Grasslands are vital ecosystems in East Africa, supporting pastoralism and providing habitat for wildlife and livestock. Pastoralism, the primary occupation for a significant portion of the population, relies on the mobility of herders to follow rainfall and grazing resources. However, increasing infrastructural development and land claims are restricting this mobility. Grasses are foundational species in these ecosystems, providing various ecosystem services. This study focuses on seven key grass species crucial to pastoralists and wildlife. Climate change poses a significant threat to these grasslands, potentially impacting the availability of grazing land and altering the mobility needs of both pastoralists and wildlife. Existing knowledge on the effects of climate change on these specific grass species in East Africa is limited, highlighting the need for this research. This study aims to address this gap by generating geospatial projections of the current and future distribution of these key grass species under a high-emission scenario (RCP8.5) at the end of the 21st century using a dynamically downscaled regional climate model to capture the region's complex climate.

The literature review highlights the importance of East African grasslands for livelihoods and biodiversity, emphasizing the role of grasses as keystone species. It notes the limitations in current understanding of climate change impacts on these grasses, particularly the lack of detailed, spatially explicit data and the limitations of coarse-resolution global climate models. While some studies exist, such as those analyzing *Cenchrus ciliaris* in North America, comparable assessments for East African species are lacking. This gap in knowledge underscores the urgency for research into the impact of climate change on the region's grasslands.

This study uses a combination of global and regional climate models and species distribution models (SDMs) to project the distribution of seven key grass species in East Africa under a high-emission scenario (RCP8.5) by the end of the 21st century. The Community Earth System Model (CESM) provides global climate data, which is dynamically downscaled using the Weather Research and Forecasting (WRF) model at 9 km resolution. This high-resolution approach is crucial for capturing the fine-scale variations and complex climate patterns of the region. The downscaling process involves using ERA5 reanalysis data to correct for biases in the global model output and then adding the projected climate changes to the ERA5 baseline. Eighteen environmental variables, including bioclimatic data, topographic features, soil properties, vegetation indices, and the human footprint index, are used as predictors in the SDMs. Two machine-learning algorithms, random forest (RF) and boosted regression trees (BRT), are employed to build the SDMs, trained using presence and absence data from multiple sources. Model performance is assessed using metrics such as True Skill Statistic (TSS) and Area Under the Curve (AUC). The final models are applied to both present and future climate data to project changes in species distribution, co-occurrence patterns, and overall range shifts. The study area is divided into physiographic units to facilitate analysis of regional variations.

The study projects significant changes in the distribution of the seven key grass species under the RCP8.5 scenario. Tree cover emerges as the most critical predictor for the distribution of all species. Temperature and precipitation-related variables are also significant predictors, but their importance varies among species. *Cenchrus ciliaris* and *Digitaria milanjiana* are projected to expand their range, while others such as *Cynodon dactylon*, *Cynodon plectostachyus*, and *Cenchrus mezianus* are projected to experience considerable range contractions. The most pronounced negative impacts are projected for the Turkana region, with potential near-complete absence of many studied species. The analysis of species co-occurrence reveals a projected decrease in species diversity in large parts of the study area, particularly the Turkana region and parts of the eastern Kenyan plains. The co-occurrence is expressed as the number of species present at a given grid point. The results show a distinct separation in the study region; western parts of the study area are projected to have an increase in co-occurrence, while eastern parts show a decrease in co-occurrence, particularly in Turkana region and eastern Kenyan plains. Changes in bioclimatic variables are projected to be within the uncertainty range of CMIP5 models, but show more realistic fine-scale patterns and maintain physical consistency between bioclimatic variables. The dynamical downscaling method, along with the two machine learning models, provides a reliable depiction of potential future habitat shifts under the high-emission scenario. A significant limitation of the study stems from the relatively sparse data coverage in some areas of the study domain. The results are therefore interpreted with caution.

The findings highlight the vulnerability of East African grasslands to climate change. The projected range shifts and reduced species co-occurrence have significant implications for pastoral livelihoods and wildlife conservation. The decline in key forage species like *Cynodon dactylon* and *Cenchrus mezianus* could severely impact livestock production, particularly during dry seasons. The projected contraction of *Themeda triandra*, a keystone species important for ecosystem function and biodiversity, is also concerning. The drastic changes in the Turkana region highlight the potential for severe ecological consequences in areas projected to experience increased precipitation and temperature seasonality. While the study focuses on climate change, other factors like land use change, human activities, and invasive species could further compound these effects. The nonlinear interplay between climate and other variables further complicates the ability to directly establish causal relationships for all observed changes in grass distribution patterns. Nevertheless, the study's findings provide valuable insights for conservation planning, resource management, and sustainable development strategies in the region.

This study provides valuable projections of future changes in the distribution and co-occurrence of key grass species in East Africa under a high-emission scenario. The findings emphasize the vulnerability of grasslands to climate change and the potential negative impacts on pastoralism and wildlife. The projected decline in species diversity, particularly in the Turkana region, calls for urgent conservation efforts and adaptive management strategies. Future studies should incorporate additional factors like land-use change, human activities, and invasive species to refine these projections and develop more comprehensive understanding of grassland evolution under future climate scenarios.

The study has some limitations. The sparse data coverage in certain regions, especially the eastern Kenyan plains, introduces uncertainty. The analysis relies on a single climate model simulation and emissions scenario (RCP8.5), limiting the generalizability of the results. Further, the model does not consider the effects of active human management practices, such as grass cultivation and reseeding of degraded rangelands. Finally, although the study uses high-resolution data and sophisticated modeling techniques, it is important to note that SDMs are inherently correlative and may not fully capture the complex interactions between species and their environments.