Ecosystem properties have a mean and a variability. The drivers of those in the face of change—nutrient levels, precipitation, dispersal limitation—is a worthy topic of investigation. This summary from the NutNet experiment explores associations between a number of such variables, with an emphasis on the p-values arising from this large, diverse dataset. Some results appear to have been reported from this experiment before—decreases in diversity and increases in biomass with Nitrogen in particular, with P playing a secondary role—and the novelty is a focus is on a derived variable, essentially the CV over time. Again, the authors find that N is doing a lot of the work, increasing the CV in biomass, and evidence for additive effects, not synergy, among the combinations of N, P and K+. This is good work.

My main critiques lie in three areas.

First, stability is a measure of resilience--the rate that a system returns to equilibrium after perturbation. The authors are not reporting mean and variance associated with a stressor (as did Tilman 1999 using biodiversity as a driver of stability/resilience to drought). The closest to this it seems is their use of precipitation as a covariate, but the nature of this hypothesis seems implicit and could use spelling out.

Given the importance of the concept of stability vs variability, I sometimes found myself wondering which property they were addressing. The authors are measuring changes in the mean and variance in biomass production, and then a ratio (effectively Coefficient of Variation). Sometimes they use stability and variability interchangeably (for example, see paragraph at 122). But in the Results they report the CV as stability, and mean and variance separately (although crucially, the former derives from the latter (e.g., line 285)).

Second, the introduction leaves unclear the nature of the hypotheses. Line 155 indicates that variance “depends” on the specific nutrient. Why (but for the nature of the experiment) would the authors predict biomass variance would differ in response to a factorial NPK fertilization? They go onto suggest it should be “partially dependent” on a number of other things. If the authors had no such explicit hypotheses before beginning the analysis, just say this was an exploratory analysis. If the hypotheses summarized in 155 are explicitly linked to earlier text in the Intro, these links need to be strengthened. For example:

321: A subsection is dedicated to precipitation, but without any context. What was the hypothesis?

Third, the results in the main text make it difficult for the reader to imagine the magnitude, slope, and proportion of variance accounted for by the different independent variables. Often, though not always, we are given just a test statistic and a p-value. Specific examples:

271: Is 14% at p<0.001 a lot? What is the effect size and how does it compare to other such studies?

299: This paragraph describes some incidental effects, but solely by their p-values. What is the magnitude of the difference in, for example “increased biomass in PK+ plots relative to controls”?

316: This paragraph leaves out even the p-values, let alone an r2. A glance at Figure 2 suggests the magnitude of this effect is small. The slope and variance accounted for should be accessible to the reader in the main text.

334: OK, these scatterplots are statistically significant. What are their r2s? How important is change in diversity to these dependent variables? How much does a decline of 1 species on average shape biomass and variation?

A more quantitative paper that describes the results in terms of magnitudes (r2s, Effect Sizes, comparisons of means) in the main text would allow the reader to better judge, not just how unlikely a result is, but its importance. Likewise, an explicit comparison of these results to others in the literature (including previous NutNet work) using similar metrics (r2s, Effect Sizes, comparisons of means) would go a long way to giving the reader context for these results.

Notes:
How much of the nutrients increasing NPP has been published by the NutNet group before? Are these results new, or updates from early NutNet publications on enrichment and diversity?