We present a modular science gateway that simplifies the deployment of machine learning workflows across heterogeneous computing environments. Designed for domain researchers without DevOps expertise, our system lowers the barrier to scalable ML by integrating Clowder, an open-source data management platform, with Ray and Kubernetes for distributed execution. The platform supports data ingestion, visualization, and metadata management, with ML-specific UI enhancements. Workflows are executed via containerized extractors that interact with a shared Ray cluster. To demonstrate real-world utility, we apply our system to the detection of ice wedge polygons from Arctic satellite imagery. Researchers can fine-tune and run inference workflows through a simple web interface, without managing infrastructure. This approach enhances accessibility and reproducibility, and promotes the reuse of data and models across research communities.

This pilot project develops a geospatial deep learning framework to assess permafrost thaw risk and identify potential relocation or neighborhood extension areas within a buffer zone of the Wainwright region. A 100 m spatial resolution grid system was overlaid across the study area and ingested multiple data layers including drained lakes, ice wedge polygons, InSAR data, contaminated sites, and infrastructure. Future work will include gravel-extraction sites, coastal erosion, thaw slumps, subsidence, ArcticDEM, and detailed infrastructure datasets. All derived features form an input matrix. These variables are combined into a composite risk index, which serves as the target for a deep learning autoencoder pipeline. After data scaling and encoding, the first draft of a deep autoencoder was applied to learn a compressed latent representation, to predict risk level. The resulting risk map is expected to provide actionable insights for community planners seeking to prioritize high risk zones and plan extension or relocation strategies under changing permafrost conditions.

Due to their weather and illumination independence and due to their large area coverage at high spatial resolution, Synthetic Aperture Radar (SAR) images have been recognized as a valuable data source for the mapping and tracking of aufeis flooding events. We modified and utilized the change detection approach of [1], based on wavelet analysis to map aufeis-related flooding on the Sagavanirktok River in northern Alaska, collected in the spring of 2015. This paper provides near real-time monitoring by generating detailed flood parameters such as flood classification probabilities, flood-related backscatter changes, and flood extent. The generated flood maps show the spatial extent and day-to-day progression of the 2015 flooding event across a 1004 km2area, that was determined by processing a series of seven TerraSAR-X datasets. From our analysis, we learned that in early to late April, the formation of aufeis from the Sagavanirktok River crossed the Dalton Highway at several points. In late April to early May, associated warm temperatures led to open water flooding, which flooded the Sagavanirktok River at several locations. It can be argued that along the Sagavanirktok River, warm temperatures led to aufeis growth with the distribution of flow moving from the western channel to the eastern channel.