IMAGe Seminar- Declaring Carbon Dioxide Sources and Sinks: A Spatial Analysis

10/27/2015 - 12:00pm
ML239- Damon Room
Sandy Burden

Sandy Burden, University of Wollongong, Australia

Statistical models play an important role in predicting the effect of changing environmental conditions
on environmentally sensitive, scientific processes of interest. For example, climate models have been
developed to help understand Earth’s atmosphere, particularly those aspects associated with global
climate change such as the global carbon cycle. One goal of these models is to obtain spatio-temporal
predictions for the location, extent, and persistence of carbon dioxide (CO2) sources and sinks, and the
movement of CO2 between these locations. This entails two challenges: Predicting areas with
substantial positive and negative CO2 flux, and quantifying the uncertainty of the predictions.

In this talk, we use a spatial analysis of remotely-sensed CO2-flux data to predict regions of positive and
negative flux that exceed a pre-determined threshold: First, we define an optimal spatial predictor using
a spatial empirical hierarchical model, and then we apply simulation-based methods to obtain
prediction regions for exceedances. The methodology is demonstrated using simulated spatio-temporal
flux data from an observing system simulation experiment (OSSE), and its dependence on the choice of
spatial predictor at various time points is assessed.

This is joint work with Noel Cressie, University of Wollongong. Keywords: Bayesian hierarchical model,
Exceedance, Prediction region, Remote sensing, Spatial flux field.

October 27, 2015, 12:00-1:00pm

Mesa Lab - Damon Room