Satellite-Based Lava Flow Morphology Classification at Mount Kīlauea, Hawai'i

Abstract

Accurately mapping lava flow morphologies is important for understanding volcanic processes evaluating hazards, and for supporting fieldwork planning. However, differentiating between the two most common basaltic morphologies, pāhoehoe and 'a'ā, in remote sensing data has been challenging due to their similar mineralogical compositions and spectral overlap. This study evaluates an object-based classification methodology using the freely available Sentinel-1 C-band Synthetic Aperture Radar (SAR) and Sentinel-2 optical imagery to classify the lava flow morphologies on the 2018 lower East Rift Zone (ERZ) eruption of Mount Kīlauea, Hawai'i. Six Segment Mean Shift (SMS) parameter sets and sixteen three-band composites were tested under two validation methods: Definition-based (Pāhoehoe vs. 'A'ā) and Textural-based (Smooth vs. Rough). Classification was done using SMS segmentation followed by K-Means clustering, with results compared to validation datasets created from high-resolution drone imagery. The highest overall accuracies (OA) were 86.31% (Definition-based) using a multi-sensor (Sentinel-1 and Sentinel-2) composite, and 88.69% (Textural-based) using Sentinel-2 RGB bands. Grey-Level Co-occurrence Matrix (GLCM) texture features (contrast, homogeneity, energy, entropy) showed consistent differences between the morphologies, supporting their potential for semi- or fully unsupervised classification, though the analysis here remained exploratory. Terrain analysis showed that Sentinel-2 was more susceptible to misclassifications related to the slope aspects Northness and Eastness, while terrain effects were negligible for Sentinel-1. The overall findings show that freely available multi-sensor data, combined with object-based segmentation and texture analysis, can reliably map lava flow morphologies, highlighting ways for improving automated classification workflows for volcanic morphological mapping.