Camera-based environmental and livestock monitoring in northern Kenyan rangelands: from repeat photography to edge-based computer vision
The topic is also suitable for GEM students in track 3 – GEM for Ecosystems & Natural Resources.
Fieldwork:
Fieldwork is foreseen, particularly in case chosen by RANGE-funded MSc students. Moreover, the study will align strongly with RANGE PhD students Benjamin Loloju (on livestock monitoring) and Bob Sammy Mwende (on environmental parameters) who initially focus their activities in Waso Ward, Samburu, Kenya (although scope exists to replicate elsewhere). [Add your text]
Northern Kenya’s dryland ecosystems are socio-ecologically complex landscapes where vegetation, water availability and livestock mobility interact to support pastoral livelihoods. Monitoring these dynamics is crucial for understanding rangeland condition, detecting environmental stress and supporting decision-making around grazing, drought management and conflict prevention.
Digital repeat-photography cameras provide an underexplored but promising source of spatial environmental information. They record rich visual signals such as vegetation spectral response, phenology, human and livestock use of waterpoints and livestock species composition. Despite existing PhenoCam networks, structured repeat camera observations are scarce in many areas around the globe (including northern Kenya), and beyond phenology, these data are rarely transformed into structured environmental indicators. Nonetheless, such observations hold great potential to complement or validate satellite observations or serve as autonomous monitoring tools in remote settings.
This MSc project investigates how information contained in RGB images can be converted into operational, scalable environmental or livestock indicators.
Examples include:
- extracting narrow-band greenness metrics from RGB cameras and comparing their temporal behaviour with Sentinel-2 GCC or other vegetation indices
- analysing visitation and livestock species composition at key waterpoints using computer vision
- assessing diurnal or seasonal movement patterns of animals captured by camera traps
- quantifying grazing pressure or early-season green-up from ground-based repeat photographs.
The research will leverage existing camera data from the RANGE project and collaborators such as local conservation partners in northern Kenya. Additional instrumentation (e.g., camera traps, low-cost edge-AI cameras) may be deployed if the student chooses.
For ambitious students, the project offers a technical extension into TinyML, comparing “unlimited” off-board computing with severely constrained edge devices to assess the trade-off between model accuracy, energy use and long-term autonomy in low-infrastructure environments.
This MSc project links to the RANGE (Resilient Approaches in Natural ranGeland Ecosystems) Project, which supports resilient livelihoods in rangeland ecosystems, as well as to an ongoing collaboration with the International Livestock Research Institute. The research will be conducted in close collaboration with local partners, aligning with ongoing initiatives to build a comprehensive understanding of dryland dynamics.
The overall objective of this research is to evaluate how digital repeat camera imagery can be used to extract meaningful environmental and livestock information at landscape-relevant scales in the Northern Kenyan rangelands.
The project will explore one or more of the following use cases:
- Vegetation vitality monitoring using repeat-RGB photography, deriving greenness proxies (e.g., GCC, GCI, other RGB-based indices) and comparing these with satellite metrics (e.g., Sentinel-2 GCC, NDVI, or EVI).
- Livestock presence, counting or species identification (e.g., cattle, goats, camels) at waterpoints or grazing sites using camera trap imagery.
- General computer-vision–based environmental indicators, such as livestock visitation patterns, rainfall-induced vegetation changes or human/livestock pressure around natural resources.
Students may analyse existing datasets collected within the RANGE project (e.g., waterpoint camera trap images, repeat photography sequences) and/or collect their own field data in collaboration with project partners in Kenya.
Depending on the student’s interest and ambition, an optional TinyML pathway is available. This involves:
- designing and training a lightweight ML model (classification, detection or feature reduction),
- deploying it on an edge device (e.g., ESP32-based camera or Raspberry Pi Zero),
- and evaluating its performance and feasibility for low-power, long-term environmental monitoring.
Supervision and support for embedded programming (C/C++ or MicroPython) will be provided given the student is interested in programming. This optional step is not required for successful thesis completion.
Muthoka, J.M., Antonarakis, A.S., Vrieling, A., Fava, F., Salakpi, E.E., & Rowhani, P. (2022). Assessing drivers of intra-seasonal grassland dynamics in a Kenyan savannah using digital repeat photography. Ecological Indicators142: 109223, https://doi.org/10.1016/j.ecolind.2022.109223
Cheng, Y., Vrieling, A., Fava, F., Meroni, M., Marshall, M., & Gachoki, S. (2020). Phenology of short vegetation cycles in a Kenyan rangeland from PlanetScope and Sentinel-2. Remote Sensing of Environment 248: 112004, https://doi.org/10.1016/j.rse.2020.112004
Although the technical component (computer vision, image analysis, and optionally TinyML deployment) is substantial, the topic is inherently interdisciplinary and therefore suitable for M-SE students. A possible link towards the social domain is to explore in greater depth the local acceptability and embedding of such cameras, given that they may collect privacy-sensitive information and/or are subject to theft and vandalism.