The Spatial and Spectral Resolution of ASTER Infrared Image Data: A Paradigm Shift in Volcanological Remote Sensing

During the past two decades, the Advanced Spaceborne Thermal Emission and Reflection Radiometer (ASTER) instrument on the Terra satellite has acquired nearly 320,000 scenes of the world’s volcanoes. This is ~10% of the data in the global ASTER archive. Many of these scenes captured volcanic activity at never before seen spatial and spectral scales, particularly in the thermal infrared (TIR) region. Despite this large archive of data, the temporal resolution of ASTER is simply not adequate to understand ongoing eruptions and assess the hazards to local populations in near real time. However, programs designed to integrate ASTER into a volcanic data sensor web have greatly improved the cadence of the data (in some cases, to as many as 3 scenes in 48 h). This frequency can inform our understanding of what is possible with future systems collecting similar data on the daily or hourly time scales. Here, we present the history of ASTER’s contributions to volcanology, highlighting unique aspects of the instrument and its data. The ASTER archive was mined to provide statistics including the number of observations with volcanic activity, its type, and the average cloud cover. These were noted for more than 2000 scenes over periods of 1, 5 and 20 years.


ASTER Instrument and History
The Advanced Spaceborne Thermal Emission and Reflection Radiometer (ASTER) instrument was launched on the NASA Terra satellite on 18 December 1999. Prior to the instrument beginning its operational phase on 4 March 2000, ASTER acquired several images including the first thermal infrared (TIR) image on 6 February ( Figure 1). That image was of Erta Ale volcano (Ethiopia), and initiated a long association of ASTER data with volcanological observations. ASTER was developed and built in Japan under the Japanese Ministry of Economy, Trade and Industry (METI), and is one of five Earth observing instruments on Terra. The combined science team of Japanese and United States investigators has changed over the years, however always maintaining a strong volcanological component [1,2]. During the past two decades, the data from ASTER have been applied to numerous questions and scales of surface processes, most notably for volcanic activity, e.g., [3][4][5][6].
ASTER was designed to observe the surface at multiple spatial and spectral resolutions as well as from different viewing geometries. It is actually a suite of three instruments with independent bore-sighted telescopes, originally having 14 spectral channels in the visible/near-infrared (VNIR), the shortwave infrared (SWIR), and the thermal infrared (TIR) regions [7]. The VNIR instrument (0.52-0.86 µm) has three spectral channels at a spatial resolution of 15 m/pixel paired with one channel oriented in a backward look direction for the creation of digital elevation models (DEMs). The SWIR instrument (1.6-2.43 µm) unfortunately failed in 2008, but originally had six channels at a spatial resolution of 30 m/pixel. Finally, and perhaps most important for many aspects of volcanological remote sensing, the TIR instrument (8.13-11.65 µm) has five channels at a spatial resolution of 90 m/pixel. During the lifetime of the mission, ASTER has acquired over 3.5 million individual scenes-approximately 22% of which were collected at night. Here, we describe the two-decade history of ASTER, specific programs to improve the observational frequency of volcanoes, and present examples of those data.
Remote Sens. 2020, 12, x FOR PEER REVIEW 2 of 39 channel oriented in a backward look direction for the creation of digital elevation models (DEMs). The SWIR instrument (1.6-2.43 μm) unfortunately failed in 2008, but originally had six channels at a spatial resolution of 30 m/pixel. Finally, and perhaps most important for many aspects of volcanological remote sensing, the TIR instrument (8.13-11.65 μm) has five channels at a spatial resolution of 90 m/pixel. During the lifetime of the mission, ASTER has acquired over 3.5 million individual scenes-approximately 22% of which were collected at night. Here, we describe the twodecade history of ASTER, specific programs to improve the observational frequency of volcanoes, and present examples of those data.  , is shown using a decorrelation stretch (DCS) of the TIR bands 14,12,10 in red, green, and blue, respectively. Color variations are mainly caused by rock and soil compositional differences, and only possible with the ASTER multispectral TIR data. The blues and purple colors are indicative of dominantly basaltic lava flows. The small cluster of white pixels in the lower central part of the image is the summit lava lake thermal anomaly. The ASTER DCS data are overlain on a visible Google Earth image for context. Image credit: NASA/METI/AIST/Japan Space Systems, and U.S./Japan ASTER Science Team.

Twenty Years of Volcanic Studies Using ASTER Data
The ASTER Science Team created useful derived science products that have benefited numerous scientific fields such as volcanology. These include a robust temperature emissivity separation algorithm for the first orbital high spatial resolution multispectral TIR data [8], programs such as the Science Team Acquisition Request (STAR) to acquire data focused on important science questions of individual users/teams, as well as the ability to generate DEMs [9]. These individual-scene DEMs of 60 by 60 km were later composited globally into the ASTER Global DEM or GDEM-version 3 of which was released in August 2019 [10,11].
Arguably, volcanology is one of the scientific disciplines on which ASTER data have had the greatest impact. In order to quantify this impact and gather all volcanology related studies together into one reference document, we have performed an extensive literature review (Appendix A). This found 271 peer-reviewed publications from 1 January 1995 to 1 December 2019, an average of nearly 11 per year ( Figure 2). Papers published prior to the Terra launch were considered precursory studies, commonly describing how the future ASTER data would be used for certain volcanic studies. This list of 271 publications was subdivided into 12 categories based on the volcanic focus of the papers. The category names and number of papers in those categories (shown in parentheses) are: Analogs (3), Calibration (4), Lava Flows (4), Gas/Plumes (30), Geothermal (9), Mapping (54), Modeling (16), Monitoring (96), Operational (4), Other (4), Precursory (16), and Topography (31). Based on this categorization, ASTER data have been primarily used to monitor volcanic activity; somewhat surprising considering the lower temporal frequency of the data. This speaks to the need for future high spatial, high spectral resolution data at temporal resolutions far better than the nominal 16 day equatorial repeat time of ASTER and Landsat, or even the 5 day resolution of the Sentinel-2 constellation. This same finding was also brought forward in the most recent Decadal Survey for Earth Science, which noted data such as these are critical for addressing two of the most important science questions related to natural hazards [12].

Twenty Years of Volcanic Studies Using ASTER Data
The ASTER Science Team created useful derived science products that have benefited numerous scientific fields such as volcanology. These include a robust temperature emissivity separation algorithm for the first orbital high spatial resolution multispectral TIR data [8], programs such as the Science Team Acquisition Request (STAR) to acquire data focused on important science questions of individual users/teams, as well as the ability to generate DEMs [9]. These individual-scene DEMs of 60 by 60 km were later composited globally into the ASTER Global DEM or GDEM-version 3 of which was released in August 2019 [10,11].
Arguably, volcanology is one of the scientific disciplines on which ASTER data have had the greatest impact. In order to quantify this impact and gather all volcanology related studies together into one reference document, we have performed an extensive literature review (Supplementary materials). This found 271 peer-reviewed publications from 1 January 1995 to 1 December 2019, an average of nearly 11 per year ( Figure 2). Papers published prior to the Terra launch were considered precursory studies, commonly describing how the future ASTER data would be used for certain volcanic studies. This list of 271 publications was subdivided into 12 categories based on the volcanic focus of the papers. The category names and number of papers in those categories (shown in parentheses) are: Analogs (3), Calibration (4), Lava Flows (4), Gas/Plumes (30), Geothermal (9), Mapping (54), Modeling (16), Monitoring (96), Operational (4), Other (4), Precursory (16), and Topography (31). Based on this categorization, ASTER data have been primarily used to monitor volcanic activity; somewhat surprising considering the lower temporal frequency of the data. This speaks to the need for future high spatial, high spectral resolution data at temporal resolutions far better than the nominal 16 day equatorial repeat time of ASTER and Landsat, or even the 5 day resolution of the Sentinel-2 constellation. This same finding was also brought forward in the most recent Decadal Survey for Earth Science, which noted data such as these are critical for addressing two of the most important science questions related to natural hazards [12].  The number of volcano-related publications per year that have incorporated ASTER data in some aspect of the work. Through 2016, the growth has been roughly exponential (dashed line). That growth has declined somewhat in the last several years, although still remaining respectable. The total number of publications shown is 271, spanning the last 25 years. made to the instrument design. Following its launch, ASTER data greatly increased both the spatial and spectral ranges, providing new capabilities from that of Landsat, significant especially for volcanology.
The need to continue (and improve upon) this class of measurement was recognized in the first Decadal Survey (DS) for Earth Science in 2007 [23]. In that report, a notional list of mission concepts was proposed, which included the Hyperspectral Infrared Imager (HyspIRI) mission that paired a hyperspectral visible/shortwave infrared (VSWIR) instrument with a multispectral TIR instrument. Many years of planning, design and science concept studies were later performed. Ultimately, however, the mission was never selected by NASA to move forward and the concept study was formally ended with the publication of the second Decadal Survey in 2018 [12].
The second report focused on science questions and key observables, around which new mission concepts could be structured. Notably, the lack of future global TIR and VSWIR image data was once again brought forward in the new DS report. For example, the requirement for infrared measurements spanned all of the working group panels, being mentioned over 190 times in the report and 12 times in the requests for information from the general science community. Focusing specifically on the reports of the Ecosystems and the Earth Surface and Interior panels, which contributed most directly to the need for TIR data, those data were key inputs for 14 different science objectives-half of which were designated as "most" or "very" important, the top categories. The science and applications summary includes the need for measurements of surface geology, active geologic processes such as natural disasters, surface/water temperature, as well as functional traits of vegetation and ecosystems. The Earth Surface and Interior panel, for example, focused two of its top-level science questions on natural disasters: data prior to the event and the outcomes following. TIR measurements are noted as vital for several of these disasters, including volcanoes, landslides and wildfires. Although temperature is an important measurement, the need for vast improvements in TIR spatial, spectral and temporal resolution data was made clear. TIR data acquired in 1-2 channels at 100 m resolution every 2 weeks no longer satisfies the requirements deemed important by the science community for the future. Hyperspectral TIR coupled with vastly improved temporal resolution at spatial scales exceeding current ASTER capabilities were recommended. The DS report lists several Designated Observables (DOs), including the Surface Biology and Geology (SBG) DO. A mission concept designed to address this DO will likely include some combination of these recommended scales of TIR data, which will allow far more detailed volcanic measurements than currently possible with either ASTER or Landsat data.

The ASTER Volcano Science Team Acquisition Request (STAR)
The recognition that ASTER data would provide a fundamentally new tool for volcanic observations was documented even before the Terra launch [7]. During that development period, the ASTER Science Team (AST) foresaw that the instrument would eventually provide data at spatial and spectral scales never before observed, routine data at night, the ability to point off-nadir to improve temporal revisit time, and therefore, required daily observation schedules. Being a scheduled instrument, unlike many other nadir-viewing systems, brought both a higher level of mission complexity as well as unique opportunities for Earth observations. Scheduling allowed specific ground targeting, a focus on larger-scale global processes, as well as an important goal of creating a global map of ASTER data.
One observational and scheduling strategy developed was the creation of the science team acquisition request or STAR. The STARs are a series of globally distributed regions of interest (ROIs) over targets with high scientific value. The ROIs had associated attributes such as seasonality, instrument gain settings, number of observation attempts per year, etc. These targets were integrated into the daily scheduling so that both STAR-focused scenes were acquired together with the many other required observations during any given orbital period. The STAR's assured a priority set of observations for high-interest science such as monitoring the global land ice inventory, change detection of large urban environments, and volcanic activity [24][25][26].
The ASTER Volcano STAR's scheduling plan was designed to allow routine observations over the world's volcanoes, which were then made available to the scientific community as quickly as possible [26]. The original plan divided the global list of the approximately 1000 active and/or potentially active volcanoes into high, medium and low priority classes. These divisions, initially dubbed class A, B, and C, respectively, varied according to the historical frequency of their eruption activity. Class A consisted of volcanoes that had several recorded eruptions during the prior decade; class B volcanoes had several recorded eruptions during the past several decades; and class C consisted of the remainder of volcanoes that had not seen activity in the prior century. The STAR designated that class A targets would be observed every 48 days during the day and every 32 days at night. Class B targets were to be observed every 3 months, both in the day and night. Finally, class C targets were to be observed once every 6 months [27,28]. Although this plan continues to provide regular data of all the volcanoes on Earth, many eruptions and the precursory activity prior, were missed due to this schedule rigidity. For the Volcano STAR, ASTER acquires~16,000 scenes per year on average, which is a combination of day and nighttime data, for a total of~320,000 individual ASTER volcano scenes over the mission lifetime. Therefore, for the 964 individual volcano ROIs in the ASTER Volcano STAR, each volcano is observed~16 times per year on average. Although an improvement over the nominal observational schedule, this frequency is still not enough to allow rapid response observations nor discrimination of short-timescale activity, especially considering that some percentage of these scenes are dominated by clouds.

The ASTER Urgent Request Protocol (URP)
Because of the lack of an adequate temporal sampling of the very restless and actively erupting volcanoes with the volcano STAR, the ASTER Urgent Request Protocol (URP) Program was proposed [2,29]. Simply, the URP is a means to improve the number of observations at the most active volcanic centers around the world. The URP integrates ASTER into a sensor web construct where all scales of activity at an erupting volcano can be captured [29,30]. The initial and most straight-forward implementation of this approach for the URP uses detection of thermally elevated pixels in high temporal resolution data to subsequently trigger more rapid scheduling and acquisition of the higher spatial/spectral resolution data from ASTER. With such a system in place, the high-frequency activity can be continually imaged throughout the eruption, with the high spatial resolution data ideal for capturing small scale changes. These ASTER data also serve as validation for the low spatial, high temporal resolution data.
The URP program has been in place as part of the ASTER sensor's operational scheduling since 2005 [2,29], responsible for over 5000 additional scenes of active volcanoes during that time (one new scene on average every day). Perhaps more importantly, the URP can be triggered manually if precursory activity is noted based on ground-based observations or reports. This allows pre-eruption data to be acquired that significantly add to the monitoring process [22].
Later expansion of the URP Program increased the original monitored area from the northern Pacific region to the entire globe [2]. The URP currently operates with two global monitoring systems using MODIS data: MODVOLC [31] and Middle InfraRed Observation of Volcanic Activity (MIROVA) [32] as well as AVHRR data focused in the north Pacific region using the Okmok algorithm [33]. New triggering systems are now being tested that will integrate ground-based thermal camera data as the source for new URP data. A trial system at Mt. Etna volcano in Italy has been ongoing since mid-2019 and will expand to Piton de la Fournaise volcano on Réunion Island in 2020 using seismic alerts as the triggering source. The addition of the URP to the ASTER observation schedule is a vast improvement from the original volcano STAR [3,27]. Importantly, however, the URP operates in tandem with the volcano STAR. The volcano STAR data represent additional scenes for the very active volcanoes, thus supplementing the URP archive. Conversely, the volcano STAR data are commonly the only information for the less active, non-thermally elevated targets, and therefore continue to represent an important source for global volcano data.

The ASTER Image Database for Volcanoes
The entire ASTER archive is available at the Land Processes Distributed Active Archive Center (LP DAAC), which can be searched using tools such as EarthDATA and GloVis (Table 1). Other web-based data repositories, however, have been created specifically for the volcano data products. The first of these is the Image Database for Volcanoes (IDV), which was created by M. Urai, and is located in and served from Japan [27,34]. A new volcano image is commonly added within a week of acquisition. The database contains all ASTER images of the 964 target volcanoes that comprise the ASTER Volcano STAR. A 20 km 2 area centered on the geographic location of volcano is shown and stored in the database. Links to each volcano are also displayed and upon selection, the best VNIR image, geographic information, and a table ordered by date of all the acquisitions are shown. The chronological list has thumbnails of the each of the three ASTER imaging systems (where available) along with metadata links for access and download. Urai [34] conducted an initial evaluation of the ASTER Volcano STAR performance using data in the IDV. As expected, volcanoes in Iceland and Kamchatka were observed the most frequently. At the time of the study, 77% of all the volcanoes in the database had at least one daytime image with <10% cloud cover. The IDV remains active as of the time of this writing, and as of the latest IDV update on 1 December 2019, the average number of scenes acquired had grown to 146.3 (daytime) and 185.4 (nighttime) per volcano.

The ASTER Volcano Archive (AVA)
The second online volcano database to appear was the AVA, which is housed and served in the United States at the Jet Propulsion Laboratory [27,28]. The AVA is the largest dedicated archive of web-accessible volcano images now containing image data other than ASTER as well as secondary derived products for each volcano ( Table 1). The archive provides capabilities to inventory and monitor properties and processes over time. These include the spectral signatures for volcanic emissions (e.g., eruption columns and plumes) and surficial deposits (e.g., lava flows, pyroclastic flows), as well as eruption precursor data. AVA has helped to improve the global monitoring and access to archival ASTER image data of more targets (~1500) than are in the ASTER Volcano STAR. Much like the IDV, volcanoes can be searched by name or location, and thumbnail images. Unlike the IDV, AVA data can be downloaded as full ASTER scenes in geoTIFF, KML, or HDF formats. These data can be also combined with the DEM files and ancillary data in an easily accessible format, which is useful in organizing the multispectral geological analysis of a particular volcano [35]. As of this writing, however, no new data have been added to the archive since late 2017. This is scheduled to be reactivated sometime in 2020.

Operational Structure of the Ongoing ASTER Volcano Observations
All ASTER data acquired as part of ongoing routine Earth observations (i.e., the Volcano STAR, global map imaging, etc.) are archived at the LP DAAC. These data are stored in several formats including raw radiance at sensor, from which numerous level 2 data products (e.g., digital elevation model, surface TIR temperature), can be ordered as on-demand products. Image data that fulfill the requirements of the Volcano STAR are also assessed weekly and stored in the IDV and the AVA archives, the latter until 2017. These volcano archives allow quick data searches on particular volcanoes and visualizations of the latest data, whereas the LP DAAC archive contains all the data, all possible level 2 options and the most up to date processing levels for each on-demand product.
Data collected as part of the URP Program are also eventually stored in the LP DAAC ASTER archive and are pulled over to the individual volcano archives as well. However, because of their expedited classification, these data are processed quickly into the Level 1BE (expedited) format and staged on the expedited data system webpage at the LP DAAC for immediate download and assessment. Within 2 h of the data being acquired, all scientists involved with the URP Program are automatically notified by email and have immediate web-based access to the new scene. Any significant changes detected in the data of a particular eruption are disseminated to the responsible monitoring agencies, local scientists working on the eruption, as well as the global community through e-mail and mailing lists. More detailed science analysis is then commonly performed over time and with the arrival of new data. All new, newly scheduled, and ongoing volcano observations of each URP volcano are tracked both through a database and email system as well as web-based map tool showing all the current targets color coded by ASTER observations status ( Figure 3). Each color coded pin is clickable allowing a query of the latest observations and links to the metadata for each scene. Also shown is the current position of the Terra satellite.
Remote Sens. 2020, 12, x FOR PEER REVIEW 8 of 39 These data can be also combined with the DEM files and ancillary data in an easily accessible format, which is useful in organizing the multispectral geological analysis of a particular volcano [35]. As of this writing, however, no new data have been added to the archive since late 2017. This is scheduled to be reactivated sometime in 2020.

Operational Structure of the Ongoing ASTER Volcano Observations
All ASTER data acquired as part of ongoing routine Earth observations (i.e., the Volcano STAR, global map imaging, etc.) are archived at the LP DAAC. These data are stored in several formats including raw radiance at sensor, from which numerous level 2 data products (e.g., digital elevation model, surface TIR temperature), can be ordered as on-demand products. Image data that fulfill the requirements of the Volcano STAR are also assessed weekly and stored in the IDV and the AVA archives, the latter until 2017. These volcano archives allow quick data searches on particular volcanoes and visualizations of the latest data, whereas the LP DAAC archive contains all the data, all possible level 2 options and the most up to date processing levels for each on-demand product.
Data collected as part of the URP Program are also eventually stored in the LP DAAC ASTER archive and are pulled over to the individual volcano archives as well. However, because of their expedited classification, these data are processed quickly into the Level 1BE (expedited) format and staged on the expedited data system webpage at the LP DAAC for immediate download and assessment. Within 2 h of the data being acquired, all scientists involved with the URP Program are automatically notified by email and have immediate web-based access to the new scene. Any significant changes detected in the data of a particular eruption are disseminated to the responsible monitoring agencies, local scientists working on the eruption, as well as the global community through e-mail and mailing lists. More detailed science analysis is then commonly performed over time and with the arrival of new data. All new, newly scheduled, and ongoing volcano observations of each URP volcano are tracked both through a database and email system as well as web-based map tool showing all the current targets color coded by ASTER observations status ( Figure 3). Each color coded pin is clickable allowing a query of the latest observations and links to the metadata for each scene. Also shown is the current position of the Terra satellite.

Statistical Analysis
All ASTER scenes used in this study were acquired from the LP DAAC through the NASA EarthData web-based search tool using an initial search refinement by date. The refined datasets were then downloaded as ASTER Level 2 surface kinetic temperature (AST_08) products. This data product allows the most efficient way to identify volcanic activity by temperature within a given scene. Scenes were automatically scanned and later visually inspected for any thermal variations. If no activity was initially identified, a more detailed investigation for thermally elevated pixels was performed for each scene. If this process yielded a negative result, the scene was classified as having no detectable volcanic activity. If volcanic (thermal) activity was present, then the type of activity was classed as a plume, lava flow, other flow, or summit hot spot. The cloud cover percentage was also visually determined for all the scenes investigated. This was then compared to the ASTER Cloud Cover Assessment Algorithm (ACCAA) value for future assessment analysis.

ASTER Capabilities and Observation Strategies
The unique instrument characteristics that make ASTER particularly well suited for volcanic observations include multispectral TIR data, routine TIR data at night, high spatial resolution data, variable gain settings to limit data saturation, off-axis pointing, and generation of along-track digital elevation models [3,36]. For example, the multispectral TIR data at a relatively high spatial resolution allowed a variety of surface materials to be distinguished and a better understanding of thermal and compositional mixing at the sub-100 m pixel scale [6,37,38] Unlike other orbiting instruments of similar spatial and temporal resolutions, ASTER routinely acquires TIR data at night. In certain special observational modes (e.g., large and highly radiant lava flows), VNIR nighttime data can also be acquired. Nighttime TIR have the advantage of reduced residual solar heating, reduced thermal topographic effects, and typically have lower cloud percentages compared to daytime scenes [3]. Subtle thermal anomalies are therefore more easily identified in nighttime TIR data. A new study by Flynn and Ramsey [39] of Fuego volcano from (1 January 2000until 30 April 2018 found that ASTER acquired 308 scenes-193 of which were collected at night. Of those nighttime scenes, 109 had visible volcanic activity present, with 47 being summit hot spots and 62 showing either lava flows or pyroclastic density currents (PDCs). These data were compared to the Guatemalan monitoring agency's weekly reports to determine the specific volcanic event that could have caused the thermal detections in the ASTER data. This data synthesis is then used to create volcanic hazard maps and cross check datasets for missed events. Without the nighttime TIR data, many of these volcanic events may not have been observed.

Multispectral TIR Data: SO 2 Plumes from Lascar Volcano, Chile
The multispectral resolution of the ASTER TIR data enabled Henney et al., [40] to detect and measure very low SO 2 concentrations (<1 g/m 2 ) released from Lascar volcano in December 2004 ( Figure 4). Of note by the authors was the high spatial resolution and radiometric sensitivity, which along with the needed multispectral resolution allowed these retrievals of very small SO 2 burdens, which are not possible with any other orbital instrument. The TIR results were also compared to ground-based measurements made in the ultraviolet (UV) spectral region during a coordinated overpass of ASTER and found to be well within the error of those instruments.

High Spatial Resolution Data: TIR Analysis of Hawaiian Volcanoes
The high spatial resolution of the ASTER data, in particular in the TIR, was documented by Patrick and Witzke [42] for long-term mapping and monitoring of the active or potentially active volcanoes on the islands of Hawai'i and Maui. The goal of the study was to determine the baseline thermal behavior over ten years (2000)(2001)(2002)(2003)(2004)(2005)(2006)(2007)(2008)(2009)(2010) in order to assess thermal changes that may precede a future eruption. They used cloud-free kinetic temperature ASTER data acquired for the five major subaerial volcanoes in Hawai'i (Kīlauea, Mauna Loa, Hualālai, Mauna Kea, and Haleakalā). The data were geolocated and stacked to create time-averaged thermal maps and to extract temperature trends over the study period. Conspicuous thermal areas were found on the summits and rift zones of Kīlauea, Mauna Loa, and the small pit craters on Hualālai. No thermal areas were detected on Haleakalā or Mauna Kea. One limiting factor noted for the lack of detections was the pixel size of the ASTER TIR, which despite being one of the highest from orbit, was seen as still too large to detect possible subtle thermal changes as well as to identify small-scale, low-temperature thermal activity.

Variable Gain Settings (VNIR/SWIR): High Temperature Monitoring of Klyuchevskoy Volcano, Russia
The ASTER VNIR and SWIR (no longer functioning) subsystems both have/had the ability to acquire data at different data gain settings. The VNIR has three settings (low, normal, and high), whereas the SWIR had four (low1, low2, normal, and high). These gain settings must be set prior to scheduling an observation, so some degree of advanced knowledge of the target is required. They were created to limit data saturation in regions of excessively high or low radiance (e.g., bright glacier and clouds surfaces, dark water body surfaces, etc.). Typically, these gain settings are uniformly set across all bands in the VNIR and the SWIR (when it was operating). However, a unique scenario was created for the volcano STAR whereby every other SWIR band was alternated between the normal and the low2 gain settings in order to maximize the possibility of capturing some unsaturated data of a future volcanic target that may have lava on the surface [25]. The saturation pixel-integrated The high spatial resolution of the ASTER data, in particular in the TIR, was documented by Patrick and Witzke [42] for long-term mapping and monitoring of the active or potentially active volcanoes on the islands of Hawai'i and Maui. The goal of the study was to determine the baseline thermal behavior over ten years (2000)(2001)(2002)(2003)(2004)(2005)(2006)(2007)(2008)(2009)(2010) in order to assess thermal changes that may precede a future eruption. They used cloud-free kinetic temperature ASTER data acquired for the five major subaerial volcanoes in Hawai'i (Kīlauea, Mauna Loa, Hualālai, Mauna Kea, and Haleakalā). The data were geolocated and stacked to create time-averaged thermal maps and to extract temperature trends over the study period. Conspicuous thermal areas were found on the summits and rift zones of Kīlauea, Mauna Loa, and the small pit craters on Hualālai. No thermal areas were detected on Haleakalā or Mauna Kea. One limiting factor noted for the lack of detections was the pixel size of the ASTER TIR, which despite being one of the highest from orbit, was seen as still too large to detect possible subtle thermal changes as well as to identify small-scale, low-temperature thermal activity.

Variable Gain Settings (VNIR/SWIR): High Temperature Monitoring of Klyuchevskoy Volcano, Russia
The ASTER VNIR and SWIR (no longer functioning) subsystems both have/had the ability to acquire data at different data gain settings. The VNIR has three settings (low, normal, and high), whereas the SWIR had four (low1, low2, normal, and high). These gain settings must be set prior to scheduling an observation, so some degree of advanced knowledge of the target is required. They were created to limit data saturation in regions of excessively high or low radiance (e.g., bright glacier and clouds surfaces, dark water body surfaces, etc.). Typically, these gain settings are uniformly set across all bands in the VNIR and the SWIR (when it was operating). However, a unique scenario was created for the volcano STAR whereby every other SWIR band was alternated between the normal and the low2 gain settings in order to maximize the possibility of capturing some unsaturated data of a future volcanic target that may have lava on the surface [25]. The saturation pixel-integrated brightness temperature for the SWIR ranged from 86 • C in band 9 (wavelength = 2.336 µm and gain = high) to 467 • C in band 4 (wavelength = 0.804 µm and gain = low2). For the VNIR, the range is 669 • C in band 3 (wavelength = 0.807 µm and gain = high) to 1393 • C in band 1 (wavelength = 0.556 µm and gain = low). By comparison, the TIR has only one gain setting and a saturation pixel-integrated brightness temperature of 97 • C.
These variable gain settings have proven quite useful for deriving accurate pixel-integrated brightness temperatures of large, highly radiant lava flows over the ASTER mission lifetime. For example, Rose and Ramsey [43] describe the use of all three ASTER subsystems to monitor the emplacement of multiple long lava flows at Klyuchevskoy volcano, Kamchatka during the 2005 and 2007 eruptions. In 2007, temperatures were extracted from 4 January to 7 June 2007 and fit into the volcanic warning color codes over that period ( Figure 5). For the first 4.5 months of the eruption, TIR data remained unsaturated and the SWIR-derived temperatures, acquired in high-gain mode, were only detectable in the highest SWIR wavelengths. This corresponded to the time period that Klyuchevskoy was designated with yellow and orange color codes, signifying increasing levels of restlessness, but no eruption. From 26 April to 10 May, all TIR and SWIR-derived temperatures became saturated due to the emplacement of new open-channel lava flow that was radiant enough to be detected at the 15 m VNIR spatial scale. VNIR-derived temperatures (852-895 • C) were detected for the first time in the eruption. These detections took place one week before the alert level was raised to red (signifying an active eruption), attesting to the importance of high-repeat, non-saturated data for the monitoring of these more remote volcanoes.
Remote Sens. 2020, 12, x FOR PEER REVIEW 11 of 39 brightness temperature for the SWIR ranged from 86 °C in band 9 (wavelength = 2.336 μm and gain = high) to 467 °C in band 4 (wavelength = 0.804 μm and gain = low2). For the VNIR, the range is 669 °C in band 3 (wavelength = 0.807 μm and gain = high) to 1393 °C in band 1 (wavelength = 0.556 μm and gain = low). By comparison, the TIR has only one gain setting and a saturation pixel-integrated brightness temperature of 97 °C. These variable gain settings have proven quite useful for deriving accurate pixel-integrated brightness temperatures of large, highly radiant lava flows over the ASTER mission lifetime. For example, Rose and Ramsey [43] describe the use of all three ASTER subsystems to monitor the emplacement of multiple long lava flows at Klyuchevskoy volcano, Kamchatka during the 2005 and 2007 eruptions. In 2007, temperatures were extracted from 4 January to 7 June 2007 and fit into the volcanic warning color codes over that period ( Figure 5). For the first 4.5 months of the eruption, TIR data remained unsaturated and the SWIR-derived temperatures, acquired in high-gain mode, were only detectable in the highest SWIR wavelengths. This corresponded to the time period that Klyuchevskoy was designated with yellow and orange color codes, signifying increasing levels of restlessness, but no eruption. From 26 April to 10 May, all TIR and SWIR-derived temperatures became saturated due to the emplacement of new open-channel lava flow that was radiant enough to be detected at the 15 m VNIR spatial scale. VNIR-derived temperatures (852-895 °C) were detected for the first time in the eruption. These detections took place one week before the alert level was raised to red (signifying an active eruption), attesting to the importance of high-repeat, non-saturated data for the monitoring of these more remote volcanoes. Thermal infrared (TIR) temperatures are denoted with purple squares; shortwave infrared (SWIR) temperatures with green triangles; and visible/nearinfrared (VNIR) temperatures with blue circles. The background colors represent the volcanic color code issued by the monitoring agencies (yellow = elevated unrest above known background levels; orange = heightened unrest with increased likelihood of eruption; red = eruption is forecast to be imminent). TIR and SWIR temperatures become saturated (and VNIR temperatures become measureable) near the time shown by the thin vertical line. This saturation indicates that a significant amount of highly radiant lava is on the surface and occurred despite adjustments to the SWIR gain settings. Importantly, this was more than two weeks before the color code is changed to red. Modified from Rose and Ramsey [43].

Figure 5.
Maximum ASTER-derived pixel-integrated brightness temperatures detected during the 2007 eruption of Klyuchevskoy volcano, Russia. Thermal infrared (TIR) temperatures are denoted with purple squares; shortwave infrared (SWIR) temperatures with green triangles; and visible/near-infrared (VNIR) temperatures with blue circles. The background colors represent the volcanic color code issued by the monitoring agencies (yellow = elevated unrest above known background levels; orange = heightened unrest with increased likelihood of eruption; red = eruption is forecast to be imminent). TIR and SWIR temperatures become saturated (and VNIR temperatures become measureable) near the time shown by the thin vertical line. This saturation indicates that a significant amount of highly radiant lava is on the surface and occurred despite adjustments to the SWIR gain settings. Importantly, this was more than two weeks before the color code is changed to red. Modified from Rose and Ramsey [43].

Off-Axis Pointing Capability: Improved Observational Frequency at Piton de la Fournaise Volcano, France
During the April-May 2018 eruption of Piton de la Fournaise volcano on Réunion Island, satellite-based surveillance of the thermal activity and emitted plumes was paired with near-real-time flow modeling to create an ensemble-based approach in response to the crisis [44]. This combined effort of four institutions in several countries using data from numerous sensors to model and forecast lava flow advance was done as a proof of concept in order to assist the small staff of the volcano observatory on the island. Rapid data acquisition was critical and provided by MODIS via the MIROVA system to determine lava discharge rates. The less frequent but higher spatial resolution ASTER data were important both for determining the length, shape and direction of the flow using the TIR data ( Figure 6), as well as the precise location of the vent using the VNIR data. The TIR images served as a validation for the MODIS data and the predicted down slope flow modeling, whereas the VNIR-derived vent location was used as the initiation point for the flow modeling, critical for model accuracy. During the 35 day eruption, a total of 11 ASTER images were acquired. This average of~1 image every 3 days was a significant improvement over the nominal 16 day repeat time for targets close to the equator. The improved temporal resolution was only made possible with the ASTER URP Program and the off-nadir pointing capability of 8 • nominally and up to 24 • for the VNIR. During the April-May 2018 eruption of Piton de la Fournaise volcano on Réunion Island, satellite-based surveillance of the thermal activity and emitted plumes was paired with near-realtime flow modeling to create an ensemble-based approach in response to the crisis [44]. This combined effort of four institutions in several countries using data from numerous sensors to model and forecast lava flow advance was done as a proof of concept in order to assist the small staff of the volcano observatory on the island. Rapid data acquisition was critical and provided by MODIS via the MIROVA system to determine lava discharge rates. The less frequent but higher spatial resolution ASTER data were important both for determining the length, shape and direction of the flow using the TIR data ( Figure 6), as well as the precise location of the vent using the VNIR data. The TIR images served as a validation for the MODIS data and the predicted down slope flow modeling, whereas the VNIR-derived vent location was used as the initiation point for the flow modeling, critical for model accuracy. During the 35 day eruption, a total of 11 ASTER images were acquired. This average of ~1 image every 3 days was a significant improvement over the nominal 16 day repeat time for targets close to the equator. The improved temporal resolution was only made possible with the ASTER URP Program and the off-nadir pointing capability of 8° nominally and up to 24° for the VNIR.  [44]. A color-coded Moderate Resolution Imaging Spectroradiometer (MODIS) radiance image produced by the Middle InfraRed Observation of Volcanic Activity (MIROVA) monitoring system is draped over a 3D Google Earth visible image for context. These KML image products are routinely produced by MIROVA. Draped on both data sets is the nighttime ASTER TIR data acquired as part of the URP Program. The ASTER image, acquired 6 days after the triggering MODIS detection, shows the spatial details of the propagating lava flow as the brighter white pixels within the colorized MODIS pixels.

Generation of Along-Track Digital Elevation Models (DEMs): Volcanoes in Guatemala, New Zealand, and Mexico
The ability to produce single 60 by 60 km scene DEMs from any daytime VNIR data is another significant capability of the instrument. Importantly, these data provided digital topography for most of the Earth's land surface at 10 m accuracy early in the mission when such information was not  [44]. A color-coded Moderate Resolution Imaging Spectroradiometer (MODIS) radiance image produced by the Middle InfraRed Observation of Volcanic Activity (MIROVA) monitoring system is draped over a 3D Google Earth visible image for context. These KML image products are routinely produced by MIROVA. Draped on both data sets is the nighttime ASTER TIR data acquired as part of the URP Program. The ASTER image, acquired 6 days after the triggering MODIS detection, shows the spatial details of the propagating lava flow as the brighter white pixels within the colorized MODIS pixels.

Generation of Along-Track Digital Elevation Models (DEMs): Volcanoes in Guatemala, New Zealand, and Mexico
The ability to produce single 60 by 60 km scene DEMs from any daytime VNIR data is another significant capability of the instrument. Importantly, these data provided digital topography for most of the Earth's land surface at 10 m accuracy early in the mission when such information was not available from other data sources such as radar. Later in the mission, these individual-scene DEMs were compiled into a seamless global data set known as the ASTER Global DEM or GDEM. The GDEM has better signal to noise and removes areas of clouds or other artificial errors commonly found in the single-scene DEMs (Figure 7). The final version (v.3) was released in August 2019 having corrections made to minor areas of known errors.
available from other data sources such as radar. Later in the mission, these individual-scene DEMs were compiled into a seamless global data set known as the ASTER Global DEM or GDEM. The GDEM has better signal to noise and removes areas of clouds or other artificial errors commonly found in the single-scene DEMs (Figure 7). The final version (v.3) was released in August 2019 having corrections made to minor areas of known errors. Whereas the GDEM provides an excellent baseline for volcanological studies reliant upon topography, it does not account for changes to the surface, either during the period used for the GDEM creation (2000-2011) as they would have been averaged out or not included in subsequent data. Dynamic topography is a characteristic of most active volcanoes and accurate knowledge of that topography becomes important for any type of lava or pyroclastic flow forecast modeling. Therefore, current single-scene DEMs continue to be an important dataset for volcanological studies using ASTER. Stevens et al., [45] noted this point in their study examining the accuracy of the ASTER single-scene DEM for Ruapehu and Taranaki volcanoes in New Zealand. They found an average rootmean-squared (RMS) error of ~10 m for the ASTER single-scene DEM compared to digitized 1:50,000 scale topographic maps, and later noted that these data will continue to be relevant for future surface change even as global DEM products from radar systems came online. A later study by Huggel et al.,[46] evaluated the ASTER DEM against such a global dataset, the Shuttle Radar Topography Mission (SRTM) DEM. The data were used for lahar modeling on Popocatépetl Volcano, Mexico. They found that although the higher number of errors in the ASTER DEM affected certain models more, both the ASTER and SRTM DEMs were feasible for lahar modeling, but that verification and sensitivity analysis of the chosen DEM is fundamental to deriving accurate hazard maps from the modeled inundation areas.

URP-Specific Results
ASTER data for a subset of volcanoes over several different time intervals were examined for this study and the significant statistics were compiled. The volcanoes were all continuously (or nearly so) active throughout these periods. The goal of this statistical analysis was to determine the improvement of the URP over the nominal Volcano STAR observations. Further details such as the number of cloudy scenes, the number of scenes with confirmed activity and the days between successive observations were also calculated. The interval periods were: one year (1 June 2018 to 31 Whereas the GDEM provides an excellent baseline for volcanological studies reliant upon topography, it does not account for changes to the surface, either during the period used for the GDEM creation (2000-2011) as they would have been averaged out or not included in subsequent data. Dynamic topography is a characteristic of most active volcanoes and accurate knowledge of that topography becomes important for any type of lava or pyroclastic flow forecast modeling. Therefore, current single-scene DEMs continue to be an important dataset for volcanological studies using ASTER. Stevens et al., [45] noted this point in their study examining the accuracy of the ASTER single-scene DEM for Ruapehu and Taranaki volcanoes in New Zealand. They found an average root-mean-squared (RMS) error of~10 m for the ASTER single-scene DEM compared to digitized 1:50,000 scale topographic maps, and later noted that these data will continue to be relevant for future surface change even as global DEM products from radar systems came online. A later study by Huggel et al.,[46] evaluated the ASTER DEM against such a global dataset, the Shuttle Radar Topography Mission (SRTM) DEM. The data were used for lahar modeling on Popocatépetl Volcano, Mexico. They found that although the higher number of errors in the ASTER DEM affected certain models more, both the ASTER and SRTM DEMs were feasible for lahar modeling, but that verification and sensitivity analysis of the chosen DEM is fundamental to deriving accurate hazard maps from the modeled inundation areas.

URP-Specific Results
ASTER data for a subset of volcanoes over several different time intervals were examined for this study and the significant statistics were compiled. The volcanoes were all continuously (or nearly so) active throughout these periods. The goal of this statistical analysis was to determine the improvement of the URP over the nominal Volcano STAR observations. Further details such as the number of cloudy scenes, the number of scenes with confirmed activity and the days between successive observations were also calculated. The interval periods were: one year ( The average percentage of scenes for all the periods that contained detectable volcanic activity was 47.6%, with a range of 19%-69%. For this same period and number of scenes, the average cloud percentage was 50.5%, with a range from 16% to 84%. Finally, during the last five years of the implementation of the global URP Program, the average days between observations at the four volcanoes studied was 6.5 days, compared to 20.2 days with the Volcano STAR Program, a nearly 200% improvement.

One-Year Analysis
Ten volcanoes were chosen for a statistical analysis over a recent one-year period ( Table 2). The volcanoes chosen represent a range of eruptive styles, compositions, and latitudinal distributions ( Figure 8). All volcanoes were active during the study period and that activity triggered the ASTER URP observations. The number of triggers (derived from MODIS data by the MODVOLC and MIROVA systems) varied from 36 for Popocatépetl to 837 for the Nyamuragira-Nyiragongo volcanoes. The number of triggers is determined by several factors-the most important of which is the style of activity (i.e., persistently active lava lake versus an intermittently active lava dome) and size of the thermal feature on the ground. These combine to determine the overall emitted radiant power. Other factors contributing to the number of ASTER URP triggers also include the latitude of the target together with the average daily cloud cover. Higher latitude volcanoes will have more overpasses due to the converging orbit tracks of the MODIS sensors and hence, more triggers. Persistently cloudy targets will commonly mask thermal activity, lowering the number of triggers. Average cloud percentage was determined by visual inspection of each URP scene done at the same time as the inspection for volcanic activity. Table 2. Statistical analysis of ten URP-monitored volcanoes active during one year (1 June 2018 to 31 May 2019). Each ASTER scene was inspected for confirmed/detected volcanic activity (e.g., thermally elevated pixels, presence of a plume) and average cloud cover. The percentage of those URP scenes with confirmed activity is also given and generally correlates to the average cloud percentage. These MODVOLC and MIROVA triggers resulted in 276 new ASTER scenes during the one-year period with confirmed volcanic activity in~45% of those scenes, with the remainder being obscured by clouds. The frequency of ASTER URP scenes increases at the higher latitude volcanoes on the list (e.g., Erebus and Shiveluch) as expected. However, Erta Ale at a much lower latitude had a similar number of observations to those two targets, a function of nearly continuous thermally elevated activity present there during the one-year period.

Five-Year Analysis
Four of the volcanoes in the prior list were then selected for a longer study period of slightly over five years (1 January 2014-31 May 2019. This period begins when the ASTER URP Program became fully global with the integration of the MODVOLC triggering system. The time frame, therefore, provides the most complete global higher temporal resolution ASTER data for any thermally elevated volcano. As such, the statistics for these observations are the most comprehensive analysis available for the planning of future orbital TIR systems, for example (e.g., expected longterm cloud cover, optimal temporal frequency, etc.).
The data for the four volcanoes chosen (Ambrym, Erta Ale, Popocatépetl, Shiveluch) are shown in Table 3. With the exception of Shiveluch, the others are more equatorial. The four do, however, represent a range of compositions (e.g., dacite, andesite, and basalt) and styles (e.g., dome-forming, lava lakes, and flows). A total of 1001 ASTER scenes were downloaded and analyzed for these four targets. Volcanic activity was confirmed in 535 of these (53.5%) with the remainder being too cloud-

Five-Year Analysis
Four of the volcanoes in the prior list were then selected for a longer study period of slightly over five years (1 January 2014-31 May 2019. This period begins when the ASTER URP Program became fully global with the integration of the MODVOLC triggering system. The time frame, therefore, provides the most complete global higher temporal resolution ASTER data for any thermally elevated volcano. As such, the statistics for these observations are the most comprehensive analysis available for the planning of future orbital TIR systems, for example (e.g., expected long-term cloud cover, optimal temporal frequency, etc.).
The data for the four volcanoes chosen (Ambrym, Erta Ale, Popocatépetl, Shiveluch) are shown in Table 3. With the exception of Shiveluch, the others are more equatorial. The four do, however, represent a range of compositions (e.g., dacite, andesite, and basalt) and styles (e.g., dome-forming, lava lakes, and flows). A total of 1001 ASTER scenes were downloaded and analyzed for these four targets. Volcanic activity was confirmed in 535 of these (53.5%) with the remainder being too cloud-covered or having no obvious activity visible in the image. Erta Ale had the highest percent of observed activity (80.2%) whereas Ambrym had the lowest (29.7%), a clear function of the semi-arid, nearly cloud-free location of Erta Ale compared to the tropical, nearly cloud-covered location of Ambrym. Also analyzed was the average days between an observation for the routine Volcano STAR (20.2 days/scene) as compared to those data from the URP Program (6.5 days/scene). This was an average improvement of 215%, with the best improvement (261%) for Ambrym volcano. The URP Program excels at acquiring more data of targets that are commonly cloudy or close to the equator, and therefore typically missed by the routine STAR observations. Table 3. Statistical analysis of four volcanoes from Table 2 active during the longer global URP period (1 January 2014-31 May 2019. Each ASTER scene was inspected for confirmed/detected volcanic activity (e.g., thermally elevated pixels, presence of a plume) and average cloud cover. The percentage of those URP scenes with confirmed activity is also given and generally correlates to the average cloud percentage. The time between scenes decreases dramatically for the URP operations as compared to the ASTER Volcano STAR. The last column shows this improvement percentage in the number of days between scenes using the URP versus the STAR scheduling.

Twenty-Year Analysis
Finally, one volcano (Shiveluch) was selected for analysis for nearly the entire period (13 April 2000-31 May 2019) that Terra has been in orbit ( Table 4). The first ASTER image acquired of Shiveluch was on 13 April 2000, only 40 days after the start of the operational phase. These early scenes of the volcanoes of Kamchatka also captured the products of a large eruption at Bezymianny volcano described by Ramsey and Dehn [47]. Shiveluch has been persistently active nearly the past twenty years (e.g., [6]) and therefore provides a good target for long-term analysis. It was also imaged more frequently early in the mission because of its location at higher latitudes as well as being one of the northern Pacific volcanoes originally monitored in the first phase of the URP Program, which began in 2005 and relying upon AVHRR data for detection triggering [29]. Table 4. Statistical analysis of a single volcano (Shiveluch) from Table 2 active during most of the entire ASTER orbital period (13 April 2000-31 May 2019. Each ASTER scene was inspected for confirmed/detected volcanic activity (e.g., thermally elevated pixels, presence of a plume, etc.) and average cloud cover. The percentage of those URP scenes with confirmed activity is also given and generally correlates to the average cloud percentage. The analysis results are shown in Table 4. A total of 815 confirmed scenes were downloaded and interrogated. Similar to the prior results, volcanic activity was confirmed in 430 (52.8%) of the scenes. We further analyzed the type of activity seen in those clear scenes. Plumes were noted in 53 (12.3%) scenes, flows in 109 (25.3%), and summit hot spots in 268 (62.3%) scenes. We also examined the impact of the URP Program to acquire more rapid data of eruption activity. The average time between ASTER Volcano STAR observations at Shiveluch is 8.8 days, much lower than the more equatorial volcanoes (~20 days). Implementation of the URP Program, however, reduced this temporal resolution slightly to 6.1 days. If one were to only consider the data following a new URP triggering event, this value decreases further to 5.1 days. If volcanic activity wanes and then renews, the first ASTER image acquired after a new triggering event may come many days to weeks after the prior image was acquired. Therefore, this lower interval is likely a better metric of the URP operational behavior. It is also close to the theoretical minimum frequency of ASTER for high latitude targets.

ASTER as a Volcanological Instrument
ASTER, unlike almost any instrument designed or chosen by a space agency in the past several decades, was first and foremost a geological sensor. The wavelength regions chosen, the placement of the multispectral bands, the multispatial resolution, and the DEM capability are all critically important for geologic surface mapping. Although it was not designed specifically for volcanology, the geologic heritage of ASTER makes the instrument's data well suited for active volcanic monitoring and data collection. Perhaps the only capability missing was a mid-infrared channel for high-temperature detections. This was offset for the first 8 years of the mission with the 2 µm region SWIR data. As originally noted by Pieri and Abrams [3] and later described by Ramsey [2] and herein, six important instrument/data characteristics made ASTER particularly well suited for spaceborne volcanology. These are the multispectral TIR data, routine TIR data at night, high spatial resolution data, variable gain settings in the VNIR and SWIR, off-axis pointing, and generation of DEMs. Individually, these characteristics all have been used for specific volcanic studies. Where combined, these capabilities become much more powerful for addressing pressing questions in volcanology [45,48].

Toward an Improving Temporal Resolution
ASTER is able to provide data with improved radiometric resolutions (absolute reflectance ≤ 1.3% and absolute temperature ≤0.3 K) [7]. Moreover, it provides data at improved spatial (15-90 m) and spectral resolutions (14 wavelength band total, with 5 in the TIR and 6 in the SWIR). This multiscale spatial resolution across a wide wavelength range provides a unique tool for volcanological applications. For example, the wavelength range coupled with the variable gain settings allows the acquisition of unsaturated data over a wide temperature range (−73-1393 • C). Furthermore, saturation in one wavelength region (e.g., the TIR) can be compensated using the nested pixels from a shorter wavelength region (e.g., the 9 SWIR or 36 VNIR pixels) to correct for that saturation (e.g., [38]). These data enable thermal anomalies of only a few degrees above the background temperature, as well as sub-pixel, highly radiant anomalies of only several m 2 , to be determined.
For the past 15 years, an ASTER-focused program called the Urgent Request Protocol (URP) has combined the rapid detection capability of higher temporal resolution instruments like MODIS with the high spatial resolution scheduled observations of ASTER. These observations have improved our knowledge of multi-year volcanic monitoring, ongoing eruption behavior, and post-eruption change. They have also become important for capturing points-in-time during any ongoing lava flow or plume forming eruption. More commonly, the ASTER URP data (as well as the entire ASTER volcanic archive) are being used for operational response to new eruptions; determining thermal trends months prior to an eruption; inferring the emplacement of new lava lobes; and mapping the constituents of volcanic plumes, to name a few (see Appendix A). These all require the higher spatial resolution data of ASTER, and in most cases, its multispectral capability.
A framework like the URP sensor web only works in its current form if there is a large enough radiant target to be detected by the operational systems using MODIS data like MODVOLC and MIROVA. The MIROVA system applies an enhanced thermal index (ETI) and therefore detects more subtle thermal signals as compared to MODVOLC [32]. Even with this enhanced detection, however, there will be a significant number of thermal events and numerous volcanoes missed every year because they are not large enough (spatially or radiantly) to be detected in a 1 km MODIS pixel. It is these smaller signals that are frequently important as they indicate the onset of renewed activity, sometimes months to years prior to an eruption [22].
There are some volcanoes whose eruptions are long-lived enough and/or whose character changes throughout their ongoing eruptions, which allow sporadic larger signals to be detected with MODIS. These then trigger subsequent ASTER data, which captures more subtle, transient activity. The best examples of such activity are typically dome-forming eruptions. Larger thermal signals are caused by dome collapse exposing the hotter material or periodic larger dome-destroying eruptions. With renewed dome growth, the cooler carapace insulates the hotter interior and limits detection by MODIS. One of the best examples of this eruptive style is Sheveluch volcano, Russia. In a recent set of three observations by ASTER over a one week period, small changes in the dome's thermal output and moderately sized block-and-ash flows were captured (Figure 9). Although interesting from a volcanological perspective for Sheveluch, it is these relatively high-frequency ASTER URP data, captured over the one to multi-year timeframe, which greatly improve our understanding of how eruptions proceed and how volcanoes reawaken.
Remote Sens. 2020, 12, x FOR PEER REVIEW 18 of 39 There are some volcanoes whose eruptions are long-lived enough and/or whose character changes throughout their ongoing eruptions, which allow sporadic larger signals to be detected with MODIS. These then trigger subsequent ASTER data, which captures more subtle, transient activity.
The best examples of such activity are typically dome-forming eruptions. Larger thermal signals are caused by dome collapse exposing the hotter material or periodic larger dome-destroying eruptions. With renewed dome growth, the cooler carapace insulates the hotter interior and limits detection by MODIS. One of the best examples of this eruptive style is Sheveluch volcano, Russia. In a recent set of three observations by ASTER over a one week period, small changes in the dome's thermal output and moderately sized block-and-ash flows were captured (Figure 9). Although interesting from a volcanological perspective for Sheveluch, it is these relatively high-frequency ASTER URP data, captured over the one to multi-year timeframe, which greatly improve our understanding of how eruptions proceed and how volcanoes reawaken.

Conclusions
Despite the two-decade archive of multispectral, multispatial resolution ASTER data (or perhaps because of its success), there have been no approved follow-on instruments by NASA or other space agencies that have similar spatial and spectral scales. Some of this land imaging is filled by sensors on the Landsat, Sentinel-2, and SPOT satellites, but the data gap discrepancy looms largest in the TIR. Currently, only the limited mission lifetime ECOSTRESS instrument on the ISS is the most similar. This gap was also noted in both the 2007 and 2018 Decadal Surveys for NASA Earth Science. In 2007, a notional mission called HyspIRI was recommended, from which ECOSTRESS derives its heritage.

Conclusions
Despite the two-decade archive of multispectral, multispatial resolution ASTER data (or perhaps because of its success), there have been no approved follow-on instruments by NASA or other space agencies that have similar spatial and spectral scales. Some of this land imaging is filled by sensors on the Landsat, Sentinel-2, and SPOT satellites, but the data gap discrepancy looms largest in the TIR. Currently, only the limited mission lifetime ECOSTRESS instrument on the ISS is the most similar. This gap was also noted in both the 2007 and 2018 Decadal Surveys for NASA Earth Science. In 2007, a notional mission called HyspIRI was recommended, from which ECOSTRESS derives its heritage.
In 2018, the focus was on specific science questions and important designated observables. TIR data was again highlighted numerous times.
What will arise from the 2018 recommendations is yet to be formulated. TIR instruments will likely get smaller and more numerous using uncooled detectors, becoming CubeSat compatible and operating in a dense sensor web network for improved response times. Although perhaps not designed specifically for volcanology, the data from these instruments will become critical for volcanic crisis response and enable never before possible measurements such as the global inventory of volcanic degassing, thermal precursory trends at every volcano, and accurate temperatures of small activity, which can be used as input to predictive flow and hazard assessment models. This next generation of high spatial, high spectral TIR data captured at ever-improving temporal resolutions will only become possible because of the ASTER instrument design and mission success, most notably for volcanic remote sensing. For example, with the Terra spacecraft's converging orbits at the poles, higher latitude volcanoes like those in Kamchatka, Iceland, and Antarctica are imaged more routinely. The addition of off-axis pointing in the ASTER design and the later establishment of the STAR and URP Programs further improved this temporal resolution to near the theoretical maximum. Multispectral TIR data are now being routinely acquired that allow us to track subtle thermal anomalies, precursory activity, explosive events, plumes, and the percentage of obscuring clouds. These data can help inform future instrument and mission design. For example, looking again at the data in Table 2, one can calculate a "miss rate percentage" between the number of MODIS-based triggers and the number of ASTER scenes acquired from those triggers. This rate varies from a low of 39.8% (Popocatépetl volcano) to a maximum of 96.2% (Erta Ale volcano), with an average of 87.6%. These high values demonstrate the amount of high spatial resolution data that are theoretically being missed because there is not a TIR sensor or sensor system with the spatial/spectral resolution of ASTER and the temporal resolution of MODIS. Perhaps even more critically, with an average cloud percentage near 50% based on our analysis of the ASTER URP archive, half of the current observation attempts fail to detect surface activity. This further highlights the need for improved temporal resolution. Ultimately, these results provide a baseline for future TIR orbital concepts that could respond to the 2018 Decadal Survey recommendations for the TIR data critically needed to address key science questions. Funding: This work was supported by NASA and several successful proposals to the Science of Terra and Aqua Program over the years, with the current being 80NSSC18K1001.

Acknowledgments:
The authors wish to thank the members of the ASTER Science Team for helpful discussions, comments, and support during the past 20 years of the mission-in particular, David Pieri at the Jet Propulsion Laboratory, who, along with Minoru Urai at the Geologic Survey of Japan, headed up the early days of volcano science with ASTER. We would like to thank Ben McKeeby for his assistance with the ASTER archival search and creation of Figure 8, as well as Tyler Leggett, Marco Michelini, James Thompson, and Daniel Williams for their assistance with the painstaking reference checking for the Appendix A. We apologize in advance for any papers inadvertently left off this list; it was not intentional. We intend to keep this list updated on the lead author's laboratory website (http://ivis.eps.pitt.edu/archives/ASTER/). Please send any future or past/missing references to the lead author and they will be added to the online table. Finally, the authors would like to thank the three anonymous reviewers, whose comments and suggestions helped to improve the quality of this manuscript.

Conflicts of Interest:
The authors declare no conflict of interest. Table A1. Reference list of 271 volcanological publications that use, contain, or make mention of (in the case of the Precursory category) ASTER data. The list spans 25 years, from 1995 to 2019, and is subdivided by category with references therein listed in chronological order. The full citations appear below.