Deep learning detection and quantification of volcanic thermal signals in infrared satellite data
Author
Saunders-Shultz, PabloChair
Lopez, TarynCommittee
Dietterich, HannahGirona, Társilo
Grapenthin, Ronni
Keyword
Volcanic activity predictionVolcanic hazard analysis
Volcanism
Volcanoes
Augustine Volcano
Pavlof Volcano
Redoubt Volcano
Shishaldin Volcano
Metadata
Show full item recordAbstract
Volcanic eruptions pose hazards to human lives and livelihoods (Loughlin et al., 2015). To mitigate these hazards, volcano monitoring groups aim to detect signs of unrest and eruption as early as possible. Prior to eruption volcanoes may show various signals of unrest, including: increased surface temperatures, surface deformation, increased seismicity, increased degassing, and more. Here we focus on one approach to monitor volcanic unrest: detecting high-temperature localized volcanic heat emissions, otherwise known as hotspots. The presence of hotspots can indicate subsurface and surface volcanic processes that precede, or coincide with, eruptions. Space-borne infrared sensors can identify hotspots in near-real-time; however, automatic hotspot detection systems are needed to efficiently analyze the large quantities of data produced. While hotspots have been automatically detected for over 20 years with simple thresholding algorithms, new computer vision technologies, such as convolutional neural networks (CNNs), enable improved detection capabilities. Here we introduce HotLINK: the Hotspot Learning and Identification Network, a CNN-based model to detect volcanic hotspots in VIIRS (Visible Infrared Imaging Radiometer Suite) imagery. We find that HotLINK achieves an accuracy of 96% when evaluated on a validation dataset of ~1,700 unseen images from Mount Veniaminof and Mount Cleveland volcanoes, Alaska, and 95% when evaluated on a test dataset of ~3,000 images from six additional Alaska volcanoes (Augustine Volcano, Bogoslof Island, Okmok Caldera, Pavlof Volcano, Redoubt Volcano, Shishaldin Volcano). Additional testing on ~700 labeled MODIS images demonstrates that our model is applicable to this sensor's data as well, achieving an accuracy of 98%. We apply HotLINK to 10 years of VIIRS data and 22 years of MODIS data for the eight aforementioned Alaska volcanoes. From these time series we find that HotLINK accurately characterizes background and eruptive periods, similar to a threshold-based method, MIROVA, but also detects more subtle warming signals, potentially related to volcanic unrest. In particular, analysis of the Mount Veniaminof record demonstrates that HotLINK is able to detect subtle hotspot signals that are coincident with elevated seismicity, potentially indicative of surface heating due to shallow magma intrusion and/or degassing. We identify three advantages to our model over its predecessors: (1) the ability to detect more subtle volcanic hotspots and produce fewer false positives, especially in daytime imagery; (2) the incorporation of probabilistic predictions for each detection that provide a measure of detection confidence; and (3) its transferability to multiple sensors and multiple volcanoes without the need for threshold tuning, suggesting the potential for global application. HotLINK is able to identify eruptions and potentially precursory warming signals in infrared satellite data, making it a valuable tool for monitoring volcanoes and tracking their heat released over time.Description
Thesis (M.S.) University of Alaska Fairbanks, 2024Table of Contents
Chapter 1: Introduction -- Chapter 2: Automatic identification and quantification of volcanic hotspots in Alaska using HotLINK: the Hotspot Learning and Identification Network -- Chapter 3: Overall conclusions -- References -- Appendix -- A. U-net code -- B. Image augmentation validation -- C. Optimizing hysteresis thresholds -- D. Optimizing MIROVA thresholds -- E. Additional HotLINK detection examples.Date
2024-05Type
ThesisCollections
Related items
Showing items related by title, author, creator and subject.
-
Three dimensional volcano-acoustic source localization at Karymsky Volcano, Kamchatka, RussiaRowell, Colin; Fee, David; Christensen, Doug; West, Michael (2013-12)We test two methods of 3-D acoustic source localization on volcanic explosions and small-scale jetting events at Karymsky Volcano, Kamchatka, Russia. Recent infrasound studies have provided evidence that volcanic jets produce low-frequency aerodynamic sound (jet noise) similar to that from man-made jet engines. Man-made jets are known to produce sound through turbulence along the jet axis, but discrimination of sources along the axis of a volcanic jet requires a network of sufficient topographic relief to attain resolution in the vertical dimension. At Karymsky Volcano, the topography of an eroded edifice adjacent to the active cone provided a platform for the atypical deployment of five infrasound sensors with intra-network relief of ~600 m in July 2012. A novel 3-D inverse localization method, srcLoc, is tested and compared against a more common grid-search semblance technique. Simulations using synthetic signals indicate that srcLoc is capable of determining vertical source locations for this network configuration to within �150 m or better. However, srcLoc locations for explosions and jetting at Karymsky Volcano show a persistent overestimation of source elevation and underestimation of sound speed by an average of ~330 m and 25 m/s, respectively. The semblance method is able to produce more realistic source locations by fixing the sound speed to expected values of 335 - 340 m/s. The consistency of location errors for both explosions and jetting activity over a wide range of wind and temperature conditions points to the influence of topography. Explosion waveforms exhibit amplitude relationships and waveform distortion strikingly similar to those theorized by modeling studies of wave diffraction around the crater rim. We suggest delay of signals and apparent elevated source locations are due to altered raypaths and/or crater diffraction effects. Our results suggest the influence of topography in the vent region must be accounted for when attempting 3-D volcano acoustic source localization. Though the data presented here are insufficient to resolve noise sources for these jets, which are much smaller in scale than those of previous volcanic jet noise studies, similar techniques may be successfully applied to large volcanic jets in the future.
-
Analysis and interpretation of volcano deformation in Alaska: Studies from Okmok and Mt. Veniaminof volcanoesFournier, Thomas J.; Freymueller, Jeffrey; Larsen, Jessica; Cervelli, Peter; Christensen, Douglas; McNutt, Stephen (2008-12)Four studies focus on the deformation at Okmok Volcano, the Alaska Peninsula and Mt. Veniaminof. The main focus of the thesis is the volcano deformation at Okmok Volcano and Mt. Veniaminof, but also includes an investigation of the tectonic related compression of the Alaska Peninsula. The complete data set of GPS observations at Okmok Volcano are investigated with the Unscented Kalman Filter time series analysis method. The technique is shown to be useful for inverting geodetic data for time dependent non-linear model parameters. The GPS record at Okmok from 2000 to mid 2007 shows distinct inflation pulses which have several months duration. The inflation is interpreted as magma accumulation in a shallow reservoir under the caldera center and approximately 2.5km below sea level. The location determined for the magma reservoir agrees with estimates determined by other geodetic techniques. Smaller deflation signals in the Okmok record appear following the inflation pulses. A degassing model is proposed to explain the deflation. Petrologic observations from lava erupted in 1997 provide an estimate for the volatile content of the magma. The solution model VolatileCalc is used to determine the amount of volatiles in the gas phase. Degassing can explain the deflation, but only under certain circumstances. The magma chamber must have a radius between ~1and 2km and the intruding magma must have less than approximately 500ppm CO2. At Mt. Veniaminof the deformation signal is dominated by compression caused by the convergence of the Pacific and North American Plates. A subduction model is created to account for the site velocities. A network of GPS benchmarks along the Alaska Peninsula is used to infer the amount of coupling along the mega-thrust. A transition from high to low coupling near the Shumagin Islands has important implications for the seismogenic potential of this section of the fault. The Shumagin segment likely ruptures in more frequent smaller magnitude quakes. The tectonic study provides a useful backdrop to examine the volcano deformation at Mt. Veniaminof. After being corrected for tectonic motion the sites velocities indicate inflation at the volcano. The deformation is interpreted as pressurization occurring beneath the volcano associated with eruptive activity in 2005.
-
Satellite thermal remote sensing of the volcanoes of Alaska and Kamchatka during 1994-1996, and the 1994 eruption of Kliuchevskoi volcanoWyatt, W. Christopher (2001-05)The Advanced Very High Resolution Radiometers on the NOAA polar orbiting satellites were used to routinely observe the volcanoes of Alaska and Kamchatka from May 1994 to July 1996, as part of the monitoring effort of the Alaska Volcano Observatory. The largest eruption observed during this period occurred at Kliuchevskoi Volcano between September 8 and October 2, 1994. Radiative temperature measurements made during this eruption were used to develop quantitative methods for analyzing volcanic thermal anomalies. Several parameters, including maximum temperature, anomalous pixels, and total volcanic signal (TVS), were compared to viewing angle and date. A new quantity, TVS7, may most effectively monitor the temporal evolution of the eruption using thermal data. By combining several observations of the thermal state of the volcano, the general nature of the volcanic activity can be described. These observations may indicate an elevation in temperature twelve to 24 hours before an ash-producing event.