Scalable Hyperspectral Inversion with Uncertainty Quantification
Abstract
The upcoming Surface Biology and Geology (SBG) mission will provide hyperspectral measurements on a global scale with a short revisit time. These observations will allow for addressing an urgent and growing need to monitor water quality in near-real time. To keep up with the massive amount of data from SBG and harness the full potential of hyperspectral observations, we propose a machine learning-based approach that directly predicts water quality parameters (e.g., Chlorophyl-a, Phycocyanin, and concentration of non-algal particles) from hyperspectral observations and is trained on synthesized radiances. We compare two machine learning algorithms for our approach: Mixture Density Networks and Deep Ensembles, both of which provide uncertainty quantifications for their predictions. Here we present the initial results from our algorithm development and show that our method can be applied globally as it is computationally efficient and can abstain from making predictions for observations that are out of distribution.
- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2022
- Bibcode:
- 2022AGUFMGC42D0742H