Automated analysis of volcano imagery with machine learning techniques

点击翻译按钮获取中文摘要

Volcanic gas emissions provide an insight into subsurface magmatic processes; their accurate measurement is therefore critical to volcano monitoring and research. Indeed, gas measurements have been pivotal to eruption forecasting in a number of places, saving many thousands of lives through subsequent timely evacuations. Due to its low background atmospheric concentration and distinctive absorption features at ultraviolet (UV) wavelengths, sulphur dioxide (SO2) is the most commonly monitored gas with remote sensing instruments. In recent years, the volcanology group at Sheffield have developed low-cost UV cameras for SO2 detection, with permanent instruments now deployed on 6 volcanoes worldwide. Vast amounts of valuable data are therefore now being collected; however, automation of retrieval algorithms for the accurate determination of SO2 emission rates remains troublesome: typically, substantial expert-user interaction is required to ensure the data are processed in a robust manner.This PhD project aims to address the above issue by developing a fully automated processing workflow for accurate retrieval of SO2 emission rates. To achieve this goal, the student will incorporate computer vision and machine learning algorithms into an easy-to-use tool that is robust to a wide range of measurement conditions. Furthermore, they will aim to accurately quantify measurement uncertainties, which are often neglected or poorly constrained. Having developed their tool, the student will analyse datasets from two active volcanoes (El Reventador, Ecuador; Kilauea, US) to further our understanding of volcanism at these highly- active sites.