Period: 01. 03. 2026 – 28. 02. 2029

Funding SOURCE: ARIS

REFERENCE NR.: J7-70247

PROJECT PROGRAMME: basic research project

COORDINATORS AND CONTACTS: assist. prof. dr. Marko Bizjak

Project website: https://gemma.feri.um.si/sphere/

LINKEDIN: /

Estimating environmental phenomena with informed deep learning over Earth Observation data

Abstract:

The escalating impacts of climate change necessitate precise, high-resolution assessments of environmental phenomena, such as Urban Heat Islands and desertification. While the surge in Earth Observation (EO) data provides critical insights, processing it presents significant challenges. Current assessment methods rely on either traditional environmental simulations—which are computationally demanding and often constrained by simplified assumptions—or Deep Learning (DL) algorithms—which require extensive labeled datasets, risk overfitting, and lack scientific interpretability. To overcome these limitations, the SPHERE (Estimating environmental phenomena with informed deep learning over Earth Observation data) project introduces an informed deep learning paradigm. This novel framework integrates the strengths of both current approaches by embedding established physical laws and domain-specific knowledge directly into data-driven DL algorithms. By fusing AI capabilities with domain expertise, SPHERE mitigates the need for massive labeled datasets, reduces computational costs, and prevents overfitting. Ultimately, the project leverages the scalability of AI alongside the explanatory depth of physical models, delivering faster, highly accurate, and scientifically interpretable environmental assessments to drive effective climate adaptation strategies.