Tradeoffs between model choice, data quality and quantity when estimating population trends and extinction risk
Population viability analysis (PVA) is commonly used to assess population trends and inform conservation management strategies. Models used in PVA generally fall within two categories, scalar (count based) or matrix (demographic) models. Both model types are often parameterized by time series data. Choice of model type, process error, measurement error, and time series length all have known impacts in populations risk assessments, but their combined impact in PVA has not been thoroughly investigated. Here, we use an age-structured population model to simulate population trajectories subject to different controlled levels of variability and uncertainty. We then evaluate the predictive capabilities of scalar and matrix models parameterized with these simulated time series to predict two common viability metrics, percent decline and quasi-extinction risk. Performance was evaluated across different time series lengths, population growth rates, and levels of process and measurement error. Specifically we ask: 1) Is there a trade-off between time series length and measurement error in the accuracy and precision of population projections generated with scalar and matrix models? and, 2) What is the influence of variability and uncertainty on the reliability of the two models’ projections?
We found that model performance varied by viability metric considered. High levels of process and measurement error reduced the reliability of the two models in predicted percent decline, with process error tending to contribute to bias in and spread of predictions more than measurement error. Somewhat surprisingly, scalar models exhibited similar or greater precision and accuracy than matrix models. Both models performed similarly well predicting quasi-extinction risk, although sometimes exhibiting a slight precautionary tendency. Increasing time series length improved precision and accuracy of predicted population trends, but only up to 15-20 years of data, and had very little effect on predicted quasi-extinction risks beyond 10 years. There was little to no trade-off between time series length and measurement error in the accuracy of the projections generated with matrix and scalar models. Finally, both models performed quite well in estimating quasi-extinction risk even under large levels of measurement and process error. This metric therefore emerges as a more robust measure of viability than percent decline. Our study provides evidence that short time series and simple (scalar) models can still be successfully used in conservation decision-making.