SATELLITE DATA FOR AGRICULTURAL ECONOMISTS: Theory and Practice

Weeks towards exams

Author
Affiliation

David Wuepper, Hadi, and Wyclife Agumba Oluoch

Land Economics Group, University of Bonn, Bonn, Germany

Published

February 5, 2026

1 Revision questions

In preparation for your examinations, accomplishing the following tasks individually or in groups can be helpful:

  1. Explain the differences and similarities between traditional machine and modern deep learning approaches to predicting phenomena using satellite images.

  2. What do you understand by the following terms as used in deep learning workflows:

  1. patch size
  2. batch size
  1. You have binary coffee plantation maps derived from ensemble of several machine learning methods, Google Embeddings, GeoTessera embeddings, and Deep Learning approaches.
  1. State and explain the evaluation metric you prioritize if the goal is to ensure smallholder farmers’ land is correctly identified. (2 marks)
  2. Discuss how the metric results could inform subsidies or incentives for sustainable coffee production. (4 marks)
  3. Explain how you would assess encroachment into protected areas using the binary maps. (4 marks)

  1. Imagine you have an image segmentation task to generate a binary map of coffee farms in Colombia. However, your model predicts more non-coffee areas as coffee farms. Describe the steps you would take to troubleshoot such a problem.

  2. A spatial machine learning model trained on cloud-free Sentinel-2 data and canopy height data for the year 2024 is applied to similar imagery for the year 2020 of the same region.

  1. What are some of the risks associated with such temporal transfer? (3 marks)
  2. How can the Area of Applicability (AoA) help mitigate these risks? (4 marks)
  3. Policy-wise, why is it important to validate temporal transfers before using predictions for land use planning? (3 marks)

  1. Discuss important ethical considerations when building predictive machine and deep learning models.

  2. Binary cocoa maps show expansion over a 5-year period in an area

  1. Describe two ways cocoa expansion can negatively affect local biodiversity. (4 marks)
  2. How can binary cocoa maps be used to monitor habitats for endangered species? (3 marks)
  3. Suggest two policy interventions that could balance cocoa expansion and biodiversity conservation. (3 marks)
  1. Establishment of rubber plantations is planned by the ministry of agriculture near wildlife corridors.
  1. Explain how binary maps can help detect potential conflicts. (4 marks)
  2. Suggest one method to quantify impact on wildlife movement. (3 marks)
  3. Recommend a policy intervention to mitigate conflict between agriculture and wildlife. (3 marks)
  1. Explain what any five of the following terms mean in deep learning as applied to spatial modeling.
  1. Patch size (2 marks)
  2. Batch size (2 marks)
  3. Sampler (2 marks)
  4. Data loader (2 marks)
  5. Augmentation (2 marks)
  6. Epoch (2 marks)
  7. Confusion matrix (2 marks)
  8. Optimizer (2 marks)

  1. Discuss the pros and cons of classical machine learning and deep learning in segmenting image for cropland classification. (10 marks)

  2. Describe the major stages of building machine learning workflow for agricultural and environmental economics. (10 marks)

All the best in your endeavors!

2 References

Cardille, J. A., Crowley, M. A., Saah, D., & Clinton, N. E. (Eds.). (2023). Cloud-based remote sensing with google earth engine: fundamentals and applications. Springer Nature. https://doi.org/10.1007/978-3-031-26588-4

Wuepper D, Oluoch W A and Hadi H 2025 Satellite Data in Agricultural and Environmental Economics: Theory and Practice Agricultural Economics 56 493–511 https://doi.org/10.1111/agec.70006

Wuepper D, Wegner J D, Mileva N, Bouchat J, Chen S, Ortiz-Bobea A and Finger R 2026 Harnessing satellite data for the next generation of agri-environmental policies Environ. Res. Lett. 21 011002 https://doi.org/10.1088/1748-9326/ae2d77