The Statistical Evaluation of Ecological Indicators

Abstract

Ecological indicators are easily measured surrogates for underlying properties or responses of a system that are difficult to measure accurately and reliably. Methods from signal detection theory can be used to assess the usefulness of such indicators, based on pilot data sets in which some gold standard of the underlying response has been measured. For responses that can naturally be dichotomized (e.g., absent vs. present, or acceptable vs. unacceptable), we can estimate an indicator's sensitivity (the probability of a positive indicator, given that the true response is positive) and specificity (the probability of a negative indicator, given that the true responses is negative). These properties, together with the prevalence of the response in the population being studied, determine the indicator's predictive value (e.g., positive predictive value is the probability of a positive response, given that the indicator is positive). Applications of this methodology are described for two examples: the use of satellite imagery to infer oceanic pigment concentrations, and the use of baseline levels of acid‐neutralizing capacity (ANC) to anticipate acidification episodes in lakes and streams.