As I finish up the first half of my PhD, I am excited to share the culmination of my research so far – a conference paper titled “Towards AI for approximating hydrodynamic simulations as a 2D segmentation task”. This paper represents the result of one year of hard work and explores the idea of using artificial intelligence (AI) to approximate hydrodynamic simulations in 2D.

Traditional predictive simulations and remote sensing techniques for forecasting floods are based on fixed and spatially restricted physics-based models. These models are computationally expensive and can take many hours to run, resulting in predictions made based on outdated data. They are also spatially fixed, and unable to scale to unknown areas. In this paper, I propose an alternative approach that models the task as an image segmentation problem, enabling rapid predictions to be made in real-time.

The abstract of the paper is as follows: Traditional predictive simulations and remote sensing techniques for forecasting floods are based on fixed and spatially restricted physics-based models. These models are computationally expensive and can take many hours to run, resulting in predictions made based on outdated data. They are also spatially fixed, and unable to scale to unknown areas. By modelling the task as an image segmentation problem, an alternative approach using artificial intelligence to approximate the parameters of a physics-based model in 2D is demonstrated, enabling rapid predictions to be made in real-time.

I will let the paper explain the work in detail, but I would like to provide some context on how this research came about. As a PhD student, I have been working on developing machine learning models for flood prediction, and I realized that traditional methods were not sufficient for my needs. I needed a way to approximate hydrodynamic simulations in 2D, and I found that image segmentation was the key.

Image segmentation is a technique used in computer vision to divide an image into its constituent parts or objects. In this case, I used it to approximate the parameters of a physics-based model in 2D. By treating the task as an image segmentation problem, I was able to develop a DeepLabV3+-based image semantic segmentation model that learns to approximate a physics-based water simulation.

The development of this model was not without its challenges. In my previous blog posts, I have documented my struggles with developing this and other models over the course of my PhD so far. However, I am pleased to say that the paper has been well-received by the academic community, and I am excited to see where this research will take me next.

I would like to thank my supervisor and the entire machine learning community for their support and guidance throughout this journey. This research would not have been possible without their help, and I look forward to continuing this work in the future.

In conclusion, this paper represents a significant milestone in my PhD journey, and I am proud to have had the opportunity to contribute to the field of machine learning for flood prediction. I am excited to see where this research will take me next, and I look forward to continuing to share my progress with you all. Thank you for reading!