Optimizing significance testing of astronomical forcing in cyclostratigraphy

David B. Kemp

Research output: Contribution to journalArticle

14 Citations (Scopus)
4 Downloads (Pure)

Abstract

The recognition of astronomically forced (Milankovitch) climate cycles in geological archives marked a major advance in Earth science, revealing a heartbeat within the climate system of general importance and key utility. Power spectral analysis is the primary tool used to facilitate identification of astronomical cycles in stratigraphic data, but commonly employed methods for testing the statistical significance of relatively high narrow-band variance of potential astronomical origin in spectra have been criticized for inadequately balancing the respective probabilities of type I (false positive) and type II (false negative) errors. This has led to suggestions that the importance of astronomical forcing in Earth history is overstated. It can be readily demonstrated, however, that the imperfect nature of the stratigraphic record and the quasiperiodicity of astronomical cycles sets an upper limit on the attainable significance of astronomical signals. Optimized significance testing is that which minimizes the combined probability of type I and type II errors. Numerical simulations of stratigraphically preserved astronomical signals suggest that optimum significance levels at which to reject a null hypothesis of no astronomical forcing are between 0.01 and 0.001 (i.e., 99–99.9% confidence level). This is lower than commonly employed in the literature (90–99% confidence levels). Nevertheless, in consonance with the emergent view from other scientific disciplines, fixed-value null hypothesis significance testing of power spectra is implicitly ill suited to demonstrating astronomical forcing, and the use of spectral analysis remains a difficult and subjective endeavor in the absence of additional supporting evidence.
Original languageEnglish
Pages (from-to)1516-1531
Number of pages16
JournalPaleoceanography
Volume31
Issue number12
Early online date5 Dec 2016
DOIs
Publication statusPublished - Dec 2016

Fingerprint

cyclostratigraphy
spectral analysis
climate cycle
hypothesis testing
geological record
Earth science
climate
history
simulation
method

Keywords

  • Milankovitch
  • spectral analysis
  • significance testing
  • sedimentation rate
  • cyclostratigraphy

Cite this

Optimizing significance testing of astronomical forcing in cyclostratigraphy. / Kemp, David B.

In: Paleoceanography, Vol. 31, No. 12, 12.2016, p. 1516-1531.

Research output: Contribution to journalArticle

Kemp, David B. / Optimizing significance testing of astronomical forcing in cyclostratigraphy. In: Paleoceanography. 2016 ; Vol. 31, No. 12. pp. 1516-1531.
@article{e477f6f9b93143fd86af0472faf11986,
title = "Optimizing significance testing of astronomical forcing in cyclostratigraphy",
abstract = "The recognition of astronomically forced (Milankovitch) climate cycles in geological archives marked a major advance in Earth science, revealing a heartbeat within the climate system of general importance and key utility. Power spectral analysis is the primary tool used to facilitate identification of astronomical cycles in stratigraphic data, but commonly employed methods for testing the statistical significance of relatively high narrow-band variance of potential astronomical origin in spectra have been criticized for inadequately balancing the respective probabilities of type I (false positive) and type II (false negative) errors. This has led to suggestions that the importance of astronomical forcing in Earth history is overstated. It can be readily demonstrated, however, that the imperfect nature of the stratigraphic record and the quasiperiodicity of astronomical cycles sets an upper limit on the attainable significance of astronomical signals. Optimized significance testing is that which minimizes the combined probability of type I and type II errors. Numerical simulations of stratigraphically preserved astronomical signals suggest that optimum significance levels at which to reject a null hypothesis of no astronomical forcing are between 0.01 and 0.001 (i.e., 99–99.9{\%} confidence level). This is lower than commonly employed in the literature (90–99{\%} confidence levels). Nevertheless, in consonance with the emergent view from other scientific disciplines, fixed-value null hypothesis significance testing of power spectra is implicitly ill suited to demonstrating astronomical forcing, and the use of spectral analysis remains a difficult and subjective endeavor in the absence of additional supporting evidence.",
keywords = "Milankovitch, spectral analysis, significance testing, sedimentation rate, cyclostratigraphy",
author = "Kemp, {David B.}",
year = "2016",
month = "12",
doi = "10.1002/2016PA002963",
language = "English",
volume = "31",
pages = "1516--1531",
journal = "Paleoceanography and Paleoclimatology",
issn = "0883-8305",
publisher = "American Geophysical Union",
number = "12",

}

TY - JOUR

T1 - Optimizing significance testing of astronomical forcing in cyclostratigraphy

AU - Kemp, David B.

PY - 2016/12

Y1 - 2016/12

N2 - The recognition of astronomically forced (Milankovitch) climate cycles in geological archives marked a major advance in Earth science, revealing a heartbeat within the climate system of general importance and key utility. Power spectral analysis is the primary tool used to facilitate identification of astronomical cycles in stratigraphic data, but commonly employed methods for testing the statistical significance of relatively high narrow-band variance of potential astronomical origin in spectra have been criticized for inadequately balancing the respective probabilities of type I (false positive) and type II (false negative) errors. This has led to suggestions that the importance of astronomical forcing in Earth history is overstated. It can be readily demonstrated, however, that the imperfect nature of the stratigraphic record and the quasiperiodicity of astronomical cycles sets an upper limit on the attainable significance of astronomical signals. Optimized significance testing is that which minimizes the combined probability of type I and type II errors. Numerical simulations of stratigraphically preserved astronomical signals suggest that optimum significance levels at which to reject a null hypothesis of no astronomical forcing are between 0.01 and 0.001 (i.e., 99–99.9% confidence level). This is lower than commonly employed in the literature (90–99% confidence levels). Nevertheless, in consonance with the emergent view from other scientific disciplines, fixed-value null hypothesis significance testing of power spectra is implicitly ill suited to demonstrating astronomical forcing, and the use of spectral analysis remains a difficult and subjective endeavor in the absence of additional supporting evidence.

AB - The recognition of astronomically forced (Milankovitch) climate cycles in geological archives marked a major advance in Earth science, revealing a heartbeat within the climate system of general importance and key utility. Power spectral analysis is the primary tool used to facilitate identification of astronomical cycles in stratigraphic data, but commonly employed methods for testing the statistical significance of relatively high narrow-band variance of potential astronomical origin in spectra have been criticized for inadequately balancing the respective probabilities of type I (false positive) and type II (false negative) errors. This has led to suggestions that the importance of astronomical forcing in Earth history is overstated. It can be readily demonstrated, however, that the imperfect nature of the stratigraphic record and the quasiperiodicity of astronomical cycles sets an upper limit on the attainable significance of astronomical signals. Optimized significance testing is that which minimizes the combined probability of type I and type II errors. Numerical simulations of stratigraphically preserved astronomical signals suggest that optimum significance levels at which to reject a null hypothesis of no astronomical forcing are between 0.01 and 0.001 (i.e., 99–99.9% confidence level). This is lower than commonly employed in the literature (90–99% confidence levels). Nevertheless, in consonance with the emergent view from other scientific disciplines, fixed-value null hypothesis significance testing of power spectra is implicitly ill suited to demonstrating astronomical forcing, and the use of spectral analysis remains a difficult and subjective endeavor in the absence of additional supporting evidence.

KW - Milankovitch

KW - spectral analysis

KW - significance testing

KW - sedimentation rate

KW - cyclostratigraphy

U2 - 10.1002/2016PA002963

DO - 10.1002/2016PA002963

M3 - Article

VL - 31

SP - 1516

EP - 1531

JO - Paleoceanography and Paleoclimatology

JF - Paleoceanography and Paleoclimatology

SN - 0883-8305

IS - 12

ER -