Nature - 15.08.2019

(Barré) #1

Letter reSeArCH


These peaks broadly correspond to peaks in the global distribution
of igneous and detrital zircon U–Pb ages^16 –^18 and, similarly to those
studies, we interpret the peaks to reflect cyclicity^4. To provide more
statistically robust (larger number of data points per calculation) char-
acterization of secular changes in the distributions of metamorphic
T/P from one cycle to the next, we binned the data about each of these
peaks, as shown in Extended Data Fig. 1. Histograms and kernel den-
sity estimates (KDE) of each distribution are shown in Fig. 2a. With
decreasing age: (1) the KDE of >2.2-Gyr-old metamorphism is narrow
and symmetric; (2) the KDE of 2.2–1.4-Gyr-old metamorphism is
skewed towards lower T/P, with three low-T/P eclogite outliers^11 ,^12 ,^19 ;
(3) the KDE of 1.4–0.85-Gyr-old metamorphism shows a distinct
low-T/P mode that is not apparent in the >1.4-Gyr-old KDEs;
(4) the KDE of 0.85–0.2-Gyr-old metamorphism is notably broader
than the older distributions, with a prominent mode at lower T/P; and
(5) the KDE of <0.2-Gyr-old (‘modern’) metamorphism is bimodal, with
distinct high-T/P and low-T/P peaks that are much more pronounced
than for the older distributions. Qualitatively, the secular evolution
in the KDEs is interpreted to represent a gradual transition from a
narrow, unimodal distribution of metamorphic T/P in the Archaean
eon, to a distinctly bimodal distribution in the modern metamorphic
rock record.
To assess this interpretation quantitatively, we fitted a Gaussian
mixing model to each distribution (Fig. 2b). First, the distributions
were assessed by whether they are non-Gaussian (that is, do not


represent a single Gaussian distribution) at a 95% confidence inter-
val, according to the Shapiro–Wilk test^20. If the data distributions were
assessed to be non-Gaussian, we fitted a bimodal mixed-Gaussian dis-
tribution (that is, two Gaussian distributions; Fig. 2b, Extended Data
Table 1) to evaluate how these compare to the bimodal distribution of
the modern metamorphic rock record (Figs.  1 , 2a). The distribution
of data older than 2. 2  Gyr is Gaussian with a mean and standard devi-
ation of 2.96 and 0.11 (for log[T/P (°C GPa−^1 )]), respectively. By con-
trast, each of the younger (<2.2 Gyr old) distributions is non-Gaussian
but is well described by a bimodal mixed-Gaussian distribution. The
difference between the two best-fit Gaussian distributions (‘low-T/P’
and ‘high-T/P’ in Fig. 2b) increases through time (Fig. 2c). Viewed
together, the KDEs and modelled mixed-Gaussian distributions show
a continuous increase in the variability of thermal gradients recorded
by metamorphic rocks, as well as a gradual emergence of a discrete
and prominent low-T/P mode of metamorphism since the end of the
Archaean eon.
The Palaeoproterozoic era (about 2  Gyr ago) is notable in that it
has three distinct outliers with low-T/P gradients comparable to
those of rocks formed in modern cold collisional environments^11 ,^12 ,^19
(Fig.  3 ). These data have been used to suggest that subduction and col-
lision similar to those of modern Earth may have been operative dur-
ing the Palaeoproterozoic era^11 ,^12 ,^19. Alternatively, the data may reflect
local, anomalous subduction–collision similar to that in modern
tectonic environments, but not representative of the dominant

Low T/P High T/P Low T/P High T/P

Gaussian
Mean = 2.96
s.d. = 0.11

0 Gyr

0.2

0.4

0.6

0.8

1.0

1.2

1.4

1.6

1.8

2.0

2.2

2.4

2.6

0.2–0
Gyr

0.85–0.2
Gyr

1.4–0.85 Gyr

2.2–1.4 Gyr

>2.2 Gyr

log[T/P (°C GPa–1)] log[T/P (°C GPa–1)]

abc

1.8 3.6

Low T/PHigh T/P

log[T/P (°C GPa–1)]

1.8 3.6 2.^23 .4

Non-Gaussian

Palaeoproterozoic

Mesoproterozoic

Neoproterozoic Palaeozoic

Mesozoic

Cenozoic

Neoarchaean

Fig. 2 | The bimodal distribution of modern metamorphism evolved
gradually since the end of the Neoarchaean era. a, Histograms and
KDEs of metamorphic T/P since the Neoarchaean era. b, Fits o f the
distributions of metamorphic T/P shown in a with bimodal Gaussian
mixing models. Global metamorphism >2.2 Gyr ago can be fitted by a
unimodal Gaussian distribution (95% confidence interval, n = 72,
p value of 0.23); metamorphism ≤2.2 Gyr ago is non-Gaussian


(95% confidence intervals; 2.2–1.4 Gyr ago, n = 106, p = 5.4 ×  10 −^5 ; 1.4–
0.85 Gyr ago, n = 45, p = 0.017; 0.85–0.2 Gyr ago, n = 232, p = 3.6 ×  10 −^7 ;
0.2–0 Gyr ago, n = 109, p = 2.2 ×  10 −^7 ), but is well described by bimodal
mixed-Gaussian distributions. c, Linear regressions of low-T/P and high-
T/P modes (from b) with 95% confidence envelopes. The best-fit bimodal
distributions of metamorphic T/P become increasingly distinct, showing
divergence and cooling of the low-T/P and high-T/P modes through time.

15 AUGUSt 2019 | VOL 572 | NAtUre | 379
Free download pdf