Vigil

About Vigil

Vigil replays real bushfires in Victoria day by day and tests how well different models predict where a fire will burn next. It's a student research project.

Don't use Vigil to make decisions about your safety. It shows past fires and experimental predictions, not current conditions. For fire information and warnings, use VicEmergency or your state's fire service, and follow official advice.

What the replay shows

Each coloured cell is a 375 m square of land, shown on the day it first burnt. Amber means it's burning that day, copper means it burnt in the last few days, and grey is the scar left behind.

The burnt area comes from MODIS satellite burned-area maps, which give each fire's final outline. The timing within that outline comes from VIIRS satellite fire detections: every cell takes the date of the nearest detection that agrees with the burned-area map to within a few days. This follows Parks (2014). A fire day runs from 6 am to 6 am, so a satellite pass at 2 am counts toward the day whose fire it saw.

What it can't show

The models

Vigil compares simple baselines (assume the fire keeps growing at its edge, optionally pushed downwind), a physics model built on the CSIRO fire-spread equations for forest, grassland and heath, and a neural network trained on past fires. Each one predicts the next day's burnt area and is tested on fire seasons it never saw. The 2019–20 Black Summer season is held back entirely as the final test. Results will appear on the Findings page.

Data and credits

Who made it

Vigil is a portfolio project by Rishi, a first-year data science student at RMIT University. It isn't affiliated with any fire agency.

Method references: Parks (2014), International Journal of Wildland Fire 23(2); Huot et al. (2022), "Next Day Wildfire Spread", IEEE Transactions on Geoscience and Remote Sensing; Cheney et al. (1998, 2012) for the grassland and dry eucalypt forest spread models.