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,Thesis/Dissertations DESCRIPTION:Faculty Supervisor:Dr. Mohammad Karim, Electrical & Computer En gineering Committee Members:Dr. Donghui Yan, MathematicsDr. Tariq Manzur, Electrical & Computer Engineering Abstract:Advances in environmental sen sing technologies have enabled the collection of high-frequency atmospheri c observations, creating new opportunities for applying machine learning t o short-term environmental prediction. Despite considerable progress in da ta-driven atmospheric modeling, comparatively little attention has been gi ven to how atmospheric predictability varies among variables governed by d ifferent physical processes, how strongly prediction depends on atmospheri c memory, and how boundary-layer regime modifies the predictive informatio n available from instantaneous observations. Understanding these relations hips is essential for developing forecasting systems that are both accurat e and physically interpretable. In this study, we systematically assess t he short-term predictability of five key atmospheric variables—solar rad iation, air temperature, wind speed, barometric pressure, and relative hum idity—using approximately 2.2 years of high-frequency meteorological obs ervations collected within the Marine Wave Boundary Layer (MWBL) at the Sc hool for Marine Science and Technology (SMAST), University of Massachusett s Dartmouth. To account for fundamentally different radiative forcing and boundary-layer dynamics, the observations were separated into daytime and nighttime regimes and analyzed independently. A unified physics-informed m achine learning framework incorporating atmospheric memory, temporal encod ing, and wind-vector decomposition was developed and evaluated using a str ict chronological train-validation-test strategy to ensure realistic predi ction conditions. The results show clear differences in predictability an d optimal model complexity across the five atmospheric variables. Air temp erature and barometric pressure exhibited strong temporal persistence and achieved high predictive performance using linear models, whereas solar ra diation and wind speed benefited from nonlinear ensemble-learning approach es capable of capturing more complex atmospheric behavior. Relative humidi ty achieved strong predictive performance but exhibited greater sensitivit y to rapid moisture variability and evolving boundary-layer conditions. Ad ditional experiments conducted comparing the complete framework with a red uced framework, in which temporal-memory features were removed,  demonstr ated that atmospheric memory contributes to the predictability of all five atmospheric variables, although its importance varies considerably among them. Comparisons between daytime and nighttime conditions further showed that boundary-layer regime modifies the predictive information available f rom instantaneous atmospheric observations while preserving the overall re lationship between atmospheric physics and model complexity. To evaluate practical applicability, model predictions were compared with independent observations collected from an ATMOS 41W all-in-one weather station deploy ed near the University of Massachusetts Dartmouth Campus Tower, providing the preliminary field-based validation of the proposed framework. The best -performing solar radiation prediction model was subsequently integrated i nto a prototype Streamlit-based real-time prediction dashboard, demonstrat ing a practical pathway from environmental data science research to operat ional coastal forecasting applications. Overall, this study demonstrates t hat atmospheric predictability is governed by the underlying physical proc esses of each atmospheric variable, that atmospheric memory provides the d ominant source of predictive information, and that boundary-layer regime m odifies predictive information without fundamentally altering the relation ship between atmospheric physics and appropriate model complexity. These f indings provide both scientific insight into coastal atmospheric predictab ility and a practical foundation for future environmental forecasting syst ems.  For further information, please contact Dr. Mohammad Karim at mkar im@umassd.edu.\nEvent page: https://www.umassd.edu/events/cms/8-14-26-vari able-dependent-predictability-of-coastal-atmospheric-parameters.php\nEvent link: https://umassd.zoom.us/j/98731680128?pwd=y211mggmO6iXQUy9DjXa5wbyPD cBqv.1 X-ALT-DESC;FMTTYPE=text/html:

Faculty Supervisor:
Dr. Mo hammad Karim\, Electrical & Computer Engineering
 
Committee Me mbers:
Dr. Donghui Yan\, Mathematics
Dr. Tariq Manzur\, Electric al & Computer Engineering
 
Abstract:
Advances in environm ental sensing technologies have enabled the collection of high-frequency a tmospheric observations\, creating new opportunities for applying machine learning to short-term environmental prediction. Despite considerable prog ress in data-driven atmospheric modeling\, comparatively little attention has been given to how atmospheric predictability varies among variables go verned by different physical processes\, how strongly prediction depends o n atmospheric memory\, and how boundary-layer regime modifies the predicti ve information available from instantaneous observations. Understanding th ese relationships is essential for developing forecasting systems that are both accurate and physically interpretable.
 
In this study\, we systematically assess the short-term predictability of five key atmosph eric variables—solar radiation\, air temperature\, wind speed\, barometr ic pressure\, and relative humidity—using approximately 2.2 years of hig h-frequency meteorological observations collected within the Marine Wave B oundary Layer (MWBL) at the School for Marine Science and Technology (SMAS T)\, University of Massachusetts Dartmouth. To account for fundamentally d ifferent radiative forcing and boundary-layer dynamics\, the observations were separated into daytime and nighttime regimes and analyzed independent ly. A unified physics-informed machine learning framework incorporating at mospheric memory\, temporal encoding\, and wind-vector decomposition was d eveloped and evaluated using a strict chronological train-validation-test strategy to ensure realistic prediction conditions.
 
The resul ts show clear differences in predictability and optimal model complexity a cross the five atmospheric variables. Air temperature and barometric press ure exhibited strong temporal persistence and achieved high predictive per formance using linear models\, whereas solar radiation and wind speed bene fited from nonlinear ensemble-learning approaches capable of capturing mor e complex atmospheric behavior. Relative humidity achieved strong predicti ve performance but exhibited greater sensitivity to rapid moisture variabi lity and evolving boundary-layer conditions. Additional experiments conduc ted comparing the complete framework with a reduced framework\, in which t emporal-memory features were removed\,  demonstrated that atmospheric mem ory contributes to the predictability of all five atmospheric variables\, although its importance varies considerably among them. Comparisons betwee n daytime and nighttime conditions further showed that boundary-layer regi me modifies the predictive information available from instantaneous atmosp heric observations while preserving the overall relationship between atmos pheric physics and model complexity.
 
To evaluate practical ap plicability\, model predictions were compared with independent observation s collected from an ATMOS 41W all-in-one weather station deployed near the University of Massachusetts Dartmouth Campus Tower\, providing the prelim inary field-based validation of the proposed framework. The best-performin g solar radiation prediction model was subsequently integrated into a prot otype Streamlit-based real-time prediction dashboard\, demonstrating a pra ctical pathway from environmental data science research to operational coa stal forecasting applications. Overall\, this study demonstrates that atmo spheric predictability is governed by the underlying physical processes of each atmospheric variable\, that atmospheric memory provides the dominant source of predictive information\, and that boundary-layer regime modifie s predictive information without fundamentally altering the relationship b etween atmospheric physics and appropriate model complexity. These finding s provide both scientific insight into coastal atmospheric predictability and a practical foundation for future environmental forecasting systems. 
 
For further information\, please contact Dr. Mohammad Karim at mkarim@umassd.edu.

Event page: https://www.umassd.edu/events/cms/8-14-26-variable-depende nt-predictability-of-coastal-atmospheric-parameters.php
Event link: https://umassd.zoom.us/j/98731680128?pwd=y211mggmO6iXQUy9DjXa5w byPDcBqv.1

DTSTAMP:20260722T010529 DTSTART;TZID=America/New_York:20260814T100000 DTEND;TZID=America/New_York:20260814T110000 LOCATION:Online via Zoom SUMMARY;LANGUAGE=en-us:Variable-Dependent Predictability of Coastal Atmosph eric Parameters Using Physics-Informed Machine Learning in the Marine Wave Boundary Layer UID:0104621f508f6113ebf2ec617c705ae2@www.umassd.edu END:VEVENT END:VCALENDAR