PhD Defence by: Casper Samsø Fibæk

Earth Observation and Deep Learning for Sustainable Development: New Approaches to Facilitate Data-Informed Decision-Making


25.11.2022 kl. 13.00 - 16.00


Earth Observation (EO) is uniquely suited to support the Sustainable Development Goals (SDGs). It plays a central role in monitoring the Indicator Framework and directly contributes to several goals at both the Biosphere and Society levels. EO can support, track, and guide efforts towards reaching an estimated 1/4th of the targets and 1/8th of all the indicators. A key contribution of EO is in enabling global comparisons and coherent monitoring efforts. This PhD project shows that these patterns can be monitored, measured, modelled, and predicted using EO data from sources such as the Copernicus Programme and Deep Learning (DL) methods.

The European Union’s Copernicus programme alone generates approximately 16 terabytes of data per day, and analysing this large amount of data is a significant challenge. Artificial Intelligence and DL are uniquely suited to take on the challenge of analysing and making sense of these datasets. Recent advances in DL algorithms, the accessibility of cloud computing, and improved processing power mean that it is increasingly viable to undertake complex global monitoring tasks without expensive setups. This PhD presents tools to assist Financial Service Providers, without subject matter experts, in setting up EO and DL analytical workflows using free and open software and data.

Despite the increased accessibility of EO data and data processing power, there are still major issues to tackle: There is a lack of easy-to-use monitoring interfaces and accessible endpoints for non-subject matter experts for information derived from satellite data. There is also a lack of labelled datasets and generic DL models tailored to ingest multi-sensor EO data. High quality, timely data with good geographical coverage is necessary to enable data-informed decision-making. Nevertheless, while countries in the Global South face major challenges, such as sustainable urban planning, less labelled data is available to inform decision-making. One way to alleviate this is by training DL models on labelled datasets from data-rich countries and carefully applying them in the Global South. Such efforts can lessen data inequality, support the SDGs, and facilitate create new business cases.

This PhD presents a series of methods and tools to create globally applicable DL models for predicting structural characteristics and population density while offering examples of how this data can be made accessible in easy-to-use decision support systems for non-EO experts. The models, tools, and datasets are available for anyone to use and further develop. The tools have been applied and are undergoing continuous development with partners in Denmark, Egypt, Ghana, and Tanzania to create business cases in the areas of Financial Inclusion and AgriTech.

Assesment committee
Professor Sebastian van der Linden | Universität Greifswald, Germany CEO Inge Sandholt | Sandholt ApS, DK
Professor Ehsan Forootan (Chair) | Department of Planning, AAU, DK


Main supervisor Professor Carsten Kessler | Department of Planning, AAU, DK 
Co-supervisor Head Of Department, Mapping & GIS Laurids Rolighed Larsen | Niras, DK 
Co-supervisor Professor Jamal Jokar Arsanjani | Department of Planning, AAU, DK

Associate Professor Lars Bodum | Department of Planning, Aalborg University, DK


Department of Planning, Aalborg University


A.C. Meyers Vænge 15, 2450 Copenhagen, Auditorium ACM15(A) 1.008 and online via Zoom

Tilmelding inden

21.11.2022 kl. 12.00

Tilmeld dig til



Se listen