Abstract

Assessing risks in environmental problems, such as contamination, heatwaves, temperature, and floods, is of utmost importance for biodiversity and human health. In many cases, this need for risk assessment can be easily translated into a “yes” or “no” problem. For example, by answering the question: does a specific pollutant, such as PM10, exceed a high threshold? For these cases, a class of mathematical models called geostatistical binary models can help us answer many questions regarding the environmental index we are observing, and they can also help us predict the occurrence of high index values in places where we do not observe it. However, fitting these models can be difficult since we usually have an imbalanced quantity of “yes” and “no”, which limits the amount of information.
This talk will discuss a novel framework for Binary geostatistical process that combines extreme value techniques and bayesian inference methods, and illustrate the versatility of this approach to predict the levels of a specific pollutant (PM10) in Mexico City over space and time.