Climatic, environmental, and phenological analyses of diverse lowland neotropical pollen rain data using ecoinformatic and machine learning tools

Analyses of modern pollen rain data from the Neotropics have traditionally been used to help interpret compositional changes observed in sedimentary pollen data. Comparisons of modern pollen assemblage data produced by different forest environments are compared to compositional changes observed within the fossil pollen record to improve interpretations of how plant communities have changed over time and under different climatic and environmental conditions. While modern pollen records provide an invaluable resource to improve paleoecological interpretations, most records are limited to short-term (1-3 year) sampling durations. These short-term sampling durations can potentially misrepresent comparisons of pollen and vegetation in different forest communities by not accounting for the full range of natural variability in phenological pollen outputs. For my dissertation, I counted and analyzed three ≥ 10-year pollen rain records obtained from two lowland Panamanian forests: Barro Colorado Island (BCI) and Parque Nacional San Lorenzo (PNSL). Together, these records represent the three longest continuous collections of airborne pollen data analyzed from the Neotropics to date. This dissertation explores novel approaches to the analysis of hyperdiverse Neotropical pollen rain assemblage data. A machine-based ecoinformatic analysis was used to correlate seasonal and annual variability in pollen abundance data to a suite of climatic variables. The analysis explores how climatic variability influences the composition of pollen assemblage data in forest sites characterized by differences in seasonality. A 15-year pollen rain collected within a forest dynamics plot located on BCI was used to analyze relationships between pollen abundances, biomass, and flowering patterns. The extended pollen rain highlights the extent to which the relationship between pollen abundance data and standing biomass can vary on a year- to-year basis and the potential for aerial pollen trapping data to supplement the study of tropical flowering patterns. Using pollen identifications from the BCI plot pollen rain, a machine learning method using convolutional neural nets was developed to fully automate the process of pollen identification.

Name thesis:degree_name

Ph.D.

Level thesis:degree_level

Dissertation

Discipline thesis:degree_discipline

Ecol, Evol, Conservation Biol

Grantor

University of Illinois at Urbana-Champaign

Year dc:date

2019

Author dc:creator

• Haselhorst, Derek Scott

Contributors dc:contributor

• Punyasena, Surangi W.
• Dalling, James W.
• Yannarell, Anthony C.
• Mander, Luke

• ecology
• Neotropical
• palynology
• phenology

Statement dc:rights

• Copyright 2019 Derek Haselhorst

Language dc:language

en

Handle dc:identifier

http://hdl.handle.net/2142/105046

OAI identifier oai:identifier

oai:www.ideals.illinois.edu:2142/105046