UM-CIP Program
Maryam Rishehri
This research investigates subsurface flow organization, solute transport, and residence-time dynamics in urban watersheds using an integrated modeling and observation framework. The project combines three-dimensional variably saturated flow simulations with the ParFlow-CLM hydrologic model, Lagrangian particle tracking using EcoSLIM, and field-based monitoring data to evaluate how storm events, land cover, and subsurface heterogeneity influence groundwater recharge and contaminant mobility in developed landscapes.
Because these simulations generate extremely large datasets, the fellowship supports the development of scalable high-performance computing workflows, including GPU-enabled simulations, parallel data processing, and distributed geospatial analysis pipelines. These computational advances will enable efficient processing of particle-trajectory datasets, execution of long transient simulations, and implementation of scenario testing across multiple urban watersheds, thereby supporting the core analytical goals of the dissertation.
Spring 2026 Presentation Slides
DS & AI Program
Evaluating ML Models for Predicting Global Deep Ocean Temperature and Salinity Profiles Trained on Deep Argo Data (Ellie Davidson)
This project evaluates and refines machine learning models, specifically deep neural networks and random forests, that have been pre-trained to predict deep ocean Temperature and Salinity (T/S) profiles below 2000 meters using upper ocean (0-2000m) data and other predictors. The evaluation is conducted by applying these models to a gridded Argo climatology, assessing the accuracy of the predicted deep-ocean profiles against established climatologies and reanalysis datasets. The project also involves working with large, multi-dimensional gridded ocean datasets and applying dimensionality reduction techniques such as principal component analysis (PCA) to support model evaluation and refinement.
Spring 2026 Presentation Slides
Identifying Linkages Between Sea Ice Retreat and Land Ice Melting Over the Antarctic: Machine Learning Approaches (Molly Balkan)
If land ice completely melts, the sea level will rise by 200ft. According to NASA, land ice has melted at a rate of 132 Gt/yr since 2000. The land ice is protected from thermal and mechanical forces by sea ice. Unfortunately, sea ice is retreating rapidly, reaching a record low (32% below average) in February of 2023, indicating a need to understand what’s causing this melting. Despite decades of research investigating this phenomenon, an in-depth understanding is lacking. This is primarily due to the limited interaction between oceanographers and glaciologists. This study bridges the knowledge gap between these two cryospheric communities by using graph-based spatial linkages. We use 2400 satellite images of sea ice extent and land ice depth data from 2000 to 2020 to develop a graph-based modeling framework that quantifies spatial heterogeneity in sea ice retreat and land ice melt. Our research discovers that there is a significant interconnection between sea ice retreat and land ice melting, especially over western Antarctica. Delaunay triangulation was used to validate points and establish spatiotemporal relations. To test the statistical significance of these linkages, Monte Carlo simulations have been used to avoid random detection.
Scalable, Secure and AI-ready Platform for Personal Informatics Project for Data Science Education (Chris Song)
This project focuses on the design and development of a web-based application to support students and instructors conducting Personal Informatics projects in the introductory data science course IS 296. Using a human-centered design approach, the system is designed to facilitate the collection, analysis, and reflection on personal data within the course environment. The application is engineered for scalable deployment beyond IS 296 and designed to support future integration of AI-powered capabilities that enable human–AI collaboration for personalized and adaptive learning experiences. The system also incorporates privacy and security protections appropriate for handling student-generated personal data.
Spring 2026 Presentation Slides
Understanding Students’ Learning Journey in Personal Informatics Project: Implications for Design (Emma Shay)
This project analyzes field data collected from the Personal Informatics project in the introductory data science course IS 296 to inform the human-centered design of an agentic AI system to support personal informatics to a broader audience. The data from Spring 2025 and Fall 2025, which includes students’ planning documents, end-of-project reflections (both qualitative and quantitative), students’ weekly reflection submissions, and creative artefacts, are used to conduct a qualitative analysis, contributing to developing a coding scheme and coding result.
Building data infrastructure and a pipeline to support a personal informatics project (Amreen Adams)
This project develops a data infrastructure to support IS 296 students’ Personal Informatics projects as part of the TEEM (Teaching to Empower) program. The infrastructure expands the types of data students can collect about themselves by providing Python libraries that connect to raw data from social media takeouts, wearable sensors, and UMBC student data. The project begins with the use of data from main social media platforms such as Instagram and YouTube, with the potential to expand to other data sources. The resulting libraries enable students to process raw data into structured, tabular formats suitable for analysis in introductory data science projects.
Spring 2026 Presentation Slides
Deepfake Legislation Policy Page (Henry Sonti)
This project analyzes and updates existing deepfake legislation, with reference to policies presented on the CISAAD project’s policy webpage. It focuses on building an international perspective, with particular attention to audio deepfakes, and compares legislative approaches across countries. The project also examines the effectiveness and scope of current policies, contributing to a clearer understanding of global strategies for regulating deepfake technology.
Spring 2026 Presentation Slides
Water Quality and Biodiversity in the Chesapeake Bay (Joseph Johnson)
This project is designed to visualize 9 years of high frequency environmental data collected in Baltimore Inner Harbor, and to relate it to a monthly dataset of aquatic invertebrate community succession at the same location between April and December each year. The goals of this data visualization are to explore correlations between environmental conditions and invertebrate animal community composition, and develop specific hypotheses that relate short and long term water quality with community make-up. The water quality data and species counts are loaded into Dataframes using the Pandas python library, then graphed using the Matplotlib library to better understand the correlations between low oxygen events and low species counts. A second goal is to investigate if interannual variation in community composition is responsive to preceding year conditions, especially in salinity. Future activity will include regressions or statistical models to test hypotheses.
Spring 2026 Presentation Slides
Spatial network of hypoxia events in Chesapeake Bay using Granger causality (Jay Gepilano)
Hypoxic events, creating low-oxygen “dead zones”, have continuously proven to be a lethal threat to a variety of species that reside within the Chesapeake Bay. Currently developed methodologies lack the ability to quantify the spatiotemporal propagation (i.e. the spread of dissolved oxygen across time and space) and effects of low-oxygen conditions to protect and preserve the Bay’s ecosystem. We showcase a new approach designed to evaluate the structure and topology of modeled hypoxia processes using the Spatial Network of Hypoxia Events (Hy-Net), which overcomes the difficulty of synthesizing complex spatiotemporal datasets for their use in regression and predictive models. By analyzing the spatiotemporal propagation of hypoxia through time series of oxygen concentrations, Hy-Net will aid in creating spatial networks via Granger causality techniques. These networks then provide forecasted dissolved oxygen values and their locations, allowing ecologists to create proactive treatment plans for the Chesapeake Bay through targeted, engineered oxygenation. Ultimately, this dynamic spatiotemporal network characterization of dissolved oxygen can be used as an input to predict and understand the distribution of the benthic community composition. This is a joint work with Ryan Langendorf, Jeremy M. Testa, Ryan Woodland, and Vyacheslav Lyubchich.
Spring 2026 Presentation Slides