Title: Advancing Water Quality Monitoring and Prediction Capability of USGS NGWOS Program with Satellite and Drone Remote Sensing Technologies
 
This project is part of the multi-institutional project:
Advancing Water Quality Monitoring and Prediction Capability of USGS NGWOS Program with Satellite and Drone Remote Sensing Technologies 
 
Project Lead: Andrew Schroth

^{University of Vermont Research Plan: Researchers at UVM will collect water samples at least 6 times per year under different seasonal and flow conditions in consultation with the University of Alabama team at predetermined locations. Those predetermined locations include Lake Champlain-Missisquoi Bay, Lake Champlain-St. Albans Bay, Lake Champlain (Main Lake), Lake Champlain-South Lake, Lake Carmi, the Missisquoi River (Swanton, VT), Winooski River (Essex, VT), CT River (West Lebanon, NH). All river locations are co-located with USGS monitoring infrastructure and lake sampling locations are co-located with Vermont Department of Environmental Conservations Long-Term Monitoring Program. Three of the lake sampling stations (Lake Carmi, Missisquoi Bay, Saint Albans Bay) will also be co-located with UVM’s YSI monitoring platforms that will be collecting high-frequency YSI EXO2 sonde profiles from the surface to a half meter from the bottom from roughly May through October. Surface profile measurements will provide high-frequency time series to further inform relationships between sonde turbidity, CDOM, and Chl-A/Phycocyanin fluorescence relative to satellite data. At each water sampling event and location, triplicate manual YSI EXO2 sonde measurements will also be collected that include turbidity, CDOM, and ChlA fluorescence. All water samples will be taken back to the University of Vermont and analyzed in PI Schroth’s laboratory for suspended sediment and DOC concentration (Shimadzu) by technician Blocher, as well as ChlA concentration and phytoplankton biomass and community composition in Co-PI Morales-Williams laboratory by the GRA and Co-PI. UVM researchers will participate in quarterly (or more frequently as needed) remote meetings with the University of Alabama research team and contribute to data analysis and publication generation. In Project Year 1, Co-PI Morales-Williams will host team members from the University of Alabama in her lab to establish comparable phytoplankton methodologies across the research team.} _{https://water.w3.uvm.edu/scripts/bib/journal_article/query2.php?id=8}

Title: Developing a Snow Modeling Foundation for Underrepresented Cold Regions in the Northeastern US
 
This project is part of the multi-institutional project:
Advancing CONUS-scale Operational Snow Modeling Capabilities
 
Project Lead(s): Dr. Katherine Hale and Dr. Beverley Wemple

^{University of Vermont Research Plan: UVM will be primarily responsible for compiling the required forcing and model validation data in historically underrepresented snow-affected regions of the Northeastern USA to better inform the NextGen NWM and thus provide more accurate forecasting of snowpack water resources across the region. UVM will work with the U of U team to generate model inputs for a process-based snow model, iSnobal, across identified testbeds. Representation of the Northeastern snowpack will be determined by comparing model outputs to a newly developed observation network in Vermont (including meteorological stations, snow scales, and UAV flights), funded through the Cold Regions Research and Engineering Lab (CRREL). We will leverage this new snow observatory network in Vermont to test iSnobal. We will focus on model performance across a range of snow accumulation and snowmelt conditions, including rain-on-snow events. The Ranch Brook watershed (9.84 km2, USGS 04288230), a forested alpine catchment on the eastern slopes of Vermont’s highest peak, will serve as our regional testbed, as it contains 13 meteorological stations which also monitor select snowpack characteristics (e.g., depth, melt, SWE), arrayed along elevational and aspect gradients. This testbed offers opportunities to extend this project and link our snow modeling advances to streamflow predictions in the NextGen framework. UVM will then develop a corresponding work plan for integration of a process-based snow modeling approach into NextGen, by identifying (1) forcing data needs to operationalize iSnobal for the Northeast, where snow observations are more limited (than the western US), (2) identify computational barriers to operationalization, and (3) build the BMI architecture for integration of iSnobal into NextGen for our testbed setting. 
 
As an associated activity to this core research, UVM will lead the development of a concept paper for an eventual CIROH Summer Institute on cold regions processes. The concept paper will identify qualified instructors, modeling challenges, and testbed sites. We envision building relationships with other CIROH-supported projects on cold-regions processes to develop this concept paper and identify key resources.} _{https://water.w3.uvm.edu/scripts/bib/journal_article/query2.php?id=9}

 
Title: Audience Segmentation to Improve Flood Inundation Mapping: Engagement and Testing with Technical Users and Impacted Communities
 
This project is part of the multi-institutional project:
Audience Segmentation to Improve Flood Inundation Mapping: Engagement and Testing with Technical Users and Impacted Communities 
 
Project Lead:  Dr. Anne Jefferson  
 
^{University of Vermont Research Plan: Despite the importance of flood forecasts to support event response and risk reduction actions, there has not been a synthesis of existing flood visualizations and testing of the prototype Flood Inundation Mapping (FIM) with different end-user groups. The goal of this project is to engage and test FIM approaches with technical users and impacted communities. The project will result in publications, presentations/briefings, and operational recommendations to improve the translation and use of FIMs. Intellectual merit: develops a novel, empirically based testing procedure to provide operational improvements that will enhance FIM forecast graphics interpretation and use in decisions. Broader Impacts: relationship building with underserved Tribes and accessibility and compliance consistent with Executive Orders (EOs), memorandum, and regulations.
 
Work to be Completed:  
 
Objective 1/: Indigenous Relationship Building and Communication Strategy: For visualizations available to Tribal Government decision makers, tribal emergency managers, and tribal community leaders, we will engage those users to assess whether they meet community expectations, such as integrating Traditional Ecological Knowledge and/or supporting the Federal trust responsibility to these communities. In Year 1, University of Vermont (UVM) will leverage the NOAA-funded Sea Grant network of resilience educators to identify where existing Sea Grant-tribal relationships have included dialogue and successful communication strategies related to inland/riverine flooding. In Year 2, UVM will contribute to the development of a strategy for flood communication to support NOAA FIM service equity priorities}.  
 
^{Objective 2: Interviews with FIM Producers and Power Users: We will facilitate semi-structured interviews through group meetings, in coordination with NOAA, with the FIMs producers and power users to better understand - from their perspective - the goals of the products, identify primary and secondary audiences of the products, and determine any design flexibility that needs to be considered as a result of planned or potential product modifications. In Year 1, UVM will contribute to participant recruitment for interviews with power users. UVM will also facilitate collaboration and information sharing with other CIROH-funded social science teams conducting focus groups, surveys, or interviews related to flood forecast communications/graphics.  
 
Objective 3: Synthesize and Diagnose Flood Visualizations: To understand the visualization and audience utilization of existing flood forecast products, we will conduct an analysis of existing literature and products to synthesize and diagnose communication challenges associated with wide-ranging approaches to flood visualizations and decision support tools.  In Year 1, UVM will play a leading role in the identification and synthesis of existing flood forecast products, drawing on the PI’s training and connections within the hydrologic science community. The flood resilience educator’s contributions will focus on flooding-related visualizations used by the Sea Grant network and its partners. UVM will contribute to the assessment and diagnosis of the existing visualizations, under the leadership of University of Minnesota (UMN). This work may continue into Year 2, in order to develop a journal article for peer-reviewed publication.  
 
Objective 4: Visualization Testing: Using a prioritized subset of diagnosed visualization problems from Objective 3, a multi-step co-production process will be used to test the efficacy of alternative prototypes and generate empirical results. In Year 1, UVM’s PI will participate in collaborative discussions with UMN and NOAA planning the testing. In Year 2, UVM’s PI will participate in discussions during and following testing. UVM’s participation in these discussions will enable greater contribution to Objective 5.  
 
Objective 5: Synthesizing results and developing best practices recommendations: Given the results of the interviews on user needs, synthesis and diagnosis of existing and new flood forecast visualizations, and potential redesign of the visualizations, we will synthesize the results into peer-reviewed papers, communications with NOAA, and recommendations for the improvement of outlook products.  In Year 1, UVM will contribute to synthesis of results from Objectives 1-3. In Year 2, UVM will contribute to synthesis of results and development of best practice recommendations from Objectives 1-4. UVM will contribute to the development of at least one article submitted for peer-reviewed publication and at least 1-2 public-facing communications of project results.} _{https://water.w3.uvm.edu/scripts/bib/journal_article/query2.php?id=10}

 
Title: Optimizing Flood Warning Information Sharing for Local Stakeholders through Science Communication Research
 
This project is part of the multi-institutional project:
Optimizing Flood Warning Information Sharing for Local Stakeholders through Science Communication Research 
 
Project Leads:  Dr. Anne Jefferson, Dr. Elizabeth Doran  
 
^{University of Vermont Research Plan: University of Vermont researchers will collaborate with researchers from RTI International to investigate local stakeholders (e.g., local officials, small business owners, homeowners, community services) as potential end-users for current and future National Water Model (NWM) and flood inundation map (FIM) forecast products. Through mixed-methods research, we will first gain a more complete understanding of existing flood information perceptions and uses across sectors and how they differ by community features (e.g., size, geography). We will then conduct workshops to explore alternative FIM data displays and information communications, gathering data to guide user-centric flood forecast design and dissemination strategies that foster trust and understanding, enabling the public to take effective action during flood events.  
 
Work In Progress: We are focusing on understanding how three types of community organizations use data to inform action, including exploring the role of flood forecast information displays (e.g., FIMS, depth modeling). Our primary audiences include local officials (e.g., utilities, public health departments), small businesses, homeowners, and community services (e.g., hospitals, shelters, food banks), and community interest organizations (e.g., large employers, faith-based organizations).  UVM researchers will conduct focus groups in two northeastern US communities who have experienced flooding in the past ten years to understand how communities use flood information for decision-making and action. Within each community, we will conduct three to four focus groups segmented by the above audiences as relevant and available. Participant recruitment will leverage local connections to emergency managers, health officials, and involved community representatives; engagement with individuals at conferences or relevant meetings; and snowball sampling. Comparing results from the northeastern US communities to focus groups conducted by RTI researchers in other parts of the US will allow UVM and RTI researchers to collaboratively understand how community use of flood information varies with geography and community size. We are supplementing the qualitative data collection with a survey that has wider reach to flood-prone communities around the U.S. to determine what processes and information uses are broadly applicable and which might be unique to a particular region, geography, level of resources, access to technology, or data literacy.  
 
In fall/winter 2024-2025, we will convene design-thinking sessions engaging the community in a dialogue centered around the question of “How are current flood information dissemination practices and communication products applied and how can processes and products be improved to promote understanding and action?” These sessions will include representatives from across audience segments, if feasible in the same communities where we conducted focus groups in Year 1. UVM researchers will continue to recruit and collect data in the northeastern United States, running at least one design-thinking session.  
 
UVM researchers will conduct data analysis and synthesis from focus groups, the survey, and the design-thinking workshops.  We will develop research products (e.g., manuscripts, white papers, conference presentations) for dissemination and will support communication of research findings to public outlets in the communities where we did our work.} _{https://water.w3.uvm.edu/scripts/bib/journal_article/query2.php?id=11}

Title: Modeling Community Trust: A Collaborative Approach to Scoping Water Forecasting Needs and NOAA Product Use in Indigenous Communities of Northeastern Oklahoma 
 
This project is part of the multi-institutional project:
Modeling Community Trust: A Collaborative Approach to Scoping Water Forecasting Needs and NOAA Product Use in Indigenous Communities of Northeast Oklahoma
 
Project Lead: Dr. Scott Merrill

^{University of Vermont Research Plan: This project will assist in the needs assessment, scoping and development of dynamic flood inundation and risk simulation models, and serious games to align with the needs of Indigenous communities in northeastern Oklahoma and beyond. Dr. Scott Merrill will serve as the PI for the UVM subaward. Dr. Merrill and Dr. Trisha Shrum (UVM) will work closely with PI Fedoroff, CUASHI researchers (Raub and Laufer), and the University of Kansas (Koliba) to assist with listening sessions, interviews, and empaneled focus groups with tribal leaders, tribal members, tribal hydrologists and planners, and NOAA NWS and RFC leads for the region. Additionally, Merrill will oversee the development of tools that will simulate flood hazard scenarios. Such tools will serve as visualization tools and discussion pieces to promote increased understanding of NOAA products and facilitate communication about the potential needs of the Tribal communities, as well as how to effectively understand and discuss flood hazards. Merrill and Shrum will lead efforts to explore how to effectively collect dynamic decision-making and behavioral data. Merrill will oversee an undergraduate student who will assist in the development of simulation models and serious games.}_{https://water.w3.uvm.edu/scripts/bib/journal_article/query2.php?id=12}

 
Title: Advancing Science to Better Characterize Drought and GroundwaterDriven Low-Flow Conditions in NOAA and USGS National-Scale Models
 
This project is part of the multi-institutional project:
Advancing Science to Better Characterize Drought and Groundwater Driven Low-Flow Conditions in NOAA & USGS National-Scale Models 
 
Project Lead: Dr. Donna Rizzo

^{University of Vermont Research Plan: We propose to develop state-of-the-art methods to improve national-level streamflow forecasts for low-flow conditions, where, in some regions, flows are dominated by baseflow from groundwater discharge rather than runoff from precipitation events. We will train machine learning algorithms using multiple observations including landcover/land-use, rainfall, soil moisture, and groundwater level datasets to identify reaches influenced by groundwater. Once these regions have been identified, we will develop methods to more accurately predict baseflow conditions in neighboring streams. We will refine and augment groundwater data with Earth observations to improve accuracy and compensate for sparse and missing groundwater data. We will leverage work being done with the USGS on Long-Short Term Memory (LSTM) models enhanced with feature selection. We will test Physics Informed Neural Networks (PINN) to ensure that our predictions are reasonable and so that they can be used reliably in sparsely gaged basins. We will develop physical models to calibrate and improve our machine learning algorithms. We will employ our resulting prediction methods to generate forcings and determine how to link these boundary conditions to the NextGen NWM using the HydroFabric framework. This work will improve the National Water Model (NWM) predictions for low-flow conditions, which to date have received less attention due to the emphasis on predicting extreme flooding events. These predictions will support operations where low-flow conditions are critical, including drought management, water supply, minimum flow rates for critical infrastructure, ecological sustainability, and river navigation. We will work with the USGS to ensure algorithms are compatible with computational frameworks/projects and that the USGS can use our results.  
 
Background: In a recent collaboration with U.S. Geological Survey (USGS), the University of Vermont (UVM) has been leveraging the USGS low flow ML modeling efforts being developed within their Water Resources Mission project – Data-Driven Drought Prediction project. Specifically, the USGS has developed and is currently testing deep learning models, known as Long Short-Term Memory (LSTM) models, to improve daily estimates of low streamflow and to forecast streamflow drought at lead times ranging from 0 days to 60 days. This UVM statement of work will leverage these on-going collaborative efforts to assess and improve LSTM model performance with a focus on forecasts for streamflow under drought conditions. In recent decades, the duration and deficit volume of streamflow droughts – defined as abnormally low streamflow and the resulting lack of water in the hydrological system (Van Loon, 2015) – have increased in the southern and western U.S. (Hammond et al., 2022). The proposed ML methods offer an approach to increase the accuracy of the NWM predictions for low-flows (i.e., streamflow drought forecasts) and expand the spatial coverage of these forecasts, which to date have received less attention due to the emphasis on predicting extreme flooding events. Under low-flow conditions, groundwater (GW) contributions to base flow become a critical forcing, and characterizing GW interactions with streamflow at a continental scale is critical.
 
Proposed Training/Testing Data: BYU has recently used ML tools that leverage Earth observational datasets to impute gaps in historical groundwater level records (S. Evans et al., 2020; S. W. Evans et al., 2020; Ramirez et al., 2022). The regional USGS LSTM models are being trained and tested using 40 years (1980-2020) of daily streamflow data from 425 streamgages within the Colorado River Basin and surrounding area. In addition to estimating low streamflows at the gages, now-casting of the latter are being assessed at ungaged locations. The LSTM input features include a large set of static watershed attributes available for the National Hydrography NHDPlus V2.1 catchments (Wieczorek et al., 2018) as well as meteorological and remotely sensed dynamic forcing inputs that have been aggregated to basin averages. Proposed Tasks: UVM will leverage the above big data and existing ML models and, in concert with BYU and UA, will work to improve low streamflow estimates as well as short- and medium-range streamflow drought forecasts. To date, UVM has been performing feature selection to rank the importance of the LSTM input features with respect to model performance accuracies. Both UVM and the USGS will present preliminary findings at the SEDHYD conference in St. Louis, MO in May, 2023. USGS has publicly released daily streamflow streamflow percentiles and drought event datasets for gages spanning all of CONUS (Simeone, 2022) and will be preparing a data release of compiled model input features, so that all data are publicly available. As a result, we expect these data will be available to our BYU-UA-UVM research team as early as late Spring 2023. LSTM model performance is currently being assessed using a variety of performance metrics.  
 
Because BYU has shown that aquifer storage curves correlate closely with long-term baseflow patterns observed in nearby streams and rivers, we hypothesize that LSTM model performance metrics will be correlated to the degree of groundwater-surface (GW-SW) water interactions. Thus, streamflows that correlate well with groundwater levels (e.g., perennial streams) may help improve forecasts of baseflow under low flow conditions. UVM proposes to expand the preliminary feature selection with an iterative clustering approach using new ML clustering tools to assess/improve LSTM model performance. Specifically, we will:  
 
1) Cluster USGS gaged watersheds based on the LSTM model performance metrics and then perform feature selection on a clustered watershed basis. In year one, we will use the upcoming 2023 USGS data release. In year two, after groundwater baseflow data have been compiled by BYU, we will repeat the cluster-feature selection analysis (i.e., re-cluster watersheds based on their degree of GW-SW water interactions and repeat feature selection).  
 
2) First perform a feature importance analysis on a watershed-by-watershed basis using a model input that varies dynamically (e.g., meteorology, degree of GW-SW water interaction, sensitivity to nearby groundwater levels), and then cluster the watersheds based on the importance/strength of the ranked input features. In this manner, we’d be investigating whether the LSTM models are able to capture SW-GW interactions.  
 
3) Tasks 1) and 2) above can be re-done on a seasonal basis to leverage, and perhaps identify, those times of the year when intermittent streams might provide added predictive value.  
 
4) Incorporate one (or more) GW-SW baseflow constraints into the loss/objective component of the USGS LSTM models. If time permits: (i) the USGS LSTM models could be re-trained using the input features selected in tasks above, and/or (ii) we might incorporate the BYU-UA groundwater baseflow data.} _{https://water.w3.uvm.edu/scripts/bib/journal_article/query2.php?id=13}

 
Title: Advancing Snow Observation Systems to Improve Operational Streamflow Prediction Capabilities   
 
This project is part of the multi-institutional project:
Advancing Snow Observation Systems to Improve Operational Streamflow Prediction Capabilities  
 
Project Lead: Dr. Christian Skalka 

Research Team: Rachael Chertok, Soheyl Faghir Hagh.

<sup>University of Vermont Research Plan:  At UVM our project has two main research goals: (1) to apply modern IoT technologies to improve NRT data reporting capabilities for distributed monitoring in remote settings, and (2) to develop an embedded machine learning-based algorithm using acoustic data for precipitation phase partitioning, to be deployed on low-cost remote platforms. All technologies are being developed on low-cost, low-power, open-source Arduino platforms. 

To accomplish goal (1) we are combining LoRA communications for LAN networking in remote settings with Satellite communications for WAN networking. This approach enables NRT reporting in essentially any remote setting since satellite communications are available anywhere on earth (unlike, e.g., cellular). And the use of LoRA enables efficient LAN communications even in light of resource constraints (e.g., power) in remote settings, supporting sensor distribution and better spatial resolution in data reporting. 

To accomplish goal (2) we are developing a novel machine learning algorithm that uses acoustic data from low-cost microphones for precipitation phase partitioning- that is, automated detection of rain, sleet, hail, or snow. Our algorithm will be embedded on devices in the field (aka “in the edge”). This is a critical enabling technology for our application, since storage or remote communication of high-bandwidth streaming datatypes such as a raw audio waveforms are ill-suited to low-powered embedded systems with typically constrained data logging capacity and/or low-bandwidth NRT reporting capability. Our method can be integrated into low-cost, low-powered embedded platform developed towards goal (1) with high temporal and spatial resolution. Our system will improve operational understanding of events such as rain-on-snow that are critical to monitoring and predicting snow and water dynamics.</sup> 
  _{https://water.w3.uvm.edu/scripts/bib/journal_article/query2.php?id=14}

Title: 3D Channel Properties for the OWP Hydrofabric
 
Principal Investigator: Dr. Rebecca Diehl 
 
Research Team: Stewart Kabis and David Baude
 
This project is part of the multi-institutional project:
Novel Geospatial Architecture of Channel and Floodplain Morphological Attributes within the OWP Hydrofabrics 

^{University of Vermont Research Plan: Realistic representation of channel morphology and hydraulic characteristics is critical to accurate hydrological and flood inundation predictions. Emerging datasets and analyses are transforming our ability to estimate these over the United States. Utilizing these advances within large-scale operational hydrological prediction frameworks will require a new paradigm in the geospatial representation of (3D) channel properties. This project is building on ongoing CIROH projects with partners at Utah State University, the University of Iowa, the University of South Carolina, and the University of Alabama, and expertise within and outside CIROH to contribute to a new geospatial framework directly linked to the OWP hydrofabric. The University of Vermont is focused on the attribution of hydraulically-relevant floodplain morphology that may inform improvements to streamflow predictions and flood inundation mapping outputs. Departures between a benchmark dataset of high-resolution topo-bathymetric surfaces, detailed stage-discharge rating curves, and calibrated inundation maps, and the OWP’s HAND-based synthetic rating curves and flood inundation maps, isolates limitations in the current representation of channel properties. The project’s outcomes are expected to transform the geospatial representation of river networks with direct and rapid translation to the NWC operational architecture and the broader hydrological community.} 
_{https://water.w3.uvm.edu/scripts/bib/journal_article/query2.php?id=15}