Discussion
The univariate analyses of the six common species showed a clear effect of NH\({}_{3}\) on all of the species. Importantly, we demonstrated an effect on the keystone species Sphagnum capillifolium . The response to NH\({}_{3}\) was in some cases very drastic, so that the effect was nonlinear over the 14-year period - there was no response to nitrogen dose after a species was extinct in a plot. This complicated the analysis, requiring the pre-extinction period to be identified for each plot, and the linear analysis applied to this subset of the data. We detected a clear effect of either NH\({}_{4}^{+}\) or NO\({}_{3}^{-}\) on all of the species considered, except Hypnum jutlandicum where responses were very variable. We observed a dramatic effect on the lichen species expected to be most sensitive to nitrogen, and this was clearly detectable using a linear model, except in the case of NO\({}_{3}^{-}\). Our results showed that some species gained from nitrogen addition, notably Hypnum jutlandicum and Eriophorum vaginatum , although this seemed to vary with nitrogen form. Similar competitive shifts have been observed in other experiments (e.g. with Polytrichum strictum out-competing Sphagnum capillifolium at Mer Bleue \citep{Juutinen2015}). The ramifications of such changes on biodiversity, peat physical structure and accumulation rates (and hence carbon balance) are hard to predict, but potentially far-reaching.
The multivariate analyses also demonstrated significant responses to the treatments. These take advantage of combining the information from all 69 species, and use the correlations present in these data, rather than treating each species as an independent variable. The disadvantage of the approach is usually in the abstract nature of the components produced. However, PRC manages to produce a component which is easily interpretable, and this confirms the trends visible in the univariate analyses and informal observation in the field. This was particularly true in the case of the NH\({}_{3}\) treatment; in the case of the NH\({}_{4}^{+}\) and NO\({}_{3}^{-}\) treatments, the interpretation was less straightforward, because of the strong leverage of Pleurozium schreberi at one end of the axis, which is an effect of PK rather than nitrogen. In the case of PLS, we established an axis which described the multivariate variation in species composition along the NH\({}_{3}\) gradient, and related this to the soil water chemistry (mainly the change in NH\({}_{4}^{+}\) and NO\({}_{3}^{-}\)). In the NH\({}_{4}^{+}\) treatment, although species composition appeared unrelated to NH\({}_{4}^{+}\) and NO\({}_{3}^{-}\), PLS did identify a vegetation axis which corresponded with Na\({}^{+}\) and Cl\({}^{-}\) in the soil solution as passive tracers for nitrogen deposition in the form of NaNO\({}_{3}\) and NH\({}_{4}\)Cl, though the interpretation of this vegetation axis was less clear. Using univariate analysis, no correspondence between species cover and soil water chemistry could be established.
Standard (multivariate) diversity indices were less useful in identifying change (data not shown). There was some evidence for a decrease in Shannon diversity in the NH\({}_{4}^{+}\) and NO\({}_{3}^{-}\) treatments (\(p=0.052\)), but not in the NH\({}_{3}\) treatment. Species evenness did not change with nitrogen dose, but did have a significant interaction with PK in the NH\({}_{4}^{+}\) and NO\({}_{3}^{-}\) treatments.
The transition of NVC communities in relation to nitrogen deposition appears to fit with the PRC results and the species responses shown in Figures 2-7. However, in most cases the fit to NVC community was quite poor even in control plots. This is mainly because NVC communities are intended for broad-scale applications at national scale. They therefore include many more species than can be found in the local species pool. Determining whether these shifts in NVC community are occurring on a wider scale would require extensive sampling across a number of sites.
A major feature of the results to be explained is the greater effect of NH\({}_{3}\) on vegetation species cover, compared to NH\({}_{4}^{+}\) and NO\({}_{3}^{-}\) at the equivalent deposition rate. For higher plants, \citet{Sheppard2011} previously attributed the greater effect of NH\({}_{3}\) to direct uptake via the stomata. This explanation cannot apply exactly to Sphagnum as they do not have true stomata, and gas exchange mainly takes place across the wet leaf surface. However, the basic proposition probably remains true: NH\({}_{3}\) deposition results in higher nitrogen concentrations at the vegetation surface and in the leaf apoplast because it deposits directly on the thin film of water on the leaf surface, without any dilution in rain water or soil water. With NH\({}_{4}^{+}\) and NO\({}_{3}^{-}\), the ions are dissolved in rain water before being sprayed on, and much of this sprayed water runs off the leaf surface (as it will usually exceed the already-saturated moss canopy interception capacity) to mix with the soil water. Partly this difference comes down to the defininition of “deposition rate”. In the case of NH\({}_{3}\) deposition, we can equate the deposition rate that we estimate to the actual addition of nitrogen to the leaf apoplast, where it has its biological effects. In the case of NH\({}_{4}^{+}\) and NO\({}_{3}^{-}\), the deposition rate that we estimate adds mainly to nitrogen in the soil water, and the fraction of this which actually ends up elevating concentrations in the leaf apoplast is much smaller. So, in the former case we have the deposition rate to the leaf apoplast, in the latter we have the deposition rate to the whole ecosystem.
To estimate the effects of the three forms in the real world, we need to factor in the magnitude of their relative deposition rates, as well their relative effects per g N deposited. We did this for Sphagnum capillifolium , as a keystone species, fundamental to the development of the peat bog system. Estimates of wet and dry deposition of NH\({}_{3}\), NH\({}_{4}^{+}\) and NO\({}_{3}^{-}\) were taken from the Concentration-Based Estimation of Deposition (CBED) model \citep{Smith2000a,Smith2001a}, based on observed atmospheric concentrations and in rainfall. The spatial distribution of peat bogs was taken from the CEH Land Cover Map 2015 (Morton et al. 2017) at 1-km resolution. At each location where peat bogs occur, the linear model coefficients for Sphagnum capillifolium, quantifying the interactions between time and NH\({}_{3}\), NH\({}_{4}^{+}\) and NO\({}_{3}^{-}\) (Table 2), were multiplied by the respective deposition of each, to give their separate effects on cover change (Figure \ref{fig:r_dc_Sphcap}). The effects are expected to be additive, so the total response to nitrogen deposition is the sum of the three effects, but this suggests that NH\({}_{4}^{+}\) deposition has the largest impact on Sphagnum capillifolium. This is because NH\({}_{4}^{+}\) and NO\({}_{3}^{-}\) have similar deposition rates on UK peat bogs, averaging 0.44 g N m\({}^{-2}\) y\({}^{-1}\) with maxima of 1.4 g N m\({}^{-2}\) y\({}^{-1}\), but NH\({}_{4}^{+}\) has a greater effect on Sphagnum capillifolium cover (-0.5 versus -0.3 % (g N m\({}^{-2}\) y\({}^{-1}\))\({}^{-1}\)). Based on the coefficients in Table 2, NH\({}_{3}\) has the least impact of the three; the deposition rate is lowest, averaging 0.16 g N m\({}^{-2}\) y\({}^{-1}\) with maximum of 0.6 g N m\({}^{-2}\) y\({}^{-1}\), and its effect on Sphagnum capillifolium cover is similar to NO\({}_{3}^{-}\) (-0.2 % (g N m\({}^{-2}\) y\({}^{-1}\))\({}^{-1}\)). When the model fit with zero values removed is used instead, this coefficient is higher (-1 % (g N m\({}^{-2}\) y\({}^{-1}\))\({}^{-1}\)), and has a bigger overall effect than NO\({}_{3}^{-}\) (at least over the 14-year time scale of the experiment). Given this level of uncertainty in our estimates, we cannot readily distinguish between the effect sizes of NH\({}_{3}\) and NO\({}_{3}^{-}\) deposition. We acknowledge that this is based on a single-site experiment, albeit a relatively long-term study which mimics realistic conditions, and some caution is needed in this extrapolation of results. However, it is effective in placing the importance of the three nitrogen forms in the context of their likely impacts at national scale. The results come from a 14-year study, but the impacts are potentially different over multiple decades of exposure in the real world. Conceivably there are short-term direct impacts of NH\({}_{3}\) and long-term indirect impacts of NH\({}_{4}^{+}\) and NO\({}_{3}^{-}\) (if these gradually accumulate in the soil, and the effects are manifested later through different pathways).
We compare our results with those from the closest comparable experiments in the literature. At Degerö Stormyr, around 50 % of the Sphagnum capillifolium cover was lost with a nitrogen deposition rate of 3 g N m\({}^{-2}\) y\({}^{-1}\) after nine years \citep{Wiedermann2007}. At Mer Bleue, 100 % of the Sphagnum capillifolium cover was lost with a nitrogen deposition rate of 6.4 g N m\({}^{-2}\) y\({}^{-1}\) \citep{Juutinen2010}. The rate and magnitude of the effects we observed was somewhat slower and smaller, typically 30 % reduction in cover at the highest dose of NH\({}_{4}^{+}\) and NO\({}_{3}^{-}\) after 14 years (Figure 3). The most likely reason for this is probably the way in which nitrogen was applied in the Whim experiment, in a large number of small doses. Considering the processes in which nitrogen concentration exerts a physiological effect in mosses, these results are entirely plausible \citep{Bridgham2002,Fritz2014}. The modelling analysis of \citet{Wu2015} discusses this from a more theoretical perspective. They used a process-based model, which included internal dynamics of nitrogen transport from the Hurley pasture model \citep{Thornley1998a}, to simulate the Mer Bleue experiment, applying the same nitrogen doses either in three-weekly or daily applications. The results showed that the effects on moss biomass were more severe when applied three-weekly, to the extent that moss died out completely in the highest simulated treatment when applied three-weekly, but maintained a stable biomass of 120 g m\({}^{-2}\) when applied daily. Our results therefore support the idea that supplying nitrogen in a realistic way produces a lesser or slower response than conventional nitrogen addition experiments. Most other experiments may contain a bias towards over-sensitivity because of the artefactual way in which nitrogen is normally applied.
The results highlight the importance of nitrogen form and concentration in determining the impact on vegetation in bog ecosystems: with the same nitrogen dose, the detrimental effects may be greater or lesser, depending on the nitrogen form and concentrations that the plants are exposed to. Weather phenomena (such as dew, fog, and low cloud) and episodic dry deposition may produce events where high concentrations of nitrogen occur on plant tissues and cause damage at low nitrogen loads. Such events are stochastic in nature, and will tend to make the relationship between nitrogen dose and damage rather variable. The background nitrogen deposition levels might also be a factor in determining the response to nitrogen, as previous exposure may have a de-sensitising effect, or result in adaptive change \citep{Nordin2005}. This is unlikely to be a factor in our experiment, where the historical background has been low.
The abundance of some species genuinely varies from year to year, depending on water table and weather variations, and this provides a source of uncertainty when trying to detect a trend against this background variability. Perhaps a more serious source of uncertainty is the visual assessment of cover, which is prone to subjective factors which make the different species appear more or less abundant to the human eye - e.g. effects of daylight, survey date in relation to flowering time, etc. To try to account for this, we also analysed the data using an alternative measure of abundance, calculated as the fraction of the 16 sub-quadrats per quadrat in which a species occurred. This should give a more conservative measure of change, dependent only on estimates of presence/absence rather than scoring cover as a percentage. The results were not markedly different to those presented here. We also modelled the proportional changes instead of the absolute changes in cover, as it might be argued that a 2 % change from 5 to 3 % has more significance than from 95 to 93 %. This was achieved using a natural log transformation of the cover data. Again, although the coefficients differed, this did not substantially change the results or conclusions drawn. More objective quantities, such as shoot length growth or gravimetric measures of biomass, are less prone to such errors (and have shown more sensitivity in other experiments), but are more difficult to carry out on a large scale to provide an adequate sample, covering all the experimental plots.