Presented by Kaiguang Zhao, PhD
OSU School of Environment and Natural Resources
Data underpins essentially all serious scientific pursuits. Letting data speak for itself is straightforward but not always easy or possible. Rather, models often come into rescue to help make sense of data, a process frequently perceived to be tricky and subjective. How to best analyze and interpret data with models –especially statistical models-- can differ greatly from one scenario to another; this is particularly true because disciplines tend to have their own established modeling paradigms and researchers have freedom to exercise personal or even idiosyncratic beliefs up to their individual intellectual depths. Such model-based inference also varies so as to tailor the specific nature of problems at hand – hypothesis-driven, question-based, or need-driven. This talk will reflect on my non-statistician views of the many aspects in tackling data with models (e.g., all models are wrong, modeling as an art, to explain or to predict, and common data phenomena such as collinearity). As a geospatial modeler, I will also go over several case studies exploring various ways to leverage statistical analytics for problems such as malaria risk assessment, mapping of ecosystem dynamics from satellite, Bayesian calibration of hydrological models, niching modeling, and climate impacts on species distribution. The purpose of this talk is not to inform best modeling practice but rather to encourage the exploration of data analytic power for environmental applications.