Verification of an Environmental Impact Assessment Using a Multivariate Statistical Model

Abstract

Environmental impact assessment is a means of preventing and mitigating the adverse effects of economic development activities on the natural environment. It is meant to ensure that decision-makers have sufficient information to consider environmental impacts before proceeding with new projects. Despite their important role in public policy, verification of environmental impact assessments has seldom been conducted. In this study, we used principal component analysis (PCA) to identify the major sources of influence on the coastal waters adjacent to a major tourist facility (an aquarium) in southern Taiwan, followed by the construction of a structural equation model (SEM) to determine the direct and indirect effects of the abiotic factors on phytoplankton and zooplankton density and diversity. Based on the loadings of principal components 1–3, we identified that river input, suspended matter, and seasonal changes were the major factors affecting the coastal area. The SEM further suggested that phytoplankton density and diversity were affected directly by seasonal changes and suspended matter, but only indirectly by river input, owing to the latter’s effect on suspended matter. In contrast, the SEM suggested that zooplankton density and diversity were affected directly by seasonal changes, but indirectly by both river input and suspended matter owing to their effects on phytoplankton density and diversity. Q2 was the season with the highest number of visitors to the aquarium, but none of the abiotic or biotic parameters showed particular differences, implying that the variations in those parameters in the adjacent coastal waters were not related to the visitors. We suggest that PCA and SEM be used in the future in other contexts to verify environmental impact assessments.

1. Introduction

The National Environmental Policy Act 1970 (NEPA) of the United States of America was the first formal incorporation of environmental impact assessment into a piece of legislation [1]. Since then, countries around the world have incorporated some form of impact assessment into their formal procedures or legislation relating to planning or other areas of environmental decision making [2]. In Taiwan, the Environmental Impact Assessment Act was promulgated by presidential order in 1994 to prevent and mitigate the adverse impact of development activity on the environment, and to ensure that decision-makers considered environmental impacts before deciding whether to proceed with new projects.

An ecosystem is usually affected by a combination of different environmental factors [3], making it difficult to determine which factors are more important than others, and to construct an ecological model that includes both physical and biological factors. Principal component analysis (PCA), coupled with structural equation modeling (SEM), has proved a reliable tool in identifying the major contributors to water quality in certain areas [4,5] and can specify the relationships among variables [6], thereby providing insight into the relationships between correlated physical, chemical, and biological variables in an ecosystem [7,8].

The National Museum of Marine Biology & Aquarium (NMMBA) is a public aquarium situated on a 58-hectare site within Kenting National Park near the southern tip of Taiwan (Figure 1). Since its completion in 2000, the NMMBA has attracted millions of visitors each year. The environmental impact assessment submitted prior to land clearing and construction [9] stated that the operation of the aquarium would not affect the adjacent coastal area. Furthermore, the Environmental Impact Comparative Analysis Report submitted prior to the construction of an additional exhibition hall at the site in 2010–2012 [10] also stated that the construction would have little effect on the coastal ecosystem. Over the years, environmental monitoring of the adjacent coastal area has been conducted seasonally, but specific verification of these two environmental impact assessments has never been conducted.

Since the operation of the aquarium could potentially increase nutrient-rich wastewater that would be drained into the coastal area, and because the construction of an exhibition hall could potentially increase the turbidity of the seawater, we would expect an impact on the phytoplankton and zooplankton in the adjacent coastal ecosystem. We hypothesize that the abiotic and biotic parameters would be different near the drainage pipe, especially during the peak season. In this study, we constructed a structural equation model (SEM) for the coastal ecosystem of NMMBA based on four years of environmental monitoring data (2011–2014) to determine the environmental factors that affect the phytoplankton and zooplankton dynamics and diversity in the area. The objective was to verify the above-mentioned environmental impact assessments, which stated that the operation of the aquarium, as well as the construction of an exhibition hall, would not affect the adjacent coastal ecosystem. The results will provide a model and case study for the verification of other environmental impact assessments elsewhere in the future.

2. Materials and Methods

2.1. Study Area

Six stations were monitored adjacent to NMMBA, of which four stations were located at a depth of around 5 m and two stations at around 20 m (Figure 1). Domestic wastewater produced by NMMBA and its visitors enters a sewage treatment plant on-site at NMMBA, from which it is recycled for irrigation, with the excess being drained into nearby coastal waters through a drainage pipe near S10 (Figure 1). Nutrient-rich waste seawater from the aquarium facilities goes through a series of artificial wetlands before entering the coastal waters, also through the drainage pipe (Figure 1). Two third-order streams, the Sichong River and the Baoli River, enter the study area from the north (Figure 1). The Sichong River goes through a forest area, whereas the Baoli River mostly goes through farmlands and residential areas. The pH and turbidity were 8.3 ± 0.1 and 23 ± 7 ntu in the Sichong River, respectively, and 8.5 ± 0.1 and 75 ± 16, respectively, in the Baoli River during the study period [11].

2.2. Measurement of Abiotic and Biotic Parameters

Between 2011 and 2014, abiotic (i.e., water temperature, salinity, transparency, turbidity, and dissolved oxygen) and biotic data (i.e., phytoplankton and zooplankton) were collected every quarter (Q1: Mar–May, Q2: Jun–Aug, Q3: Sep–Nov, Q4: Dec–Feb), on 16 occasions altogether, at a depth of 1 m below the water surface at 6 stations off the NMMBA (Figure 1). Water depth at the stations was ~5–25 m, and their distance from the shore ranged from 0.2 to 1.5 km. Temperature and salinity were measured in situ using a handheld meter (YSI Model-30, YSI, Yellow Springs, OH, USA), and a multiparameter meter (YSI Model-556, YSI, OH, USA) was used to measure dissolved oxygen (DO) and pH. Turbidity was measured using a turbidimeter (2100P Turbidimeter, Hach, Loveland, CO, USA) [12]. Nutrients (NO₂-N + NO₃-N, NH₃-N, PO₄-P, and SiO₂-Si) were also measured according to [13]: nitrate was reduced to nitrite using a cadmium redactor, and nitrite was determined via diazotation with sulfanilamide and coupling with N-(1-naphtha)-ethylenediamine dihydrochloride [14]; ammonia was determined using the indophenol blue spectrophotometric method [15]; phosphate was determined using the molybdate–antimony method [16]; silicate was determined using the silicomolybdate method, which was reduced by a metal–oxalic acid solution [17]. The nitrate concentration in this study was nitrate + nitrite; however, the nitrite concentration in the area was very low [9,18] and, thus, negligible. Phytoplankton samples were taken using a Niskin water sampler, and 1 L of each sample was preserved using formalin to a final concentration of 5%. In the laboratory, 200 mL of each such sample was filtered onto a 0.45-µm piece of filter paper, which was dried in an oven at 50 °C for 24 h. The filter paper was mounted onto a slide using immersion oil, and then, examined for phytoplankton taxa under a light microscope [8]. Individual phytoplankters were identified to the lowest taxonomic level possible. Zooplankton were collected using horizontal tows at cruising speeds of about 2 knots for 10 min, using a Norpac net (0.45-m diameter opening, 330-μm mesh size) equipped with a flowmeter (Model 438 110, Hydro-bios, Kiel, Germany). Zooplankton samples were preserved onboard immediately in 5% borax-buffered formalin in seawater [19]. Zooplankton were later examined under a dissecting microscope (Askania model GSZ 2) and their different classes identified.

2.3. Data Analysis

The biotic and abiotic parameters were analyzed using principal component analysis to identify their influences on the coastal area. All data were ln-transformed before analysis. Principal components with eigenvalues larger than 1 and components explaining at least 10% of the variability were considered to represent the factors that influence the study area. The interpretation of each PC axis was determined based on the factor loadings of the variables [8] and the temporal and spatial variations of each score, which were analyzed using two-way ANOVA and multiple comparisons (Supplementary Materials). For explanation of the method, please refer to https://www.statistixl.com/features/principal-components/ (accessed on 28 June 2022).

After the main environmental forces had been identified, we introduced phytoplankton and zooplankton densities and diversity indices [20] to predict the interactions between environmental factors and the biotic parameters in a conceptual model (Figure 2). We assumed that the environmental factors could affect both phytoplankton and zooplankton, and that phytoplankton and zooplankton could affect each other reciprocally. Lastly, we used LISREL 8 [21] to verify the model. Detailed assumptions and the concept of SEM are explained in [6,22,23].

3. Results

Table 1. Ranges and means (±SD) of the hydrological and biological variables in the study area from 2011 to 2014 (N = 96).

	Range	Mean ± SD
Temperature (°C)	24.1–31.6	27.8 ± 2.1
Salinity	18.8–34.6	32.9 ± 1.8
Dissolved oxygen (mg L−1)	4.7–10.0	6.7 ± 0.7
pH	7.87–8.40	8.18 ± 0.11
Transparency (m)	1.0–20.0	8.1 ± 4.4
Turbidity (ntu)	0.20–16.4	0.90 ± 2.02
Suspended solids (mg L−1)	1.84–16.70	6.55 ± 3.25
NO3-N (mg L−1)	0.001–0.193	0.024 ± 0.033
NH3-N (mg L−1)	0.001–0.289	0.018 ± 0.042
PO4-P (mg L−1)	0.002–0.008	0.004 ± 0.002
SiO2-Si (mg L−1)	0.020–1.382	0.121 ± 0.217
Chl a (µg L−1)	0.02–2.53	0.33 ± 0.50
Phytoplankton density (cell L−1)	260–180,400	27,537 ± 36,921
Phytoplankton species number	4–23	13 ± 4
Phytoplankton Shannon–Weaver diversity index	0.17–2.53	1.57 ± 0.50
Zooplankton density (ind 100 m−3)	299–93,231	12,826 ± 20,171
Zooplankton class number	9–25	19 ± 4
Zooplankton Shannon–Weaver diversity index	0.62–2.30	1.60 ± 0.35

summarizes the means (±SD) and ranges of the biotic and abiotic variables measured quarterly at all six stations at the study site off the NMMBA between 2011 and 2014. The water temperature ranged from 24.1 to 31.6 °C, typical of a subtropical region. The salinity varied greatly between 18.8 and 34.6, as did various nutrient concentrations, indicating a strong effect of freshwater input. The transparency, turbidity, and suspended solids also fluctuated greatly, suggesting that suspended matter may play an important role in the region (Table 1). Spatial and/or temporal variation in phytoplankton and zooplankton were also apparent in both the taxon number and diversity index (Table 1).

The first three principal components explained 63.69% of the abiotic variables at the study site (

Table 2. Loadings of principal components 1–3 (after varimax rotation) for abiotic variables measured in the study area (N = 96).

Variable	PC 1	PC 2	PC 3
Transparency	−0.277	−0.639	−0.057
Temperature	0.211	0.311	−0.472
Salinity	−0.822	−0.090	0.358
NO3-N	0.763	0.226	0.122
PO4-P	0.186	0.012	0.094
SiO2-Si	0.916	0.141	0.071
NH3-N	0.800	0.195	0.174
Turbidity	0.683	0.092	−0.412
Chl a	−0.024	0.728	−0.083
Suspended solids	0.180	0.643	−0.143
Eigenvalues	3.393	1.574	1.401
Total variance (%)	33.94	15.74	14.01

). The eigenvalues of PC1, PC2, and PC3 were larger than 1 and explained 33.94%, 15.74%, and 14.01% of the variance, respectively. The factor loadings for salinity (−0.822), nutrients such as NO₃-N, SiO₂-Si, and NH₃-N (0.763~0.916), and turbidity (0.683) in PC1 (Table 2) indicated that low salinity was coupled with high turbidity and high nutrient content, which suggests that PC1 represents the influence of river input (i.e., high river discharge may have reduced the salinity while increasing the nutrient load and turbidity of the coastal seawater). Both the temporal (two-way ANOVA, DF = 3, F = 3.76, p = 0.014 *) and spatial (two-way ANOVA, DF = 5, F = 3.91, p = 0.003 *) variabilities of the principal component scores of PC1 were significantly different between seasons and sites, respectively (

Table 3. Results from two-way ANOVA tests and all pairwise multiple comparisons (Tukey’s HSD Test) on principal components 1 to 3. * p < 0.05. Different superscripts indicate significant differences.

(A) PC1
		Two-way ANOVA		Multiple Comparison
	DF	F	p
Season	3	3.76	0.014 *	Q3 a Q2 a Q1 ab Q4 b
Station	5	3.91	0.003 *	S1 a S10 b S4 b S7 b S11 b S2 b
Season × Station	15	1.39	0.175
(B) PC2
		Two-way ANOVA		Multiple Comparison
	DF	F	p
Season	3	7.98	<0.001 *	Q2 a Q4 a Q3 a Q1 b
Station	5	1.36	0.252
Season × Station	15	0.27	0.996
(C) PC3
		Two-way ANOVA		Multiple Comparison
	DF	F	p
Season	3	12.77	<0.001 *	Q1 a Q4 b Q3 bc Q2 c
Station	5	0.72	0.611
Season × Station	15	0.70	0.778

A). Further analysis using Tukey’s test revealed that Q4 (the dry season) was significantly different, in this respect, from Q2 and Q3 (the rainy seasons) [11] and that station S1 (closest to the river mouths: Figure 1) was significantly different from other stations (Table 3A).

PC2 showed a negative factor loading for transparency (−0.639) coupled with positive loadings for chl a (0.728) and suspended solids (0.643) (Table 2), which suggests that PC2 represents the suspended matter in the water. The temporal variability in the principal component scores of PC2 was significant (two-way ANOVA, DF = 3, F = 7.98, p < 0.001 *) (Table 3B). Further analysis using Tukey’s test revealed that Q1 (the dry season) [11] was significantly different from other seasons, whereas there was no significant difference between the sampling stations (Table 3B).

PC3 showed a negative factor loading for water temperature (−0.472) and turbidity (−0.412), while the loading for salinity was high (0.358) (Table 2). The temporal variability of the principal component scores of PC3 was significant (two-way ANOVA, DF = 3, F = 12.77, p < 0.001 *) (Table 3C). Multiple comparisons using Tukey’s test revealed that Q1 (the dry season) was significantly different from other seasons, Q2 was significantly different from Q4, and Q3 and Q4 were not significantly different from each other (Table 3C). This suggests that the water temperature per se, not the rainy/dry season alternation, was mainly responsible for the differences, and that PC3 represents the changes in season.

Based on the loadings of principal components 1–3, we identified that river input, suspended matter, and seasonal changes were the main factors influencing the biotic and abiotic parameters in the study area. We modified the simple conceptual model of Figure 2 into a more detailed model (Figure 3), with the influence of river input expressed in terms of salinity, turbidity, and various nutrients; the influence of suspended matter expressed in terms of transparency, chl a, and suspended solids; and the influence of the changes in season expressed in terms of temperature, salinity, and turbidity. The phytoplankton and zooplankton densities and taxon numbers are treated in terms of their respective Shannon–Weaver diversity indices.

Using the 18 kinds of observed biotic and abiotic data obtained from six stations during 16 cruises (N = 96) in 2011–2014, a final SEM model was constructed (Figure 4). The weighted-least-squares Chi-square test (χ² = 86.17, df = 58, p = 0.0096) was significant, possibly due to the large sample sizes [24]. The low root-mean-square error of approximation (RMSEA = 0.071) suggests that the model was statistically significant and substantively meaningful [23]. Other model-fit indices were the standardized root-mean-square residual (SRMR) = 0.074; the goodness of fit index (GFI) = 0.88; and the adjusted goodness of fit index (AGFI) = 0.81, all indicating that the model was moderately acceptable [23].

The model showed no direct influence of river input on the planktonic communities (Figure 4). River input, however, did affect the amount of suspended matter (0.45), which negatively influenced the phytoplankton density and species number. The changes in season affected both phytoplankton and zooplankton densities, while the phytoplankton density also affected zooplankton density, showing a bottom-up effect in this coastal area [25].

Between 2011 and 2014, the highest average number of visitors entering NMMBA was in Q2, significantly greater than in other quarters (two-way ANOVA, F = 90.90, p < 0.001 *). The average number of visitors in Q3 was also significantly higher than in Q1 and Q4, whereas the latter two quarters were not significantly different from each other (Figure 5). There were no significant differences in visitor numbers between different years.

4. Discussion

Environmental monitoring has become a common practice in safeguarding natural ecosystems. In many countries, when a development project might be detrimental to the surrounding environment, governmental regulations require an environmental impact assessment in advance to aid in decision-making [2]. After the environmental assessment has been conducted, typically with the inclusion of baseline environmental monitoring, and a conclusion of mild impact has been reached, the project will ordinarily be approved by government regulators and carried out. After the project has been completed, environmental monitoring will usually be continued for some time to ensure that the impact is acceptable, but long-term verification of the conclusions of environmental impact assessments has rarely been conducted.

In this study, we examined the temporal and spatial changes in abiotic and biotic parameters in the coastal area off the National Museum of Marine Biology & Aquarium, a major tourist facility in southern Taiwan. The PCA scores led us to conclude that river input, suspended matter, and the changes in season were the main factors influencing the study area. The input of freshwater from two rivers caused a simultaneous decrease in salinity and increase in turbidity and nutrient concentrations, with the greatest impact being detected at the sampling station closest to the river mouths. The abiotic and biotic parameters were different in the dry season (Q4 and Q1) when compared to the wet season (Q2 and Q3), suggesting that river input was the main contributor to the variations in the coastal marine environment. Similar results have also been reported at a locality along the west coast of central Taiwan, where river discharge contributed the most to the phytoplankton dynamics in the area [8].

Diatoms are the most dominant phytoplankton group in the study area [26,27]. Diatoms produce extracellular acidic polysaccharides [28] in the form of transparent exopolymer particles (TEPs) that can slow the sinking of solid particle aggregations and prolong pelagic residence time [29]. Our results showed parallel dynamics between chl a and suspended solids, which increased and decreased together, but an opposite trend in transparency (Table 2). This suggests that in the study area, the phytoplankton and suspended solids might be trapped by TEPs [30], thereby decreasing the transparency of the water. Such a phenomenon has been reported recently, based on satellite observations of the Bohai and Yellow Seas [31]: low transparency in winter and spring and high transparency in summer and autumn were strongly correlated with the total suspended matter.

The final SEM (Figure 4) indicated a significant influence of the abiotic environment factors on phytoplankton density and diversity, but not on zooplankton density and diversity, a result similar to that found at the above-mentioned site in central Taiwan [8]. The high river input during the rainy season, despite bringing in higher concentrations of various nutrients that could enhance phytoplankton growth [26], also increased the concentration of suspended matter. This could have negatively affected the phytoplankton density and diversity owing to a concomitant decrease in light penetration [32]. Furthermore, strong katabatic wind occurred sporadically in Q4 in the study area [33], and the wind-mixed layer could reach the bottom and cause resuspension of the bottom sediments. Had we not considered the indirect effects of river input on phytoplankton via suspended matter, we would have concluded that nutrient concentrations did not affect phytoplankton when, in fact, they did. Similar situations have also been reported in other estuaries [34,35], although phytoplankton can sometimes increase under conditions of low light penetration [36].

Suspended matter did not directly affect the zooplankton during this study; however, the SEM results revealed an indirect influence through phytoplankton density, a clear bottom-up effect [37]. Contrary to our results, other studies have shown that abiotic factors such as temperature are usually the main drivers of plankton communities in aquatic ecosystems [38,39]. Our SEM model clearly shows both direct and indirect influences among the biotic and abiotic parameters, and thus, enhances our understanding of the complex ecosystem in the area. Since the model is robust, we expect to be able to further modify it and confirm additional interactions by considering other factors as data become available. This would greatly improve our understanding of the disturbances faced in the area, regardless of whether they represent natural or anthropogenic perturbations.

During the monitoring period, Q2 was always the season with the highest number of visitors to the NMMBA, and we would expect higher nutrient concentrations during the quarter near S10. However, none of the abiotic or biotic parameters showed any significant increase or decrease at the station during Q2 compared to other quarters. This implies that the observed changes in various environmental parameters in the coastal waters adjacent to the NMMBA were not related to the visitors. Based on the results from the multivariate statistical model analysis, we verified that the environmental impact assessment concluded before the construction of the NMMBA was reliable and accurate. We suggest that using the methods of PCA, to determine sources of influence, and SEM, to confirm their direct and indirect effects, should be applied to other areas of interest to verify environmental impact assessments in the future.