

. Both y

and x

are three-dimensional

ﬁelds. Variables y

and x

are not usually obtained at the same location and some

interpolation is needed. Furthermore, observations could provide a different vari-

able than the one needed by the model. Some transformation of model results (or

alternatively of observations) is then needed before they can be compared with

data. This is represented by H(x

), where H is called the observation operator. The

difference between the ﬁrst guess obtained from the model forecast and observa-

tions, y − H(x

), is called observational increment or innovation. The analysis x

then obtained by adding this innovation to the model forecast, using some weight

W that is determined by accounting for uncertainties in model results as well as in

observations. This could be mathematically represented by

= x

+ W(y

− H(x

)) (1)

The sequence of obtaining a ﬁrst guess from model results and then updating

using data to provide the analysis can be repeated from time to time in an “analysis

cycle”. Such a technique is often called sequential assimilation (e.g. Talagrand

1997; Kalnay 2003).

In weather forecasting, model and data assimilation schemes (for instance the

formulation of W and H) are improved regularly, providing potential inconsis-

tencies between different periods. Reanalyses of the past 50 years, using the same

procedure during all periods covered (e.g. Kalnay et al. 1996), have provided a very

valuable and widely used comprehensive set of atmospheric ﬁelds.

Such a procedure could, in theory, be applied to any period and thus also for the

past millennia. However, several difﬁculties arise, a major one being the amount of

data. Hundreds of thousands of atmospheric and surface observations are used for

the analyses in weather forecasting, covering all areas of the world, and providing a

very detailed knowledge of the atmospheric state, whereas only a few tens up to

9781405159050_4_007.qxd 6/3/08 4:44 PM Page 179