Predictive distribution modeling with enhanced remote sensing and multiple validation techniques to support mountain bongo antelope recovery


  • Editor: Iain Gordon

  • Associate Editor: Hervé Fritz

Lyndon D. Estes, Woodrow Wilson School, Princeton University, Princeton, NJ 08544, USA. Tel: +1 202 431 0496; Fax: +1 609 258 6082


Understanding endangered species' spatial ecologies is fundamental to designing effective recovery strategies. Transferable predictive distribution models (PDMs), based on predictors describing the ranges and scales of relevant environmental gradients, can provide this understanding. Using such models for rare species such as the mountain bongo Tragelaphus eurycerus isaaci, an endangered antelope being restored within its endemic range in Kenyan montane forests, is difficult because the species' rarity and challenging terrain complicate data collection. To help overcome data limitations, we used advanced remote sensing (RS) and multiple validation techniques to improve bongo PDMs, which were developed using logistic regression and the information-theoretic approach. We derived predictors using RS, including a new technique for measuring micro-scale vegetation structure, and assessed predictive performance using bootstrapping and independent observations. Terrain ruggedness was the strongest habitat-use predictor, followed by soil moisture availability, distance from law enforcement outposts, vegetation structural complexity, and vegetation edge density. Prediction accuracy generally ranged between 73 and 89%, but terrain ruggedness limited model transferability. The more direct RS-based vegetation predictor improved model transferability. Bongo restoration efforts should focus on high probability areas delineated via a composite of all tested models. The techniques used – particularly RS – enhanced inference quality and the transferability of distribution models, and can be applied to other critical species and ecosystems.