When: Friday, October 31, 3:00 PM
Where: Tyler 055
Arup Mazumder: Structured Noise in AMSR-E SST Fields and Its Impact on Their Deconvolution
Abstract:
AMSR-E sea surface temperature (SST) fields are significantly oversampled, with a footprint of approximately 45×65 km but gridded at 10×10 km resolution. This oversampling suggests that, with the aid of additional high-resolution data—such as MODIS SST observations—deconvolution could potentially produce a true 10×10 km AMSR-E SST product. However, attempts to deconvolve AMSR-E fields using a U-Net trained on simulated SST fields (from LLC4320) failed, revealing a critical issue: the presence of structured noise. Further investigation shows, a distinct “lump” at wavelengths of 30–100 km, absent in MODIS-derived SST fields. This raised concerns about its impact on mesoscale SST analyses. Through spectral analysis of AMSR-E minus convolved MODIS SST fields, we confirmed that this noise persists across multiple overpasses, indicating a systematic, non-geophysical origin. These anomalies manifest as elongated features, approximately 1000 km in length and 100 km in width, within AMSR-E SST fields. Given AMSR-E’s 6.9 GHz channel susceptibility to interference and the structured, streaky nature of these anomalies, they are consistent with radio frequency interference (RFI). The 6.9 GHz band is vulnerable to emissions from geostationary and low Earth orbit (LEO) satellites, particularly near the Equator, making RFI a plausible explanation for these mesoscale patterns. Such contamination has serious implications for mesoscale oceanography, potentially distorting SST gradients and spectral analyses. This study characterizes these anomalies and examines their impact on SST retrievals, emphasizing the need for careful evaluation of AMSR-E data in mesoscale studies. Understanding these artifacts is essential for improving the interpretation of AMSR-E SST and assessing their reliability in oceanographic applications.
Bio:
Arup Mazumder is a Ph.D. student in Computer Science and Statistics at the University of Rhode Island, where his research lies at the intersection of artificial intelligence and ocean remote sensing. He earned his Master’s degree in Computer Science from Bauhaus University Weimar, Germany, where he worked on human–computer interaction and machine learning. His current work focuses on developing deep learning and transformer-based architectures to denoise and deconvolve satellite-derived sea surface temperature data. His broader research interests include generative modeling, structured noise characterization, and AI-driven environmental data reconstruction.
Priyankan Kirupaharan: Understanding the Accessibility of Single-User Virtual Reality Environments for Adults with Intellectual and Developmental Disabilities
Abstract:
In this paper, we aim to understand accessibility issues for people with intellectual and developmental disabilities (I/DD) with single-user VR applications. To this end, we recruited eight participants with I/DD for this study. We asked each participant to use a single-user VR application (on Meta Quest 2) and then conducted semi-structured interviews about their experiences. A subsequent thematic analysis of our interviews resulted in identifying several accessibility problems in using VR for people with I/DD. Overall, we found that participants had difficulty: perceiving (including comprehending) the various elements of the virtual environment and using physical controllers to engage with (i.e., act within) the virtual environment. The participants then suggested potential improvements to make the virtual environments more accessible. Based on these findings, we call for further research in four broad areas to foster an accessible VR experience for people with I/DD.
Bio:
Priyankan Kirupaharan is a PhD candidate working with the Accessible and Socially Aware Technologies (ASSET) Lab. My research focuses on human-computer interaction, accessibility, and immersive technologies, with the goal of understanding and addressing the needs of marginalized populations.
Yusra Suhail: Font Size Normalization – An Investigation of How Size is Perceived by Different Populations
Abstract:
User perception of font size has influenced how researchers, typographers, and
designers normalize fonts. To identify which font attributes contribute most to
readers’ perceptions of font size across age groups and language backgrounds, we
conduct a modified discrete choice experiment with crowdworkers. We develop
new calibration and measurement methods to convert the size of font attributes in
pixels to centimeters on a remote crowdworker’s screen. Results show that while
character height was the most influential attribute, age and language background
influence users’ perception differently. For participants with a Chinese language
background, character weight was most important. Character weight and character
spacing were more important for adults over 40 than for those under 40. Character width was particularly more important for participants with a language background featuring long descenders, such as Hindi and Arabic. Our results provide the first evidence on personalizing font size normalization across age groups and language backgrounds
Bio:
Yusra Suhail is a PhD student working with the Human-Centered Experiential Technologies (HAX) lab. Her research focuses on human-computer interaction, readability, and personalization. Her work integrates user research and design to build public systems that adapt to user needs and promote more accessible and intuitive web experiences. Yusra is also a part of the Machine Learning for Socio-technical Systems lab, where she uses her software engineering expertise to refine machine learning libraries and create tools and LLM agents that can be utilized by both large language models and human users.