Back to search

University of Minnesota

Towards Accessible and Reliable Polar Science Data: Infrastructure,Cleaning, and Imputation

Abstract

dc:description.abstract

Polar science increasingly relies on massive observational and environmental datasets, including reanalysis products, satellite imagery, and in-situ measurements. Although these datasets are publicly available in unprecedented volume, the ability to actually use them for scientific analysis is severely hindered by two fundamental challenges: data accessibility and data reliability. On the accessibility side, many archived datasets are simply too massive for interactive use, forcing scientists to wait hours or even days to download data and write substantial custom scripts before any meaningful analysis can begin. On the reliability side, these datasets often suffer from a range of data quality issues, including inaccuracies, inconsistencies, and, most pervasively, incompleteness. Missing values are especially common and can arise from sensor failures, battery depletion, cloud occlusion, and the harsh operating conditions of polar environments. Together, these challenges make the main bottleneck in polar science not merely the availability of data, but the accessibility and reliability of the data used for scientific discovery. The goal of this dissertation is to advance accessible and reliable polar science data through three complementary systems and techniques. For data accessibility, we present Polaris, an interactive and scalable data infrastructure for polar science. Polaris is designed around the workload characteristics of polar scientists, whose analyses are predominantly spatio-temporal, aggregate-heavy, and sensitive to data importance across space and time. By combining hierarchical spatio-temporal indexing with precomputed aggregates, Polaris enables highly interactive access to massive polar datasets and significantly improves their usability for scientific analysis. For data reliability, we present two contributions that address different facets of the data quality problem. First, we present Sparcle, a novel framework that injects spatial awareness into the core engine of rule-based data cleaning systems as a means of boosting their accuracy. Sparcle extends the notion of exact co-occurrence to spatial neighborhood and distance-aware reasoning, thereby improving the accuracy of data cleaning for datasets with spatial attributes. Lastly, we propose Twilight, a new framework for multivariate time-series data imputation that exploits exogenous reanalysis data to impute polar science observational data. Twilight is motivated by the observation that reanalysis datasets (e.g., ERA5) provide data that is synonymous to many polar science data, yet have mismatched spatial-temporal scales and systematical bias hence cannot directly substitute the to-be-imputed data. Twilight designs a single uniformed Transformer model that resolves the mismatching and bias of reanalysis information, and integrates it with the temporal continuity and inter-variable correlation to improve the imputation accuracy under the most challenging scenarios. In summary, this dissertation establishes a unified foundation for accessible and reliable polar science data by improving how such data are accessed, cleaned, and completed. These contributions help reduce key barriers in polar data analysis and enable more effective data-driven scientific discovery.

Author and committee

dc:creator, dc:contributor.*
Author dc:creator
  • Huang, Yuchuan

Rights

Language dc:language.iso
en

Identifiers

dc:identifier.*
Dc Identifier Citation
Huang, Yuchuan. (2026). Towards Accessible and Reliable Polar Science Data: Infrastructure,Cleaning, and Imputation. Retrieved from the University Digital Conservancy, https://hdl.handle.net/11299/283099.

Chain of custody

source
Harvested from
University of Minnesota
Base URL
conservancy.umn.edu/server/oai/request
Last updated
2026-11-19
Source record
OAI-PMH GetRecord
related terms
citation

Huang, Yuchuan. Towards Accessible and Reliable Polar Science Data: Infrastructure,Cleaning, and Imputation. 2026. https://hdl.handle.net/11299/283099