T236 Alkaline Lakes on Earth and Beyond: Mineralogy, Geochemistry, and Habitability

Wednesday, 25 September 2024
8:00 AM - 12:00 PM
210D

SPECTRAL MAPPING OF MINERAL PATTERNS AT LAKE NATRON, TANZANIA USING HYPERSPECTRAL REMOTE SENSING AND MACHINE LEARNING METHODS: IMPLICATIONS FOR MARS

Paper number: T236-8
Program: Technical Sessions
Time: 9:40 AM – 9:55 AM

Authors

Abstract

Prospecting for paleolake deposits on Mars relies on remote sensing and was greatly aided by the CRISM hyperspectral spectrometer on the MRO orbiter. On Earth, the Hyperion hyperspectral instrument on the EO-1 satellite provided similar coverage. The spectral signatures of minerals are used to map mineral patterns of the planetary surface, helping reconstruct its history.

We employed spectral mapping combined with fully unsupervised machine learning methods to assist with a source-to-sink analysis of saline-alkaline paleolake deposits of Lake Natron, Tanzania, an analog site for Jezero Crater, Mars. Over three field seasons we collected 149 samples from Pleistocene alkaline paleolake deposits of the Moinik Formation (MF; clays and tuffs), Humbu Formation sediments (HF), recent volcanic ash, interbedded lavas, and younger sands and alluvium. We collected VNIR spectra with a field spectrometer and mineral compositions of samples in the lab using XRD.

We applied cascaded, unsupervised Kohonen Self-Organizing Maps (SOM) and Decision Tree Analysis to classify our samples, using mineralogy determined by XRD. The Elbow and Silhouette methods were used to identify the optimal number of clusters. We also developed a spectral index database using our spectral data following the 12 CRISM spectral parameters listed by Viviano-Beck et al., 2014. We employed a Learning Vector Quantization (LVQ) neural network method with 10-fold cross-validation to identify the best spectral indices for our study area. These spectral indices were then used to generate index maps using Hyperion hyperspectral data of the study area.

The most important CRISM spectral indices include BD530 (Band depth at 530 nm), BD2290 (Band depth at 2290 nm), and DOUB2200H (Si-OH band depths around 2200 nm). Our index images show that the HF, tuffs, and interbedded lavas are rich in ferric iron minerals, while the MF is rich in Si-OH and Al-OH rich minerals such as chert, clay minerals, and Mg-carbonate. Areas with younger sand alluvium had high values in the DOUB2200H index map. This study demonstrates the efficacy of fully unsupervised machine learning models for spectral mapping of paleolake deposits on Earth and their potential applications in mapping similar environments on Mars.

Geological Society of America Abstracts with Programs. Vol. 56, No. 5

doi: 10.1130/abs/2024AM-403410

© Copyright 2024 The Geological Society of America (GSA), all rights reserved. Permission is hereby granted to the author(s) of this abstract to reproduce and distribute it freely, for noncommercial purposes. Permission is hereby granted to any individual scientist to download a single copy of this electronic file and reproduce up to 20 paper copies for noncommercial purposes advancing science and education, including classroom use, providing all reproductions include the complete content shown here, including the author information. All other forms of reproduction and/or transmittal are prohibited without written permission from GSA Copyright Permissions.