Annual Reviews in Control: Philippe de Larminat
Annual Reviews in Control: Philippe de Larminat
Annual Reviews in Control: Philippe de Larminat
Review
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
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
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
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.
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.