![Hier klicken, um den Treffer aus der Auswahl zu entfernen](images/unchecked.gif) |
Titel |
Dimensionality reduction and network inference for sea surface temperature
data |
VerfasserIn |
Fabrizio Falasca, Annalisa Bracco, Athanasios Nenes, Constantine Dovrolis, Ilias Fountalis |
Konferenz |
EGU General Assembly 2017
|
Medientyp |
Artikel
|
Sprache |
en
|
Digitales Dokument |
PDF |
Erschienen |
In: GRA - Volume 19 (2017) |
Datensatznummer |
250142076
|
Publikation (Nr.) |
EGU/EGU2017-5651.pdf |
|
|
|
Zusammenfassung |
Earth’s climate is a complex dynamical system. The underlying components of the system
interact with each other (in a linear or non linear way) on several spatial and time scales.
Network science provides a set of tools to study the structure and dynamics of such systems.
Here we propose an application of a novel network inference method, δ-MAPS, to investigate
sea surface temperature (SST) fields in reanalyses and models. δ-MAPS first identifies the
underlying components (domains) of the system, modeling them as spatially contiguous,
potentially overlapping regions of highly correlated temporal activity, and then infers the
weighted and potentially lagged interactions between them. The SST network is represented
as a weighted and directed graph. Edge direction captures the temporal ordering of
events, while edge weights capture the magnitude of the interaction between the
domains.
We focus on two reanalysis datasets (HadISST and COBE ) and on a dozen of runs of the
CESM model (extracted from the so-called large ensemble). The networks are built using 45
years of data every 3 years for the total dataset temporal coverage (from 1871 to
2015 for HadISST, from 1891 to 2015 for COBE and from 1920 to 2100 for CESM
members).
We then explore similarities and differences between reanalyses and models in terms of
the domains identified, the networks inferred and their time evolution.
The spatial extent and shape of the identified domains is consistent between observations
and models.
According to our analysis the largest SST domain always corresponds to the El Niño
Southern Oscillation (ENSO) while most of the other domains correspond to known climate
modes. However, the network structure shows significant differences.
For example, the unique role played by the South Tropical Atlantic in the observed
network is not captured by any model run.
Regarding the time evolution of the system we focus on the strength of ENSO: while we
observe a positive trend for observations and model members in the second half of the 20th
century and the first decade of the 21st century, the trajectory of different ensemble members
differs significantly in the projections.
Fully coupled climate models represent a primary tool for predicting the evolution of the
climate system under increasing anthropogenic emissions: our findings help identifying
important limitations in the models’ ability to reproduce historical climate records and
therefore predict future climate. |
|
|
|
|
|