Although Artificial Intelligence (AI) projects are common and desired by many institutions and research teams, there are still relatively few success stories of AI in practical use for the Earth science community. Many AI practitioners in Earth science are trapped in the prototyping stage and their results have not yet been adopted by users. Many scientists are still hesitating to use AI in their research routine. This paper aims to capture the landscape of AI-powered geospatial data sciences by discussing the current and upcoming needs of the Earth and environmental community, such as what practical AI should look like, how to realize practical AI based on the current technical and data restrictions, and the expected outcome of AI projects and their long-term benefits and problems. This paper also discusses unavoidable changes in the near future concerning AI, such as the fast evolution of AI foundation models and AI laws, and how the Earth and environmental community should adapt to these changes. This paper provides an important reference to the geospatial data science community to adjust their research road maps, find best practices, boost the FAIRness (Findable, Accessible, Interoperable, and Reusable) aspects of AI research, and reasonably allocate human and computational resources to increase the practicality and efficiency of Earth AI research.

Seismicity during explosive volcanic eruptions remains challenging to observe through the eruptive noise, leaving first‐order questions unanswered. How do earthquake rates change as eruptions progress, and what is their relationship to the opening and closing of the eruptive vent? To address these questions for the Okmok Volcano 2008 explosive eruption, Volcano Explosivity Index 4, we utilized modern detection methods to enhance the existing earthquake catalog. Our enhanced catalog detected significantly more earthquakes than traditional methods. We located, relocated, determined magnitudes and classified all events within this catalog. Our analysis reveals distinct behaviors for long‐period (LP) and volcano‐tectonic (VT) earthquakes, providing insights into the opening and closing cycle. LP earthquakes occur as bursts beneath the eruptive vent and do not coincide in time with the plumes, indicating their relationship to an eruptive process that occurs at a high pressurization state, that is, partially closed conduit. In contrast, VT earthquakes maintain a steadier rate over a broader region, do not track the caldera deflation and have a larger b ‐value during the eruption than before or after. The closing sequence is marked by a burst of LPs followed by small VTs south of the volcano. The opening sequence differs as only VTs extend to depth and migrate within minutes of the eruption onset. Our high‐resolution catalog offers valuable insights, demonstrating that volcanic conduits can transition between partially closed (clogged) and open (cracked) states during an eruption. Utilizing modern earthquake processing techniques enables clearer understanding of eruptions and holds promise for studying other volcanic events.

IntroductionThe degree of openness of a volcanic system is recognized as a major determinant of behavior (e.g., Vergniolle & Métrich, 2021; Seropian et al., 2021; Roman et al., 2019). Whether magma or gas can escape determines the state of stress, and ultimately the eruptibility. Similarly, when an eruption begins, the opening of the vent is the most important factor in any subsequent dynamics. Eruptions are often prolonged and complex sequences of events with eruptive plumes separated by quiescent periods, that might be indicative of resealing, changes in magma fragmentation, or exhaustion of supply. Characterizing the eruptive processes is difficult as high temporal resolution of co-eruptive observations requires techniques that can track behavior continuously during one of the most difficult-to-observe periods of the volcanic cycle.Volcano seismicity provides a promising road into solving the problem since often the earthquake record provides continuous and highly resolved information. However, seismicity during long-lived, explosive eruptions is practically invisible because of the extreme noise levels of the eruption itself. In a companion paper, we have set out a workflow to solve the detection problem utilizing modern earthquake detection methods (Garza-Giron et al., submitted). As we showed in that paper, we can use a combination of traditional, machine learning and template matching approaches to expand the co-eruptive seismic catalog of the 2008 Okmok Volcano eruption by about a factor of 10. In this paper we now use that catalog to address the first-order questions about co-eruptive seismicity that have not been previously accessible:When does the overall earthquake rate increase or decrease in context of the eruption? How does the seismicity evolve as the volcano opens and reopens to erupt material?Okmok Volcano is a 10 km wide basaltic-andesitic caldera located on Umnak Island, in the Aleutian arc of Alaska (Figure 1). For over a century, most of the eruptions, the last of which occurred in 1997, had their source at an intra-caldera cone (Cone A; Figure 1 inset) and were mostly Hawaiian to Strombolian (Coats, 1950; Grey, 2003). The 12 July 2008 eruption, which was given a scale of 4 in the Volcanic Explosivity Index (VEI), marked a change in this behavior because of the interactions between magma and water, making new intra-caldera maars and developing a new tephra cone during a large phreato-magmatic eruption (Larsen et al., 2015). Since the island has a topographical regional-scale tilt toward the northeast, the northern sector is characterized by larger bodies of surface and groundwater, with approximately 1010 kg of water available for the 2008 eruption (Unema et al., 2016). Multiple geophysical studies, most of which focused on modeling the source of geodetic deformation, have found the location of a shallow (2-4 km) magma reservoir at approximately the same location in the caldera (Figure 1; Mann et al., 2002; Fournier et al., 2009; Biggs et al., 2010; Freymueller and Kaufman, 2010; Lu and Dzurisin, 2010; Masterlark et al., 2010; Albright et al., 2019; Xue et al., 2020; Wang et al., 2021). Furthermore, different authors have shown that inflation cycles started immediately after the end of the deflationary eruptive periods in 1997 and 2008, suggesting a quick replenishment of the shallow magma reservoir to compensate the pressure gradient (Lu et al., 2005; Lu and Dzurisin, 2010; Freymueller and Kaufman (2010); Wang et al., 2021).During the 6 months preceding the eruption, the Alaska Volcano Observatory (AVO) detected only 9 low magnitude (M<=2.6) earthquakes, although many of the stations in the region had outages during those months. Most of the inter-eruptive seismicity is localized in a geothermal field on the isthmus of Umnak Island inland from Inanudak Bay (Figure 1). The only precursory activity to the eruption came on 12 July 2008, when the seismic network at Okmok Volcano recorded the onset of a ~4.5 hour-long earthquake swarm (Larsen et al., 2009; Johnson et al., 2010) after which explosive activity commenced. The short sequence of precursory earthquakes was reanalyzed by Ohlendorf et al. (2014) using the AVO catalog, and the earthquakes originated at approximately 3 km depth beneath the intra-caldera cone known as Cone D (Figure 1 inset). The beginning of the eruption was marked by a large-scale sub-Plinian explosion that released a ~16 km above sea level (ASL) high dark plume, consistent with a VEI 4 eruption (Newhall and Self, 1982). This plume was accompanied by more than 12 hours of continuous high-amplitude seismic eruption tremor (Larsen et al., 2009). Tremor continued at variable levels throughout the 40-day-long eruption and emanated mainly from a new intra-caldera cone (Haney, 2010; Haney, 2014). This new cone, to the north of Cone D and built during the 2008 eruption, was subsequently named Ahmanilix, which means ‘surprise’ in the language of the Unangan people whose ancestral lands included Umnak Island (Larsen et al., 2015). After the initial plume, the activity continued by the opening, and perhaps widening, of new vents in a westward alignment from the north-west of Cone D. On July 19, the network recorded high-amplitude continuous tremor that lasted ~20 hours, and is thought to be related to the initiation of the temporary drainage of the perennial North Cone D Lake (hereby called North Lake) (Figure 1). The drainage of the lake was verified by satellite imagery until August 1 when standing water was observed again at the lake (Larsen et al., 2015). Whether the lake refilled before August 1 is unknown. Moreover, Larsen et al. (2015) reported that between July 24 and August 1 the North Vent structure, directly to the north of Ahmanilix, widened and there was an increase in number and size of reflectors observed in Synthetic Aperture Radar (SAR) images, accompanied by an increase in ash production from August 1 until August 3, confirmed by AVO staff in the field. From August 3 until August 19, when the last emission of ash was reported and the eruption officially ended, the plumes decreased in number and size.

By providing unrivaled resolution in both time and space, volcano seismicity helps to chronicle and interpret eruptions. Standard earthquake detection methods are often insufficient as the eruption itself produces continuous seismic waves that obscure earthquake signals. We address this problem by developing an earthquake processing workflow specific to a high-noise volcanic environment and applying it to the explosive 2008 Okmok Volcano eruption. This process includes applying single-channel template matching combined with machine-learning and fingerprint-based techniques to expand the existing earthquake catalog of the eruption. We detected an order of magnitude more earthquakes, then located, relocated, determined locally calibrated magnitudes, and classified the events in the enhanced catalog. This new high-resolution earthquake catalog increases the number of observations by about a factor of 10 and enables the detailed spatiotemporal seismic analysis during a large eruption.

Small unmanned aircraft systems (sUAS) are becoming prominent components of many humanitarian assistance and disaster response (HADR) operations. Pairing sUAS with onboard artificial intelligence (AI) substantially extends their utility in covering larger areas with fewer support personnel. A variety of missions, such as search and rescue, assessing structural damage, and monitoring forest fires, floods, and chemical spills, can be supported simply by deploying the appropriate AI models. However, adoption by resource-constrained groups, such as local municipalities, regulatory agencies, and researchers, has been hampered by the lack of a cost-effective, readily-accessible baseline platform that can be adapted to their unique missions. To fill this gap, we have developed the free and open-source ADAPT multi-mission payload for deploying real-time AI and computer vision onboard a sUAS during local and beyond-line-of-site missions. We have emphasized a modular design with low-cost, readily-available components, open-source software, and thorough documentation (https://kitware.github.io/adapt/). The system integrates an inertial navigation system, high-resolution color camera, computer, and wireless downlink to process imagery and broadcast georegistered analytics back to a ground station. Our goal is to make it easy for the HADR community to build their own copies of the ADAPT payload and leverage the thousands of hours of engineering we have devoted to developing and testing. In this paper, we detail the development and testing of the ADAPT payload. We demonstrate the example mission of real-time, in-flight ice segmentation to monitor river ice state and provide timely predictions of catastrophic flooding events. We deploy a novel active learning workflow to annotate river ice imagery, train a real-time deep neural network for ice segmentation, and demonstrate operation in the field.

Volcanic eruptions eject ash and gases into the atmosphere that can contribute to significant hazards to aviation, public and environment health, and the economy. Several volcanic ash transport and dispersion (VATD) models are in use to simulate volcanic ash transport operationally, but none include a treatment of volcanic ash aggregation processes. Volcanic ash aggregation can greatly reduce the atmospheric budget, dispersion and lifetime of ash particles, and therefore its impacts. To enhance our understanding and modeling capabilities of the ash aggregation process, a volcanic ash aggregation scheme was integrated into the Weather Research Forecasting with online Chemistry (WRFChem) model. Aggregation rates and ash mass loss in this modified code are calculated in line with the meteorological conditions, providing a fully coupled treatment of aggregation processes. The updated-model results were compared to field measurements of tephra fallout and in situ airborne measurements of ash particles from the April–May 2010 eruptions of Eyjafjallajökull volcano, Iceland. WRF-Chem, coupled with the newly added aggregation code, modeled ash clouds that agreed spatially and temporally with these in situ and field measurements. A sensitivity study provided insights into the mechanics of the aggregation code by analyzing each aggregation process (collision kernel) independently, as well as by varying the fractal dimension of the newly formed aggregates. In addition, the airborne lifetime (e-folding) of total domain ash mass was analyzed for a range of fractal dimensions, and a maximum reduction of 79.5 % of the airborne ash lifetime was noted.

Over the past decade Unmanned Aircraft Systems (UAS, aka “drones”) have become pervasive, touching virtually all aspects of our world. While UAS offer great opportunity to better our lives and strengthen economies, at the same time these can significantly disrupt manned flight operations and put our very lives in peril. Balancing the demanding and competing requirements of safely integrating UAS into the United States (US) National Airspace System (NAS) has been a top priority of the Federal Aviation Administration (FAA) for several years. This paper outlines efforts taken by the FAA and the National Aeronautics and Space Administration (NASA) to create the UAS Traffic Management (UTM) system as a means to address this capability gap. It highlights the perspectives and experiences gained by the University of Alaska Fairbanks (UAF) Alaska Center for Unmanned Aircraft Systems Integration (ACUASI) as one of the FAA’s six UAS test sites participating in the NASA-led UTM program. The paper summarizes UAF’s participation in the UTM Technical Capability Level (TCL1-3) campaigns, including flight results, technical capabilities achieved, lessons learned, and continuing challenges regarding the implementation of UTM in the NAS. It also details future efforts needed to enable practical Beyond-Visual-Line-of-Sight (BVLOS) flights for UAS operations in rural Alaska.

Drone Construction and Racing for PreCollege Students Engaging precollege students early in their academic development is an important factor in ensuring their continued interest and focus in education. Science, Technology, Engineering, and Math (STEM) activities, and in particular, those involving unmanned aircraft systems (UAS, or ‘drones’) can provide exciting and valuable outlets for young students who may be considering a technical career path in engineering or related field. Advances in technology over the past decade have dramatically decreased drone prices and increased their availability, making these commonplace in today’s society. Drones are used for everything from personal entertainment to covering news and sporting events, conducting vital scientific research, monitoring critical infrastructure, and even providing emergency services. In addition, recent relaxations in the regulatory framework governing the rules for UAS operations by the Federal Aviation Administration (FAA) have made it much easier to provide meaningful drone activities to precollege students. The erosion of these previous barriers has resulted in a much more conducive environment for college and precollege teachers to engage in drone-centric educational activities. As a result, many colleges and precollege schools are beginning to actively partner with various government agencies and corporate sponsors to bring UAS STEM educational experiences to interested students. One example of this is a program instituted at UNIVERSITY this past year, in partnership with the FAA and the local school district. This program, titled Drone Camp, provided 5th and 6th grade students from the local community an opportunity to learn how to build and pilot small quadcopters, such as those commonly seen in popular Drone Racing League (DRL) events across the country. Held at UNIVERSITY and taught by UAS RESEARCH CENTER personnel, the one-week camp was sponsored by the FAA with the intent of increasing educational opportunities for young students. Elements of instruction included: (1) Basics of flight; (2) Applications of UAS in research and public service; (3) Job opportunities in UAS-related fields; (4) Familiarization with DRL-type small quadcopters; (5) Construction and basic operation of these; and (6) Participation in DRL races on an indoor closed course. In addition to the above skills, students were exposed to basic concepts of teamwork in sharing tools, common equipment, and instructor resources needed for the construction and racing of their drones. They also worked with each other in small groups to provide feedback on construction and preparation of drones, piloting techniques, and race results. Finally, the students were provided a forum to interact directly with experienced educational program coordinators from the FAA for the duration of the camp. As a result of its participation in this effort, UNIVERSITY has experienced significant interest in its educational programs by the students and their families. UNIVERSITY has recently kicked off a DRL club for the community and plans to increase the number and scope of racing events, including the eventual incorporation of more complex engineering-centric competitions. This paper will outline the long term motivation for UNIVERSITY’s involvement in this Drone Camp and related activities, as well as skills learned by the students participating. It will also detail lessons learned from this first event, including student feedback, and provide a look at future outreach activities to be conducted over the next couple years.

Remote sensing data and the application of geo-spatial technologies have progressively been built into real-time volcanic hazard assessment. Remote sensing of volcanic processes provides a unique synoptic view of the developing hazard, and provides insights into the ongoing activity without the need for direct, on-the-ground observations. Analysis and visualization of these data through the geospatial tools, like Geographical Information Systems (GIS) and new virtual globes, brings new perspectives into the decision support system. In this chapter, we provide examples of (i) how remote sensing has assisted in real-time analysis of active volcanoes; (ii) how by combining multiple sensors at different spatial, spectral and temporal resolutions one is able to better understand a given hazard, leading to better communication and decision making; and (iii) how visualizing this in a common platform, like a GIS tool or virtual globe, augments effective hazard assessment system. We will illustrate how useful remote sensing data can be for volcanic hazard assessment, including the benefits and challenges in real-time decision support, and how the geo-spatial tools can be useful to communicate the potential hazard through a common operation protocol.

Explosive volcanic eruptions can inject large amounts of ash and gases into the atmosphere. Such volcanic aerosols can have a significant impact on the surrounding environment, and there is the need to closely investigate their effects on meteorology on local, regional, and even continental scale. This work presents a study of the 2010 Eyjafjallajokull volcanic eruption the resulting ash dispersion and its radiative feedback effects on the meteorological conditions with the Weather Research Forecasting model with on-line Chemistry (WRF-Chem). Two model runs, one meteorology-only simulation (without chemistry) and one that considers gas- and aerosol chemistry as well as direct- and semidirect aerosol feedbacks were performed and compared. Results for daily values show that aerosol radiative feedback effects can cool the atmosphere close to the surface on average by 1 degrees C with maximum cooling exceeding even 2 degrees C for the considered episode. Near-surface atmospheric wind speed changed on average by 0.5 m/s with maximum values above 2 m/s. Furthermore, the presence of ash aerosols affected the vertical shape of the profiles of wind speed and temperature and resulted in a better agreement with radiosonde measurements when radiative feedback effects were considered. Although the modeling of the dispersion of volcanic ash clouds is subject to large uncertainties, we have demonstrated that the WRF-Chem model can reproduce observations at surface levels and vertical profiles more realistically when radiative feedback effects are considered in the simulations.

The editors of a new book describe how to characterize uncertainty in natural hazards, the incorporation of uncertainty into modeling, its contribution to better decision-making, and research needs.