Annual Reviews in Control 42 (2016) 114–125

Contents lists available at ScienceDirect

Annual Reviews in Control

journal homepage: www.elsevier.com/locate/arcontrol

Review

Earth climate identification vs. anthropic global warming attributionR

Philippe de Larminat
Professor (retired) from Ecole Centrale, Nantes, France

a r t i c l e i n f o a b s t r a c t

Article history: Based on numerical models and climate observations over past centuries, the Intergovernmental Panel on
Received 3 February 2016 Climate Change (IPCC) attributes to human activity most of the warming observed since the mid-20th
Revised 14 September 2016
century. In this context, this paper presents the first major attempt for climate system identification – in
Accepted 27 September 2016
the sense of the systems theory – in the hope to significantly reduce the uncertainty ranges. Actually, cli-
Available online 25 October 2016
matic data being what they are, the identified models only partially fulfill this expectation. Nevertheless,
Keywords: despite the dispersion of the identified parameters and of the induced simulations, one can draw robust
System identification conclusions which turn out to be incompatible with those of the IPCC: the natural contributions (solar
Climate activity and internal variability) could in fact be predominant in the recent warming. We then confront
Global warming our work with the approach favored by IPCC, namely the “detection and attribution related to anthropic
Detection and attribution climate change”. We explain the differences first by the exclusion by IPCC of the millennial paleoclimatic
Anthropogenic
data, secondly by an obvious confusion between cause and effect, when the El Niño index is involved in
detection and attribution.
© 2016 The Authors. Published by Elsevier Ltd on behalf of International Federation of Automatic
Control.
This is an open access article under the CC BY-NC-ND license
(http://creativecommons.org/licenses/by-nc-nd/4.0/).

1. Introduction the anthropic attribution of climate change. This last point is clear
through the title of the following major publication: “Good prac-
The climatic process is a highly complex system, on which sci- tice guidance paper on detection and attribution related to anthro-
entists experienced in systems theory have much to say. This con- pogenic climate change”, by Hegerl et al. (2010). The lack of refer-
cerns particularly the global climate modeling and the attribution ence to identification is puzzling, knowing that the D&A has close
of the recent warming to human activity. This analysis involves cli- relationships with it, and that the respective findings are mutually
matic observations, present and past, direct and indirect. Then, a inconsistent.
preferred approach would rely on dynamical systems identification, This paper describes the first significant work on the identifi-
a theory which is well known to all systems scientists, but has not cation of the climate system. It summarizes some findings from
been applied so far to the climate science. our book “Climate Change, identification and projections” (de Larmi-
Actually, bibliographic searches based on the key words system nat, P., ISTE/Wiley, 2014). It adds news developments about its re-
identification, climate, global warming, return strictly nothing re- lationship with the D&A, and further elaborates on the differences
lated to identification of the climatic process. But if the key words between our conclusions and those of the IPCC.
detection and attribution are added, there are now dozens of pa- The latest IPCC Assessment Report is the fifth (AR5, 2013): 1550
pers regarding the attribution of climate change to human activity. pages, 9200 publications quoted. A synthesis is made in the Sum-
Conversely, the sole couple of keywords detection and attribution mary for Policy Makers (SPM, 2013). One of its main conclusions is
addresses references exclusively relates to anthropogenic climate that “it is extremely likely that human influence has been the dom-
change. inant cause of the observed warming since the middle of the 20th
In fact, it appears that “Detection and Attribution” (D&A) is an century”. It is mainly supported by Chapter 10 of AR5: “detection
emerging theory, born in the early 21th, dedicated exclusively to and attribution - from global to regional”. But these conclusions are
infirmed by those based on identification: from the millenary cli-
mate observations, it appears that the recent warming is due pri-
R marily to natural causes (solar activity and random variations), and
This research did not receive any specific grant from funding agencies in the
public, commercial, or not-for-profit sectors that one cannot reject the hypothesis that the human contribution
E-mail address: [email protected] be negligible.

http://dx.doi.org/10.1016/j.arcontrol.2016.09.018
1367-5788/© 2016 The Authors. Published by Elsevier Ltd on behalf of International Federation of Automatic Control. This is an open access article under the CC BY-NC-ND
license (http://creativecommons.org/licenses/by-nc-nd/4.0/).
 P. de Larminat / Annual Reviews in Control 42 (2016) 114–125 115

The purpose of this paper is to clarify the causes of this contra-

diction. It is organized as follows.
Section 2 presents the climatic data. Inputs: representative sig-
nals of human, solar and volcanic activities; output: the global sur-
face temperature.
Section 3 describes the fundamental features of the Earth’s cli-
mate system and the mathematical structure of an identifiable
model.
The main results of identification are presented in Section 4,
obtained by the Output Error method (OE), as well as the conclu-
sions of the statistical analysis (reported in appendix) and hypoth-
esis testing.
Section 5 presents and criticizes outcomes of the D&A: first,
the observation periods (from a few dozens to about one hundred
years) are too short and therefore lead to underestimate the inter-
nal variability, increasing the risk of a false detection of the human
contribution in global warming. Fig. 1. Four reconstructed temperatures.
Furthermore, this recent period is characterized by the simulta-
neous increase of global temperature and of atmospheric content
in CO2, while the major past climate events (Medieval Warm Pe-
2.2. Global mean temperature
riod, Little Ice Age) are the only ones that may allow highlighting
the solar contribution. Finally, D&A studies which involve the El As an output, the global climate indicator is the mean surface
Nino index make a fundamental methodological error, namely con- temperature. Fig. 1 presents a catalogue of four reconstructions:
fusion between cause and effect in the climate process. The general Ljungqvist (2009); Loehle (2007); Moberg et al. (2005); Mann,
conclusions are drawn in Section 6. Bradley, and Hughes, (1999). All four, except Mann, are quoted in
the AR5 (Chapter 5: paleoclimate archives). They are aligned on –
and extended by – modern measurements from 1850 (HadCrut4:
thick black curve).
2. Input-output data of climatic process The further coming back in time, the more rare and inaccurate
the proxies are. Some series start at the (symbolic) year 10 0 0. For
2.1. Causes and effects
reasons of accuracy and availability, we will exploit temperatures
The Earth’s climate is a complex natural system on which we reconstructions reduced to the second millennium. Moberg and
can observe a large amount of signals, among which it is not al- Mann reconstructions are restricted to the northern hemisphere,
ways easy to distinguish which are causes and which are effects. Ljungqvist and Loehle to extra tropical zones. Nevertheless, the dif-
For systems scientists, the question of causality makes sense only ferences between modern temperatures of the northern and south-
if the concerned system (or subsystem) is clearly delimited; know- ern hemispheres are much lower than the observed disparities be-
ing that for coupled systems, the same signal often is both a cause tween the reconstructions above, which allow considering that the
for a subsystem and an effect for another. The answer is unam- North/South or other climatic differences are dominated by errors
biguous when the causality can play only in one direction, for ex- due to proxies and reconstruction techniques.
ample between solar activity and terrestrial climate. It is much less We note that the curve of Mann, called Hockey Stick Graph, de-
obvious when it comes to variables internal to the climate system, viates significantly from others, which will reflect on the results of
such as the phenomena of oceanic oscillations and the associated the identification.
ENSO (El Niño South Oscillation) index. We will further comment
on this point in Section 5.
One of the available means to assess the relative contributions
of the different causes is the theory of dynamical systems identifi- 2.3. Anthropic indicator: CO2 atmospheric concentration
cation, in particular the branch dedicated to the determination of
causal models from observed input output signals. Typical causal Human activity has an impact on the emissions of greenhouse
dynamic models correspond to linear transfer functions, rational or gases (GHG), industrial aerosols, land use changes, etc. From C, the
not, and more generally to state-space models. atmospheric concentration of CO2 , we define a global indicator of
Concerning the whole climate system, it is clear that the global human activity as:
temperature is an effect. The major independent causes – on what
temperature has no action in return – are the solar activity, the u1 = log2 (C/C0 )
volcanism and, to a large extent, human activities.
The issue of available climate data is crucial, both for identifi- where C0 is the preindustrial concentration (ante 1750), about
cation and for detection and attribution. The reader must there- 280 ppm (parts per million).
fore get a precise idea of the datasets that we have gathered and Several reasons motivate this formula. First, the action of CO2
used. Knowing the large time scales involved in the climate sys- is reportedly predominant. Also, others anthropogenic actions are
tem, identification requires input-output data whose period widely cross-correlated and may tend to mutually compensate. Moreover,
exceeds those of the ’historical’ measures – which, according to cli- the CO2 -induced greenhouse effect is widely admitted to follow a
matologists, start between 1850 and 1880. Paleoclimatology allows logarithmic law. Finally, CO2 doubling is often considered as the
reconstructing past climate data from substitution measures or unit of variation; hence the interest of the base 2 logarithm.
proxies (tree rings, isotopes stored in sediments, ice cores, etc.) The Fig. 2 shows the signal u1 , resulting from the connection of
accessible reconstructions, available in public data bases (NOAA, modern atmospheric measures with the archives extracted from
NASA, Hadley Center, etc.) are far from overlapping perfectly, and Arctic or Antarctic ice cores (source: NOAA and CDIAC). Note that
are not always well connected to the historical data. u1 = 0.5 ↔ C = C0 + 41 %
116 P. de Larminat / Annual Reviews in Control 42 (2016) 114–125

shows the 3-components decomposition of the TSI reconstruction

SORCE/TIM (SOlar Radiation and Climate Experiment/Total Irradi-
ance Monitor) by Kopp and Lean (2011). This composite recon-
struction is based both on modern satellite measurements and on
sunspots observed since 1610 (invention of the Galileo telescope).
Our decomposition principle is outlined in annex A.
In all signals and proxies related to solar activity, one can find,
with different relative magnitudes, the same cyclic component and
the same LF component (HF component being widely considered
a measurement noise). The mutual ratio between the cyclical and
the LF components is extremely variable, depending not only on
Fig. 2. Anthropic signal: u1 = log 2 (C/C0 ).
the nature of the signals (associated to mechanisms 1 to 5 above),
but also on the given reconstructions of a same signal, in particular
2.4. Solar activity the TSI.
The cyclic components of all the TSI reconstructions have in-
The manifestations of solar activity are multiple and likely to deed roughly the same maximum excursion (1 Wm–2 in the 20th
impact the climate through extremely various mechanisms. century), but from 17th to 20th, according to reconstructions, LF
components range from 1 to 10 Wm–2 (see, for example, Lean
1. First of all, there is obviously the total energy flux (TSI: To- (2004) or Shapiro et al. (2011). These differences come from the
tal Solar Irradiance), weakly variable around 1367 Wm–2 , and absence of TSI measures over sufficiently long periods allowing cal-
which directly affects the Earth’s radiative balance. ibration with a good precision of the LF components of proxies re-
2. The IR/UV spectral distribution acts differently, through the lated to solar activity.
stratospheric creation of a GHG (the ozone). On the other hand, it is recognized that the cyclic variations (11
3. The solar magnetism modulates cosmic radiations, which are years) detectable on the global temperature are practically negligi-
likely to act on the formation of condensation nuclei and thus ble (less than one-tenth of degree). If we admit that the recon-
on terrestrial cloud cover and its albedo effect. This modulation struction SORCE/TIM is valid and that solar activity occurs on the
is found in proxies consisting of cosmogenic isotopes (or cos- climate exclusively through the energy factor (mode 1), we should
monucleids: 12 Be, 14 C). concluded, like IPCC does, that the impact of solar activity is al-
4. The solar wind, and its known role in boreal Auroras. most zero.
5. The background radio noise, not or little studied in the climate Some mechanisms involved in low-frequency (through the he-
context... liocentric magnetic field, for example) are not yet well enough un-
derstood to include them in the physical models. It does not imply
All the signals related to solar activity clearly appears as a
that they do not exist. Our identification is precisely intended to
sum of three components: (1) cyclic, period around 11 years, (2)
determine if the climatic contributions of the low-frequency solar
high-frequency (HF), (3) low-frequency (LF). For example, Fig. 3
activity is significant, if not predominant.
If therefore one accepts either that LF variations of TSI can be
very superior to those of Fig. 2, or that other mechanisms of so-
lar activity may predominate over the energy factor, the just way
to highlight it will be to restrict reconstructions and/or proxies of
solar activity to their only LF components.

2.5. Low Frequencies indicators for solar activity

Given the multiplicity of manifestations and mechanisms of so-

lar activity, there would be no reason to favor one physical signal
rather than another (TSI, heliocentric magnetic field, potential so-
lar modulation…), or one proxy in preference to another, especially
after reduction to low frequencies component.
For various needs, it is however preferred to convert everything
in terms of TSI. We arbitrarily adopted here for reference signal the
SORCE/TIM reconstruction, widely used by IPCC (AR5), and limited
here to its LF component (Fig. 3-b). After calibration and alignment
on this component, Fig. 4 combines the LF components of the four
following proxies:

– The SORCE/TIM LF component itself.

– The series of sunspots groups number (SGN), reconstructed
since 1610, recently reviewed and corrected by the Royal Obser-
vatory of Brussels (Clette, Svalgaard, Vaquero, & Cliver, 2015).
– The cosmogenic series of Usoskin, Korte, and Kovaltsov (2008),
initially expressed in CRII (Cosmic Ray Induced Ionization Rate
Reconstruction), based on concentrations of 14 C (years 5 to
2005).
– The cosmogenic series of Delaygue and Bard. (2010), initially
Fig. 3. SORCE/TIM decomposition. expressed in 10Be anomalies (years 695-1982).
 P. de Larminat / Annual Reviews in Control 42 (2016) 114–125 117

Fig. 5. Volcanic activity.

Fig. 4. Four low frequency index for solar activity.

3.3. Proposed structure

Making connections Usoskin-SORCE/TIM, Delaygue-SORCE/TIM,
Usoskin-SGN, Delaygue-SGN, this produces a catalogue of four so- Our climate model is organized into two coupled subsystems
lar indicators (Fig. 4). (Fig. 6).
The heat accumulation subsystem includes all the thermal iner-
2.6. Volcanic activity tia in which the flow  of energy balance (in Wm–2 ) accumulates,
circulates and diffuses. It mainly consists in oceanic mass, the su-
Volcanic activity is estimated through the Aerosols Optical perficial thermal inertias of the continents being negligible.
Depth (AOD) produced by eruptions (Crowley & Unterman, 2013). The thermal inertia of the atmosphere is also negligible. It accu-
Fig. 5 shows this indicator u3 , centered on its mean value and mulates virtually no energy, so that the radiative balance  at the
sign changed, because volcanism is known to impact the tempera- Top Of the Atmosphere (TOA) is therefore fully transmitted, with-
tures downwards. out significant delay, into the oceanic thermal inertia. Immediacy
is to be understood here with respect to the climate time scale, i.e.
3. An identifiable model less than one year (time unit adopted in climatology).
The thermal exchanges between atmosphere, ocean and space
3.1. General circulation models (GCM) are driven by the three aforementioned independent causal inputs
u, referred to as forcing factors, and also by the resulting global
GMC’s are knowledge models, which can be simulated by finite surface temperature TG , which depend itself on the oceanic tem-
elements, including thousands of interconnected cells (atmospheric perature TO .
and oceanic). The equations that govern each cell and their mutual
3.4. Mathematical expression of the model
interactions are firstly those of fundamental physics: mass and en-
ergy conservation, fluid mechanics, thermodynamics, radiation, etc.
The accumulation of heat in the oceanic thermal inertia is de-
They incorporate also some amount of more or less empirical
scribed by:
representations concerning for example the clouds, their genesis
and their radiation and absorption properties, or the heat trans- IO dTO /dt =  (1)
fer at the Earth’s surface by convection, evapotranspiration, etc. where IO is in (W/m2 )/(K/year).
The associated parameters are somewhat arbitrary and usable as The thermal exchanges (Fig. 6) expresses into algebraic equa-
adjustment variables. The overabundance of adjustable parameters tions. Regardless of their complexity, one can admit the existence
gives designers the ability to get any desired result, in particular of linear approximations. Then,  is classically written as:
accurate reproduction of the warming in the last quarter of XXth
century.  ≈ α1 u1 + α2 u2 + α3 u3 − λ (TG − TE ) (2)
Note finally that the GCMs spontaneously reproduce the fluc- Recall that the ui are the deviations of the forcing indicators
tuations of the atmospheric or oceanic chaos, at the origin of the compared to their nominal values. The products α i ui give the re-
climate internal variability (see Section 3.6). sulting radiative forcings, where the α i ’s are the associated coeffi-
cients of radiative forcing.
3.2. Energy balance models (EBM) TE is the hypothetical surface temperature at equilibrium, when
u1 = u2 = u3 = 0, and assuming the absence of internal variability.
The large numerical models are obviously the only ones able to
reflect the regional climatic particularities. However, if interest is
primarily focused on global behaviors, the climate system makes
no exception to the vast majority of complex systems, which lend
themselves to representation by input-output models, so-called
‘black box’ (or ‘grey box’ models, when they incorporate some
macroscopic physical laws).
This is the case of so-called Energy Balance Models, some of
which have been developed by IPCC (e.g. Meinshausen, Raper, &
Wigley, 2011). These models reproduce the global behavior of large
digital models, on which they are tuned, but their complexity is
not yet reduced enough to make them identifiable from the avail-
able input-output climate observations. Fig. 6. Structure of an energy balance model.
118 P. de Larminat / Annual Reviews in Control 42 (2016) 114–125

This TE is unknown, due to the lack of measurements in the pre- and where 0.3 is the mean terrestrial albedo. Dividing (9) by α 1 ,
industrial era. we obtain the following bounds on sensitivity irradiation:
λ is a climatic feedback coefficient, expressing the variation of
0.27 ◦C/W m−2 < S2 < 1.62 ◦C/W m−2 (10)
the radiative balance under the effect of the deviation TG – TE . All
the above parameters are widely used in the climate science. They Using our solar indicators, based on SORCE/TIM, we shall see
directly result from the universally accepted concept of energy bal- that the identified lower bound for S2 will be much greater than
ance. the above upper bound, except when using the hockey stick curve.
In (2), the global surface temperature TG results from the ther-
mal version of the Ohm law: 3.6. Internal variability

TG = TO + r  (3)
The natural variability of the climate results from variations of
where r denotes the thermal resistance through the surface, de- causes other than human (solar, volcanic). It includes also, but is
pending on the convection characteristics of the atmospheric and not restricted to internal variability, which mainly comes from tur-
oceanic limit layers, and on the evapotranspiration properties of bulences, inherent to any fluid flows. The chaos of the atmospheric
the surface. Eliminating  between Eq. (1 to 3) easily leads to the and oceanic circulations originates from a tingling of independent
monodimensional state space model: causes: the famous flaps of butterfly’s wings, able to initiate – or
not – meteorological or climatic events). Globally, internal vari-
Tclim dTO /dt = −TO + TE +  Si ui
(4) ability is equivalent to an additive disturbance v acting on the
TG = (1 − ρ )TO + ρ TE + ρ  Si ui
output:
where:
(TG − T0 ) = G(s )  [Si ui ] + v
Tclim = IO (1 + r λ )/λ, ρ = rλ/(1 + rλ ), (5) This disturbance is not a white noise (an independent se-
quence) and its spectrum will be taken into account, if not in the
Si = αi /λ, i = [1 : 3] (6) method of identification, at least in the calculation of the variance
of the estimated parameters (see Appendix B).
An equivalent formulation is the following:
4. Identification
(TG − T0 ) = G(s ) (S1 u1 + S2 u2 + S3 u3 ) (7)
where 4.1. Specificity of the climate process identification
1 + ρ sTclim
G (s ) = (8) In system identification, it is usually recommended to save part
1 + sTclim
of the data for model validation purposes. To do this, it would be
is a transfer function of unit dc gain (G(0) = 1) and the Si ’s are the necessary, either to have multiple experiments, or to split a single
equilibrium sensitivity coefficients with respect to the forcing fac- one, but of long duration.
tors ui . History, climate or other, does not repeat itself. Regarding the
The structure (7, 8) results from reasonable physical assump- low frequencies involved in the climate process, one millennium is
tions. The most important point is that the transfer function G(s) barely enough, and splitting would make it unusable for the iden-
is common to the three entries. This comes from the fact that the tification, in particular with regard to solar activity. In this situ-
radiative balance is generated by the atmospheric machinery, as- ation, validations through the statistical calculation of the uncer-
suming a negligible thermal inertia, and therefore modeled by al- tainty ranges according to rigorous methods is therefore of partic-
gebraic equations, such as 2 and 3. The transfer function G(s) re- ular importance.
duces here to a first-order filter (with numerator), which appears Another difficulty comes from the discrepancies in the recon-
to be sufficient to describe the climatic transients in response to structions of temperature and of solar irradiance (Figs. 1 and 4).
the low frequency components of the excitations ui . It is not neces- Depending on the selected data, some uncertainty ranges will
sary to introduce a transfer function G(s) of higher order, as we did reveal incompatible with each other (see below Fig. 10). In absence
in our book (de Larminat, 2014). Even without formal evaluation of objective criteria for rejecting some data rather than others, the
(AIC criterion, for example), it appears through our identification only objective attitude is to present all of the results obtained from
attempts that higher order models are over-parameterized, regard- the 16 possible combinations (4 temperature × 4 solar irradiances),
ing the input output data, which would require – as we previously and to leave observers their freedom of judgment.
did – an arbitrary tuning of the denominator the denominator of
G(s), in order to reduce the number of free parameters. 4.2. Method

3.5. A priori estimates Since several decades, identification of dynamic systems is a

mature discipline: Åström and Eykoff (1971); de Larminat and
The model will be identified under the reduced form (7, 8). In Thomas (1977); Söderström and Stoica (1988); Walter and Pron-
its last report, IPCC gives bounds for equilibrium sensitivity to CO2 zato (1997); Ljung, (1999); de Larminat (2009), chapter 13; Landau
doubling: (2001).
Here, observation data are poor in events and severely dis-
1 ◦C < S1 (Prob. > 95 % ), S1 < 6 ◦C (Prob. > 90 % ) (9)
turbed by noises and internal variability. Then, the simplest and
IPCC does not directly provide such bounds for the solar sen- most robust method is the Output Error method (OE).
sitivity S2 . Nevertheless, we can deduce it, knowing that (6) im- The principle is easily understandable, even by non-experts: it
plies S2 = (α 2 /α 1 ) S1 . According to IPCC, the radiative forcing to consists in simulating the model (here the Eq. (7)), fed by the
CO2 doubling is α 1 ≈ 3.7 Wm − 2 ± 10%. In the IPCC models, so- recorded input signals ui , and in tuning the parameters of the
lar activity is assumed to be acting through the energetic factor model until the deviation between the simulated and observed
only, then α 2 = 1, after converting the TSI into Net Solar Irradiance: output be minimized in the mean square sense. Using the OE
NSI = (1 − 0.3)/4 × TSI, where the factor 1/4 is for Earth sphericity method, the observation data speak freely, without any constraints
 P. de Larminat / Annual Reviews in Control 42 (2016) 114–125 119

Fig. 7. Free identification. Fig. 8. Constrained identification.

that could impact the results, nor even structures or variances of

4.4. Tests of hypothesis
noises.
Actually, like for many other identification methods (AR, AR-
Another way to practice the identification techniques is to per-
MAX, B&J, PEM, etc., see Ljung (1999)) the statistical characteris-
form forced optimizations, under assumptions of which we want
tics of the noises and disturbances are extracted from the observed
to test the relevance. According to the IPCC, solar activity acts ex-
data themselves (here from the correlation function of the residu-
clusively through the total flux of solar irradiance, with a rather
als). So, we do not need to make use of the confidence ranges that
insignificant solar sensitivity, most likely less than 1.6 2 °C/Wm–2
are often provided with the climate data.
(Eq. (10)).
In our case, the residual output error will be far from a white
As expected, performing the identification under this constraint
noise (Fig. 7-b). The method is therefore not statistically optimal.
(S2 < 1.62 °C/Wm − 2 ) leads to an insignificant solar contribution
This does not prevent computing the variance of the estimator (un-
(Fig. 8-d), and recent warming mostly attributed to the anthropic
der the classical approximations: normality, ergodicity, asymptotic
factor (frame c).
convergence, etc.) See Appendix B.
Some indications tend to dismiss this hypothesis:

– In the free identification, solar activity contributes to explain

the medieval warm period and the little ice age. It is not so in
4.3. First set of results forced identification (Fig. 9-d).
– As a result, the error output visibly increases over these peri-
Consistently with our principles, we have dealt with all the ods.
combinations of data sets, but for the sake of illustration, this sec- – A significant cross-correlation (not shown here) appears be-
tion is limited to the identification results from our first combi- tween the solar activity indicator and the output error, sign of
nation of temperature and irradiance, namely Moberg × Usoskin- a causality not taken into account.
SORCE/TIM.
Fig. 7 shows (a) the simulated output of the optimized model, Visual assessments being not formal proofs, this must be con-
(b) the output error and (c, d, e) the respective contributions of the firmed by hypothesis testing, in order to confirm whether the er-
three forcing factors. In these frames, the light grey curves show ror output really significantly increases, implying rejection of the
the observed output (Moberg). hypothesis. The statistical tests, based on the estimated variances
It can be seen that the output error is large, but comparable (Appendix B) show that: the hypothesis of a low sensitivity to solar
with the millennial simulations of IPCC (see e.g. AR5, 2013; Fig. activity must be rejected with a probability level greater than 90%.
1 (b), p. 78). This comes not only from output noise, reconstruction Similarly one can test the hypothesis of a low anthropogenic
errors, and identification errors, but mainly from the internal cli- sensitivity, in the limit S1 = 0. Then, the reproduction of recent
mate variability (Section 3.6), irreducible by nature. This last com- warming is significantly degraded, but the estimated internal vari-
ponent being not a white noise, it not an independent identically ability remains comparable to the millennial values. The hypothesis
distributed sequence (idd). test confirms: with a 90% probability level, one cannot reject the
Yet in this first case, the recent anthropogenic contribution is hypothesis of a zero anthropogenic contribution. In other words:
found to be less than the contribution of solar activity. Reflect- The selected combination of observation data invalidates the claim of
ing the predominance of internal variability in the error output, the IPCC, that the anthropogenic contribution to recent warming is
the natural contribution (solar and volcanic activities, plus internal predominant with 95% probability level.
variability) becomes clearly much greater than the anthropogenic
contribution in the recent warming. 4.5. Exhaustive results
Recall that the contributions shown in Fig. 7 are specific to our
first combination of data. One should therefore not yet generalize Previous results were one-off, based on one particular set of
these conclusions. data. Instead of detailing similar results, contributions and tests
120 P. de Larminat / Annual Reviews in Control 42 (2016) 114–125

Fig. 10. Anthropic contribution to climate change.

For further investigations, it would be of interest to use more

Fig. 9. Confidence intervals. refined analyses, for instance (Faivre, Iooss, Mahévas, Makowski, &
Monod, 2013).
Anyway, one can think that the above conclusions are robust:
they remain practically the same as in our previous works (de
relating successively to each set, Fig. 9 limits to the estimates of Larminat, 2014), despite substantial quantitative changes, consec-
S1 and S2 , along with their 90% confidence intervals, calculated utive to the structure adopted for the model, the frequencies de-
from the 16 combinations of data. There are grouped into four sets, composition of solar activity and the selected paleoclimatic data.
relating to the respective reconstructions of Moberg, Ljungqvist,
Loehle and Mann. Each group is subdivided according to the four 5. Detection and attribution
reconstructions of solar irradiance (Section 2.5). The grey segments
show the ranges of sensitivity according to IPCC (Eqs. 9 and 10). 5.1. Introduction: “Good practice guidance paper on detection and
Fig. 9-b clearly shows a total mismatch between confidence in- attribution” (GPGP)
tervals resulting from the use of the Mann’s reconstruction and
those from the first three. At the same time, this last alone could Generally speaking, detection recognizes if there actually exists
confirm the IPCC parametric ranges, both for the anthropic and the a phenomenon potentially masked by random fluctuations; while
solar sensitivity. the attribution specifies, or even quantifies the causes to effect re-
At the outset, there is no reason to discard the reconstruction of lashionship. D&A operates on the basis of climate observations and
Mann: scientific truth does not determine by majority. The hockey from some predetermined climatic models (GMC or EBM). Unlike
stick graph had appeared repeatedly in previous reports of the identification, it does not intend to rebuild or to retune these mod-
IPCC, and it is still widely used in many vulgarization reports on els. Note that the human-induced climate change is set as a prin-
climate change. But one cannot ignore that it has been the subject ciple. This appears in the complete title: “Good practice guidance
of serious controversies; and it must also be reported also that the paper on detection and attribution related to anthropogenic climate
IPCC no longer mentions it anywhere in the 1550 pages of his 2013 change” (Hegerl, 2010), which echoes the mission of IPCC: “assess
report, even in chapter 5 on Paleoclimate Archives. the scientific, technical and socio-economic information relevant
If we limit ourselves to the three other reconstructions for the understanding of the risk of human-induced climate change.”
(Moberg, Ljungqvist, Loehle), then our conclusions are as follows: (www.ipcc.ch/Procedures/Principles/Role).
About anthropic sensitivity: This ‘Good Practice Guidance Paper’ (GPGP), was intended for
The partial conclusions of 4.3 and 4.4 don’t generalize. In the the drafters of the future AR5 report being prepared at that time.
whole, the lower bound S1 > 1 °C can neither be confirmed, nor GPGP is partly a compendium of definitions and language ele-
invalidated. It depends also of the TSI reconstructions. But in all ments: it specifies that external forcing refers to causes external to
cases, the upper bound S1 < 6 °C is far from being reached. the climate system seen as a whole (typically human, solar and
About solar sensitivity: volcanic actions), while the term external drivers is for more gen-
The first three reconstructions lead to reject the IPCC hypothesis eral use (e.g., ‘the reduction of sea ice might act as an external driver
of a low sensitivity to solar activity. on polar bear populations’).
Two explanations may combine. The first could be the exis- It also introduces the concept of confounding factor “which may
tence of mechanisms of solar action other than energy factor, the mask or shears the effects of external forcings and drivers”. A long
only one retained by IPCC; the second could be a too low eval- list includes the following items: “...model errors and uncertainties;
uation assessment of the LF component in the SORCE/TIM recon- improper or missing representation of forcings in climate and im-
struction, used as reference in Section 2.5. If one adopts a recon- pact models; structural differences in methodological techniques; un-
struction where the LF component is higher, such as Shapiro et al., certain or unaccounted for internal variability...”. Each of the points
(2011), then the identified sensitivity range might become compa- above would have deserved further clarifications: by itself, warn-
rable with that of the IPCC. ings against confusions do not ensure that the D&A provides ad-
Now, the most important is not the value of the coefficient S2 , equate tools to avoid them. In particular, “unaccounted for internal
but the assessment of whether or not the contribution of natu- variability” is mentioned as one such confounding factor. Actually,
ral factors prevails on the human contribution. In Fig. 10, twelve it seems at the contrary that it is accounting for ENSO index which
simulations (discarding again those from Mann) show the anthro- would constitute an obvious confusion, this between cause and ef-
pogenic contributions on the last century. On average, the assertion fect (Section 5.5).
of the IPCC that most of the warming observed during the second GPGP lists various methods for D&A: attribution to external
half of the 20th century is of human induced is not confirmed. forcings (single-step or multi-step: depending on whether the
 P. de Larminat / Annual Reviews in Control 42 (2016) 114–125 121

chain of causality is single or multiple); associative pattern attribu- then nothing else than the contributions of the various inputs to
tion; assignment to a change in climatic conditions “which may be the output, simulated by a model fully specified a priori, including
the final step in Multi-Step Attribution”. This last mention is impor- G(s) and sensitivities: Xi = G(s) Si ui .
tant: the GPGP points out that ‘the overall assessment will generally Then, the vector aˆ, obtained by optimal fingerprinting (Eq. 14)
be similar to or weaker than the weakest step’. would allow a readjustment of the initial sensitivities, by multiply-
The SPM describes about twelve climatic effects, all attributed ing the a priori given values Si by the factors aˆi . The main differ-
to anthropogenic influence: in the first place the global warm- ence between our identification and D&A is that the last kept the
ing itself. Then, come the melting of sea ice, the raising of the transfer function G(s) outside the scope of the identification. In the
sea level, the frequency of extreme events, the heat waves, etc. GCM, the sensitivities do not appear explicitly. The D&A therefore
In chapter 10 of AR5, about sixty contributors provide dozens of cannot lead to identification strictly speaking. Besides this point,
other examples, giving material to about 700 bibliographical refer- there is no fundamental difference in nature between deductions
ences. However most of the submitted attributions are multi-step based on D&A and those from system identification, by hypothesis
ones and involve anthropic global warming as an initial step. If the testing’s and confidence intervals.
weakest link lies in the attribution of global warming to human By nature, this identification has the advantage of taking into
activity, then the reality of all these attributions becomes highly account uncertainties on the transients modeled by the G(s), while
questionable, and their accumulation cannot in any way be con- the D&A only claims being robust: “Attribution does not require, and
sidered as multiple evidence of human influence on climate. nor does it imply, that every aspect of the response to the causal factor
Subsequently, we will thus focus exclusively on the assessment in question is simulated correctly” (AR5, p 873). If one admits this,
of the validity of the attribution of global warming by mean of the it is therefore necessary to look elsewhere to understand why the
D&A. conclusions between D&A and identification are so much opposed.
Minor technical differences, such as OE vs. BLUE do not bring sig-
5.2. D&A and fingerprinting nificant explanations. Ultimately, the explanation would first of all
be in the data used, their period and their natures.
One of the main tools used in D&A is called ‘optimal fingerprint-
ing’, a concept introduced in the 1990’s (Hasselman, 1993). The 5.4. Observation periods
principle is as follows (see, for example, Hegerl and Zwiers 2011).
Regarding global temperature, fingerprints (or patterns) are de- Almost all of D&A studies focus on climate observations start-
fined as the changes in the simulated temperature in response to ing after 1850, sometimes even well afterwards (1979). These du-
observed variations of each external forcing or driver, considered rations are much too short.
independently. The used simulation models are either some large A first reason lies in estimation of the statistical characteristics
digital general circulation models (GCM), either simple energy bal- of internal climate variability (the C matrix of Eq. (14)), which is
ance models. Unlike the approach of identification, these models essential to detect whether a variation emerges or not above the
are a priori fixed, and the D&A is not intended to revisit them. level of internal variability. This variability can be first assessed by
Making explicit linear hypothesis: means of general circulation models simulations, which inherently
y = X1 + X2 + X3 + v (12) can reproduce the atmospheric and oceanic chaos, primary respon-
sible for internal variability. Then it raises the question of the abil-
where y is the observed global temperature, Xi the fingerprints ity of these models to actually reproduce the variance and espe-
associated with each indicator of forcing (e.g., human, solar and cially the low-frequency spectrum of this variability.
volcanic activities), and v results from internal variability or from These characteristics can also be determined empirically, from
any unlisted causes. Introducing possible model errors, Eq. (12) be- the residues vˆ = y − X aˆ. In AR5, chapter 10, contributors show
comes: some embarrassment and inconsistency on this subject: “It is dif-
y = Xa + v (13) ficult to evaluate internal variability on multi-decadal time scales
(1950-2010) in observations, given the shortness of the observa-
where X = [X1 X2 X3 ], and where a is a vector of scaling factors, tional record…” (p. 881), in contradiction with: “… it is difficult to
each nominally equals to 1. An estimate of a may be obtained by validate climate models’ estimates of internal variability over such
some linear regression (e. g. BLUE: Best Linear Unbiased Estimate): a long period” (1861-2010: p. 882.)
The second reason is that paleoclimatic data do exist, and thus
aˆ = (X T C −1 X )−1 X T C −1 y (14) cannot be ignored. Millennial observations suggest a causal rela-
where the matrix C is the covariance of internal variability signal, tionship between solar activity and global temperature. The role of
the determination of which we will return on. The variance of the identification, as well as of D&A, is to assess the validity of this
estimate aˆ is given by the expression (XT C − 1 X) − 1 , from which one suggestion trough statistical analysis. Limiting itself to a centennial
can deduce the confidence intervals associated with estimates aˆ. time scale, the D&A leaves virtually no possibility to attribute the
Depending on whether the estimated intervals include (or not) the recent warming to solar activity rather than to human influence.
values 1 or 0, changes in y will be detected (or not), and will be On the other hand, treatment of the millennial data by D&A would
attributed (or not) to the corresponding forcing factor. almost certainly lead to conclusions similar to ours.

5.3. Relations with identification 5.5. El Niño: a confusion factor between cause and effect

The D&A allows involving the large General Circulation Models, Internal variability mainly arises from Ocean chaos, through the
and thereby addressing regional or local phenomena, ignored by thermohaline circulation, also called meridional overturning circula-
the ‘black boxes’ models. But if one restricts to the fundamental tion (MOC). The driving force of this circulation is the downwelling
question of the attribution of global warming, the D&A is strongly of the cooled and densified waters in Polar Regions. Those disperse
related with our identification, so much that both methods might and spread at the bottom of the oceans before resurfacing, among
be expected to lead to the same conclusions. many others places, in the vicinity of the eastern coast of Pacific.
Actually, let’s take again our model structure (Eq. 7): The chaotic variations of the atmospheric circulation can inhibit for
(TG − T0 ) = G(s) (S1 u1 + S2 u2 + S3 u3 ). The associated fingerprints are a more or less long time these cold upwelling, whose absence then
122 P. de Larminat / Annual Reviews in Control 42 (2016) 114–125

on the global or regional climatic state, which depends itself of the

real external causes.
It is not because oceanic chaos is involved in the onset of
El Niño that it is akin to an independent perturbation: the up-
welling’s are not only due to chance, but are modulated by the
mean level of the upper thermocline limit, which mainly depends
on the amount of heat in the ocean, itself determined by the ra-
diative balance, and therefore by the external forcings. One might
consider ENSO as a cause (in the sense of an external driver) only
for some subsystem, to be delimited through decomposition of the
climate process in interconnected sub-systems, involving various
feedback loops. Hence, dealing with El Niño in the same way than
an external forcing is a methodological error, which is obvious to
any expert in systems science.
Yet, in AR4 (2013), IPCC quote numerous works on the depen-
dence of ENSO on solar activity, and more generally of many other
tropical Pacific climatic signals, as attested by the following quota-
tions:
1. van Loon, Meehl, and Shea, 2007: “This then is physically
consistent with the mechanisms that link solar forcing to a
strengthening of the climatological mean circulation and pre-
cipitation features in the tropical Pacific”;
2. van Loon and Meehl, 2008 “in solar peak years the sea level
pressure (SLP) is, on average, above normal in the Gulf of Alaska
and south of the equator”;
3. White and Liu, 2008 “we find most El Nino and La Nina
Fig. 11. From Imbers et al. (2013). episodes from 190 0–20 05… Here we find these alignments
(Top) The variations of the observed global mean surface temperature anomaly replicated in both coupled general circulation model and con-
(HadCRUT3), and the best multivariate fits using the method of Lean, Lockwood,
ceptual model driven by 11-yr solar forcing”;
Folland and Kaufmann.
(Below) The contributions to the fit from (b) El Niño-Southern Oscillation (ENSO),
4. Meehl, Arblaster, Matthes, Sassi, and van Loon, 2009 “One of
(c) volcanoes, (d) solar forcing, (e) anthropogenic forcing and (f) other factors (At- the mysteries regarding Earth’s climate system response to vari-
lantic Multi-decadal Oscillation (AMO) for Folland and a 17.5-year cycle, semi- ations in solar output is how the relatively small fluctuations
annual oscillation (SAO), and Arctic Oscillation (AO) from Lean). of the 11-year solar cycle can produce the magnitude of the
observed climate signals in the tropical Pacific associated with
such solar variability”;
induces a more or less temporary warming. It is the El Niño phe-
5. Tung and Zhou, 2010 “It is noted that previous reports of a
nomenon, and La Niña refers to the inverse situation.
coldtongue (La Niña–like) response to increased greenhouse or
The intensity of El Niño phenomena is assessed through the
to solar-cycle heating were likely caused by contaminations due
ENSO index (El Niño South Oscillation Index, or SOI), defined from
to the dominant mode of natural response in the equatorial Pa-
a batch of measures: pressure, atmospheric and oceanic temper-
cific”;
atures, wind speeds, etc. A strong correlation between ENSO and
6. Roy and Haigh, 2010 “An important aspect of our paper is to
short-term global temperature variations is established, and makes
point out that the timing is crucial to show how this produces
the interest of this index.
apparent discrepancies between different analyses and how it
The ENSO index is equated with an external driver in many
may be used to test mechanisms proposed to explain solar-
studies: Lean (2009), Lockwood (2008), Folland et al., 2013, Kauf-
climate links, in the context of ENSO variability”;
mann, Kauppi, Mann, & Stock (2011). Their works was collected by
7. Roy and Haigh (2012). Both the SLP and SST signals vary coher-
Imbers, Lopez, Huntingford, and Allen (2013), and reported in the
ently with the solar cycle and neither evolves on an ENSO-like
AR5. Our Fig. 11 reproduces figure 10.6 of AR5. It can be criticized
time scale;
in several respects.
8. Bal, Schimanke, Spangehl, and Cubasch, 2011 “there is evidence
The anthropogenic contribution to global warming appears to
for a La Niña-like response assigned to solar maximum condi-
be predominant (frame e). Among them, the lowest is that from
tions”;
Folland (green), which deals with the longest duration of observa-
9. Haam and Tung, 2012 “The solar peak years can coincide with
tion (1890-2010); the others start in the 1950 s or later. Visibly, the
cold ENSO by chance, even if the two time series are indepen-
more the duration is restricted to the recent period, the more the
dent”;
correlation between warming and human activity predominates.
10. Hood and Soukharev, 2012 “the tropical lower stratospheric re-
Conversely, there is little doubt that an attribution based on mil-
sponse is produced mainly by a solar-induced modulation”;
lennial observations would reverse the findings of the D&A, in ac-
11. Misios and Schmidt, 2012. “the tropical Pacific Ocean should
cordance with the results of the identification.
warm when the sun is more active”.
Finally, there is something more questionable: the assimilation
of the ENSO index (or similar: AMO, SAO, AO) to some external All the above references are quoted from AR5 (Box 10.2: The
forcings. Recall that these indexes consist in batches of climatic ob- Sun’s Influence on the Earth’s Climate). They take place after the
servations. Each of them is an effect of the external forcings, and a following statement: « it can be difficult to discriminate the solar-
combination of effects cannot in any case be considered as a cause forced signal from the El Niño-Southern Oscillation (ENSO) signal ».
for the climate system as a whole. Actually, as said in 3.6, inter- Yet, this wording seems to go against the substance of the quoted
nal variability results from a tingling of elementary independent works (except the ninth): the question is not to discriminate be-
causes, but as soon as a measurable effect appears, it depends also tween two more or less similar signals, but to recognize that the
 P. de Larminat / Annual Reviews in Control 42 (2016) 114–125 123

ENSO signal is not an independent external forcing, and that it is – f(θ ) = [f1 (θ )ft (θ )fN (θ )]T is the series of the outputs, which
inconsistent to include it, like in Fig. 11, with the same status as would result from the simulation of the exact model driven by
the real external forcings. It has the effect of minimizing the con- the observed inputs.
tribution of solar activity, already underestimated by omitting the – ν = [ν 1 ν t ν N ]T represents the noises and disturbances act-
millennial observations. In contrast, the above proposed identifica- ing on the output, including those coming from input errors.
tion method clearly detects a strong contribution of solar activity This sequence is assumed to be identically distributed, ergodic
in the recent climate variations. and centered, but not independent.

The OE estimate of θˆ is that which minimizes the criterion:

6. Conclusions
J (θ ) = y − f (θ ) .
2

General circulation models are powerful tools for the study of

Around optimal estimate, the first order development of the
the climate system. Unfortunately, they still include many empir-
function f(θ ) is written as:
ical representations concerning phenomena to which the climate
is extremely sensitive, mainly those involved in the generation of f (θ ) ∼ f (θˆ ) + Fθ (θ − θˆ )
clouds. It is also possible that some climate mechanisms be still
completely unknown, in particular those related to solar activity. where Fθ is the Jacobian matrix of partial derivatives of f. At the
These models should therefore absolutely be validated by obser- optimum, the gradient Jθ of J(θ ) is zero. It develops as:
vations. Regarding global temperatures, models must be assessed Jθ = 2FθT (y − f (θˆ )) = 2FθT (v + f (θ ) − f (θˆ )) = 0
according to their ability to predict, or at least to reproduce the
past evolutions. The approach by “Detection and Attribution” of the Replacing f(θ ) by its development gives:
global warming is an attempt in this direction, but it not convinc-
2Fθ (v + Fθ (θ − θˆ )) ∼ 0
T
ing for the reasons set out in Sections 5.4 and 5.5 and that we
recall: Define Fθ+ as the pseudo-inverse of Fθ : Fθ+ = (FθT Fθ )−1 FθT . Pre-
The first reason is the systematic restriction of observations to multiplying the equation above by (FθT Fθ )−1 leads to the error es-
the period ‘history’ (post 1850), or even to the period ‘satellite’ timation:
(post 1979), thus leaving out paleoclimatic data, that are essential
despite their imperfections.
θˆ − θ ∼ Fθ+ v
The second is a flagrant systemic error, relative to the use of The error variance Vθ θ = E[(θˆ − θ )(θˆ − θ )T ] is there-
variability index (ENSO, AMO, or other). When there is confusion fore approximated by: Vθ θ ∼ E (Fθ+ v vT Fθ+T ) = Fθ+Vvv Fθ+T , where
between cause and effect, it is difficult to give some credit to the
Vνν = E(v vT ).
resulting conclusions.
In order to perform this expression, the Jacobian matrix Fθ is
The third is that the findings of D&A are contradicted by those
computed through finite differences, carrying out n simulations,
of the identification of the climate system (Sections 2 to 4). Ob-
successively varying each component: θˆi → θˆi + δθi . Under the er-
servation data being what they are, findings of identification are
godicity hypothesis, the variance matrix Vvv can be calculated us-
far from perfect. Yet parametric confidence intervals and hypothe-
ing the autocorrelation function of the residuals vˆ = y − f (θˆ ):
sis testing related to identification allow direct comparison with
1
the D&A. It appears that, despite dispersion of the observation N

data, the thesis of a predominant contribution of human activity Vvv (i, j ) ∼ vˆ t vˆ t+i− j
N
to global warming should be revisited in favor of dominant natural t=1

contributions: solar activity and internal variability. Once determined variance Vθ θ , one can perform any classical
analyses: uncertainty ranges, parametric tests, etc.
Appendix A. Three component decomposition
References
The decomposition of a (solar) signal y into its three compo-
SPM, Alexander, L. V., Allen, S. K., Bindoff, N. L., Bréon, F. M., Church, J. A.,
nents is obtained here by a Raugh-Tung-Striebel smoother (see e.g. Cubasch, U., et al. (2013). Summary for policymakers (2013).
Simo Särkkä, 2013), based on the following stochastic model: AR5, Stocker, T. F., Qin, D., Plattner, G. K., Tignor, M., Allen, S. K., Boschung, J.,
et al. (2013). Contribution of working group I to the fifth assessment report of the
y = y1 + y2 + y3 , intergovernmental panel on climate change.
Åström, K. J., & Eykhoff, P. (1971). System identification – a survey. Automatica, 7(2),
dy1 3 /dt 3 = w1 , 123–162.
dy2 2 /dt 2 + (2π /11 )y2 = w2 , Bal, S., Schimanke, S., Spangehl, T., & Cubasch, U. (2011). On the robustness of the
solar cycle signal in the Pacific region. Geophyshical Research Letters, 38, L14809.
y3 = w3 , Clette, F., Svalgaard, L., Vaquero, J. M., & Cliver, E. W. (2015). Revisiting the sunspot
number. In The solar activity cycle (pp. 35–103). New York: Springer.
where the wi are assumed Gaussian independent white noises, Crowley, T. J., & Unterman, M. B. (2013). Technical details concerning development
of a 1200-yr proxy index for global volcanism. Earth System Science Data, 5.
whose power spectra are the synthesis parameters of the
de Larminat, P. (2009). Automatique appliquée. Paris, Lavoisier: Wiley and Sons.
algorithm. de Larminat, P. (2014). Climate change, identification and projections London.
de Larminat, P., & Thomas, Y. (1977). Automatique des systèmes linéaires 2. Identifica-
tion, Flammarion, Paris.
Appendix B. Estimating the parametric uncertainty variance Delaygue, G., & Bard, E. (2011). An Antarctic view of Beryllium-10 and solar activity
for the past millennium. Climate Dynamics, 36(11–12), 2201–2218.
Let’s explicit the dependence of the observed output y of a pro- Faivre, R., Iooss, B., Mahévas, S., Makowski, D., & Monod, H. (2013). Analyse de
sensibilité et exploration de modèles: Application aux sciences de la nature et de
cess in relation to a vector θ of parameters by writing: l’environnement Editions Quae, Paris.
y = f ( θ ) + v,
Folland, C. K. (2013). High predictive skill of global surface temperature a year
ahead. Geophyshical Research Letters, 40, 761–767.
Haam, E., & Tung, K. K. (2012). Statistics of solar cycle-La Niña connection: Corre-
where: lation of two autocorrelated time series. Journal of the Atmospheric Sciences, 69,
2934–2939.
– y = [y1 yt yN ]T is the sequence of observed outputs, from Hasselmann, K. (1993). Optimal fingerprints for the detection of timedependent cli-
time t = 1 to t = N mate change. Journal of Climate, 6(10), 1957–1971.
124 P. de Larminat / Annual Reviews in Control 42 (2016) 114–125

Hegerl, G. C., Hoegh-Guldberg, O., Casassa, G., Hoerling, M. P., Kovats, R. S., Parme- Meehl, G. A., Arblaster, J. M., Matthes, K., Sassi, F., & van Loon, H. (2009). Amplifying
san, C., et al. (2010). Good practice guidance paper on detection and attribution the Pacific climate system response to a small 11-747year solar cycle forcing.
related to anthropogenic climate change. In Meeting report of the intergovern- Science, 325, 1114–1118.
mental panel on climate change expert meeting on detection and attribution of Meinshausen, M., Raper, S. C. B., & Wigley, T. M. L. (2011). Emulating coupled at-
anthropogenic climate change (p. 8). IPCC working group i technical support unit. mosphere-ocean and carbon cycle models with a simpler model, MAGICC6–Part
Bern, Switzerland: University of Bern. 1: Model description and calibration. Atmospheric Chemistry and Physics, 11(4),
Hegerl, G., & Zwiers, F. (2011). Use of models in detection and attribution of climate 1417–1456.
change. Wiley Interdisciplinary Reviews: Climate Change, 2(4), 570–591. Misios, S., & Schmidt, H. (2012). Mechanisms involved in the amplification of the
Hood, L. L., & Soukharev, R. E. (2012). The lower-stratospheric response to 11-yr 11-yr solar cycle signal in the Tropical Pacific ocean. Journal of Climate, 25,
solar forcing: Coupling to the troposphere-ocean response. Journal of the Atmo- 5102–5118.
sphere Sciences., 69, 1841–1864. Moberg, A. (2005). 2,000-Year Northern hemisphere temperature reconstruction.
Imbers, J., Lopez, A., Huntingford, C., & Allen, M. R. (2013). Testing the robustness of IGBP PAGES/World data center for paleoclimatology data contribution series #
the anthropogenic climate change detection statements using different empiri- 2005-019. NOAA/NGDC Paleoclimatology program, Boulder, CO.
cal models. Journal of Geophysical Research Atmospheres, 118(8), 3192–3199. Roy, I., & Haigh, J. D. (2010). Solar cycle signals in sea level pressure and sea surface
Kaufmann, R. K., Kauppi, H., Mann, M. L., & Stock, J. H. (2011). Reconciling an- temperature. Atmospheric Chemistry and Physics, 10, 3147–3153.
thropogenic climate change with observed temperature 1998–2008. Proceed- Roy, I., & Haigh, J. D. (2012). Solar cycle signals in the Pacific and the issue of tim-
ing of the National Academy of Sciences of the United States of America, 108, ings. Journal of the Atmospheric Sciences, 69(4), 1446–1451.
11790–11793. Särkkä, S. (2013). Bayesian filtering and smoothing (vol. 3). Cambridge University
Kopp, G., & Lean, J. L. (2011). A new, lower value of total solar irradiance: Evidence Press.
and climate significance. Geophysical Research Letters, 38(1). Shapiro, A. I., Schmutz, W., Rozanov, E., Schoell, M., Haberreiter, M., Shapiro, A. V.,
Landau, I. D. (2001). Identification des systèmes. Paris: Hermès science publications. & Nyeki, S. (2011). A new approach to the long-term reconstruction of the solar
Lean, J. L., & Rind, D. H. (2009). How will Earth’s surface temperature change in irradiance leads to large historical solar forcing. Astronomy & Astrophysics, 529,
future decades? Geophyshical Research Letters, 36, L15708. A67.
Lean, J. (2004). Solar irradiance reconstruction. IGBP PAGES/World data center for pa- Söderström, T., & Stoica, P. (1988). System identification. Prentice-Hall, Inc.
leoclimatology data contribution series # 2004-035. NOAA/NGDC Paleoclimatology Tung, K.-K., & Zhou, J. (2010). The Pacific’s response to surface heating in 130 yr
program, Boulder, CO. of SST: La Niña-like or El Niño-like? Journal of the Atmospheric Sciences, 67,
Ljung, L. (1999). System identification: Theory for the user. 2649–2657.
Ljungqvist, F. C. (2009). N. hemisphere extra-tropics 2,0 0 0 yr decadal temperature Usoskin, I. G., Korte, M., & Kovaltsov, G. A. (2008). Role of centennial geomagnetic
reconstruction. IGBP PAGES/World data center for paleoclimatology data contribu- changes in local atmospheric ionization. Geophysical Research Letters, 35, L05811.
tion series # 2010-089. NOAA/NCDC Paleoclimatology program, Boulder, CO. van Loon, H., & Meehl, G. A. (2008). The response in the Pacific to the sun’s decadal
Lockwood, M., & Froḧlich, C. (2008). Recent oppositely directed trends in solar cli- peaks and contrasts to cold events in the Southern Oscillation. Journal of Atmo-
mate forcings and the global mean surface air temperature: II. Different recon- spheric and Solar-Terrestrial Physics, 70, 1046–1055.
structions of the total solar irradiance variation and dependence on response van Loon, H., Meehl, G. A., & Shea, D. J. (2007). Coupled air-sea response to solar
time scale. Proceeding of the Royal Society London A, 464, 1367–1385. forcing in the Pacific region during northern winter. Journal of the Atmospheric
Loehle, C. (2007). A 2000-years global temperature reconstruction based on Sciences, 112, D02108.
non-treering proxies. Energy & Environment, 18(7 + 8). Walter, E., & Pronzato, L. (1997). Identification of parametric models from experimental
Mann, MichaelE., Bradley, RaymondS., & Hughes, MalcolmK. (1999). Northern hemi- data. Springer Verlag.
sphere temperatures during the past millennium : Inferences, uncertainties, and White, W. B., & Liu, Z. Y. (2008). Non-linear alignment of El Niño to the 11-yr solar
limitations. Geophysical Research Letters, 26, 759–762. cycle. Geophysical Research Letters, 35, L19607.
 P. de Larminat / Annual Reviews in Control 42 (2016) 114–125 125

Philippe de Larminat (Graduate Engineer, 1964, Ph. D., 1972). He was Professor at the Institut National des Sciences Appliquées (Rennes, France)
and École Centrale (Nantes). He is the author of 6 books and more than 100 papers in journals and international conferences. Since 2001, he is an
independent consultant and author of several patents (e.g. power plants control, satellite guidance).
His research interests include mathematical modeling, system identification, signal processing and control theory. Since 2012, he conducts a pioneer
work on identification of the Earth climate system.