The Global-Local Loop

In a native multi-source world, a key issue is to adequately and optimally exploit all available geospatial, even rare, sources. Community and discipline bias exist and siloed reasoning has dominated for numerous years.
The geospatial-related communities should now devote more research efforts in higher cross-fertilization, barely documented fusion configurations and how to take advantage of main outputs of other communities or disciplines for key applications.

arXiv paper ISPRS Paper

Which limitations today ?

1. Insufficient and asymmetric exploitation of the various categories of data sources. Such exploitation depends on the application and the domain the contributors fall into. This is a community-based bias: every community defines what is auxiliary source. Each geographical extent and data source has its specific contribution. Many approaches limit themselves to two of them and narrow down the full potential. It often comes from the fact that one spatial extent is often a master extent, from which the core information is extracted. The slave one(s) is(are) only inserted for a specific step (supervision or alternative input source), without (i) being appropriately handled and (ii) retroaction loop, for the benefit of this slave extent.

2. Limited two-way interactions. Few initiatives adopt a back-and-forth or retroaction strategy. Slave level(s) can help improving the master one in terms of spatial, temporal, semantic information. This can be cast into the lifecycle data assessment or critical analysis of the sources paradigms.

Which situation today ?

We evaluated the current level of interactions between data sources here. One may have first a look on how we categorized such sources here.

Extending AI models

First, among the well-known AI generic model-centric issues, few are highly relevant in our context:

• Handling unseen sources.
• Federated and knowledge-based learning.
• Handling domain shifts, domain generalization issues.
• Mitigating bias transfer.
• Estimating uncertainty and moving to interpretability.

Real-world benchmarking and validation

Heavily related to the previous point, key aspects are:

• Consistent and muti-faceted benchmarks.
• Spatially and temporally consistent reference data across data sources and extents for validation.
• Handling genuine applications (economics, history, social sciences) to help understanding complex cases.

Adopting a user-centric perspective

Pure predictive performance on mainstream data sources is over, which calls for:

• Disantangling data producer, model designer, and end-user perspectives.
• Adopting a critical analysis of the sources or a hermeneutics perspectives.
• Quo vadis human-centric machine learning ?
• Enforcing source diversity.

Discoverability, and reuse of existing research

Open models and data does not suffice to comply with FAIR principles.

• Closing the gap in the discoverability and comparability of available benchmarks, models, and algorithms across research communities;
• Improving the tools for providing curating and comparable research resources, and user feedbacks on such resources.

Key references

• Arribas-Bel, D., 2014. Accidental, open and everywhere: Emerging data sources for the understanding of cities. Applied Geography, 1, 45-53.
• Baek, E., Park, K., Ko, J., hwan Oh, M., Gong, T., Kim, H.-S., 2025. AI should sense better, not just scale bigger: Adaptive sensing as a paradigm shift. NeurIPS.
• Editorial, 2016. So long to the silos. Nature Biotechnology, 34, 357.
• Koldasbayeva, D., Tregubova, P., Gasanov, M., Zaytsev, A., Petrovskaia, A., Burnaev, E., 2024. Challenges in data-driven geospatial modeling for environmental research and practice. Nature Communications, 15(10700).
• Kommers, C. et al., 2026. Computational Hermeneutics: Evaluating Generative AI as a Cultural Technology. Fron- tiers in Artificial Intelligence, 9(1753041).
• Larivière, V., Haustein, S., Börner, K., 2015. Long-Distance Interdisciplinarity Leads to Higher Scientific Impact. PLOS ONE, 10(3), 1-15.
• Shi, F., Evans, J., 2023. Surprising combinations of research contents and contexts are related to impact and emerge with scientific outsiders from distant disciplines. Nature Communications, 14, 1641.
• Stammler-Gossmann, A., 2024. Knowledge exchange in the arctic environmental studies: Bridging science and the local community in dialogue. Polar Science, 41, 101103.