Student information

MSc thesis topic: Cashew Mapping across Continents with Location Information

Cashew mapping emerges as a critical endeavor, particularly in regions like Africa and Southeast Asia, where it stands as a vital cash crop cultivated predominantly by smallholders. Despite its economic significance, the expansion of cashew cultivation has, in some instances, come at the expense of environmental degradation, contributing to deforestation in tropical countries. However, the lack of precise location information on cashew farms presents a significant challenge, exacerbated by the complexities of mapping this crop using remote sensing imagery. Existing efforts to map cashew farms have been limited in scale and accuracy, hindered by factors such as climate variations, seasonality, and planting patterns. While deep learning methods have been explored, their effectiveness at large scales remains questionable. In addressing these challenges, this thesis focuses on leveraging location information, particularly through pretrained location encoders, to enhance the accuracy and scalability of cashew mapping within a deep learning framework.
The Master Thesis will work with Sentinel-2 image data with dense cashew labels from India (Southeast Asia), Côte d'Ivoire & Bénin (West Africa), Tanzania (East Africa) with different spatial and environmental conditions.

This Master Thesis builds on previous work by R. Masolele and a previous Thesis exploring the potential of Cashew mapping across continents.

The focus of this thesis specifically is the role of location information through (pretrained) location encoders and their integration in a deep model architecture.

This Thesis is jointly supervised by R. Masolele (expert in Commodity crop Mapping) & M. Rußwurm (expert in location encoding) from the GRS Lab.

Research Questions

  • Will location information improve the accuracy of cashew mapping?
  • How to implement a deep location–vision model for cashew mapping.
  • How does the performance vary between continents with and without location encoding?

Objectives

  • Implement a joint location-vision model in Pytorch by combining existing implementations
  • Train and evaluate this model within and across countries in Africa and Asia

Requirements

  • Experience with Pytorch, Google Earth Engine (required).
  • Attended the Deep Learning & Remote Sensing courses (required).
  • Attended the Advanced Remote Sensing course (optional).

Expected reading list before starting the thesis research

  1. Masolele, R.N., Marcos, D., De Sy, V. et al. Mapping the diversity of land uses following deforestation across Africa. Sci Rep 14, 1681 (2024). https://doi.org/10.1038/s41598-024-52138-9
  2. Yin et al., (2023). Mapping smallholder cashew plantations to inform sustainable tree crop expansion in Benin. Remote Sensing of Environment, 295, 113695. Retrieved 2023-08-24, from doi: 10.1016/j.rse.2023.113695
  3. Klemmer, K., Rolf, E., Robinson, C., Mackey, L., & Rußwurm, M. (2023). arXiv
    preprint arXiv:2311.17179.

Theme(s): Sensing & measuring; Integrated Land Monitoring