Introduction

Deposition of anthropogenic nitrogen (N) has increased since the industrial revolution as a result of fossil fuel burning and agricultural use \citep{Fowler2005}. Peat bog communities are adapted to conditions of very low nitrogen availability, and hence are likely to be sensitive to additional nitrogen inputs \citep{Bobbink1998}. Western Europe, and particularly the UK, have large areas of peatland in relatively close proximity to emission sources of nitrogen pollution, and hence these areas may be particularly vulnerable.
Demonstrating the true long-term effects of nitrogen pollution on vegetation is difficult. Much debate in the literature concerns the validity of short-term field experiments in estimating the long-term response, and the utility of glasshouse experiments for estimating the response of vegetation in the field \citep[e.g.][]{Wiedermann2009a,Armitage2012,Limpens2012,Phoenix2012a}. Several short-term manipulative experiments have demonstrated rapid loss of Sphagnum moss communities when relatively large doses are applied over a short time \citep{Limpens2012}. Results are not altogether consistent, and some studies have shown positive responses or no detectable effect on areal cover \citep{Gunnarsson2000,Saarnio2003}. Relatively few experiments have run for more than five years, but two experiments are of particular note. At Mer Bleue in Canada, treatments of up to 6.4 g N m\({}^{-2}\) y\({}^{-1}\) in NH\({}_{4}\)NO\({}_{3}\) solution were applied for up to 12 years \citep{Bubier2007,Larmola2013}. Fertilization was applied every third week from early May to late August, i.e. five to six applications per year. \citet{Juutinen2010} found that Sphagnum cover decreased rapidly in relation to nitrogen addition. Effects were clear after three years, and Sphagnum remained absent from the higher nitrogen treatments (\(>=\) 3.2 g N m\({}^{-2}\) y\({}^{-1}\)) after 5 years. Dwarf shrubs and the moss Polytrichum strictum Brid. benefitted from the nitrogen addition \citep{Juutinen2015}. At Degerö Stormyr in Sweden, treatments of 3 g N m\({}^{-2}\) y\({}^{-1}\) of NH\({}_{4}\)NO\({}_{3}\) solution were applied for eight years \citep{Wiedermann2007}. One third of the seasonal dose was applied directly after snowmelt in May, with four further monthly applications covering the short growing season \citep{Eriksson2010}. No effects were seen on the vegetation in the first four years, but after eight years Sphagnum cover was reduced from 100 % to 41 %.
A serious issue in almost all experiments is the application of nitrogen in relatively few but large doses (e.g. in monthly watering treatments), thereby exposing the plants to unrealistically high concentrations of nitrogen in solution on the leaf and root surface \citep{Pitcairn2006,Pearce2008a,Wu2015}. This may have toxic effects, unrepresentative of nitrogen deposition in real ecosystems. In observational studies, where trends in peatland vegetation were studied over time in areas with different nitrogen deposition rates, relatively little sensitivity to nitrogen has been seen \citep{Hajkova2011}. The above points suggest that, as well as long-term experiments, consideration needs to be given to the nitrogen concentration experienced at the vegetation surface, as well as the total nitrogen deposition per year \citep[see the modelling analysis of][]{Wu2015}. It is conceivable that nearly all experiments contain the same artefact, and that real-world ecosystems are less sensitive to nitrogen than experiments suggest.
The experiment at Whim bog in central Scotland \citep{Leith2004,Sheppard2004} is a globally unique opportunity to investigate the effect of different forms of nitrogen when applied at realistic rates. Here, nitrogen treatment is applied near-continuously over the year, via automated sprayers mimicking realistic rain events for wet deposition, and as a gas plume for dry deposition. Because the nitrogen is applied as many (\(>\)100) small application events per year, the effect of this is to allow manipulation of the nitrogen deposition whilst maintaining realistic nitrogen concentrations in solution at the leaf and soil surface. This paper reports results from this long-term experiment, where 14 years of treatment data are now available. Nitrogen has been deposited in three different forms (as ammonia (NH\({}_{3}\)) gas, as ammonium (NH\({}_{4}^{+}\)) solution or nitrate (NO\({}_{3}^{-}\)) solution) \citep{Sheppard2004}. Ambient nitrogen inputs at the site are relatively low (0.5-0.8 g N m\({}^{-2}\) y\({}^{-1}\)), and the site had not received obvious damage prior to the experiment, so it is reasonably representative of similar sites across Europe.
Previously, \citet{Sheppard2011} showed that very high doses of NH\({}_{3}\) produced visible damage, mortality and reduced cover of Calluna vulgaris (L.) Hull, but could not detect effects of wet-deposited NH\({}_{4}^{+}\) and NO\({}_{3}^{-}\). Effects on cover change of other species were largely unclear, although trends could be discerned \citep{Sheppard2008,Sheppard2011,Sheppard2014}. Here, we build on this work in several ways. Firstly, we present a longer-term analysis, with six additional years of experimental treatment. Secondly, we use improved statistical methods; \citet{Sheppard2011,Sheppard2014} analysed cover data using ANOVA to identify differences between treatment groups. However, nitrogen deposition rate is a continuous variable, and we are interested in the dose-response relationship of the vegetation, rather than distinguishing groups (as in the ANOVA model). Here, we apply a linear mixed-model approach \citep{Pinheiro2006}. This allows us to treat nitrogen deposition as a continuous variate, accounting appropriately for the correlation in residuals which arises from making repeated measurements on the same locations (quadrats nested within plots, nested within blocks). Thirdly, a phosphorus and potassium addition treatment was included in the experiment, but was excluded from most previous analyses. We include this interaction in our analysis.
Fourthly, we apply multivariate analyses of species cover \citep[principal response curves, PRC, and partial least squares regression, PLS,][]{VandenBrink1999,Mevik2007}, which have greater statistical power in detecting changes in community composition. The advantage of these over univariate methods is that they use all information on all the species present simultaneously, and in doing so they evaluate the effects of pollution at the community level. The main drawback of multivariate methods is that the results are more abstract than those of univariate methods. Results are often presented as biplots, which can be hard to interpret, especially for policy-makers unfamiliar with the approach. Including time dependence in multivariate responses often produces biplots that are too cluttered to allow easy interpretation of the changes in treatment effects over time. The “principal response curve” method \citep[PRC, ][]{VandenBrink1999}, was developed to address these problems. The PRC method focusses on the time-dependent, community-level treatment effects, yielding a “principal response curve” of the community for each treatment. The PLS method allows the multivariate matrix of plant species cover change to be analysed in terms of change in another multivariate matrix, which in this context is the set of variables describing the changes to the below-ground chemical environment.
The aims of this paper were to:
  1. quantify the response of the key species to enhanced nitrogen deposition, using appropriate analysis techniques, including the effects of phosphorus and potassium in modulating the response;
  2. quantify the time-dependent, community-level effects of nitrogen deposition;
  3. relate vegetation change to changes in the below-ground chemical environment.
  4. estimate the likely effect of nitrogen deposition in different formson the key peat-forming Sphagnum species across the UK, based on the results from the long-term experiment.