AGEOS formation 2024

A look back at the introductory webinars

Prior to this face-to-face training, four introductory webinars were organised between May and July 2024. These sessions, each lasting 2 hours, enabled AGEOS participants to familiarise themselves with the key concepts and tools used in the project. For more details on these webinars, a dedicated article is available here.

Two concrete case studies: Mondah and Ipassa

The training focused on two specific case studies: the creation of land use maps for the Mondah region, and biomass mapping in Ipassa. These case studies enabled the engineers to immerse themselves in real-life problems and apply deep learning techniques to real data.

An intensive, collaborative programme

For ten days, participants had the opportunity to explore in depth data processing pipelines designed to automate and optimise the analysis of satellite images. The two main tools developed by LSCE researchers are gee-fetch, which enables data to be downloaded and pre-processed from Google Earth Engine, and SPROUT, a deep learning pipeline using convolutional neural networks (CNNs) to analyse satellite data.

Although the programme was tailored to the needs of the participants, it covered several key aspects:

• Introduction to data download pipelines
• Presentation of datasets and observations
• Application of deep learning models for soil analysis and canopy height
• Comparison of the results and discussion of the height and biomass maps with the results obtained by Kayrros, a key partner in the field.

A time for reflection and exchange was also devoted to the relationship between canopy height and biomass, with a day organised at the offices of Kayrros in Paris on 16 December, offering participants an additional perspective on the methods used by this pioneering company in spatial data analysis.

Flexible and constantly evolving training

LSCE researchers, in particular David Purnell (post-doctoral fellow at OFVi) and Gabriel Belouze (PhD student), played a crucial role in developing easy-to-use, open-source pipelines. The aim was to ensure that engineers could not only learn how to use the tools, but could also adapt the modules in real time to best suit the specifics of their case studies.

Exchanges between researchers and participants considerably enriched the learning process. Thanks to the flexible and collaborative approach of this training course, the engineers were able to deepen their knowledge while responding to local issues, directly linked to their missions in the field.

The rest of the training: distance support

The training does not end with the ten days spent at LSCE. Remote follow-up, via regular exchanges by email and videoconferencing, is planned to ensure a lasting appropriation of the tools and support in the next stages of the soil and biomass mapping projects.

Conclusion

This training course marked an important stage in the application of advanced technologies for soil and biomass analysis in Central Africa. As a result, AGEOS is strengthening its in-house capabilities and now has a team of engineers trained in the latest deep learning methods for processing and analysing satellite data, with the aim of better understanding and managing terrestrial ecosystems.