CIOREVIEW >> Data Analytics >>

Data-driven Discovery in Earth Geosciences by ML

By CIOReview | Tuesday, June 4, 2019

FREMONT, CA: The earth, its oceans, and the atmosphere form a complex interacting ecosystem. Many processes that aid in the explanation of the Earth’s geosystem characteristics range from atomic to planetary in the spatial scale and milliseconds to billions of years in the temporal scale. Interactions among the physical, chemical, and biological entities have a significant influence on the complex geosystems that make up Earth. The increasing urbanization in hazard-prone areas is much impactful, and still, the exposure continues. Geoscientists face challenges in extracting useful information to gain new insights from the data, with simulations. With rapidly advancing Machine Learning (ML) technologies, some hope can be restored. 

The convergence of supercomputers with progressive ML algorithms and the readily available datasets in geosciences have resulted in an enormous advantage. This advantage, when applied in the field, can unleash insights in the three following categories:

• Complicated predictive tasks, which cannot be described by a set of explicit commands, are carried out autonomously.

• Numerical simulations and interaction theories are represented with modeling and inverse problems with ML technologies.

• Unraveling new patterns, unanticipated structures, and relationships.

ML in solid Earth geosciences are advancing rapidly, but still prevalent in early stages, with uneven progress. Many problems can be solved with existing datasets from decades of data sources, of which the majority is unexplored. The latest data sources, like detection and ranging (LiDAR), fiber-optic sensors demand new pathways for access in both volume and character.

By utilizing ML and extracting value from the geoscientific data to the maximum extent will require new methods and gateways for initiating combinations between data-driven methods, ML algorithms capable of leveraging in a weak, limited, or biased labels and physical modeling. Funding that targets the inter-linking of disciplines is lacking, but if improved in the field of data science and ML, will provide the opportunities to reveal route maps to the Earth’s treasure trough.