Risk assessment of the impact of future land use and future climate change on the distribution and diversity patterns of plant species used for nutrition (Burkina Faso)

Species distribution modeling, impact of land use and climate change on diversity patterns of plant species used for nutrition purposes Species records According to Zizka et al. (in press) 1,013 plant species (49%) in Burkina Faso have a recorded use. Traditional medicine (34%), human nutrition (19.4%) and animal fodder (17.6%) are the most important use categories. Amongst the 1,013 plant species, Zizka et al. (in press) identified 401 species with a nutrition purpose. The species modeled in our study were selected from the data compiled by Heubes et al. (2013) for the investigation dealing with the analysis of the impact of climate and land use changes on overall plant diversity in Burkina Faso. These data consist of observation and collection plant data from the West African Vegetation Database (Janßen et al. 2011); SIG-IVOIRE, Senckenberg Zoological and Botanical Collection Database SeSam, PhD and master thesis (see Acknowledgements in Heubes et al. 2013). Data originate from the period 2001-2010. Most of the studies focused on complete species inventories, reducing the risk of taxonomic bias in the species data. Besides data from Burkina Faso and plant data from the adjacent countries Mali, Niger, Benin, Togo, Ghana and Ivory Coast were introduced. The data from the adjacent countries were used to cover the broadest environmental range of each species in future predictions. Even if a species does not occur in Burkina Faso today, it might get suitable habitat conditions there in the future. In the final model we only considered species with more than 14 occurrence points as small amounts of occurrence points influence the accuracy of models (Stockwell & Peterson 2002). The species occurrence points were aggregated within grid cells of 5x5 arc minutes which represents approximately a resolution of 10×10 km. A total of 263 nutrition species were finally analyzed. Climate data For representing the current climate conditions (2000) we downboaded the bioclim data of Worldclim (Hijmans et al. 2005; http://www.wordclim.org). The bioclim variables are derived from monthly temperature and rainfall values in order to generate biologically meaningful variables. They present annual trends, seasonality and extreme environmental factors. For future climate climate onditions (2050) the Miroc3.2medres data (Center for Climate System Research, Japan) were introduced as Miroc3.2medres data had been used to generate the land use simulations of LandSHIFT (see chapter below). A pre-selection of the ecologically important bioclimatic variables for the study area resulted in seven bioclimatic variables. Amongst them only the two variables mean annual temperature (Bio1) and annual precipitation (Bio12) had correlations smaller than 0.6. Thus, only these two variables were applied as climate variables. High correlations between variables are problematic for the final model results (Table 1.26).  Land use data The dynamic and spatially explicit land use and land cover model LandSHIFT (Schaldach et al. 2011) was employed in the species distribution models. LandSHIFT uses biophysical and socio-economic drivers to simulate agricultural development on regional and continental scales and has been tested for the African continent in previous studies (e. g. Alcamo et al. 2011). The basemap of LandSHIFT is the Global Land Cover 2000 dataset (Mayaux et al. 2004). The three principal land use categories in Burkina Faso were (from north to south): Grassland, cropland and woodland/shrubland. Classes with minor, spot-like representation were barren, water and cities. LandSHIFT predictions were run under two model assumptions: technological change produced by mechanization and fertilization and no-technological change. The assumptions of the technological change scenario result in higher crop yields and in less demand for arable land (Heubes et al. 2013). Four different models were run in our study: 1.) current climate and current land use (no-technological change - technological change) and 2.) future climate and future land use (no-technological change - technological change) (Table 1.26). The spatial resolution of the LandSHIFT model is 0.01°. It was resampled to the target resolution of 10x10km. The SRES storyline for the future land use predictions was the IPCC A2. The IPCC SRES A2 scenario assumes intermediate levels of CO2 emissions and an intermediate increase in temperature being congruent with the current global trends. Species distribution models The analysis was carried out with BIOMOD under the free software environment R (R Development Core Team 2011, Package ‘Biomod2’). BIOMOD is a computer platform for ensemble forecasting of species distributions and enables the treatment of a range of methodological uncertainties in models and the examination of species-environment relationships. It includes the ability to model species distributions with several techniques, test models with a wide range of approaches, project species distributions into different environmental conditions such as climate and land use change scenarios (Thuiller et al. 2009). Within BIOMOD the regression methods GLM and GAM, the machine learning methods ANN and GBM, the classification methods CTA and MDA and the maximum entropy method MaxEnt were selected. The models were fitted with the climate parameters annual precipitation and mean temperature and the LandSHIFT scenarios under the assumptions of technological or no-technological change. We calibrated the models with current climate and current land use ‘no-technological change’. Then the models were run with current climate and current land use ‘technological change’ and future climate and future land use (technological and no-technological change). By calculating consenus models for each species the level of uncertainty was decreased based on the weighted averages methodology (Marmion et al. 2009). The diversity maps were generated by summing up the single model results for 2000 and 2050 according to the assumptions of technological or no-technological (ArcGIS 10.2., Esri Redlands). Thus, four diversity maps were obtained, two for 2000 and two for 2050.