Predictive modelling of terrestrial reptile species richness for conservation in Saudi Arabia

Studying species richness distribution patterns depends on data quality, which imposes challenges for conservation action. Species distribution modelling is a powerful tool to overcome some of these hurdles, and can gives us ecological and biogeographical insights about species distributions patterns and habitat suitability even in poorly sampled habitats. The main objective for this thesis was to model the distribution of 62 terrestrial reptile (lizard and snake) species to obtain the pattern of species richness and habitat suitability across Saudi Arabia. I used both species distribution models (SDMs) built for Egypt (a better-studied area) and then spatially transferred into Saudi conditions, as well as models built with local data. Maxent software was used to interpolate (based on Saudi data) and transfer (based on Egyptian data) models. The accuracies of these models were then evaluated on-the-ground using data from field work conducted in the summer of 2017 and 2018 in Tabuk Province. The detectability and occupancy of each species in the field was estimated using PRESENCE software. I then assessed the conservation value of the proposed and current Protected Area network in Saudi (managed by the Saudi Wildlife Authority) and identified potential new important areas of conservation planning under various different assumptions (socioeconomic and uncertainty analysis) using Zonation software. The maximum species richness of reptiles was predicted to occur in the central plateau, north-western borders, south-western areas, and in the coastal areas of Saudi Arabia. I identified areas across Saudi Arabia that are considered to be under-sampled relative to these predictions. The transferred Egyptian models yielded very effective predictions, analogous to those generated by local independent models based on Saudi data. Some species were predicted better by the transferred Egyptian model than the equivalent Saudi model. Model accuracy across species as estimated using the independent field data varied among species and correlated positively with accuracy estimated using partitioning of the original Saudi data, but validation using independent data gave lower accuracy estimates overall than data partitioning. Ground-truthing validation showed that the transferred Egyptian model was able to predict presences better than the Saudi model built from local data, and that the Saudi model was better at capturing the predicted areas of high species richness. Species detectability in the field was weakly positively correlated with model accuracy calculated from the new independent field data, hinting that models work better for species which are easy to survey. The directed field validation successfully led us to record species within Saudi Arabia that had not been recorded before, and extend the known range for a species that is located completely outside the Protected Areas and had not been recorded for 40 years. I found that the conservation value of current and proposed Saudi’s PAs was significantly better than the areas outside them across all Zonation models. I proposed possible extensions of some PAs to cover top-ranking areas predicted to be suitable for reptile conservation. SDM predictions appear to be a very effective tool for exploring spatial patterns of species diversity, especially when there is a paucity of data. When testing SDMs, using new data, collected independently of the data used to build the models, seems to be the best practice to evaluate model predictive performance, but caveats with such data must be considered as they may not be entirely statistically independent. Incorporating different scenarios in conservation assessments and planning can help increase its reliability in identifying potential important areas for conservation.