Climate informatics:

Core problem:
climate system, knowledge graph, dependencies between climate variables depending on different
regimes of the response like mean/extreme and timescales, covariance of information, uncertainty
estimation, predictability estimation, a priori knowledge integration,time-lagged correlations in
current network method,balancing of dominating signal of spatial autocorrelations and possible
teleconnection, combination of data driven models and physical models,including non-climatic
variables in the network like anthropomorphic drivers, statistical relational learning,
the application of complex network theory to climate science
framework of self-organised criticality, or
the description of the corresponding dynamics by means of complex networks.
Bayes provides framework for incorporating prior knowledge.
I Examples of prior knowledge in climate prediction include:
I predictor should explain significant variance
I predictor should be in the tropics or same hemisphere
I predictor should be large scale and persistent
I dynamical models may simulate approximately correct relations
Spatio-temporal correlations:
Networks with complex topology are a new approach to model spatio-temporal dynamics. Such complex networks are systems composed of a very large number (thousands and
even millions) of nodes and are characterised by a complex topology and by complex types of
interactions. The dynamics of nodes can be regular or chaotic; the topology can be random or
somehow ordered, can include short- or long-range interaction; the types of interaction can be
linear or non-linear, constant or changing in time,etc.The emergence of synchronous behaviour
insuchnetworksisoneofthenewandchallengingproblemsinthestudyofdynamicalsystems. The
application of complex network theory to climate science is a very young field,
where only few studies have been reported recently .It has already been shown that this
approachenablesnovelinsightsintothetopologyanddynamicsoftheclimatesystemovermany
spatial scales ranging from local to global properties, and one can associate them with teleconnection patterns in the atmosphere, e.g. the North Atlantic Oscillation (NAO). On the global
scale, climate networks were found to possess small-world properties due to long-range connections (edges linking geographically very distant vertices), that stabilise the climate system
and enhance the information transfer within it. Donges et al. [
51
] develop a rather general
method to extract and characterise such a complex network from climate data sets. It is
important to emphasise that nonlinear characteristics,such as mutual information are essential
here Jim enez et al. [
52
] analyse seismic data from the network perspective. Based on a subtle
clustering analysis, they give arguments that the occurrence of a specific earthquake in Spain
might have been due to a reservoir impoundment.
small world networks

Thus, one way to apply networks

to the climate system is to assume that climate is
represented by a grid of oscillators, with each one
representing a dynamical system varying in some
complex way. In what we are then interested is the
collective behavior of these interacting dynamical
systems and the structure of the resulting network
A nonlinear measure that could
be used instead is the mutual information, but the
problem is that its accurate estimation requires much
more data than we have available.
Sub problems:
model uncertainty estimation based on unconstrained simulations and then observational
verification
cluster analysis to identify previously unknown climate features: by associating model proxies
with the observed phenomenons
prediction of extreme events or extreme stresses due to climate change on regional scale to
decadal scale
modelling the cascading uncertainty