Assessing long-term nutrient and lime enrichment effects on a subtropical South African grassland

Nutrient enrichment influences grassland ecosystem structure, typically manifested by reduced species richness and increased productivity. Quantifying the long-term impacts of nutrient enrichment on grasslands contributes to understanding eutrophication effects on grassland, particularly for grasslands adapted to low soil nutrient status. Overextended time periods, nutrient enrichment may modify soil fertility. The Ukulinga Grassland Nutrient Experiment situated on a natural C4 grassland was set up in 1951 on the Ukulinga research farm, South Africa. Continuously applied treatments on plots measuring 2.7 × 9 m include combinations of nitrogen at 0, 7, 14 and 21 g m−2 per annum, phosphorus at 0 and 2.8 g m−2 per annum and lime at 0 and 225 g m−2 applied every five years. Nitrogen sources included ammonium sulphate (acidifying) and limestone ammonium nitrate (less acidifying). Grass species composition was influenced by both nitrogen forms. In contrast, forb species composition was more sensitive to LAN. We found evidence of nitrogen limitation on aboveground net-primary production. However, species richness (for grass and forbs) declined with increasing ammonium sulphate levels, owing to increased soil acidity and high aluminium concentrations. Aluminium toxicity can affect overall species composition by replacing Al-sensitive species with Al-tolerant species on site.

Human activities result in excess amounts of nutrients reaching natural environments, mainly through fertilisation and atmospheric depositions associated with industries (Peñuelas et al. 2013). A continued increase in nutrient enrichment in natural environments threatens biodiversity and ecosystem services (Sala et al. 2000;Tilman et al. 2001;Bobbink et al. 2010;Clark et al. 2017). The main causes of this are the altered competitive relationships in highly productive environments, which exclude species that are incapable of competing for resources, such as light (Harpole and Tilman 2007;DeMalach and Kadmon 2017). Furthermore, increased availability of nutrients also affects belowground growth and competition (Rajaniemi 2003) and increases litter, thus lowering the success of seedling establishment . In addition, studies show that long-term nitrogen addition affects seed germination and soil seed banks (Xia and Wan 2013;Basto et al. 2015a).
It is well known that, apart from water, nitrogen (N) and phosphorus (P) are the two most limiting nutrients in grassland productivity (Fay et al. 2015;Harpole et al. 2017). These nutrients strongly influence productivity, as well as species composition and diversity (Ceulemans et al. 2013;Avolio et al. 2014;Soons et al. 2017;DeMalach 2018;Riesch et al. 2018). Limitations of N and P affect plant groups differently. For example, N deficiency suppresses the development of grasses, whereas P deficiency curbs that of legume and forb species (Lambers et al. 2011;Kidd et al. 2017;Zarzycki and Kopeć 2020). N is reported to be the most limiting in terms of productivity in terrestrial ecosystems (Vitousek and Howarth 1991;Bobbink et al. 2010;Fay et al. 2015). Therefore, understanding the factors that limiting and colimiting nutrients have on species composition and diversity in grassland ecosystems is important. This will allow the development of strategies to counteract these potential changes caused by nutrient enrichment Schelfhout et al. 2017). Research indicates that high N supply rates negatively affect species diversity, whereas et al. 2009). Therefore, the use of long-term studies, such as the Ukulinga Grassland Nutrient Experiment (UGNE), to evaluate these changes is pertinent to understand the relationships that occur.
One of these widely discussed relationships is the plant productivity-plant richness (PRR) relationship. The main findings regarding this relationship are as follows: (1) a monotonic relationship, showing that there is either a negative or positive relationship between the number of species and biomass (Roscher et al. 2005;Balvanera et al. 2006;Duffy et al. 2017), (2) a non-significant hump-shaped/monotonic relationship (Waide et al. 1999;Mittelbach et al. 2001;Gillman et al. 2006) and (3) a hump-shaped relationship, whereby richness increases and then decreases with a rise in productivity (Grace et al. 2014). To date, research published on the UGNE demonstrates a significantly negative relationship between species richness and biomass (Ward et al. 2017). This is opposite to the most cited pattern of the PRR that shows a positive/unimodal relationship (Ward et al. 2017). Therefore, additional clarification on the current relationship between these two variables at the UGNE is necessary.
Previous studies have shown that soil pH influences species richness at the UGNE Ward et al. 2017). This is linked to nitrogen form applied, with ammonium sulphate lowering the soil pH and having a stronger negative effect on forb richness than grass richness . Liming reduced the negative effects of N enrichment, but less so for ammonium sulphate relative to limestone ammonium nitrate Silvertown et al. 2006). Future studies need to include not only N form but also N application rate, which is less emphasised. Identifying a pH threshold where species richness does not decline any further below a certain threshold is also beneficial in understanding the relationship between soil pH and species richness. A meta-analysis showed a threshold of pH 4.9 for subtropical grasslands and savanna (Azevedo et al. 2013). A previous study at UGNE indicated a threshold pH of approximately 4.5 (Ward et al. 2017). Soil pH is identified as an indicator of level of acidity in soil and has an important role in predicting plant presence and structure mainly because it is correlated to various soil nutrients and acidifying pollutants, such as sulphur and aluminium (Zelm et al. 2007;Peppler-Lisbach and Kleyer 2009;Van Kozlov and Zvereva 2011). Acidic soils, also referred to as ultisols/oxisols, are common in the subtropics and tropics and they generally have pH levels lower than 5.5 (Sade et al. 2016). Acidic soils occur naturally or through anthropogenic processes, such as the use of excessive fertilisers and the burning of fossil fuels that release nitrogen and sulphur dioxide that undergo further processes resulting in acid rain (Bojórquez-Quintal et al. 2017). Acid rain contributes towards acidifying the soil and such soils are characterised of consisting of high levels of certain metals, such as iron (Fe) and aluminium (Al) (Gupta et al. 2013). Aluminium is a metal that is known to be toxic to plants (Sade et al. 2016) and has no particular biological role in the soil (Poschenrieder et al. 2008). In highly acidic soil, Al is abundant in its trivalent (Al 3+ ) form and can pose a serious threat to production by inhibiting root growth (Kopittke and Blamey 2016). This supports an investigation into the effect of soil pH and aluminium soil content on species richness at the UGNE, because this may potentially be contributing towards low species richness observed in acidic soils.
In this paper, we present the findings of an experiment initiated in 1951 that involved a combination of nutrient enrichment strategies, with two forms of nitrogen (limestone ammonium nitrate and ammonium sulphate), phosphate and lime at varying application levels and rates. The experiment was first established at a time when applying nutrients to increase forage production from natural grassland was of interest (Morris and Fynn 2001). Nutrient application to natural grassland is no longer commonly used in South Africa. However, the 70-year-old experiment allows for important ecological questions to be addressed. We asked the following questions: 1. Are there any differences in natural grassland species composition, above-ground biomass, species richness, Shannon-Wiener diversity index (H') and Pielou's evenness (J'), including comparisons within and between grasses and forbs following 70 years of nutrient and lime enrichment? 2. What is the current relationship between aboveground biomass and total richness? 3. What is the relationship between soil pH and (a) grass richness, (b) forb richness and (c) total richness? 4. Is there a relationship between soil pH and aluminium soil content?

Study site
The UGNE is located at Ukulinga Research Farm, University of KwaZulu-Natal, Pietermaritzburg, South Africa (29°24' E, 30°24' S). The long-term experiment is located at an altitude ranging from 838 to 847 m above sea level (Ward et al. 2017). The mean annual precipitation and temperature of the area is approximately 838 mm and 18°C, respectively (Ward et al. 2020). The soil is known to be derived from shales under the Westleigh classification (Soil Classification Working Group 1991). In addition, the soils are relatively infertile and considered to be acidic (Soil Classification Working Group 1991). The vegetation of the UGNE and surrounding area is classified as KwaZulu-Natal Hinterland Thornveld of Sub-Escarpment Savanna (Mucina and Rutherford 2006  ) and grazing has been excluded since 1951.

Experimental design and nutrient enrichment
The UGNE involves the manipulation of nitrogen (N), phosphorus (P) and lime. There are 96 plots and each plot is 9.0 × 2.7 m in size with a 1 m spacing between plots. The experiment was replicated in three blocks, each block containing 32 plots, resulting in a 4 × 2 3 factorial design. Firstly, the two forms of nitrogen applied were limestone ammonium nitrate and ammonium sulphate (henceforth LAN and AS, respectively). Four levels of N fertiliser were applied annually on the plots (0 (control), 7.1, 14.1, and 21.2 g m −2 ) for both LAN and ASU. In addition, the N treatments were applied either alone or in combination with P and L Both LAN and ASU were applied twice a year in October and December per plot. Secondly, in terms of P addition, the application was in the form of super-phosphate at two levels (0 (control) and 33.6 g m −2 ). Phosphorus was applied once a year in October. Thirdly, the lime treatments were applied every five years at two levels (0 (control) and 225 g m −2 ) (Le Roux and Mentis 1986;Tsvuura and Kirkman 2013), with the last application being in 2016.

Sampling
All the data used in this study were from one growing season (2019). Soil samples were collected from a depth ranging from 0 to 15 cm in each plot. The topsoil region is reported to be the most active site where fine roots absorb nutrients (Xia and Wan 2013;Basto et al. 2015b;Li et al. 2017). Two samples were collected per plot and sent to the Institute for Commercial Forestry Research in Pietermaritzburg for comprehensive soil analyses, including pH (KCl) and aluminium (µmol g −1 ). Aboveground net-primary production (ANPP) was measured at the end of the growing season in March 2019. We used ANPP here for ease of comparison with previous studies on the same experiment. Aboveground net-primary production was estimated in each plot by mowing a 2.13 m wide strip per plot at the end of the growing season. All plant material present (grasses and forbs) was mown to ground level. The fresh material in each plot was collected and weighed on site. The dry biomass was determined with the use of a grab sample that was dried at 60 °C for two days. Species composition was determined using cover abundance, for each plot. Cover abundance is described as the cover proportion of each plant species present in a quadrat as a percentage (Peratoner and Pötsch 2019). Each plot was sampled in January using four systematically placed quadrats (1 m × 1 m). The species richness, which is the simplest indicator of species diversity was defined as the number of plant species (Keylock 2005)

Shannon-Wiener diversity index (H') calculation:
where, pi = proportional abundance of species i and b = base of logarithm.
Pielou's evenness (J′) calculation: where, H′ = Shannon-Wiener diversity index (H′) and S = total number of species in a sample, across all samples.

Statistical analyses
All statistical analyses were performed using R Studio (version 4.0.2), with packages 'vegan' for analysis and 'ggplot2' for plotting (RStudio team 2020). Prior to performing the canonical correspondence analyses and diversity analyses, we transformed and manipulated the cover plant cover abundance dataset. We used the 'dplyr' and 'tidyverse' packages (Wickhman et al. 2019a;Wickham et al. 2019b) to filter the replicated data and calculate the average values for each species per plot, to avoid pseudoreplication. We used CCA to investigate the relationships between the plant community composition of grasses and forbs separately, with the inclusion of environmental factors. The lengths of the calculated gradients were <4. This permitted an additional analysis of the species composition using a CCA (Jongman et al. 1995). This form of eigenvalue ordination analysis uses the selected environmental matrix to explain the variation in the plant community matrix. In each CCA we separated the categorical factors into their specific levels using the 'ellipses' function in the 'vegan' package. Species occurring in less than 10% of the plots were removed in the forb CCA, to reduce clutter. We used permutation tests to observe whether each model CCA (CCA 'terms' and CCA 'axes') explained more variance of the species abundance matrix than expected by chance. Both grass and forb CCA's had p < 0.05, suggesting that the model was indeed significant. The biplot scores for constraining variables and eigenvalues for constrained axes (with p-values) are shown in Table 1 and Table 2, respectively.
The effects of nutrients (LAN, ASU, and P) and lime on ANPP, grass richness, forb richness, Shannon-Wiener diversity index (H') and Pielou's evenness (J) were assessed using generalised linear models (GLM). This is similar to methods used in other studies Ward et al. 2017). We used a Poisson distribution for richness, a Gaussian distribution for diversity and evenness and a Gamma distribution for ANPP as the error distribution. For each GLM we tested for overdispersion and used diagnostic model misfits' plots for model validation. For overdispersion measures, values greater than 1.1 indicated mild overdispersion and therefore transformations were necessary. The model misfit plots assisted in visualizing, (1) constant variance of fitted values against residuals, (2) the normality of residuals, (3) residuals vs leverage for outliers and (4) scale-location plots. We also conducted Breusch-Pagan tests to check for homoscedasticity. No transformations were required for ANPP, grass richness, diversity, and evenness. However, a box-cox transformation was performed on the forb richness dataset to obtain normality and constant variance. We also performed a simple linear regression of firstly ANPP and total species richness and secondly soil pH and aluminium. Polynomial regressions were performed on (1) soil pH and grass richness and (2) forb richness. The function 'specnumber' was used to calculate the number of species in the R software.

Species composition
Along axis one of the CCA, there is a shift in grass species composition as influenced by N form and levels (Figure 1a; Figure 1b). The species composition changes along axis two are related to P level ( Figure 1c) and to a lesser degree, lime (Figure 1d). The occurrence of bare ground is more associated with plots with high nutrient application level than lower levels. The N applications affected forb species composition differently to grass composition. Limestone ammonium nitrate shifted forb composition along axis one (Figure 2a). This change also resulted in higher variability (Figure 2a). Unlike limestone ammonium nitrate, ammonium sulphate shifted forb species composition along axis two (Figure 2b). Phosphate and lime had a similar effect on forb species composition change along axis two (Figure 2c and Figure 2d).

Aboveground biomass (ANPP)
We tested the effects of LAN, ASU, P, and lime on ANPP. The main effect of LAN significantly affected the ANPP (Table 1a). The two-way interaction between limestone ammonium nitrate and lime was significant (Table 1a). Limestone ammonium nitrate application resulted in peak increase at the lowest level (level 1) and then decreased as LAN level increased ( Figure 3a). In addition, the interaction between LAN and lime was similar to that of LAN level two only ( Figure 3b). Aboveground biomass was separately regressed on total species richness and there was a non-significant relationship (p = 0.10) with a correlation coefficient of −2.121 (Supplementary Figure S1).

Species richness and diversity
Species richness differed greatly between functional groups of grasses and forbs on the UGNE. For grass richness, the values ranged from 1 to 8, whereas forb richness varied greatly, with values ranging from 1 to 22. There was a significant main effect of ASU on grass richness (Table 1b). We found a significant effect of ASU reducing grass species richness at levels two and three (Figure 4a). The GLM indicated significant main effects of LAN, ASU and lime on forb richness (Table 1c). Similarly, to ANPP, there was a significant increase of forb richness at level one, due to limestone ammonium nitrate (Figure 4b). There was a sharp decline in forb richness with increasing ASU level ( Figure 4c). Adding lime increased forb richness at the UGNE (Figure 4d). In the GLM test for the effects of 1) LAN and lime, 2) ASU and lime and 3) P and lime on forb richness, the overall effects were significant (Table 1c). Of these effects, the interaction between ASU and lime showed a clear increase in forb richness with addition of lime ( Figure 5b). Overall, the addition of lime improved the negative effect that LAN, ASU and P had on forb richness. There was also a significant three-way interaction effect of LAN, P and lime on forb richness (Table 1c). This interaction caused forb richness to be similar for the following interactions, 1) LAN, P, and no lime and 2) LAN, P and lime. The interaction with LAN, no P and no lime, resulted in a peak in forb richness at LAN level one, whereas the last interaction of LAN, no P and lime slightly increased forb richness at LAN level three (Figure 5d). This suggests that forb richness at the highest LAN level is increased in the presence of lime (and absence of P).
We tested the effects of limestone ammonium nitrate, ammonium sulphate, phosphate and lime on the grass and forbs species diversity and evenness separately.  Table 1: Generalised linear model analyses on the effects of nutrients. These nutrients are in the following form and levels; limestone ammonium nitrate (0 (control), 7.1, 14.1, and 21.2 g m −2 ), ammonium sulphate (0 (control), 7.1, 14.1, and 21.2 g m −2 ), phosphorus (0 (control), 33.6 g m −2 ) and lime (0 (control), 225 g m −2 ) on (a) Above-ground Net biomass (ANPP), (b) Grass species richness (count) and, (c) forb species richness at the UGNE. * = significant difference observed effects were non-significant for all nutrients and lime for both groups (Table 2). However, there was a marginal significant effect of ASU (p = 0.055 and the interaction between ASU and lime on forb evenness (p = 0.075).

Soil pH and species richness
There was a significantly positive correlation between soil pH and grass species richness (r 2 = 0.30; p < 0.0001), and this relationship was best explained by a polynomial regression. A similar trend was observed for the relationship between soil pH and forb species richness, whereby the polynomial regression (r 2 = 0.50; p < 0.0001) explained the relationship better than a quadratic (r 2 = 0.46; p < 0.001) or simple linear regression (r 2 = 0.26; p < 0.001).
The addition of the two N nutrient forms yielded different responses in terms of richness for both grass and forbs. Limestone ammonium nitrate addition level did not show   Figure 1: A CCA of grass species community composition in 2019 at the UGNE. Axes one and two account for 53.51 and 27.09% of the total variability of the dataset and 57.39 and 29.05% of the variability in the environmental dataset, respectively. The plots shown here are separated into the three nutrient treatments (limestone ammonium nitrate, ammonium sulphate and phosphate) and lime treatment for ease of interpretation. Key: limestone ammonium nitrate and ammonium sulphate: (0 (none); 1= 7.1 g m −2 ; 2 = 14.1 g m −2 and 3 = 21.2 g m −2 ), phosphate: (0 (none); 1 = 33.6 g m −2 ) and lime: (0 (none); 1 = 225 g m −2 ). Full species names are shown in Supplementary Table S3 any clear trend (Figure 6a; Figure 6e). Although the highest ASU level strongly reduced the richness on those plots (Figure 6b; Figure 6f). This trend was also true for total richness. The application of lime contributed to increasing the soil pH and the species richness for both grass and forbs, but this was more prominent for forb species (Figure 6d; Figure 6h). Nutrient enrichment by P did not show any clear trend (Figure 6c; Figure 6g). The grass species occurring in soils with the lowest pH frequently included Panicum maximum Jacq. and Eragrostis curvula (Schrad.) Nees. In these plots, these were commonly the only grass species present. In contrast to the lowest pH soils, control plots had more grass species, including Aristida junciformis Trin. & Rupr. Subsp., Themeda triandra Forssk., Tristachya leucothrix Trin. ex Nees, Hyparrhenia hirta (L.) Stapf and Setaria nigrirostris (Nees) T. Durand & Schinz. Cephalaria pungens Szabó was the most common forb species found in low pH soils. A wide variety of forb species belonging to a range of plant families occurred in control plots.

Soil pH, total richness, and soil aluminium
A regression on total species richness indicated that 54% of the variance in total species richness is explained by soil pH at the UGNE. A threshold of approximately 4.6 (pH) was observed (Figure 7a). Soil pH was a good predictor of aluminium content in the soil at the UGNE. The correlation coefficient between the two variables was −2.81, suggesting that when soil pH increases, the aluminium content in the soil decreases. A hyperbola best explained the relationship (r = −2.814: p < 0.0001) (Figure 7a).
Aluminium ranged from 7.597 (µmol g −1 ) to 49.759 (µmol g −1 ) and pH ranged from 3.205 to 6.840. The two N nutrient forms affected the relationship between  pH and aluminium differently. Limestone ammonium nitrate application level did not have a noticeable effect (Figure 7b), a similar effect was observed for P ( Figure 7b). In contrast, a high-level application of ASU was clearly associated with plots having low pH soils (pH <4, with one exception) and high aluminium content (Figure 7b). Generally, plots with lime had high pH values (pH >5, with some exceptions) and lower aluminium content (Figure 7b). The exceptions referred to here are associated with plots with interacting nutrients having been applied.

Discussion
Previously, the species occurring most frequently at the UGNE were Setaria sphacelata, Eragrostis curvula, Tristachya leucothrix, Themeda triandra, Eragrostis plana and Panicum maximum Tsvuura and Kirkman 2013). These species all accounted for the majority of the herbaceous ANPP Tsvuura and Kirkman 2013). We assessed the species composition separately for grasses and forbs and found differences in how the two functional groups respond to nutrient and lime addition. The latest data reveal that N form affected the two functional groups differently. Grass species composition was influenced by both LAN and ASU. In contrast, forb species composition was mainly influenced by LAN. In both cases, plots with the highest LAN level application had more bare ground than the lower levels. Phosphate and lime enrichment explained less of the species composition for both functional groups. In this experiment, short species (c. <1.5 m in height), such as T. triandra, T. leucothrix, Brachiaria serrata (Thunb.) Stapf (B. serrata) and S. nigrirostris declined in abundance with an increasing N level addition. Similar results were previously observed on the UGNE . These short species were replaced by taller grass species, such as P. maximum, P. aequinerve and C. validus that are known to respond positively to nutrient enrichment . However, in this experiment not all tall grasses (i.e H. hirta and E. curvula) were competitive in nutrient rich plots. This could suggest that other traits  or environmental factors are involved in determining species replacement. For example, grasses such as P. maximum and H. hirta have tall broad or narrows leaves that increase their competitive abilities over shorter species, such as T. triandra in fertile soils ).
In addition, a low specific leaf area is associated with slow growth rates and a reduction in competitive ability in fertile soils (Chapin 1980;MacDougall and Turkington 2004). Furthermore, shade tolerance may explain success in fertilised plots, but other models emphasise that long-term dominance in a fertilised site requires shade tolerance and a tall growth form (Huston and Smith 1987). This supports observations indicating that tall broad-leaved species are the most dominant in productive sites (Gaudet and Keddy 1995;Lepš 1999). The observed pattern of species replacement may also be related to accessibility to light. Nitrogen addition tends to decrease the amount of photosynthetically active radiation reaching the ground, due to a high biomass, thus reducing light accessibility to shorter species (Tsvuura and Kirkman 2013b;Ward et al. 2017). In this experiment, nitrogen was the primary limiting nutrient as demonstrated in other studies (Tilman 1987;Vitousek and Howarth 1991;Fynn and O'Connor 2005;LeBauer, D;Treseder 2008;van Dobben et al. 2017). More specifically, limestone ammonium nitrate was the only nitrogen form to result in significant changes in ANPP. When LAN was applied at the lowest level, it resulted in an increase in ANPP relative to the control plots. However, when LAN was applied with lime it resulted in the highest ANPP, this indicates that N and lime are both limiting, but in a hierarchical and interactive manner. Previously, N and P were identified as limiting factors on the same experiment .
Nutrients and lime had a significant effect on grass and forb richness. This negative effect of both N forms on richness had previously been reported on the site (Scott and Booysen 1956;Ward et al. 2017). The decline in species richness with N addition is a common response recorded in other countries (Tilman 1987;Jenkinson et al. 1994;Inouye and Tilman 1995;Shaver et al. 2001;Crawley et al. 2005;Silvertown et al. 2006;Socher et al. 2012). Another study revealed that the effects of fertilisation by acidic ammonium vs limestone ammonium nitrate differed (Ward et al. 2017). We had a similar result, however, there was a slight difference whereby the lowest level of LAN increased forb richness. This increase was also prevalent when LAN was combined with lime. The addition of lime alone or with LAN, ASU or P increased forb richness, but not grass richness. This suggests that forb species may be more sensitive to nutrient and lime enrichment than grass species. This sensitivity can be linked to the effects of soil pH (Crawley et al. 2005;Silvertown et al. 2006;Ward et al. 2017). Liming ameliorated the negative effects of both forms of nitrogen on forb richness, but less so for ammonium sulphate (Fynn and O'Connor 2005). This is consistent with the current study.
Aboveground-net primary productivity was not a good predictor of total species richness. This contrasts a previous study at the UGNE where a negative relationship was observed between species richness and ANPP (Ward et al. 2017). However, the authors indicate that the relationship was only marginally significant (p = 0.056). A negative relationship was also reported in the Park Grass Experiment (Crawley et al. 2005). In an earlier study at a SOIL pH UGNE, a non-significant relationship was observed in one year and a significant relationship the following, suggesting that rainfall variability may be responsible for the differences (Tsvuura and Kirkman 2013). This non-significant relationship between ANPP and richness experimentally demonstrated here contrasts with the commonly recorded pattern of a positive/unimodal relationship (Waide et al. 1999;Mittelbach et al. 2001;Cornwell and Grubb 2003). We expected a negative relationship, based on previous studies assuming interspecific competition for light under high nutrients to be the main cause (Tsvuura and Kirkman 2013). Our results are in support of findings of a review of experiments in 48 sites on five continents (Adler et al. 2011). Some studies indicate that long-term studies are not good predictors of the relationship between ANPP and species richness because they are normally affected by small term perturbations, whereas the pattern of species richness in natural grasslands is influenced by long-term ecological and evolutionary processes, such as dispersal and speciation, respectively (Gough et al. 1994;Henry et al. 2004). Furthermore, the richness of a region reflects the level of plant biomass allowed by a local climate and soil conditions, rather than the cause of increased biomass (Huston 1997). In addition, covarying environmental factors can limit the positive effects of species richness (Ma et al. 2010). This could explain why species richness is weakly correlated with biomass in natural communities (Li et al. 2020) and potentially the UGNE.
A very interesting result in this study was that the two N forms did not affect soil pH and both forb and grass species in a similar manner. Ward et al (2017) previously recorded a significant positive relationship between soil pH and total species richness at the UGNE. We too found a similar positive relationship at the site. In addition, a similar result was found at the Park Grass Experiment (Crawley et al. 2005;Silvertown et al. 2006). Unlike most studies we assessed the effect of pH on forb and grass species richness and found clear differences between the two functional groups. The most noteworthy effects on soil pH were associated with liming (Crawley et al. 2005;Ward et al. 2017) and extreme acidification by ammonium sulphate decreased species numbers Ward et al. 2017).
The most important causes of soil acidification on agricultural land are the application of ammoniumbased fertilisers and urea, elemental sulphur fertiliser and the growth of legumes (Bolan and Hedley 2003). Ammonium salts strongly acidify soils through the process of nitrification (Goulding 2016). Another response of acidic soils is the dissolution of Al minerals that caus concentrations of soluble Al that are rhizotoxic to various plant species (Kopittke and Blamey 2016). In this study, the highest ammonium sulphate application increased the soils acidity (pH <4). This effect would strongly increase the solubility of Al. Kopittke and Blamey (2016) recorded Al toxicity occurring in soils with pH between 4.5 and 5.0. These acidic and high Al content soils are suitable for certain species and this has implications for biodiversity. Aluminium-sensitive species can be replaced by more Al-tolerant species (Sade et al. 2016). For example, in this study, acidic soils were dominated by P. maximum and E. curvula. Both species are considered somewhat Al-tolerant (Miles and De Villiers 1989;Almeida etal. 2000) and nitrophilous (Tsvuura and Kirkman 2013;Noukeu et al. 2019). These advantages and morphological features, such as tall broad leaves, enhance their ability to capture light and outcompete neighbouring species in acidic soils (Wilson and Tilman 1991;Fynn and O'Connor 2005). Cephalaria pungens was the most dominant forb species, however, its dominance has not yet been linked to acidic soils or Al-tolerance, to our knowledge. A strategy to alleviate the negative effects of acidity is liming, because it can provide a resistance against Al toxicity (Sade et al. 2016). However, liming may not be feasible because of costs and in situations where acidity occurs at deep soil layers (Sade et al. 2016). This study shows that plots that have not received nutrients in 70 years had the highest pH, lowest Al content and ultimately more species.

Conclusions
Our findings reveal that changes in species composition, aboveground net-primary productivity, species richness and soil variables have been influenced by nutrients and lime enrichment over the past 70 years. High nutrient conditions may directly influence species compositional shifts from short species like T. triandra and T. leucothrix to more physiologically tolerant species, such as P. maximum.
However, there appears to be other important traits that enhance a plants ability to outcompete neighbouring species. Specifically, limestone ammonium nitrate and lime were considered most limiting in terms of aboveground net-primary productivity. Phosphorus and limestone ammonium nitrate did not influence species richness as strongly as ammonium sulphate. The latter nutrient increased soil acidity, creating a more suitable environment for high aluminium concentrations and reducing both grass and forb species richness. The soil pH threshold for the lower limit for species richness was approximately 4.6, as in previous work (4.5) at UGNE. Aluminium toxicity poses a threat to agricultural production, because it inhibits root growth and ultimately overall plant growth.
Funding information -This research was funded in part by the University of KwaZulu-Natal and the Agricultural Research Council, South Africa.