Working Group on risk

Cette conférence sera animée par le professeur Sreekar VADLAMANI, CAM Bangalore & Lund Univ. Dpt. of Statistics

Abstract

Analysing high dimensional spatio--temporal data has been a central theme of research in areas ranging from natural sciences to social sciences.

Specifically, identifying interesting patterns and clusters hidden in such complex databases is undoubtedly a challenging and critical task. Common clustering algorithms like k-means perform the task of clustering without utilising additional information like the physics of the data. Often, when studying natural phenomenon like weather, the data is supported by well established physical theories and models. A data analyst may want to use this information to his/her advantage by appropriately incorporating the physical theories in the (clustering) algorithms. In this aspect, researchers have used graphical models to understand complex data structures in various different contexts, and gained significant understanding of many different phenomena.

Taking cue from the confidence shown by researchers on graphical models, we fit a Markov random field model to Indian monsoon during the monsoon season, and identify chief spatio--temporal patterns in rainfall over the Indian landmass. In the process, we also identify what are called the "active" and "break" spells during the monsoon season which have long been considered the primary indicators of good or bad monsoon.