What is it about?
When a volcano erupts near a populated area, emergency responders need lava flow forecasts fast. The standard simulation tool for this, MrLavaLoba, took nearly 30 minutes per run. A reliable probabilistic hazard map requires thousands of runs, making real-time forecasting impractical during an active eruption. Flowy reimplements the MrLavaLoba method from scratch in modern C++. The physics and statistics are identical to the original, but the code runs up to 400 times faster. A simulation that previously took 23 minutes finishes in about 10 seconds. This speed comes from algorithmic improvements and careful engineering, not from simplifying the model. With Flowy, scientists can generate full ensemble-based hazard maps in minutes rather than days, giving civil protection authorities actionable forecasts while lava is still advancing.
Featured Image
Photo by Toby Elliott on Unsplash
Why is it important?
Probabilistic lava hazard assessment requires running thousands of stochastic simulations with varying source parameters. At 23 minutes per run, the original MrLavaLoba made this infeasible during a crisis. Decisions about evacuation boundaries and barrier placement were made on incomplete data. Flowy removes that constraint. The 400x speedup means a 10,000-run ensemble completes in under two hours on a single machine. For the first time, statistically robust probabilistic forecasts become available on timescales relevant to emergency management. The code is open-source and reads the same input formats as MrLavaLoba, so existing workflows transfer directly.
Perspectives
We started this project after seeing the gap between what probabilistic forecasting requires (thousands of runs) and what the existing code could deliver (one run per 23 minutes). The question was whether a rewrite could close that gap. The answer turned out to be software engineering, not hardware. The original Python/C code had accumulated years of incremental additions. Starting fresh in C++ with better data structures and memory layout gave us the 400x improvement without changing the underlying stochastic model. Every output matches MrLavaLoba to numerical precision. The recent eruptions near Grindavik, Iceland made the practical stakes clear. Flowy now runs operationally for real-time hazard assessment.
Rohit Goswami
University of Iceland
Read the Original
This page is a summary of: Flowy: High performance probabilistic lava emplacement prediction, Computer Physics Communications, October 2025, Elsevier,
DOI: 10.1016/j.cpc.2025.109745.
You can read the full text:
Resources
Contributors
The following have contributed to this page
