Search USGSSearch

SEARCH · Search USGS

Results for “AI Applications”

Search indexed USGS publications on groundwater, aquifers, geologic maps, mineral resources and earthquakes. Explore source records by subject and place.

Quote a phrase for an exact phrase match. Source license links do not imply unrestricted reuse.

At least 19 recordsLinked to original sources

Harnessing geospatial artificial intelligence and deep learning for landslide inventory mapping: Advances, challenges, and emerging directions

Recent advancements in artificial intelligence (AI) and deep learning enable more accurate, scalable, and automated mapping. This paper provides a comprehensive review of the applications of AI, particularly deep learning, in landslide inventory mapping. In addition to examining commonly used data sources and model architectures, we explore innovative strategies such as feature enhancement and fusion, attention-boosted techniques, and advanced learning approaches, including active learning and transfer learning, to enhance model adaptability and predictability. We also highlight the remaining challenges and potential research directions, including the estimation of more diverse variables in landslide mapping, multimodal data alignment, modeling regional variability and replicability, as well as issues related to data misinterpretation and model explainability. This review aims to serve as a useful resource for researchers and practitioners, promoting the integration of deep learning into landslide research and disaster management.

Remote Sensing

Challenges and solutions for automated avian recognition in aerial imagery

Remote aerial sensing provides a non-invasive, large geographical-scale technology for avian monitoring, but the manual processing of images limits its development and applications. Artificial Intelligence (AI) methods can be used to mitigate this manual image processing requirement. The implementation of AI methods, however, has several challenges: (1) imbalanced (i.e., long-tailed) data distribution, (2) annotation uncertainty in categorization, and (3) dataset discrepancies across different study sites. Here we use aerial imagery data of waterbirds around Cape Cod and Lake Michigan in the United States to examine how these challenges limit avian recognition performance. We review existing solutions and demonstrate as use cases how methods like Label Distribution Aware Marginal Loss with Deferred Re-Weighting, hierarchical classification, and FixMatch address the three challenges. We also present a new approach to tackle the annotation uncertainty challenge using a Soft-fine Pseudo-Label methodology. Finally, we aim with this paper to increase awareness in the ecological remote sensing community of these challenges and bridge the gap between ecological applications and state-of-the-art computer science, thereby opening new doors to future research.

Remote Sensing in Ecology and Conservation

Experiments with microcomputer-based artificial intelligence environments

The U.S. Geological Survey (USGS) has been experimenting with the use of relatively inexpensive microcomputers as artificial intelligence (AI) development environments. Several AI languages are available that perform fairly well on desk-top personal computers, as are low-to-medium cost expert system packages. Although performance of these systems is respectable, their speed and capacity limitations are questionable for serious earth science applications foreseen by the USGS. The most capable artificial intelligence applications currently are concentrated on what is known as the "artificial intelligence computer," and include Xerox D-series, Tektronix 4400 series, Symbolics 3600, VAX, LMI, and Texas Instruments Explorer. The artificial intelligence computer runs expert system shells and Lisp, Prolog, and Smalltalk programming languages. However, these AI environments are expensive. Recently, inexpensive 32-bit hardware has become available for the IBM/AT microcomputer. USGS has acquired and recently completed Beta-testing of the Gold Hill Systems 80386 Hummingboard, which runs Common Lisp on an IBM/AT microcomputer. Hummingboard appears to have the potential to overcome many of the speed/capacity limitations observed with AI-applications on standard personal computers. USGS is a Beta-test site for the Gold Hill Systems GoldWorks expert system. GoldWorks combines some high-end expert system shell capabilities in a medium-cost package. This shell is developed in Common Lisp, runs on the 80386 Hummingboard, and provides some expert system features formerly available only on AI-computers including frame and rule-based reasoning, on-line tutorial, multiple inheritance, and object-programming. ?? 1988 International Association for Mathematical Geology.

Mathematical Geology

Prevalence of antibodies to type A influenza virus in wild avian species using two serologic assays

Serologic testing to detect antibodies to avian influenza (AI) virus has been an underused tool for the study of these viruses in wild bird populations, which traditionally has relied on virus isolation and reverse transcriptase-polymerase chain reaction (RT-PCR). In a preliminary study, a recently developed commercial blocking enzyme-linked immunosorbent assay (bELISA) had sensitivity and specificity estimates of 82% and 100%, respectively, for detection of antibodies to AI virus in multiple wild bird species after experimental infection. To further evaluate the efficacy of this commercial bELISA and the agar gel immunodiffusion (AGID) test for AI virus antibody detection in wild birds, we tested 2,249 serum samples collected from 62 wild bird species, representing 10 taxonomic orders. Overall, the bELISA detected 25.4% positive samples, whereas the AGID test detected 14.8%. At the species level, the bELISA detected as many or more positive serum samples than the AGID in all 62 avian species. The majority of positive samples, detected by both assays, were from species that use aquatic habitats, with the highest prevalence from species in the orders Anseriformes and Charadriiformes. Conversely, antibodies to AI virus were rarely detected in the terrestrial species. The serologic data yielded by both assays are consistent with the known epidemiology of AI virus in wild birds and published reports of host range based on virus isolation and RT-PCR. The results of this research are also consistent with the aforementioned study, which evaluated the performance of the bELISA and AGID test on experimental samples. Collectively, the data from these two studies indicate that the bELISA is a more sensitive serologic assay than the AGID test for detecting prior exposure to AI virus in wild birds. Based on these results, the bELISA is a reliable species-independent assay with potentially valuable applications for wild bird AI surveillance.

Alaska, California, Delaware, Georgia, Michigan, M

Exposure-related effects of Pseudomonas fluorescens ( Pf -CL145A) on juvenile unionid mussels

The exposure-related effects of a commercially prepared spray-dried powder (SDP) or freeze-dried powder (FDP) formulation of Pseudomonas fluorescens (strain CL145A) on the survival of seven species of newly metamorphosed (<72 hours old) freshwater unionid mussels was evaluated. Forty unionid mussels of each species were randomly distributed to test chambers and each species independently exposed for 24 hours to a static dose of either SDP (four species: Lampsilis cardium , Lampsilis siliquoidea , Lampsilis higginsii , and Ligumia recta ) or FDP (three species: Obovaria olivaria , Actinonaias ligamentina , and Megalonaias nervosa ). Each test chamber was assigned to one of six treatment groups ( n = four chambers per group) by using a randomized block design. The six treatment groups included (1) an untreated control group, (2) groups that received applications with nominal target active ingredient (AI) concentrations of 50, 100, 200, and 300 milligrams per liter (mg/L), and (3) a group that received an application with a nominal target AI concentration of 300 mg/L of heat-deactivated test article (300 HD). After a 24-hour exposure period, water inflow to the test chambers was restored, and the unionid mussels were maintained for an additional 7 days before they were assessed for survival. Mean survival of four unionid mussels species exposed to SDP varied among species and treatment groups when compared to the untreated control groups. The results indicate that exposure to SDP-formulated P. fluorescens up to the maximum label concentration (100 mg/L AI) and up to three times the maximum label exposure duration (8 hours) is not likely to affect the survival of L. siliquoidea and L. higginsii . Low mean survival in the L. recta control group (25.0 percent) indicates that results for L. recta should be interpreted with caution. Mean survival of the L. cardium was significantly lower in all treated groups (14.4 to 40.6 percent) compared to the control group (68.8 percent). These results indicate that further investigation on the impact of SDP-formulated P. fluorescens on L. recta and L. cardium is warranted. Mean survival of three unionid mussels species exposed to FDP was not significantly different in the 50-, 100-, and 200-mg/L AI treatment groups and the 300 mg/L heat-deactivated treatment groups when compared to the control groups. Mean survival of O. olivaria and M. nervosa was significantly lower in the 300-mg/L AI treated groups (38.1 and 48.1 percent, respectively) compared to the control groups (71.9 and 88.1 percent, respectively). The results indicate that exposure to FDP-formulated P. fluorescens up to the maximum label concentration (100 mg/L AI) and up to three times the maximum label exposure duration (8 hours) is not likely to affect the survival of O. olivaria , A. ligamentina , and M. nervosa .

Open-File Report

Uranium—Deposits, production and resources, market dynamics, and supply chain risks

Introduction Interest in nuclear power for the generation of electricity has risen with the increase in the need for more diverse baseload power, enhanced energy security, and the development of new technologies, such as small modular reactors (SMRs), which could provide power for remote areas, industrial applications, and artificial intelligence (AI) data centers. In 2024, the U.S. Department of Energy received $2.7 billion in congressional funding to bolster the domestic uranium production and nuclear fuel supply chain and address reliance on imports from foreign suppliers. In 2025, the U.S. Government issued several Executive and Secretary’s orders aimed at revitalizing the U.S. nuclear sector. If SMRs are to be as widely deployed in the United States and worldwide as envisioned, demand for uranium (nuclear reactor fuel) will likely increase. After the Fukushima nuclear accident in 2011, the market spot price of uranium began a decline, followed by a decrease in U.S. and global uranium exploration and mine development expenditures that led to a uranium supply deficit until 2020, when prices started to recover, prompting a resurgence in uranium exploration and development. In January of 2024, the uranium spot price rose to a 17-year high $106 (U.S. dollars) per pound of U3O8 (triuranium oxide, commonly known as “yellowcake”), which is expected to increase uranium exploration, mine development, and uranium production domestically and worldwide.

Fact Sheet

Knowledge-based geographic information systems (KBGIS): New analytic and data management tools

In its simplest form, a geographic information system (GIS) may be viewed as a data base management system in which most of the data are spatially indexed, and upon which sets of procedures operate to answer queries about spatial entities represented in the data base. Utilization of artificial intelligence (AI) techniques can enhance greatly the capabilities of a GIS, particularly in handling very large, diverse data bases involved in the earth sciences. A KBGIS has been developed by the U.S. Geological Survey which incorporates AI techniques such as learning, expert systems, new data representation, and more. The system, which will be developed further and applied, is a prototype of the next generation of GIS's, an intelligent GIS, as well as an example of a general-purpose intelligent data handling system. The paper provides a description of KBGIS and its application, as well as the AI techniques involved. ?? 1988 International Association for Mathematical Geology.

Mathematical Geology

Cholinesterase activity in black-crowned night-herons exposed to fenthion-treated water

Fenthion, ( 0,0 -Dimethyl 0 -[3-methyl-4-(methylthio)phenyl] phosphorothioate), a widely used mosquito control agent, has caused wildlife mortality. To simulate a shallow wetland environment, an exposure chamber was used containing water treated with fenthion at 1 and 10 times the field application rate of 112 g active ingredient (AI)/ ha. This system permitted an evaluation of exposure routes and the effects of fenthion in a representative species of wading bird, the black-crowned night-heron ( Nycticorax nycticorax ). The results suggested that herons received only a dermal exposure, and that their brain acetylcholinesterase activity was not significantly inhibited. In contrast, however, plasma butyrylcholinesterase activity was inhibited, suggesting the herons were exposed to the insecticide. The application rates and types of exposures were not life-threatening in this species.

Archives of Environmental Contamination and Toxico

Remote sensing-based 3D assessment of landslides: A review of the data, methods, and applications

Remote sensing (RS) techniques are essential for studying hazardous landslide events because they capture information and monitor sites at scale. They enable analyzing causes and impacts of ongoing events for disaster management. There has been a plethora of work in the literature mostly discussing (1) applications to detect, monitor, and predict landslides using various instruments and image analysis techniques, (2) methodological mechanics in using optical and microwave sensing, and (3) quantification of surface geological and geotechnical changes using 2D images. Recently, studies have shown that the degree of hazard is mostly influenced by speed, type, and volume of surface deformation. Despite available techniques to process lidar and image/radar-derived 3D geometry, prior works mostly focus on using 2D images, which generally lack details on the 3D aspects of assessment. Thus, assessing the 3D geometry of terrain using elevation/depth information is crucial to determine its cover, geometry, and 3D displacements. In this review, we focus on 3D landslide analysis using RS data. We include (1) a discussion on sources, types, benefits, and limitations of 3D data, (2) the recent processing methods, including conventional, fusion-based, and artificial intelligence (AI)-based methods, and (3) the latest applications.

Remote Sensing

Chapter 12 - Explainable AI for understanding ML-derived vegetation products

Current machine learning applications and algorithms have developed promise to produce autonomous systems that automatically perceive, learn, predict, and act on their own. However, the effectiveness of these systems is limited by the machine's current inability to explain their decisions, algorithmic paths, and actions to human users. The purpose of this chapter is to apply explainable artificial intelligence (XAI) to black-box models using an example of the U.S. Geological Survey's LANDFIRE Existing Vegetation Type (EVT). This chapter also demonstrates the tools developed to assist scientists/analysts in understanding and trusting prediction outcomes of vegetation type that streamline development of the LANDFIRE EVT product.

Book chapter

Aquatic invasive species in the Chesapeake Bay drainage—Research-based needs and priorities of U.S. Geological Survey partners and collaborators

Executive Summary The U.S. Geological Survey (USGS) is revising the Chesapeake Bay-based science plan to align it with recent U.S. Department of Interior and USGS science priorities that include, as stated in the plan, providing “an integrated understanding of the factors affecting fish habitat, fish health, and landscape conditions” in Chesapeake Bay and its watershed. A report of partner agencies’ needs and priorities related to aquatic invasive species (AIS) science was identified as an informational gap; a report would help to further development of the science program related to aquatic animal health and habitat. This objective was addressed through review of pertinent documentation and conversations with representatives of State, Federal, and regional agencies with vested interests in AIS management in Chesapeake Bay and the Chesapeake Bay drainage area, and this document was produced to summarize the related findings. All agencies and organizations (13) reported that AIS are of general concern, with most stakeholder groups reporting AIS-related issues to be of high priority, including invasive fishes and invertebrates, invasive plants, and microbes including aquatic animal pathogens. Invasive fishes are of great concern to all partner agencies. Channa argus (northern snakehead) and Ictalurus furcatus (blue catfish) are high priority and represent the two most named AIS of concern for these agencies. Nine of 10 stakeholder groups listed northern snakehead as a high priority species, and 6 listed blue catfish as a high priority species as well. Pylodictis olivaris (flathead catfish), invasive crayfish species , and dressenid mussels were also prioritized by multiple partner groups, each receiving specific mention by at least 3 of the 10 stakeholder groups in discussions or documents. Invasive carp, such as Hypophthalmichthys molatrix (silver carp), also received mention by multiple agencies (3 of the 10 stakeholder groups) because these fish represent priority AIS in nearby watersheds and a threat for introduction and dissemination within the Chesapeake Bay watershed from these neighboring regions. Invasive plants are among priority species, and Hydrilla verticillata (hydrilla) topped the list. Hydrilla was reported as a priority species by 5 of the stakeholder groups queried. Trapa natans and T. bispinosa (water chestnut), Phragmites australis (common reed), and Lythrum salicaria (purple loosestrife) were also among the aquatic invasive plants that were prioritized by multiple partner agencies. Multiple stakeholder groups (5 of the 10 groups) also considered Didymosphenia geminata (didymo) and various aquatic animal pathogens among their priority AIS for management considerations. Science needs that were recurrently indicated by stakeholders to support management of invasive species include Technology to enhance biosurveillance capability, such as reliable environmental DNA based detection methodology; Risk assessment modeling to forewarn of and prioritize AIS-related threats; Increased information and intervention methods related to vectors and pathways of AIS introductions; Increased information about the biology and life history of AIS, including information related to trophic interactions, health and disease, and distribution and abundance; and Potential applications of mitigation strategies, including genetically based biocontrol mechanisms. Potential next steps to address the science needs include Development of biosurveillance and risk assessment tools for identification of AIS in proactive management; Development of proactive management techniques to prevent AIS introductions through recognized vectors and pathways; Development of interagency biosurveillance programs to best utilize personnel, funds, and other resources among interested agencies and organizations; Investigations to address life history, consequences, and movement/dissemination of top priority invasive species in the region; Investigations to determine the potential for novel mitigation technologies, such as the application of synthetic biological (genetic) control methods; and Investigations with focus on emerging and high priority AIS in the region, including fishes (blue catfish, flathead catfish, northern snakehead), invertebrates (invasive crayfish and mollusks) and plants (hydrilla, water chestnut, phragmites).

Virginia, Maryland, Delaware, West Virginia, Penns

Effects of the biopesticide Zequanox® on reproduction and early development of the fathead minnow ( Pimephales promelas )

The biopesticide, Zequanox®, is registered for dreissenid mussel control in open water systems in the United States. Previous toxicity trials with nontarget organisms, including several young-of-the-year fish species and invertebrates, demonstrated selectivity of Zequanox for dreissenid mussels, but data are lacking on the treatment-related effects on reproduction and early life stage development of fish. The present study evaluated the effects of Zequanox on spawning and early life stages of the fathead minnow, Pimephales promelas , after exposure to the maximum approved concentration [100 mg active ingredient (AI)/L] and exposure duration (8h) for open water application. The results showed no significant treatment-related effect of Zequanox on survival, condition, or cumulative egg production (21 d) in adult fathead minnow. Eggs (≤24 h old) exposed to Zequanox developed to the eyed-stage at a similar rate to that of untreated eggs. Additionally, Zequanox did not have a significant effect on survival and growth (90 d) of newly hatched larvae (≤24-h old). Zequanox may be an option for control of dreissenid mussels in localized open water habitats where concerns exist regarding reproduction and recruitment of cyprinids and related species.

Management of Biological Invasions

Modeling, mapping, and measuring the risk of freshwater invasive species across Alaska

Freshwater ecosystems of the Alaskan Arctic and Subarctic provide resources that are culturally, ecologically, and economically invaluable. Presently, these regions are relatively free of the impacts from invasive species compared to southern latitudes. To date, there have been relatively few verified introductions of aquatic invasive species (AIS) to freshwater ecosystems in Alaska. The expanding list and distribution of AIS has led to significant negative ecological and economic impacts (e.g., waterweed Elodea nuttalli ; E. canadensis and northern pike Esox Lucius introduced outside its native range in Alaska). Escalating human activity across Alaskan lands and waters, coupled with rapidly shifting environmental conditions, increases the potential for new species introductions and subsequent establishment. Creating a proactive framework for well-informed decision-making and action can improve the effectiveness of prevention efforts and bolster decision support tools that help resource managers direct limited resources. Prioritizing AIS that may be introduced and become established, as well as the locations at highest risk of invasion, is foundational to building a proactive invasive species management framework in Alaska. This project sought to identify and prioritize AIS known to be invasive in the contiguous United States, evaluate current and future habitat suitability for AIS in Alaska, and assess potential for AIS to be transported to habitats across Alaska, utilizing similar assessment methods as implemented for Bering Sea marine invasive species and non-native plants in Alaska. To accomplish this goal, the objectives of the project were to: 1) develop a formal ranked list of potential AIS to freshwater systems of Alaska; 2) assess the level of establishment risk for potential AIS by developing habitat suitability models for waterbodies across Alaska; and 3), identify potential pathways and specific vectors for high-risk AIS to invade Alaska and develop a framework for how vector analysis will be completed to understand transport risk. Overall, our goal is horizon scanning which is defined by Roy et al. (2019) as “a systematic examination of potential threats and opportunities, within a given context, and likely future developments, which are at the margin of current thinking and planning.” The scans include pathway analyses and risk screening of species present at pathway origin points, with a focus on identifying species at high risk of being introduced, becoming established, spreading, and causing harm. We refined a list of 28 AIS from a list of hundreds based on characterizations of species’ invasiveness and species’ proximity to Alaska (USGS 2020; GBIF 2022). Next, we evaluated the relative invasiveness of individual species to create an initial AIS ranking. We sought to characterize habitat suitability of AIS by selecting variables that were continental in scale, covering North America to include Alaska as well as the lower 48 states comparing natural discharge, sub-basin average terrain slope (degrees), average silt fraction, average organic carbon, lithological class, and human footprint in sub-basin in 2009. We estimated AIS habitat suitability across the entire state of Alaska using the physiological tolerances of the AIS (Appendix 2). We also evaluated pathways and vectors for the introduction of AIS (Appendix 2). Many pathways and vectors considered did not meet the criteria for Alaska or freshwater systems. Of the 28 ranked species that we categorized as very high, high, and moderate levels of invasiveness; all three risk groups included fish and mollusks (Appendix 2). One commonality of the very high-invasiveness-ranked species was the availability of Ecological Risk Screening Summary documents (USFWS, 2022) produced by U.S. Fish and Wildlife Service (USFWS), except for the goldfish ( Carassius auratus ) and the New Zealand mudsnail ( Potamopyrgus antipodarum ). The Ecological Risk Screening Summary is now available for New Zealand mudsnails. In general, fish species often ranked very high or high in invasiveness and included sportfish and aquarium fish, suggesting the importance of pathways such as aquarium trade, fishing industry, intentional (but illegal) introductions of sportfishes and aquarium fishes for establishment. The technique we used for habitat suitability models necessitated aquatic environmental datasets that were continental in scale, which was often interpolated from very coarse resolution source data layers, particularly in Alaska. Better spatial data representing aquatic environments would likely improve this approach. While the lack of introductions in Alaska and nearby provinces and states is encouraging, the lack of occurrence data for the focal species also created complications for habitat suitability modeling. Despite the challenges, the habitat suitability models indicated limited suitability for warmwater species while some species, such as Brook trout (S alvelinus fontinalis ), have high habitat suitability across Alaska no matter what threshold approach is taken. Some environmental predictors were more important than others. Specifically, the most important predictor variable, ‘frost free days,’ was critical for 15 out of 28 species as expected due to harsh winter conditions in Arctic and Subarctic regions. The second most important predictor was ‘subbasin land surface runoff’, a variable that indicates the amount of discharge and runoff, while the third most important predictor was ‘snow cover’ another indication of winter conditions. Overall, the ability to understand the effect of future climate scenarios on the establishment of AIS was challenging. A detailed dataset of freshwater temperatures and water chemistry (e.g., pH, calcium) would greatly improve the ability to predict invasiveness of freshwater species to Alaska’s ecosystems on a regional basis. Future studies may benefit from a more focused geographic scope examining a group of subbasins or a regional basin rather than the entire state. These drainages could be selected based upon the mostly likely locations of introduction pathways. The two most prevalent pathway risks for AIS are in-state transfer and stowaways/contaminants. Although there are examples of introductions from other pathways, the risk is somewhat mitigated by Alaska’s climate and regulations. However, variable application of protocols for inspection and cleaning of fishing gear, watercraft, and other similar items while traveling into Alaska as well as transferring from waterbody to waterbody within the state creates a substantial risk in introducing invasive species. We plot cumulative invasive vulnerability for all subbasins and for the top 10% of subbasins (Appendix 3).

Alaska

Modeling waterfowl habitat selection in the Central Valley of California to better understand the spatial relationship between commercial poultry and waterfowl

Wildlife researchers frequently study resource and habitat selection of wildlife to understand their potential habitat requirements and to conserve their populations. Understanding wildlife spatial-temporal distributions related to habitat have other applications such as to model interfaces between wildlife and domestic food animals in order to mitigate disease transmission to food animals. The highly pathogenic avian influenza (HPAI) virus represents a significant risk to the poultry industry. The Central Valley of California offers a unique geographical confluence of commercial poultry and wild waterfowl, which are thought to be a key reservoir of avian influenza (AI). Therefore, understanding spatio-temporal distributions of waterfowl could improve our understanding of potential risk of HPAI exposure from a commercial poultry perspective. Using existing radio-telemetry data on waterfowl (U.S. Geological Survey) in combination with habitat and vegetation data based on Geographic Information Systems (GIS), we are developing GIS-based statistical models that predict the probability of waterfowl presence (Habitat Suitability Mapping). Near-real-time application can be developed using recent habitat data derived from Landsat imagery (acquired by satellites and publicly available through the U.S. Geological Survey) to predict temporally- and spatially-varying distributions of waterfowl in the Central Valley. These results could be used to provide decision support for the poultry industry in addressing potential risk of HPAI exposure related to waterfowl proximity.

California

Application of microwave digestion to the analysis of peat

A microwave digestion technique for the dissolution of peat is described and compared with a dry ashing method and a nitric - perchloric - hydrofluoric acid wet digestion. Peat samples with different organic matter contents were used and Ca, Mg, Fe, AI, Na, K, Mn, Zn, Cu and Li were determined by atomic absoprtion spectrometry. The results obtained using the three dissolution techniques were in good agreement. The microwave method has the advantage of digesting the samples in less than 2 h and uses less acid than the conventional wet digestion method. Keeping the volume of the acid mixture as small as possible minimises contamination and leads to lower blank values.

The Analyst

Segment anything model can not segment anything: Assessing AI foundation model's generalizability in permafrost mapping

This paper assesses trending AI foundation models, especially emerging computer vision foundation models and their performance in natural landscape feature segmentation. While the term foundation model has quickly garnered interest from the geospatial domain, its definition remains vague. Hence, this paper will first introduce AI foundation models and their defining characteristics. Built upon the tremendous success achieved by Large Language Models (LLMs) as the foundation models for language tasks, this paper discusses the challenges of building foundation models for geospatial artificial intelligence (GeoAI) vision tasks. To evaluate the performance of large AI vision models, especially Meta’s Segment Anything Model (SAM), we implemented different instance segmentation pipelines that minimize the changes to SAM to leverage its power as a foundation model. A series of prompt strategies were developed to test SAM’s performance regarding its theoretical upper bound of predictive accuracy, zero-shot performance, and domain adaptability through fine-tuning. The analysis used two permafrost feature datasets, ice-wedge polygons and retrogressive thaw slumps because (1) these landform features are more challenging to segment than man-made features due to their complicated formation mechanisms, diverse forms, and vague boundaries; (2) their presence and changes are important indicators for Arctic warming and climate change. The results show that although promising, SAM still has room for improvement to support AI-augmented terrain mapping. The spatial and domain generalizability of this finding is further validated using a more general dataset EuroCrops for agricultural field mapping. Finally, we discuss future research directions that strengthen SAM’s applicability in challenging geospatial domains.

Remote Sensing

Effects on birds of fenthion aerial application for mosquito control

Effects on birds of an aerial application of fenthion, a potent organophosphorus cholinesterase (ChE)-inhibiting insecticide, were assessed on four study sites 1.8 to 3.6 km 2 in size. These sites were located within 121.5 km 2 of wet meadows treated with 47 g of fenthion (AI) per ha in ultra-low-volume formulation. Assessment methods were searches for sick or dead birds, measurements of brain ChE activity in specimens found dead or collected alive at different time intervals, and counts of bird populations. After treatment, 99 birds and 15 mammals were found sick or dead; 106 of these were on one site. Brain ChE activity in dead birds was depressed sufficiently to indicate that death was caused by an anti-ChE substance. Brain ChE activity in three common bird species collected alive showed the greatest reduction 2 days postspray. Two of these species had ChE activity that was still significantly ( P <0.05) depressed 15 days postspray. Bird populations declined most where mortality was heaviest. Fenthion sprayed for mosquito control was life threatening to many birds inhabiting treated meadows.

Journal of Economic Entomology