Artificial Intelligence (AI) Use Case Inventory

By tracking current and planned Artificial Intelligence (AI) use cases – and sharing those with relevant internal and external audiences through Interior's Use Case Inventory – the Department will be able to identify and prioritize deployment of uses that are proven to enhance mission delivery, while also spurring the development of new use case ideas for development and testing. In addition to promoting the creation of new use cases, Interior will also work with partners within the Department, across the federal government, and in the academic, nonprofit, and private sectors to collaborate on adapting use cases to fit Interior’s mission. The Department does not currently have any AI use cases that are rights or safety impacting.   

Use Case Name Agency Bureau Use Case Topic Area Summary Stage of Development
DOIChatGPT AI Chatbot Department of the Interior  OCIO Mission-Enabling (internal agency support) Create for DOI users a secure space to chat with a Large Language Model without leaving our security boundary.

This is a customized version of Microsofts Govchat repo deployed as an azure app service on the internal network with entra authentication, using DOIChatGPT APIM as the back end AI service provider.		 Implementation and Assessment
DOIChatGPT API Management Instance Department of the Interior  OCIO Mission-Enabling (internal agency support) An Azure OpenAI instance secured behind an Azure API Management instance hosted within our internal network. This will create a backbone of secure access to Azure OpenAI permitted by APIM subscription keys that also supports cost tracking per project key.

This is just a service layer to enable other project activities.		 Implementation and Assessment
DOIChatGPT API Management Development Environment Department of the Interior  OCIO Mission-Enabling (internal agency support) This will be a public facing version of the DOIChatGPT APIM and Azure OpenAI that is used for training and proof of concept validation (with sanitized data) prior to migrating to production DOIChatGPT APIM. Implementation and Assessment
Cost Estimation Using EDL Database Data Department of the Interior  OS - Office of the Secretary of the Interior Mission-Enabling (internal agency support) This use case aims to leverage AI/ML for estimating environmental remediation costs based on historical data within the EDL (Environmental Data Liability) database. The model would analyze archived and current records to provide estimates that aid budget planning and contractor evaluations.

Problem Being Solved: To enhance budget estimation accuracy for environmental remediation activities by comparing new sites with archived data of similar sites to estimate total cleanup costs.

Goal: The goal is to improve cost estimation accuracy, enhance contractor proposal evaluation, and aid in long-term financial planning.


 		 Initiated
Adobe Firefly Generative Images Department of the Interior  NPS Mission-Enabling (internal agency support) Adobe's Firefly generative image AI model was used within Photoshop to create artistic sketches of different nature scenes. These sketches were used in the StoryMap created by San Juan Island National Historical Park as artistic and narrative elements. The images were generated in Photoshop and then manually edited to ensure accuracy. Using this AI model created results that were quicker and cheaper compared to traditional artistic methods. The StoryMap with the sketches can be viewed here: https://arcg.is/1uCy0e Operation and Maintenance
RAG knowledge base of agent procedures Department of the Interior  OCIO Government Services (includes Benefits and Service Delivery) Custom AI Solutions for Internal and Customer Support
Description:    We are addressing the challenge of providing quick, accurate, and consistent responses to both internal customer service agents and external customers. Currently, agents must manually search through extensive documentation, which can lead to delays in service response times, inconsistencies in information shared, and a potential for human error. Additionally, external customers often need direct access to information on our website but must wait for human support, which impacts customer satisfaction.
 
By implementing a customized AI chatbot, we aim to streamline access to knowledge for agents, reducing their workload and improving efficiency. Additionally, a customer-facing chatbot would offer external customers instant answers to common questions, further enhancing customer experience and reducing the volume of support inquiries handled by human agents.		 Implementation and Assessment
Use of AI to Enhance Flash Flood Forecast Tool Department of the Interior  NPS Mission-Enabling (internal agency support) University of Illinois created a model to predict rainfall on a watershed scale in Great Smoky Mountains National Park.  They are now working with IBM to create a system that will use this model to provide forecasts of flooding events with a goal of a 24+ hour lead time.   The developers are using AI to improve on the Quantitative Precipitation Forecast of the National Weather Service to thus improve on the accuracy of their tool.  The NPS is cooperating as an end-user of the flood forecast tool, but not as a developer or direct user of the AI. Implementation and Assessment
Liable Party Research Department of the Interior  ONRR Mission-Enabling (internal agency support) Supervised machine learning model for scraping leasing documents for current liable party ownership. Implementation and Assessment
Data Extraction Using MS Power Automate AI Functionality Department of the Interior  ONRR Mission-Enabling Train pre-built Machine Learning model to extract data from different document formats and convert into a consistent structured format to improve business operations. Acquisition and/or Development
CESU project to detect of bird nests using deep learning to support annual colonial bird monitoring Department of the Interior  NPS Mission-Enabling (internal agency support) This work is already being accomplished via a cooperative agreement with a university (Florida International University)

The National Park Service (NPS), has been monitoring colonial nesting birds monthly via low level aerial photography from a helicopter platform in Biscayne National Park since 2010. These photographs are processed to determine the number of active nests for specific focal species: Double-crested Cormorants, Great Egrets, Great White Herons, Great Blue Herons, White Ibises, and Roseate Spoonbills. NPS now has 13 years of data which is composed of 64,350 uniquely identified nests and 57,793 birds representing.

This monitoring is time consuming and needs to be accomplished consistently and accurately. Currently, the aerial photos are processed by individuals. Which takes substantial time and effort. We propose to explore the creation of automatic script that would allow the active nest to be identified using software. The goal of this project is to explore software programs and create automatic script that can identify potential bird nests on the photograph and to highlight this area on the photograph. By developing a robust object detection model, researchers can monitor bird colonies and their populations more efficiently, e.g. by consistently identifying nests with software which will allow the tracking of nesting patterns with less person effort needed to accomplish this task. It will also help to control observer detection variation over time.		 Implementation and Assessment
Image Generation and Audio Video Editing Department of the Interior  ONRR Mission-Enabling To efficiently create and edit images, audio, and video. Operation and Maintenance 
Use of AI for developing bioacoustic and remote camera imagery wildlife species classifiers for noni Department of the Interior  NPS Mission-Enabling (internal agency support) University of Illinois created a model to predict rainfall on a watershed scale in Great Smoky Mountains National Park. They are now working with IBM to create a system that will use this model to provide forecasts of flooding events with a goal of a 24+ hour lead time. The developers are using AI to improve on the Quantitative Precipitation Forecast of the National Weather Service to thus improve on the accuracy of their tool. The NPS is cooperating as an end-user of the flood forecast tool, but not as a developer or direct user of the AI. Implementation and Assessment
Machine Learning Model Optimization Department of the Interior  ONRR Mission-Enabling Alteryx machine learning model optimization and fitting helps users select which model is optimal to fit to an underlying dataset. It also can help refine the model, recommending variable removal if autocorrelation issues are detected in the dataset. Implementation and Assessment
NPS Acquisition Management Workload Software Department of the Interior  NPS Government Services (includes Benefits and Service Delivery) Provide assistance for employees working on procurment packages by drafting documents and summarizing content.  Implementation and Assessment
Caseguard Studio Department of the Interior  NPS Mission-Enabling Our office uses CaseGuard Studio to streamline and enhance the accuracy of video and audio redactions for law enforcement and public records. This tool addresses the challenge of protecting sensitive information while ensuring compliance with privacy laws and FOIA requirements, improving efficiency and reducing manual workload. Implementation and Assessment
Azure based Internal Controls Testing App for Award Descriptions Department of the Interior  OS - Office of the Secretary of the Interior Mission-Enabling PGM will work with Microsoft to develop an application that will live the the secure Azure environment and allow bureaus to conduct their own testing of project descriptions as a learning tool. The application will be locked to the same prompts for review of project descriptions but will allow bureau financial assistance policy personnel to run their own tests to compare to FY 2023 results and look for improvements. Acquisition and/or Development
Internal Controls Testing - SAM.gov certifications Department of the Interior  OS - Office of the Secretary of the Interior Mission-Enabling Federal financial assistance regulation and DOI PGM policy requires awarding officers check SAM.gov, a Federal public website, to ensure that the entity funds are to be award to is eligible to receive Federal funds and that the POCs on the awards are not debarred. The use case involves using a large language AI model to assess and score SAM.gov pdf documents against the date the award was made. Acquisition and/or Development
Potential Applications of AI Models to UAS Post-fire Mapping Data Analysis and Image Processing Department of the Interior  OS - Office of the Secretary of the Interior Mission-Enabling Evaluating AI Models for UAS Post-Fire Mapping Analysis Focusing on the Following Issues...
Post-Fire Damage Assessment:   Efficiently allocate resources for future rehabilitation efforts.
Vegetation Recovery Prediction:   Assess burn severity to estimate ecosystem regeneration speed.
Soil Erosion Risk Mapping:  Identify vulnerable areas and guide erosion control measures.
Habitat Restoration Planning:   Prioritize areas for reforestation, invasive species removal, and soil stabilization.
Water Quality Monitoring:  Predict sediment runoff, nutrient levels, and contaminants for future water management.
Structural Evolution Analysis and Prediction:  Analyze fire spread and growth to aid risk assessment and response planning.
Environmental Impact Assessment: Aid in planning rehabilitation efforts and assessing ecosystem resilience.
Risk Communication and Public Awareness: Inform communities about rehabilitation efforts and safety precautions.
Ongoing Post-Fire Rehabilitation Treatment Effects:  Assist BAER teams in assessing burned landscapes and monitoring ecological effects.
Cross-Disciplinary Integration:   Bridge communication gaps between experts in various fields for wildfire risk assessment.
Literature Reanalysis:  Assist researchers in developing insights, identifying gaps, and informing future research.
Assessing Damage and Recovery:   Identify fire-damaged objects (e.g., infrastructure, trees) to prioritize rehabilitation.
Monitoring Ecosystem Recovery:   Identify plant species, soil erosion patterns, and ecological indicators for restoration planning.
Identifying Hazards and Safety Risks:   Identify post-fire hazards (e.g., unstable terrain, debris) to ensure safety during rehabilitation.
Identifying Cultural and Archeological Objects: Identify relevant objects (e.g., artifacts, disturbances) for avoidance during rehabilitation.
Customized Rehabilitation Plans:   Detect specific objects (e.g., utility poles, contaminated soil) for tailored rehabilitation plans.		 Initiated
Proposal for the I-NEPA System: Leveraging Artificial Intelligence (AI) for Enhanced Efficiency in E Department of the Interior  OS - Office of the Secretary of the Interior Government Services (includes Benefits and Service Delivery) I-NEPA aims to streamline and enhance environmental reviews, foster improved public engagement, and facilitate advanced searching, tracking, and reporting of both past and present NEPA actions throughout the Department of the Interior. I-NEPA will leverage the Retrieval Augmented Generation (RAG) process, combining NEPA and policy documents and user input into a large language model (LLM) prompt to get tailored output for each project. Implementation and Assessment
Public Comment Analysis Tool (PCAT): Leveraging Artificial Intelligence (AI) for Enhanced Efficiency Department of the Interior  OS - Office of the Secretary of the Interior Government Services (includes Benefits and Service Delivery) Federal agencies struggle with efficiently processing and analyzing public comments, particularly for NEPA actions involving controversial topics with millions of responses. Manual processing is labor-intensive and introduces delays and inconsistencies. PCAT seeks to address these challenges by automating the analysis process steps only to improve accuracy and efficiency. Implementation and Assessment
Pilot: Using Machine Learning to Harvest Data for Standardized Species Conservation Department of the Interior  FWS Science & Space The U.S. Fish and Wildlife Service is in the process of reimagining and creating a new database to hold, depict, and disseminate information about the species we manage. This effort will greatly expand and standardize the data related to species and their conservation. We have worked with contractors to develop a preliminary tool that explores the possibiliy of using machine learning and large language modelling to harvest species information related to their biological and physical needs into standardized pick-lists. This includes a way to then vet the harvested data. As part of this we are evaluating the infrastructure needed to containerize tool like this in our environment. Initiated
Summarization of documents and output to ECOSphere species workflow Department of the Interior  FWS Science & Space The U.S. Fish and Wildlife Service has a substantial number of documents that we would like to take advantage of NLP or ML process in order to summarize their content.  Afterwards, we aim to integrate that summmarized data, using some sort of tailored AI and framework or API, into our species workflows in ECOSPhere as it relates to species information.  Initiated
Prediction of Suitable Habitat for ESA-listed Species Department of the Interior  FWS Science & Space The U.S. Fish and Wildlife Service currently uses species distribution models (i.e., machine learning algorithms) to predict potential habitat for ESA-listed threatened and endangered species.   These habitat predictions are vetted by USFWS field biologists and, in some cases, are used as ranges in IPaC to inform users of possible impacts on ESA-listed species.  Species distribution model outputs have also been used to identify suitable habitat on the landscape for potential reintroduction efforts or suitable areas for additional surveys, with the potential to locate new, previously undocumented populations. Implementation and Assessment
Computer Vision Model to Rapidly Identify Habitat Change Department of the Interior  FWS Science & Space The U.S. Fish and Wildlife Service is planning to partner with the Chesapeake Conservancy, an NGO, to develop a computer vision model to monitor for early detection of habitat loss across the landscape, a significant threat to biodiversity, including threatened and endangered species.  By using a computer vision model one can rapidly identify and flag areas where habitat loss may be occurring due to natural or human-caused disturbances.  Early detection can facilitate rapid responses, when appropriate, or allow practitioners to accurately calculate habitat loss over time.  More accurate estimates of habitat loss allow for better management decisions and potentially shorter recovery times for threatened and endangered species. Acquisition and/or Development
Applying Deep Learning to Detect and Classify Ocean Wildlife Department of the Interior  FWS Mission-Enabling Deep learning methods are being advanced to automate data processing and improve the cost-efficiency of remote sensing technologies for surveys covering broad geographic areas and generating very large image datasets. The FWS is partnering with the Bureau of Ocean Energy Management (BOEM), U.S. Geological Survey (USGS), academic institutions, and private contractors to accomplish these objectives. Initial focus has been on marine bird and other wildlife surveys given overlapping agency requirements for these data. A workflow is now being established to advance the imagery data from the sensors, to an AI detection model in the aircraft, to a species classification algorithm that is able to distinguish and count each species, and then a georeferenced location is obtained for each bird. More specifically, a cutting-edge artificial intelligence/deep learning algorithm has been developed to automatically detect seabirds in imagery. This detection algorithm has 80 to 90% accuracy across a wide range of seabird species. In FY25, we will continue to iterate on this model using a human-in-the-loop technique to improve model generalization. Acquisition and/or Development
Enhancing Migratory Bird Surveys with Thermal Imagery and Deep Learning Department of the Interior  FWS Mission-Enabling The U.S. Fish and Wildlife Service (USFWS) Division of Migratory Bird Management is aiming to improve upon migratory bird surveys by utilizing aerial remote sensing combined with deep learning (a form of artificial intelligence) analyses to automate survey counts. The goal is to provide accurate wildlife counts while simultaneously reducing risk to pilots by allowing aerial surveys to occur at higher altitudes. In partnership with the College of William and Mary, USFWS has previously demonstrated that thermal remote sensing technology, coupled with deep learning, can provide accurate counts of sandhill cranes (Antigone canadensis) at night during their critically important migratory stopover in the Platte River Valley of Nebraska Acquisition and/or Development
SOCS documents analysis Department of the Interior  BOEM Mission-Enabling The BOEM Status of the Outer Continental Shelf (SOCS) environmental information portal contains over 900 BOEM documents including Environmental Impact Statements (EISs), Environmental Assessments (EAs), Categorical Exclusion Reviews (CERs), Biological Assessments (BAs), Biological Opinions (BiOPs), National Historic Preservation Act (NHPA) Section 106 documents, policy documents, environmental guidance, reusable content, and environmental studies. Some potential use cases for integrating LLM/AI into SOCS:
 - Generate resource affected sections for different resources based on existing recent EISs and pull in new information to keep the analysis current. For example, if a wind energy lease is being contemplated in the vicinity of an already analyzed wind energy lease, the information would be pulled from the existing environmental analyses as a starting point, incorporating any new information in the system. The drafted information would include citations and a references cited section in a particular format. 
- Have LLM/ summarize key findings from multiple studies on similar resources. For example, multiple studies addressing noise or sound impacts to different species could be summarized with references to the source publications.		 Implementation and Assessment
ONRR Video Hosting Platform [OVHP] Department of the Interior  ONRR Mission-Enabling ONRR's Video Hosting Platform is a secure, internal-only video hosting site hosting training videos, all-hands meetings, health & wellness, SEP events and more.

AI-generation of captions/subtitles, metadata, and chapters to make video hosting, editing, and searching more efficient. AI-generated captions support Section 508 requirements.		 Operation and Maintenance 
Video Creation and Editing Department of the Interior  ONRR Mission-Enabling To efficiently generate captions, subtitles, text-to-speech, video, photo, and audio. Operation and Maintenance 
Predictive AI Applications for Wildlife Monitoring: SeeOtter, a custom built software solution for a Department of the Interior  FWS Science & Space Marine Mammals Management is tasked with Stock Assessments under the Marine Mammal Protection Act, Section 117 and sea otter population monitoring has shifted to image-based surveys, relying on on a YoloV5, AI-assisted process to procure a data set of sea otter observations from photos. Department of Interior agencies, USFWS, USGS, and NPS have all adopted an image-based methodology for sea otter population monitoring, but the model requires continued maintenance and refinement with new surveys, sensor upgrades, and model updates. Unfortunately, there are no AI expert permanent staff tased with this project across Alaska DOI programs and the need has been fulfilled through contractors. SeeOtter was a custom developed software by Collin Power and Evan Wetherington, under contract to USFWS Sea Otter Program, to turn hundreds of thousands of images from across the Gulf of Alaska into datasets usable for statistical analyses by USGS and Stock Assessment Reporting by USFWS and NPS. Glacier Bay NPS assisted in developing an SOP for the AI-assisted processing for sea otter data and Collin Power, under contract with USGS built the system out through a Center for Data Integration Project to make it more accessible to interested users. We would like SeeOtter to be a tool availabel to Partners, across DOI as well as Tribes and researchers, who are interested in applying a predictive AI model for counting objects from imagery.

USGS also have plans for large scale photo surveys coming up in 2025/26 and we will be working collaboratively to figure out the support for image processing following the SeeOtter SOP that has been developed in a partnership among USFWS, USGS, and NPS.		 Operation and Maintenance
Non-Generative AI use for Trust Information Analysis and Reporting Tool Department of the Interior  BTFA   The use of Non-Generative AI for identifying   blank pages, document classification, title extraction, date extraction. This endeavor is using Azure Computer Vision and Document Intelligence Read  to help with identifying blank pages. Azure AI Search Services (This is on hold right now and not listed under AI Tool Name) for document classification, title extraction and date extraction. Acquisition and/or Development
Intelligent Optical Character Recognition Department of the Interior  IBC   IOCR is machine learning software that replaces the need for collecting documents and manually entering them into systems.  The software learns over time based on volume fed through the software and corrects common errors automatically, such as 0 vs O.  Invoice processing is one of the immediate applications considered for IOCR.  IOCR will speed the flow of data to the certifying officer for review and ultimate payment.  This capability also has broader potential to enhance data extraction, validation, and preservation.  IOCR will be deployed in conjunction with Robotics Process Automation to allow for more comprehensive automation solutions. The same robust testing and validation of automations prior to deployment to production will be applied to IOCR. Implementation and Assessment
Seasonal/Temporary Wetland/Floodplain Delineation using Remote Sensing and Deep Learning Department of the Interior  BOR Mission-Enabling Reclamation was interested in determining if recent advancements in machine learning, specifically convolutional neural network architecture in deep learning, can provide improved seasonal/temporary wetland/floodplain delineation (mapping) when high temporal and spatial resolution remote sensing data is available? If so, then these new mappings could inform the management of protected species and provide critical information to decision-makers during scenario analysis for operations and planning. Implementation and Assessment
Data Driven Sub-Seasonal Forecasting of Temperature and Precipitation  Department of the Interior  BOR Mission-Enabling Reclamation has run 2, year-long prize competitions where particants developed and deployed data driven methods for sub-seasonal (2-6 weeks into future) prediction of temperature and precipitation across the western U.S. Particpants outperformed benchmark forecasts from NOAA. Reclamation is currently working with Scripps Institute of Oceanography to further refine, evaluate, and pilot implement the most promising methods from these two copmetitions. Improving sub-seasonal forecasts has significant potential to enhance water management outcomes.  Implementation and Assessment
Data Driven Streamflow Forecasting Department of the Interior  BOR Mission-Enabling Reclamation, along with partners from the CEATI hydropower industry group (e.g. TVA, DOE-PNNL, and others) ran a year-long   evaluation of existing 10-day streamflow foreasting technologies and a companion prize competition open to the public, also focused on 10-day streamflow forecasts. Forecasts were issued every day for a year and verified against observed flows. Across locations and metrics, the top perfoming foreacst product was a private, AI/ML forecasting company - UpstreamTech. Several competitors from the prize competition also performed strongly; outperforming benchmark forecasts from NOAA. Reclamation is working to further evaluate the UpstreamTech forecast products and also the top performers from the prize competition.  Implementation and Assessment
Snowcast Showdown  Department of the Interior  BOR Mission-Enabling Reclamation partnered with Bonneville Power Administration, NASA - Goddard Space Flight Center, U.S. Army Corps of Engineers, USDA - Natural Resources Conservation Service, U.S. Geological Survey, National Center for Atmospheric Research, DrivenData, HeroX, Ensemble, and NASA Tournament Lab to run the Snowcast Showdown Prize Competition. In this competition, particiapnts were asked to develop mehtods to estimate distributed snow information by blending observations from different sources   using machine learning methods that provide flexible and efficient algorithms for data-driven models and real-time prediction/estimation. Winning methods are now being evaluated and folded into a follow-on project with NOAA's River Forecast Centers.  Implementation and Assessment
PyForecast  Department of the Interior  BOR Mission-Enabling Pyforecast is a statistical/ML water supply forecasting software developed by Reclamation that uses a range of data-driven methods.  Implementation and Assessment
Improved Processing and Analysis of Test and Operating Data from Rotating Machines Department of the Interior  BOR Mission-Enabling This project is exploring a better method to analyze DC ramp test data from rotating machines. Previous DC ramp test analysis requires engineering expertise to recognize characteristic curves from DC ramp test plots. DC ramp tests produce a plot of voltage vs current for a ramping voltage applied to a rotating machine. By using machine learning/AI tools, such as linear regression, the ramp test plots can be analyzed by computer software, rather than manual engineering analysis, to recognize characteristic curves. The anticipated result will be faster and more reliable analysis of field-performed DC ramp testing. Acquisition and/or Development
Improving UAS-derived photogrammetric data and analysis accuracy and confidence for high-resolution data sets using artificial intelligence and machine learning Department of the Interior  BOR Mission-Enabling UAS derived photogrammetric products contain a large amount of potential information that can be less accurate than required for analysis and time consuming to analyze manually. By formulating a standard reference protocol and applying machine learning/artificial intelligence, this information will be unlocked to provide detailed analysis of Reclamation's assets for better informed decision making. Implementation and Assessment
Photogrammetric Data Set Crack Mapping Technology Search  Department of the Interior  BOR Mission-Enabling This project is exploring a specific application of photogrammetric products to process analysis of crack mapping on Reclamation facilites.  This analysis is time consuming and has typically required rope access or other means to photograph and locate areas that can now be reached with drones or other devices.  By formulating a standard reference protocol and applying machine learning/AI, this information will be used to provide detailed analysis of Reclamation assets for better decision making.  Implementation and Assessment
Effects of vehicle traffic on space use and road crossings of caribou in the Arctic Department of the Interior  USGS Mission-Enabling Assessing the effects of industrial development on wildlife is a key objective of managers and conservation practitioners. However, wildlife responses are often only investigated with respect to the footprint of infrastructure, even though human activity can strongly mediate development impacts. In Arctic Alaska, there is substantial interest in expanding energy development, raising concerns about the potential effects on barren-ground caribou (Rangifer tarandus granti). While caribou generally avoid industrial infrastructure, little is known about the role of human activity in moderating their responses, and whether managing activity levels could minimize development effects. To address this uncertainty, we examined the influence of traffic volume on caribou summer space use and road crossings in the Central Arctic Herd within the Kuparuk and Milne Point oil fields on the North Slope of Alaska. Implementation and Assessment
Automated Walrus Haulout Monitoring Department of the Interior  USGS Mission-Enabling The purpose of this software is to provide a framework for using pre-trained image classification convolutional neural network CNN models to (1) make predictions on unlabled image datasets to provide data for further analysis of walrus (Odobenus rosmarus) coastal haulout occupation and (2) make predictions on image datasets where keywords have been manually (i.e. by human reviewers) added to the image ITPC metadata indicating the content of the images, including the presence and absence of walruses so that the model's performance can be evaluated by comparing its predictions to the human-assigned class labels using confusion matrices and standard classifier evaluation metrics such as accuracy, precision, recall, F1-score (a composite of precision and recall). Acquisition and/or Development
Forecasting Earthquake Ground Motion Time Series Department of the Interior  USGS Mission-Enabling Development of a deep learning models to generate earthquake ground motion time series for potential application to Earthquake Early Warning, Operational Aftershock Forecasting, and the National Seismic Hazard Model. Acquisition and/or Development
CONUS EcoFlows Planning & Prototype Department of the Interior  USGS Mission-Enabling To support development of national-scale ecological-flow response models, a benchmark estimate of typical, long-term unaltered flow conditions is needed as a reference benchmark to evaluate flow departures from "normal". This model uses unaltered, near-reference USGS gages from across the US, and associated natural landscape datasets, to predict long-term average monthly flows. These models, trained in near-reference sites, are then used to predict the hypothetical near-reference long-term flow conditions of sites with anthropogenic influences, where we have paired biological datasets. By comparing observed flows to these 'natural' expectations, we calculate flow departures from a 'natural normal' and which is used in later ecological modeling. Implementation and Assessment
Prioritized Constituents: Sediment Department of the Interior  USGS Mission-Enabling Regional prediction of suspended sediment concentration in unmonitored rivers to characterize sediment transport in the Delaware, Illinois, and Colorado River Basins. Implementation and Assessment
Avian population estimates from passive acoustic monitoring Department of the Interior  USGS Mission-Enabling Reliable estimates of avian abundance from acoustic recordings Acquisition and/or Development
FEMA mixed population flood-frequency analysis Department of the Interior  USGS Mission-Enabling Classification of historical floods based on causal mechanisms to support improved estimation of flood reoccurrence intervals Acquisition and/or Development
Nutrient, Salinity, and temperature model development Department of the Interior  USGS Mission-Enabling Multiple projects designed to simulate nutrients (phosphorus and nitrate), temperature, and salinity in streams across the U.S. using machine learning approaches. Initiated
Data-Driven Streamflow Drought Department of the Interior  USGS Mission-Enabling Prototype streamflow drought forecasts using data-driven, machine learning approaches for USGS gage locations across the continental U.S. Implementation and Assessment
AI/ML for aquatic science Department of the Interior  USGS Mission-Enabling This project aims to develop novel computational frameworks and AI algorithms for individual fish recognition, by leveraging AI, computer vision and deep learning. The main objectives of this project include:
(1) Develop baseline AI models by exploiting visual features and pre-trained deep learning models.
(2) Improve individual fish recognition performance, as well as handling new individuals and exploring dynamic environments.
(3) Evaluate melanistic markings associated with blotchy bass syndrome to assess the capacity for AI detection of diseased fish.
(4) Evaluate deep learning models for individual recognition and respiration rate (ventilate rate) using video data collected in laboratory settings and natural streams.		 Initiated
National-Extent Groundwater Quality Prediction for the National Water Census and Regional Integrated Department of the Interior  USGS Mission-Enabling The primary objectives of this project are to (1) provide Nationally consistent predictions of groundwater quality (salinity and nutrients) relevant for human and ecological uses and its influence on surface-water, and (2) develop strategies for integrating these predictions into comprehensive water-availability assessments including the National Water Census and regional Integrated Water Availability Assessments. Implementation and Assessment
Use of artificial intelligence tools for optimization and documentation for computer codes Department of the Interior  USGS Mission-Enabling For the USGS National Seismic Hazard Model and other earthquake hazards research, computer codes are needed that implement earthquake rupture forecasts and ground-motion models. This project uses ChatGPT to suggest optimizations and documentation for computer codes. Implementation and Assessment
Improved earthquake detection for research studies Department of the Interior  USGS Mission-Enabling Deep learning methods are being used to improve detection of earthquakes to provide more complete, high-resolution catalogs that are used in research to better understand earthquake occurrence, rupture processes and seismic hazard. Acquisition and/or Development
NGWOS External R&D - Using advanced computing techniques for image-based monitoring Department of the Interior  USGS Mission-Enabling Multiple projects conducting research on AI/ML techniques for image-based water monitoring Acquisition and/or Development
Water Use Model Development Department of the Interior  USGS Mission-Enabling Goal to develop process-based and machine learning models to estimate multiple categories of water use across the U.S. Implementation and Assessment
National Temperature Observations Department of the Interior  USGS Mission-Enabling The objectives of this project are to reduce the burden on Science Centers for the collection, storage, analysis, and processing of quality assurance data with the expectation this will lead to an increase of deployed sensors in the water temperature network. More specifically the project will (1) modify software to allow for  processing and storage of discrete water temperature data collected during streamflow measurements, (2) implement workflows and QA checks in data collection software that supports new temperature policies and procedures (3) create a pilot program to support Science Centers in accomplishing 5-pt temperature checks. Initiated
Downscaling and assimilation of meteorology data for local estimates of irrigation water demand. Department of the Interior  USGS Mission-Enabling We found that in the context of satellite remote sensing for estimates of crop water use, high-frequency and low resolution gridded meteorological reanalysis data contributes about half the total error in our estimates. This is due to well-documented localized effects of terrain and land use that aren't accounted for in meteorological forcing datasets such as NLDAS-2 or ERA-5. Our work suggests that assimilating local meteorological station data alongside land use and terrain information may improve the accuracy of our estimates of local-scale atmospheric demand (i.e., reference evapotranspiration, a function of temperature, wind speed, humidity, and solar radiation). The relationship between the reanalysis products and observed local meteorology appears complex and non-linear. We've found that using deep learning techniques, including Long-Short Term Memory (LSTM) time series modeling along Graph Neural Networks to assimilate local meteorological observations offers promise to improve our local-scale estimates atmospheric demand, and thus irrigated crop water use. We hope to use the information we've gathered to add value to existing meteorological forcing datasets by developing software that reliably downscales low spatial resolution datasets to make more accurate estimates of irrigation water use. Such a product would be especially useful in areas without the highly-developed meteorology and climatology datasets that are exclusive to the United States, such as PRISM and GridMET. Initiated
National-Extent Groundwater Quality Prediction for the Integrated Water Availability Assessments Department of the Interior  USGS Mission-Enabling The primary objectives of this project are to (1) provide Nationally consistent predictions of groundwater quality (salinity and nutrients) relevant for human and ecological uses and its influence on surface-water, and (2) develop strategies for integrating these predictions into comprehensive water-availability assessments including the National Water Census and regional Integrated Water Availability Assessments. Implementation and Assessment
Building capacity for assessment and prediction of post-wildfire water availability Department of the Interior  USGS Mission-Enabling Model development to forecast water quality impacts of wildfires in the western U.S. focused on suspended sediment and salinity Initiated
Mapping sagebrush from drones to satellites Department of the Interior  USGS Mission-Enabling Vegetation maps are critical tools for that are widely used in applications including habitat modeling, evaluating effectiveness of habitat restoration, and understanding the ecological implications of biological invasions. We are using machine learning and imagery from unmanned aerial vehicles (UAV), aircraft, and satellites to extend presence modeling to map fractional cover of sagebrush in the Dakotas, where accurate maps of sagebrush are needed to identify seasonal habitats of sage-grouse for the Bureau of Land Management. Acquisition and/or Development
Delineating sub-surface drainage using satellite imager Department of the Interior  USGS Mission-Enabling Knowing subsurface drainage (tile-drain) extent is integral to understanding how landscapes respond to precipitation events and subsequent days of drying, as well as how soil characteristics and land management influence stream response. Consequently, a time series of tile-drain extent would inform one aspect of land management that complicates our ability to explain streamflow and water-quality as a function of climate variability or conservation management. Operation and Maintenance
Vegetation mapping on the Hawaiian island of Lanai Department of the Interior  USGS Mission-Enabling Creating high-resolution vegetation mapping approach that combines satellite imagery, machine learning, and expert knowledge to accurately classify plant species across the Hawaiian island of Lanai, producing detailed maps that can support conservation planning and monitoring of both native and invasive species. Implementation and Assessment
Machine Learning for automatic fracture mapping and rock identification Department of the Interior  USGS Mission-Enabling Machine learning algorithms are being used to improve detection and characterization of faulting after major surface rupturing earthquakes and identify and collect imagery of fragile geologic features with application to assessment of earthquake hazards. Acquisition and/or Development
Reinforcement Learning for Helmholtz Coil Operation and Simulation Department of the Interior  USGS Mission-Enabling The USGS is using AI/ML to optimize performance of its magnetic observatories. For example, reinforcement learning (RL) can significantly aid in the operation of a Helmholtz coil by optimizing its performance in generating uniform magnetic fields. Acquisition and/or Development
Deep-learning Integration into NEIC Operations Department of the Interior  USGS Mission-Enabling The National Earthquake Information Center (NEIC) is improving its earthquake detection and processing systems by leveraging artificial intelligence and machine learning. Implementation and Assessment
Predictions of PFAS Concentrations in Groundwater Department of the Interior  USGS Mission-Enabling A model of PFAS occurrence in groundwater at the depths of public and private drinking water supplies across the CONUS has been created and is expected to receive national attention. The model predicts occurrence of PFAS (detect or non-detect) in 1x1 km grid cells, and leverages the best data available in 2023. By updating this model, the goals are to (1) provide concentration predictions, instead of occurrence predictions, and (2) improve model accuracy owing to a larger set of sample data to train the model with. Concentration estimates would allow for better leveraging of resources to areas with predicted significant impacts. Initiated
NGWOS External R&D - Using advanced computing techniques for mobile monitoring platforms Department of the Interior  USGS Mission-Enabling This project is designing a networking approach for underwater robotic sensor platforms so that they can work together without direct human intervention to adjust their monitoring operations to changing bathymetry and water conditions using artificial intelligence algorithms and underwater communications. Acquisition and/or Development
Population and critical habitat modeling of overwintering monarch butterflies Department of the Interior  USGS Mission-Enabling Monarch butterflies in the western United States overwinter at very specific locations across coastal California. As monarch population decline it become important to identify the characteristics of what makes an overwintering grove a suitable habitat. Understanding the land cover and climatic factors that influence site selection by monarch can aid land managers in both making decisions to support exisiting critical habitat, and identify previously unknown locations where monarchs overwinter Implementation and Assessment
Automating the Detection and Classification of Wildlife in Aerial Imagery Department of the Interior  USGS Mission-Enabling The US Geological Survey (USGS), Bureau of Ocean Energy Management (BOEM), and US Fish and Wildlife Service (FWS) are partnering on a multi-year effort to develop deep learning algorithms and tools for the detection and classification of seabirds and other marine wildlife in aerial imagery. The tools and workflows developed by this project will be used by BOEM to assess wildlife populations as part of planning and monitoring for offshore energy development. Implementation and Assessment
Machine Learning algorithm for stream velocity prediction Department of the Interior  USGS Mission-Enabling The algorithm will be used to develop and incorporate a time-of-travel web-based application that will allow users to estimate travel times in a spill response scenario with greater accuracy. Acquisition and/or Development
Machine Learning for streamflow forecasting Department of the Interior  USGS Mission-Enabling In the Pacific Northwest, where the Willamette River is located, floods are becoming more common and severe. To help prevent flooding, the area has 13 dams built to control floods. This project, a collaboration with Portland State University, aims to build a smart system to predict river flows accurately and quickly across the Willamette River Basin by using state-of-the-art statistical and computational models as well as new hydrologic observations. The project aims to improve our understanding of how to use new river flow observational data that USGS provides to make better predictions. Acquisition and/or Development
Automated otolith aging using image processing Department of the Interior  USGS Mission-Enabling Fisheries managers and researchers often need to know the age of fish for population estimates, stock assessment, and similar projects. Fish otoliths (an ear bone) often accumulated rings annual (similar to tress). This process traditionally is done manually and can vary across individual agers. We are training an image process ML program to automate this process to see if we can reduce variability across individual agers and automate the aging process of counting otolith rings possibly saving time. Implementation and Assessment
Automating blood smear cell counts using machine learning Department of the Interior  USGS Mission-Enabling We exposed largemouth bass to an immunogen (poly I:C) to stimulate an antiviral response and collected blood smears from 38 of these fish. The blood smear slides were scanned at 83x, digitized using an Aperio ScanScope CS, and uploaded for labeling using SageMaker Ground Truth software. We are working with USGS Cloud Hosting Solutions to label training sets of images that include lymphocytes, monocytes, and granulocytes. We are in the process of validating and testing the ability of the model to accurately recognize and distinguish WBC images. To accomplish this, we are comparing manual cell counts among the 3 human readers, manual cell counts to automated cell counts by the model, and automated cell counts of novel tiles taken from the training slides. Implementation and Assessment
Machine learning for tsunami source zones Department of the Interior  USGS Mission-Enabling State of the art tsunami hazard analysis for coastal communities and infrastructure is computationally demanding. Ml will be used to select the most representative source zones (among thousands of offshore earthquake ruptures) Implementation and Assessment
Shoreline Modeling Department of the Interior  USGS Mission-Enabling Exploring the prospect of using AI/ML models to predict shoreline change and compare accuracy to traditional models. Examples of AI/ML models include LSTM, CNN, and Transformers Implementation and Assessment
Determining the resource potential of critical minerals in seafloor massive sulfide deposits Department of the Interior  USGS Mission-Enabling ML to input data on seafloor massive sulfide geochemstry to predict composition. Implementation and Assessment
Oceanographic, coastal, and geomorphic change analysis: data generation, QC/QA, and data management Department of the Interior  USGS Mission-Enabling Machine learning to quantify coastal/marine change across broad scales. QC/QA processes in place to assess data robustness. Verified data will be used by USGS projects for forecasting trends (ie, shorelines, role of permafrost) in a variety of coastal/marine settings for US coasts. Implementation and Assessment
Synthesizing mapping and monitoring data to inform prairie dog management in National Parks Department of the Interior  USGS Mission-Enabling Synthesizing Mapping and Monitoring Data to Understand Fluctuations in Prairie Dog Colony Size and Densities in Theodore Roosevelt National Park. There have been efforts to capture variation in the size and extent of prairie dog colonies at Theodore Roosevelt National Park in some form since the 1940s. Prairie dog colonies have been mapped semi-annually since the 1990s, but shifting priorities and a largely static budget have made it difficult for park staff to continue mapping. Furthermore, little research has been conducted with the existing mapping data to assess prairie dog habitat quality and the factors that affect colony size fluctuations. This project is aimed at developing more cost-efficient methods for prairie dog colony mapping (that is, remote sensing techniques) and developing indices and models that can help managers derive population inferences based on colony area. Park staff need modeling and remote analysis tools they can use to assess variability in prairie dog populations within the park. These tools should include a method for remotely assessing prairie dog colony size and a predictive tool that can help park managers understand the relationship between colony size and prairie dog population size. Implementation and Assessment
Quantifying the effects of land-use change and bioenergy crop production on pollinators, wildlife, a Department of the Interior  USGS Mission-Enabling The annual migration of monarch butterflies in North America represents a biological phenomenon unique to our planet,
covering more than 4,000 kilometers and requiring multiple generations of monarchs to complete. The monarch was proposed for listing under the Endangered Species Act in 2014 because of significant population declines and extinction risk. Disappearance of milkweed, the essential host plant for monarch larvae, has been implicated in the decline of the eastern monarch population. The objective of this study is to test the effectiveness of using uncrewed aircraft systems (UAS) and artificial neural networks to quantify the density of common and showy milkweed in working grasslands of Minnesota and North Dakota. First, NPWRC scientists will develop a machine learning algorithm for detecting milkweeds from UAS-collected aerial images. Second, NPWRC will validate the algorithm by comparing plot-level counts of milkweed estimated from UAS images to field count data across a range of milkweed densities. Lastly, NPWRC will take steps towards facilitating the integration of this technology into the Integrated Monarch Monitoring Program by estimating the number of spatially independent UAS images required for achieving accurate and precise estimates of milkweed across entire fields. In fiscal year 2019, preliminary UAS flights were completed; however, this project was put on hold because of a DOI ruling that grounded all UAS flights. No USGS UAS flights will be completed until this ruling is changed.		 Implementation and Assessment
Computationally efficient emulation of spheroidal elastic deformation sources using machine learning Department of the Interior  USGS Mission-Enabling Elastic continuum mechanical models are widely used to compute deformations due to pressure changes in buried cavities, such as magma reservoirs. In general, analytical models are fast but can be inaccurate as they do not correctly satisfy boundary conditions for many geometries, while numerical models are slow and may require specialized expertise and software. To overcome these limitations, we trained supervised machine learning emulators (model surrogates) based on parallel partial Gaussian processes which predict the output of a finite element numerical model with high fidelity. Implementation and Assessment
Wildlife species recognition and distance from camera estimation Department of the Interior  USGS Mission-Enabling Reliable population estimates of animal density. Acquisition and/or Development
Automating blood smear cell counts using machine learning Department of the Interior  USGS Mission-Enabling We exposed largemouth bass to an immunogen (poly I:C) to stimulate an antiviral response and collected blood smears from 38 of these fish. The blood smear slides were scanned at 83x, digitized using an Aperio ScanScope CS, and uploaded for labeling using SageMaker Ground Truth software. We are working with USGS Cloud Hosting Solutions to label training sets of images that include lymphocytes, monocytes, and granulocytes. We are in the process of validating and testing the ability of the model to accurately recognize and distinguish WBC images. To accomplish this, we are comparing manual cell counts among the 3 human readers, manual cell counts to automated cell counts by the model, and automated cell counts of novel tiles taken from the training slides. Initiated
Machine Learning to evaluate water quality Department of the Interior  USGS Mission-Enabling Examining the effect of physicochemical and meteorological variables on water quality indicators of harmful algal blooms in a shallow hypereutrophic lake using machine learning techniques. Implementation and Assessment
Ecological niche models for bat species Department of the Interior  USGS Mission-Enabling We are trying to understand what environmental factors determine the presence and absence of bat species across their range. Implementation and Assessment
HotLINK: The volcanic hotspot learning and identification network Department of the Interior  USGS Mission-Enabling An increase in volcanic thermal emissions can indicate subsurface and surface processes that precede, or coincide with, volcanic eruptions. Operation and Maintenance
Development of a Strategic Framework for Use and Implementation of Machine Learning in Energy Resour Department of the Interior  USGS Mission-Enabling The overarching objective for this project is development of a strategic framework for integrating ERP science with traditional information technology related platforms. The proposed framework includes (1) adoption of ML pipelines/models in ERP project workflows, tailored to specific needs of ERP project scientists; (2) modernization of key ERP data assets through API extension to meet the needs for data accessibility in big data platforms, analytics libraries, and machine learning models; and (3) technology transfer, targeted training, and multi-disciplinary career development for existing geospatial ERP workforce. Adoption of the proposed strategic framework will help us achieve this 21st century science vision and position the ERP to more effectively deliver its unique data-driven information products. Acquisition and/or Development
Quantifying Watershed Controls on Fine Sediment Flux to Lake Tahoe, California/Nevada Department of the Interior  USGS Mission-Enabling The variability in precipitation state impacts the fine sediment (FS, < 16 um) flux from upland areas to Lake Tahoe influencing lake clarity. We used supervised random forest regression models to estimate watershed parameters of importance that drive sediment flux. Implementation and Assessment
Seismology of Magmatic Injection Department of the Interior  USGS Mission-Enabling USGS staff are working with a student from Baylor University to use machine learning and network covariance to o understand the nature and dynamics of seismic sources associated with magmatic injection and magmatic transport. This information is necessary to understand volcanic systems. Seismic investigations are also being done for magma plumbing. Implementation and Assessment
Earthquake Catalog Development Department of the Interior  USGS Mission-Enabling Using AI/ML to develop more complete and robust earthquake catalogs, including focal mechanisms. This includes volcanic earthquake catalog enhancement using integrated detection, matched-filtering, and relocation tools. Implementation and Assessment
Seedling Identification and Percent Growth Analysis Department of the Interior  USGS Mission-Enabling his project aims to automate the extraction of alphanumeric labels and analyze seedling growth in petri dish images. Labels identifying seed type, treatment, and replication are extracted using Optical Character Recognition (OCR), saving time and reducing human error. Additionally, k-means clustering is applied to segment seedlings from the background, enabling quantification of percent growth over time. The process addresses challenges such as varied image orientations, lighting condidtions, and label placements. By automating label extraction and seedling measurement, the workflow accelerates data analysis, improves accuracy and supports scalable environmental and toxicological research. Implementation and Assessment
Gulf Coast Geologic Energy Machine Learning Department of the Interior  USGS Mission-Enabling Machine learning has been used to predict expected ultimate recovery of shale oil wells in a previously studied assessment unit. These predictions use multi-layer perceptron based artificial neural networks along with decline-curve based estimated ultimate recoveries and a feature database of geological, reservoir, and well completion parameters. Researchers are developing a ML model using elemental data to predict total organic carbon. Implementation and Assessment
Flow Photo Explorer to estimate flow Department of the Interior  USGS Mission-Enabling Full USGS streamgages can be expensive for cooperators, especially on internment streams, so a lower cost, low maintenance method for determining if a stream is flowing and relative flow volumes is needed. Implementation and Assessment
Predicting Sparse (Geothermal) Resources Availability by using Machine Learning Department of the Interior  USGS Mission-Enabling This research is developing the machine learning (ML) tools (a subset of artificial intelligence) to predict the availability of sparse natural resources (e.g., geothermal, minerals) at regional levels by providing careful consideration to mathematical and geostatistical practices that adhere to geoscience processes. The associated challenges include developing new ML metrics for evaluating model performance that work with sparse natural resources, addressing the extreme mathematical sparsity of these resources at the regional scale, and engineering new evidence layers to inform modeling workflows. The goals of this work include increasing the explainability, reproducibility, and accessibility of the assessment modeling process. Acquisition and/or Development
National Wildlife Disease Database Department of the Interior  USGS Mission-Enabling The National Wildlife Health Center (NWHC) has contracted the Pacific Northwest National Laboratory (PNNL) to build a national wildlife disease database (NWDD).  Funded through the American Rescue Plan Act of 2021 (ARPA), the NWDD will bring together various wildlife health data streams across informational domains (i.e., laboratory results, environmental observations, news media, etc.) to provide situational awareness and advanced analytics to natural resource authorities around the country.  As part of the NWDD, PNNL plans to integrate their Canvas software.   Canvas is a machine learning, AI, and data science tool that can visualize and contextualize information from one or more sources. Acquisition and/or Development
Landform Mapping GeoAI Department of the Interior  USGS Mission-Enabling NASA is helping the USGS is developing a faster, smarter process for creating a national landform dataset using machine learning and active learning techniques. These tools help prioritize which data experts should label, saving time and improving accuracy, especially for rare or complex features. Since the data involves 3D shapes, the project is exploring advanced methods like topological analysis and deep learning to better analyze the data. A basic user interface will also be developed to make labeling easier and lay the foundation for future tools like natural language queries. Initiated
Using Machine Learning Methods for Automatic Discovery and Catalog of North Dakota Stock Ponds and Other Impoundments Department of the Interior  USGS Mission-Enabling The purpose of this project is to address the challenges in cataloging stock ponds and other impoundments by developing a machine learning (ML) workflow that processes satellite imagery for the extraction of their location and type. This workflow will be specifically designed to distinguish between multiple types of stock ponds, stock dams and dugouts, and other impoundments. The second goal is to utilize this ML workflow to build a new, comprehensive catalog of stock ponds and other impoundments. This catalog will be unique in its inclusion of smaller stock ponds, which are often omitted in existing databases, thereby offering a more accurate and complete understanding of stock pond and impoundment distribution and types. Initiated
Improved point cloud classification of 3DEP lidar data using Deep Learning models Department of the Interior  USGS Mission-Enabling Point clouds are remote sensing data that is represented by points in space that can represent earths surface. The class of the point can identify the surface as belonging to buildings, trees, or roads for example. This study is testing methods for enhancing 3DEP lidar point data classification to include more surface types at a finer resolution. The tests are using deep learning (DL) models to refine and enrich the 3DEP data classification. Initiated
Knowledge Graph development Department of the Interior  USGS Mission-Enabling Research and development of knowledge graph designs to make information more interoperable based on logical relationships automatically identified from text. The research work includes using AI for natural language processing. Initiated
Predicting inundation dynamics of small forested wetlands Department of the Interior  USGS Mission-Enabling This project aims to help land managers in the Upper Midwest understand the wetting/drying dynamics of small wetlands relevant to amphibians. The approach will leverage field observations of wetland water levels as training data in a random forest model. Acquisition and/or Development
Hydrograhy feature extraction from remotely sensed data Department of the Interior  USGS Mission-Enabling Surface water features shown on maps, also known as hydrography, show the available water that can greatly influence the quality of life and the environment. Mapping hydrography is a complex task because the amount and distribution of surface water continuously varies with weather conditions. This research is testing machine learning techniques to develop models that predict the location of surface water from remotely sensed elevation and image data. Machine learning techniques are methods that train a model based on existing hydrography features in training area. After training a model, it can be applied to other areas where hydrography data do not exist, and the model can potentially be applied to new remote sensing data to help update hydrography features over time. Initiated
Using machine learning to detect invasive bullfrogs Department of the Interior  USGS Mission-Enabling Detecting bullfrogs along their invasion front in order to inform removal efforts Implementation and Assessment
Deep Learning application for automated mapping of surficial landforms, surficial geological deposit Department of the Interior  USGS Mission-Enabling Florence Bascom Geoscience Center (FEDMAP-funded): The Bascom Geoscience Center (FEDMAP) is using the deep-learning capabilities implemented within the ESRI ArGIS Pro software platform to automate the process of mapping surficial landforms and related surficial geologic deposits from lidar-derived topography. We are also using this capability to train models that can identify abandoned mine sites from lidar-derived topography. Implementation and Assessment
Zero shot segmentation to expedite Quaternary geologic mapping Department of the Interior  USGS Mission-Enabling Geosciences and Environmental Change (GEC) Science Center (FEDMAP-funded): The construction of detailed geologic maps requires a lot of manual GIS data input to outline the extent of interpreted geologic features. We are working to extend and adapt the Segment Anything Model for identification of Quaternary geologic features in order to expedite the process of creating GIS data for geologic maps. Acquisition and/or Development
Oil Spil Response for Ice-Covered Rivers Department of the Interior  USGS Mission-Enabling This project uses backscatter data from the SAR (synthetic aperture radar instrument) on the Sentinel-1 satellite to determine ice phenology for large rivers and lakes based on machine learning method (i.e., a Random Forest Classifier). The goal of this DOI Inland Oil Spill Preparedness Program (IOSPP) funded work is to provide rapid, near real-time information to oil spill response crews concerning about the safety of ice-covered areas (FY2023-25). Operation and Maintenance
Modeling Sediment Abundance in the Eastern Snake River Plain Aquifer Using Supervised Learning Department of the Interior  USGS Mission-Enabling This project aims to model sediment abundance in the eastern Snake River Plain aquifer near the Idaho National Laboratory. We are employing supervised learning techniques to predict sediment presence at various depths within boreholes using natural gamma ray readings. The model is trained with independently derived estimates of sediment probability to enhance its accuracy. This initiative will significantly aid scientists in understanding and managing sediment distribution in the aquifer. Initiated
Pacific Northwest Stream Flow Permanence Department of the Interior  USGS Mission-Enabling We are using machine learning approaches, specifically the random forest algorithm, to provide spatially explicit estimates of the presence of year-round surface flow in streams across large (several US states) geographic extents. Empirical random forest models include discrete flow/no flow observations as the response variable and a broad suite of physio climatic covariates. The models are used to inform management decisions that require streamflow classification of perennial versus non-perennial which is the charge of many land steward agencies including the Bureau of Land Management, U.S. Forest Service and State and private forests. Funding for this project is congressional allocated funds (FY2023-25). Operation and Maintenance
SAMPLE Toolbox Department of the Interior  USGS Mission-Enabling A toolbox for land managers to develop plans for monitoring vegetation Acquisition and/or Development
Mapping wildfire fuels in previously burned landscapes Department of the Interior  USGS Mission-Enabling The goal is to understand how land management treatments affect the probability of reburning. Acquisition and/or Development
Inventorying landforms with convolutional neural networks Department of the Interior  USGS Mission-Enabling Geosciences and Environmental Change (GEC) Science Center (FEDMAP-funded): LiDAR derived topographic data images earth's surface in unprecedented detail and is available across most of the country. This data reveals landforms relevant for understanding many scientific phenomena (e.g., patterned ground) and natural hazards (e.g., karst). We are developing simple pipelines to train and deploy convolutional neural networks on high resolution topographic data in order to efficiently identify and inventory these features. Implementation and Assessment
Lava lake thermal pattern classification using self organizing maps and relationships to eruption pr Department of the Interior  USGS Mission-Enabling We apply a machine learning algorithm called self-organizing maps (SOM) to thermal infrared time-lapse images . Implementation and Assessment
Advancing image-based surveys to support sea duck conservation along the Pacific Flyway Department of the Interior  USGS Mission-Enabling For most of their annual cycle, North American sea ducks are densely distributed in estuaries and along the coastal nearshore where they are susceptible oil spills, energy development, changing ocean conditions, and other potential threats. Observer-based aerial surveys have been an important tool for evaluating coastal distributions and estimating population abundances to understand sea duck responses to their changing environment. However, safety, expense, observer bias and lack of methodological consistency are rising concerns associated with observer-based surveys, making it imperative to transition to more sustainable methods. Digital aerial surveys (DAS) that automate counts from aerial imagery using convolutional neural network (CNN) models are one way to improve survey safety and count accuracy. We are developing a standardized DAS for the lower Pacific Flyway to help maximize safety, while improving data consistency and model accuracy among important regions within the Flyway. Acquisition and/or Development
Probabilistic source classification of large tephra producing eruptions using supervised machine lea Department of the Interior  USGS Mission-Enabling USGS has produced a model that accurately and confidently identifies large tephra-producing eruption volcanic sources in the Alaska-Aleutian arc using only in situ geochemistry. The model is a voting ensemble classifier comprised of six conceptually different machine learning algorithms trained on proximal tephra deposits that have had their source positively identified. Eruptive products from Alaska's Aleutian Arc-Alaska Peninsula and Wrangell volcanic field were used as a test environment for 11 supervised classification algorithms, trained on nearly 2000 electron probe microanalysis measurements of glass major oxides, representing 10 volcanic sources. Implementation and Assessment
InSAR and other geodetic studies at Volcanoes Department of the Interior  USGS Mission-Enabling The USGS uses InSAR (Inferometric Synthetic Aperture Radar) to map ground deformation and track volcanic activity globally. Artificial Intelligence approaches are being used to recognize transient signals in combined InSAR and GPS data that may be indications of impending hazardous volcanic activity. Implementation and Assessment
Climate Futures for Lizards and Snakes in Western North America Department of the Interior  USGS Mission-Enabling Identifying new management challenges to reptiles based on shifting environmental conditions Operation and Maintenance
Predicting inundation dynamics of small forested wetlands Department of the Interior  USGS Mission-Enabling This project aims to help land managers in the Upper Midwest understand the wetting/drying dynamics of small wetlands relevant to amphibians. The approach will leverage field observations of wetland water levels as training data in a random forest model. Acquisition and/or Development
Intelligent National Map project Department of the Interior  USGS Mission-Enabling The Intelligent National Map is a vision by the U.S. Geological Survey (USGS) to make mapping smarter by using advanced technology like artificial intelligence (AI). This project aims to create maps that can update themselves automatically, detect changes in the environment, and provide better, faster information for decisions about land, water, and natural resources. Initiated
Machine-learning model to delineate sub-surface agricultural drainage from satellite imagery Department of the Interior  USGS Mission-Enabling We trained a UNet machine-learning model, a convolutional neural network designed to highlight objects of interest within an image, to delineate tile-drain networks in panchromatic satellite imagery without additional data on soils, topography, or historical tile-drain extent. This was done by training the model to match the accuracy of human experts manually tracing the surface representation of tile drains in satellite imagery. Our approach began with a library of images that were used to train and quantify the accuracy of the model, with model performance tested on imagery from two areas that were not used to train the model. Satellite imagery included acquisition dates from 2008 to 2020. Training imagery was from agricultural areas within the US Great Lakes basin. Validation imagery was from the upper Maumee River, tributary to western Lake Erie, and an Indiana, Ohio-River headwater tributary. Our analysis of the satellite imagery paired with meteorological and soil data found that during spring, a combination of relatively high solar radiation, intermediate soil-water content and bare fields enabled the best model performance. Each area of interest was heavily tile-drained, where better understanding the movement of water, nutrients, and sediment from fields to downstream water bodies is key to managing harmful algal blooms and hypoxia. The trained UNet model successfully identified tile drains visible in the validation imagery. https://doi.org/10.1002/jeq2.20493 Implementation and Assessment
Environmental streamflows in the United States: historical patterns and predictions Department of the Interior  USGS Mission-Enabling It is important that environmental streamflow assessments by water managers consider changes in climate, land use, and water management; this cannot be done effectively without understanding historical variability and changes in environmental streamflows. Estimates of environmental streamflows also are needed for ungaged streams and machine-learning methods are likely useful for this. We are analyzing historical change and variability at hundreds of streamflow gages across the United States for a suite of environmental streamflows and using machine-learning methods to estimate environmental streamflows for thousands of ungaged stream reaches. Acquisition and/or Development
PROSPER Department of the Interior  USGS Mission-Enabling Machine learning approach to estimating the annual probability of streamflow permanence at a sub-reach (10m) scale. Operation and Maintenance
[Un]supervised clustering of [non-]earthquake signals commonly recorded on regional seismic networks Department of the Interior  USGS Mission-Enabling Surficial mass movements (SMMs), such as landslides and rockfalls, have seismic signatures distinct from other routinely recorded seismic sources like earthquakes and explosions. However, overlaps between the characteristics of these signals can still make it difficult to discriminate between source types during operational seismic monitoring. This ambiguity motivates the development of automated techniques for seismic signal classification. Furthermore, seismic differentiation within the broad class of SMMs has additional scientific and rapid response value. Examination of SMM seismic waveforms highlights a particularly strong contrast between the signals produced by primarily vertical processes versus processes that have a non-negligible horizontal component such as landslides and avalanches. A machine learning (ML) classification scheme is being investigated for differentiating between seismic signals generated by falls versus slides. Additionally included are shallow earthquakes and blasts because these are most similar to SMM signals and are commonly recorded on regional seismic networks. These classes, therefore, are the most useful for automated classification. Test datasets derive from waveform picks in the Exotic Seismic Events Catalog, a diverse collection of non-earthquake seismogenic surface events, and the Advanced National Seismic System (ANSS) Comprehensive Earthquake Catalog. In development is a shallow, feature-based approach to classification using statistical metrics extracted from waveforms. Feature importance metrics provide insight into the ML method, which leverage both unsupervised techniques and supervised techniques. Initiated
Extracting robust, searchable data from narrative geologic descriptions Department of the Interior  USGS Mission-Enabling Geosciences and Environmental Change (GEC) Science Center (FEDMAP-funded): Geologic map units are often described in narrative text descriptions. These descriptions sometime contain useful data (e.g., about lithology, unit thickness), but it is hard for users to operationalize that information since is not codified in any standard way. We are using natural language processing and large language models to parse readily queryable information out of thousands of descriptions in a national database of geology. Implementation and Assessment
Classifying GPS data to understand flight behavior of birds. Department of the Interior  USGS Mission-Enabling This project will help understand under what circumstances eagles are more likely to collide with wind turbines. Implementation and Assessment
Whole-lake indexing of round goby abundances with photographic catch data Department of the Interior  USGS Mission-Enabling USGS is responsible for monitoring abundances of prey fish across the entirety of the Great Lakes. This project utilizes autonomous vehicles and artificial intelligence to quantify abundances of one of the most abundant prey fishes in the Great Lakes, an invasive species called Round Goby. Robotic surveys of Roung Goby are carried out across three of the five Great Lakes each year. The work also characterizes surface geology, which influence round goby abundances. Implementation and Assessment
Predicting PFAS in shallow soils in northern New England Department of the Interior  USGS Mission-Enabling This project leverages statewide shallow soil data and machine learning methods (boosted regression tree models) to predict PFAS in soils across Maine, New Hampshire, and Vermont. Implementation and Assessment
Improving accuracy and precision of sonar-based estimates of fish abundance Department of the Interior  USGS Mission-Enabling USGS uses sonar to monitoring prey fish populations to support fisheries management decision making for the eight US Great Lakes states, Fish and Wildlife Service, and numerous tribes. Sonar-based estimates of fish abundance are prone to inaccuracies that can limit their utility. New technologies are being strategically cultivated to improve the accuracy and precision of USGS's annual prey fish abundance estimates. Artificial intelligence is being used to accelerate data processing. Implementation and Assessment
Classifying GPS data to understand flight behavior of birds. Department of the Interior  USGS Mission-Enabling This project will help understand under what circumstances eagles are more likely to collide with wind turbines. Implementation and Assessment
Machine learning-based landscape feature classification using satellite and airborne imagery Department of the Interior  USGS Mission-Enabling We use existing machine learning classifiers on satellite and airborne imagery to enhance the accuracy of habitat and land cover classifications. Implementation and Assessment
Predicting PFAS occurrence in groundwater using machine learning Department of the Interior  USGS Mission-Enabling This project leverages USGS groundwater data and machine learning methods (boosted regression tree models) to predict PFAS occurrence in groundwater at the depths of drinking water supplies across the conterminous U.S. A paper with the results of this effort was published in early FY25. A second iteration of this project will aim to predict concentration ranges instead of only occurrence, and commenced at the beginning of FY25. Implementation and Assessment
Machine Learning Image Classification Department of the Interior  USGS Mission-Enabling The mission of the PLACE (Patterns in the Landscape - Analyses of Cause and Effect) project is to inform land managers, planners and researchers about historical and current changes to human and natural environments. PLACE focuses on floods, droughts, and fires, which are increasing in severity, extent and frequency around the globe. The PLACE project utilizes random forest machine learning to classify wetlands and soil moisture at large scales, and to quantify causal processes behind wildfire. Implementation and Assessment
Zero shot segmentation to expedite Quaternary geologic mapping Department of the Interior  USGS Mission-Enabling The construction of detailed geologic maps requires a lot of manual GIS data input to outline the extent of interpreted geologic features. We are working to extend and adapt the Segment Anything Model for identification of Quaternary geologic features in order to expedite the process of creating GIS data for geologic maps. Acquisition and/or Development
Inventorying landforms with convolutional neural networks Department of the Interior  USGS Mission-Enabling LiDAR derived topographic data images earth's surface in unprecedented detail and is available across most of the country. This data reveals landforms relevant for understanding many scientific phenomena (e.g., patterned ground) and natural hazards (e.g., karst). We are developing simple pipelines to train and deploy convolutional neural networks on high resolution topographic data in order to efficiently identify and inventory these features. Implementation and Assessment
Tracking wetlands and water movement across watersheds Department of the Interior  USGS Mission-Enabling Accurate prediction of flood and drought impacts requires understanding upstream surface water storage dynamics and storage capacity. We use machine learning algorithms to classify satellite imagery into open and vegetated water extent. We then use machine learning and deep learning algorithms to relate daily river discharge to meteorology and surface water storage dynamics. This work is funded by USEPA, Office of Research and Development. Acquisition and/or Development
Everglades-Flux, Digital Surveys Department of the Interior  USGS Mission-Enabling We are creating a program that can automatically process Normalized Difference Vegetation Index images and come up with a true value of live vegetation. It will also gap fill missing data via AI/ML programs. Acquisition and/or Development
Cooperative Agreement Oakridge National Lab-USGS Research:  Using Artificial Intelligence to Improve Department of the Interior  USGS Mission-Enabling The research objectives of this project will focus on developing workflow processes and tools that will form the framework for implementation of data management and improvements to data management lifecycles. This includes removing dependencies on personnel to perform manual tasks that enable discovery, access, and formatting of potentially desirable data to users, and instead automate the long-term management of those data to ensure that they are readily usable under various automation scenarios. 

Specific objectives of this project aligned with AI use cases include:
1.  Research and design an algorithm to characterize and visualize data contents in USGS data repositories and DOE ARM holdings by adapting AI and machine learning approaches.
2.  AI Tool development to help scientists generate standardized metadata (ISO/FGDC)
3.  Using AI principles, automate methods for performing FAIR (Findable, Accessible, Interoperable, and Reusable) data assessments using training data developed for USGS's State of
the Data Rubric		 Initiated
Improved earthquake detection for research studies Department of the Interior  USGS Mission-Enabling Deep learning methods are being used to improve detection of earthquakes to provide more complete, high-resolution catalogs that are used in research to better understand earthquake occurrence, rupture processes and seismic hazard. Acquisition and/or Development
Machine learning for tsunami source zones Department of the Interior  USGS Mission-Enabling State-of-the=art tsunami hazard analysis for coastal communities and infrastructure is computationally demanding. In order to select the most representative source zone among the thousands of possible offshore earthquake ruptures, unsupervised machine learning is needed.   Using this type of AI will more accurately determine representative earthquake ruptures from offshore tsunami wave heights than previous interpretive methods that are subject to significant uncertainty. The result will yield a transparent and consistent methodology for tsunami source selection. Acquisition and/or Development
Mendenhall postdoctoral fellow using machine learning Department of the Interior  USGS Mission-Enabling The Global Marine Minerals Project, with the Coastal and Marine Hazards and Resources Program, currently employs a Mendenhall postdoctoral fellow who is using a machine learning approach to predict the composition of seafloor massive sulfide deposits. The fellow proposed this work in their submission in 2023, which was accepted for funding. They are currently executing the work and preparing publicaitons. Implementation and Assessment
Critical Mineral Assessment with AI Support (CriticalMAAS) Department of the Interior  USGS Mission-Enabling An AI-assisted workflow could enable the USGS to accomplish its mission, produce high-quality derivative products from raw input data, and deliver timely assessments that reduce exploration risk and support decisions affecting the management of strategic domestic resources. Acquisition and/or Development
Shoreline modeling Department of the Interior  USGS Mission-Enabling We're trying to evaluate the prospect of using AI/ML methods/models (e.g., LSTM, CNN, Transformers) to predict shoreline evolution and compare their accuracy to traditional physics-based models. Implementation and Assessment
Cell Phone Application for Oil Spill Detection Department of the Interior  USGS Mission-Enabling We would like to develop a model that can be used to interpret cell phone images to predict oil in environmental samples. The tool can be rapidly deployed for use in the field by the oil spill responder community. Acquisition and/or Development
Wave runup and total water level observations from time series imagery at several sites with varying Department of the Interior  USGS Mission-Enabling Tool is used for separation (segmentation) of land and water in images. The resulting mask is used to calculate water levels. Tool will be used to compare to forecasted water levels and may be displayed on USGS webpages Implementation and Assessment
Development of a Ploidy Distinction Application: A Machine Learning Approach for Discriminating Trip Department of the Interior  USGS Mission-Enabling A machine learning approach is being researched as a way for fisheries personnel in the field to determine the ploidy of wild-caught invasive carp. The use of machine learning in biomedical imaging accumulates substantial volumes of data via digital images, applies computational tools for processing these data, and can combine images from various microscopes by enhancing and equalizing image resolution. Thus, a computational model was trained with selected brightfield microscopic images of whole blood smears made from known diploid and triploid Grass Carp. Digital images were generated by using two types of microscopes: (1) a laboratory-dedicated, high caliber scope, and (2) a portable microscope meant for use in the field. Initiated
Storm Induced Erosion Response Network Department of the Interior  USGS Mission-Enabling Provide the next generation of Total Water Level and Coastal Change Forecast Initiated
Coastal Ecosystem Prediction System Department of the Interior  USGS Mission-Enabling The USGS is partnering with NOAA National Centers for Coastal Ocean Science to co-develop an integrated, national-scale framework for projecting wetland vulnerability to sea level rise. Multi-model ensemble predictions will show a range of possible future conditions and inform the protection of coastal communities, economies and their natural resources. Initiated
National Wildlife Disease Database Department of the Interior  USGS Mission-Enabling The National Wildlife Health Center (NWHC) has contracted the Pacific Northwest National Laboratory (PNNL) to build a national wildlife disease database (NWDD).  Funded through the American Rescue Plan Act of 2021 (ARPA), the NWDD will bring together various wildlife health data streams across informational domains (i.e., laboratory results, environmental observations, news media, etc.) to provide situational awareness and advanced analytics to natural resource authorities around the country.  As part of the NWDD, PNNL plans to integrate their Canvas software.   Canvas is a machine learning, AI, and data science tool that can visualize and contextualize information from one or more sources. Acquisition and/or Development
Sediment Transport in Coastal Environments (funded by San Francisco Bay-Delta PES) Department of the Interior  USGS Mission-Enabling Machine learning based time-series imputation of oceanographic time-series Implementation and Assessment
Machine learning based shoreline time-series imputation, classification and forecasting (time-series Department of the Interior  USGS Mission-Enabling Supports basic data generation and QC/QA procedures, for large scale and short-term forecasting of shoreline trends Implementation and Assessment
National Oceanographic Partnership Program (NOPP) Department of the Interior  USGS Mission-Enabling Machine Learning based coastal sediments assessment and prediction Implementation and Assessment
RSCC and TCA projects. Department of the Interior  USGS Mission-Enabling Machine Learning (ML) methods for identifying, assessing, and quantifying coastal features and habitats and coastal change hazards. Implementation and Assessment
Machinine Learning based shoreline detection and sea ice dynamics using coastal cameras Department of the Interior  USGS Mission-Enabling Supports basic data generation and QC/QA procedures, for large scale and short-term forecasting of shoreline trends and sea-ice dynamics in coastal environments Acquisition and/or Development
Using Machine Learning in USGS StreamStats to make suspended sediment and bedload predictions Department of the Interior  USGS Mission-Enabling A tool for resource managers who need estimates of suspended sediment and bedload in Minnesota rivers without sampling data. Operation and Maintenance
Merbok Supplemental Department of the Interior  USGS Mission-Enabling Machine Learning based shoreline detection and mapping, automated data suitability analyses from satellite imagery Implementation and Assessment
Annual National Land Cover Database Deep Learning Department of the Interior  USGS Mission-Enabling Using geospatial artificial intelligence methods to map annual land cover and other land surface characteristics for the Nation. Operation and Maintenance
Event and sequence (life-stage, behavior, activity, movement modality) identification, segmentation, Department of the Interior  USGS Mission-Enabling Animal movement data reflect various activities from individuals, including the use and availability of habitats and resources necessary for survival, behaviors such as feeding, flying, or nesting, and interactions with neighbors, predators, and prey. These biological patterns and processes are not readily observable from remotely tracked individuals and require identification from complex data streams. Complete representation of desired activites and behaviors may require multiple individual tools to satisfy all objectives. Initiated
Data-driven approaches to filling missing time-series data within the San Francisco Bay-Delta Department of the Interior  USGS Mission-Enabling In measurements of natural systems, incomplete time-series data are the rule, not the exception. Environmental time-series data may suffer from gaps at a variety of time scales, significantly reducing the number of observations to understand phenomena, identify change, calibrate models, and predict future behavior. Data may be missing because of sensor degradation (e.g., biofouling, mechanical issues, power failure), failure to meet basic quality assurance checks, or resampling of paired observations to a common timestamp.

Decades of water-quality time-series data (e.g., turbidity, salinity, temperature) collected throughout the San Francisco Bay-Delta (SFBD) by a variety of agencies including USGS, CA DWR, USBR, USFWS, and USACE are used as a critical indicator of estuary health. These data are no exception to the rule and contain numerous gaps. While bad data points can be relatively straightforward to identify (e.g., as anomalously high or low values), they are challenging to replace and are often flagged and left blank, which can bias long-term observational records.

We will test and develop several (~5) proposed methods to fill gaps in water-quality time-series data and quantify the uncertainty in those estimates. The methods will range in complexity, from linear regression to machine learning and deep learning approaches (Lepot and others, 2017), and will characterize uncertainty in the filled data (Cox and others, 2003).  For this proposal, we will initially restrict our scope to turbidity data, because of the relatively high biofouling rate of the optical sensors typically used to measure turbidity, as well as its ecological significance. In addition, we are already actively implementing one of the methods to fill gaps in turbidity data collected in a south S.F. Bay salt-marsh tidal creek (Figure 1) as part of a project investigating temporal variability in sediment delivery to marshes funded by the S.F. Bay Regional Monitoring Program.		 Acquisition and/or Development
Seabird and Marine Mammal Surveys Near Potential Renewable Energy Sites Offshore Central and Souther Department of the Interior  USGS Mission-Enabling Using rapidly developing machine- learning (ML) techniques, the USGS WERC team is developing new methods to automate the detection and counts of seabirds and marine mammals from digital imagery. Implementation and Assessment
Data-driven approaches to filling missing time-series data within the San Francisco Bay-Delta Department of the Interior  USGS Mission-Enabling In measurements of natural systems, incomplete time-series data are the rule, not the exception. Environmental time-series data may suffer from gaps at a variety of time scales, significantly reducing the number of observations to understand phenomena, identify change, calibrate models, and predict future behavior. Data may be missing because of sensor degradation (e.g., biofouling, mechanical issues, power failure), failure to meet basic quality assurance checks, or resampling of paired observations to a common timestamp.

Decades of water-quality time-series data (e.g., turbidity, salinity, temperature) collected throughout the San Francisco Bay-Delta (SFBD) by a variety of agencies including USGS, CA DWR, USBR, USFWS, and USACE are used as a critical indicator of estuary health. These data are no exception to the rule and contain numerous gaps. While bad data points can be relatively straightforward to identify (e.g., as anomalously high or low values), they are challenging to replace and are often flagged and left blank, which can bias long-term observational records.

We will test and develop several (~5) proposed methods to fill gaps in water-quality time-series data and quantify the uncertainty in those estimates. The methods will range in complexity, from linear regression to machine learning and deep learning approaches (Lepot and others, 2017), and will characterize uncertainty in the filled data (Cox and others, 2003).  For this proposal, we will initially restrict our scope to turbidity data, because of the relatively high biofouling rate of the optical sensors typically used to measure turbidity, as well as its ecological significance. In addition, we are already actively implementing one of the methods to fill gaps in turbidity data collected in a south S.F. Bay salt-marsh tidal creek (Figure 1) as part of a project investigating temporal variability in sediment delivery to marshes funded by the S.F. Bay Regional Monitoring Program.		 Acquisition and/or Development
Image classification from images on robotic platforms Department of the Interior  USGS Mission-Enabling Tool identifies different terrains based on training on labeled images. Application developed for NASA Mars rovers, but applicable elsewhere. Implementation and Assessment
ChatGPT to write Python scripts for ArcGIS Pro Maps to be CVD-Friendly Department of the Interior  USGS Mission-Enabling I do not know how to code, but we are attempting to automate the process of changing colors of planetary geologic map units within ArcGIS pro so that they are all color-vision deficiency friendly. This process is very time-consuming when done by hand, so we're using ChatGPT to help write scripts about this. Acquisition and/or Development
Analytical Chemistry & Geologic Resources Identification from Laser Spectroscopy Department of the Interior  USGS Mission-Enabling Determining composition of geologic, mineral, and resource targets from laser-based spectroscopic platforms Implementation and Assessment
Analytical Chemistry & Geologic Resources Identification from X-rays Department of the Interior  USGS Mission-Enabling Determining chemistry and rock composition from X-ray-based analytical chemistry platforms Implementation and Assessment
Retrieval-Augmented Generation Using USGS Documentation Department of the Interior  USGS Mission-Enabling Generative AI chatbots continue to make inroads across our lives. This project seeks to test creating and using a RAG model on USGS documentation with the ultimate goal of creating an internal only chat bot that has knowledge of internal USGS documentation. Initiated
Foundation Models to Advance Earth Science Department of the Interior  USGS Mission-Enabling Advance the understanding of Earth's conditions and processes by developing and deploying generalist AI models (Foundation Models) trained on Earth Observations from field, suborbital, and orbital sensors. These models, along with the resulting insights, will empower scientists and land user managers to achieve more while also advancing the broader field of AI and machine learning science. Implementation and Assessment
Rangeland Condition Monitoring Assessment and Projection (RCMAP) Department of the Interior  USGS Mission-Enabling Rangelands, comprised of open grasslands and/or shrublands, occupy huge swathes of land in the US, typically where climate and/or soils are too harsh for either forest or agriculture. Rangeland ecosystems provide critical wildlife habitat (e.g., greater sage grouse, pronghorn, black-footed ferret), forage for livestock, carbon sequestration, provision of water resources, and recreational opportunities. At the same time, rangelands are vulnerable to climate change, fire, and anthropogenic disturbances. The arid-semiarid climate in most rangelands fluctuates widely, impacting livestock forage availability, wildlife habitat, and water resources. Many of these changes can be subtle or evolve over long time periods, responding to climate, anthropogenic, and disturbance driving forces. To address the need for long-term tracking of vegetation change, scientists from the USGS and Bureau of Land Management (BLM) developed the Rangeland Condition Monitoring Assessment and Projection (RCMAP) project. RCMAP provides maps of vegetation cover at yearly time-steps, a critical refence to advancing science in the BLM and assessing Landscape Health standards. RCMAP quantifies the percent cover of ten rangeland components (annual herbaceous, bare ground, herbaceous, litter, non-sagebrush shrub, perennial herbaceous, sagebrush, shrub, and tree cover and shrub height) at yearly time-steps across the western U.S. using field training data, Landsat imagery, and machine learning. Operation and Maintenance
Global Food Security-Support Analysis Data (GFSAD) Project Department of the Interior  USGS Mission-Enabling Climate variability and ballooning populations are putting unprecedented pressure on agricultural croplands and their water use, which are vital for ensuring global food and water security in the twenty-first century. In addition, the COVID-19 pandemic, military conflicts, and changing diets have added to looming global food insecurity. Therefore, there is a critical need to produce consistent and accurate global cropland products at fine spatial resolution (e.g., farm-scale, 30m or better), which are generated consistently, accurately, and routinely (e.g., every year).

Therefore, the overarching goal of this continuity NASA MEaSUREs (Making Earth System Data Records for Use in Research Environments) proposal is to develop a comprehensive global food security-support analysis data (GFSAD) project that will produce multiple cropland models, maps, and monitoring tools leading to a wide array of products using machine learning algorithms (MLAs), Artificial Intelligence (AI), and satellite sensor big-data analytics through cloud-computing. The GFSAD project, will produce four distinct Landsat-derived global cropland products. These products are:
1. Global Cropland Extent Product @ 30m (LGCEP30-2020, LGCEP30-2025).
2. Global Rainfed and Irrigated Product @ 30m (LGRIP30-2020, LGRIP30-2025).
3. Global Cropping Intensity Product @ 30m (LGCIP30-2020 & LGRIP30-2025) &
4. Global Crop Type Product @ 30m for USA, Canada, and India (LGCTY30-2020USACAN, LGCTY30-2025USACAN; LGCTY30-2020India, LGCTY30-2025India).

The GFSAD Data is released through NASA's LP DAAC (links below):
https://lpdaac.usgs.gov/news/release-of-gfsad-30-meter-cropland-extent-products/
https://lpdaac.usgs.gov/news/release-of-lgrip30-data-product/
https://lpdaac.usgs.gov/products/gfsad1kcdv001/
https://lpdaac.usgs.gov/products/gfsad1kcmv001/

GFSAD Data can be visualized at:
https://www.usgs.gov/apps/croplands/app/map

GFSAD website for further information:
www.usgs.gov/wgsc/gfsad30		 Acquisition and/or Development
Extracting analysis ready information from narrative geologic descriptions Department of the Interior  USGS Mission-Enabling I work in the geologic mapping group in the USGS on a new national synthesis of geologic maps.  The different rocks depicted on geologic maps are typically described in narratives of 1-2 paragraphs, but often contain sparse details users would like to be able to search for/operate on. For example, how thick a unit is, analytical measurements associated with that unit, the presence of certain of-interest materials.

From some experimenting with the web interface it seems the doi-hosted chatgpt does a great job extracting this sort of information with relatively simple prompts. We were hoping to use this programmatically to more systematically parse information in our geologic map synthesis database. At present, this is ~10k descriptions that consist of on average ~250 characters of text, and I would seek to run a few prompt versions on each unit (plus some testing) - so perhaps a few 10s of thousands of requests. This would then be evaluated for accuracy against a subset of human-derived extractions.		 Initiated
21st Century Prospecting: AI-assisted Surveying of Critical Mineral Potential Department of the Interior  USGS Mission-Enabling Based on the Congressional mandate to assess critical minerals distributions in the US, the USGS Mineral Resources Program has partnered with DARPA. The objective of this partnership is to accelerate advances in science for understanding critical minerals, assessing unknown resources, and increase mineral security for the Nation.  As part of an extensive suite of tools, large language models are being used to extract information from text in scanned paper geologic maps as well as mining reports (mostly government documents).

(This is a re-submission as part of the new use case process)		 Acquisition and/or Development
Articulate Training Develpment AI Add On Department of the Interior  OS - Office of the Secretary of the Interior Mission-Enabling The Office of Emergency Management's Interior Operations Center uses Articulate to create online training courses for OEM staff. Articulate has introduced an AI add on that will expedite course creation and editing. The AI is a realistic text to speech solution but in the future will also assist with course content creation. Initiated
Machine Learning Applied to Geotechnical Engineering: Statistical Methods Applied to Seismic Analysis 1 Department of the Interior  BOR Mission-Enabling Purpose is to create machine learning models that predict soil liquefaction during a seismic event and compare to the geotechnical case history database to gain confidence in the machine learning estimates. The expected benefits are the existence of another tool for geotechnical staff to predict the probability of liquefaction and reducing the amount of more complex approaches to be conducted. Implementation and Assessment
Machine Learning Applied to Geotechnical Engineering: Statistical Methods Applied to Seismic Analysis 1 Department of the Interior  BOR Mission-Enabling Purpose is to create machine learning models to predict crest movement of an embankment dam during a seismic event. The expected benefits are the creation of a screening-level tool to be used prior to large investments in field exploration, lab testing, and large scale modeling efforts. Acquisition and/or Development
Machine Learning Working Group Department of the Interior  BOR Mission-Enabling Description: Perform networking and reach out to Reclamation employees to gauge staff experience and interest in creating machine learning models.  Also document potential applications of machine learning to various problems encountered by staff at Reclamation in their daily work.   
How it helps: This gives us a better idea of the in-house capabilities of staff at Reclamation and of the potential uses that machine learning can play to help solve problems faced by Reclamation.     
Stage of development: Ongoing discussion and information gathering with staff is occurring. 		 Acquisition and/or Development
Machine Learning for Chemical Savings at Reverse Osmosis Plants Department of the Interior  BOR Mission-Enabling Chemical use for cleaning membranes in reverse osmosis (RO) water treatment facilities can  contribute significant costs to the treatment process. Reclamation’s Desalination and Water  Purification Research Program (DWPR) funded GHD to apply machine learning techniques to optimize Clean In Place chemical usage at RO plants. The techniques analyze water chemistry data to inform application of chemicals and result in less chemical per unit of water produced without sacrificing operational uptime or membrane life. The project was implemented at full scale operational facilities of the Water Replenishment District located in Southern California. The project demonstrated that machine learning can be a useful tool to optimize cleaning frequency of RO ystems. Implementation and Assessment
Machine Learning Refines Quagga Habitat Suitability Department of the Interior  BOR Mission-Enabling Extensive research has been conducted to prevent the spread of quagga mussels, a costly and damaging invasive species. Notably, several detected introductions failed to establish populations in hypothetically suitable waterbodies by pH and calcium levels. To better parameterize quagga habitat suitability, we collected ecological data at 20 stations across four invaded and two uninvaded, connected waterbodies in Arizona during 2021-2023. Data were analyzed by gradient boosted machine, an ensemble machine learning algorithm that aggregates iteratively optimized decision trees. Results identified water conditions and plankton taxa that further differentiated quagga-invaded from uninvaded stations, within a system of pH- and calcium-suitable waterbodies. Implementation and Assessment
Generative AI to Improve Visitor Experience on NPS.gov Department of the Interior  NPS Mission-Enabling This is a proof of concept to us generative AI to extract data from NPS.gov and the NPS API to bring forth relevant content to visitors based on topics of interest allowing for improved trip planning.   This proof of concept enriches structured data without requiring parks to create new content and reduced the time and labor cost of re-creating content.  Initiated
Autonomous Drone Inspections Department of the Interior  BSEE Science & Space BSEE has entered into an Interagency Agreement with the Massachusetts Institute of Technology (MIT) Lincoln Laboratory (LL) to research autonomous unmanned aerial systems (UAS) and sensor technologies, with the goal of extending BSEE’s inspection capabilities to previously inaccessible environments on the Outer Continental Shelf (OSC). This research aims to build upon and leverage the work conducted by NASA's Advanced Supercomputer Division on corrosion models developed for the Level 1 Survey Report's Corrosion Level Classifications. Acquisition and/or Development
Well Activity Report Classification Department of the Interior  BSEE Science & Space Researching the use of Masked Language Models and Convolutional deep neural networks to identify classification systems for significant well event using data from Well Activity Reports.  Acquisition and/or Development
Level 1 Survey Report Corrosion Level Classification Department of the Interior  BSEE Science & Space Offshore operators conduct Level 1 surveys annually to report on platform structural integrity, as mandated by 30 CFR 250.901(a)(7), and submit these surveys to BSEE. Each survey includes a corrosion assessment of the platform with accompanying photos. Each area is assigned a coating grade and are key indicators of a platforms overall structural health. Currently, BSEE manually reviews each report to determine if a platform requires further audits, a process that is both time and labor intensive. To expedite this, BSEE has been collaborating with NASA's Advanced Supercomputer Division to develop a machine learning model designed to flag potential mislabeling of corrosion within the survey reports. The current version of the model has shown promising accuracy in identifying sections with potential misratings, especially in areas with the most severe corrosion. BSEE and NASA will continue to focus on further improving the model's accuracy while also exploring ways to integrate it into BSEE's IT infrastructure and operational workflows for potential future implementation. Acquisition and/or Development
Well Risk Assessment Department of the Interior  BSEE Science & Space NASA's Advanced Supercomputer Division will utilize the work performed in the sustained casing pressure research to explore the development of machine learning models to identify various precursors of risk factors for wells. By identifying these risk factors it would help inform BSEE engineers of potential problems with the well during its various stages of development. Currently, BSEE and NASA are focusing on identifying the precursors to sustained casing pressure and evaluating their contribution to the overall risk of a well. Acquisition and/or Development
Sustained Casing Pressure Identification Department of the Interior  BSEE Science & Space Well casing pressure requests are submitted to BSEE to determine whether a well platform is experiencing a sustained casing pressure (SCP) problem. SCP is usually caused by gas migration from a high-pressured subsurface formation through the leaking cement sheath in one of the wells casing annuli, but SCP can also be caused by defects in tube connections, downhole accessories, or seals. Because SCP can lead to major safety issues, quickly identifying wells with SCP could greatly mitigate accidents on the well platforms. BSEE entered into an Inter-Agency Agreement with NASA's Advanced Supercomputing Division to help research the use of various AI techniques. Acquisition and/or Development
ROV Smart Touch Subsea Pipeline Inspections Department of the Interior  BSEE Science & Space The BSEE funded project with the University of Houston is researching the development a state-of-the-art robotic system, "Smart Touch," to enhance subsea pipeline inspections by integrating advanced robotics and machine learning technologies. The system uses Remote Operated Vehicles (ROVs) equipped with stress wave-based smart touch sensors, video cameras, and scanning sonars to detect flange loosening and pipeline leaks autonomously. By leveraging machine learning for active sensing and robotic navigation, the project aims to improve inspection efficiency, reduce reliance on human operators, and mitigate risks associated with subsea operations. Collaborations with industry will support development, testing, and potential commercialization. Acquisition and/or Development

Was this page helpful?

Please provide a comment