Beruflich Dokumente
Kultur Dokumente
S0167-2789(15)30164-0
http://dx.doi.org/10.1016/j.physd.2016.11.007
PHYSD 31867
To appear in:
Physica D
Abstract
Motivated by an important geophysical significance, we consider the influence
of stochastic forcing on a simple three-dimensional climate model previously
derived by Saltzman and Sutera. A nonlinear dynamical system governing three physical variables, the bulk ocean temperature, continental and
marine ice masses, is analyzed in deterministic and stochastic cases. It is
shown that the attractor of deterministic model is either a stable equilibrium or a limit cycle. We demonstrate that the process of continental ice
melting occurs with a noise-dependent time delay as compared with marine ice melting. The paleoclimate cyclicity which is near 100 ky in a wide
range of model parameters abruptly increases in the vicinity of a bifurcation point and depends on the noise intensity. In a zone of stable equilibria,
the 3D climate model under consideration is extremely excitable. Even for a
weak random noise, the stochastic trajectories demonstrate a transition from
small- to large-amplitude stochastic oscillations (SLASO). In a zone of stable
cycles, SLASO transitions are analyzed too. We show that such stochastic
transitions play an important role in the formation of a mixed-mode paleoclimate scenario. This mixed-mode dynamics with the intermittency of
large- and small-amplitude stochastic oscillations and coherence resonance
are investigated via analysis of interspike intervals. A tendency of dynamic
paleoclimate to abrupt and rapid glaciations and deglaciations as well as its
transition from order to chaos with increasing noise are shown.
November 2, 2016
(1)
(2)
a3 c1 y a1 (1 a2 y)y
c0 x(y) + c1 y
, z(y) =
.
a0 a3 c0
Substituting these functions into the second equation (1), we get g(y) = 0,
where
g(y) = b0 x(y) + b1 (1 b2 y b3 y 2 b4 x2 (y))y b5 z(y).
In Fig. 1, plots of the modified function g(v) are shown in terms of the
scaled variable v from (2) for three values of the parameter , which represents one of the main process parameters - the reverse relaxation time of the
bulk ocean temperature to its mean value.
5
13
x 10
g
0
10
15
20
0.1
0.2
0.3
0.4
0.5
Figure 1: Plots of g(v) for various : = 0.1 104 y1 (blue), = = 0.1903 104
y1 (red), and = 0.3 104 y1 (black). The rest system parameters are [8, 13]: a0 = 0,
a1 = 0.145 y1 , a2 = 1.86 1017 kg1 , a3 = 1.265 1014 y1 , b0 = 0.276 106 y1 ,
b1 = 3.77 104 y1 , b2 = 1.58 1018 kg1 , b3 = 3.77 1034 kg2 , b4 = 0.7152 1038
kg2 , b5 = 0.697 1013 y1 , c0 = 0.792 1023 y1 , c1 = 28.65 1023 y1 .
Note that in the whole zone under consideration, the system (1) possesses
the trivial unstable equilibrium M0 (0, 0, 0). For < = 0.1903 104 y1 ,
the function g(v) has two positive roots, and consequently, the system has two
non-trivial equilibria M1 (
u1 , v1 , w1 ) and M2 (
u2 , v2 , w2 ), where u2 < u1 < 0.
Moreover, the first equilibrium M1 is unstable whereas the second one, M2 ,
is stable.
With increase of the parameter , two positive roots of the function g(v)
approach and merge at the point = . For > , the function g(v) does
not have positive roots, and the point M0 is a single equilibrium. The value
= represents the bifurcation point, where an annihilation of M1 and
M2 is accompanied by the birth of a limit cycle. Note that this stable cycle
represents a single attractor for > .
Fig. 2 shows extrema of u- and v-coordinates of attractors and repellers
for the deterministic system versus . Here, the stable limit cycles, unstable
and stable equilibria, are shown by the (black) solid, (red) dashed and (blue)
solid lines, respectively. As is easy to see, the bifurcation point divides
the stable climate states describing by stable equilibria or limit cycles. Thus,
the system (1) exhibits a steady-state regime with the equilibrium M2 in the
6
interval < (blue line), and a self-oscillatory regime with the stable limit
cycle for > (black lines).
In the case of stationary regime with = 0.18 104 y1 , some peculiarities of the phase portrait are shown in Figure 3. Note that the unstable
equilibrium M1 belongs to the separating surface which detaches the long
and short transients of phase trajectories to the stable equilibrium M2 . In
other words, if we start from a point lying above M1 , we have got a long
excursion (going through the upper part) to M2 . In the opposite case, when
we start from a point lying below M1 , we shortly reach the stable equilibrium
M2 .
In the zone of self-oscillations, a dependence of the spatial arrangement
of limit cycles on the parameter is illustrated in Fig. 4. An important
point is that the dispersions of all parameters shown in Fig. 4 increase with
decreasing the operating parameter (with increasing the relaxation time
of the bulk ocean temperature to its mean value). To say this another way,
small relaxation times lead to narrower cycles and smaller deviations of the
ocean temperature and ice masses from their equilibrium values. Note that
in the zone of cycles, the amplitudes of oscillations of u- and v-coordinates
vary slightly while the amplitude of w-coordinate decreases more than twice.
The time series of oscillations are plotted in Figure 5. It is evident that
oscillations of the ocean temperature and marine ice mass occur nearly in
opposite phases. An important point is that the continental ice melting
happens with time delays in comparison with the marine ice melting with
increase in the ocean temperature perhaps due to the time-lag effect. In
addition, the onset of continental ice melting occurs only a short time before
the highest temperature in the ocean attains. Beyond that point, the melting
of continental ice takes place abruptly for a broad range of whereas the
rate of marine ice melting decelerates in the vicinity of zero.
The period T of these oscillations as a function of the parameter is
shown in Figure 6. Note that T unlimitedly increases as the parameter
approaches to the bifurcation value . It is of prime importance that the
period is of the order of 102 ky for a broad range of . This is in agreement
with a number of previous studies [6, 7, 8, 41, 42].
3. Stochastic analysis
In order to study an influence of possible unpredictable random fluctuations in the ocean temperature dynamics let us introduce the stochastic
7
term in the third equation (1). This equation is most sensitive to stochastic
forcing in comparison with the rest equations governing the continental and
marine ice masses. As noted by Saltzman and Sutera in their preliminary
study [13], the system solution in this case is irregular and structurally stable
in the vicinity of deterministic solution. So, in order to study this behaviour
in greater detail, we have
x = a0 x + a1 (1 a2 y)y a3 z
y = b0 x + b1 (1 b2 y b3 y 2 b4 x2 )y b5 z
z = c0 x + c1 y z + (t).
(3)
which appears at > (see the limit cycle for = 0.191 104 y1 plotted
by the blue thick curves in Fig. 8).
Consider now how these peculiarities of the deterministic phase portrait clarify the probabilistic mechanism of LASO generation. In Fig. 9,
the deterministic (black) and stochastic (green) trajectories are plotted for
= 0.18 104 and = 0.19 104 y1 . The possible scenaria of SASO and
noise-induced LASO are shown in the left and right panels, respectively.
Figures 9(a,c) (SASO regime) demonstrate that the random trajectories
starting from the stable equilibrium M2 do not cross the separatrix that
passes through the unstable equilibrium M1 . So, for a weak noise, these
trajectories are localized near the stable equilibrium at any time. As noise
increases, the random trajectories begin to intersect the separatrix that leads
to the formation of repeating large-amplitude excursions (see the green curves
in Figures 9(b,d)). Analyzing the corresponding time series (see Fig. 7)
one can conclude that increasing noise generates a mixed-mode stochastic
process with intermittency of SASO and LASO. At first, LASO look like
rare spikes separating the long time intervals with SASO. Note that the
interspike intervals decrease with increasing noise.
Let sequences of random spike times, tk , of stochastic model are fixed
as moments of threshold crossings, u = 0. The corresponding sequence of
interspike intervals (ISI) is k = tk tk1 . We quantify the intermittency
of SASO and LASO by means of value m = E of ISI and dispersion D =
E( m)2 . In Fig. 10, the functions m() and D() are shown for various
values of the parameter . As one can see, due to the effect of stochastic
forcing at = 0.18 104 y1 , the mean value of ISI decreases to the value
2 105 y that is close to the period of deterministic oscillations in the region
> . For = 0.1 104 y1 , that is far from the bifurcation point as
a result of noise, the mean value of ISI increases up to 1 106 y. In other
words, if the present-day climate system is represented by near 100 ky ice
age cycles, it is probably located on the left or right side of the bifurcation
point = whereas its deviations from these cycles occurring previously
(in Early and Middle Pleistocene [8]) can be explained by changes in and
approaching to the system bifurcation point.
As one can see from Fig. 10, the function D() has a pronounced minimum. This means that a phenomenon of coherence resonance [43] occurs in
the presence of noise-induced generation of LASO.
Consider now how the noise governs a spatial configuration of LASOs. In
Fig. 11, the time series of all scaled deviations are plotted along with their
9
corresponding mean values. It is worth noting that the mean values (dashed
lines) of all functions decrease with increasing noise. In other words, the
presence of noise (external forcing) leads to some irregularities in melting and
freezing of continental and marine ice. However, the downward deviations
(black curves in figures 11(a,b)) are much more than the upward ones. The
effect of v-amplitude jittering in the vicinity of zero demonstrates a local
stabilization of marine ice mass near its equilibrium value with v = 0. From
the physical point of view Fig. 11 shows the possibility of great climate
deviations from its equilibrium.
Some geometrical details of the beat-type climate dynamics are shown
in Figs. 12(a,b). If we fix a temperature deviation (w u panel) one can
see a broad range of u-deviations (predominantly directed to the negative
side) with a rise in noise. Figure 13 illustrates that the mean value of the
continental ice decreases with increasing the noise intensity. Also, Fig. 12(a)
demonstrates that v is localized in the vicinity of zero when u substantially
changes in the wide interval 4 . u . 1. It is significant that these deviations, illustrated in Figs. 12(a,b) by the horizontal tails directed to the left,
are shifted in the direction of ice mass decreasing (u < 0). This behaviour
demonstrates a tendency to global warming (the continental ice decreases).
An interesting point is that there is a rather narrow domain of u, 1 . u . 0,
where w undergoes essential deviations 20 . w . 0 from the equilibrium
point w = 0 and v is positive (two saturated tails consistent of random
trajectories in Figs. 12(a,b)). These tails show a potential tendency of the
climate system to its rapid glaciations within a broad range of temperature
deviations w in the presence of stochastic forcing. Such a shift of random
states with increasing noise can be clearly seen in the u-variable pdf plot (see
Fig. 12(c)).
3.2. Stochastic phenomena in a zone of cycles: generation of SASO, randomization of interspike intervals
Now we study an influence of noise in the zone > , where the deterministic system possesses the stable limit cycle. Consider how the noise
affects the deterministic periodic oscillations near the bifurcation point .
In Fig. 14, the time series of u(t) are plotted for = 0.2 104 y1 , = 0,
= 0.005, and = 0.02. An important point is that the deterministic spikes
are randomly shifted and interspike intervals of slow monotonic growth of
u(t) are replaced by stochastic oscillations. Moreover, the effect of stochastic
forcing leads to the appearance of abrupt random changes in the continental
10
ice mass (abrupt climate glaciations and deglaciations as in the aforementioned case of stochastic forcing in the zone of equilibria).
Under the noise influence, the monotonic deterministic dynamics within
interspike intervals transforms into SASO. Along with the generation of
SASO, a decrease of interspike intervals is observed. The mean values and
dispersions of ISIs versus noise intensity are plotted in Fig. 15. Again, this
figure confirms that near 100 ky ice age cycles exist in a broad range of noise
intensities. As one can see, due to the stochastic forcing for = 0.2104 y1 ,
the mean value of ISIs decreases and stabilizes in the zone 0.02 < < 0.04
near the value 2 105 y that is close to the period of deterministic oscillations
in the wide zone . Further increase of noise leads to the growth of mean
values of ISIs. Note that for = 1 104 y1 , that is far from the bifurcation
point , the mean value of ISIs monotonically grows with increasing noise.
Consider now how the noise deforms the spatial configuration of the
stochastically forced cycles. Some geometrical details of this deformation
are shown in Figs. 16(a,b). Note that this deformation is similar to the case
of = 0.18 104 y1 considered above. In addition to the description of
Fig. 12(a), here a broader range of u-deviations (5 . u . 1) exists for a
fixed (close to zero) value of w. A shift of the u-coordinate pdf to the left
(in section w = 0) is clearly seen in Fig. 16(c).
3.3. Noise-induced chaotization
In the analysis of nonlinear stochastic oscillations, along with the study
of frequency and spatial properties of the solutions, the study of dynamics
of the mutual arrangement of random trajectories in stochastic flows plays
an important role. For this study, the Lyapunov exponents are traditionally
used [44]. The Lyapunov exponent depends on the noise intensity. As is
seen from Fig. 17, at first, () decreases, after that it begins to increase
and changes its sign from minus to plus. By this is meant that the climate
system transits from order to chaos. It should be noted that the smaller the
value of parameter , the smaller the noise value at which a transition to
chaos occurs.
A transition from order to chaos plays an important role in analyzing the
possibilities of predictive accuracy of climate changes. In a zone of chaos,
the interval of reliable forecast is significantly shortened.
11
4. Conclusion
Let us summarize the main conclusions of our deterministic and stochastic
analyses of a simple three-dimensional climate model previously derived in
[13].
The deterministic system exhibits two basic dynamic regimes: (i) stable
stationary state or (ii) periodic self-oscillations. In (i), a mutual arrangement
of the stable equilibrium M2 and unstable equilibrium M1 define the system
dynamics. The bifurcation point separates regimes (i) and (ii). At this
point, equilibria M1 and M2 merge, annihilate, and the stable cycle appears.
Geometrically, in regime (i), the point of unstable equilibrium M1 belongs
to the separatrix which divides the whole domain of phase trajectories into
two regions with the long or short time excursions of a phase trajectory to
the stable equilibrium M2 . Besides, the larger relaxation times of the bulk
ocean temperature to its mean value are described by broader cycles and
correspond to greater deviations of the ocean temperature and ice masses
and .
It is shown that the mean values of all departures from equilibrium values
and decrease when the noise intensity increases. As this takes place,
,
stochastic forcing induces essential negative deviations in the continental and
marine ice masses which are vastly greater than their positive deviations.
Moreover, a local stabilization of the marine ice mass in the vicinity of its
equilibrium is illustrated. One of the main results of our model calculations
is that the climate system has a potential possibility to abrupt glaciations
and deglaciations with increasing the noise intensity (stochastic forcing).
Finally, in the present paper, on the base of Lyapunov exponents, it
is shown that noise-induced intermittency of large- and small-amplitude
stochastic oscillations causes a transition from order to chaos. In the climate models, such noise-induced chaotization signals that the time interval
of predictability of the climate changes is shortened.
This article, continuing the research work on stochastic phenomena carried out earlier for a two-dimensional climatic model, expands our understanding of the possible mechanisms of generation of complex mixed-mode
oscillations in higher dimension models. In addition to the aforementioned
conclusions, let us especially emphasize the following general implications of
the results coming from the present mathematical and computational analyses. It is demonstrated that (i) there is a possibility of great climate deviations from its equilibrium, (ii) near 100 ky ice age cycles exist in a broad
range of noise intensities, (iii) there is a potential tendency of the climate
system to its rapid glaciations in the presence of stochastic forcing, and (iv)
the closer the system to its bifurcation point the smaller noise leads to a
transition from order to chaos.
Acknowledgements
This work was supported by the Ministry of Education and Science of the
Russian Federation (project no. 1.849.2017). We are grateful to M. Crucifix
for his thoughtful comments which have helped in improving the manuscript.
References
References
[1] N. Zeng, Quasi-100 ky glacial-interglacial cycles triggered by subglacial
burial carbon release, Clim. Past 3 (2007) 135-153.
13
[2] C.L. Prescott, A.M. Haywood, A.M. Dolan, S.J. Hunter, J.O. Pope,
S.J. Pickering, Assessing orbitally-forced interglacial climate variability
during the mid-Pliocene Warm Period, Earth and Planetary Science
Letters 400 (2014) 261-271.
[3] Y. Ashkenazy, D.R. Baker, H. Gildor, Simple stochastic models for
glacial dynamics, J. Geophys. Res. 110 (2005) C02005.
[4] G. Matteucci, Orbital forcing in a stochastic resonance model of the
Late-Pleistocene climatic variations, Climate Dynamics 3 (1989) 179190.
[5] C. Wunsch, Quantitative estimate of the Milankovitch-forced contribution to observed Quaternary climate change, Quaternary Science Reviews 23 (2004) 1001-1012.
[6] A.J. Ridgwell, A.J. Watso, Is the spectral signature of the 100 kyr glacial
cycle consistent with a Milankovitch origin?, Paleoclimatology 14 (1999)
437-440.
[7] J.R. Petit, J. Jouzel, D. Raynaud, N.I. Barkov, J.M. Barnola, I. Basile,
M. Bender, J. Chappellaz, M. Davis, G. Delaygue, M. Delmotte, V.M.
Kotlyakov, M. Legrand, V.Y. Lipenkov, C. Lorius, L. Pepin, C. Ritz, E.
Saltzman, M. Stievenard, Climate and atmospheric history of the past
420,000 years from the Vostok ice core, Antarctica, Nature 399 (1999)
429-436.
[8] B. Saltzman, Dynamical Paleoclimatology, San Diego, 2002.
[9] T.S. Ledley, S. Chu, The initiation of ice sheet growth, Milankovitch solar radiation variations, and the 100 ky ice age cycle, Climate Dynamics
11 (1995) 439-445.
[10] J.W. Crowley, R.F. Katz, P. Huybers, C.H. Langmuir, S.-H. Park,
Glacial cycles drive variations in the production of oceanic crust, Science
347 (2015) 1237-1240.
[11] B.S. Cramer, J.D. Wright, D.V. Kent, M.-P. Aubry, Orbital climate
forcing of 13 C excursions in the late Paleocene-early Eocene (chrons
C24nC25n), Paleoceanography 18 (2003) 1097.
14
15
16
17
u
1
10
10
10
0.5
0.5
10
18
w
M0
M1
M2
Figure 3: Projections of the phase trajectories on the u w plane for = 0.18 104 y1 .
The stable equilibrium M2 is shown by the black circle whereas the unstable equilibria M0
and M1 are indicated by the open circles.
19
w
0
=0.191 104
=0.5 104
=1 104
1
=1.5 104
0.5
=0.191 104
=0.5 104
=1 104
=1.5 104
0.5
w
0.5
0.5
=0.191 104
=0.5 104
1
=1 104
4
=1.5 10
0.5
0.5
20
1
continental ice
marine ice
ocean temperature
continental ice
10
t, years
a)
x 10
ocean temperature
10
t, years
b)
x 10
1
continental ice
marine ice
ocean temperature
continental ice
1
0
marine ice
marine ice
ocean temperature
0.5
1.5
2.5
3.5
t, years
c)
4.5
1
0
5
5
x 10
0.5
1.5
2.5
t, years
d)
3.5
4.5
5
5
x 10
Figure 5: The time series for a) = 0.2 104 y1 , b) = 0.5 104 y1 , c) = 1 104
y1 , and d) = 1.5 104 y1 .
x 10
T
9
8
7
6
5
4
3
2
1
0.2
0.4
0.6
0.8
1.2
1.4
1.6
1.8
21
0.5
0.5
1.5
0
t x 1010
6
a)
0.5
0.5
1.5
0
t x 1010
6
b)
Figure 7: u-time series for the 3D Saltzman-Sutera model: a) with = 0.18 104 y1 and
= 0.001 (green), = 0.003 (blue), = 0.01 (red); b) = 0.19 104 y1 and = 0.0001
(green), = 0.001 (blue), = 0.01 (red). A noise-induced generation of LASO-oscillations
when the system has the only stable equilibrium.
22
w
0
M1
1
M2
1
a)
w
0
M1
M2
1
b)
Figure 8: uw-projections of the phase trajectories (black thin lines) for the 3D SaltzmanSutera deterministic model with a) = 0.18 104 y1 and b) = 0.19 104 y1 . The
stable equilibrium M2 is shown by the black circle and unstable equilibrium M1 is plotted
by the red circle. The blue thick closed curves are the u w-projections of the limit cycle
for = 0.191 104 y1 .
23
0
M1
M1
2
1
a)
b)
M1
M1
M2
M2
1
c)
d)
13
10
12
10
10
11
10
10
0.02
0.04
0.06
0.08
10
10
0.02
0.04
0.06
0.08
Figure 10: The mean value m and dispersion D of interspike intervals for = 0.1 104
y1 (blue) and = 0.18 104 y1 (red) versus noise intensity.
24
2
0
x 10
v
0.4
0.2
0.2
0.4
x 10
10
0
x 10
Figure 11: Time series of u, v and w for = 0.18 104 y1 , = 0 (red), = 0.02 (blue),
and = 0.05 (black). Their mean values are shown by the dashed lines of the same color.
25
v
0.5
=0.1
=0.05
=0.01
equilibrium
0.5
5
a)
w
0
10
=0.1
=0.05
=0.01
equilibrium
20
b)
=0.01
=0.05
=0.1
0
4
c)
Figure 12: Projections of the random trajectories for system (3) shown in the dimensionless
variables (2) on the phase planes w = 0 (a), v = 0 (b) and pdf of the u-coordinate (c) for
= 0.18 104 y1 .
26
u
0.4
0.6
0.8
1
0
0.02
0.04
0.06
0.08
Figure 13: The mean values of u-coordinate for = 0.1104 y1 (blue) and = 0.18104
y1 (red).
27
1
0
a)
x 10
u
1
b)
x 10
Figure 14: The time series (blue) for = 0.2 104 y1 with a) = 0.005, and b) = 0.02
(deterministic time series are plotted by the red color).
28
12
10
6
10
10
10
5
10
10
10
0.02
0.04
0.06
0.08
10
0.02
0.04
0.06
0.08
Figure 15: The mean value m and dispersion D of interspike intervals for = 0.2 104
y1 (red) and = 1 104 y1 (blue) versus noise intensity.
29
v
=0.1
=0.05
=0.01
cycle
0.5
0.5
6
a)
w
10
=0.1
=0.05
=0.01
cycle
10
20
b)
u
=0.01
=0.05
=0.1
0
4
c)
Figure 16: The random trajectories (a,b) and pdf (c) of the u-coordinates of intersection
points with w = 0 for = 0.2 104 y1 .
30
x 10
2
=0.1 104
=0.2 104
=0.3 104
=1 104
0.02
0.04
0.06
0.08
Figure 17: Chaotization. The largest Lyapunov exponent for = 0.1 104 y1 (red),
= 0.2 104 y1 (green), = 0.3 104 y1 (black), and = 1 104 y1 (blue).
31
Highlights
The article deals with a problem of excitability, mixed-mode oscillations and chaotization in a
three dimensional climate model.
A limit cycle and stable equilibria represent the attractors of deterministic model.
The formation of a mixed-mode regime of stochastic oscillations is revealed.
A system transition from order to chaos with increasing the noise intensity is found.