Abstract
Understanding the importance of climate in determining species distribution and how it might change as a function of spatial grain size is a vital issue for species distribution modelling (SDM), yet it is often not accounted for in models and has not been extensively addressed in under sampled areas in tropical forests. Using extensive field sampled vegetation plots data on species occurrences and current climate conditions we modelled 150 plant species in the Okavango River Basin, to map their current projected suitable climate space at 2km2, 5km2, 10km2, 20km2 and 50km2 pixel resolution. Relationships between the variable importance scores and variable identity and their interaction with predictor spatial grain were investigated using Generalised Linear Models and post-hoc analysis. We found variation in the relative influence of temperature and precipitation variables across the spatial grains. The importance of the determinants of species distribution may change between species but such changes are less determined by the predictor’s spatial grain. Potential evapotranspiration consistently exhibited the greatest influence in determining species and richness distribution across spatial grains. We found that the spatial grain of predictors had no effect on the model predictive power and that varying predictor spatial grains had only negligible effects on the model performance measured by AUC and Kappa statistics. The spatial grains of climatic predictors used showed no effect on species richness pattern either. Our results indicate that in areas with relatively low topographic variation, modelling at coarse spatial grain for conservation purposes can be acceptable. Moreover, we show that in tropical areas that have comparatively homogeneous climatic conditions along large spatial extents the variable importance is not influenced by predictor spatial grain. For projections of contemporary species suitable climate space in relatively flat and topographically homogeneous areas which often have a climatically homogenous landscape, more attention must be given to the identity of the selected predictor variables for modelling species distributions than to their spatial grain size. We suggest that in species distribution modelling for conservation planning, assessment of the input datasets spatial grain should be informed and guided by knowledge of the landscape level topographic conditions, as protocol.
Highlights
Species Distribution Models (SDM) have been a critical technique for investigating a variety of ecological, conservation and management issues (Garcia et al, 2012; Van Echelpoel et al, 2015; DellaSala et al, 2018) as well as in studies of paleoclimate (Reichgelt et al, 2018)
There was no effect of spatial grain in the performance of the models, as evident from the reported consistent Area Under the curve (AUC) values (Table S1)
Our analysis (Figure 3) shows PET having the largest effect in determining the species suitable climate space of plants of the ORB at fine predictor spatial grain sizes (2, 5 km2)
Summary
Species Distribution Models (SDM) have been a critical technique for investigating a variety of ecological, conservation and management issues (Garcia et al, 2012; Van Echelpoel et al, 2015; DellaSala et al, 2018) as well as in studies of paleoclimate (Reichgelt et al, 2018). Ecologists and conservation practitioners are still faced with problems of spatial and temporal scale variations in the datasets used in operating SDMs.There is no consensus about the most suitable predictor spatial grain that can improve the accuracy of SDM prediction (Raven, 2002; Bradter et al, 2013; Fournier et al, 2017) and there is a lack of full understanding of the effects of building SDMs at multiple predictor spatial grains. This has an impact on management planning and conservation decision making (Porfirio et al, 2014; Yates et al, 2018). The uncertainty in predictor variables can emanate from spatial scaling (Scott et al, 2002), brought about by the disparity between the spatial grain size of the climate data and that of the species modeled, which may lead to uncertainty in the magnitude, rate and direction of changes of biodiversity imposed by the changing climate (Garcia et al, 2012; Wan et al, 2016)
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
