Preface to the Second Workshop on Knowledge Discovery from Climate Data: Prediction, Extremes, and Impacts | doi.page

Preface to the Second Workshop on Knowledge Discovery from Climate Data: Prediction, Extremes, and Impacts

2010pp. xxix–xxix

Abstract

The analysis of climate data, both observed and model-generated, poses a number of unique challenges: (i) massive quantities of data are available for mining, much of which has never been analyzed due to lack of a suitable methodology, (ii) the data is spatially and temporally correlated so that the IID assumption underlying many learning algorithms does not apply, (iii) the data-generating processes are known to be non-linear, (iv) the data is potentially noisy, uncertain and highly imbalanced, and (v) extreme events are known to exist within the data. These characteristics present fundamental research challenges to the data mining community. We posit that the interface with climate data sciences will not only present a real-world data challenge bereft of such issues, but also provide an application that has the potential to impact physical and natural systems in the world. Climate data mining is based on spatio-temporal data and inherits the attributes of space-time data mining. In addition, climate relationships are nonlinear, spatial correlations can be over long range (teleconnections) and have long memory in time. The processes are nonlinear, even chaotic or sensitive to initial conditions, with seasonal effects, as well as with low-frequency, even 1/f, noise. Thus, in addition to new or state of the art tools from temporal, spatial and spatio-temporal data mining, new methods from nonlinear modeling and analysis are motivated along with analysis of massive data for teleconnections and long-memory dependence.

Abstract

Related Papers