The Earth system is globally connected and assessing its risk requires a globally connected view. At reask we are building all our models with this obsession in mind. The result is a unique framework allowing our partnering organisations to (i) quantify the correlation in the risk they carry globally and (ii) gain some early indication about their risk profile in the near future (seasonal to inter-annual probabilistic risk prediction).
How we go about it:
Automated Knowledge Discovery: teleconnections such as El Niño–Southern Oscillation (ENSO) or the North Atlantic Oscillations (NAO) are known to impact our climate globally and over a range of time scales. The question we have been (re)asking is how can the influence from such natural modes of climate variability best be identified and measured? We found our answer from recent advances in the field of machine learning / pattern recognition and the result is a series of non-linear global climate indicators that inform all our risk models.
Links to risks of interest and uncertainty modelling: History does not provide a reliable picture of the risk we face today and none of the past relationships we might have identified should be taken as set in stone. To avoid over-fitting the recent past we rely on two key tools: (i) use of alternative versions of history generated from high resolution numerical model simulations and (ii) a Bayesian approach to modelling the predictor / risk relationships at the core of all our models.