Predicting tree distributions in an East African biodiversity hotspot: model selection, data bias and envelope uncertainty

Abstract

Article:Platts, Philip J. orcid.org/0000-0002-0153-0121, McClean, Colin J. orcid.org/0000-0002-5457-4355, Lovett, Jon C. et al. (1 more AbstractThe Eastern Arc Mountains (EAMs) of Tanzania and Kenya support some of the most ancient tropical forest on Earth. The forests are a global priority for biodiversity conservation and provide vital resources to the Tanzanian population. Here, we make a first attempt to predict the spatial distribution of 40 EAM tree taxa (38 species), using generalised additive models, plot data and environmental predictor maps at 1 km resolution. The results of three modelling experiments are presented, investigating predictions obtained by (1) two different procedures for the stepwise selection of predictors, (2) down-weighting absence data, and (3) incorporating autocovariate terms to describe fine-scale spatial aggregation. In response to recent concerns regarding the extrapolation of model predictions beyond the restricted environmental range of training data, we also demonstrate a novel graphical tool for quantifying envelope uncertainty in restricted range niche-based models (envelope uncertainty maps). We find that even for taxa with very few documented occurrences useful estimates of distribution can be achieved. Initiating selection with a null model is found to be useful for explanatory purposes, whilst beginning with a full predictor set can over-fit the data. We show that a simple multimodel average of these two best-model predictions yields a superior compromise between generality and precision. Down-weighting absences shifts the balance of errors in favour of higher sensitivity, reducing the number of serious mistakes (i.e., falsely predicted absences); however, response functions are more complex, exacerbating uncertainty in larger models. Spatial autocovariates help describe fine-scale patterns of occurrence and significantly improve explained deviance, though if important environmental constraints are omitted then model stability and explanatory power can be compromised. We conclude that the best modelling practice is contingent both on the 60 | Platts PhD Thesis intentions of the analyst (explanation or prediction) and on the quality of distribution data.Generalised additive models have potential to provide valuable information for conservation in the EAMs, but methods must be carefully considered, particularly if occurrence data are scarce.

Frequency, not relative abundance, of temperate tree species varies along climate gradients in eastern North America

“…1 is truncated to zero when a plot is outside the estimated climatic range limits for a given species. There has been relatively little work on defining statistical estimation methods for climatic range limits (Parmesan et al 2005, Platts et al 2008). In our likelihood framework, we define a climatic range limit as the value along a climate gradient beyond which the probability of observing a nonzero abundance within a plot is set to some arbitrarily defined but very small likelihood.…”

Reconstructing historical snow depth surfaces to evaluate changes in critical demographic rates and habitat components of snow‐dependent and snow‐restricted species

“…For each accuracy assessment, we assessed bias with the slope and relative precision with the correlation coefficient (r) and calculated the root mean square error (RMSE) for predicted values. We also used envelope uncertainty maps (EUM) to investigate the extent to which calibration sites captured the range of environmental conditions across our study region (Platts et al 2008). All statistical analyses were performed in Program r (R Development Core Team 2010).…”

“…The EUM is a contribution-weighted average of distance maps associated with each model parameter. This method estimates prediction uncertainty by calculating the proportional distance of each grid cell across a study region from the calibration envelope with respect to each covariate in a model, and uses the average of these distance maps, weighted according to the relative contribution of covariates in the model (drop in explained deviance with covariate removed; Platts et al 2008). Dormann (2007) recommends that model predictions should not be extrapolated beyond 1 ⁄ 10th of the parameter range; therefore, caution is advised for regions where the EUM >0AE1 since this indicates that ‡1 predictors was extrapolated beyond the 1 ⁄ 10th-level (Platts et al 2008).…”

The Impact of Environmental Factors on Distribution of Ferula Ovina (Boiss.) Boiss. In Northwest Iran

“…Therefore, identification of effective variables on spatial distributions of plant communities is an essential issue in ecology (Araujo and Guisan, 2006;Bagheri et al, 2017). Often, there are many combinations of predictor variables that can describe the species distributions, especially when environmental variables are correlated, which leads to uncertainty on the effects of each variable (Platts et al, 2008;Murray and Conner, 2009). Therefore, determining the plant habitats and effective variables on their distributions is the first step in rangeland management (Bagheri et al, 2017).…”