Cinque Terre

Identifying vulnerable hotspots of land cover or climate dependent bird species in Africa, using species distribution modelling

by Serjay Stolyarov (MSc GIS)

external supervisor: Dr Graeme Buchanan

Birds act as indicators of the health of our environment, alerting us when ecosystems are out of balance. They occur nearly everywhere, but due to differing ecological requirements, particular species are uniquely distributed. Knowledge of the occurrence of birds can be used to identify the places that are most important for the conservation of biodiversity.

from "State of Africas birds" (Birdlife International, 2013)

Overview

Birds are crucial parts of ecosystems and good indicators of the state of natural environments of the planet. As empirical evidence suggests, bird's populations around the world are declining due to habitat loss from agricultural activities, deforestation and increasing effects from climate change, particularly in areas of Africa which are greatly abundant and rich in biodiversity. Africa is home for very large number of endemic birds, distributions and environments of which we still do not fully understand. The aim of this project is, using SDM methods, to identify the species whose ranges appear to be limited by land cover (captured by NDVI) and those limited by climate, and identify those areas in Africa which might be hotspots for particularly responsive species. The identification of these areas will guide conservation efforts for either management to reduce loss of natural habitat, or introduce adaptive management for climate change.

Source of PICTURES

Species survival and existence, in accordance to ecological niche theory, relies on certain ecological and environmental conditions. With advances in computation technology and Geographical Information Systems (GIS), the ability to model the relationships between those environmental conditions and species niches has been increasing. Using species distribution models (SDM) it is possible to quantify relative influence of environmental variables such as annual precipitation, temperature, elevation, slope, vegetation and other drivers on the species existence in environmental and geographical space.

Methodology

This project investigates the use of unclassified cumulative NDVI data from SPOT to represent land cover and vegetation, in contrast to climatic variables of temperature and precipitation from WorldClim, along with GMTED2010 elevation data, and assess their relative influence across African bird species. Mahalanobis Distance (MD) was chosen as a modelling method (Figure 1).

All of the bird species presence data is derived from EOO maps kindly provided by BirdLife International. The origin of the distribution maps are derived from a variety of sources, including observer records, distribution atlases, field guides and expert opinion. To reduce commission errors there was a step to estimate species maximum potential occurrence within its EOO by creating ESHs (Beresford et al., 2011). Raster of species ESHs are converted to points to represent species presence localities.

Datasets

For each species, three sets of environmental MD distributions were predicted based on: climate + elevation, NDVI + elevation and all variables combined (Figure 1). Each month middle decal of SPOT VEGETATION NDVI was compiled to form 6 year mean product for the period of 1999-2004. To avoid over fitting, four variables were selected for each of the models. For climate: Annual Mean Temperature, Temperature Seasonality, Annual Precipitation, and Precipitation Seasonality. For NDVI: Minimum, Maximum, Mean and Coefficient of Variance. Elevation (GMTED2010) was also included in all of the models.

Modelling and model evaluation was done using R 3.11 statistical software running on a high performance computing cluster �Ultra� at Edinburgh University EPCC. To assess accuracy, three measures were used: Area Under the Receiver Operating Curve (ROC) known as AUC, Kappa statistic and True Skills Statistic. Max Specificity plus Sensitivity threshold was used to gain accuracy results. TSS measure was chosen to select between NDVI or climate responsive(Allouche et al. 2006).

Mahalanobis Distance

Figure 1: Illustrated process of SDM using Mahalanobis Distance

Results

Accuracy Measures

In total, 1503 African bird species were analysed. The mean of all accuracy measures across all modelled species showed consistent contrast between three types of models (Figure 2). The combined model with climate, NDVI and elevation variables, on average showed the highest accuracy response across all modelled African bird species (55%-65%)(Figure 3). However, this was not shown to be consistent for all of the species, with 5-10% giving a higher response to NDVI models than both climate and combined approach (Figure 3). Similarly, 25-30% of species did not show a better model performance when combining NDVI variables with bioclimatic. Climatic models on average show higher response, with only 22-25% of species distribution being predicted better by NDVI variables (Figure 4).

Figure 2: Mean values of accuracy measures for all bird species

Figure 3: Comparison of three model performances for all bird species

Figure 4: Comparison of NDVI and climate model performances for all bird species

Species Distributions

TSS measure was selected to distinguish between NDVI responsive species and climatically responsive (Figure 5). Since the species could be equally responsive to climatic and NDVI variables, particularly as NDVI can be a function of climate, it is useful to investigate TSS measure differences. A difference of 0.05 in TSS value was selected to separate NDVI and climate responsive from equally responsive (Figure 5, 6). 892 (59%) species have shown a better response to climatic variables, 211 (14%) of species being particularly land cover influenced and 400 (27%) equally responsive (Figure 5).

Bivariate map shows relative abundance of unique NDVI or climate responsive species (Figure 7). All the mapping was done by summing the corresponding species ESHs to calculate species densities. Separate plots of total count and proportions were further made for equally, NDVI and climate responsive bird species to identify hotspots (Figure 8, 9, 10). Majority of the climate responsive species are concentrated along the equator, in West and Central African tropical rainforests, where species diversity is large.

Figure 5: TSS measure for climate and NDVI models, highlighting response

Figure 6: Histogram of highlighted response

A high climatic effect is also seen around Saharan desert where species diversity is low (Figure 7b, 10b). NDVI and climate equally good explain species ranges particularly below Sahara (Figure 8b). Many NDVI responsive species are distributed in patchy fashion in South East Africa, in addition to Central Africa, along with climate responsive (Figure 7a, Figure 9a), but proportionally are much more in Madagascar and South Africa (Figure 7a, 9b).

This study highlights key distributions of bird species which are relatively more influenced by climate or by vegetation cover, and brings this understanding to highlight conservation efforts. High concentration of NDVI responsive birds in Madagascar leads to suggest that there is a high influence of vegetation cover on bird distributions, leading to direct danger from habitat loss. Areas corresponding with high density of climatically responsive species at Central Africa, are particularly at risk of potential habitat loss due to expansion and conversion of natural habitats for anthropogenic purposes.

Figure 7: a) NDVI responsive species. b) Climate responsive species. Bi-plot of relative abundance against total modelled species

Figure 8: Equally responsive species a) Total count density of unique species. b) Relative abundance of unique species

Figure 9: NDVI responsive species a) Total count density of unique species. b) Relative abundance of unique species

Figure 10: Climate responsive species a) Total count density of unique species. b) Relative abundance of unique species

Land Cover Assessement

Habitat cover, derived from GLC2000 over the areas with high concentration of NDVI and climate responsive species, was analysed (Figure 11). Due to vast difference in concentrations, visual assessment was made to highlight areas of interest. Evergreen lowland forest is a dominant habitat for high proportions of climate responsive species. While arid and rocky areas are hosts to low proportions but highly climate dependant species.

NDVI responsive species have a varied land cover use. Heavier reliance on croplands, grasslands and shrublands observed in NDVI responsive species with significant but lower proportions concentrated around South Africa. Madagascar hosts particularly higher concentrations of NDVI responsive species with most preferring forest or shrubland habitats.

Figure 11: GLC2000 land cover of areas of interest for Climate responsive: a) Proportional concentration > 0.9. b) Proportional concentrations 0.8-0.9. And NDVI responsive: c) Proportional concentrations > 0.4. d) Proportional concentrations 0.15-0.4.

Species Vulnerability Assessement

In addition to identifying the distribution and hotspots of either climate limited bird species or land cover limited, study also assesses overall climatic vulnerability, developed by Foden et al., (2013) , plus the IUCN threatened classification (IUCN, 2015).

Only those bird species which were classified overall climatically vulnerable (Foden et al., 2013), on IUCN threatened list and either NDVI responsive or climate responsive, were mapped to distinguish hotspots (Figure 13, 14). Proportionally, both NDVI responsive and climate responsive have near the same percentage of threatened and climatically vulnerable (Figure 12).

Combining the findings with vulnerability traits produced by Foden et al (2013), as well as IUCN classifications, hotspots of particular attention are visible. Majority of high concentrations of NDVI or climate responsive species fall into already recognised endemic biodiversity hotspots of Africa (Brooks et al., 2002), including Madagascar, Upper Guinea forests, Cameroon lowland forests and Albertine Rift, most of which fall into Endemic Bird Areas, as classified by BirdLife International. Many endemic hotspots of biodiversity have been found to be at a great risk of extinction due to past and on-going habitat loss.

vulnerability comparison

Figure 12: Vulnerability Comparison

The general trend of bird species in Africa is negative, with only 7 species down listed on IUCN threatened list, while 25 up listed (BirdLife International, 2013). Important Bird Areas (IBA) and Protected Areas (PA) serve as a proven conservation tool, although only 14% of threatened African species ranges fall into PAs and 30% in IBAs (Beresford et al. 2011). This study suggests focusing on Protected Areas expansions in Madagascar and areas of high climate response.

Figure 13: Vulnerable NDVI responsive species: a) Total count of unique IUCN threatened b) Total count of unique climatically vulnerable

Figure 14: Vulnerable climate responsive species: a) Total count of unique IUCN threatened b) Total count of unique climatically vulnerable

Conclusion

This study contributes to understanding of bird distributions over continental Africa. Combining vulnerability assessment with species response distributions will hopefully be beneficial to conservationists in concentrating on providing more efficient practices for particular hotspots in more detail as identified in this study. The study highlights areas of Madagascar and to lower proportions, South African regions with particularly high concentrations of NDVI responsive bird species, while central African rainforests, Afromontanes, West African rainforests and arid areas around Saharan Desert show higher concentrations of climate responsive species. Out of 1503 modelled species, 892 (59%) were particularly climate responsive, 400 (27%) are equally responsive and 211 (14%) are particularly NDVI responsive. Additionally combined modelling approach, for 55% - 65% of species, yielded higher model performance, while 25% - 30% of distributions were better explained by climate alone and 5% - 10% by NDVI alone. Both NDVI responsive and climate responsive are proportionally equally vulnerable under IUCN threatened list and climatic vulnerability traits. Vulnerability assessment of hotspots and past research highlights a dire need to improve PA networks in order to preserve threatened species, many of which are endemic.

Key References

Allouche, O., Tsoar, A., Kadmon, R., 2006. Assessing the accuracy of species distribution models: prevalence, kappa and the true skill statistic (TSS). J. Appl. Ecol. 43, 1223�1232. doi:10.1111/j.1365-2664.2006.01214.x

Beresford, A.E., Buchanan, G.M., Donald, P.F., Butchart, S.H.M., Fishpool, L.D.C., Rondinini, C., 2011. Poor overlap between the distribution of Protected Areas and globally threatened birds in Africa: Protected Areas and threatened African birds. Anim. Conserv. 14, 99�107. doi:10.1111/j.1469-1795.2010.00398.x

BirdLife International, 2013. State of Africa�s birds 2013: Outlook for our changing environment.

BirdLife International and NatureServe, 2014. Bird species distribution maps of the world.

Brooks, T.M., Mittermeier, R.A., Mittermeier, C.G., Da Fonseca, G.A., Rylands, A.B., Konstant, W.R., Flick, P., Pilgrim, J., Oldfield, S., Magin, G., others, 2002. Habitat loss and extinction in the hotspots of biodiversity. Conserv. Biol. 16, 909�923.

IUCN, 2015. The IUCN Red List of Threatened Species. Version 2015.1 http://www.iucnredlist.org.

Foden, W.B., Butchart, S.H.M., Stuart, S.N., Vi�, J.-C., Ak�akaya, H.R., Angulo, A., DeVantier, L.M., Gutsche, A., Turak, E., Cao, L., Donner, S.D., Katariya, V., Bernard, R., Holland, R.A., Hughes, A.F., O�Hanlon, S.E., Garnett, S.T., Sekercioglu, �.H., Mace, G.M., 2013. Identifying the World�s Most Climate Change Vulnerable Species: A Systematic Trait-Based Assessment of all Birds, Amphibians and Corals. PLoS ONE 8, e65427. doi:10.1371/journal.pone.0065427