Town Hall on Machine Learning for Land DA
Wednesday, February 22 | 10:00-12:00 EST // 16:00-18:00 CET
Registration: Open until Wednesday, February 15, 2023
Organizers†: Istem Fer (1), Jana Kolassa (2), Natasha MacBean (3), Hannah Liddy (4)
(1) Finnish Meteorological Institute, Finland; (2) NASA GSFC, USA; (3) Western University, Canada; (4) Columbia University/NASA GISS, USA
† Organized by the AIMES Land Data Assimilation Working Group
Overview: Earth observations of the land surface have proliferated over the past decade, and the urgency to use these data to improve land modeling and data assimilation (DA) systems has opened up avenues into the data sciences. The application of machine learning algorithms has led to the improvement of existing (physically-based) retrieval algorithms and, in the context of DA, have been used to construct observation operators that when trained on the model can simultaneously be used to bias-correct the observations. Machine learning has further led to improvements in the analysis of model-data mismatch and has the potential to reduce the costs of computationally expensive DA systems. However, in this nascent area of research, there is a need to exchange knowledge about emerging tools and methods for machine learning that can be used to improve land DA systems. The Land DA Community is organizing this town hall to (1) showcase recent developments in machine learning, (2) provide an opportunity to share and discuss research with your colleagues, and (3) identify next steps that the land DA community should take to advance the development of machine learning tools to improve land DA systems.
In particular, we ask participants to consider:
- In what ways can new machine learning tools be modified and combined with existing land DA systems?
- How can the land DA community support the scattered development of methodologies and foster critical understanding of the trade-offs of methodological choices?
Through posters and lightning talks, attendees can share their methodologies, address challenges, and meet others working in this area. Machine learning applications in land DA is a new area of research, and we are aiming to help shape the dialogue and exchange knowledge about best practices moving forward.
- Learning the spatial variability of photosynthesis parameters Shanning Bao* (1, 2); Nuno Carvalhais (1); (1) Max Planck Institute for Biogeochemistry, Germany, (2) National Space Science Center Chinese Academy of Sciences, China
- Systematic land-model calibration Linnia Hawkins* (1), Daniel Kennedy (2), Katie Dagon (2), Dave Lawrence (2), Pierre Gentine (1); (1) Columbia University, United States, (2) NCAR, United States
- Using History Matching to optimize land surface model performance Nina Raoult* (1) and Philippe Peylin (2); (1) University of Exeter, U.K., (2) Laboratoire des Sciences du Climat et de l’Environnement (LSCE), France
- Scalable land model calibration via likelihood-based emulation Andrew Roberts* (1); Michael Dietze (1); Jonathan Huggins (1); Istem Fer (2), (1) Boston University, (2) Finnish Meteorological Institute
- CASM: A long-term Consistent Artificial-intelligence base Soil Moisture dataset based on machine learning and remote sensing Olya Skulovich* (1), Pierre Gentine (1); (1) Columbia University, United States
- Deep learning optimises model prediction of soil carbon sequestration with big data Feng Tao* (1); Yiqi Luo (2), (1) Tsinghua University, China (2) Cornell University, USA
- Archetype biophysical parameter trajectories Feng Yin*(1); Philip Lewis (1), (1) Department of Geography, University College London