Constraining the thermal evolution of Earth's crust through machine learning: Development of fully automated digital fission track analysis

Fission track thermochronology is a temperature-sensitive geological dating technique that provides unparalleled insights into the thermal and tectonic evolution of the Earth’s crust with widespread applications. The method is based on the formation of radiation damage zones, called fission tracks, from the decay of 238U in natural minerals. The retention and length of fission tracks are sensitive to temperatures common in the shallow parts of the Earth’s crust (up to 5km depth).

Thus, by quantifying the number and length distribution of fission tracks through digital microscopy and determining the 238U content via mass spectrometry, the detailed thermal history of a rock can be determined.

Since 2003, the Melbourne Thermochronology Research Group has developed Fission Track Studio (https://fissiontrackstudio.com/), an image analysis software suite that has brought significantly increased automation to the laborious collection and analysis of fission track data. However, persistent difficulties remain in identifying certain classes of tracks and a considerable degree of analyst review is still required to correct for deficiencies in the present algorithms.

This team wishes to collaborate with MDAP to develop a radically new approach to digital fission track analysis based on machine learning. Using an existing database of more than 30,000 photomicrograph stacks with corresponding expert-reviewed image sets, an artificial neural network approach will enable fully automated fission track image analysis to be achieved. Such an advance will revolutionise the methodology by significantly reducing analytical time, removing the influence of observer bias and allowing the wider, non-specialist geoscience community to utilise this powerful technique.