 |
| Titel |
Generating geomorphological catalogues using neural networks: Seamounts in the Atlantic Ocean |
| VerfasserIn |
Andrew Valentine, Lara Kalnins, Chantal van Dinther, Jeannot Trampert |
| Konferenz |
EGU General Assembly 2013
|
| Medientyp |
Artikel
|
| Sprache |
Englisch
|
| Digitales Dokument |
PDF |
| Erschienen |
In: GRA - Volume 15 (2013) |
| Datensatznummer |
250080389
|
|
|
|
|
|
| Zusammenfassung |
| We recently introduced the idea that neural networks may be used to construct catalogues of
geomorphological features, by extrapolating from the characteristics of a set of
hand-selected examples (Valentine et al., 2012). These learning algorithms are
inspired by the complex pattern identification and recognition capabilities of the
human brain and remove the need to develop an a priori model of the feature of
interest.
In order to demonstrate this approach, and to develop a clearer understanding of its
possibilities and pitfalls, we concentrate on the problem of identifying seamounts — isolated
topographic highs of volcanic origin — in the world’s oceans. The distribution
of seamounts in time and space can provide important constraints on the tectonic
history and evolution of the Earth and has been studied using several conventional
approaches (e.g. Kim & Wessel, 2011). However, these typically perform poorly in the
Atlantic, where the slow spreading rate results in a rough ‘background’ seafloor that
produces many false positives. The learning algorithm approach should improve this, as
it attempts to encapsulate more complex information about the seamount and its
surroundings.
We present an overview of our work to date, with a focus on results from a systematic
search for seamounts in the Atlantic. We compare the performance of our approach in
detecting seamounts in bathymetric, free-air gravity anomaly and vertical gravity gradient
(VGG) datasets to examine the particular strengths and weaknesses of each data
type and to assess the potential benefits of assimilating information from two or
three data types simultaneously. We compare the resulting seamount database with
existing catalogues, examining the variations in measures such as total count, height
distribution, and spatial and temporal distribution across the Atlantic, and comment
on the potential implications for our understanding of the tectonic history of the
region.
Kim, S.-S. & Wessel, P., 2011. New global seamount census from altimetry-derived gravity data, Geophysical
Journal International, 186, pp.615–631.
Valentine, A., Kalnins, L. & Trampert, J., 2012. Hunting for seamounts using neural networks: learning
algorithms for geomorphic studies, EGU General Assembly, Abstract EGU2012-4560.
Valentine, A. & Trampert, J., 2012. Data-space reduction, quality assessment and searching of seismograms:
Autoencoder networks for waveform data. Geophysical Journal International, 189, pp.1183–1201. |
|
|
|
|
|
|