BEGIN:VCALENDAR VERSION:2.0 X-WR-CALNAME:EventsCalendar PRODID:-//hacksw/handcal//NONSGML v1.0//EN CALSCALE:GREGORIAN BEGIN:VTIMEZONE TZID:America/New_York LAST-MODIFIED:20240422T053451Z TZURL:https://www.tzurl.org/zoneinfo-outlook/America/New_York X-LIC-LOCATION:America/New_York BEGIN:DAYLIGHT TZNAME:EDT TZOFFSETFROM:-0500 TZOFFSETTO:-0400 DTSTART:19700308T020000 RRULE:FREQ=YEARLY;BYMONTH=3;BYDAY=2SU END:DAYLIGHT BEGIN:STANDARD TZNAME:EST TZOFFSETFROM:-0400 TZOFFSETTO:-0500 DTSTART:19701101T020000 RRULE:FREQ=YEARLY;BYMONTH=11;BYDAY=1SU END:STANDARD END:VTIMEZONE BEGIN:VEVENT CATEGORIES:College of Engineering,Graduate Studies,Lectures and Seminars,Th esis/Dissertations DESCRIPTION:Faculty Supervisor:Dr. Gokhan Kul, Computer & Information Scien ce/Data Science Committee Members: Dr. Gavin Fay, SMAST Fisheries & Oceano graphyDr. Firas Khatib, Computer & Information Science/Data ScienceDr. Ash okkumar Patel, Computer & Information Science/Data Science Abstract: Marin e ecosystem models such as Atlantis valuable insight into multispecies int eractions , spatial dynamics and fisheries management, but their high comp utational cost limits rapid scenario analysis and real-time decision makin g. This study presented a data-driven ecosystem emulator for the Northeast U.S. Atlantis model using automated machine learning approaches. A large -scale dataset (~3.4 million records) spanning 1964-2020 was constructed , integrating bio ass of species functional groups , spatial polygons , tem poral indices and environments variables such as temperature and salinity.  The emulator framework employed automated machine learning techniques, including Random Forest and Extra tress regression, with model selection a nd hyperparameter optimization performance using AutoML strategies. In add ition, AutoKeras was utilized to explore neural network architectures in a n automated manner, enabling data-driven model allocation, lagged variable s to capture ecological inertia and time-aware transformations. Model perf ormance was evaluated using out-of-sample temporal validation, recursive b ack testing, and ecological plausibility assessments. Result demonstrated string predictive performance, with species -level R2 values frequently ex ceeding 0.90 and overall model accuracy approaching 94%. The emulator achi eved high computational efficiency, with end-to-end prediction completed i n under few seconds, substantiality reducing runtime compared to tradition al Atlantis simulations. This work established a scalable and efficient Au toML-driven alternative to process-based ecosystem model, enabling rapid b iomass estimation and supporting data-driven fisheries management and ecos ystem analysis.  For further information please contact Dr. Gokhan Kul at gkul@umassd.edu \nEvent page: https://www.umassd.edu/events/cms/emulatin g-marine-ecosystem-dynamics-a-machine-learning-approach-for-biomass-predic tion-and-forecasting.php\nEvent link: https://teams.microsoft.com/meet/210 134622120814?p=HF1gXRarGipViGBpda X-ALT-DESC;FMTTYPE=text/html:
Faculty Supervisor:
Dr. Gokhan Kul\, Computer & Information Science/Data Science
\n

Committ ee Members:
Dr. Gavin Fay\, SMAST Fisheries & Oceanography
Dr. Firas Khatib\, Computer & Information Science/Data Science
Dr. Ashokk umar Patel\, Computer & Information Science/Data Science

\n

Abstract: Marine ecosystem models such as Atlantis valuable insight into multispeci es interactions \, spatial dynamics and fisheries management\, but their h igh computational cost limits rapid scenario analysis and real-time decisi on making. This study presented a data-driven ecosystem emulator for the N ortheast U.S. Atlantis model using automated machine learning approaches. A large -scale dataset (~3.4 million records) spanning 1964-2020 was const ructed \, integrating bio ass of species functional groups \, spatial poly gons \, temporal indices and environments variables such as temperature an d salinity.  The emulator framework employed automated machine learning t echniques\, including Random Forest and Extra tress regression\, with mode l selection and hyperparameter optimization performance using AutoML strat egies. In addition\, AutoKeras was utilized to explore neural network arch itectures in an automated manner\, enabling data-driven model allocation\, lagged variables to capture ecological inertia and time-aware transformat ions. Model performance was evaluated using out-of-sample temporal validat ion\, recursive back testing\, and ecological plausibility assessments. Re sult demonstrated string predictive performance\, with species -level R2 v alues frequently exceeding 0.90 and overall model accuracy approaching 94% . The emulator achieved high computational efficiency\, with end-to-end pr ediction completed in under few seconds\, substantiality reducing runtime compared to traditional Atlantis simulations. This work established a scal able and efficient AutoML-driven alternative to process-based ecosystem mo del\, enabling rapid biomass estimation and supporting data-driven fisheri es management and ecosystem analysis. 

\n

For further information pl ease contact Dr. Gokhan Kul at gkul@umass d.edu 

Event page: https://www.umassd.edu/events/cms/emulati ng-marine-ecosystem-dynamics-a-machine-learning-approach-for-biomass-predi ction-and-forecasting.php
Event link: https://teams.microsoft .com/meet/210134622120814?p=HF1gXRarGipViGBpda

DTSTAMP:20260421T151412 DTSTART;TZID=America/New_York:20260511T160000 DTEND;TZID=America/New_York:20260511T170000 LOCATION:Online/Teams (https://teams.microsoft.com/meet/210134622120814?p=H F1gXRarGipViGBpda) SUMMARY;LANGUAGE=en-us:Emulating Marine Ecosystem Dynamics: A Machine Learn ing Approach For Biomass Prediction And Forecasting UID:040cbd30c1740addfa22b4b1b701e07d@www.umassd.edu END:VEVENT END:VCALENDAR