Interpreting the results of oceanic mesoscale enrichment experiments: Caveats and lessons from limnology and coastal ecology

Hale and Rivkin (2006; hereafter HR) provide a critique of the statistical treatment of our results (Oliver et al. 2004; hereafter O2004 as in HR) from the Southern Ocean Iron Experiment (SOFeX) study and question our conclusion that differences in bacterial properties between iron-enriched (''fertilized'') and non-iron-enriched (''unferti-lized'') water masses were due to the iron addition. HR made the important point that iron enrichment studies have not been properly replicated. As a result, appropriate statistical analyses cannot be performed, and statistically rigorous inferences about responses and causative factors cannot be made. They also noted that few studies in the history of open-ocean iron fertilization experiments have addressed the issue of replication—or rather, the lack thereof. HR singled out our paper and that by Arrieta et al. (2004) for criticism because, unlike the vast majority of the iron enrichment literature, we actually performed statistical analyses of our results—just not the appropriate ones, according to HR. HR's critique of our analyses suggests some other analytical approaches and experimental and sampling designs that might be used in future studies. HR point out that time-series measurements, as performed in a fertilized water mass (''patch''), may not be independent of each other, and should not be treated as replicates in a statistical analysis. This is especially important when considering oceanographic patches tagged with geochemical tracers and followed in the Lagrangian frame of reference. Such patches will exhibit some degree of biogeochemical and physical continuity or integrity over time. Whether the surrounding ''control'' patches exhibit similar coherence, and are subject to the same caveats about statistical replication, is an interesting question not explored by HR. It is likely, for example, that the tagged patch will flow past several distinct patches with different properties. Measurements within each of the distinct, nonexperimental patches would be independent. In the future, mesoscale enrichment studies should perhaps tag both the (replicated!) fertilized and unfertilized (control) patches, to provide better experimental controls on the treatment effects. But the logistics of following and sampling them would be daunting and at present are not practicable, given the overwhelming limitations in conducting open-ocean field work at this scale. HR used ANCOVA to reanalyze our time-series results and concluded that ''ANCOVA results for bacterial properties during SOFeX were generally similar to those reported by O2004 using two-sample t-tests.'' Next, they asked three questions about our results, ''considering the following in turn: (1) whether the magnitude of …