TCEM_SS5 | THERE ARE DATA, BIG DATA AND THERE IS DETERMINISTIC FLOOD MODELLING
Theme:
Theoretical, Computational and Experimental Methods (TCEM)
Short description on the session:
Objections to data-driven flood mapping may begin as ontological (and related to the subject of knowledge) before epistemological – most people working in physical sciences would agree that is inherently human to attempt to know their environment before acting upon it. Hence, predictions should be supported by some sort of understanding of the processes underlying the phenomena to be predicted. It thus may feel unnatural to some that, in many fields, future science is essentially about sensing vast amounts of data from a system, learning from its connections and patterns and acting upon that system without explanatory and predictive physical-based models.
In the context of flood risk, the epistemological argument against data-driven flood mapping is still largely about the pitfall of being unable to generate predictions that are not already encoded in the values or connections of the elements in the data-set.
On the side of the growing community of data scientists, the usual objections against deterministic flood modelling are that it is computationally demanding and requires a large number of (calibrated) parameters. It is thus deemed unsuitable to nowcasting or to generate probabilistic flood maps.
The many conversations we had on these topics with many scientists of different backgrounds usually end with the tepid consensus that both, deterministic modelling and data-driven predictions, have their place and should be complementary. We labour under the impression that this consensus is born out of a diplomatic instinct rather than clear views on what may be the role of data-driven and deterministic modelling.
We thus propose this session with the aim of clarify the arguments about the scope and the ambition of data-driven flood risk mapping and of deterministic modelling of flood processes. We welcome work about:
Theoretical, Computational and Experimental Methods (TCEM)
Short description on the session:
Objections to data-driven flood mapping may begin as ontological (and related to the subject of knowledge) before epistemological – most people working in physical sciences would agree that is inherently human to attempt to know their environment before acting upon it. Hence, predictions should be supported by some sort of understanding of the processes underlying the phenomena to be predicted. It thus may feel unnatural to some that, in many fields, future science is essentially about sensing vast amounts of data from a system, learning from its connections and patterns and acting upon that system without explanatory and predictive physical-based models.
In the context of flood risk, the epistemological argument against data-driven flood mapping is still largely about the pitfall of being unable to generate predictions that are not already encoded in the values or connections of the elements in the data-set.
On the side of the growing community of data scientists, the usual objections against deterministic flood modelling are that it is computationally demanding and requires a large number of (calibrated) parameters. It is thus deemed unsuitable to nowcasting or to generate probabilistic flood maps.
The many conversations we had on these topics with many scientists of different backgrounds usually end with the tepid consensus that both, deterministic modelling and data-driven predictions, have their place and should be complementary. We labour under the impression that this consensus is born out of a diplomatic instinct rather than clear views on what may be the role of data-driven and deterministic modelling.
We thus propose this session with the aim of clarify the arguments about the scope and the ambition of data-driven flood risk mapping and of deterministic modelling of flood processes. We welcome work about:
- conceptual or numerical advances, High Speed Computing and related innovation in flood computing;
- surrogate models, Artificial Neural Networks, Support Vector Regression, decision analysis methods other data-driven techniques;
- hybrid techniques and articulation of data and models, including data-assimilation and Bayesian calibration techniques.
Conveners:
Useful information:
Rui M. L. FERREIRA, Professor, CERIS - Instituto Superior Tecnico, Universidade de Lisboa, Email: ruimferreira@tecnico.ulisboa.pt, Phone: + 351 91 740 59 18
João LEAL, Professor, Department of Engineering and Science, Agder University, Email: joao.leal@uia.no
Leonardo MOURA, Engineer, Instituto Superior Técnico; Universidade de Lisboa, Portugal, Email: leonardo.moura@tecnico.ulisboa.pt
Useful information:
| Duration of session: |
2 hours |
| Duration of oral presentation: | 15 minutes (12 presentation + 3 discussion) |
| Total number of presentations: | 6 - 8 |
| Length of abstract: | 2 pages, following the template of the Congress |
| Special issue on a journal: | Yes, by the Convener; it can be supported by the Organizing Committee |
