PS 22-48 - Tools for conservation: Estimating the population size of heterogeneously distributed plants

Abstract: Population size is a key ecological parameter and conservation metric for assessing extinction risk. However, for many plant populations estimating total population size is surprisingly challenging. Populations are often too large, and conservation resources too limited, to make counts of the entire population feasible. Current approaches to estimate population size from surveys often assume simple distribution patterns (e.g. homogenous density) that may be inaccurate when factors like micro-climates, soil chemistry, and individual heterogeneity in demography lead to heterogenous population distributions. The goal of this study is to develop statistical methodology that estimates the size of heterogenous plant populations using transect survey data. We do this by developing a spatial gaussian process model in a Bayesian framework that estimates latent autocorrelation structure of sampled locations to then predict the density at unsampled locations. To test our approach, we simulate a population with heterogeneous densities under various scenarios of autocorrelation (clumped vs random distributions) and sampling effort. We found that regardless of levels of autocorrelation, more sampled transects resulted in a better predictions of population density. Yet, a decrease in sampled transects created more drastic differences in prediction quality for populations with low autocorrelation and random distributions than populations with high autocorrelation and clumped distributions. This result highlights the need for appropriate sampling effort dependent on how plants are distributed on a landscape. Our approach could be applied to estimate total population size in endemic species with limited ranges, focal populations of wide-ranged species, or any sedentary species with presence and absence data. This project demonstrates statistical methodology for estimating population sizes that allow for more biological complexity by including heterogeneous densities while encompassing commonly used field survey data.