Abstract

Scientific knowledge is acquired in geophysics generally without the benefit of controlled experiments. In this paper, we discuss how scientific inference based on observations occurs in geophysical contexts. We develop a specific approach that uses approximate simultaneity of proposed cause and effect phenomena to infer causality. The approach applies equally well to effect phenomena that follow the cause with a known time delay. We find that, in general, establishing a causal relationship between two phenomena based on simultaneity requires knowledge of how often simultaneity of these phenomena occurs in the absence of causality. We then extend the discussion to using numerical simulations in the scientific inference process. Numerical simulations of physical processes, because they can simulate the values of observations, are often used to infer what physical processes are occurring in nature. We discuss agreement between model output and observations as a basis for inferring the physical processes underlying the observations. We find that an important factor to consider, which we here call the “confusion factor,” is how often it may occur that insufficient model representations of the physical processes nevertheless lead to agreement between model computations and observations. We suggest that models of intermediate or low complexity may have a significant role to play when using geophysical simulations to reach scientific conclusions.

Key Points

A Bayesian approach is developed to analyze scientific inference in geophysics Simultaneity of proposed cause and effect phenomena does not imply causality Geophysical simulations of low or intermediate complexity are useful