NLDL 2024: My rainfall radar paper is out!
Towards AI for approximating hydrodynamic simulations as a 2D segmentation task
......that's the title of the conference paper I've written about my rainfall radar research that I've been doing as part of my PhD, and now that the review process is complete I'm told by my supervisor that I can now share it!
This paper is the culmination of one half of my PhD (the other is multimodal social media sentiment analysis and it resulted in a journal article). Essentially, the idea behind the whole project was asking the question of "Can we make flood predictions all at once in 2D?".
The answer, as it turns out, is yes*.... but with a few caveats and a lot more work required before it's anywhere near ready to be coming to a smartphone near you.
It all sort of spiralled from there - and resulted in the development of a DeepLabV3+-based image semantic segmentation model that learns to approximate a physics-based water simulation.
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 These models are computationally expensive and can take many hours to run, resulting in predictions made based on outdated data.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'll let the paper explain the work I've done in detail (I've tried my best to make it understandable by a wide audience). You can read it here:
https://openreview.net/forum?id=TpOsdB4gwR
Long-time readers of my blog here will know that I haven't had an easy time of getting the model to work. If you'd like to read about the struggles of developing this and other models over the course of my PhD so far, I've been blogging about the whole process semi-regularly. We're currently up to part 16:
Speaking of which, it's high time I wrote another PhD update blog post, isn't it? A lot has been going on, and I'd really like to document it all here on my blog. I've also been finding it's been really useful to get me to take a step back to look at the big picture of my research - something that I've found very helpful in more ways than one. I'll definitely discuss this and my progress in the next part of my PhD update blog post series, which I tag with PhD to make them easy to find.
Until then, I'll see you in the next post!