Carbon Accounting for Drone Image Datasets and AI Models

Design a framework to estimate and document emissions of drone imagery datasets and AI models.

MSc Thesis Wageningen University & Research Open

Overview

Drone imagery is increasingly used to train AI models for agriculture, ecology, and environmental monitoring. However, the carbon emissions linked to producing, storing, and using these datasets and models are rarely quantified.

Description

Large collections of drone (UAV) images are becoming a key resource for AI-driven analysis in domains such as precision agriculture, biodiversity monitoring, and land-use assessment. While these datasets enable powerful models, they also introduce hidden environmental costs related to data acquisition flights, storage, processing, and model training and evaluation.

This thesis focuses on estimating and structuring the carbon emissions associated with drone image datasets and the AI models built on top of them. The goal is to design a dataset-centric framework, conceptually similar to a dataset hub, where drone image datasets are documented together with their associated carbon footprints and the emissions of AI models trained or evaluated using them.

Objectives

This project is connected to ongoing research in Green AI and sustainable data-centric AI. Rather than optimizing models alone, the thesis focuses on emissions at the dataset and usage level. The following aspects are of interest:

  1. Carbon estimation for drone image datasets - define how to estimate emissions linked to drone flights, data transfer, storage, and preprocessing of image datasets.
  2. Carbon accounting for AI models using drone imagery - quantify emissions related to training and evaluating AI models that rely on these datasets.
  3. Design of a dataset and model documentation framework - propose a structured way to document datasets, models, and their associated carbon emissions, inspired by dataset cards and model cards.
  4. Comparative evaluation and decision support - explore how carbon metrics can support more sustainable choices between datasets, models, or training strategies.

Tasks

The work in this master thesis entails:

  • Conducting a focused literature study on carbon accounting for AI, data-centric AI, and sustainability of drone-based sensing.
  • Defining a methodology to estimate carbon emissions for drone image datasets and AI model usage, including assumptions and limitations.
  • Designing and implementing a prototype dataset and model registry that stores metadata and carbon estimates.
  • Applying the framework to one or more example datasets and models, and analyzing the results and trade-offs between accuracy and emissions.

Literature

  • Doornbos, J., & Babur, O. (2025). Ending Overfitting for UAV Applications - Self-Supervised Pretraining on Multispectral UAV Data. ISPRS Annals of the Photogrammetry, Remote Sensing and Spatial Information Sciences, X-2-W2-2025, 31-39. doi:10.5194/isprs-annals-X-2-W2-2025-31-2025
  • Strubell, E., Ganesh, A., & McCallum, A. (2019). Energy and policy considerations for deep learning in NLP. ACL 2019, 3645-3650.
  • Lacoste, A., Luccioni, A., Schmidt, V., & Dandres, T. (2019). Quantifying the carbon emissions of machine learning. arXiv:1910.09700.
  • Henderson, P., Hu, J., Romoff, J., Brunskill, E., Jurafsky, D., & Pineau, J. (2020). Towards the systematic reporting of the energy and carbon footprints of machine learning. Journal of Machine Learning Research, 21(248), 1-43.
  • Verdecchia, R., Sallou, J., & Cruz, L. (2023). A systematic review of Green AI. WIREs Data Mining and Knowledge Discovery, 13(4), e1507. doi:10.1002/widm.1507
  • Duran, P., Castano, J., Gomez, C., & Martinez-Fernandez, S. (2024). GAISSALabel: A Tool for Energy Labeling of ML Models. ACM Conferences. doi:10.1145/3663529.3663811

Requirements

Courses:

  • Programming in Python (INF-22306)
  • Machine Learning (FTE-35306)
  • Big Data (INF-33806)
  • Optional: Software Engineering

Required skills and knowledge:

  • Programming and data analysis skills
  • Interest in AI, sustainability, and carbon accounting
  • Willingness to work with datasets, metadata, and experimental evaluation
  • Affinity with drone analytics or geospatial data is a plus

Keywords: Green AI, carbon emissions, drone imagery, datasets, data-centric AI, sustainable machine learning, environmental impact

Contact Person(s)

 June Sallou  ·   Onder Babur

Interested? The official listing is available on the JobTeaser platform.