Coastal AI
Machine Learning (ML) is part of my everyday activities. I started the endeavor in 2017 with a ML camp at DHI where we learned the fundamentals and applied on various environmental applications. I evolved from using simple regression models to more complex deep neural networks such as GANs. llllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllll..loading 90%
In the news !
In April 2026, at the 6th Baltic Earth Conference in Heringsdorf, Germany, we presented our ongoing research on the use of Generative Adversarial Networks (GANs) for downscaling climate model outputs (e.g., ERA5, NORA3) from coarse spatial resolutions (0.25–1°) to kilometer-scale grids (poster below).
This work aims to enable the downscaling of climate change projections to resolutions suitable for regional and local coastal applications, including wave dynamics, hydrodynamics, and sediment transport modeling.
Such high-resolution datasets are essential for improving the accuracy of coastal impact assessments and for supporting the development of adaptation and mitigation strategies. Ultimately, these advances contribute to better anticipating the effects of climate change and preparing coastal communities for future conditions, including extreme events such as flooding and acute erosion.
Coastal Machine learning applications
(2024) I implemented a genetic programming algorithm to predict dune equilibrium dimensions. New predictors outperform classic predictors in the litterature, see in DUNAMICS
(2018) Time series of water levels in this Danish lagoon are incomplete due to sensor problems and weather conditions. Using a Multi Layer Perceptron, I was able to reconstruct time series of water levels for two stations in the lagoon.
(2025-2026) I am applying Random Forest algorithms to predict sedimentation, erosion rates, and mean depth evolution in various areas of the Danish Harbor of Hvide Sande.
(2019) Using a Multi Layer Perceptron, I was able to replicate sediment transport rates generated by the DHI sand transport routine (STP) routine with good accuracy. This helps to reduce computing times by avoiding running the STP tables.
(2025-2026) In this application, I am using Generative Adversarial Networks (GAN) to downscale wind and pressure fields over various geographical areas and thereby augmenting the resolution by a factor ten.