Advancing ecosystem understanding: Identifying the drivers of habitat degradation, species distributions and species vulnerabilities in East African grasslands.

Environments are changing at an accelerated rate, as a consequence of hu- man activity. Many questions remain unanswered regarding the drivers of this change in the landscape, the mechanisms by which species are affected, and patterns of consequential species vulnerabilities. Here I use remote sensing and machine learning to investigate pathways of savannah degradation; use Bayesian species distribution models with data integration to test predictors of range shifts in savannah birds; and evaluate a common climate change vul- nerability assessment framework based on simulated data and foundational concepts. I find that the most degraded savannah sites are those that de- cline in resistance over time and tend to exhibit lower rainfall and higher human and livestock density. However, I show that the same sites do not lose their recovery potential, giving hope for their eventual restoration under correct management. I find that degradation has increased across the whole landscape, and that this increase was lowest for national parks and wildlife management areas, underlining the effectiveness of these management strate- gies for mitigating current degradation trends. Next, I find little support for broad trait-range shift relationships across taxa, for either local extinctions, local colonisations, or total change. This calls into question the usefulness of traits in vulnerability assessments of taxa, where they are applied to wider taxonomic groups. However, I also identify strong species-specific relation- ships among the results, suggesting that more research into those individual species might reveal important trait relationships. Finally, I show that vulner- ability frameworks based on separately assessed species sensitivity, exposure, and adaptive capacity, such as many trait-based approaches, are fundamen- tally unable to accurately predict true vulnerability of species. I showcase how recent advances in species distribution modelling can be applied to develop revised vulnerability metrics.