Contained within the fossil pollen and spore record is one of the most comprehensive histories of vegetation and its response to long term environmental change. However, scientists have been unable to efficiently use this record because pollen is still studied much in the same low technology way it was a century ago: by a highly trained expert using a transmitted light microscope. The proposed research intends to transform the study of pollen and spores into a high throughput and precise science by developing an automated system capable of identifying and classifying extremely diverse pollen and spore samples. This system will establish standards approaches for automated classification and be based on recent advances in machine learning and computer vision. The system will be developed and tested on 900 samples of pollen material collected over seventeen years from pollen traps placed in two tropical forests in Panama. Automation will allow, for the first time, a detailed study of the seasonal production of pollen in response to multiple El Nino-Southern Oscillation, or ENSO, events.
The results of the proposed research will increase the quantity and quality of tropical pollen data and promote needed research on the large scale dynamics of plant communities that is recorded in fossil pollen records. This research represents a fundamental transformation in the way scientists think about and approach the analysis of pollen data. The machine learning software that will be developed will be publically available, with the hope that other researchers will then adopt the methods and standards and establish objective measures for consistent and reliable pollen identifications. Educational goals of this project include undergraduate research training for underrepresented and first generation college students.