Learning about Interacting Networks in Climate: the LINC...

1
Learning about Interacting Networks in Climate: the LINC project 1. What is LINC? LINC is a 4-year Marie Curie Initial Training Network (ITN) funded by EU 7FP. Marie Curie ITNs aim at improv- ing the career perspectives of young re- searchers through a transnational net- working training program. The LINC consortium includes 6 aca- demic partners and 3 private-sector part- ners with expertise ranging from complex systems (networks, nonlinear time series analysis), environment and geosciences (climate, ocean and atmosphere dynam- ics) to commercial applications. The LINC network is training 15 young re- searchers and will organize workshops, schools and a final conference. The next LINC school (open to ex- ternal participants) will take place in April 2013 in The Netherlands. 2. Interdisciplinary research The complex network paradigm has proven to be a fruitful tool for the investigation of complex systems in various areas of science, e.g., the internet, neural networks, social networks, etc. The application of complex network theory to climate science is advancing the understanding of the Earth complex climate phenomena, such as El Ni˜ no-Southern Oscillation (ENSO). This has a huge economic and social impact for present and future generations, and can underpin advances in areas such as energy, environment, agricultural and marine sciences. Given the complexity of the inter-relations between the subsystems which constitute our climate, it is important to approach the problem from an interdisciplinary perspective. Veronika Stolbova (at Potsdam Institute for Climate Impact Re- search, Germany) is using the network approach for studying the Indian Monsoon. The anal- ysis reveals important patterns in the sense of water transport through the network of precip- itation and could be useful for prediction of the onset of Indian Summer Monsoon and switching to Winter Monsoon. Enrico Ser Giacomi (at Universi- tat de les Illes Balears, Spain) is studying geophysical transport networks that represent con- nectivity patterns given by fluid advection in the Earth’s atmo- sphere, focusing on surface cur- rents in the Mediterranean Sea. By simulating the motion of > 1.000.000 particles during sev- eral months (1 left, 6 right), the research could characterize the relationship between transport barriers and avenues, and net- work quantifiers. Liubov Tupikina (at Potsdam In- stitute for Climate Impact Re- search, Germany) is study- ing the Indian Summer Mon- soon (ISM) rainfall by using paleorecords which could yield light on monsoon variability dur- ing past periods of Earth cli- mate: the Medieval Warm Period (9001100 AD) and the Little Ice Age (15151715 AD). Giulio Tirabassi (at Universitat Politecnica de Catalunya, Spain) is studying the seasonal cycle effects in similarity measures of surface air temperature anoma- lies. The map of seasonal cycle phases (in the plot, the lag-times with respect to a node in the Mongolian Desert) reveals clear climate regions, the ocean mem- ory, and the 6 months symmetry between the NH and SH. Qingyi Feng (at Utrecht Univer- sity, The Netherlands) is study- ing the propagation of North At- lantic Multidecadal SST Anoma- lies. The analysis reveals that networks constructed from mu- tual information capture the main features of westward propagat- ing patterns existing in an ideal- ized model (left) and in the ob- servations (right). Fernando Arizmendi (at Univer- sidad de la Republica, Uruguay) is studying climate networks build from time series of the ed- die geopotential height at 200 mb. The Area Weighted Con- nectivity (AWC, in the plot) rep- resents the fraction of Earth that each node is connected to. The analysis shows that the most connected regions have certain interdecadal variability.” Veronica Martin Gomez (at Universidad de la Republica, Uruguay) is studying network distances using two datasets (HadSST and ERSSTv3b) from El Nino3.4, the Tropical North Atlantic (TNA), the Indian Ocean Dipole and the rainfall over northeast of Argentina (PCP) in- dices covering the period 1901- 2006. The analysis reveals strong periods of synchroniza- tion, for example, from 1911 to 1919 and from 1968 to 1974. Juan Ignacio Deza (at Univer- sitat Politecnica de Catalunya, Spain) is studying climate net- works constructed from surface air temperature anomalies using ordinal time-series analysis. The map shows the worldwide corre- lation of ENSO, as measured with mutual information. Victor Rodriguez (at Universi- tat de les Illes Balears, Spain) is studying a method to infer structure in networks based on Expectation Maximization (EM) with the aim of applying it to networks whose nodes are em- bedded in a metric space. The method has been tested when the connectivity depends on the distance between nodes in a 1D space (left). Combining EM and Metric Multidimensional Scaling allows recovering the connectiv- ity function (right). Alexis Tantet (at Utrecht Univer- sity, The Netherlands) is study- ing community detection by the Infomap algorithm in networks built from detrended monthly anomalies of SST (HadISST Hadley dataset, 1870 -2011). Intra-seasonal variability pre- viously filtered so as to focus on interannual to decadal time- scales. the analysis reveals that the largest community (in red) corresponds to ENSO, the dom- inant pattern of variability. Yang Wang (at Bar Ilan Univer- sity, Israel) is analyzing climate networks built from daily tem- perature data at 1000 hPa iso- bar. The plot shows the probabil- ity density function of the mean weight of negative links in the globe (left) and the dependence of the weight of negative links on the distance (right), in the SH during SH summer months. Miguel Bermejo (at Climate Risk Analysis, Germany) is analyzing the possible existence of heavy tails in climate data to try to answer the question does cli- mate have heavy tails?”. The figure displays a time series of a Levy process that allows the existence of large values, which could be a relevant feature in climate modeling. Hisham Ihshaish (at VORtech, The Netherlands) is developing efficient parallel graph-analyzing algorithms that exploit the ben- efits of nowadays supercomput- ing architectures. Parallel pro- cessing is crucial to overcome the resource limitations (pro- cessing and memory) of single processors. The figure depicts two main approaches of paral- lelization in the context of graph computation. Marc Segond (at Ambrosys, Germany) is working on cast- ing the processes and net- work topologies developed by the LINC partners into software using standardized processes. This includes creating a frame- work that integrates various tools and algorithms in several programming languages and make then communicate in order to interchange data. He is also working on exploring climatolog- ical data using machine learn- ing and data analysis techniques such as Genetic Programming, Ant Colony Algorithms, Innova- tive Graph Mining Algorithms or Frequent Pattern Mining. 3. Contact information: climatelinc.eu, [email protected]

Transcript of Learning about Interacting Networks in Climate: the LINC...

Page 1: Learning about Interacting Networks in Climate: the LINC ...cris/Talk/poster_potsdam_2013_v02.pdf · 7FP. Marie Curie ITNs aim at improv-ing the career perspectives of young re-searchers

Learning about Interacting Networksin Climate: the LINC project

1. What is LINC?LINC is a 4-year Marie Curie InitialTraining Network (ITN) funded by EU7FP. Marie Curie ITNs aim at improv-ing the career perspectives of young re-searchers through a transnational net-working training program.

The LINC consortium includes 6 aca-demic partners and 3 private-sector part-ners with expertise ranging from complex

systems (networks, nonlinear time seriesanalysis), environment and geosciences(climate, ocean and atmosphere dynam-ics) to commercial applications.The LINC network is training 15 young re-searchers and will organize workshops,schools and a final conference.The next LINC school (open to ex-ternal participants) will take place inApril 2013 in The Netherlands.

2. Interdisciplinary researchThe complex network paradigm has proven to be a fruitful tool for the investigation of complex systems in various areas of science, e.g., the internet, neural networks, social networks, etc. Theapplication of complex network theory to climate science is advancing the understanding of the Earth complex climate phenomena, such as El Nino-Southern Oscillation (ENSO). This has ahuge economic and social impact for present and future generations, and can underpin advances in areas such as energy, environment, agricultural and marine sciences. Given the complexityof the inter-relations between the subsystems which constitute our climate, it is important to approach the problem from an interdisciplinary perspective.

Veronika Stolbova (at Potsdam

Institute for Climate Impact Re-search, Germany) is using thenetwork approach for studyingthe Indian Monsoon. The anal-ysis reveals important patternsin the sense of water transportthrough the network of precip-itation and could be useful forprediction of the onset of IndianSummer Monsoon and switchingto Winter Monsoon.

Enrico Ser Giacomi (at Universi-tat de les Illes Balears, Spain) isstudying geophysical transportnetworks that represent con-nectivity patterns given by fluidadvection in the Earth’s atmo-sphere, focusing on surface cur-rents in the Mediterranean Sea.By simulating the motion of >1.000.000 particles during sev-eral months (1 left, 6 right), theresearch could characterize therelationship between transportbarriers and avenues, and net-work quantifiers.

Liubov Tupikina (at Potsdam In-stitute for Climate Impact Re-search, Germany) is study-ing the Indian Summer Mon-soon (ISM) rainfall by usingpaleorecords which could yieldlight on monsoon variability dur-ing past periods of Earth cli-mate: the Medieval Warm Period(9001100 AD) and the Little IceAge (15151715 AD).

Giulio Tirabassi (at Universitat

Politecnica de Catalunya, Spain)is studying the seasonal cycleeffects in similarity measures ofsurface air temperature anoma-lies. The map of seasonal cyclephases (in the plot, the lag-timeswith respect to a node in theMongolian Desert) reveals clearclimate regions, the ocean mem-ory, and the 6 months symmetrybetween the NH and SH.

Qingyi Feng (at Utrecht Univer-

sity, The Netherlands) is study-ing the propagation of North At-lantic Multidecadal SST Anoma-lies. The analysis reveals thatnetworks constructed from mu-tual information capture the mainfeatures of westward propagat-ing patterns existing in an ideal-ized model (left) and in the ob-servations (right).

Fernando Arizmendi (at Univer-

sidad de la Republica, Uruguay)is studying climate networksbuild from time series of the ed-die geopotential height at 200mb. The Area Weighted Con-nectivity (AWC, in the plot) rep-resents the fraction of Earth thateach node is connected to. Theanalysis shows that the mostconnected regions have certaininterdecadal variability.”

Veronica Martin Gomez (at

Universidad de la Republica,Uruguay) is studying networkdistances using two datasets(HadSST and ERSSTv3b) fromEl Nino3.4, the Tropical NorthAtlantic (TNA), the Indian OceanDipole and the rainfall overnortheast of Argentina (PCP) in-dices covering the period 1901-2006. The analysis revealsstrong periods of synchroniza-tion, for example, from 1911 to1919 and from 1968 to 1974.

Juan Ignacio Deza (at Univer-

sitat Politecnica de Catalunya,Spain) is studying climate net-works constructed from surfaceair temperature anomalies usingordinal time-series analysis. Themap shows the worldwide corre-lation of ENSO, as measuredwith mutual information.

Victor Rodriguez (at Universi-

tat de les Illes Balears, Spain)is studying a method to inferstructure in networks based onExpectation Maximization (EM)with the aim of applying it tonetworks whose nodes are em-bedded in a metric space. Themethod has been tested whenthe connectivity depends on thedistance between nodes in a 1Dspace (left). Combining EM andMetric Multidimensional Scalingallows recovering the connectiv-ity function (right).

Alexis Tantet (at Utrecht Univer-

sity, The Netherlands) is study-ing community detection by theInfomap algorithm in networksbuilt from detrended monthlyanomalies of SST (HadISSTHadley dataset, 1870 -2011).Intra-seasonal variability pre-viously filtered so as to focuson interannual to decadal time-scales. the analysis reveals thatthe largest community (in red)corresponds to ENSO, the dom-inant pattern of variability.

Yang Wang (at Bar Ilan Univer-

sity, Israel) is analyzing climatenetworks built from daily tem-perature data at 1000 hPa iso-bar. The plot shows the probabil-ity density function of the meanweight of negative links in theglobe (left) and the dependenceof the weight of negative linkson the distance (right), in the SHduring SH summer months.

Miguel Bermejo (at Climate Risk

Analysis, Germany) is analyzingthe possible existence of heavytails in climate data to try toanswer the question does cli-mate have heavy tails?”. Thefigure displays a time series ofa Levy process that allows theexistence of large values, whichcould be a relevant feature inclimate modeling.

Hisham Ihshaish (at VORtech,

The Netherlands) is developingefficient parallel graph-analyzingalgorithms that exploit the ben-efits of nowadays supercomput-ing architectures. Parallel pro-cessing is crucial to overcomethe resource limitations (pro-cessing and memory) of singleprocessors. The figure depictstwo main approaches of paral-lelization in the context of graphcomputation.

Marc Segond (at Ambrosys,

Germany) is working on cast-ing the processes and net-work topologies developed bythe LINC partners into softwareusing standardized processes.This includes creating a frame-work that integrates varioustools and algorithms in severalprogramming languages andmake then communicate in orderto interchange data. He is alsoworking on exploring climatolog-ical data using machine learn-ing and data analysis techniquessuch as Genetic Programming,Ant Colony Algorithms, Innova-tive Graph Mining Algorithms orFrequent Pattern Mining.

3. Contact information: climatelinc.eu, [email protected]