Evaluating the effectiveness of M-AMBI with other biotic indexes in a temperate estuary

Abstract

Need for a scalable and widely applicable index has been increasingly important. This study evaluates the applicability of the M-AMBI, a potential comprehensive index, at small spatial scales. M-AMBI was compared to regional indices (EMAP-E and GOM B-IBI), assessing response to natural environmental gradients and low oxygen stress. Results indicate poor agreement between indices with M-AMBI and GOM B-IBI showing positive correlation but significant disagreement in habitat condition. EMAP-E had no agreement. Indices showed similar patterns of better habitat scores in higher salinities. M-AMBI also showed a negative relationship with sediment organic matter and total nitrogen. DO influenced all indices with M-AMBI the most sensitive. However, mismatches between DO and index score were observed further calibration may be needed before adoption into programs. Overall, the M-AMBI demonstrates potential at smaller, local scales, but additional studies are needed to validate its performance in different coastal environments and under different conditions.

1. Introduction

Estuaries and coastal ecosystems play a vital role in providing ecosystem services and direct benefits to humans, like protecting coastlines, nutrient cycling, and fisheries productivity (Barbier et al., 2011). Benthic invertebrates are regularly used as biological indicators in these systems as they demonstrate a range of responses to disturbance (Pearson and Rosenberg, 1978; Dauer, 1993; Borja et al., 2000) and compound the effects of stressors over time due to their relatively sedentary life histories (Pearson and Rosenberg, 1978; Dauer, 1993), presenting an integrated measure of the conditions at a specific site. Indices of biologic integrity (IBI) are widely used as benthic monitoring tools (Borja et al., 2015; D’Alessandro et al., 2020). Initially developed for freshwater (Cairns et al., 1968; Chandler, 1970; Hilsenhoff, 1982; Cairns and Pratt, 1993), many marine indices have been introduced in the last two decades (Borja et al., 2019). Translational differences between indices, however, hinder the aggregation of large volumes of data and limit utility of the information beyond the scope of the individual study or assessment (Borja et al., 2014).

Indices are calculated from metrics based on benthic community abundances which generate a numerical index score that can be used to determine ecological status and infer benthic habitat condition from impairment thresholds. Each index is calculated differently and uses different impairment criteria, making direct comparisons between them challenging. This poses a problem for integrating secondary data, and for monitoring programs that span large geographic areas or cross jurisdictional boundaries (Pelletier et al., 2018). For example, in the US Gulf of Mexico (GoM), federal and state agencies operate alongside estuary programs, nonprofits, and other stakeholders with little consistency in benthic assessment (Engle et al., 1994; Malloy et al., 2007; EPA, 2010; TetraTech, 2011; EPA, 2015; Nestlerode et al., 2020; Deitch et al., 2021; Wiesenburg et al., 2021). As a consequence, evaluating environmental impacts from the Deepwater Horizon disaster in 2011 has been constrained by limited baseline data (Joye, 2015; Martin et al., 2023). Subsequently, a scalable and widely applicable marine benthic index is needed for cohesive monitoring and for relevance to interests beyond the GoM. A potential option that could be used in current and future efforts, as well applied to historical data when available, is the multivariate AMBI (M-AMBI).

The M-AMBI is an extension of the AZTI’s Marine Biotic Index (AMBI) and combines AMBI BC (Biotic Condition) with Shannon diversity (H) and species richness (S) (Muxika et al., 2007; Sigovini et al., 2013; Borja et al., 2019). The AMBI sorts taxa into one of five ecological groups (EG) following the Pearson-Rosenberg (1978) successional model (Borja et al., 2000; Borja, 2004), and the proportions of each EG are used to calculate AMBI BC (Borja et al., 2000). Gillett et al. (2015) introduced US specific EG values to include additional taxa and reflect response to a variety of disturbances, grouped by national as well as east, west, and gulf coast values. Further recommendations by Pelletier et al. (2018) for applying M-AMBI in the US include calculations by discrete salinity zones, minimum grab size, and substituting % oligochaetes for species richness in tidal freshwater habitats. The US Environmental Protection Agency adopted M-AMBI in 2015, replacing the multiple regional indices used in the National Coastal Condition Assessment (NCCA) program with a single index (EPA, 2015). Utility of M-AMBI has been demonstrated at national and regional scales; however, further work is needed to evaluate its applicability in smaller, local monitoring needs. Furthermore, validation datasets did not include the Gulf of Mexico (Gillett et al., 2015; Pelletier et al., 2018) highlighting the value of studies applying M-AMBI in the GoM specifically.

The Environmental Monitoring and Assessment Program-Estuaries (EMAP-E) index for the Gulf of Mexico was used prior to the US EPA adopting the M-AMBI at a national level (EPA, 2010). It is based on the index developed by Engle et al. (1994) and revised by Engle and Summers (1999). It is calculated with an expected Shannon diversity score from observed salinities, and abundances of specific taxa and groups (EPA, 2010). The Gulf of Mexico Benthic Index of Biological Integrity (GOM B-IBI) was developed for the Gulf of Mexico Alliance (GOMA) (TetraTech, 2011) and incorporates a sensitive Beck’s index (Beck, 1955), and abundance of certain taxa. Calculations are defined by salinity classes: low salinity (LS), high salinity non-Florida (HS-xF), and high salinity south Florida (HS-sF); however, due to insufficient data representation for north Florida, a separate high salinity class for the region could not be established, potentially limiting its effectiveness (Nestlerode et al., 2020) and warranting further studies in the northern GoM.

Objectives of this study were to investigate the applicability of the US adapted M-AMBI at smaller, local scales by comparing it to regional indices (EMAP-E and GOM B-IBI) using Pensacola Bay, Florida in the northern Gulf of Mexico as a case study. More specifically, we sought to evaluate agreement between index numerical scores and the assignment of benthic habitat condition. Additionally, the study aimed to compare how indices respond to natural estuarine gradients and low oxygen stress. We did not expect to find high levels of agreement between indices as they are calculated with different metrics (Table 1), and we expected the M-AMBI to exhibit greater sensitivity to oxygen demand compared to other indices and the most degraded habitat scores in samples associated with low bottom water DO.

Table 1.

Comparison of index calculations (metrics and calculation steps) and benthic habitat condition thresholds for M-AMBI, EMAP-E, GOM B-IBI LS (low salinity) and HS-SF (high salinity south Florida), and relevant references for each.

Index	Calculations		Benthic Habitat Condition			References
	Metrics	Steps	Poor	Moderate	Good
M-AMBI	AMBI BC (Biotic Coefficient)	Calculate metrics and standardize	< 0.39	0.39 – 0.53	> 0.53	Sigovini et al. 2013 NCCA (2015) Pelletier et al. 2018
	Species richness (S)	Set reference values
	Shannon diversity index (H)	Factor analysis (FA)
		Normalize scores (0 to 1)
EMAP-E	Proportion of expected Shannon’s (H)	Calculate metrics	< 3	3 – 4.5	> 4.5	NCCA (2010)
	Mean abundance Tubificidae	Univariate formula
	Percent abundance Capitellidae
	Percent abundance Bivalvia
	Percent abundance Amphipoda
GOM B-IBI LS	Percent Bivalvia	Calculate metrics	< 45	45 – 55	> 55	TetraTech (2011)
	Percent Spionidae	Univariate formula
	Percent predators
	Percent Tolerant
	Becks (sensitive) Biotic Index
GOM B-IBI HS-SF	Number Bivalvia	Calculate metrics
	Percent Polychaeta	Univariate formula
	Percent interface feeders
	Percent tolerant
	Becks (sensitive) Biotic Index

2. Methods

2.1. Study area and approach

Pensacola Bay (Fig. 1) is a relatively shallow microtidal estuary system in Northwest Florida (USA) (Collard, 1991) and consists of several sub-estuaries including Escambia Bay and Pensacola Bay proper (Schroeder and Wiseman, 1999; Hagy and Murrell, 2007; Murrell et al., 2009). It is a river-dominated system that receives significant freshwater inputs in the late spring and summer, contributing to saline stratification (Hagy and Murrell, 2007). The seven stations utilized in this study were selected based on an established Pensacola Bay survey design. They are located along the estuarine salinity gradient from station PB02 near the Escambia River to station PB08 near the Gulf of Mexico (Hagy et al., 2006; Hagy and Murrell, 2007; Nestlerode et al., 2020; Duvall et al., 2022). Stations were evaluated both individually and grouped by potential management areas: upper bay (stations PB02–04), middle bay (PB05–06), and lower bay (PB07–08).

Fig. 1.

Map of Pensacola Bay. Points refer to sampling stations along the salinity gradient from north near the Escambia River down to the Gulf of Mexico (Adapted from Nestlerode et al. 2020)

Water quality, sediment, and benthic invertebrates were collected in July, October, and November in 2016, and then monthly from May through November in 2017 (N = 10 for each measurement for each station). Benthic indices were calculated for each sample and assigned habitat condition (poor, moderate, good). Agreement between indices was evaluated and index response to dissolved oxygen quantified. Benthic habitat condition assignments were also evaluated with respect to changes in benthic community structure and natural gradients in the estuary.

2.2. Sample collection and processing

Water quality (temperature (°C), salinity, and DO (mg L⁻¹)), was measured with a single cast of a Sea-Bird CTD (Model SBE 25 and SBE25plus) at the time of sampling. Values were taken from the minimum recorded depth (~0.2 m from the bottom). The data from the last two observations in October and November 2017 were lost due to CTD data corruption, and missing values were estimated using regression models based on monthly CTD measurements from 2015 to 2017.

Sediment was collected with a petite box corer (0.023m², Wildco/Science First, Yulee, FL USA). A single sediment core was collected from the grab for each parameter. Sediment bulk density (BD), percent organic matter (% OM), percent calcium carbonate (% CaCO₃), total nitrogen (N mg g⁻¹), and organic carbon (C mg g⁻¹) were processed at the EPA’s GEMMD laboratory (Gulf Breeze, FL). BD was measured as water loss by volume while %OM and % CaCO₃ were estimated by loss on ignition at 550 °C and 1000 °C (Heiri et al., 2001; Percival, 2017). Sediment C and N were measured via combustion using a Carlo Erba FLASH 2000 organic elemental analyzer (CE Elantech, Lakewood NJ, USA). Grain size analysis was performed by the Texas A&M Coastal Geology Laboratory (Galveston, TX USA). Samples were treated with a solution of sodium metaphosphate for dispersal and the slurry poured into a Malvern Mastersizer 2000© (Malvern Panalytical, Westborough, MA USA) for analysis via laser diffraction. Sediment Chlorophyll-a analysis was performed by the Nutrient Analytical Services Laboratory, University (Cambridge, MD USA) via acetone extraction followed by fluorometric analysis (Arar and Collins, 1997). Quality assurance included the collection of two additional cores taken from one site on each sampling date at random: a QC duplicate taken from the same grab as the sample replicate to determine within-grab variability (% coefficient of variation (CV) < 10) and a site duplicate taken from a separate grab to assess within-site variability (% CV < 25). <10 % of all sediments exceeded quality assurance criteria.

Benthic macroinvertebrates were collected with the box corer described above. Two grabs were taken for each benthic sample. Identifications were made by the Versar Benthic Laboratory (Versar Inc., Columbia, MD USA) following World Register of Marine Species (WoRMS) nomenclature (WoRMS, 2021). Abundances were normalized to represent the number of individuals per m². Quality control procedures conducted by Versar included an evaluation of sorting and identifications on 10 % of samples, falling within acceptable criteria of 5 % or fewer errors. As sampling was carried out before the 0.03 m² recommendation by Pelletier et al. (2018) for calculating the M-AMBI, grabs were evaluated as a single 0.046 m² sample for each station for each sampling event (N = 10 for each station). This process allowed us to assess the broader responses of the benthic community to variations in sediment and water while not specifically focusing on temporal trends.

2.3. Data analysis

Statistical computations and figures were performed using R 4.1.0 (R, 2020). Differences in water and sediment parameters were assessed via ANOVA followed by Tukey’s post hoc, transforming as needed to meet assumptions, or nonparametric Kruskal-Wallis followed by Dunn’s multiple comparisons test. Nonmetric multidimensional scaling (NMDS) was applied to macroinvertebrate abundances using the vegan package (V 2.6.4) (Oksanen et al., 2020). A Hellinger transformation was first applied using the “decostand” function. The ordination was then constructed using Bray-Curtis dissimilarity scores condensed to 2 dimensions. Model fit was evaluated by inspecting the Shepard diagram (stress plot) and stress score, which at 0.714 fell below the suggested 0.2 criteria for model fit (Clarke, 1993). The “envfit” function in vegan was used to correlate environmental parameters with the NMDS ordination via correlation and significance determined via permutation tests (N = 999) with the ordination axes as explanatory variables. Vectors were then projected onto the ordination using plot functions in ggplot2 (V 3.4.0) (Wickham, 2016) for correlations that were significant at p <0.05 and species scores at p < 0.02. To detect differences in benthic communities, permutational multivariate analysis of variance (PERMA-NOVA) was applied to abundances using the “adonis” function in the EcolUtils package (V 2.0) (Salazar, 2021). Bray-Curtis dissimilarities (from Hellinger transformed abundances) were modeled using bay region and habitat condition as explanatory variables followed by permutations (N = 999) to determine significance.

The M-AMBI was calculated for each sample by first coding taxa by their ecological group (EG, i.e. tolerance value), using the “US Gulf Coast Values” where EG designation differed from the “US Hybrid Value” (Gillett et al., 2015). We include the US Gulf Coast EG values in Appendix A. For species missing EG designation, the EG value for the genus was used when available with no sample exceeding 50 % of taxa without an assigned EG. AMBI was calculated for each sample using the formula given by Borja et al. (2000). Species diversity (S) and Shannon Diversity Index (H′) were calculated using the vegan package. Samples were then labeled according to Venice salinity zone (Muxika et al., 2005) based on bottom water salinities averaged over the study period: PB02 mesohaline (5 to <18), PB03-PB06 polyhaline (18 to <30), and PB07-PB08 euhaline (30 to <40). The M-AMBI score for each sample was calculated following Pelletier et al. (2018) using the R script “fun. mambisimpl.R” given by Sigovini et al. (2013). Furthermore, the original script was updated following changes to the function “factor.scores” in the psych package (V 2.2.9) that were made after the script was published (Sigovini et al., 2020). The output was then corrected for over and under projections where M-AMBI values >1 were scaled to 1 and values < 0 were scaled to 0. M-AMBI scores were then used to assign benthic habitat condition: poor (<0.39), moderate (0.39–0.53), and good (>0.53) (Table 1). Three levels of habitat quality were used for M-AMBI instead of the five-level system suggested by Muxika et al. (2007) to more easily compare with other indices and for consistency with the US national monitoring program (EPA, 2015).

The NCCA index EMAP-E was calculated using the formulas in EPA (2010). Bottom water salinity measured at the time of benthic sampling was used for the expected Shannon’s value needed to calculate the index. Transformations of abundances were performed as recommended and all parameters were scaled to mean = 1 and standard deviation = 0 prior to index calculation. Benthic habitat condition assignments followed EPA (2010): poor (<3), moderate (3–4.5), good (>4.5).

The Gulf of Mexico index (GOM B-IBI) was calculated by first coding each taxon by tolerance value (TetraTech, 2011). Tolerance values were obtained by request of the Mississippi DEQ Project Manager as recommended by TetraTech (2011). Tolerance value were developed from National Coastal Assessment data from the Gulf of Mexico collected 2000 through 2006 from indicator taxa analysis (ITA) (Dufrˆene and Pierre, 1997) and Constancy/Fidelity (Boesch, 1977) to assess response of individual taxa to organic pollution tolerance/sensitivity (TetraTech, 2011). Functional group classifications were assigned according to those given by TetraTech (2011), which were assigned using available literature following the nomenclature described by Tenore et al. (2006). Groups include conveyer belt species (conv), interface and water column feeders (infc), predators (pred), scavenger/browsers (scbrw), and subsurface feeders (subsrf). Sites were classified as low salinity (LS, 0.5–18; Site PB02) and high salinity (HS, 18–40; Sites PB03 to PB08) by average bottom water salinities. The index established two high salinity classes, one for south Florida (below 28.0°N) and another for non-Florida Gulf states, however no class was established for north Florida and the panhandle as not enough data were available. Both high salinity metrics were calculated for comparison, however the non-Florida calculations fell within 0.01 of 60 and did not give any resolution between sites; thus, the high salinity south-Florida metric is reported. Values were assigned habitat condition according to the 25th and 75th percentiles of the samples: poor (<45), moderate (45–55), good (>55).

Agreement between numerical index scores was evaluated by Spearman correlation by first scaling and centering values and then identifying significant relationships. As indices have the same directionality (i.e., higher scores relate to better habitat condition), monotonic relationships were evaluated without further adjustment. Overall benthic habitat condition assignments were compared using Cochran-Mantel-Haenszel tests with package vcd (V 1.4.11) for each pair of indices, stratified by station, followed by Cohen’s Kappa to quantify pairwise agreements between indices for each station. Linear mixed models were constructed with the “lmer” function in the lme4 package (V 1.1.30) (Bates et al., 2014) to evaluate how each index responds to changes in dissolved oxygen and other water and sediment parameters, with index score set against water and sediment and sampling station as mixed effects. Explanatory variables were centered and scaled prior to modeling, and likelihood ratios were used to determine significance of predictor variables while Akaike Information Criterion (AIC) values were used to evaluate model fit between the full model and reduced predictors in the final model. Model R² values were computed with the r2glmm package (V 0.1.2) (Jaeger, 2017) using the measure described by Edwards et al. (2008).

3. Results

3.1. Site conditions

Bottom water temperatures ranged from 21.4 °C to 31.4 °C with an average of 26 °C across all stations (Table 2). As expected, bottom water salinities (Fig. 2) varied significantly among bay regions (Kruskal-Wallis followed by Dunn’s, X² ₌ 49.5, p ≪ 0.001). Although hypoxia (DO ≤ 2 mg L⁻¹) was observed at least once at every station (Fig. 2), within-site variability limited the ability to detect significant differences between stations (Kruskal-Wallis followed by Dunn’s post hoc, X² ₌ 10.33, p = 0.11). DO values in the middle bay trended lower, particularly for site PB05 which averaged 2.5 mg L⁻¹ (±2 SD) compared to 4.8 (±1.4 SD) and 3.8 (±2 SD) for lower and upper bays respectively. Sediments were predominantly muddy sand, comprised mostly of clay and silt with sand making up roughly 10 % of sediments in the upper bay and decreasing down the transect (Table 2). Sediment BD was highest in the upper bay while % CaCO₃ and % OM increased in the lower bay along with total nitrogen. Sediment chlorophyll a was variable with mean observations at PB04 and PB08 roughly three times higher than other stations. Additionally, phaophyton levels were elevated at station PB08.

Table 2.

Water and sediment parameters by station. Values represent means (± standard deviation) for the study period. Superscripts denote statistical significance (α = 0.05 for ANOVA followed by Tukey’s posthoc (parametric) and Kruskal Wallis followed by Dunn’s posthoc (non-parametric)). The r² and p value (“p val) from correlations with the NMDS ordination for each variable are given.

	P02	P03	P04	P05	P06	P07	P08	r2	p val
Temperature °C	26.62 ± 3.24	26.39 ± 2.75	26.77 ± 2.81	27.20 ± 2.21	26.33 ± 2.39	26.18 ± 2.24	25.95 ± 2.28	0.06	0.14
Salinity	17.87 ± 8.77	20.02 ± 9.37	20.86 ± 8.84	27.69 ± 1.96	28.28 ± 2.77	31.36 ± 2.05	32.41 ± 1.59	0.41	0.001
DO (mg L−1)	3.86 ± 2.19	3.50 ± 2.14	3.73 ± 2.06	2.48 ± 2.00	3.00 ± 1.63	4.03 ± 1.71	4.85 ± 1.45	0.20	0.002
Bulk Density	0.38 ± 0.03a	0.32 ± 0.06b	0.29 ± 0.01b	0.22 ± 0.01c	0.22 ± 0.01cd	0.21 ± 0.02d	0.25 ± 0.04d	0.48	0.001
% Organic Matter	10.77 ± 1.15a	11.78 ± 1.82b	12.76 ± 0.71b	13.97 ± 0.81b	13.85 ± 0.51c	14.67 ± 0.47cd	13.61 ± 1.39d	0.41	0.001
% CaCO 3	5.61 ± 0.59a	7.12 ± 0.63b	6.56 ± 0.50b	9.29 ± 0.90c	11.19 ± 1.03d	14.73 ± 0.90e	14.16 ± 2.35e	0.68	0.001
% Silt	24.20 ± 7.87a	15.97 ± 4.46b	15.89 ± 3.77b	12.40 ± 3.52b	12.70 ± 3.50b	15.29 ± 3.98b	27.06 ± 8.37a	0.20	0.001
% Clay	65.88 ± 7.23a	73.60 ± 5.08a	75.24 ± 3.66a	80.29 ± 2.77b	79.23 ± 2.98b	77.76 ± 2.60c	66.96 ± 8.03c	0.20	0.001
% Sand	9.92 ± 1.41a	10.43 ± 1.44ab	8.87 ± 1.37bc	7.30 ± 2.68bc	8.07 ± 2.78cd	6.95 ± 1.71d	5.98 ± 2.57d	0.28	0.001
Total N (mg g−1)	3.50 ± 1.02a	4.36 ± 1.12a	5.24 ± 1.83ab	4.87 ± 1.23abc	4.02 ± 0.48bc	5.52 ± 1.04cd	5.82 ± 1.09d	0.10	0.02
Organic C (mg g−1)	34.83 ± 6.09	35.34 ± 5.74	34.22 ± 3.34	29.96 ± 10.78	32.65 ± 4.29	32.09 ± 5.88	35.53 ± 12.46	0.03	0.40
Chlorophyll A (mg m−3)	35.97 ± 15.04a	76.19 ± 46.85a	124.71 ± 76.58b	58.21 ± 41.62a	40.33 ± 19.76a	58.36 ± 28.18a	147.33 ± 87.30b	0.13	0.01
Phaophyton (mg m−3)	113.19 ± 39.20a	127.93 ± 50.53a	133.39 ± 34.63a	126.70 ± 47.86a	121.09 ± 32.14a	133.73 ± 28.76a	197.81 ± 61.25b	0.18	0.002

Fig. 2.

Bottom water salinity and dissolved oxygen from CTD casts taken across the study. Boxes represent upper quartiles, medians, and lower quartiles while whiskers give the upper and lower extremes for upper (stations PB02 - PB04), middle (PB05, PB06), and lower (PB07, PB08). No significant differences were observed in DO between stations (Kruskal-Wallace followed by Dunn’s pairwise posthoc tests, X² ₌ 49.5, p < <0.001 (salinity); X² = 10.33, p=0.11 (DO)).

3.2. Macroinvertebrates

Across the study 130 taxa were identified of which 61 were represented by class polychaeta at 80 % and 85 % of total abundances in the upper and mid bays and 60 % in the lower bay. Species richness was highest in the lower bay, roughly double the observations for upper and mid bays (Table 3). Shannon diversity (H′) was also highest in the lower bay at 2.07, and then followed the salinity gradient down to 0.82 at PB02 near the river (Table 3). Taxa sensitivity was assessed using ecological group (EG) classification (Gillette et al. 2015) and tolerance values (TetraTech, 2011), which differed slightly in the number of taxa missing designation and the tolerances assigned. Communities in the upper bay were mostly represented by tolerant taxa while more sensitive taxa were found mid-bay (Table 3). Lower bay communities had the highest proportion of sensitive taxa, which was more pronounced for the ecological group classification than tolerance grouping. For functional groups, the percentage of interface/water column feeders in the upper bay was roughly half that of the mid and lower bays at 22 % compared to 54 % (Table 3). Conversely, subsurface feeders were the most abundant in the upper bay representing 55 % of the benthic community compared to 16 % in the middle and 18 % in the lower bay. Overall, there were relatively few scavengers/browsers in the bay while predators were the most abundant in the middle bay at roughly 25 % of the community.

Table 3.

Benthic community by station and bay region: upper (stations P02 - P04), middle (P05, P06), and lower (P07, P08). Species is the total number of species observed for each station and bay region. Shannon diversity scores (H) represent mean values (± standard deviation); superscripts denote statistical significance (α = 0.05 for Kruskal Wallis followed by Dunn’s posthoc). Sensitive (EG) represents the percent of community abundances classified as EG “I” (Gillett et al. 2015). Sensitive (T. val) reports the percent of abundances in designation “1” (Tetratech 2011). Following Tenore (2006): Infc = percent of abundances that were interface and water column feeders, Pred = percent predators, Scbrw = percent scavenger/browsers, and Subsrf = percent subsurface feeders.

	S	H	% Sensitive (EG)	% Sensitive (T. val)	%Infc	%Pred	%Scbrw	%Subsrf
Upper	45	1.10 ± 0.51	0.26	0.21	21.88	20.67	0.14	55.12
P02	36	0.82 ± 0.40a	0.30	0.20	18.98	28.27	0.30	48.55
P03	27	0.98 ± 0.35a	0.09	0.14	22.16	14.79	0	62.78
P04	24	1.50 ± 0.53b	0.47	0.37	29.45	11.46	0	57.78
Middle	49	1.81 ± 0.46	1.14	1.95	53.58	25.77	0	16.02
P05	31	1.60 ± 0.47b	1.19	2.53	48.51	29.17	0	20.24
P06	39	2.02 ± 0.37c	1.08	1.25	59.68	21.68	0	10.93
Lower	87	2.07 ± 0.40	9.17	2.29	53.96	14.69	0.83	18.23
P07	56	2.07 ± 0.45c	5.96	1.70	60.43	16.81	0.21	10.64
P08	58	2.07 ± 0.37c	12.24	2.86	47.76	12.65	1.43	25.51

Nonmetric multidimensional scaling (NMDS) ordination indicates benthic communities in the upper, middle, and lower bay are significantly distinct (Fig. 3) supported by PERMANOVA via Bray-Curtis followed by post hoc pairwise permutations (F₍₂₎ = 16.6, p < 0.001, R² ₌ 0.33). The upper bay is dominated by Ancistryosyllis jonesi, Tubulanus sp., Capitella capitata, Streblospio benedicti, Carinoma tremaphoros, and Mediomasus ambiseta (Fig. 3a). These taxa represent relatively tolerant species within class Polychaeta (as subsurface feeders) and Paleonemertea (predators). In the middle bay Middle bay, Cossura soyeri and Aricidea (Strelzovia) sp. are prevalent (Fig. 3a); both Polychaetes classified as interface and water column feeders. C. soyeri, in particular, is tolerant. Communities in the lower bay are more diverse with representatives in class Malacostraca, Gastropoda, Sipunculidea, and Bivalvia identified in addition to Polychaeta: Leucon sp., Nassarius acutus, Phascolion (Phascolion) strombus strombus, Nuculanidae, Parvilucina crenella, and Paraprionospio alata (Fig. 3a); all except for P. alata are classified as interface and water column feeders. Environmental parameters fit to the ordination are given by Fig. 3b, with upper bay communities strongly influenced by dense sandy sediments and the lower bay by high %OM, N, and %CaCO₃. Sediments in the middle bay are characterized by high clay and low silt. There also appears to be a relationship between mid-bay communities and low dissolved oxygen, which was significant at p = 0.002 and an R² of 0.2 but the relationship is weaker than observed for other environmental parameters (Table 2).

Fig. 3.

NMDS of benthic community. Points represent site scores colored by bay region: upper (stations PB02 - PB04), middle (PB05, PB06), and lower (PB07, PB08). Species vectors (top) represent species scores correlated to the ordination at p < 0.02. Environmental parameters (bottom) were fit to the ordination through correlations and plotting only those with p < 0.05. Black vectors represent environmental factors significant to explaining M-AMBI as determined by linear mixed models.

Benthic communities categorized as “good” and “poor” benthic habitat condition for M-AMBI were different (PERMANOVA via Bray-Curtis followed by pairwise permutations: F₍₂₎ = 2.4, p = 0.007, R² ₌ 0.07). For GOM B-IBI, communities designated as “moderate” showed significant differences from “good” and “poor” communities, while no differences were detected for EMAP-E (PERMANOVA via Bray-Curtis followed by post hoc pairwise permutations: F₍₂₎ = 2.5, p < 0.01, R² = 0.07 and F₍₂₎ = 1.7, p = 0.08, R² ₌ 0.02). Between numerical index scores, Spearman correlation of M-AMBI and GOM B-IBI suggests a moderately positive monotonic relationship with correlation coefficient of 0.32 (p < 0.01) but no correlation between M-AMBI and EMAP-E was detected (p = 0.2). There was a negative correlation between EMAP-E and GOM B-IBI (r² ₌ −0.53 p <_<0.001). Comparing benthic habitat condition assignments via Cochran-Mantel-Haenszel tests, results suggest slight overall relationships between M-AMBI and EMAP-E and GOM B-IBI (CMH M² ₌ 11.90, p = 0.02, M² ₌ 9.20, p = 0.01), but not between EMAP-E and GOM B-IBI (CMH M² ₌ 3.2, p = 0.20). However, pairwise comparisons by station via Cohen’s Kappa indicate no agreement between index habitat condition assignments with most scores falling below the minimum criteria for agreement (Landis and Koch, 1977) and only one value greater than the criteria for substantial agreement (Table 4).

Table 4.

Summary of each benthic index including numerical index score (mean ± standard deviation across the study period), benthic habitat condition with total number of samples in each level, and Kappa score for each pairwise comparison with values above the minimum threshold for agreement bolded.

	Numerical Index Scores			Benthic Habitat Condition									Cohen’s Kappa
Site	M-AMBI	GOM B-IBI	EMAP-E	M-AMBI			EMAP-E			GOM B-IBI			M-AMBI vs EMAP-E	M_AMBI vs GOM B-IBI	EMAP-E vs GOM B-IBI
				Good	Mod	Poor	Good	Mod	Poor	Good	Mod	Poor
PB02	0.40 ± 0.13	57.90 ± 3.28	4.14 ± 0.20	2	5	3	1	9	0	6	4	0	−0.13	−0.18	0.14
PB03	0.33 ± 0.07	51.71 ± 3.17	4.10 ± 0.35	0	3	7	1	9	0	2	8	0	0.04	−0.05	−0.15
PB04	0.39 ± 0.11	50.10 ± 2.47	4.15 ± 0.20	0	6	4	0	10	0	0	10	0	0.00	0	0
PB05	0.38 ± 0.10	47.89 ± 7.14	4.33 ± 0.15	0	6	4	2	8	0	2	2	6	0.23	0.06	0.00
PB06	0.46 ± 0.08	46.12 ± 2.61	4.35 ± 0.11	2	6	2	1	9	0	0	6	4	0.09	−0.07	0.13
PB07	0.46 ± 0.10	49.41 ± 8.35	4.30 ± 0.14	3	4	3	1	9	0	3	2	5	0.02	0.71	−0.01
PB08	0.50 ± 0.09	54.04 ± 5.90	4.26 ± 0.15	5	3	2	1	9	0	4	5	1	−0.03	0.05	−0.18

Linear mixed models (LMM) were built from significant explanatory variables, as determined by likelihood ratio tests (p < 0.05) and comparing AIC values between the full model and reduced model (Table 5). All indices responded to changes in dissolved oxygen, with M-AMBI exhibiting the most significant response and predicting a 0.03 increase in M-AMBI for every 1 mg L⁻¹ increase in DO (±0.01). The M-AMBI also responded to changes in sediment % OM and N while EMAP-E responded to % CaCO₃, C, and Cl-A.

Table 5.

Linear mixed models with index score set against water and sediment as explanatory variables and sampling station as random effects: IBI = Var 1 + Var 2 + … + (1|Station). Bold p val (p values) represent significant relationships (α = 0.05) determined from log-likelihood ratio’s of standardized variables. Coefficients (“Coef.”) represent unstandardized values ± 95% confidence intervals (CI) for expected change in index score for one unit change in each parameter.

	M-AMBI			EMAP-E			GOM B-IBI
	p val	Coef	± CI	p val	Coef	± CI	p val	Coef	± CI
Salinity	<0.001	0.01	± 0.003	0.01	−0.01	± 0.007	<0.001	0.24	± 0.190
Dissolved Oxygen (mg L−1)	<0.001	0.03	± 0.010	0.01	−0.04	± 0.023	0.04	0.88	± 0.634
Bulk Density	0.57			0.87			0.31
% Organic Matter	0.02	−0.02	± 0.016	0.18			0.91
% CaCO3	0.35			<0.001	0.04	± 0.021	0.74
% Silt	0.76			0.62			0.20
% Clay	0.76			0.62			0.20
% Sand	0.76			0.62			0.20
Total N (mg g−1)	0.01	−0.02	± 0.016	0.2			0.09
Organic C (mg g−1)	0.69			0.01	0.00	± 0.006	0.29
Chlorophyll A (mg m−3)	0.40			0.01	0.00	± 0.001	0.12
Phaophyton (mg m−3)	0.55			0.22			0.26

4. Discussion

Benthic macroinvertebrates play a crucial role in assessing the health of estuaries and coastal ecosystems due to their predictable responses to disturbance and numerous marine benthic indices that have been developed in recent decades (Diaz et al., 2004; Birk et al., 2012; Borja et al., 2019). However, before creating new indices it is important to evaluate and adapt existing ones, considering metrics and performance in different situations (Diaz et al., 2004; Borja and Dauer, 2008). This approach helps to avoid limiting data beyond project scope or biasing conclusions because of index selection (Gillett et al., 2015; Froján et al., 2016; Nestlerode et al., 2020). The M-AMBI (multivariate AMBI) is a robust measure of benthic habitat condition with potential to meet this need, however more work is needed to validate the index under diverse conditions (Borja et al., 2009) and at different spatial and temporal scales (Pelletier et al., 2018). As such, the purpose of this study was to compare M-AMBI to regional indices (EMAP-E and GOM B-IBI) in a small temperate estuary in the Gulf of Mexico.

We generally observed poor agreement between indices. Numerical index scores for M-AMBI and GOM B-IBI were positively correlated which could indicate they are capturing similar response patterns in the benthic community. When translated into habitat condition, however, the indices disagree significantly as indicated by the very low Kappa scores. The EMAP-E was negatively correlated with GOM B-IBI and had no agreement in benthic habitat condition with either index. It had low resolution with most of the samples assigned “moderate” and no samples assigned “poor”. Disagreements between indices can occur due to differences in how taxa that are used as metrics respond to disturbance (Berthelsen et al., 2018) and impairment thresholds (Borja and Dauer, 2008). Univariate measures like EMAP-E and GOM B-IBI are calculated from weighted abundances of only a few taxonomic groups and are susceptible to bias from low sample size and patchy distributions. For example, in our study two groups used by the EMAP-E, Tubificidae and Amphipoda, were absent or near absent from samples and may therefore affect calculation of benthic habitat scoring. Nestlerode et al. (2020) also failed to find agreement between EMAP-E and GOM B-IBI, suggesting computational issues with the later due to low representation of north Florida during index development (TetraTech, 2011). In contrast, M-AMBI considers the benthic community as a whole by integrating species richness and Shannon diversity with weighted tolerances (AMBI) (Muxika et al., 2007; Pelletier et al., 2018), countering taxonomic limitations of EMAP-E and GOM B-IBI. However further evaluation of the M-AMBI is necessary as US validation data did not include the Gulf of Mexico (Pelletier et al., 2018).

To better understand index performance, natural environmental gradients in the Pensacola Bay estuary were used to contextualize the benthic community and compare community metrics (e.g., species richness, Shannon diversity) to habitat condition grades. As expected, the variable salinity of the upper bay is reflected as a species-poor benthic community dominated by subsurface feeders and low percentage of sensitive taxa. Conversely, the higher more stable salinity of the lower bay region supports a complex, heterogeneous benthic community. All of the indices we evaluated attempt to limit salinity bias in different ways. The M-AMBI and GOM B-IBI use discrete salinity zones with different calculations for each, while the EMAP-E uses the salinity of the individual sample in index calculation. All indices showed a similar pattern for salinity with better scores in higher salinities, while the M-AMBI also had a negative relationship with sediment total nitrogen and percent organic matter which could explain occurrences of poor habitat condition at lower bay stations where values are higher. The GOM B-IBI assigned a high number of sites in the upper bay as “good” and the lower bay as “poor” while the EMAP-E indicate little difference in habitat condition between regions, both of which are in conflict with the benthic community metrics. The M-AMBI followed the other benthic metrics with better habitat scores in higher salinities, although more studies are needed to determine if there is a low salinity bias or if “poor” habitat condition assignments in the upper bay is due to anthropogenic disturbance. The AMBI can have difficulty with natural disturbance (Muniz et al., 2005; Muxika et al., 2005) and naturally low abundance or species richness as is commonly seen in the lower salinities of estuaries (Gillett et al., 2015). The US M-AMBI extension was developed to ameliorate that bias (Pelletier et al., 2018), however, further calibration may be needed regionally or locally before adoption into monitoring programs.

Hypoxia is a stressor of concern in Pensacola Bay, which commonly stratifies over the summer resulting in episodic hypoxia in the middle bay (Hagy and Murrell, 2007). Each index responded to changes in DO, with M-AMBI being the most sensitive; however, mismatch where scores were low and DO was high or vice versa were observed. Dissolved oxygen was also significantly correlated to the NMDS, although we expected to see a stronger correlation compared to sediment parameters. It’s possible there could be a confounding effect from one or more co-occurring stressors we were unable to decouple from dissolved oxygen. Multiple stressors present a challenge to evaluating index response to hypoxia, particularly with single time point observations which assume that dissolved oxygen taken at the time of sampling is representative of conditions at that site, and thus may not adequately capture hypoxic conditions that can change quickly (Duvall et al., 2022). This presents further challenges to determining if benthic measurements are made following an episode of hypoxia and the community is in recovery or if observed low DO represents the onset of hypoxia and the benthos is in an early state of response (Coffin et al., 2018). This could account for some of the mismatch we observed between M-AMBI score and DO.

4.1. Management considerations

The M-AMBI has been adopted by the US EPA’s NCCA for evaluating marine benthic habitat across large stretches of coastline, albeit at lower spatial and temporal resolution than are sometimes needed by managers (EPA, 2015). To be useful for natural resource managers, indices need to respond to a variety of anthropogenic stressors across a range of natural environmental conditions, as well as demonstrate utility in different coastal environments (Borja and Dauer, 2008). This study demonstrates that M-AMBI can adequately handle natural gradients and respond to disturbance at a smaller scale, in large part, due to well calibrated EG values. The international EGs applicable to any soft-bottom coastal community have led to broad use of AMBI and M-AMBI. However, these EG assignments can hinder index performance as species sometimes exhibit dissimilar tolerances to the same stressor in different regions (Gillett et al., 2015; Robertson et al., 2016; Pelletier et al., 2018). Furthermore, EG values were originally developed to reflect benthic response to nutrient enrichment (Borja et al., 2000) and may not adequately capture other stressors. Subsequently, Gillett et al. (2015) compiled region-specific EG values for east, west, and gulf coasts in the United States while considering benthic response to all stressor types. Additionally, components of M-AMBI were further calibrated against sediment gradients, toxicity, and hypoxia (Pelletier et al., 2018). The relationship between M-AMBI and multiple stressors has been further validated in the US at national and ecoregion scales (Pelletier and Charpentier, 2023). Yet additional calibration of EG values or metrics may be needed at the local level, which would allow for catering the M-AMBI to individual needs while preserving the underlying calculations and maintaining comparability.

Abbreviations:

AMBI

AZTI Marine Biotic Index

M-AMBI

Multivariate AMBI

EMAP-E

Environmental Monitoring and Assessment Program-Estuaries

GOM B-IBI

Gulf of Mexico Benthic Index of Biotic Integrity

Data availability

Data will be made available on request.