Simultaneous analysis of impacts of human abundance and quasi-cyclic climate conditions in tornado counts with hierarchical Bayesian models

Our previous work using hierarchical Bayesian models (HBMs) has confirmed these tools are capable of modeling non-stationarity in tornado count records arising from societal and climate shifts, but our models to date have examined these factors independently. Climate factors that were found to be important are ENSO and NAO; furthermore, significant regional variability was evident in their influence on tornado reports (Wikle and Anderson 2003). In regards to societal factors, we found the assumption that low population density might influence less the counts of damaging (EF2 or higher) compared to nuisance tornadoes (EF0, EF1) is not valid in all regions (Anderson et al. 2007). In particular, this assumption does not hold in the vicinity of Oklahoma City and Tulsa.

We will present new work in which HBMs are devised that enable simultaneous modeling of societal and climate factors. One critical enabling step is the use of nighttime light intensity maps to identify regions of contemporary population abundance. Furthermore, we examine seasonal counts to infer contemporaneous and lag dependence on climate indices. These developments move our modeling approach in the directions of use within climate projections and seasonal forecasts.