Data Science MS Thesis Defense by Ajmal Abbas

Monday, May 11, 2026 at 4:00pm to 5:00pm

College of Engineering

Data Science MS Thesis Defense

Title: Emulating Marine Ecosystem Dynamics: A Machine Learning Approach for Biomass Prediction and Forecasting

by Ajmal Abbas

Date & Time: Monday, May 11, 2026 at 4:00 PM

Meeting Link: Join: https://teams.microsoft.com/meet/210134622120814?p=HF1gXRarGipViGBpda

Meeting ID: 210 134 622 120 814   Passcode: Ut7pS6RH

Committee chair:

Dr. Gokhan Kul

Committee Members:

Dr. Gavin Fay (co-advisor)

Dr. Ashok Kumar Patel

Dr. Firas Khatib

Abstract: Marine Ecosystem models such as Atlantis provide valuable insights into multi-specific interactions, spatial dynamics and fisheries management, but their high computational cost limits rapid scenario analysis and real time decision-making. This study presented a data driven ecosystem emulator for the Northeast U.S. Atlantis model using machine learning and deep learning techniques. A large-scale dataset (~3.4 million records) spanning 1964-2020 was constructed from over 500 Atlantis model run scenarios, integrating biomass of species functional groups (guilds), spatial polygon, fishery removals and environments variables including temperature, salinity and primary production. The Emulator adopted a hybrid modeling framework combining Random Forest regression for accurate biomass prediction and Long Short -Term memory Networks for temporal forecasting of ecosystem dynamics. Feature Engineering incorporated spatial biomass distribution, lagged variables to capture ecological inertia and time aware transformations. Model evaluation included out of sample temporal validation, recursive back testing and ecological plausibility assessments to ensure robustness and realism. Result demonstrated strong predictive performance, with guild-level R2 values frequently exceeding 0.90 and overall model accuracy approaching 94%. The forecasting component produced stable, biologically consistent trajectories without unrealistic oscillations. Importantly, the emulator achieved high computational efficiency, with end-to-end prediction and forecasting completed in under few seconds, substantially reducing runtime compared to the original Atlantis simulations. This work established a scalable and efficient alternative to process-based ecosystem models, enabling rapid scenario testing and supporting data -driven fisheries management and environmental decision making.

Meeting Link:  Join: https://teams.microsoft.com/meet/210134622120814?p=HF1gXRarGipViGBpda
Meeting ID: 210 134 622 120 814  
Passcode: Ut7pS6RH

Teams
Gokhan Kul
gkul@umassd.edu
https://teams.microsoft.com/meet/210134622120814?p=HF1gXRarGipViGBpda