Akila Sampath

iHARP Research Assistant | UMBC Ph.D. Candidate

Email: asampath@umbc.edu

Research Interests

  • Artificial Intelligence
  • Machine Learning
  • Causal Inference

Successfully defended the Ph.D. proposal in December 2024

Short Biography

Akila Sampath is a PhD student at IS department at the University of Maryland Baltimore County. Her PhD advisor is Prof. Jianwu Wang, and her co-advisor is Prof. Vandana Janeja. Her research focuses broadly on the application of AI to the Arctic climate. She currently works as a research assistant at iHARP. The goal of her thesis work is to build a physics-guided ML model to exploit causalities between Arctic climate phenomena.

Research Summary

Akila is developing a physics-informed machine learning model to predict snow depth, which is an important factor in Arctic sea ice loss. And sea ice loss in the Arctic is a major sign of rapid environmental change. Traditional climate models have had trouble capturing the complex interactions between the atmosphere, ocean, and sea ice. Machine learning can analyze large datasets and find patterns, but most models in this area don’t use domain knowledge from physics. Akila’s research fills this gap. This makes it more accurate and helps us understand seasonal changes in snow depth better. Her work aims to improve how we understand climate shifts and support better decision-making for climate policy.

Publications
  • Akila Sampath, Omar Faruque, Azim Khan, Vandana Janeja, and Jianwu Wang. 2024. Physics-Informed Machine Learning for Sea Ice Thickness Prediction. In 2024 IEEE International Conference on Knowledge Graph (ICKG), December 11, 2024. IEEE, Abu Dhabi, United Arab Emirates, 325–333. https://doi.org/10.1109/ICKG63256.2024.00048
  • Sahara Ali, Uzma Hasan, Xingyan Li, Omar Faruque, Akila Sampath, Yiyi Huang, Md Osman Gani, and Jianwu Wang. 2024. Causality for Earth Science — A Review on Time-series and Spatiotemporal Causality Methods. https://doi.org/10.48550/arXiv.2404.05746
  • Sudip Chakraborty, Chhaya Kulkarni, Atefeh Jabeli, Akila Sampath, Gehan Boteju, Jianwu Wang, and Vandana Janeja. 2023. Understanding the Role of 2019 Amazon Wildfires on Antarctic Ice Sheet Melting Using Data Science Approaches. Accepted 2023 ACM SIGKDD International Conference on Knowledge Discovery and Data Mining Workshop: Fragile Earth: AI for Climate Sustainability – from Wildfire Disaster Management to Public Health and Beyond, 2023 [Conference/Workshop Paper]
Internships/ Fellowships
  • 2025 Summer – Center for Learning the Earth with Artificial Intelligence and Physics (LEAP) at Columbia University
    • Akila will work on understanding cloud microphysical processes in climate models, focusing on applying data science and machine learning techniques to climate modeling.
Proposal Defense

Successfully Defended On: December 4, 2024

Abstract

The Arctic region is undergoing rapid environmental changes, with sea ice loss being a prominent indicator. Accurate prediction of sea ice thickness and extent is crucial for understanding climate change impacts and informing policy decisions. Traditional climate models often struggle to capture the complex interactions between atmospheric, oceanic, and sea ice processes. In recent years, machine learning has emerged as a powerful tool for analyzing large datasets and making accurate predictions. However, machine learning models, while capable of capturing complex patterns, can sometimes produce unrealistic or physically implausible results. To address this limitation, we propose a novel approach that combines the power of machine learning with the rigor of physical principles. By integrating physics-informed techniques into machine learning models, we aim to develop more accurate and reliable predictions of sea ice conditions.
Physics-Informed Machine Learning (PIML) is a novel approach that blends traditional scientific models with the pattern-recognition capabilities of machine learning to improve predictions of sea ice thickness. By leveraging both data-driven and physics-based knowledge, PIML offers more accurate and reliable predictions, even in data-scarce environments. This approach is particularly well-suited for studying complex systems like the Arctic sea ice. Our research introduces Physics-Encoded Neural Networks (PeNNs) to predict snow density. PeNNs embed physical laws directly into the neural network architecture, enabling the estimation of hidden physical properties from observable data. This makes PeNNs highly effective for tracking snow and ice conditions with precision. Additionally, we developed a physics-guided model (PGM) to investigate the causal link between sea ice loss and increased turbulence in the Beaufort Gyre through wind current. By incorporating physical constraints, the PGM ensures that predictions align with known scientific principles, aiding in the understanding of these causal relationships. Our proposed models offer a more comprehensive approach to studying Arctic sea ice. By integrating physics into machine learning, we aim to create tools that are both accurate and interpretable. This research provides valuable insights into Arctic climate change and demonstrates how physics can enhance the capabilities of machine learning models, leading to a deeper understanding of the impacts of sea ice loss.

Committee

  • Dr. Jianwu Wang – Chair/Advisor (UMBC)
  • Dr. Vandana Janeja – Co-Chair (UMBC)
  • Dr. Houbing Song – Committee Member (UMBC)
  • Dr. James Foulds – Committee Member  (UMBC)
  • Dr. Donald. K. Perovich – Committee Member (Dartmouth College)
  • Dr. Nicole Schlegel – Committee Member (NOAA)

Last updated 7 May 2025