Alaska: Google Summer of Code (GSoC)

<img src="https://raw.githubusercontent.com/uaanchorage/GSoC/main/figures/Alaska.png" width="150" height="150" align="left" style="padding:10px;"/> 2026 will be our third time participating in the Google Summer of Code (GSoC) as a mentoring organization, and we are already excited about the potential and opportunities. We had four enthusiastic contributors as a first-time organization in 2024 and we doubled in 2025 with our excellent 8 contributors. We consist of stable open-source projects in production use in research and relatively young projects. We also have mentors who have participated in several GSoC instances over the past several years and have been active in open-source software development for decades.

We represent the 49th state, Alaska. Anchorage, the largest city in Alaska, has a vibrant open-source community. Through this GSoC initiative, researchers from the University of Alaska Anchorage (UAA) and University of Alaska Fairbanks (UAF) join hands with the Alaska-based software experts that are part of open-source entities such as Healthy and Alaska Developer Alliance to provide a perfect mentoring experience for interested contributors globally. We provide a glimpse of this northern state and its tech landscape to the Lower 48 and the outside world through this open-source remote summer coding program organized and funded by Google. Our projects focus on healthcare, climate science, polar science, and other research fields critical to the Circumpolar North and the rest of the world.

Please first refer to the contributor guidelines to get started! It puts you on the right track with application details and a standard template. Please also refer to our Acceptable and Ethical AI Use Policy to make sure your use of AI/ML/LLM tools such as ChatGPT falls under the acceptable use. A rubrics is provided as reference material on being a competitive applicant for Alaska in GSoC. Some helpful pointers on effective communication are also provided.

Please avoid sending individual private emails (or social media messages!!) to mentors. However, the mentors' emails with each project idea are listed below in case the mentor initiates or recommends an email communication later. If you are proposing your own idea, make sure that idea is relevant to Alaska and has a potential mentor from the list of mentors below.

Project Ideas

Many of the ideas proposed here have a research component. Contributors who work on these ideas have the potential to author a research paper (as the first author, working with the researchers from the University of Alaska) or become co-authors in our ongoing research papers. We strongly encourage those interested in higher studies or research careers to apply for their GSoC with us.

---

[1] Automated coastline extraction for erosion modeling in Alaska.

Mentors: Frank Witmer (fwitmer -at- alaska.edu) and Ritika Kumari (rkjane333 -at- gmail.com)

Overview: The rapidly warming Arctic is leading to increased rates of coastal erosion, placing hundreds of Alaska communities at the frontline of climate change. Understanding current rates of coastline change and accurately forecasting future changes is critical for communities to mitigate and adapt to these changes. Current modeling approaches typically use a simple linear model based solely on historical coastline positions to measure rates of change and extrapolate them into the future. In doing so, these models fail to capture the dynamic effects associated with decreasing sea ice, increasing annual wave energy, and increasing temperatures. To improve the quality of these coastal models, we need to increase the quantity of digitized coastlines, but manual photointerpretation is slow and laborious.

Current Status: An initial model and pipeline have been developed to automatically extract coastlines from PlanetLabs imagery. An auto-download script is available to retrieve PlanetLabs imagery (3-5m spatial resolution) by specifying any timeframe, cloud coverage percentage, and geometry. Additionally, NDWI with a majority sliding window has been introduced, allowing a specific threshold for each window to improve water detection accuracy. The DeepWaterMap algorithm was originally trained with the Global Surface Water (GSW) dataset at 30 m resolution from Landsat imagery, but the model did not not work well applied to PlanetLabs imagery. We are working to re-train the model using PlanetLabs imagery automatically labeled using the NDWI thresholding method. This project extends and expands on the progress made in 2024 and 2025.

Potential areas of improvement:

• Improve training data by incorporating the PlanetLabs Usable Data Mask (UDM) data.
• Data Expansion (Deering 2017–2019 and Beyond): Currently using data from 2017 to 2019 for Deering; we plan to include more recent data to extend the time series.
• Improved Cliff Area Segmentation: Enhance segmentation performance specifically in steep or cliff-like coastal areas.
• Handling Challenging Conditions: Improve segmentation in regions with water shadows, buildings, satellite artifacts, and other data quality issues.
• SWIR and Elevation Data Integration: Investigate combining short-wave infrared (SWIR) data and elevation data (e.g., DEMs) to further refine segmentation accuracy.

Expected Outcomes: A finished model with high accuracy that automatically extracts a vectorized coastline representation from PlanetLabs satellite imagery. Then, the model can be applied to large amounts of imagery to model coastline changes over time.

Required Skills: Python

Code Challenge: Experience with multi-band satellite imagery, geospatial data processing, and machine learning.

Source Code: https://github.com/fwitmer/CoastlineExtraction

Discussion Forum: https://github.com/fwitmer/CoastlineExtraction/discussions

Effort: 350 Hours

Difficulty Level: Intermediate

---

[2] BHV: Behavioral Health Vault.

Mentors: Mohamed Abdullah F (abdullahfakrudeen2020 -at- gmail.com) and Pradeeban Kathiravelu (pkathiravelu -at- alaska.edu)

Overview: The goal of this project is to provide a digitization approach to record the journey of recovery of people with serious mental illnesses, substance-use disorders, and other social determinants affecting well-being. BHV (Behavioral Health Vault) aims to complement traditional Electronic Health Records (EHRs) by storing patient provided images such as photographs, sketches, and scanned drawings along with associated textual narratives, which may be provided directly by patients or recorded by social workers during interviews. BHV is a minimal, Python-based application that enables healthcare networks to store and retrieve patient-provided images. BHV aims to provide them access to upload, view, and edit their own images and narratives. It will provide admin-level access to system administrators, allowing them to view the entire ecosystem, upload images on behalf of users, along with the narrative, edit images on behalf of users, and delete images or narrations on behalf of users or as a moderation action. Importantly, BHV will also include an optional research module integrating a fuzzy logic–based color-emotion association model (inspired by Color-Emotion Associations in Art: Fuzzy Approach). This enables researchers to explore correlations between color palettes in user-submitted images and self-reported emotions supporting studies on emotional recovery, addiction patterns, and social determinants of mental health. This user-driven data (images + emotional self-reports) helps researchers study how environmental factors, social contexts, and personal narratives affect mental health. The system is built for participatory behavioral health research, emphasizing accessibility, transparency, and ethical data handling.

Current Status: Currently, we have Beehive project. However, it has a complex architecture, which has made it a difficult installation for community health clinics that lack the technical expertise. The goal of BHV is to provide the same functionality while minimizing the complex tech stack of Beehive. The eventual goal of BHV is to replace Beehive and be the "Beehive-2.0."

Expected Outcomes: The Behavioral Health Vault (BHV) will deliver a secure yet minimal system designed for ease of use and installation in healthcare and research environments. The signup and login process will be intentionally simple, allowing users to register using a basic email and password combination or Using Google OAuth would be nice, without unnecessary authentication complexity. The platform will prioritize a lightweight, modular architecture that avoids bloat, ensuring that community clinics and behavioral health researchers can deploy the system effortlessly, even with limited technical expertise. To achieve this, BHV will be installable and runnable from a single command, without the need to start separate services for the frontend, backend, and database. Images will be stored efficiently within the local file system, accompanied by a database index that enables fast search and retrieval of user submissions and narratives. The design will follow a privacy