Research programme
PA-AutoML
Creation of a framework for environmental parameter estimation that benefits from the consistency of physics-based theory-driven models and the accuracy of the machine-learning-based data-driven models
- Duration
- 2021 - 2025
- Contact
- Mitra Baratchi
- Funding
- NWO-Klein
To address emergent global environmental problems, geoscientists envision using globally available Earth Observation (EO) data to create a "Digital Twin of Earth" —a comprehensive model of environmental processes. There are currently two sets of modelling paradigms available to create such a digital twin.
On the one hand, there exists a group of widely-adopted theory-driven physical models. Due to their ill-posed nature, these models often lead to imprecise results. Although they do maintain the consistency according to the governing physical rules of the Earth system, their usability in extracting global models is limited due to their inaccuracy.
Data-driven models acquired by using machine learning algorithms, on the other hand, offer substantial benefits in improving the modelling accuracy by capturing patterns in global EO data. They may, however, lead into physically inconsistent results.
In this project we work on a framework for environmental parameter estimation that benefits from the consistency of physics-based theory-driven models and the accuracy of the machine-learning-based data-driven models. In collaboration with researchers at the European Space Agency and Leiden Institute of Environmental Sciences, this project focuses on developing automated machine learning pipelines for retrieval of biophysical parameters. Specifically, we aim to enhance the use of raw EO data while addressing data imperfection issues (e.g., cloud cover, inconsistent resolution), as well as, incorporating radiative transfer models, a specific class of physical models commonly used for estimating biophysical parameters.