21st Conference on Climate Variability and Change
Seventh Conference on Artificial Intelligence and its Applications to the Environmental Sciences

J2.3

Major Influences of Circulation Patterns on Temperatures in the Italian Side of the Greater Alpine Region: an Investigation via Neural Network Modeling

Antonello Pasini, CNR, Rome, Italy; and R. Langone

Neural networks (NNs) have been shown their ability in attribution studies at global scale: they are able to attribute the recent global warming mainly to anthropogenic forcings and to reconstruct the global temperatures of the last 140 years in a very satisfying manner. As well known, however, when passing from the global scale to a regional level, climate natural variability masks any direct link between global forcings and regional temperatures, so that the influence of circulation patterns is the key element for “attributing” regional climate.

In this framework, adopting the HISTALP database and concentrating on the time series related to its SW region, we study the influence of several circulation patterns (North Atlantic Oscillation, East Atlantic pattern, Arctic Oscillation, Scandinavian pattern, East Atlantic/West Russian pattern, Atlantic Blocking Index, European Blocking Index, El Niño Southern Oscillation) on annual and seasonal temperatures in this region.

We choose a NN model (endowed with different transfer functions) previously developed and applied to environmental studies, that has already shown its ability in finding nonlinear relationships in the complex climate system. Here, through a particular leave-one-out training/validation procedure (due to the quite short time series analyzed - 50 years), NNs are able to recognize the major roles of certain circulation patterns and to reconstruct temperatures better than multilinear regressions. In doing so, we find a difference in the “complexity” of the influences in different seasons, too.

This kind of studies is very useful for establishing major influences on regional temperatures and, therefore, makes available a crucial information for a downscaling activity. In fact, in doing so, we have to choose those GCMs which are able to reconstruct the behavior of these patterns. Only in this manner we could achieve a correct downscaling for successfully reconstructing climate at regional level and possibly supplying reliable future scenarios at this scale.

extended abstract  Extended Abstract (240K)

wrf recording  Recorded presentation

Joint Session 2, Applications of artificial learning techniques in climate variability, especially as it relates to the urban environment
Wednesday, 14 January 2009, 8:30 AM-10:00 AM, Room 125A

Previous paper  Next paper

Browse or search entire meeting

AMS Home Page