Integrating machine learning and process-based models for Earth and Environmental Sciences

Start: Jun 4, 2024
End: Jun 6, 2024
Location: Max Planck Institute for Biogeochemistry
Host: Martin Jung

	June 4, Tuesday	June 5, Wednesday	June 6, Thursday
09:30 - 11:00	1) Martin Jung: Welcome and Introduction (10’) 2) Chaopeng Shen: “Differentiable modeling to improve global hydrologic modeling and water management” (30’) 3) Olivier Bonte: “Differentiable modelling for terrestrial evaporation: the GLEAM perspective” (10’)	Talks & discussion 1) Patrick Gallinari: “Operator learning with neural fields : solving differential equations on general geometries” (30’) 2) Qi Yang: “Advancing flexibility and efficiency in ecosystem data assimilation via know ledge- guided machine learning” (10’)	Talks & discussion 1) Basil Kraft: “Hybrid (?) data - driven river routing” (10’) 2) Vitus Benson: “Hybrid Atmospheric Transport” (10’) 3) Maximilian Gelbrecht: “Towards stable hybrid atmospheric models: PseudoSpectralNet” (10’) Joint recap & synthesis of break out discussion
11:00 -11:30	Coffee	Coffee	Coffee
11:30 - 12:30	Talks & discussion 1) Zhenong Jin: „Knowledge -guided machine learning for the next generation of agroecosystem prediction” (30’) 2) Alexander Winkler: “Hybrid Modelling of Coupled Climate-Carbon-Cycle Interactions Using Neural ODEs” (10’)	Talks & discussion 1) Nathan Doumèche: “Some statistical insights on Physics - Informed Machine Learning” (30’) 2) Lazaro Alonso: “Introduction to mixed -mode automatic differential in Julia” (10’)	Joint recap & synthesis of break out discussion
12:30 - 13:30	Lunch	Lunch	Lunch
13.30 - 15:00	Posters 1) Bernhard Ahrens: „Parameter learning with a soil carbon model” 2) Shijie Jiang: „Hybrid and interpretable machine learning for understanding extreme hydroclimatic events” 3) Jiaxin Xie: „Can data streams from empirical estimation improve hybrid modeling?” 4) Maha Badri & Philipp Hess: “Towards a Hybrid Vegetation Model” 5) Nuno Carvalhais: “Integrating machine learning and mechanistic models for process abstraction and spatial parametrization towards an improved representation of carbon cycle dynamics across scales” 6) Laurent Bataille: tbd	Talks & discussion 1) Christian Reimers: “Integrating prior knowledge into Stable Diffusion For Time Series Prediction” (20’) 2) Markus Reichstein: “Some reflections on critical aspects of hybrid modelling” (20’)	Planning out specific activities and collaborations
15:00 - 16:00	Talks & discussion 1) Zavud Baghirov: “Hybrid Global Modeling of Water - Carbon cycles” (20’) 2) Manuel Alvarez Chaves: “E valuating physics-based representations of hydrological systems through hybrid models and Information Theory” (20’)	Coffee Break & break out group discussion	Concluding remarks (until 15:30)
16:00 -.17:00	Joint discussion with coffe and snacks	break out group discussion
17:00 - 18:00		Hike to Vollradisroda
18:30 - 22:00		Dinner & synthesis

Relevant literature shared by participants

Yin et al. (2022): Augmenting Physical Models with Deep Networks for Complex Dynamics Forecasting

Serrano et al. (2023): Operator Learning with Neural Fields: Tackling PDEs on General Geometries

Doumèche et al. (2024): Physics-informed machine learning as a kernel method

Doumèche et al. (2023): Convergence and error analysis of PINNs

Doumèche et al. (2023): Human spatial dynamics for electricity demand forecasting: the case of France during the 2022 energy crisis

Shen et al. (2023): Differentiable modelling to unify machine learning and physical models for geosciences

Feng et al. (2022): Differentiable, Learnable, Regionalized Process-Based Models With Multiphysical Outputs can Approach State-Of-The-Art Hydrologic Prediction Accuracy

Tsai et al. (2021): From calibration to parameter learning: Harnessing the scaling effects of big data in geoscientific modeling

Bindas et al. (2024): Improving River Routing Using a Differentiable Muskingum-Cunge Model and Physics-Informed Machine Learning

Feng et al. (2023): Deep Dive into Global Hydrologic Simulations: Harnessing the Power of Deep Learning and Physics-informed Differentiable Models

Son et al. (2024): Integration of a Deep-Learning-Based Fire Model Into a Global Land Surface Model Bao et al. (2023): Toward Robust Parameterizations in Ecosystem-Level Photosynthesis Models

Yang et al. (2023): A flexible and efficient knowledge-guided machine learning data assimilation (KGML-DA) framework for agroecosystem prediction in the US Midwest

Liu et al. (2024): Knowledge-guided machine learning can improve carbon cycle quantification in agroecosystems

Kraft et al. (2022): Towards hybrid modelling of the global hydrological cycle
Reichstein et al. (2019): Deep learning and process understanding for data-driven Earth system

science

ElGhawi et al. (2023): Hybrid modelling of evapotranspiration: inferring stomatal and aerodynamic resistances using combined physics-based and machine learning