Results

Figure \ref{fig:Cover_change_example} shows some typical raw data from example quadrats over the course of the experiment. The plots show the decline in the spatial extent and percent cover of the dominant ericaceous shrub, Calluna vulgaris . At 16 m from the NH\({}_{3}\) fumigation source, receiving approximately 7 g N m\({}^{-2}\) y\({}^{-1}\) as dry deposition, the decline was dramatic and rapid, with no living Calluna vulgaris left in the quadrat after five years of treatment. At 32 m from the NH\({}_{3}\) fumigation source, receiving approximately 3 g N m\({}^{-2}\) y\({}^{-1}\), the decline was still marked but less severe, with some living fraction remaining after 13 years.
Figure \ref{fig:dc_vsDose_Calluna_vulgaris} shows the change in cover of Calluna vulgaris in response to nitrogen deposition after 13-14 years of treatment. This shows a clear decline in cover in response to NH\({}_{3}\), with the cover of green Calluna vulgaris reduced to near zero within a few years of treatment commencing with the highest doses, with a similar pattern only slightly delayed in the lower doses. This is reflected in the results of the univariate linear mixed model, with a significant interaction term between time and NH\({}_{3}\) (\(p=0.015\)) in the model (Table 1). The interpretation of this interaction term is that the decline in cover over time was greater at higher levels of NH\({}_{3}\) (by -0.13 % (g N m\({}^{-2}\) y\({}^{-1}\))\({}^{-1}\)). The effects of NH\({}_{4}^{+}\) and NO\({}_{3}^{-}\) are much less clear, although the trends in cover were always negative at the high levels of nitrogen deposition. In terms of statistical significance, the interaction between time and NO\({}_{3}^{-}\) shows borderline significance (\(p=0.046\)) and PK significantly exacerbates this interaction, i.e. the decline over time with NO\({}_{3}^{-}\) is greater in the presence of PK. (\(p=0.018\), Table 1).
Figure \ref{fig:dc_vsDose_Sphagnum_capillifolium} shows the equivalent change in cover of Sphagnum capillifolium . As with Calluna vulgaris , there was a clear decline in cover in response to NH\({}_{3}\), reduced to zero within a six years of treatment commencing with all doses, but showing a slight increase in the control. For NH\({}_{4}^{+}\) and NO\({}_{3}^{-}\) treatments, there is considerably more scatter in the data, but show almost complete loss of Sphagnum capillifolium cover in the high doses, and substantial losses at all doses except the controls. Because the effect of NH\({}_{3}\) was so abrupt, the linear model does not pick up the effect of nitrogen in this decline; since there was no Sphagnum capillifolium after year 6 except in the control plots, there was no further response to nitrogen to detect. However, if the data are truncated to the years where Sphagnum capillifolium was still present in plots, the interaction between NH\({}_{3}\) and time becomes larger (-1 % (g N m\({}^{-2}\) y\({}^{-1}\))\({}^{-1}\))) and borderline statistically significant (\(p\) = 0.06). With NH\({}_{4}^{+}\) and NO\({}_{3}^{-}\), although the data are more scattered, the response was more gradual, and these are picked up as clearer linear effects. Table 2 shows highly significant interactions between time and both NH\({}_{4}^{+}\) and NO\({}_{3}^{-}\) (\(p<0.01\)). Again PK appears to exaggerate the effect, significantly so in the case of NO\({}_{3}^{-}\).
The dominant sedge, Eriophorum vaginatum L., showed a strong positive response to NH\({}_{3}\) (\(p<<0.01\), Table 3), with cover increasing by up to 40 % (Figure \ref{fig:dc_vsDose_Eriophorum_vaginatum}. Cover decreased with nitrogen dose in the NH\({}_{4}^{+}\) and NO\({}_{3}^{-}\) treatments, though this is only significant in the case of NO\({}_{3}^{-}\) (\(p=0.02\), Table 3). The effect of PK was to change the decrease with NH\({}_{4}^{+}\) and NO\({}_{3}^{-}\) into an increase, and this is manifested in the interaction term for time and PK (\(p<0.01\), Table 3).
Cladonia portentosa (Dufour) Coem. is the most common lichen species at the site, and lichens might be expected to be some of the most sensitive to nitrogen deposition. In two of the treatments (NH\({}_{3}\), and NO\({}_{3}^{-}\) with PK), it died out almost completely after four years (Figure \ref{fig:dc_vsDose_Cladonia_portentosa}). In the other treatments, the effect was similar but less extreme. However, despite the visually clear results, this effect is not picked up clearly by the linear model, for the same reasons as discussed above: the effect of time was not linear, as Cladonia portentosa was often already at 0 % within four years, with no subsequent change over the remaining nine years. Also, Cladonia portentosa was generally rarer, and there happened to be none in most control plots. The option of truncating the data to plots and years when it was present left too few observations to give a satisfactory model fit. Although it detects effects of time, and the interaction between time and NH\({}_{3}\) and NH\({}_{4}^{+}\), the univariate linear modelling approach could therefore be misleading, in detecting no other statistically significant effects (Table 4).
The change in cover of Pleurozium schreberi , a moss which is common at the site but not usually dominant is shown in Figure \ref{fig:dc_vsDose_Pleurozium_schreberi}. Here, the decline with nitrogen over time was similar across the treatment forms without PK, and is statistically significant in the NO\({}_{3}^{-}\) treatment (\(p<<\) 0.01, Table 5). The more striking result is the response to PK with intermediate nitrogen addition, where the cover is increased (in absolute terms, by around 50 % of the quadrat area on average). The effect of high nitrogen addition with PK was similar to the other treatments, though with the nonlinearities noted for Cladonia portentosa above. The combination of PK addition without nitrogen addition was not available, so interpolating or extrapolating this relationship is not easy. Figure \ref{fig:dc_vsDose_Hypnum_jutlandicum} shows the change in cover of Hypnum jutlandicum , the only other commonly-occurring moss species in the experimental plots. The clearest response this shows is the decrease in cover over time with NH\({}_{3}\) addition, which the linear modelling shows as statistically significant (\(p<<\) 0.01, Table 6). Responses across the other treatments are variable and equivocal.
Consistent trends with nitrogen were discernible across species. All species showed a decline with NH\({}_{3}\) treatment, except for Eriophorum vaginatum which increased. In the absence of PK, all species declined with NH\({}_{4}^{+}\) and NO\({}_{3}^{-}\), except for Calluna vulgaris and Hypnum jutlandicum . The effect of PK was not consistent across species, being unclear in most cases (Calluna vulgaris , Sphagnum capillifolium and Cladonia portentosa ) but reversing the decline with nitrogen to become an increase (Eriophorum vaginatum ), and enhancing growth at low nitrogen whilst having no effect at high nitrogen (Pleurozium schreberi ).
The multivariate analyses show significant community-level responses. The principal response curves for the treatments show the time-dependent, community-level effects of nitrogen form and dose, with and without addition of PK (Figures \ref{fig:PRC_dry} and \ref{fig:PRC_wet_1x5}). The \(y\) axis shows the treatment-time coefficients \(c_{dt}\) from the first PRC component, representing the multivariate measure of plant species composition which best captures the time-dependent response of the vegetation to treatments. \(c_{dt}\) values are relative to the control, so by definition the control values form a horizontal line on the plot at \(c_{dt}=0\). Figure \ref{fig:PRC_dry} shows a coherent response to the NH\({}_{3}\) treatment, with a clear decline over time, followed by an apparent recovery phase after 2011. The responses closely follow the nitrogen dose, except that the second highest dose produces the greatest response. The right-hand panel depicts the loadings, \(b_{s}\), for each species. These relate to the correlation between cover of each species and the overall response pattern specified by the principal response curves. \(b_{s}\) values near zero indicate that the cover of species \(s\) does not differ between treatments or is uncorrelated with the overall response pattern. This shows that the response to NH\({}_{3}\) over time was characterised by a reduction in Sphagnum capillifolium , Calluna vulgaris , and Pleurozium schreberi , together with an increase in Eriophorum vaginatum (in live, dead and litter forms). In the NH\({}_{3}\) treatment data set, 22 % of the total variance in species cover could be attributed to the treatment groups (including the interaction with time), and 16 % to effects of time per se . Monte Carlo permutation tests indicated that the PRC components were highly significant (\(p<0.01\)).
Figure \ref{fig:PRC_wet_1x5} shows similar responses to the NH\({}_{4}^{+}\) and NO\({}_{3}^{-}\) treatments, with a general decline over time, followed by a possible recovery phase after 2011 in three or more cases. The significant exception is when PK is added at intermediate levels of nitrogen, where community change takes a quite different trajectory. This is dominated by an increase in Pleurozium schreberi . Without PK, the response again follows the nitrogen dose, with the smallest response produced by adding the lowest dose of 1.6 g N m\({}^{-2}\) y\({}^{-1}\), and the largest response with the highest dose (6.4 g N m\({}^{-2}\) y\({}^{-1}\)). The species loadings (right-hand panel) share some similarities with the NH\({}_{3}\) treatment, showing a loss of Sphagnum capillifolium and gains of Eriophorum vaginatum (live dead and litter), dead Calluna vulgaris with nitrogen addition but are somewhat skewed by the PK response of Pleurozium schreberi . 18 % of the total variance in species cover could be attributed to the treatment groups, and 13 % to effects of time. The PRC components were again highly significant.
Figure \ref{fig:PLS} shows the PLS ordination of the samples in relation to the loadings for the plant species (right-hand) and the chemical ions. The first axis is very similar to that identified by PRC on the NH\({}_{3}\) treatment, and differentiates between the quadrats with high NH\({}_{3}\) deposition (with increased Eriophorum vaginatum (live, dead, and litter) and dead Sphagnum capillifolium ), and those with no NH\({}_{3}\) deposition (with increased Calluna vulgaris , Sphagnum capillifolium and Pleurozium schreberi . Negative values on this axis (high NH\({}_{3}\) deposition, high Eriophorum vaginatum ) are associated with high NO\({}_{3}^{-}\) and NH\({}_{4}^{+}\) ion concentrations in the soil water. The second axis appears to differentiate within the NH\({}_{4}^{+}\) treatment, showing quadrats with high doses of NH\({}_{4}\)Cl and high Cl concentrations on the left. However, the interpretation of this in terms of species composition is not very clear. and the magnitude and variability in NO\({}_{3}^{-}\) and NH\({}_{4}^{+}\) ion concentrations was much less in the NH\({}_{4}^{+}\) and NO\({}_{3}^{-}\) treatments. So, PLS can distinguish an axis of vegetation variation which corresponds to high NO\({}_{3}^{-}\) and NH\({}_{4}^{+}\) ion concentrations in the soil water in the NH\({}_{3}\) treatment, but species composition change in the NH\({}_{4}^{+}\) and NO\({}_{3}^{-}\) treatments was not clearly related to soil water chemistry.
There appeared to be some change to the NVC community classification due to nitrogen deposition, in all treatments (Figure \ref{fig:Whim_NVC}), although the goodness-of-fit to NVC classes mostly ranged from poor to fair, with a few plots where this was good. In the control plots, the communities stayed relatively constant through time. In NH\({}_{3}\) treatment plots, transition from M17 and M18 to M19 appears to be driven mostly by the increase in Eriophorum vaginatum and the decrease in Calluna vulgaris . There was weaker evidence of community transition to M15. In the NH\({}_{4}^{+}\) and NO\({}_{3}^{-}\) treatments, there was a general background successional trajectory towards the M19 community from M18, again mainly driven by changes in Eriophorum vaginatum and Calluna vulgaris . In the NH\({}_{4}^{+}\) treatment, increasing the dose shifted plots towards M17, again mostly driven by changes in Eriophorum vaginatum and Calluna vulgaris . This shift towards M17 only generally applied to the plots without PK addition. In a slight contrast in the NO\({}_{3}^{-}\) treatment, the shifts towards M17 also applied to the plots with the PK addition, again mostly driven by changes in Eriophorum vaginatum and Calluna vulgaris . It should be noted that a degree of subjectivity is involved in community classification and there are limitations on results from a single site.