Effects of Variability, Error, and Time Series Length on Population Predictions

Global biodiversity is fundamental for human welfare as ecosystem services, agricultural crops, and even human health depend on its preservation. Identifying strengths and shortcomings of current conservation risk assessment practices is of great importance given the current biodiversity crisis. Population viability analyses (PVA) are quantitative tools used to perform such risk assessments by modeling population dynamics. Variation in PVAs includes the type of population model used and the quantity and quality of the data available to parameterize the models. In this dissertation, I use computer simulations to evaluate the performance of two commonly used models (Matrix and Scalar) that have different data requirements. The first study evaluates the combined effects of environmental variability, measurement error and data quantity on the percent population decline estimates generated using matrix and scalar PVA for two different life histories. It was observed that measurement error has a stronger detrimental effect on projected decline than data quantity. Additionally, scalar models projected declines quite well relative to matrix models. The latter tend to be over-precautionary. The second study expanded the findings from the first by evaluating if the same results hold on a widely used decision-making framework—Red List Category of the International Union for the Conservation of Nature (IUCN). Results supported previous conclusions and provided evidence that scalar models can successfully be used in conservation decision-making with moderate levels of variability and uncertainty. The third study evaluates the reliability of population decline projections and their subsequent use in the Red List in the face of variability and uncertainty across life histories with different generation times. It was observed that increments in generation time could either increase or decrease the reliability of model predictions depending on the underlying growth rate. Specifically, both matrix and scalar models are more suitable for organisms with “slow” life histories than organisms with “fast” life histories. Overall, these studies highlight three important conclusions regarding the use of matrix and scalar models under high levels of variability and uncertainty: 1) percent population declines estimated with scalar models tend to be more accurate than for matrix models, 2) matrix models tend to over-estimate population declines, and 3) variability leads to greater errors in IUCN Red List classifications than measurement error for both models