Improving population estimates of difficult-to-observe species: A dung decay model for forest elephants with remotely sensed imagery

Abstract

Accurate and ecologically relevant wildlife population estimates are critical for species management. One of the most common survey methods for forest mammals – line transects for animal sign with distance sampling – has assumptions regarding conversion factors that, if violated, can induce substantial bias in abundance estimates. Specifically, for sign (e.g. nests, dung) surveys, a single number representing total time for decay is used as a multiplier to convert estimated sign density into animal density. This multiplier is likely inaccurate if not derived from a study reflecting the spatiotemporal variation in decay times. Using dung decay observations from three protected areas in Gabon, and a previous study in Nouabalé-Ndoki National Park (Congo), we developed Weibull survival models to adaptively predict forest elephant (Loxodonta cyclotis) dung decay based on environmental variables from field collected and remotely sensed data. Seasonal decay models based on remotely sensed covariates explained 86% of the variation for the wet season and 79% for the dry season. These models included canopy cover, cloud cover, humidity, vegetation complexity and slope as factors influencing dung decay. With these models, we assessed sensitivity of elephant density estimates to spatiotemporal environmental heterogeneity, showing that our methods work best for large-scale studies >50 km2. We simulated decay studies with and without these variables in four Gabonese national parks and reanalyzed two previous surveys of elephants in Minkébé National Park, Gabon. Disregarding spatial and temporal variation in decay rate biased population estimates up to 1.6 and 6.9 times. Our reassessment of surveys in Minkébé National Park showed an expected loss of 78% of forest elephants over ten years, but the elephant abundance was 222% higher than previously estimated. Our models incorporate field or remotely sensed variables to provide an ecological context essential for accurate population estimates while reducing need for expensive decay field studies.