ABOUT DATA QUEST

Can AI, primed with data from multiple satellites and local sensor networks, detect bushfires earlier, predict fire behaviour and help emergency services respond more effectively to protect homes, people and valuable natural capital?

More than a hack: a research sprint

• A one week sprint with super motivated teams + talented mentors + huge resources + a deeply focused, interdisciplinary environment.

• Unique outcomes to problems that are difficult to tackle in traditional academic / industrial research settings.

• Developing crucial capabilities for fighting future bushfires.

• Researchers are supported with access to datasets, industry experts and compute.

In 2020 we ran our first thematic Wildfires research program in Australia - the Bushfire Data Quest.

Our talented interdisciplinary teams tackled community-defined challenges and produced amazing results at TRL-3. The four data quest teams were focused on three challenges that are connected/adjacent but different (and that use the same data sets:

1. Order of magnitude improvement in fuel moisture mapping.

   The team build a suite of regression models against ground-truth Live Fuel Moisture Content measurements. Proved concept, however, absolute calibration requires much more extensive ground measurements - exposed data gaps.

2. Early detection from geostationary orbit

   The team applied ML Super Resolution and Image Stacking / Subtraction to historic fire. Earliest detection is at higher signal-to-noise ratio than standard hotspot product.

3. ML fire progression models

   The team searched for signatures of extreme fires in multi-band (optical + infrared) Himawari-8 data. Using PCA and colour-ratio analysis they uncovered potential signatures of PyroCB events before the clouds were visible.