Machine Learning Components in Coastal Morphodynamic Models
Abstract
Numerical models in geomorphology often use observational data to validate model predictions by comparing model output to an observed landscape. But perhaps there are other uses in models for the variety of data available on geomorphic processes? I will discuss examples from coastal settings where we use data to develop machine learning parameterizations of small-scale processes within morphodynamic models. We have experimented with both deterministic and probabilistic machine learning-based parameterizations, and there are benefits to both approaches for providing morphodynamic insight. Various reasons can be used to justify the use of a given machine learning technique (and its output), including maximize predictability, emulation of more computationally greedy model components, the need for smooth, continuous, parsimonious expressions for further analysis (e.g., examination of phase portraits), and explicit inclusion of uncertainty (e.g., for ensemble output). Our work with probabilistic machine learning parameterizations of small-scale geomorphic processes pushes us to ask questions about how to explicitly account for uncertainty in geomorphic models.
- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2019
- Bibcode:
- 2019AGUFM.U34B..04G
- Keywords:
-
- 3305 Climate change and variability;
- ATMOSPHERIC PROCESSES;
- 3337 Global climate models;
- ATMOSPHERIC PROCESSES;
- 1824 Geomorphology: general;
- HYDROLOGY;
- 1916 Data and information discovery;
- INFORMATICS