CIG webinar held to introduce AVNI software

Jan. 13, 2022
  • Open-source Python package with APIs to handle data and compute intensive queries
  • Introduce storage formats or classes for models and processed seismic data
  • Interactive web-based visualization tools for data and model exploration 
  • Formulate and benchmark forward solvers for rapid data validation of models 

Modeling the interior structure of terrestrial planets has become one of the most computationally-intensive, big-data problems in the physical sciences with demonstrated utility in assessing hazard, locating explosions and characterizing plate tectonics. Multi-disciplinary advancements have led to a proliferation of dynamical simulations and model snapshots from seismic tomography. Reconciling seismic models and data with simulations from geodynamics, mineral physics and geochemistry is crucial for robust thermo-chemical interpretations. Such cross-disciplinary initiatives have been impeded by discrepant spatial scales, observational or theoretical assumptions, and lack of data validation algorithms. AVNI is an Analysis and Visualization toolkit for plaNetary Inferences that will handle model and data queries from the 3D reference Earth model (REM3D) project.

We present methods and data formats that facilitate rapid prototyping of multi-scale models by reconciling and assimilating features ranging from reservoir (~0.1 - 10 km) to global scales (~500 - 5000 km). Our approach involves three complementary aspects: (1) Code repositories comprising modular libraries with model classes and scalable HDF5 formats for archival, (2) API (Application Programming Interface) calls for querying model and data evaluations with fast, benchmarked forward solvers, (3) Web-based applets for visualization and outlier analyses. Both (1) and (2) are utilized by (3) and can be accessed on the client side with Jupyter notebooks and command-line tools. 

AVNI aids reconciliation of measurements made using different techniques by identifying (in)consistent features and subsequently models them using a flexible scheme that permits almost instantaneous forward calculations of data. The methodology employs in-memory and filesystem data storage, providing rapid and scalable filtering of Earth models and calculation of seismic observations. By coupling existing, reconciled observations with predictions for arbitrary locations, this application will be a useful tool for identifying regions of scientific interest, validating new techniques, planning future seismic deployments, and testing hypotheses about the Earth's deep interior.