Although many of us still joke about how inaccurate we think they are, the fact is that weather models are improving all the time. Certain things, however, like lightning, are trickier to predict. The basic science behind the formation of lightning may be nailed down – nuances aside – but that doesn’t mean we can precisely foresee their appearance.

As spotted by ScienceMag, a new study outlines a model that does a fairly good job at doing just that. Writing in the Journal of Geophysical Research: Atmospheres, the team stress that it is, as ever, early days, and the model has some apparent glitches and rough patches, but it’s a solid start on a new way of looking at lightning flashes, which can damage property, planes, and people.

The problem, in a nutshell, is that the processes that produce clouds capable of generating lightning occur on a very small scale – too small, in fact, for the current weather models, at their present spatial resolution, to capture.

The ingredients are simple enough: moisture, a buoyant lifting force (a warmer ground surface, say) and a mass of rising, then unstable, air. Ice particles and water droplets of varying sizes bump into each other, and their electrical charges change.

Their different densities separate them in the cloud layer, which also separates the charges. When the charge segregation becomes too great, a lightning bolt appears to neutralize the electrical imbalance.

Scientists have tried to model various parts of this cycle before. The new study points to one that even tries to simulate the evolving electric field in a storm in order to predict the point when lightning makes its flashy debut.

None, however, have been that successful.

Thanks to the work of the UK Met Office – already working wonders predicting rainfall and the spread of cholera in Yemen – a new model was devised that could image these processes happening on a scale of around 8 kilometers (5 miles), improving the resolution of pre-existing models by an order of magnitude.

This wasn’t easy: they had to remove parts of the pre-existing models, add in some others, and try to see what they could fit together in a virtual jigsaw that permitted this resolution. Eventually, they settled on focusing on precise modeling of the convection part of the process, while keeping the ice formation more ambiguous.

They couldn’t just test it on upcoming events, though, so this time around, they decided to train it on five years of old data to see how well it then simulated the planet’s lightning.

It seemed to work wonders, though: it managed to accurately pick out lightning hotspots around the globe while getting regional intricacies right. Compared to the records of the World Wide Lightning Location Network (WWLLN), a ground-based array, and the spaceborne Lightning Imaging Sensor and the Optical Transient Detector (LIS/OTD), the model, for the most part, was on the money.

Seasonal and annual patterns match up, marking the first time a global simulation has managed to succeed in doing so. Additionally, it also replicated the fact that lightning peaks in the late afternoon and early evening over land, which is about the time the soil has heated up enough to cause an updraft.

There was one key problem, though: it appears to overestimate the lightning flash rate over the oceanic equatorial regions, and it’s unclear as to why this may be. Still, this is impressive stuff, and things can only get better from now as they build on the foundations of this model.