New Machine Learning Approach for Inverse Modeling of Atmospheric Sources from Discrete Hit or Miss Data
Abstract
Inverse modeling is used to infer information about atmospheric sources by combining sensor measurements with atmospheric models. The measurements used in the inversion usually describe quantities with values that vary continuously, like air concentrations or surface deposition. However, binary or discrete data may be available from atmospheric sensors that register only hits or misses, but standard inversion algorithms may not be designed to handle non-continuous data. Here we present a new algorithm based on ensemble machine learning to infer locations of atmospheric sources from discrete sensor data. The method is probabilistic in nature, yet requires no assumptions about the forms of underlying probability distribution functions. The new algorithm is tested on a hypothetical release of a radioactive tracer and is compared to a previous method that incorporates discrete data into a traditional cost function and is optimized using using neural networks.
- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2019
- Bibcode:
- 2019AGUFMGC43D1364L
- Keywords:
-
- 0555 Neural networks;
- fuzzy logic;
- machine learning;
- COMPUTATIONAL GEOPHYSICS;
- 1626 Global climate models;
- GLOBAL CHANGE;
- 1942 Machine learning;
- INFORMATICS;
- 4313 Extreme events;
- NATURAL HAZARDS