Pollen–vegetation relationships at a tropical cloud forest's upper limit and accuracy of vegetation inference

• Published in: Review of palaeobotany and palynology Vol. 184; pp. 1 - 13
• Main Authors: Crausbay, Shelley D., Hotchkiss, Sara C.
• Format: Journal Article
• Language: English
• Published: Elsevier B.V 15.09.2012
• Subjects: animals; bogs; data collection; fossils; highlands; lakes; modern pollen rain; paleoecology; palynology; Plantago; Poaceae; pollen; pollen–vegetation relationship; pollination; ROC analysis; spores; surface soils; tropical montane cloud forest; tropical montane cloud forests; upper forest line; wetland plants; wind; Hawaii; upper forest line; surface soils; ROC analysis; pollen–vegetation relationship; tropical montane cloud forest; modern pollen rain
• DOI: 10.1016/j.revpalbo.2012.07.010
• ISSN: 0034-6667

Palaeoecological records are increasingly needed from tropical montane systems. To infer past tropical vegetation dynamics, understanding modern pollen–vegetation relationships is required, but understanding effects of different sampling media between calibration datasets and palaeorecords is also needed. This is especially true when palaeorecords are derived from bogs or lakes with boggy shores and common wetland plants share the same pollen taxon with important upland plants that distinguish tropical vegetation types (e.g., Poaceae and Plantago). We assessed modern pollen–vegetation relationships around an upper cloud forest line in the Hawaiian Islands and tested the utility of a modern pollen calibration dataset derived from 88 surface soil samples when applied in 10 test wetland sites more typical of palaeorecords. We assessed over- and under-representation of pollen/spore taxa with a direct comparison to plant abundance and derived several metrics from the pollen/spore assemblages — analogs, ordinations, relative abundance of life forms, and ratios of life forms. We used the Receiver Operator Characteristic (ROC) to (1) compare metric performance at distinguishing vegetation around the upper forest line, (2) assess whether excluding wetland taxa significantly affected metric performance, and (3) test the accuracy of vegetation inference. Pollen–vegetation relationships were influenced by great ecological breadth and over- or under-representation of pollen and spores, which could be explained by pollination syndrome (wind vs. animal), grain/spore mass and upslope transport in winds. However, we found no evidence that upslope transport significantly blurred the upper-forest-line signal here, likely because winds are predominantly perpendicular to slope, and vertically constrained by the trade-wind inversion. Pollen from Poaceae and Plantago characterizes vegetation around this Hawaiian upper forest line and dominates wetland assemblages. Removing wetland taxa from the modern pollen calibration dataset levied no cost on a metric's performance, and greatly reduced the incidence of inaccurate vegetation inference in test sites. Minor error rates remained when rare, over-represented, or ecologically broad types were used in isolation. Overall, this study demonstrates that inferring forest line position from fossil pollen/spore assemblages requires careful consideration because (1) differences in sampling media between the modern calibration dataset and palaeorecords create opportunity for inaccurate vegetation inference and (2) some metrics perform better than others. ► Modern pollen rain distinguished vegetation around a tropical upper forest limit ► Some metrics (ordination, ratios, and sums across life forms) performed best ► Pollen taxa indicative of subalpine–alpine vegetation were abundant in wetlands ► Excluding wetland pollen taxa from pollen assemblages levied no cost on performance ► Excluding wetland pollen taxa greatly reduced inaccurate vegetation inference