CPR coccolithophore occurrence in the North Atlantic rose 15-fold in 1965–2010 without a step at the 1993 counting change, but did not follow a fitted CO2 relation in 2011–2022

Question: Rivero-Calle et al. (2015) reported a tenfold increase in Continuous Plankton Recorder (CPR) coccolithophore occurrence in the North Atlantic in 1965-2010, with CO2 and the AMO as the best predictors. Does the rise replicate in public CPR data? Does the 1993 start of coccolithophore counting explain it? Does a 1965-2010 CO2 relation predict 2011-2022? Data: 188,628 CPR samples (CC BY GBIF copy, 1946-2020; OBIS copy to 2022) in 40-66 N, 70 W-10 E; Mauna Loa CO2; AMO; ERSSTv5; MODIS-Aqua monthly PIC for 2003-2020. Method: month-balanced annual frequency; cell-month standardized logistic index; step test at 1992 against 42 placebo break years; logit OLS fitted on 1965-2010 and used to predict later years; comparison with 60 other long-recorded phytoplankton categories. The analysis plan was written before the series were computed. Results: Occurrence increased from 1.2% (1965-1970) to 18.2% (2000-2010) of samples, in all three regions. The 1992 step (x1.63) was smaller than 18 of 42 placebo steps. CO2 slopes were similar inside each recording rule. But occurrence stopped increasing after 2007 and decreased after 2016. In 2017-2020 it was 0.20, against 0.41 from the CO2 model (ratio 0.48, AR(1)-adjusted z = -2.8). In 2021-2022 it was 0.19-0.23 (OBIS), against 0.47-0.49. Most other phytoplankton categories also decreased after 2016. Relative to their median, coccolithophores changed by -0.19 logit (95% CI -0.48 to +0.11) from 2007-2013 to 2017-2020, while the CO2 model implies +0.87. The 1965-2010 rise was much larger than in other categories (+3.2 vs +0.39 logit). Exploratory: the standardized index increased in short steps (1972-74, 1998-2000, 2005-07) and was flat between them. CPR presence was not positively related to matched satellite PIC, so PIC does not confirm the rise or the fall. Conclusion: the CPR rise is real in the data and is not caused by the counting change, but the CO2 relation did not predict the next decade. Limitations: presence index with a 270 um mesh; analyst changes cannot be tested; short test period, with few samples in 2020; linear models, not the original random forests.

coccolithophores · Continuous Plankton Recorder · Coccolithaceae · Emiliania huxleyi · North Atlantic · ocean acidification · CO2 fertilization · out-of-sample test · recording change · replication · particulate inorganic carbon · MODIS · phytoplankton time series