| Popis: |
climate models and/or statistical downscaling techniques produces projected data for which there is no matching observed data sets for comparison, especially over Africa. The untested assumption is that the high-resolution climate models and statistical downscaling techniques generate more realistic historical climate and climate change projections, than do the low-resolution models that are evaluated by observational data. The influence of the resolution of climatic predictors on predicting species distributions and how the predictions might change, therefore remains inadequately addressed in species distribution modelling. To address the deficiency, this thesis aims to advance knowledge of the sensitivity of predictions of species distribution to different spatial resolutions of climatic predictor data using the high-value biodiversity area of Okavango River Basin (ORB) in Southern Africa as a study system. The steps involved are: Step 1: Comparing climate datasets across different spatial resolution to establish the degree of difference in spatial distribution of climate features. I did the comparisons between, Global Climate Model (GCM) climatologies and climate projections of rainfall and 2-meter air temperature and AFRICLIM historical baseline (WorldClim interpolated observations) and AFRICLIM future projection (statistically downscaled and bias corrected CORDEX projections), over the Southern Africa domain which includes the ORB. I then proceeded the work to the analysis of the variability in changes in future climate simulations from the GCM and AFRICLIM climate datasets for the period 2071-2100. From this first analytical step I found that the fine resolution dataset benefits from the addition of spatial detail such as more structure reflecting the topography of the region which is not clear in the coarse resolution dataset. Climatologies obtained showed that the rainfall gradient depicted in the course spatial resolution models is smoother along the south east of Southern Africa where steep topography leads to a sharp gradient in the observed field. In contrast the fine spatial resolution observational product displayed a clear zonal rainfall gradient from the southern Namibia area to central Botswana, this in agreement with the observed climatology. I therefore decided it was worth proceeding to test whether this seemingly added value presents any benefit in modelling species distribution under contemporary climate envelopes obtained from both fine and coarse models. Step 2: Creating species distribution models to explore the sensitivity of SDM predictions to different spatial resolutions of climatic data. I used field sampled vegetation plots data on species occurrences and current climate conditions to model one hundred and fifty terrestrial plant species in the ORB, to map their contemporary projected distributions at 2km×2km, 5km×5km, 10km×10km, 20km×20km and 50km×50km spatial resolution. I investigated relationships between the climatic variable importance scores and climatic variable identity and their interaction with climatic predictor spatial resolutions using Generalised Linear Models and Multiple Comparison Analysis (Post-hoc analysis). I found that the relative influence of temperature and precipitation variables varied with spatial resolution. The importance of the determinants of species distribution change between species but such changes are less determined by the predictor’s spatial resolution. Potential evapotranspiration consistently showed the greatest influence in determining species and richness distributions, across the fine to coarse spatial resolutions. Through the thesis I found that the spatial resolution of predictors had no effect on the model predictive power and that differing predictor spatial resolutions had only negligible effects on the model performance measured by Area Under the curve (AUC) of the Receiver Operator Characteristic Curve statistics. The spatial resolution of the used climatic predictors exhibited no effect on species richness pattern either. Through the sensitivity analysis, I conclude that the use of climatic predictors at different spatial resolutions will not intrinsically bias future species distribution models in certain landscapes. As a result, I believe that modelling at coarser spatial resolutions for conservation purposes in areas with relatively homogenous topographic and climatic conditions can be acceptable. This result will help conservation planning in situations where no high-resolution climatology exists. Step 3: Projecting future suitable climate space for a subset of the contemporary ORB conservation relevant plants, to assess the potential implications of climate change on them and on conservation and management planning in general. In doing so, I designed and applied a novel comprehensive climate change vulnerability assessment framework. I found different responses to climate change at species level and suggest that conservation and management actions should therefore be approached at the level of species, or subspecies, and not assume that multiple species will react in the same way as each other to climate change. |
