- Email: [email protected]
Temporal stability of vegetation indicators of wetland condition
Ecological Indicators 34 (2013) 69–75
Contents lists available at SciVerse ScienceDirect
Ecological Indicators journal homepage: www.elsevier.com/locate/ecolind
Original article
Temporal stability of vegetation indicators of wetland condition Elizabeth Deimeke a,1 , Matthew J. Cohen a,∗ , Kelly C. Reiss b a b
School of Forest Resources and Conservation, University of Florida, Gainesville, FL 32611, United States Environmental Engineering Sciences, University of Florida, Gainesville, FL 32611, United States
a r t i c l e
i n f o
Article history: Received 1 November 2012 Received in revised form 12 April 2013 Accepted 26 April 2013 Keywords: Wetland condition Biological assessment Index of biological integrity IBI Temporal stability
a b s t r a c t Vegetation indices are widely employed to evaluate wetland ecological condition, and are expected to provide sensitive and specific detection of environmental change. Most studies evaluate the performance of condition assessment metrics in the context of the data used to calibrate them. Here we examined the temporal stability of the Florida Wetland Condition Index (FWCI) for vegetation of depressional forested wetlands by resampling sites in 2008 that were previously sampled to develop the FWCI in 2001. Our objective was to determine if FWCI, a composite of six vegetation-based metrics, provides a robust measure of condition given inter-annual variation in environmental conditions (i.e., rainfall) between sampling periods. To that end, we sampled 22 geographically isolated wetlands in north Florida that spanned a wide land use/land cover intensity gradient. Our results suggested the FWCI is robust. We observed no significant paired difference in FWCI across or within land use categories, and the relationship between FWCI in 2001 and 2008 was strong (r2 = 0.88, p < 0.001). This was despite surprisingly high composition change. Mean Jaccard community similarity within sites between years was 0.30, suggesting that most of the herbaceous taxa were replaced, possibly because of different antecedent rainfall conditions or sampling during different phenological periods; both are contingencies to which condition indices must be robust. We did observe some evidence of convergence toward the mean in 2008, with the fitted slope relating 2001 and 2008 FWCI scores significantly below one (0.63, 95% CI = 0.53–0.73). The most variable FWCI component metric was the proportional representation of obligate wetland taxa, suggesting that systematic changes may have been induced by different hydrologic conditions prior to sampling; notably, however, FWCI computed without this component still exhibited a slope significantly less than 1 (0.72, 95% CI = 0.61–0.88). Moreover, there was evidence that species lost from reference sites (higher condition) were replaced by taxa of lower floristic quality, while species lost from agricultural sites (consistently the lowest condition land use category) were replaced by species of higher quality. A significant positive association between FWCI and the ratio of coefficients of conservatism (CC) of species lost to those gained suggests some overfitting in FWCI development. However, despite modest evidence of overfitting, FWCI provides temporally consistent estimates of wetland condition, even under conditions of substantial taxonomic turnover. © 2013 Elsevier Ltd. All rights reserved.
1. Introduction Wetlands provide a variety of important goods and services (Costanza et al., 1997; MEA, 2000), delivery of which may be affected by anthropogenic impacts. In recognition of their value and modern decline in extent, wetlands are increasingly assigned special protections against these impacts. For example, beginning in 1988 there has been an effort by national policies to encourage no net loss of wetlands in the United States. A result of this attention to protecting wetland ecosystems is the widespread development and
∗ Corresponding author. Tel.: +1 352 846 3490; fax: +1 352 846 1277. E-mail addresses: [email protected] (E. Deimeke), mjc@ufl.edu (M.J. Cohen), kcr@ufl.edu (K.C. Reiss). 1 Present address: 633 Moreland Avenue #1, Atlanta, GA 30307, United Sates. 1470-160X/$ – see front matter © 2013 Elsevier Ltd. All rights reserved. http://dx.doi.org/10.1016/j.ecolind.2013.04.022
maturation of assessment tools that are used to evaluate wetland condition, impact, restoration and mitigation. Wetland condition assessments measure site departure from conditions observed in minimally impacted settings (Karr and Chu, 1997). This facilitates understanding of the variety of wetlands in the landscape and prioritizes sites for conservation or restoration efforts, but the central use of these assessment scores is in determining the level of mitigation necessary if the wetland was degraded or removed from the landscape (Fennessy et al., 2004). Many wetland condition assessment tools are based on rapid site reconnaissance combined with landscape setting (e.g., Brown and Vivas, 2005; Mack, 2006; Reiss and Brown, 2007), but more intensive assessments of condition typically investigate the biota that inhabit a site. Indices of biological integrity (IBIs) compare the composition of minimally impacted reference sites to those that span a gradient of human impacts (Karr and Chu, 1997); multiple
70
E. Deimeke et al. / Ecological Indicators 34 (2013) 69–75
axes are possible, including the prevalence of certain taxonomic groups and/or functional guilds. Examples include the use of benthic invertebrates for the assessment of stream condition (e.g., Southerland et al., 2008) and the use of vegetation and/or fish as an index of lake condition (e.g., Beck and Hatch, 2009). Strong benthic invertebrate metrics have not been consistently developed for wetland systems, so most attention has been focused on the autotrophic communities (e.g., diatoms – Lane and Brown, 2007, vascular plants – Lopez and Fennessy, 2002; Miller et al., 2006). These have generally shown strong covariance with human disturbance gradients, and benefit from their basis in minimally ambiguous taxonomic inventories. The use of vegetative communities in wetland condition assessment implicitly assumes stability in the quality if not the composition of wetland communities. These communities are temporally dynamic with natural background rates of species recruitment and displacement (e.g., by competition, herbivory, successional trajectories, or environmental variability). They are also subject to changes in response to anthropogenic stressors. Clearly, the goal is enumerate vegetation indices that are sensitive to human disturbances but also specific to those changes; in other words, condition assessment scores should respond predictably to one set of community-level changes or stressors. Indeed, it would be highly informative and would explicitly validate the approach were condition scores to remain constant at a site with unchanging land use in spite of composition changes over time. Despite the critical importance of vegetation indices for evaluating wetland condition, there is little evidence to evaluate their sensitivity and specificity. The objective of this study was to assess changes in vegetation condition scores over time, comparing an early set of measurements from 2001 with more recent measurements in 2008 at sites where land use remained unchanged. Our focus was on the Florida Wetland Condition Index (FWCI), a composite of six vegetation-based metrics developed to assess geographically isolated depressional forested wetlands (Reiss, 2004, 2006). If, as we predict, FWCI is a robust measure of wetland condition (i.e., both sensitive and specific), both overall condition scores and the six individual component metrics from 2008 should not be systematically different from 2001, even where taxonomic changes occurred. 2. Methods 2.1. Study area and site selection Twenty-two palustrine depressional forested wetlands (principally cypress domes) in north Florida were visited between July and October in 2008 (Fig. 1); all sites had been sampled in 2001/2002 (hereafter 2001) also between May and August as part of a statewide effort that culminated in development of the FWCI (Reiss and Brown, 2007). Sampling Julian dates differed between sites, with all sites sampled later in the year in 2008 than in 2001. Mean Julian day for 2001 sampling was 178, while it was 261 for 2008; the difference ranged from 4 to 152 days, with a mean of 83 days later in 2008. Sites were located in three land use settings identified in 2001 as “reference” (n = 12), “agricultural” (n = 4), and “urban” (n = 6) based on land use within a 500 m wetland buffer (Reiss, 2004); land use was verified visually during 2008 site visits and no changes in site land use categorization were observed. 2.2. Florida Wetland Condition Index: past and present All sites were evaluated using the FWCI, developed in 2001 using biological observations at these and other sites across Florida (Reiss, 2006). For the vegetation survey, wetland boundaries were approximated using hydrology and vegetation (Reiss, 2006). Four
Fig. 1. Location of 22 isolated wetland sites in north Florida, USA. Gray circles represent agricultural sites (n = 4), white circles are reference standard sites (n = 12), and black circles are urban sites (n = 6). All sites fell within the same statewide ecoregion (North Region; Lane, 2000).
1-m wide belt-transects radiating in cardinal directions from the center to edge of each wetland were used to measure community composition; no attempt was made to ensure consistent transect length between sampling periods. Each transect was subdivided into 5-m sections in which presence/absence of each taxa was recorded, along with ancillary information about each species (growth form: aquatic, fern, grass, herb, sedge, shrub, tree, or vine; annual/perennial; evergreen/deciduous; and native/exotic) based on Wunderlin and Hansen (2008). FWCI scores were computed from six floristic metrics as described and defined in Reiss (2004): (1) proportion of tolerant indicator species; (2) proportion of sensitive indicator species; (3) floristic quality assessment index (FQAI); (4) proportion of exotic species; (5) proportion of native perennial species; and (6) proportion of species with obligate or facultative-wetland status. All metrics excluding FQAI were calculated as site proportions in which the number of taxa in each metric (e.g., tolerant, sensitive, exotic) (N) was divided by site species richness (R). FQAI is the sum of coefficients of conservatism (CC) scores for all species divided by site species richness (i.e., omitting any effect of species richness on floristic quality):
FQAI =
(C1 + C2 + . . . Cn ) R
CC scores were determined by a survey of expert Florida botanists in 2001 according to species fidelity to a gradient of wetland conditions, and range from 0 (opportunistic invaders) to 10 (species with high affinity for reference standard wetland conditions) modeled on earlier FQAI work by Wilhelm and Ladd (1988). Values of CC scores reported in Reiss (2004) were used to compute FQAI in this study. Vegetation not identified to species level was excluded from metric calculations because CC, nativity status and tolerance/sensitivity classifications, all integral to FWCI calculation, are species specific. The original implementation of FWCI scaled each
E. Deimeke et al. / Ecological Indicators 34 (2013) 69–75
71
metric from 0 to 10. The same metric scaling used in 2001 was used for 2008 so results for each site are directly comparable. Final FWCI scores were the sum across the six metrics, with scores ranging from 0 to 60, with higher values indicating superior condition. 2.3. Data analysis We used paired t-tests to test the hypothesis that mean FWCI scores and individual metrics would not be significantly different between sampling years (2001 vs. 2008). We used ordinary least squares linear regression to determine the strength and slope of the association between FWCI measured in 2001 and 2008. Analysis of variance (ANOVA) was used to test differences in mean condition scores by land use for FWCI; post hoc Scheffe comparisons were used to evaluate pairwise differences (e.g., reference vs. urban). We also evaluated changes within each of the individual FWCI metrics to further explore sources of temporal variation. All calculations were completed using Microsoft Excel v. 12.0.0. To test whether species changes over time were neutral with respect to floristic quality, we evaluated the relationship between coefficients of conservatism of species lost versus species gained as a function of site FWCI score. Specifically, we regressed FWCI versus the ratio of CC for species lost to CC of species gained. A positive relationship indicates that high condition sites lost species with relatively higher CC than those they gained (i.e., overall quality declined), while low condition sites lost species of lower floristic quality than those they gained. Conceptually, this indicates the degree of convergence of FWCI scores over time, which could occur either because of broad landscape degradation or due to over-fitting (excess specificity) of the original FWCI metrics to the calibration data. Stable FWCI scores over time may indicate limited variation in community composition over time, or, alternatively, that in spite of community composition shifts, community assembly selects for new taxa of similar quality. To evaluate the source of FWCI stability, we evaluated vegetation composition similarity between years (2001 vs. 2008) as well as within years (overall and within land use categories) using Jaccard’s index of similarity (Wallwork, 1976): J=
j r
where j is the number of species common to both sites or sampling years and r is the combined species richness. Low similarity values were interpreted as changes in vegetation composition; within site dissimilarity between 2001 and 2008 was evaluated in reference to between site differences. We evaluated similarity using both the entire community and for those more common taxa that occurred at a minimum of five sites in either year. Finally, ordinary least squares linear regression was used to examine the relationship between Julian day sampling difference and similarity between years. 3. Results Land use did not change during the seven-year interval between sampling periods at any site. Although human disturbance scores were not recalculated, visual inspection on aerial imagery and during field visits confirmed land use categories in 2008 were the same as those assigned in 2001. The 2001 sites were sampled, on average, 83 days earlier in the year than in 2008 (±41 SD) with a range of four to 152 days. The mean Julian day of sampling was 178 (end of June) in 2001, and 261 (mid-September) in 2008. Antecedent rainfall in north Florida was different between the sampling periods, with annual rainfall prior to sampling in summer 2001 (96 cm at Gainesville Regional Airport, the approximate center of the study
Fig. 2. Mean FWCI scores by land use from the 2001 and 2008 assessments. For both years, letters indicate significant differences between land uses. Error bars represent one standard error (n = 22; 2001: ANOVA, F = 18.9, p < 0.005; 2008: ANOVA, F = 20.6, p < 0.0001). All comparisons across years were not significant.
site) well below normal (ca. 122 cm), in contrast with normal rainfall prior to the 2008 sampling period (123 cm). There was broad evidence of significant community changes between 2001 and 2008. During the 2001 survey, 248 different taxa were encountered, whereas only 167 were found in 2008. Moreover, the overlap in this composition was relatively modest, with only 109 species found in both years. Consequently, we observed a significant (p = 0.007) decline in mean species richness between 2001 and 2008, with values (±SD) of 33.1 ± 14.0 in 2001 versus 27.1 ± 9.0 in 2008. Species richness was significantly negatively correlated with wetland condition (FWCI) in 2008 (r = −0.62, p = 0.002), but this association was not significant in 2001. There was evidence of land use effect on richness across dates. Specifically, in 2001, reference sites had the lowest mean richness (30.5 ± 13.6), followed by urban (33.8 ± 6) and agricultural (39.8 ± 23) sites; in 2008, the same trends held, but the differences were statistically significant (p = 0.009) with reference site richness (22.1 ± 6.4) lower than agricultural sites (34.2 ± 9.9) and urban sites (32.3 ± 7.8), which were not different. Loss of richness between 2001 and 2008 occurred most in reference sites (mean decline = 8.9 species, p = 0.001), while richness declines in urban (mean = 2.3 species) and agricultural (mean = 5.5 species) sites were not significant. 3.1. Florida Wetland Condition Index The mean FWCI score in 2008 across the 22 sites was 36.5 (±13.6), where a value of 60 denotes the highest possible wetland condition; this compared favorably with the mean value for the same sites from 2001, which was 37.4 (±18.3). However, the range (48.5 versus 58.4) and standard deviation (13.7 versus 18.5) of FWCI scores in 2008 was narrower than for 2001. Despite this decline in the spread of FWCI scores, mean FWCI scores were not significantly different between sampling periods (p = 0.44) despite evidence that community composition varied. Within sampling years, FWCI differed significantly between land uses (Fig. 2). In both 2001 and 2008, agricultural sites had the lowest FWCI scores (13.1 ± 17.4 and 17.4 ± 13.2, respectively), followed by urban sites (28.4 ± 11.2 and 30.9 ± 8.6), and reference sites (49.1 ± 10.4 and 45.7 ± 5.5) (ANOVA, F = 20.6, p < 0.0001). All pairwise comparisons of land use were significant in 2001, but in 2008 significant pairwise differences were observed for agricultural versus reference sites, and reference versus urban sites, but not for agricultural versus urban sites (p = 0.15). In comparing FWCI scores within land use category across
72
E. Deimeke et al. / Ecological Indicators 34 (2013) 69–75 Table 1 Jaccard similarity indices (mean ± 95% CI) between sites (for 2001 and 2008, grouped by land use) and within sites (2001 vs. 2008, grouped by land use).
Fig. 3. FWCI scores in 2001 vs. 2008; linear regression was strongly siginficant (p < 0.0001), but the fitted slope (95% CI = 0.53–0.73) is significantly less than one (dashed line).
years, no changes were statistically significant. There was, however, a trend where agricultural sites (low FWCI) increased from 2001 to 2008 (mean = 4.4) and reference sites decreased (mean = −3.3); this movement toward the global mean in 2008 is further explored below. FWCI condition scores were highly correlated between sampling periods (Fig. 3; R2 = 0.88, p < 0.001); however, the fitted slope was significantly below one (0.68 ± 0.13, 95% CI), implying a systematic convergence in scores toward median values. To further underscore this trend, we note that the change in FWCI score between 2001 and 2008 was significantly negatively correlated with the 2001 score (FWCI = 11.46 − 0.32 × FWCI2001 ; r2 = 0.54, p < 0.001), suggesting a unit change in predicted score for every three unit change in 2001 FWCI score. The standard error of the linear regression estimate was 5.6 suggesting scores can be accurately predicted within ±6 FWCI points; the standard error obtained using the 1:1 line was 7.9. 3.2. Community similarity Jaccard metrics of community composition comparing sites within and across land use and time suggest general low similarity (Table 1). Overall, Jaccard similarity averaged 0.3 (±0.03 SD) between 2001 and 2008 with no apparent effect of land use (Table 1). To place the implied community changes in context, we also evaluated Jaccard similarity across sites, but within land use, for both 2001 and 2008. The results indicate extremely low similarity in both 2001 (mean = 0.18 ± 0.01) and 2008 (mean = 0.19 ± 0.01).
Land use
Between sites (2001)
Within sites (2001 vs. 2008)
Between sites (2008)
Reference Agriculture Urban
0.22 ± 0.01 0.17 ± 0.06 0.21 ± 0.03
0.31 ± 0.04 0.26 ± 0.11 0.28 ± 0.04
0.21 ± 0.01 0.22 ± 0.07 0.28 ± 0.03
Overall
0.18 ± 0.01
0.30 ± 0.03
0.19 ± 0.01
While within site similarity was much higher than across site similarity, the implied marked temporal community shifts reinforce the general findings that the species present at the study sites changed substantially between 2001 and 2008. When rare species (those present at less than five sites) were omitted, the similarity metrics rose (mean = 0.35 ± 0.03), but the general trend was retained. One explanation for low similarity is sampling during different phases of wetland phenology. Sites were sampled later in the year in 2008 than in 2001 (mean = 83 days). However, we observed no correlation between Jaccard similarity and the Julian day difference between sampling, which ranged from 4 to 152 days (r2 = 0.03, p = 0.46). As such, low similarity cannot be explained by the different sampling dates. 3.3. Individual FWCI metrics In spite of large differences in composition, we observed relatively consistent FWCI scores. To better understand this important finding, we evaluated the six equally weighted metrics that comprise FWCI between sampling years (Table 2). In general, the individual metrics aligned over time, with relatively small absolute changes for all metrics, except wetland status. Moreover, the magnitude of disagreement between sampling periods was relatively consistent across land use categories. The correlation between metric values in 2001 and 2008 was generally strong, and the slope of the relationship was always near 1, with the exception of the % sensitive metric which was higher in 2008. The main exception to broad agreement across years was for wetland status, for which the correlation between 2001 and 2008 across all sites was not statistically significant and the fitted slope was much less than 1. We observed almost no change in the proportion of tolerant or sensitive species, despite significant turnover in the composition. The metric focused on the proportion of exotic taxa was also unchanged between years, though in this case the representative non-native taxa were relatively well conserved. The native perennial metric was modestly different between sampling years with values of 7.5 in 2001 and 8.1 in 2008 (p = 0.02). This shift may be due to differences in sampling season or wetter antecedent hydrologic conditions in 2008, which would reduce the number of annuals. While the shifts in wetland status were large for many
Table 2 Summary of changes in individual components of FWCI between 2001 and 2008. The mean absolute error (MAE) denotes the magnitude of disagreement between sampling periods, summarized for within and across land use categories. The Pearson correlation coefficient indicates the consistency of values between 2001 and 2008; significant correlations (p < 0.05) are in bold and where correlations were not significant the p-value is presented in parentheses. The linear slope for the overall association is also shown; bold values are significantly different (p < 0.05) from 1.0. Metrica
%Tolerant %Sensitive % Exotic FQAI %Native Perennial %Wetland Status a
Overall
Agricultural
Reference
Urban
MAE
r
Slope
MAE
r
MAE
r
MAE
r
1.2 1.5 0.9 1.4 1.1 2.8
0.90 0.89 0.91 0.87 0.91 0.36 (0.09)
1.15 1.25 0.98 1.18 1.01 0.56
1.2 2 0.4 0.6 1.1 3.1
0.70 0.80 0.76 0.79 0.77 0.38 (0.26)
1.4 1.5 0.8 1.7 0.9 2.6
0.68 0.80 0.30 (0.14) 0.46 0.11 (0.46) 0.05 (0.69)
0.9 1.1 1.3 1.3 1.3 2.9
0.91 0.94 0.73 0.42 0.05 (0.77) 0.32 (0.21)
All metrics are reported after rescaling (Reiss, 2004), which created a range from 0 to 10 for 2001 data but a lower range for some metrics for 2008.
E. Deimeke et al. / Ecological Indicators 34 (2013) 69–75 Table 3 Mean CC scores of species lost (CClost ) or gained (CCgained ) between 2001 and 2008. Mean values for both lost and gained taxa differed significant by land use in both years (p < 0.0001). However, a significant difference between gained and lost values was observed only for reference sites (p = 0.03). Values of CClost :CCgained departed significantly from one for reference sites only. Values are means (±95% CI). Land use
Mean CCgained
Mean CClost
CClost :CCgained
Agriculture Reference Urban Overall
2.84 (1.04) 3.88 (0.36) 3.33 (0.47) 3.54 (0.47)
2.56 (1.29) 4.49 (0.29) 3.18 (0.40) 3.78 (0.44)
0.89 (0.27) 1.21 (0.18) 0.99 (0.21) 1.09 (0.13)
Fig. 4. Significant positive relationship between FWCI scores from 2001 and the ratio of CC scores of species lost to species gained from 2001 to 2008 (p = 0.0009). This suggests higher condition sites gained species of lower conservatism coefficient that they lost, and vice versa.
of the sites, it is important to note that they were not directional; the mean change was small (−0.5) despite large absolute shifts. While these changes may be responding to hydrology, it is likely that this is relatively localized, and not systematic in response to drought conditions that prevailed during the 2001 sampling. When we computed FWCI without wetland status as a subcomponent, we obtained results between measurements in 2001 and 2008 that were similar in strength to the full FWCI (r2 = 0.89, p < 0.001), but where the slope was closer to, but still significantly less than, unity (0.72, 95% CI = 0.61–0.84). The last individual metric, the FQAI was significantly lower in 2008 (4.6 ± 2.2) compared with 2001 (5.7 ± 3.3) (p = 0.001), suggesting a modest shift toward species with lower CC. This overall effect was largely a result of changes in species composition within reference sites (Table 3), which was the only land use category to show significant differences between the CC for species gained (3.88) versus species lost (4.49) (p = 0.03). Despite significant species turnover, CC for urban sites did not change (CClost :CCgained ∼ 1.0; Table 3) and agricultural sites showed a nonsignificant trend toward higher CC. Combining these effects, we observed a statistically significant positive relationship between FWCI in 2001 and the ratio of CClost :CCgained (Fig. 4). This suggests convergence toward mean FQAI as sites with high FWCI scores tended to lose species with higher CC than were recruited. Similarly, sites with low FWCI scores tended to lose species with lower CC scores than those recruited. 4. Discussion The most important finding of this work is that wetland condition, as measured by FWCI, did not change substantially between sampling years. Because land use categories determined in 2001
73
were field verified as unchanged in 2008, the strong correlation observed in FWCI scores between years supports the contention that land use is the commanding variable controlling wetland condition (Brown and Vivas, 2005). It also suggests that this metric is robust to both variation in the sampling time within a year and the natural interannual variation that is characteristic of these geographically isolated wetlands. Notably, neither urban nor agricultural sites showed a downward trajectory in condition as might be expected as these wetlands continue to adjust to their relatively new surroundings. This suggests that even these systems are at something that approximates a dynamic equilibrium with regard the quality of vegetation community despite dramatic climate variation between the two sampling events. Despite relatively consistent FWCI scores over time, this study also observed surprising dissimilarity in vegetation composition between sampling years. Indeed, our results suggest that vegetation differed in all land use classes by ca. 70% between 2001 and 2008, a result that was robust to the exclusion of rare species (those occurring at less than five sites). One potential explanation for this difference is evaluator bias. While we cannot rule this possibility out entirely, we note that both evaluators were trained botanists, and consulted the same taxonomic references. Perhaps more importantly, the species lists from 2001 were available to evaluators in 2008, which would generally lead to higher similarity given prior information to guide identification of unknown species. This substantial shift in species composition is likely real, and is comparable to other studies using Jaccard index similarities in wetlands, albeit when comparing across, not within, sites. For example, Hartzell et al. (2007) reported Jaccard index similarities for vegetation of 38% between created and natural depression wetlands. What is notably similar to our results is that this large discrepancy in taxonomic composition was not evident in the twelve condition metrics used to compare the two systems. Similarly, invasion of two taxa, non-native Phragmites australis and Typha spp., in coastal wetlands on the Great Lakes caused a 19% similarity (Jaccard’s index) between 2001 and 2004 (T’ulbure et al., 2007). While the authors correctly point to this striking dissimilarity, FQAI and CC scores were actually greater after the two species were established. These results illustrate that the list of species that occupy a site at any given time may be highly dynamic, responding to water level and seasonal variations that are independent of the anthropogenic disturbance gradients that are the focus of condition assessment; this underscores the utility of condition scores that integrate over the entire community. This does not discount the insights that can be gleaned by turnover in species composition, but supports a cautious approach to inference about site condition based on that evidence alone. Because biological assessment cannot feasibly be implemented in the same phenological period for all sites, the need to develop and employ metrics that are robust to natural variation induced by seasons is clear. Species lists are unlikely to provide that robustness, but our work suggests that FWCI does. While FWCI scores between 2001 and 2008 were strongly correlated, we did observe significant evidence of a convergence of FWCI scores across sites in 2008. That is, sites with high scores in 2001 tended to score lower in 2008, while sites with low scores in 2001 tended to score higher in 2008. This convergence effect, manifest as a decline in range between 2001 and 2008 and a fitted slope between 2001 and 2008 significantly less than 1 (Fig. 3), results in a loss of extreme wetland condition scores. The most likely explanation for this effect is statistical over-fitting of 2001 FWCI scores. The 5th to 95th percentile of individual FWCI metrics were normalized to occupy the entire score range of 0–10 (Reiss, 2004). The current scores were derived as FWCI would be field applied, without stretching minimum or maximum scores to the 0 or 10 endpoints. FWCI development was based on a snapshot in time, and the normalization process establishes conditions at that time
74
E. Deimeke et al. / Ecological Indicators 34 (2013) 69–75
as the benchmark; observed variation in composition underscores that these systems are compositionally dynamic, making it highly likely that the FWCI formulation is over-specified to one set of conditions; this effect is modest in magnitude for any given site, but operationally relevant when sampling a population of wetlands. The convergence in scores in 2008 might occur in response to systematic shifts in vegetation that would affect all six elements of the FWCI. It was striking, therefore, to observe that five metrics (i.e., tolerant indicator species, sensitive indicator species, exotic species, FQAI, and native perennial species) were highly correlated between years with slopes in almost all cases not significantly different from 1, contributing to inference of overall stability in total FWCI scores. However, the sixth metric, the average wetland status of the species present within each site, showed significant, albeit directionally inconsistent, disagreement between sampling years. While changes were large at many sites there was no systematic loss or gain of obligate and facultative wetland species. Rather, both loss and gain occurred between sampling years. It is likely this effect can be ascribed to variation in hydrologic conditions just prior to sampling. The 2001 survey occurred in the midst of a regional drought in which annual rainfall was well below normal. In 2008, the annual rainfall was near the long term mean of 123 cm yr−1 . However, there was also within-year hydrologic variation, with the latter half of the 2001 sampling occurring under wetter conditions and the early part of the 2008 sampling occurring during wetter conditions. In fact, when sites were separated into 2001 and 2002 sampling years, those sampled in 2002 were significantly more similar to 2008 (mean similarity = 0.41) than the 2001 sites (mean similarity = 0.29) (p = 0.02). This may be due to easing of the drought in 2002 creating hydrologic conditions more similar to those in 2008, and thus creating greater species composition similarity. However, it may also be due to shorter time lag between 2002 and 2008 than between 2001 and 2008. While a non-uniform shift in the overall importance of obligate and facultative wetland species likely cannot account for convergence of FWCI scores, within year variation is an area that merits continued attention. Examining the FQAI metric revealed that most of the species lost were of comparable quality to those gained, suggesting robust correspondence between quality scores and land use stressors. While the mean CC of species lost or gained was similar (3.78 vs. 3.54, respectively), the ratio of CClost :CCgained was significantly positively correlated with overall 2001 FWCI scores. That is, wetlands with high FWCI scores, predominantly reference sites, tended to lose species of high CC, and gain species of modestly lower CC, suggesting an overall loss of species quality within reference sites. Conversely, wetlands with low FWCI scores, mainly agricultural sites, tended to lose species of low CC, and gain species of higher CC, which implies an overall gain of species quality within agricultural sites. The ultimate effect of this shift in vegetation was a loss of sites with extreme CC, contributing to an overall convergence of FWCI scores in 2008, and an overall moderation of vegetative quality in these sites. It is unclear if other assessment methods show similar convergence over time because the temporal stability of assessment metrics is not widely reported. Additional research is warranted on the repeatability of CC scores as indicators of wetland condition and the role of FQAI metrics in assessing wetland condition. This is particularly important because species CC scores are assigned by a panel of experts and are thus highly variable (Cohen et al., 2004). It is rare that wetland condition is repeatedly assessed over time, particularly without a change in disturbance, environmental impacts, or land use. Consequently, most literature focuses on condition due to gradients in anthropogenic disturbance, i.e., land use change or logging (Lopez and Fennessy, 2002; Mack, 2007; Miller and Wardrop, 2006; Reiss and Brown, 2007). The current
study fills an important knowledge gap regarding condition variability over time in the absence of changing driving forces, and demonstrates a need for long-term monitoring of condition and vegetation beyond what is currently mandated for mitigation wetlands (i.e., created, restored, preserved or enhanced wetlands; Reiss et al., 2007), which are often compared against other regional reference wetlands or a hypothetical reference condition in determining mitigation success. Understanding natural variability in condition and vegetation will further reinforce the validity of using benchmark sites against which mitigation wetlands can be compared, particularly in light of the low site community similarity between years. The fact that the condition scores were relatively consistent across years in spite of marked changes in composition means that wetland condition is conserved by a “portfolio” effect wherein some sub-components may decline in response to natural environmental variation, but others increase. This lends strong support to the multi-variate nature of FWCI and other wetland assessment tools, and suggests that evaluating the portfolio of services may be more informative than focusing on one service or metric. Expanding this research beyond a small sample of wetlands in one wetland region and of one specific wetland class (i.e., palustrine geographically isolated depressional forested wetlands) would strengthen our collective understanding of temporal variability in the wetland plant community. Expanding the focus beyond vegetation to other trophic levels with established IBIs would also enhance knowledge of natural variability for wetland biotic communities along a human disturbance gradient.
5. Conclusions As a foremost conclusion, this research suggests support for the use of vegetation based condition metrics in assessing Florida’s geographically isolated wetlands. This is most well supported by the observation that index values were robust to a high level of taxonomic turnover. However, several more nuanced conclusions remain, specifically with regard to shifting mean CC value for both reference and impacted wetlands. Significantly lower species richness and community dissimilarity between years also expands the uncertainty in adopting IBI metrics that are based on straight richness or diversity calculations.
Acknowledgements This project was funded through support from the UF-College of Agriculture and Life Sciences and the Florida Department of Environmental Protection (#WM-683). We gratefully acknowledge the field assistance of Chad Foster, Jorge Guevara, Jeff Hull, Larry Korhnak, Dina Liebowitz, Danielle Liebenow, Lauren Long, Danielle Watts, and Justin Vogel, as well as Mark Brown, the principal investigator for the FWCI project. We also acknowledge private landowners that permitted repeated site access.
References Beck, M.W., Hatch, L.K., 2009. A review of research on the development of lake indices of biotic integrity. Environ. Rev. 17 (NA), 21–44. Brown, M.T., Vivas, M.B., 2005. Landscape development intensity index. Environ. Monit. Assess. 101, 289–309. Cohen, M.J., Carstenn, S., Lane, C.R., 2004. Floristic quality indices for biotic assessment of depressional marsh condition in Florida. Ecol. Appl. 14 (3), 784–794. Costanza, R., d’Arge, R., deGroot, R., Farber, S., Grasso, M., Hannon, B., Limburg, K., Naeem, S., O’Neill, R.V., Paruelo, J., Raskin, R.G., Sutton, P., van den Belt, M., 1997. The value of the world’s ecosystem services and natural capital. Nature 387, 253–260. Fennessy, M.S., Jacobs, A.D., Kentula, M.E., 2004. Review of Rapid Methods for assessing Wetland Condition. EPA/620/R-04/009. U.S. Environmental Protection Agency, Washington, DC.
E. Deimeke et al. / Ecological Indicators 34 (2013) 69–75 Hartzell, D., Bidwell, J.R., Davis, C.A., 2007. A comparison of natural and created depressional wetlands in central Oklahoma using metrics from indices of biological integrity. Wetlands 27, 794–805. Karr, J.R., Chu, E.W., 1997. Biological monitoring: essential foundation for ecological riskassessment. Hum. Ecol. Risk Assess. 3, 993–1004. Lane, C.R., 2000. Proposed ecological regions for freshwater wetlands of Florida. University of Florida, Environmental Engineering Sciences, Gainesville, FL (Masters Thesis for Masters of Science). Lane, C.R., Brown, M.T., 2007. Diatoms as indicators of isolated herbaceous wetland condition in Florida, USA. Ecol. Ind. 7, 521–540. Lopez, R.D., Fennessy, M.S., 2002. Testing the floristic quality assessment index as an indicator of wetland condition. Ecol. Appl. 12, 487–497. Mack, J.J., 2006. Landscape as a predictor of wetland condition: an evaluation of the landscape development intensity index (LDI) with a large reference wetland dataset from Ohio. Environ. Monit. Assess. 120, 121–241. Mack, J.J., 2007. Developing a wetland IBI with statewide application after multiple testing iterations. Ecol. Indic. 7, 864–881. MEA (Millenium Ecosystem Assessment), 2000. Ecosystems and Human WellBeing: Wetlands and Water Synthesis. World Resources Institute, Washington, DC. Miller, S.J., Wardrop, D.H., 2006. Adapting the floristic quality assessment index to indicate anthropogenic disturbance in central Pennsylvania wetlands. Ecol. Indic. 6, 313–326. Miller, S.J., Wardrop, D.H., Mahaney, W.M., Brooks, R.P., 2006. A plant-based index of biological integrity (IBI) for headwater wetlands in central Pennsylvania. Ecol. Indic. 6, 290–312.
75
Reiss, K.C., 2004. Developing biological indicators for isolated forested wetlands in Florida. University of Florida, Environmental Engineering Sciences (Dissertation for Doctor of Philosophy). Reiss, K.C., 2006. Florida Wetland Condition Index for depressional forested wetlands. Ecol. Indic. 6, 337–352. Reiss, K.C., Brown, M.T., 2007. Evaluation of Florida Palustrine Wetlands: application of USEPA levels 1, 2, and 3 assessment methods. EcoHealth 4, 206–218. Reiss, K.C., Hernandez, E., Brown, M.T., 2007. An evaluation of the effectiveness of mitigation banking in Florida: Ecological success and compliance with permit criteria. FDEP Final Report #WM881. U.S. EPA Region Four #CD 96409404-0. Southerland, M.T., Rogers, G.M., Kline, M.J., Morgan, R.P., Boward, D.M., Kazyak, P.F., Klauda, R.J., Stranko, S.A., 2008. Improving biological indicators to better assess the condition of streams. Ecol. Indic. 7, 751–767. T’ulbure, M.G., Johnston, C.A., Auger, D.L., 2007. Rapid invasion of a Great Lakes coastal wetland by non-native Phragmites australis and Typha. J. Great Lakes Res. 33, 269–279. Wallwork, J.A., 1976. The Distribution and Diversity of Soil Fauna. Academic Press, London. Wilhelm, G., Ladd, D., 1988. Natural area assessment in the Chicago region. In: Transactions of the 53rd North American Wildlife and Natural Resource Conference, Louisville, Kentucky, Wildlife Management Institute, Washington, DC, USA, pp. 361–375. Wunderlin, R.P., Hansen, B.F., 2008. Atlas of Florida Vascular Plants. Institute for Systematic Botany, University of South Florida, Tampa [S. M. Landry and K. N. Campbell (application development), Florida Center for Community Design and Research] http://www.plantatlas.usf.edu/
Contents lists available at SciVerse ScienceDirect
Ecological Indicators journal homepage: www.elsevier.com/locate/ecolind
Original article
Temporal stability of vegetation indicators of wetland condition Elizabeth Deimeke a,1 , Matthew J. Cohen a,∗ , Kelly C. Reiss b a b
School of Forest Resources and Conservation, University of Florida, Gainesville, FL 32611, United States Environmental Engineering Sciences, University of Florida, Gainesville, FL 32611, United States
a r t i c l e
i n f o
Article history: Received 1 November 2012 Received in revised form 12 April 2013 Accepted 26 April 2013 Keywords: Wetland condition Biological assessment Index of biological integrity IBI Temporal stability
a b s t r a c t Vegetation indices are widely employed to evaluate wetland ecological condition, and are expected to provide sensitive and specific detection of environmental change. Most studies evaluate the performance of condition assessment metrics in the context of the data used to calibrate them. Here we examined the temporal stability of the Florida Wetland Condition Index (FWCI) for vegetation of depressional forested wetlands by resampling sites in 2008 that were previously sampled to develop the FWCI in 2001. Our objective was to determine if FWCI, a composite of six vegetation-based metrics, provides a robust measure of condition given inter-annual variation in environmental conditions (i.e., rainfall) between sampling periods. To that end, we sampled 22 geographically isolated wetlands in north Florida that spanned a wide land use/land cover intensity gradient. Our results suggested the FWCI is robust. We observed no significant paired difference in FWCI across or within land use categories, and the relationship between FWCI in 2001 and 2008 was strong (r2 = 0.88, p < 0.001). This was despite surprisingly high composition change. Mean Jaccard community similarity within sites between years was 0.30, suggesting that most of the herbaceous taxa were replaced, possibly because of different antecedent rainfall conditions or sampling during different phenological periods; both are contingencies to which condition indices must be robust. We did observe some evidence of convergence toward the mean in 2008, with the fitted slope relating 2001 and 2008 FWCI scores significantly below one (0.63, 95% CI = 0.53–0.73). The most variable FWCI component metric was the proportional representation of obligate wetland taxa, suggesting that systematic changes may have been induced by different hydrologic conditions prior to sampling; notably, however, FWCI computed without this component still exhibited a slope significantly less than 1 (0.72, 95% CI = 0.61–0.88). Moreover, there was evidence that species lost from reference sites (higher condition) were replaced by taxa of lower floristic quality, while species lost from agricultural sites (consistently the lowest condition land use category) were replaced by species of higher quality. A significant positive association between FWCI and the ratio of coefficients of conservatism (CC) of species lost to those gained suggests some overfitting in FWCI development. However, despite modest evidence of overfitting, FWCI provides temporally consistent estimates of wetland condition, even under conditions of substantial taxonomic turnover. © 2013 Elsevier Ltd. All rights reserved.
1. Introduction Wetlands provide a variety of important goods and services (Costanza et al., 1997; MEA, 2000), delivery of which may be affected by anthropogenic impacts. In recognition of their value and modern decline in extent, wetlands are increasingly assigned special protections against these impacts. For example, beginning in 1988 there has been an effort by national policies to encourage no net loss of wetlands in the United States. A result of this attention to protecting wetland ecosystems is the widespread development and
∗ Corresponding author. Tel.: +1 352 846 3490; fax: +1 352 846 1277. E-mail addresses: [email protected] (E. Deimeke), mjc@ufl.edu (M.J. Cohen), kcr@ufl.edu (K.C. Reiss). 1 Present address: 633 Moreland Avenue #1, Atlanta, GA 30307, United Sates. 1470-160X/$ – see front matter © 2013 Elsevier Ltd. All rights reserved. http://dx.doi.org/10.1016/j.ecolind.2013.04.022
maturation of assessment tools that are used to evaluate wetland condition, impact, restoration and mitigation. Wetland condition assessments measure site departure from conditions observed in minimally impacted settings (Karr and Chu, 1997). This facilitates understanding of the variety of wetlands in the landscape and prioritizes sites for conservation or restoration efforts, but the central use of these assessment scores is in determining the level of mitigation necessary if the wetland was degraded or removed from the landscape (Fennessy et al., 2004). Many wetland condition assessment tools are based on rapid site reconnaissance combined with landscape setting (e.g., Brown and Vivas, 2005; Mack, 2006; Reiss and Brown, 2007), but more intensive assessments of condition typically investigate the biota that inhabit a site. Indices of biological integrity (IBIs) compare the composition of minimally impacted reference sites to those that span a gradient of human impacts (Karr and Chu, 1997); multiple
70
E. Deimeke et al. / Ecological Indicators 34 (2013) 69–75
axes are possible, including the prevalence of certain taxonomic groups and/or functional guilds. Examples include the use of benthic invertebrates for the assessment of stream condition (e.g., Southerland et al., 2008) and the use of vegetation and/or fish as an index of lake condition (e.g., Beck and Hatch, 2009). Strong benthic invertebrate metrics have not been consistently developed for wetland systems, so most attention has been focused on the autotrophic communities (e.g., diatoms – Lane and Brown, 2007, vascular plants – Lopez and Fennessy, 2002; Miller et al., 2006). These have generally shown strong covariance with human disturbance gradients, and benefit from their basis in minimally ambiguous taxonomic inventories. The use of vegetative communities in wetland condition assessment implicitly assumes stability in the quality if not the composition of wetland communities. These communities are temporally dynamic with natural background rates of species recruitment and displacement (e.g., by competition, herbivory, successional trajectories, or environmental variability). They are also subject to changes in response to anthropogenic stressors. Clearly, the goal is enumerate vegetation indices that are sensitive to human disturbances but also specific to those changes; in other words, condition assessment scores should respond predictably to one set of community-level changes or stressors. Indeed, it would be highly informative and would explicitly validate the approach were condition scores to remain constant at a site with unchanging land use in spite of composition changes over time. Despite the critical importance of vegetation indices for evaluating wetland condition, there is little evidence to evaluate their sensitivity and specificity. The objective of this study was to assess changes in vegetation condition scores over time, comparing an early set of measurements from 2001 with more recent measurements in 2008 at sites where land use remained unchanged. Our focus was on the Florida Wetland Condition Index (FWCI), a composite of six vegetation-based metrics developed to assess geographically isolated depressional forested wetlands (Reiss, 2004, 2006). If, as we predict, FWCI is a robust measure of wetland condition (i.e., both sensitive and specific), both overall condition scores and the six individual component metrics from 2008 should not be systematically different from 2001, even where taxonomic changes occurred. 2. Methods 2.1. Study area and site selection Twenty-two palustrine depressional forested wetlands (principally cypress domes) in north Florida were visited between July and October in 2008 (Fig. 1); all sites had been sampled in 2001/2002 (hereafter 2001) also between May and August as part of a statewide effort that culminated in development of the FWCI (Reiss and Brown, 2007). Sampling Julian dates differed between sites, with all sites sampled later in the year in 2008 than in 2001. Mean Julian day for 2001 sampling was 178, while it was 261 for 2008; the difference ranged from 4 to 152 days, with a mean of 83 days later in 2008. Sites were located in three land use settings identified in 2001 as “reference” (n = 12), “agricultural” (n = 4), and “urban” (n = 6) based on land use within a 500 m wetland buffer (Reiss, 2004); land use was verified visually during 2008 site visits and no changes in site land use categorization were observed. 2.2. Florida Wetland Condition Index: past and present All sites were evaluated using the FWCI, developed in 2001 using biological observations at these and other sites across Florida (Reiss, 2006). For the vegetation survey, wetland boundaries were approximated using hydrology and vegetation (Reiss, 2006). Four
Fig. 1. Location of 22 isolated wetland sites in north Florida, USA. Gray circles represent agricultural sites (n = 4), white circles are reference standard sites (n = 12), and black circles are urban sites (n = 6). All sites fell within the same statewide ecoregion (North Region; Lane, 2000).
1-m wide belt-transects radiating in cardinal directions from the center to edge of each wetland were used to measure community composition; no attempt was made to ensure consistent transect length between sampling periods. Each transect was subdivided into 5-m sections in which presence/absence of each taxa was recorded, along with ancillary information about each species (growth form: aquatic, fern, grass, herb, sedge, shrub, tree, or vine; annual/perennial; evergreen/deciduous; and native/exotic) based on Wunderlin and Hansen (2008). FWCI scores were computed from six floristic metrics as described and defined in Reiss (2004): (1) proportion of tolerant indicator species; (2) proportion of sensitive indicator species; (3) floristic quality assessment index (FQAI); (4) proportion of exotic species; (5) proportion of native perennial species; and (6) proportion of species with obligate or facultative-wetland status. All metrics excluding FQAI were calculated as site proportions in which the number of taxa in each metric (e.g., tolerant, sensitive, exotic) (N) was divided by site species richness (R). FQAI is the sum of coefficients of conservatism (CC) scores for all species divided by site species richness (i.e., omitting any effect of species richness on floristic quality):
FQAI =
(C1 + C2 + . . . Cn ) R
CC scores were determined by a survey of expert Florida botanists in 2001 according to species fidelity to a gradient of wetland conditions, and range from 0 (opportunistic invaders) to 10 (species with high affinity for reference standard wetland conditions) modeled on earlier FQAI work by Wilhelm and Ladd (1988). Values of CC scores reported in Reiss (2004) were used to compute FQAI in this study. Vegetation not identified to species level was excluded from metric calculations because CC, nativity status and tolerance/sensitivity classifications, all integral to FWCI calculation, are species specific. The original implementation of FWCI scaled each
E. Deimeke et al. / Ecological Indicators 34 (2013) 69–75
71
metric from 0 to 10. The same metric scaling used in 2001 was used for 2008 so results for each site are directly comparable. Final FWCI scores were the sum across the six metrics, with scores ranging from 0 to 60, with higher values indicating superior condition. 2.3. Data analysis We used paired t-tests to test the hypothesis that mean FWCI scores and individual metrics would not be significantly different between sampling years (2001 vs. 2008). We used ordinary least squares linear regression to determine the strength and slope of the association between FWCI measured in 2001 and 2008. Analysis of variance (ANOVA) was used to test differences in mean condition scores by land use for FWCI; post hoc Scheffe comparisons were used to evaluate pairwise differences (e.g., reference vs. urban). We also evaluated changes within each of the individual FWCI metrics to further explore sources of temporal variation. All calculations were completed using Microsoft Excel v. 12.0.0. To test whether species changes over time were neutral with respect to floristic quality, we evaluated the relationship between coefficients of conservatism of species lost versus species gained as a function of site FWCI score. Specifically, we regressed FWCI versus the ratio of CC for species lost to CC of species gained. A positive relationship indicates that high condition sites lost species with relatively higher CC than those they gained (i.e., overall quality declined), while low condition sites lost species of lower floristic quality than those they gained. Conceptually, this indicates the degree of convergence of FWCI scores over time, which could occur either because of broad landscape degradation or due to over-fitting (excess specificity) of the original FWCI metrics to the calibration data. Stable FWCI scores over time may indicate limited variation in community composition over time, or, alternatively, that in spite of community composition shifts, community assembly selects for new taxa of similar quality. To evaluate the source of FWCI stability, we evaluated vegetation composition similarity between years (2001 vs. 2008) as well as within years (overall and within land use categories) using Jaccard’s index of similarity (Wallwork, 1976): J=
j r
where j is the number of species common to both sites or sampling years and r is the combined species richness. Low similarity values were interpreted as changes in vegetation composition; within site dissimilarity between 2001 and 2008 was evaluated in reference to between site differences. We evaluated similarity using both the entire community and for those more common taxa that occurred at a minimum of five sites in either year. Finally, ordinary least squares linear regression was used to examine the relationship between Julian day sampling difference and similarity between years. 3. Results Land use did not change during the seven-year interval between sampling periods at any site. Although human disturbance scores were not recalculated, visual inspection on aerial imagery and during field visits confirmed land use categories in 2008 were the same as those assigned in 2001. The 2001 sites were sampled, on average, 83 days earlier in the year than in 2008 (±41 SD) with a range of four to 152 days. The mean Julian day of sampling was 178 (end of June) in 2001, and 261 (mid-September) in 2008. Antecedent rainfall in north Florida was different between the sampling periods, with annual rainfall prior to sampling in summer 2001 (96 cm at Gainesville Regional Airport, the approximate center of the study
Fig. 2. Mean FWCI scores by land use from the 2001 and 2008 assessments. For both years, letters indicate significant differences between land uses. Error bars represent one standard error (n = 22; 2001: ANOVA, F = 18.9, p < 0.005; 2008: ANOVA, F = 20.6, p < 0.0001). All comparisons across years were not significant.
site) well below normal (ca. 122 cm), in contrast with normal rainfall prior to the 2008 sampling period (123 cm). There was broad evidence of significant community changes between 2001 and 2008. During the 2001 survey, 248 different taxa were encountered, whereas only 167 were found in 2008. Moreover, the overlap in this composition was relatively modest, with only 109 species found in both years. Consequently, we observed a significant (p = 0.007) decline in mean species richness between 2001 and 2008, with values (±SD) of 33.1 ± 14.0 in 2001 versus 27.1 ± 9.0 in 2008. Species richness was significantly negatively correlated with wetland condition (FWCI) in 2008 (r = −0.62, p = 0.002), but this association was not significant in 2001. There was evidence of land use effect on richness across dates. Specifically, in 2001, reference sites had the lowest mean richness (30.5 ± 13.6), followed by urban (33.8 ± 6) and agricultural (39.8 ± 23) sites; in 2008, the same trends held, but the differences were statistically significant (p = 0.009) with reference site richness (22.1 ± 6.4) lower than agricultural sites (34.2 ± 9.9) and urban sites (32.3 ± 7.8), which were not different. Loss of richness between 2001 and 2008 occurred most in reference sites (mean decline = 8.9 species, p = 0.001), while richness declines in urban (mean = 2.3 species) and agricultural (mean = 5.5 species) sites were not significant. 3.1. Florida Wetland Condition Index The mean FWCI score in 2008 across the 22 sites was 36.5 (±13.6), where a value of 60 denotes the highest possible wetland condition; this compared favorably with the mean value for the same sites from 2001, which was 37.4 (±18.3). However, the range (48.5 versus 58.4) and standard deviation (13.7 versus 18.5) of FWCI scores in 2008 was narrower than for 2001. Despite this decline in the spread of FWCI scores, mean FWCI scores were not significantly different between sampling periods (p = 0.44) despite evidence that community composition varied. Within sampling years, FWCI differed significantly between land uses (Fig. 2). In both 2001 and 2008, agricultural sites had the lowest FWCI scores (13.1 ± 17.4 and 17.4 ± 13.2, respectively), followed by urban sites (28.4 ± 11.2 and 30.9 ± 8.6), and reference sites (49.1 ± 10.4 and 45.7 ± 5.5) (ANOVA, F = 20.6, p < 0.0001). All pairwise comparisons of land use were significant in 2001, but in 2008 significant pairwise differences were observed for agricultural versus reference sites, and reference versus urban sites, but not for agricultural versus urban sites (p = 0.15). In comparing FWCI scores within land use category across
72
E. Deimeke et al. / Ecological Indicators 34 (2013) 69–75 Table 1 Jaccard similarity indices (mean ± 95% CI) between sites (for 2001 and 2008, grouped by land use) and within sites (2001 vs. 2008, grouped by land use).
Fig. 3. FWCI scores in 2001 vs. 2008; linear regression was strongly siginficant (p < 0.0001), but the fitted slope (95% CI = 0.53–0.73) is significantly less than one (dashed line).
years, no changes were statistically significant. There was, however, a trend where agricultural sites (low FWCI) increased from 2001 to 2008 (mean = 4.4) and reference sites decreased (mean = −3.3); this movement toward the global mean in 2008 is further explored below. FWCI condition scores were highly correlated between sampling periods (Fig. 3; R2 = 0.88, p < 0.001); however, the fitted slope was significantly below one (0.68 ± 0.13, 95% CI), implying a systematic convergence in scores toward median values. To further underscore this trend, we note that the change in FWCI score between 2001 and 2008 was significantly negatively correlated with the 2001 score (FWCI = 11.46 − 0.32 × FWCI2001 ; r2 = 0.54, p < 0.001), suggesting a unit change in predicted score for every three unit change in 2001 FWCI score. The standard error of the linear regression estimate was 5.6 suggesting scores can be accurately predicted within ±6 FWCI points; the standard error obtained using the 1:1 line was 7.9. 3.2. Community similarity Jaccard metrics of community composition comparing sites within and across land use and time suggest general low similarity (Table 1). Overall, Jaccard similarity averaged 0.3 (±0.03 SD) between 2001 and 2008 with no apparent effect of land use (Table 1). To place the implied community changes in context, we also evaluated Jaccard similarity across sites, but within land use, for both 2001 and 2008. The results indicate extremely low similarity in both 2001 (mean = 0.18 ± 0.01) and 2008 (mean = 0.19 ± 0.01).
Land use
Between sites (2001)
Within sites (2001 vs. 2008)
Between sites (2008)
Reference Agriculture Urban
0.22 ± 0.01 0.17 ± 0.06 0.21 ± 0.03
0.31 ± 0.04 0.26 ± 0.11 0.28 ± 0.04
0.21 ± 0.01 0.22 ± 0.07 0.28 ± 0.03
Overall
0.18 ± 0.01
0.30 ± 0.03
0.19 ± 0.01
While within site similarity was much higher than across site similarity, the implied marked temporal community shifts reinforce the general findings that the species present at the study sites changed substantially between 2001 and 2008. When rare species (those present at less than five sites) were omitted, the similarity metrics rose (mean = 0.35 ± 0.03), but the general trend was retained. One explanation for low similarity is sampling during different phases of wetland phenology. Sites were sampled later in the year in 2008 than in 2001 (mean = 83 days). However, we observed no correlation between Jaccard similarity and the Julian day difference between sampling, which ranged from 4 to 152 days (r2 = 0.03, p = 0.46). As such, low similarity cannot be explained by the different sampling dates. 3.3. Individual FWCI metrics In spite of large differences in composition, we observed relatively consistent FWCI scores. To better understand this important finding, we evaluated the six equally weighted metrics that comprise FWCI between sampling years (Table 2). In general, the individual metrics aligned over time, with relatively small absolute changes for all metrics, except wetland status. Moreover, the magnitude of disagreement between sampling periods was relatively consistent across land use categories. The correlation between metric values in 2001 and 2008 was generally strong, and the slope of the relationship was always near 1, with the exception of the % sensitive metric which was higher in 2008. The main exception to broad agreement across years was for wetland status, for which the correlation between 2001 and 2008 across all sites was not statistically significant and the fitted slope was much less than 1. We observed almost no change in the proportion of tolerant or sensitive species, despite significant turnover in the composition. The metric focused on the proportion of exotic taxa was also unchanged between years, though in this case the representative non-native taxa were relatively well conserved. The native perennial metric was modestly different between sampling years with values of 7.5 in 2001 and 8.1 in 2008 (p = 0.02). This shift may be due to differences in sampling season or wetter antecedent hydrologic conditions in 2008, which would reduce the number of annuals. While the shifts in wetland status were large for many
Table 2 Summary of changes in individual components of FWCI between 2001 and 2008. The mean absolute error (MAE) denotes the magnitude of disagreement between sampling periods, summarized for within and across land use categories. The Pearson correlation coefficient indicates the consistency of values between 2001 and 2008; significant correlations (p < 0.05) are in bold and where correlations were not significant the p-value is presented in parentheses. The linear slope for the overall association is also shown; bold values are significantly different (p < 0.05) from 1.0. Metrica
%Tolerant %Sensitive % Exotic FQAI %Native Perennial %Wetland Status a
Overall
Agricultural
Reference
Urban
MAE
r
Slope
MAE
r
MAE
r
MAE
r
1.2 1.5 0.9 1.4 1.1 2.8
0.90 0.89 0.91 0.87 0.91 0.36 (0.09)
1.15 1.25 0.98 1.18 1.01 0.56
1.2 2 0.4 0.6 1.1 3.1
0.70 0.80 0.76 0.79 0.77 0.38 (0.26)
1.4 1.5 0.8 1.7 0.9 2.6
0.68 0.80 0.30 (0.14) 0.46 0.11 (0.46) 0.05 (0.69)
0.9 1.1 1.3 1.3 1.3 2.9
0.91 0.94 0.73 0.42 0.05 (0.77) 0.32 (0.21)
All metrics are reported after rescaling (Reiss, 2004), which created a range from 0 to 10 for 2001 data but a lower range for some metrics for 2008.
E. Deimeke et al. / Ecological Indicators 34 (2013) 69–75 Table 3 Mean CC scores of species lost (CClost ) or gained (CCgained ) between 2001 and 2008. Mean values for both lost and gained taxa differed significant by land use in both years (p < 0.0001). However, a significant difference between gained and lost values was observed only for reference sites (p = 0.03). Values of CClost :CCgained departed significantly from one for reference sites only. Values are means (±95% CI). Land use
Mean CCgained
Mean CClost
CClost :CCgained
Agriculture Reference Urban Overall
2.84 (1.04) 3.88 (0.36) 3.33 (0.47) 3.54 (0.47)
2.56 (1.29) 4.49 (0.29) 3.18 (0.40) 3.78 (0.44)
0.89 (0.27) 1.21 (0.18) 0.99 (0.21) 1.09 (0.13)
Fig. 4. Significant positive relationship between FWCI scores from 2001 and the ratio of CC scores of species lost to species gained from 2001 to 2008 (p = 0.0009). This suggests higher condition sites gained species of lower conservatism coefficient that they lost, and vice versa.
of the sites, it is important to note that they were not directional; the mean change was small (−0.5) despite large absolute shifts. While these changes may be responding to hydrology, it is likely that this is relatively localized, and not systematic in response to drought conditions that prevailed during the 2001 sampling. When we computed FWCI without wetland status as a subcomponent, we obtained results between measurements in 2001 and 2008 that were similar in strength to the full FWCI (r2 = 0.89, p < 0.001), but where the slope was closer to, but still significantly less than, unity (0.72, 95% CI = 0.61–0.84). The last individual metric, the FQAI was significantly lower in 2008 (4.6 ± 2.2) compared with 2001 (5.7 ± 3.3) (p = 0.001), suggesting a modest shift toward species with lower CC. This overall effect was largely a result of changes in species composition within reference sites (Table 3), which was the only land use category to show significant differences between the CC for species gained (3.88) versus species lost (4.49) (p = 0.03). Despite significant species turnover, CC for urban sites did not change (CClost :CCgained ∼ 1.0; Table 3) and agricultural sites showed a nonsignificant trend toward higher CC. Combining these effects, we observed a statistically significant positive relationship between FWCI in 2001 and the ratio of CClost :CCgained (Fig. 4). This suggests convergence toward mean FQAI as sites with high FWCI scores tended to lose species with higher CC than were recruited. Similarly, sites with low FWCI scores tended to lose species with lower CC scores than those recruited. 4. Discussion The most important finding of this work is that wetland condition, as measured by FWCI, did not change substantially between sampling years. Because land use categories determined in 2001
73
were field verified as unchanged in 2008, the strong correlation observed in FWCI scores between years supports the contention that land use is the commanding variable controlling wetland condition (Brown and Vivas, 2005). It also suggests that this metric is robust to both variation in the sampling time within a year and the natural interannual variation that is characteristic of these geographically isolated wetlands. Notably, neither urban nor agricultural sites showed a downward trajectory in condition as might be expected as these wetlands continue to adjust to their relatively new surroundings. This suggests that even these systems are at something that approximates a dynamic equilibrium with regard the quality of vegetation community despite dramatic climate variation between the two sampling events. Despite relatively consistent FWCI scores over time, this study also observed surprising dissimilarity in vegetation composition between sampling years. Indeed, our results suggest that vegetation differed in all land use classes by ca. 70% between 2001 and 2008, a result that was robust to the exclusion of rare species (those occurring at less than five sites). One potential explanation for this difference is evaluator bias. While we cannot rule this possibility out entirely, we note that both evaluators were trained botanists, and consulted the same taxonomic references. Perhaps more importantly, the species lists from 2001 were available to evaluators in 2008, which would generally lead to higher similarity given prior information to guide identification of unknown species. This substantial shift in species composition is likely real, and is comparable to other studies using Jaccard index similarities in wetlands, albeit when comparing across, not within, sites. For example, Hartzell et al. (2007) reported Jaccard index similarities for vegetation of 38% between created and natural depression wetlands. What is notably similar to our results is that this large discrepancy in taxonomic composition was not evident in the twelve condition metrics used to compare the two systems. Similarly, invasion of two taxa, non-native Phragmites australis and Typha spp., in coastal wetlands on the Great Lakes caused a 19% similarity (Jaccard’s index) between 2001 and 2004 (T’ulbure et al., 2007). While the authors correctly point to this striking dissimilarity, FQAI and CC scores were actually greater after the two species were established. These results illustrate that the list of species that occupy a site at any given time may be highly dynamic, responding to water level and seasonal variations that are independent of the anthropogenic disturbance gradients that are the focus of condition assessment; this underscores the utility of condition scores that integrate over the entire community. This does not discount the insights that can be gleaned by turnover in species composition, but supports a cautious approach to inference about site condition based on that evidence alone. Because biological assessment cannot feasibly be implemented in the same phenological period for all sites, the need to develop and employ metrics that are robust to natural variation induced by seasons is clear. Species lists are unlikely to provide that robustness, but our work suggests that FWCI does. While FWCI scores between 2001 and 2008 were strongly correlated, we did observe significant evidence of a convergence of FWCI scores across sites in 2008. That is, sites with high scores in 2001 tended to score lower in 2008, while sites with low scores in 2001 tended to score higher in 2008. This convergence effect, manifest as a decline in range between 2001 and 2008 and a fitted slope between 2001 and 2008 significantly less than 1 (Fig. 3), results in a loss of extreme wetland condition scores. The most likely explanation for this effect is statistical over-fitting of 2001 FWCI scores. The 5th to 95th percentile of individual FWCI metrics were normalized to occupy the entire score range of 0–10 (Reiss, 2004). The current scores were derived as FWCI would be field applied, without stretching minimum or maximum scores to the 0 or 10 endpoints. FWCI development was based on a snapshot in time, and the normalization process establishes conditions at that time
74
E. Deimeke et al. / Ecological Indicators 34 (2013) 69–75
as the benchmark; observed variation in composition underscores that these systems are compositionally dynamic, making it highly likely that the FWCI formulation is over-specified to one set of conditions; this effect is modest in magnitude for any given site, but operationally relevant when sampling a population of wetlands. The convergence in scores in 2008 might occur in response to systematic shifts in vegetation that would affect all six elements of the FWCI. It was striking, therefore, to observe that five metrics (i.e., tolerant indicator species, sensitive indicator species, exotic species, FQAI, and native perennial species) were highly correlated between years with slopes in almost all cases not significantly different from 1, contributing to inference of overall stability in total FWCI scores. However, the sixth metric, the average wetland status of the species present within each site, showed significant, albeit directionally inconsistent, disagreement between sampling years. While changes were large at many sites there was no systematic loss or gain of obligate and facultative wetland species. Rather, both loss and gain occurred between sampling years. It is likely this effect can be ascribed to variation in hydrologic conditions just prior to sampling. The 2001 survey occurred in the midst of a regional drought in which annual rainfall was well below normal. In 2008, the annual rainfall was near the long term mean of 123 cm yr−1 . However, there was also within-year hydrologic variation, with the latter half of the 2001 sampling occurring under wetter conditions and the early part of the 2008 sampling occurring during wetter conditions. In fact, when sites were separated into 2001 and 2002 sampling years, those sampled in 2002 were significantly more similar to 2008 (mean similarity = 0.41) than the 2001 sites (mean similarity = 0.29) (p = 0.02). This may be due to easing of the drought in 2002 creating hydrologic conditions more similar to those in 2008, and thus creating greater species composition similarity. However, it may also be due to shorter time lag between 2002 and 2008 than between 2001 and 2008. While a non-uniform shift in the overall importance of obligate and facultative wetland species likely cannot account for convergence of FWCI scores, within year variation is an area that merits continued attention. Examining the FQAI metric revealed that most of the species lost were of comparable quality to those gained, suggesting robust correspondence between quality scores and land use stressors. While the mean CC of species lost or gained was similar (3.78 vs. 3.54, respectively), the ratio of CClost :CCgained was significantly positively correlated with overall 2001 FWCI scores. That is, wetlands with high FWCI scores, predominantly reference sites, tended to lose species of high CC, and gain species of modestly lower CC, suggesting an overall loss of species quality within reference sites. Conversely, wetlands with low FWCI scores, mainly agricultural sites, tended to lose species of low CC, and gain species of higher CC, which implies an overall gain of species quality within agricultural sites. The ultimate effect of this shift in vegetation was a loss of sites with extreme CC, contributing to an overall convergence of FWCI scores in 2008, and an overall moderation of vegetative quality in these sites. It is unclear if other assessment methods show similar convergence over time because the temporal stability of assessment metrics is not widely reported. Additional research is warranted on the repeatability of CC scores as indicators of wetland condition and the role of FQAI metrics in assessing wetland condition. This is particularly important because species CC scores are assigned by a panel of experts and are thus highly variable (Cohen et al., 2004). It is rare that wetland condition is repeatedly assessed over time, particularly without a change in disturbance, environmental impacts, or land use. Consequently, most literature focuses on condition due to gradients in anthropogenic disturbance, i.e., land use change or logging (Lopez and Fennessy, 2002; Mack, 2007; Miller and Wardrop, 2006; Reiss and Brown, 2007). The current
study fills an important knowledge gap regarding condition variability over time in the absence of changing driving forces, and demonstrates a need for long-term monitoring of condition and vegetation beyond what is currently mandated for mitigation wetlands (i.e., created, restored, preserved or enhanced wetlands; Reiss et al., 2007), which are often compared against other regional reference wetlands or a hypothetical reference condition in determining mitigation success. Understanding natural variability in condition and vegetation will further reinforce the validity of using benchmark sites against which mitigation wetlands can be compared, particularly in light of the low site community similarity between years. The fact that the condition scores were relatively consistent across years in spite of marked changes in composition means that wetland condition is conserved by a “portfolio” effect wherein some sub-components may decline in response to natural environmental variation, but others increase. This lends strong support to the multi-variate nature of FWCI and other wetland assessment tools, and suggests that evaluating the portfolio of services may be more informative than focusing on one service or metric. Expanding this research beyond a small sample of wetlands in one wetland region and of one specific wetland class (i.e., palustrine geographically isolated depressional forested wetlands) would strengthen our collective understanding of temporal variability in the wetland plant community. Expanding the focus beyond vegetation to other trophic levels with established IBIs would also enhance knowledge of natural variability for wetland biotic communities along a human disturbance gradient.
5. Conclusions As a foremost conclusion, this research suggests support for the use of vegetation based condition metrics in assessing Florida’s geographically isolated wetlands. This is most well supported by the observation that index values were robust to a high level of taxonomic turnover. However, several more nuanced conclusions remain, specifically with regard to shifting mean CC value for both reference and impacted wetlands. Significantly lower species richness and community dissimilarity between years also expands the uncertainty in adopting IBI metrics that are based on straight richness or diversity calculations.
Acknowledgements This project was funded through support from the UF-College of Agriculture and Life Sciences and the Florida Department of Environmental Protection (#WM-683). We gratefully acknowledge the field assistance of Chad Foster, Jorge Guevara, Jeff Hull, Larry Korhnak, Dina Liebowitz, Danielle Liebenow, Lauren Long, Danielle Watts, and Justin Vogel, as well as Mark Brown, the principal investigator for the FWCI project. We also acknowledge private landowners that permitted repeated site access.
References Beck, M.W., Hatch, L.K., 2009. A review of research on the development of lake indices of biotic integrity. Environ. Rev. 17 (NA), 21–44. Brown, M.T., Vivas, M.B., 2005. Landscape development intensity index. Environ. Monit. Assess. 101, 289–309. Cohen, M.J., Carstenn, S., Lane, C.R., 2004. Floristic quality indices for biotic assessment of depressional marsh condition in Florida. Ecol. Appl. 14 (3), 784–794. Costanza, R., d’Arge, R., deGroot, R., Farber, S., Grasso, M., Hannon, B., Limburg, K., Naeem, S., O’Neill, R.V., Paruelo, J., Raskin, R.G., Sutton, P., van den Belt, M., 1997. The value of the world’s ecosystem services and natural capital. Nature 387, 253–260. Fennessy, M.S., Jacobs, A.D., Kentula, M.E., 2004. Review of Rapid Methods for assessing Wetland Condition. EPA/620/R-04/009. U.S. Environmental Protection Agency, Washington, DC.
E. Deimeke et al. / Ecological Indicators 34 (2013) 69–75 Hartzell, D., Bidwell, J.R., Davis, C.A., 2007. A comparison of natural and created depressional wetlands in central Oklahoma using metrics from indices of biological integrity. Wetlands 27, 794–805. Karr, J.R., Chu, E.W., 1997. Biological monitoring: essential foundation for ecological riskassessment. Hum. Ecol. Risk Assess. 3, 993–1004. Lane, C.R., 2000. Proposed ecological regions for freshwater wetlands of Florida. University of Florida, Environmental Engineering Sciences, Gainesville, FL (Masters Thesis for Masters of Science). Lane, C.R., Brown, M.T., 2007. Diatoms as indicators of isolated herbaceous wetland condition in Florida, USA. Ecol. Ind. 7, 521–540. Lopez, R.D., Fennessy, M.S., 2002. Testing the floristic quality assessment index as an indicator of wetland condition. Ecol. Appl. 12, 487–497. Mack, J.J., 2006. Landscape as a predictor of wetland condition: an evaluation of the landscape development intensity index (LDI) with a large reference wetland dataset from Ohio. Environ. Monit. Assess. 120, 121–241. Mack, J.J., 2007. Developing a wetland IBI with statewide application after multiple testing iterations. Ecol. Indic. 7, 864–881. MEA (Millenium Ecosystem Assessment), 2000. Ecosystems and Human WellBeing: Wetlands and Water Synthesis. World Resources Institute, Washington, DC. Miller, S.J., Wardrop, D.H., 2006. Adapting the floristic quality assessment index to indicate anthropogenic disturbance in central Pennsylvania wetlands. Ecol. Indic. 6, 313–326. Miller, S.J., Wardrop, D.H., Mahaney, W.M., Brooks, R.P., 2006. A plant-based index of biological integrity (IBI) for headwater wetlands in central Pennsylvania. Ecol. Indic. 6, 290–312.
75
Reiss, K.C., 2004. Developing biological indicators for isolated forested wetlands in Florida. University of Florida, Environmental Engineering Sciences (Dissertation for Doctor of Philosophy). Reiss, K.C., 2006. Florida Wetland Condition Index for depressional forested wetlands. Ecol. Indic. 6, 337–352. Reiss, K.C., Brown, M.T., 2007. Evaluation of Florida Palustrine Wetlands: application of USEPA levels 1, 2, and 3 assessment methods. EcoHealth 4, 206–218. Reiss, K.C., Hernandez, E., Brown, M.T., 2007. An evaluation of the effectiveness of mitigation banking in Florida: Ecological success and compliance with permit criteria. FDEP Final Report #WM881. U.S. EPA Region Four #CD 96409404-0. Southerland, M.T., Rogers, G.M., Kline, M.J., Morgan, R.P., Boward, D.M., Kazyak, P.F., Klauda, R.J., Stranko, S.A., 2008. Improving biological indicators to better assess the condition of streams. Ecol. Indic. 7, 751–767. T’ulbure, M.G., Johnston, C.A., Auger, D.L., 2007. Rapid invasion of a Great Lakes coastal wetland by non-native Phragmites australis and Typha. J. Great Lakes Res. 33, 269–279. Wallwork, J.A., 1976. The Distribution and Diversity of Soil Fauna. Academic Press, London. Wilhelm, G., Ladd, D., 1988. Natural area assessment in the Chicago region. In: Transactions of the 53rd North American Wildlife and Natural Resource Conference, Louisville, Kentucky, Wildlife Management Institute, Washington, DC, USA, pp. 361–375. Wunderlin, R.P., Hansen, B.F., 2008. Atlas of Florida Vascular Plants. Institute for Systematic Botany, University of South Florida, Tampa [S. M. Landry and K. N. Campbell (application development), Florida Center for Community Design and Research] http://www.plantatlas.usf.edu/