Ecological thresholds under atmospheric nitrogen deposition for 1200 herbaceous species and 24 communities across the US

Abstract

Nitrogen (N) emissions and atmospheric deposition have increased significantly during the last century and become a stressor for many N-sensitive plant species. Understanding individual and community herbaceous plant species thresholds to atmospheric N deposition can inform emissions reduction policy. Here we present results using Threshold Indicator Taxa Analysis (TITAN) applied to more than 1200 unique plant species and 24 vegetation communities (i.e., alliances) across the United States (US), to assess vulnerability to N deposition. Alliance-level thresholds (change points) for species decreasing in abundance along the gradient ranged from 1.8−14.3 kg N ha⁻¹ yr⁻¹ and tended to be lower in the west than the east, which suggests that eastern communities, where N deposition has been historically higher, may have already lost many sensitive species. For the species that were present in more than one alliance, over half had a variable response to the N deposition gradient, suggesting local factors affect vulnerability. Significant progress has been made during the past 30 years to reduce N emissions, which has reduced the percentage of plots at risk to N deposition from 72% to 35%. Nevertheless, over a third of plots remain at risk, and an average reduction of N deposition of 20% would protect half of the plots where N deposition exceeds community thresholds. Further, the alliance-and species-level change points determined in this study may be used to inform N critical loads.

1. Introduction

Nitrogen (N) deposition is recognized as a key stressor to biodiversity in the United States (US) and globally (Pardo et al., 2011; Butchart et al., 2010; Sala et al., 2000). The mechanisms by which elevated N deposition can influence plants, and subsequently biodiversity, vary widely (Bobbink et al., 1998; Bobbink et al., 2010; Dise 2011), and the interaction between those mechanisms can be complex. For example, elevated N deposition can lead to soil eutrophication, which may promote growth of weedy species and in turn reduce light levels to neighbors (Hautier et al., 2009). As well, increases in foliage nutrient richness in some species can increase pest pressures (Throop and Lerdau, 2004). Soil acidification (Driscoll et al., 2003), can lead to foliar nutrient imbalance and damage from frost (Schaberg et al., 2002). To further complicate the impacts of elevated N deposition to ecosystems, other abiotic factors within the ecosystem, such as soil content and quality, can mediate the effects of N deposition (Clark et al., 2019), meaning that not all ecosystems respond the same way to increased atmospheric N. For example, bog habitats may show an increase in vascular plant composition, while heathlands may show a transition from heather to grass dominance (summarized in Bobbink R and Hettelingh J-P, 2011). While responses and mechanisms differ, there is ample evidence of negative impacts of elevated N deposition to biodiversity that have led to, and continue to lead to, air quality policies regulating atmospheric N.

In the US, atmospheric deposition of N and sulfur (S) has decreased owing to air quality policies enacted since the 1990s (Burns et al., 2011; Lloret and Valiela, 2016). Deposition of oxidized nitrogen (NO_×), predominantly from fossil fuel combustion (Howarth et al., 2002), has decreased across many eastern regions of the US, but trends are relatively flat in the Midwest and West (Lloret and Valiela, 2016). Furthermore, reduced forms of nitrogen (NH_×), predominantly from agricultural sources, have increased in more than half of US States and contributes to the majority of inorganic N deposition nationally (Li et al., 2016). As such, atmospheric deposition of N in many areas remains elevated above levels shown to have negative effects on many sensitive plant species (EPA, 2017b; Pardo et al., 2011). For example, epiphytic lichens were shown to have a negative response to N deposition > 3.1 kg N ha⁻¹ yr⁻¹ in Californian oak and chaparral forests (Fenn et al., 2008), where 40–50% of the forests received deposition in excess of this threshold (Fenn et al., 2010). While reducing N deposition to protect the most sensitive species may be challenging, or even unrealistic in many regions, it may be necessary to protect biodiversity in habitats where these species dominate, such as tundra heath (Gordon et al., 2001) and alpine grasslands. Similarly, herbaceous species in the US have demonstrated negative responses to N at deposition levels ranging from 3.2–17.6 kg N ha⁻¹ yr⁻¹ (Clark et al. 2019), as has overall plant species richness in open (8.7 kg N ha⁻¹ yr⁻¹) and closed canopy (13.4 kg N ha⁻¹ yr⁻¹) ecosystems (Simkin et al. 2016); N deposition exceeds these values in many areas.

One policy-relevant tool that scientists have developed for assessing vulnerability to atmospheric deposition is ‘critical loads’ (De Vries et al. 2015, Blett et al. 2014), which are defined as “quantitative estimates of exposure to one or more pollutants below which significant harmful effects on specified sensitive elements of the environment do not occur according to present knowledge” (Nilsson and Grennfelt, 1988). Simply put, they are quantitative thresholds where specified harmful effects begin to occur. Critical loads are currently used extensively across Europe (Bobbink and Hettelingh, 2011) and are implemented under the UNECE and the European Union to assess compliance with transboundary transport of air pollutants among countries under the Air Convention (ECE 2013) and National Emissions Ceiling Directive (EU 2016), respectively (de Vries et al., 2015). Within the US, critical loads are being used by the US Environmental Protection Agency (EPA) to help inform the current review of the secondary standards under the National Ambient Air Quality Standards (NAAQS, EPA 2017a), and are currently used on an ad hoc basis in National Parks (URL: www.nps.gov/subjects/air/critical-loads.htm) and National Forests (McDonnell et al. 2019; URL: www.srs.fs.usda.gov/airqualityportal/critical_loads) to assess air quality effects (McDonnell et al. 2019). While critical loads for acidification and eutrophication were once typically based on simple mass balance models, with the later addition of empirical studies included for critical loads of N, more recently there has been an interest in moving toward biodiversity-based critical loads (Hettelingh et al., 2017), which often consider changes in plant species diversity within habitats using probability of species occurrence.

Critical loads are typically calculated at the regional, biome, or habitat-level in the US and Europe (Bobbink et al., 2003; Bobbink and Hettelingh, 2011; Pardo et al., 2011). However, it has been recognized that there is wide variation in species-level vulnerability to N deposition for trees (Horn et al. 2018), and for herbs (e.g., UK: Payne et al., 2013; Ireland: Wilkins et al., 2016; and the US: Clark et al., 2019) across habitats. Although critical loads are commonly targeted at protecting the structure and function of a given habitat, variation in species-level vulnerability can also be considered, particularly if there is an intent to protect specific species for cultural or conservational reasons. In the US, Clark et al. (2019) examined vulnerability of herbaceous species to N and S deposition by evaluating the association of deposition and probability of species occurrence in 15,136 plots. They found a wide range of response to N deposition among species (e.g., from 3.2 kg N ha⁻¹ yr⁻¹ for Cirsium arvense to 17.6 kg N ha⁻¹ yr⁻¹ for Solidago canadensis), and within species across their range (e.g., 80% of species had responses that varied by >2 kg N ha⁻¹ yr⁻¹ across their range). Although the information from Clark et al. (2019) provided detailed information on plant species probability of occurrence to deposition and various co-factors, only 198 of roughly 4000 species were able to be confidently assessed (or roughly 5%) because of the data-intensive approach. Many species were too rare to be assessed, and rare species are often important from a conservation and natural heritage perspective.

Here we advance the work from Clark et al. (2019), that explored the response of forb and grass species to increasing N deposition, and Simkin et al. (2016), that explored changes in species richness, by determining the point along the N deposition gradient at which herbaceous species in the US show the most abrupt change in abundance. We used Threshold Indicator Taxa Analysis (TITAN), which has previously been used in Europe to assess the response of plant communities to environmental stressors (Payne et al., 2020; Payne et al., 2013; Sabatini et al., 2019; Wilkins et al., 2016; Wilkins and Aherne, 2016) and in North America for ectomycorrhizal fungi (Steidinger et al., 2020). Using the dataset of forb and grass coverage at plots across the US, previously used by both Simkin et al. (2016) and Clark et al. (2019), we applied TITAN to identify the point along the N deposition gradient at which the rate of change in abundance of those species was maximized (Baker and King 2010). TITAN identifies, separately by species, the univariate change point along the N deposition gradient at which any systematic difference in abundance is maximized, whether the difference in cover on either side of this change point is statistically significant, and, if so, the abruptness of the difference. This approach enabled us to increase the number of species examined from 198 in Clark et al. (2019) to more than 1200. Further, TITAN evaluates the multiple change points of the species assemblage within a community to infer a community-level change point. This community change point provides information about the community-level response to N deposition accounting for the variable response within a species across their range. These results greatly expand our understanding of species- and community-level vulnerability to N deposition in the US and could be used to develop empirical critical loads to atmospheric N deposition.

2. Methods

2.1. Data sources

The vegetation abundance and N deposition data used in our analysis were obtained from Simkin et al. (2016). Vegetation data was compiled from twelve sources; to be included in the Simkin dataset, sampling methods met the following criteria: vegetation plots were 100–700 m², were sampled after 1989 (up to 2012), included a complete inventory of species from graminoid and forb functional groups, and had percent cover measurements. Subspecies and variants were grouped so that all abundance results were given to the species rank. All plots were classed by vegetation community into ‘alliances’ as defined by the National Vegetation Classification (USNVC 2017), either at the time of sampling or when the data were compiled. Alliances are distinguished by vegetation and site characteristics, and key diagnostic species, but a given species can be present in any number of alliances. From the full Simkin dataset (n=15,136), 12,225 vegetation plots were retained for our analysis, representing 26 unique alliances (Figure 1). The alliances represented in the dataset were predominantly from forested habitats, with a limited number of herbaceous and shrubland habitats.

Figure 1:

Vegetation plot locations (n = 12,225) across the United States (Simkin et al. 2016), coloured by vegetation alliance. The vegetation alliance names are given as a short form in the legend; see Table 1 for full alliance names.

In Clark et al. (2019), species were only included if they occurred at five or more plots in a given alliance and encompassed a deposition range of at least 7 kg N ha⁻¹ yr⁻¹. In our analysis, plant species were included if they occurred at 10 or more plots for a given alliance with no additional restriction on the length of the deposition gradient. Out of 26 alliances, two were not analyzed as they either had very few sites (Quercus virginiana - Sabal palmetto forest alliance, n sites = 16) or very few species (Symphoricarpos albus Shrubland Alliance, n species = 6), and were unlikely to produce change point results. Alliances with 20–30 plots were analyzed but did not yield significant results. In contrast, the Quercus alba - (Quercus rubra, Carya spp.) forest alliance consisted of 3557 plots and was trimmed to a random selection of 2000 plots owing to computational constraints. For this alliance, only plants occurring at 50 or more plots were included in the analysis.

Positive indicator species (PI) are those that are considered representative of a given alliance from the USNVC description of the alliance and were flagged in the dataset. Invasive species (In), as well as those considered threatened or endangered (USDA, NRCS 2020) were also flagged. There were 2489 unique species identified across all of the vegetation plots included in this analysis; only six were classified as endangered, and another four as threatened, but all of these species were trimmed from the analysis as they did not meet the minimum occurrence criteria (i.e., >= 10) within a given alliance. Species that were neither PI nor In for a given alliance were assumed to be ‘neutral species’. Neutral species’ ranges may overlap multiple alliances and they may be PI for a different alliance.

Total (wet and dry) N deposition from the Simkin dataset represented long-term annual averages estimated by combining Community Multiscale Air Quality (CMAQ) 2002–2012 modelled dry deposition estimates to the National Atmospheric Deposition Program (NADP) 1985–2011 observation-based wet deposition estimates (see Simkin et al., 2016 for further details), and will be referred to here as ‘long-term N deposition’. Although plant communities can respond on shorter time scales (Bowman et al., 2006; Clark and Tilman, 2008), in the current study long-term N deposition was deemed to be appropriate given the unknown lag in ecosystem response and span of years where vegetation plots were sampled.

2.2. Plant species abundance change points

We used the statistical tool Threshold Indicator Taxon Analysis (TITAN; version 2; Baker and King, 2010) to detect abrupt changes in species’ distribution along the N deposition gradient, and to identify the location (point) along the gradient of those changes. Further, we used TITAN to detect potential ecological thresholds to N deposition at the alliance level. TITAN uses indicator species scores (i.e., IndVal scores; Dufrene and Legendre, 1997) as a metric that includes both abundance and relative frequency, and assesses the indicator species scores using non-parametric change-point analysis to determine where changes in a species’ distribution is maximized. The individual species change points are compared for a given community (in this case, vegetation alliance), and a community-level ecological threshold is identified at the point along the gradient where the most synchrony in individual species change points is observed.

The analysis was run separately for each alliance, as TITAN is univariate and grouping sites by alliance helps to minimize confounding variables. For each alliance, TITAN first determines whether each species in that alliance is increasing or decreasing in abundance as the N deposition gradient increases, and groups the species as ‘decreasers’ or ‘increasers’ accordingly. Then, for each species, TITAN identifies the location along the N deposition gradient where the rate of change in abundance is maximized. These individual species change points are referred to as z− change points for the decreasers and z+ change points for the increasers. Further, for each group (decreasers and increasers), a community-level decreasing (z−) change point and a community-level increasing (z+) change point is determined, based on the location along the gradient where the individual taxa change points synchronize (i.e. the maximum sum of the z scores at each candidate change point). So, for each alliance, there can be two community-level change points: one representing the location along the gradient with the greatest rate of change for taxa decreasing in abundance, and one representing the location along the gradient with the greatest rate of change for taxa increasing in abundance.

TITAN’s robustness comes from using bootstrapped resampling to estimate the confidence of the change point through metrics of purity and reliability (Baker and King 2010). Purity is defined as the percentage of bootstrap replicate results that agree with observed response direction (i.e., increasing or decreasing abundance). Reliability is defined as the percentage of bootstrap replicates that have a response magnitude at a given change point that differs significantly from what would be expected from random permutation.

We applied TITAN to species abundance datasets for 24 vegetation alliances against a long-term average annual N deposition gradient. We set the number of bootstrap replicates to 500 and both purity and reliability criteria to 95%, as recommended by Baker and King (2010). As purity and reliability measures represent certainty in the results, this meant that we only accepted individual species change points with bootstrap replication results that were pure and reliable 95% of the time. Community-level change points were based only on species that met that level of certainty. Plant species can show a unimodal response to nutrients, but TITAN does not reliably characterize unimodal relationships since unimodal abundance distributions do not meet purity criteria (Baker et al. 2020). As such, plant species that had a strong unimodal response (increasing and decreasing abundance) across the observed N deposition within a given alliance were potentially not captured in our analysis.

We summarized our community change point results to the ecoregion level for comparison to previous studies conducted in the US (i.e., Simkin et al., 2016; Pardo et al., 2011). The sites within each alliance were assigned the community change point of the alliance, and then the sites (independent of alliance) were grouped by ecoregion. The range and mean of the community change points within each ecoregion were determined to provide a cursory (non-statistical) comparison with the thresholds presented in Pardo et al. (2011) and Simkin et al. (2016). It is notable that the mean is the arithmetic average of all community change point plots within each ecoregion; however, it is not a weighted value or representative of the habitat cover within each ecoregion, but simply represents the average of the plots with available data.

2.3. Nitrogen deposition in excess of the community change point

To determine which plots received N deposition in excess of their alliance-specific community change points, we used simple exceedance calculations, i.e., by subtracting the community change point from N deposition (excess N = deposition – community change point). We used two different summaries of total deposition to address different research questions. First, the community change point was subtracted from the long-term annual N deposition (i.e., from Simkin et al., 2016), which provided an estimate of risk to changes in community abundance under long-term average conditions. Second, the community change point was subtracted from the most recent available three-year (2016–2018) annual average total N deposition estimate from the NADP’s total deposition (TDEP) maps (Schwede and Lear, 2014; URL: https://nadp2.slh.wisc.edu/ntn/maps.aspx) at each plot. This provided an assessment of risk of change in species abundance in vegetation alliances under current conditions.

For the decreasing (z−) community change points, an excess (positive value) indicates that N deposition was higher than the change point, and therefore the site may be at risk of losing sensitive species due to N deposition. A deficit (negative value), on the other hand, indicates that the N deposition was below the community change point, and so the plot is less likely at risk of losing species. For the increasing (z+) community change points, the reverse is true (i.e., an excess (positive value) means that species are being gained while a deficit means that they are not).

For the sites where the three-year (2016–2018) TDEP was in excess of the z− community change points, the percent excess was calculated (i.e., % excess = TDEP in excess of community change point/TDEP*100). The percent excess was averaged across the sites to estimate the average percent reduction needed so that N deposition remains below the z− community change point countrywide.

3. Results

3.1. Community (alliance) change points

For the 24 vegetation alliances assessed in this study, the number of plots analyzed ranged from 25 to 1998, and the number of species occurring at 10 or more plots within an alliance ranged from 6 to 426 (Table 1). Long-term average N deposition (1985–2012) across all plots ranged from 1.3 kg N ha⁻¹ yr⁻¹ to 18.8 kg N ha⁻¹ yr⁻¹ (Table 1). Community (alliance) change points were determined for 23 vegetation alliances (Table 2, Figure 2). No pure and reliable individual species change points could be determined for the Betula papyrifera Forest Alliance, and so a community change points could not be determined. Species within that alliance may have change points across the N deposition gradient, but they were not captured in this analysis, possibly because there were few plots to assess (n = 28).

Table 1:

Summary statistics and TITAN results for National Vegetation Classification alliances, showing the number of plots, the number of positive indicators (PI) listed for each alliance, the number of species occurring more than 10 times and how many of those were PIs, and the range in long-term average nitrogen deposition.

USNVC Alliance (total number of PIs)	n plots	n species >10 (PI)	N Deposition Range (kg N ha−1 yr−1)
Forest and Woodland Habitats
Abies lasiocarpa - Picea engelmannii Forest Alliance (33)	355	86 (6)	1.7–6.0
Acer saccharum - Betula alleghaniensis - (Fagus grandifolia) Forest Alliance (15)	953	293 (6)	4.0–15.6
Acer saccharum - Tilia americana - (Quercus rubra) Forest Alliance (29)	556	146 (24)	4.7–14.4
Betula papyrifera Forest Alliance (13)	28	7 (4)	4.3–11.4
Fagus grandifolia - Quercus rubra - Quercus alba Forest Alliance (26)	101	48 (12)	7.0–18.8
Liriodendron tulipifera Forest Alliance (25)	38	22 (6)	8.4–15.6
Pinus contorta Forest Alliance (29)	613	178 (9)	1.4–4.8
Pinus palustris Woodland Alliance (12)	352	287 (18¥)	5.6–15.1
Pinus ponderosa Forest Alliance (48)	251	57 (9)	1.5–6.0
Pinus strobus Forest Alliance (21)	344	184 (4)	4.1–14.4
Pinus taeda Forest Alliance (13)	684	379 (1)	6.6–18.6
Populus tremuloides Forest Alliance (30)	681	166 (8)	1.3–12.9
Pseudotsuga menziesii Forest Alliance (47)	344	167 (18)	1.4–7.5
Pseudotsuga menziesii Giant Forest Alliance (41)	118	21 (8)	1.4–5.6
Pseudotsuga menziesii Woodland Alliance (47)	654	100 (12)	1.5–4.4
Quercus alba - (Quercus rubra, Carya spp.) Forest Alliance (25)	1998	314* (8)	6.6–18.2
Quercus prinus - Quercus (alba, falcata, rubra, velutina) Forest Alliance (29)	65	28 (7)	7.0–15.3
Quercus spp. - Pinus (rigida, echinata) Forest Alliance (23)	1011	426 (9)	7.5–17.9
Thuja occidentalis Forest Alliance (22)	35	24 (10)	4.2–9.1
Tilia americana - Fraxinus americana - (Acer saccharum) Woodland Alliance (29)	121	89 (15)	4.7–14.1
Shrubland Habitats
Artemisia tridentata Shrubland Alliance (27)	332	76 (13)	1.3–4.6
Grassland Habitats
Andropogon gerardii - (Sorghastrum nutans) Herbaceous Alliance (17)	236	156 (11)	6.4–14.7
Festuca idahoensis Herbaceous Alliance (31)	330	127 (14)	1.8–4.6
Schizachyrium scoparium - Bouteloua curtipendula Herbaceous Alliance (10)	323	142 (7)	6.5–14.3

^*occurring >50 times;

^¥the number of present indicators is greater than possible indicators because the value includes multiple species of a taxa that was only defined to the genus rank in the indicator species list.

Table 2:

The number of species decreasing in abundance (z– species), the number of species increasing in abundance (z+ species), and the decreasing and increasing community-level change points (z– community change point and z+ community change point, respectively; units are kg N ha⁻¹ yr⁻¹ and showing 0.05−0.95 quantiles in brackets) for each alliance. PI = number of positive indicator species; In = number of invasive species.

USNVC Alliance	n z- species (PI|In)	n z+ species (PI|In)	z- community change point	z+ community change point
Forest and Woodland Habitats
Abies lasiocarpa - Picea engelmannii Forest Alliance	16 (0|1)	26 (2|2)	2.0 (2.0–2.8)	4.3 (3.7–4.4)
Acer saccharum - Betula alleghaniensis - (Fagus grandifolia) Forest Alliance	107 (5|13)	75 (0|4)	8.0 (7.3–8.9)	8.9 (8.8–10.7)
Acer saccharum - Tilia americana - (Quercus rubra) Forest Alliance	34 (3|1)	60 (12|4)	9.2 (8.5–9.8)	10.9 (9.7–11.0)
Betula papyrifera Forest Alliance	0	0	ND	ND
Fagus grandifolia - Quercus rubra - Quercus alba Forest Alliance	4 (1|0)	9 (1|1)	14.3‡ (10.2–16.0)	16.2 (15.3–17.4)
Liriodendron tulipifera Forest Alliance	2 (0|0)	0	12.1‡ (10.2–12.2)	ND
Pinus contorta Forest Alliance	21 (2|6)	68 (2|1)	2.8 (2.4–2.9)	3.4 (3.4–4.2)
Pinus palustris Woodland Alliance	196 (15|7)	32 (1|3)	9.7 (9.3–9.7)	12.0 (9.6–12.3)
Pinus ponderosa Forest Alliance	13 (3|2)	15 (1|2)	2.1 (1.9–2.3)	4.0 (3.6–5.1)
Pinus strobus Forest Alliance	45 (0|4)	81 (3|2)	8.7 (8.7–9.0)	10.4 (9.6–11.5)
Pinus taeda Forest Alliance	144 (0|10)	42 (0|4)	8.8 (8.0–9.4)	11.3 (11.2–16.6)
Populus tremuloides Forest Alliance	26 (2|9)	72 (4|13)	3.5 (3.6–4.8)	8.3 (6.9–8.5)
Pseudotsuga menziesii Forest Alliance	31 (7|12)	31 (5|2)	2.8 (2.2–3.5)	4.6 (4.6–4.7)
Pseudotsuga menziesii Giant Forest Alliance	4 (3|1)	4 (1|0)	2.5‡ (1.7–2.6)	2.9‡ (2.5–4.4)
Pseudotsuga menziesii Woodland Alliance	14 (2|1)	57 (4|7)	2.4 (2.2–2.6)	2.6 (2.6–2.8)
Quercus alba - (Quercus rubra, Carya spp.) Forest Alliance	64 (4|3)	68 (0|5)	10.2 (10.2–10.7)	15.4 (14.4–17.4)
Quercus prinus - Quercus (alba, falcata, rubra, velutina) Forest Alliance	8 (1|0)	0	9.8 (9.8–11.7)	ND
Quercus spp. - Pinus (rigida, echinata) Forest Alliance	104 (7|6)	102 (0|9)	9.2 (9.1–9.6)	11.2 (11.1–12.4)
Thuja occidentalis Forest Alliance	0	2 (0|0)	ND	7.0‡ (6.6–7.6)
Tilia americana - Fraxinus americana - (Acer saccharum) Woodland Alliance	43 (4|1)	18 (3|4)	9.4 (8.8–9.7)	11.4 (9.8–11.4)
Shrubland Habitats
Artemisia tridentata Shrubland Alliance	9 (2|4)	45 (6|11)	1.8 (1.7–2.0)	2.5 (2.3–2.7)
Grassland Habitats
Andropogon gerardii - (Sorghastrum nutans) Herbaceous Alliance	39 (1|4)	35 (0|10)	8.0 (6.8–8.6)	10.6 (8.2–11.5)
Festuca idahoensis Herbaceous Alliance	14 (1|5)	43 (10|2)	2.4 (2.4–2.5)	4.2 (2.8–4.3)
Schizachyrium scoparium - Bouteloua curtipendula Herbaceous Alliance	43 (3|7)	23 (1|10)	11.2 (11.2–11.7)	11.8 (11.8–12.1)

^‡community change point(s) based on fewer than five species

Figure 2:

Community (alliance) level change points for decreasing (z− CP; A, B) and increasing (z+ CP; C, D) species, with points scaled by the proportion of species significantly changing in abundance across the nitrogen deposition gradient (A, C) and showing 0.05–0.95 quantile ranges (horizontal lines) based on bootstrap replicates (B, D). The vegetation alliance names are given as a short form; see Table 1 for full alliance names.

The z− community change points across the alliances ranged from 1.8 kg N ha⁻¹ yr⁻¹ (Artemisia tridentata Shrubland Alliance, Figure 3a) to 14.3 kg N ha⁻¹ yr⁻¹ (Fagus grandifolia - Quercus rubra - Quercus alba Forest Alliance, Figure 3b), while the z+ community change points across the alliances ranged from 2.5–16.2 kg N ha⁻¹ yr⁻¹, for the same two alliances, respectively. Community change points were determined from significantly changing individual taxa, that comprised 8–68% of the total taxa assessed in an alliance (Figure 2a, c). The 0.05–0.95 quartile ranges around the change points were all under 5 kg N ha⁻¹ yr⁻¹ (Figure 2b, d), with the exception of the decreasing community change point for the Fagus grandifolia - Quercus rubra - Quercus alba Forest Alliance and the increasing community change point for the Pinus taeda Forest Alliance.

Figure 3:

Individual species change points (filled circles; sized in proportion to z standardized indicator species score, see Baker and King, 2010) and community-level change points (dashed vertical line) along a total nitrogen (N) deposition gradient for the vegetation alliance with the (A) lowest and (B) highest decreasing community-level change points. Plots (C) and (D) are representative of alliances that have a range of N deposition between the decresasing and increasing community change points greater than and less than 2 kg N ha⁻¹ yr⁻¹, respectively. Black = decreasing species; red = increasing species. Horizontal lines represent 5–95% quantiles from the bootstrapped change point distribution. ‡ = Positive indicator species; * = invasive species.

The community change points tended to group into two clusters, for both decreasing and increasing species (Figure 2). In the low-deposition cluster for decreasing species (1.8–2.8 kg N ha⁻¹ yr⁻¹), community change points were based on a lower proportion of decreasing species, while the high-deposition cluster (8.0–14.3 kg N ha⁻¹ yr⁻¹) were based on a high proportion of decreasing species. For the increasing change points, both the low deposition (2.5–4.6 kg N ha⁻¹ yr⁻¹) and high deposition (7.0–16.2 kg N ha⁻¹ yr⁻¹) change point clusters were based on a similar proportion of increasing species.

Twenty alliances demonstrated a community change point for both the decreasers and increasers; the z− community change point was lower than the z+ community change point in all cases, with the difference between the z− community change point and z+ community change point within 2 kg N ha⁻¹ yr⁻¹ in all but 6 alliances (e.g. Figure 3).

There was a spatial trend in the community change points associated with the distribution of alliances (Figure 4). Both z− and z+ community change points were significantly lower in the west compared with the east, using the 100^th meridian as the dividing line (t = 331.41, P < 0.001). One exception to the east–west spatial trend is in Minnesota, where we see the highest to lowest community change points existing in the same state (Figure 4). The community change points in Minnesota showed a north-south trend, with lower community change points in the north and higher community change points in the south.

Figure 4:

Vegetation plot locations showing (A) z− community change points (i.e., for decreasing species) and (B) z+ community change points (i.e., for increasing species) along the nitrogen deposition gradient.

Summarizing the community change points to the Level 1 Ecoregion scale (Table 3) found that Eastern temperate forests had the highest average community change points, with a mean z− community change point of 9.8 kg N ha⁻¹ yr⁻¹ and a mean z+ community change point of 13.7 kg N ha⁻¹ yr⁻¹. Northwestern forested mountains and Temperate sierras had the lowest community change points. Compared to existing critical load suggestions in the US, the ecoregion-summarized community change points were generally lower than the species richness critical loads of Simkin et al. (2016), and within the critical load ranges proposed by Pardo et al. (2011).

Table 3:

Ecoregion level 1 TITAN community change points and critical loads of nitrogen deposition (CLN) from Pardo et al. (2011), and Simkin et al. (2016) for total herbaceous richness (units kg N ha⁻¹ yr⁻¹).

Level 1 ecoregion	CLN (Pardo et al., 2011)	CLN (Simkin et al., 2016)	z– community change point (present study)	z+ community change point (present study)
Northern forests	7–21 (hardwood forest alteration of herbaceous understory)	8.0–18.9	3.5–10.2 (mean=7.5, n=424)	6.7–15.4 (mean=9.8, n=424)
Northwestern forested mountains	4–10 (alpine grassland species composition change)	8.0–19.6	1.8–3.5 (mean=2.4, n=6174)	2.5–8.3 (mean=3.6, n=6174)
Marine west coast forests	No vascular plant CL	10.4–15.0	2.8–2.8 (mean=2.8, n=3)	4.6–4.6 (mean=4.6, n=3)
Eastern temperate forests	<17.5 (hardwood forest declines in species-rich genera)	6.6–19.3	3.5–14.3 (mean=9.8, n=5578)	7.0–16.2 (mean=13.7, n=5578)
Great plains	5–15 (tallgrass prairie community shifts)	8.3–19.6	1.8–11.2 (mean=3.1, n=161)	2.5–15.4 (mean=4.9, n=161)
North American desert	3–8.4 (warm desert decrease of native forbs)	8.3–17.0	1.8–3.5 (mean=2.4, n=74)	2.5–8.3 (mean=3.7, n=74)
Temperate Sierras	No vascular plant CL	8.6–14.8	2.11–2.11 (mean=2.11, n=3)	3.95–3.95 (mean=3.95, n=3)

3.2. Nitrogen deposition in excess of the community-level change points

For communities with species that were decreasing in abundance, long-term average N deposition (i.e., from Simkin et al., 2016) was in excess of the z− community change points for 72% of the study plots (Figure 5a, positive values). The most recent N deposition estimates available for TDEP (2016–2018) were in excess of the z− community change points change point at 35% of the study plots (Figure 5c, positive values). The percent excess for sites where TDEP 2016–2018 was in excess of the z− community change point ranged from 0.01–73.5% (mean = 23.5 %, median 20.2%; Figure S2).

Figure 5:

Excess nitrogen deposition for decreasing community-level change points under (A) the long-term (1985–2012) average annual total N deposition and (C) the TDEP 2016–2018 average annual total nitrogen deposition, and excess nitrogen deposition for increasing community-level change points under (B) the long-term (1985–2012) average annual total N deposition and (D) the TDEP 2016–2018 average total deposition. Positive values represent a potential change in the community (i.e., either a decrease in z– species or an increase in z+ species).

For communities with species that were increasing in abundance, long-term N deposition was in excess of the z+ community change point at 23% of the plots (Figure 5b; positive values), and current N deposition was in excess at 12% of the plots (Figure 5d; positive values).

3.3. Species-level change points across alliances

At the individual species level, we determined a change point, or lack thereof, for 1268 unique species (Table S1). Individual species change points for decreasing species ranged from 1.3 kg N ha⁻¹ yr⁻¹ (for Sisymbrium altissimum in the Artemisia tridentata Shrubland Alliance; Figure 3a) to 16.8 kg N ha⁻¹ yr⁻¹ (for Euonymus americanus in the Fagus grandifolia - Quercus rubra - Quercus alba Forest Alliance; Figure 3b). Individual species change points for increasing species ranged from 1.7 kg N ha⁻¹ yr⁻¹ (for Fragaria virginiana in the Pseudotsuga menziesii Woodland Alliance) to 18.0 kg N ha⁻¹ yr⁻¹ (for Polygonum virginianum in the Fagus grandifolia - Quercus rubra - Quercus alba Forest Alliance; Figure 3b). Across all alliances, 56% of individual species change points for decreasing species fell below the community level change point of their alliance, and 53% of individual change points for increasing species were above the community level change point of their alliance. However, there is much variation in this within individual alliances, and examples can be found where there is a less balanced distribution of individual species change points on either side of the community change point (see Figures 3 and S1 for examples).

A number of the individual species occurred across more than one distinct vegetation alliance; out of all unique species assessed, 469 species were present in only one alliance, and 799 species were present in two or more alliances. For the species that occurred across multiple alliances, 68% had variable responses to N deposition, i.e., increases in abundance, decreases, and no significant change between different alliances (Table 4).

Table 4:

Species responses across vegetation alliances to atmospheric nitrogen deposition.

Total unique species	1268
Species occurring in one alliance only	469
Decreasing:	169
Increasing:	104
No change:	196
Species occurring in more than one alliance	799
Always decreasing	89
Always increasing	58
Always no change	107
Mixed response	545
Increase and decrease	22
Decrease more than increase (>2/3)	4
Increase more than decrease (>2/3)	3
Decrease and no change	172
Increase and no change	202
Increase, decrease, and no change	149

In all but four of the alliances that had species decreasing in abundance (i.e., z– change point), at least one of those species was a positive indicator for the alliance (Table 1); however, most alliances had positive indicator species that were also increasing in abundance (i.e., had a z+ change point). Similarly, invasive species were shown to increase (e.g. Andropogon gerardii - (Sorghastrum nutans) Herbaceous Alliance) and decrease (e.g. Pinus taeda Forest Alliance) in abundance across the N deposition gradient (Table 1).

4. Discussion

4.1. Community change in response to nitrogen deposition

Community level change points were found for 23 of the 24 alliances assessed (Table 2). The lowest community change points observed in this study, for both decreasing and increasing species, were below those found in the similar TITAN studies carried out in Europe (Payne et al., 2020; Wilkins et al., 2016). However, it is difficult to compare this study to those conducted previously, as we considered many different habitat types. We found that the community change points tended to cluster (Figure 2) around 2–4 kg N ha⁻¹ yr⁻¹ and again around 8–12 kg N ha⁻¹ yr⁻¹. Both the low and high clusters included multiple habitat types (low: forest, herbaceous, shrubland, and woodland; high: forest, herbaceous, and woodland). Most of the alliances assessed in this study were forested habitats (Figure 1), and while forest alliances dominated by coniferous species were generally more sensitive to N deposition than deciduous broadleaf-dominated forests, consistent with the European empirical critical loads work (Bobbink and Hettelingh 2011), this was not always the case. Rather than habitat type, the community change point clusters reflected a spatial trend; all habitats in the low-range cluster, with the exception of Populus tremuloides Forest alliance, were located in western US. In contrast, all of the alliances in the high-range cluster were located in eastern US (Figure 4).

The spatial trend showing lower community change points in the west of the country compared to the east may be a result of long-term N deposition trends. It is likely that eastern communities (where N deposition has been historically higher) have already lost N sensitive species and those that remain are less vulnerable to elevated N deposition, thus having higher z− and z+ change points; these alliances have decreasing community change points of 8.0–14.3 kg N ha⁻¹ yr⁻¹ (with the exception of Populus tremuloides Forest alliance) and increasing community change points of 7.0–16.2 kg N ha⁻¹ yr⁻¹ (Figure 2). Conversely, the western alliances have not lost sensitive species or experienced excessive elevated deposition that favours opportunistic species, and so the community change points remain lower, with decreasing community change points of 1.8–2.8 kg N ha⁻¹ yr⁻¹ and increasing community change points of 2.5–4.6 kg N ha⁻¹ yr⁻¹ (Figure 2). In Minnesota, where there is an exception to this west-east trend and the low to high community change points follow a north to south pattern, the vegetation dataset had a relatively high number of plots (n = 2179) and alliances (n = 10) represented for this state; this variety in alliance types may be contributing to the wide range in community change points. The north-south trend is partly explained by the range in climate and soil types in the state, where the northern portions of the state are colder and soils are sandier than the southern portions due to glaciation in the north. Areas that are colder and with sandier soils have been reported to be more sensitive to species loss following N fertilization (Clark et al., 2007). The north-south trend may also be explained by land-use variation both within the state and in neighbouring states and the Canadian provinces to the north. There may be historically higher N deposition towards the south of the state due to a greater population density and a greater prevalence of agricultural practices in surrounding states compared with the predominantly forested region at the north end of the state, and so N sensitive species may already be lost, driving community change points higher.

For all alliances with a community change point for both the decreasers and increasers, the z− community change point was lower than the z+ community change point, suggesting that N sensitive species decrease in abundance at lower deposition levels prior to increases in abundance for N tolerant species (Figure 3, Figure S1). However, Simkin et al. (2016), using the same vegetation data to explore how edaphic and climatic conditions affected the vulnerability of the herbaceous community to N deposition, found that richness tended to increase at lower N deposition across the sites and decrease at higher deposition using presence-absence data. This might suggest that increasing community change points should be lower than decreasing community change points. However, our contrary results may be explained by our use of species abundance; the complete disappearance of the z− species (i.e., abundance = 0) would occur at a higher N deposition than the z− community change point, potentially higher than the z+ community change point. In other words, the local extirpation of the z− species is overlapped by the local recruitment of the z+ species, resulting in the initial increase in richness, and eventual decrease further along the gradient. The z− and z+ community change points were also within 2 kg N ha⁻¹ yr⁻¹ for 70% of the alliances and this similarity between the community change points for decreasers and increasers suggests that there is an ecological threshold where the communities shift from nitrogen sensitive to nitrogen tolerant species as N deposition increases. Both N sensitive and N tolerant species may be present as this shift is occurring, resulting in a peak in species richness during this change.

We found that roughly half of the alliances had more z+ species than z− species (Table 2) within the N deposition range assessed. This trend was different than that found in the TITAN studies of Payne et al. (2020) and Wilkins et al. (2016) conducted on European datasets, in which nearly all habitats assessed had a greater number of N sensitive species declining than N tolerant species inclining. In our study, the dominance of increasing species was particularly evident in the low N deposition change point cluster (Figure 2), suggesting that species in the low deposition areas initially respond with increased cover. This result is consistent with Simkin et al. (2016), who found a positive relationship between richness and N deposition for 64% of 44 community gradients assessed; species richness likely reflects the cumulative response of the different z− and z+ species within each alliance, which may explain why Simkin et al. (2016) found that richness was more often increasing with N deposition in the US. Thus, the work presented here provides more information on the species changes driving the total richness trends, and may help to identify the impacts of N deposition, such as the loss of key indicator species, despite an increase in total richness.

4.2. Nitrogen deposition in excess of community change thresholds

Nitrogen deposition in excess of a community change point (Figure 5) suggests that the herbaceous community at a site has already experienced, or is at risk of, a shift in species composition. Long-term N deposition estimates were shown to be in excess of the z− community change points at nearly three-quarters of the study plots, while more recent N deposition (i.e., TDEP 2016–2018 averages) were shown to be in excess of the z− community change points at one third of the plots. Similarly, long-term N deposition was in excess of the z+ community change point at twice as many plots as recent N deposition. If the long-term N deposition values are considered as ‘past’ N deposition, this suggests fewer plots are at risk of community change in species abundance due to N deposition now than they were previously, though many remain vulnerable.

Although the decrease in the percentage of plots at risk is encouraging, our results suggest that N deposition may still be changing herbaceous communities. Considering the percent excess for sites where recent N deposition estimates were above the z− community change points, our results suggest that a reduction of N deposition of approximately 20% would protect half of the plots for which N deposition currently exceeds the change point (Figure S2).

4.3. Individual species change

At the individual species level, we were able to assess five times more species from the Simkin et al. (2016) dataset than was possible in the Clark et al. (2019) study. As a number of the individual species occurred across more than one distinct vegetation alliance, we were able to compare variation in species change points in different habitats. We found that the direction of the response (i.e., increasing, decreasing, or no significant change) for 68% species that occurred across multiple alliances varied (Table 4), which has been found in similar studies (Payne et al. 2020). For example, Pteridium aquilinum, a common bracken fern with a very widespread distribution, had a z+ CP in the Pseudotsuga menziesii Forest Alliance (N dep. range = 1.4–7.5 kg N ha⁻¹ yr⁻¹), no significant CP in the Quercus prinus - Quercus (alba, falcata, rubra, velutina) Forest Alliance (N dep. range = 7.0–15.3 kg N ha⁻¹ yr⁻¹), and a z− CP in the Acer saccharum - Betula alleghaniensis - (Fagus grandifolia) Forest Alliance (N dep. range = 4.0–15.6 kg N ha⁻¹ yr⁻¹); it is a positive indicator species in all three alliances.

The apparent inconsistent response of so many species between alliances potentially reflects their exposure to different regions of the N deposition gradient, as not all alliances experience the same range in N deposition. Thus, these could be species that showed a unimodal relationship in Clark et al. (2019), where sites were not separated into alliances and the full range of the N deposition gradient was captured simultaneously for each species. It is important to note the varying response trend is not consistent among all species, but it does highlight the need to consider the N deposition exposure in a particular alliance when assessing species-level change points.

While we did not conduct this analysis with any specific species of concern in mind, we did flag species that were positive indicators of the alliances, and also known invasive species (USDA, NRCS 2020; Table S1). Extensions of this work could also flag species that are listed on the USFS Regional Forester Sensitive Species list (RFSS). Our results show that most alliances assessed had at least one positive indicator decreasing in abundance; however, the same was true for positive indicators increasing in abundance. Likewise, invasive species were shown to both increase in some alliances and decrease in others. Again, these observations may reflect the variation in N deposition gradient between the alliances. This demonstrates the importance of determining responses to N deposition, and subsequent critical loads, at the community level and at regional scales.

4.4. Comparison to critical loads

When community change points were summarised at the Level 1 ecoregion scale, the community change points tended to be lower than the Simkin et al. (2016) species richness critical loads, particularly for the z− community change points (Table 3). This makes sense, as the changes in abundance would occur before the absence of a species (i.e., decline in richness).

The ecoregion-summarized community change points fell below or within the critical load of N deposition (CLN) ranges for US habitats provided by Pardo et al. (2011), who synthesized empirical CLN ranges from a number of research studies. The TITAN results presented here may allow for a revision of the empirical critical load ranges to include protection from changes in plant species abundance or herbaceous community composition. Developing critical loads requires the designation of one or more indicators of ‘significant harmful effects’. The shift in community composition suggested by the community change point may be considered a significant harmful effect, but other results may be considered as well, such as the decrease in abundance of positive indicator species (e.g., Aherne et al., 2021; Wilkins et al. 2016), or protection from increase of invasive or weedy species. For example, in this study the Artemisia tridentata Shrubland Alliance had an increase in a number of invasive species; setting the CLN to the z+ community change point of 2.5 kg N ha⁻¹ yr⁻¹ would deter those invaders.

One of the advantages of TITAN is that uncertainty around the change points is assessed, and represented as 0.05–0.95 quantiles determined from bootstrapping, and it is also notable that quantile ranges for individual species can be quite wide (see examples in Figure 3). In contrast, the quantile ranges for community change points are narrower (Figure 2b, d). Thus, while our results can be used to inform critical loads for both individual species and communities (alliances), quantile ranges should be considered in each case. The narrower quantile ranges suggest greater confidence in community change points, although they should be considered with more caution if they are based on few species (Figure 2a, c).

5. Summary

For the 24 vegetation alliances assessed, the community-level change points for decreasing species (z− community change point) ranged from 1.8–14.3 kg N ha⁻¹ yr⁻¹ and were lower in western US. This suggests that eastern communities, where N deposition has been higher historically, may have already lost the more sensitive species, and N-tolerant species remain. While the z+ community change point was consistently greater than the z− community change point, the difference between the two was > 2 kg N ha⁻¹ yr⁻¹ in only 30% of the alliances, suggesting that, in most of the alliances, there is a more distinct threshold of N deposition where the vegetation in an alliance shifts from N sensitive to N tolerant assemblages. Current N deposition estimates suggest that there has been a 37% reduction in the number of plots where N deposition is in excess of the community change point for decreasing species (from 72% to 35%) during the past 2+ decades. While these reductions are encouraging, many areas are still at risk; this suggests that further policies are required to reduce N deposition to protect sensitive habitats and species. Our results suggest that a reduction of 20% would protect half of the plots where current N deposition exceeds the decreasing community change point. The work presented here may help us to understand the changes in species richness demonstrated in Simkin et al. (2016) and Clark et al. (2019); a more thorough comparison of all three approaches would further our understanding of the relationship between N deposition and herbaceous communities. Irrespective, alliance-and species-level change points may be used to inform critical loads as they are developed in the US and elsewhere.