In 2023, SMHI was awarded development access to use the GPU partition of LUMI supercomputer. The purpose was to develop its own handwritten text recognition (HTR) models, integrated into an application known as Dawsonia. Dawsonia combines computer vision algorithms and machine learning HTR models to digitize tabular data. Throughout its development period and after the compute allocation in LUMI ended, Dawsonia received expert assistance on technical improvements from members of ENCCS.
In 2025, a web based demo app was deployed on HuggingFace, built using Dawsonia . The goal of this application was to present to the wider audience the capabilities of the project. The demo also provides the current developers of Dawsonia an interactive tool to inspect the correctness of the results, by overlaying the original images with its digital equivalent. It was revealed during RISE Computer Science open house and is now publicly available for you to try.
We presented the method as a poster during the 2025 Nordic Workshop on AI for Climate Change. The workshop acted as a melting pot for people from industry and researchers trying to make a difference. The attendees demonstrated their knowledge in analyzing and finding solutions for climate change while making use of scientific AI models. It may sound as an oxymoron to have AI and tackling climate change mentioned in the same sentence. However, to quote the opening talk of in the workshop, “just like science, AI is neither good or bad”. By making targeted studies, by being mindful of the energy impact and optimizing it, these researchers showed how AI and scientific machine learning models can be used to reap a net benefit.
Acknowledgements
We thank our contacts in SMHI for the continued dialogue and for open-sourcing the application. Martin Simonsson (RISE, ex-ENCCS) is gratefully acknowledged for his guidance during the development phase of Dawsonia. The web based demo was a derived from the solid foundations of a similar application by Swedish National Archives following its own HTR project supported by ENCCS.
