Proposal Description

This session consists of three talks and one discussion, centered around the broad themes of forecast and variability, in the context of environmental data analysis, climate sciences and climate change. One of the speakers will discuss the possibilities of deconstructing the frequency spectrum a system predictor in a way that rebrands it with a spectrum similar to the response being modeled. Any forecasting problem where the system predictors represent global modes of variability while the response variables are local, require such spectral modifications to maximzse the forecastability of the system. In other words, predicting seasonal drought anomalies for, say, Adelaide, should be undertaken using a localized version of global climate variability indices such as the NINO3.4, the localized NINO3.4 being a modification such that its frequency spectrum resembles that of the response. This modification is demonstrated to enhance drought forecasts across seasonal to decadal timescales throughout Australia. Another speaker will discuss hierarchical Bayesian modeling to understand tropical cyclones and hurricanes and the damage they cause, with a specialized case study based on the Atlantic basin. Bayesian modeling of a sequence of extreme events, which may themselves be less extreme in a changing climate regime, leads to accurate forecasting and elicitation fo the predictive distributions of interest. The third talk will focus on the understanding of climate variability and uncertainty with process understanding and the aid of data-driven sciences. The focus here is on modern machine learning techniques for decomposition of sources of variability of data, for better physical modeling of the underlying processes, and accurate forecasts and prediction. The session will also include a discussion of the talks by an expert on both environmental data sciences and machine learning, who will provide critical evaluation of the proposed methodologies contained in the talks, and propose open questions and challenges.