detecting long-term change in complex communities: a case study from the rocky intertidal zone

1813

Ecological Applications, 15(5), 2005, pp. 1813–1832q 2005 by the Ecological Society of America

DETECTING LONG-TERM CHANGE IN COMPLEX COMMUNITIES:A CASE STUDY FROM THE ROCKY INTERTIDAL ZONE

JOHN R. STEINBECK,1,4 DAVID R. SCHIEL,2 AND MICHAEL S. FOSTER3

1Tenera Environmental, 141 Suburban Road, Suite A2, San Luis Obispo, California 93401 USA2School of Biological Sciences, University of Canterbury, Private Bag 4800, Christchurch 1, New Zealand

3Moss Landing Marine Labs, 8272 Moss Landing Rd., Moss Landing, California 95039 USA

Abstract. Despite the recognition of the usefulness of BACI designs for assessingenvironmental impacts, there are few examples because of the need for repetitive samplingover long time periods. Our examination of the application of a BACI design to detect theimpacts of elevated seawater temperature from the Diablo Canyon Power Plant in centralCalifornia on rocky intertidal communities showed statistically significant changes in alarge percentage of the species analyzed. The statistical power of the analysis resulted fromboth the large numbers of surveys before and during plant operation and from other designfeatures that made the study resilient to the effects of two ‘‘100-year’’ storms, several ENSOwarming events, and the highly variable nature of the impacts. The large data set from thestudy required the development of decision rules for determining the appropriate surveys,stations, and species to analyze. BACI analyses were used to test the effects of the thermalplume on 47 algal and 50 invertebrate data sets. There were statistically significant effectsfor 79% of the algal and 60% of the invertebrate data sets. At the impact sites, there wasa loss of cover by foliose algae and increases in crustose forms. Many invertebrates,particularly grazing gastropods, increased in abundance. Multivariate analysis of the com-munity showed that there was continual change in impact sites that lasted throughout thestudy. The nature, magnitude, and spatial extent of the effects identified from the study arebeing used to determine appropriate plant modifications or mitigation for the effects ofdischarge. This study illustrates many of the problems in analyzing environmental effectsand clearly demonstrates the need for long-term monitoring. This was especially true forthis study, where storms and ENSO events affected our ability to analyze data from someof the stations, and points out the importance of having redundancies built into monitoringprograms. The complex interactions among the direct effects of the discharge, indirectcommunity-level effects, and variation due to oceanographic conditions provide usefulinsights for planning impact assessments and other ecological studies, and help contributetoward science-based regulation and management.

Key words: BACI (before–after, control–impact); community analysis; decision rules; impactanalysis; intertidal communities; power plant; sampling design; thermal effect.

INTRODUCTION

Increased development and associated impacts tonatural environments are consequences of the growthof human populations. As a result, several approachesand statistical designs for assessing the effects of im-pacts on natural communities have been proposed(Green 1979, 1993, Stewart-Oaten et al. 1986, Under-wood 1992, Wiens and Parker 1995). All such ap-proaches and designs are complicated by several fac-tors. One of the most important of these is that theimpacts are almost always unreplicated (Hurlbert 1984,Eberhardt and Thomas 1991). Furthermore, the chang-es due to the impacts must be detected against a back-ground of often considerable natural spatial and tem-poral variability. This is particularly the case in tem-perate rocky shore communities because they are high-

Manuscript received 29 June 2004; revised 26 January 2005;accepted 1 February 2005. Corresponding Editor: J. B. Zedler.

4 E-mail: [email protected]

ly diverse, with a wide range of perennial andephemeral species that vary in abundance seasonally,annually, and over multiple spatial scales. However,rigorous approaches to impact assessment are neededif regulation is going to be ‘‘science-based.’’

One of the most powerful methods for impact de-tection is the Before–After, Control–Impact (BACI) de-sign, which uses sampling of control and impact sitesthrough time to provide replication within the ‘‘before’’and ‘‘after’’ periods (Stewart-Oaten et al. 1986, Schroe-ter et al. 1993). BACI overcomes the lack of spatialreplication of the impact by using repeated samplingof control and impact sites through time to providereplication within the ‘‘before’’ and ‘‘after’’ periods.However, the initial selection of sites to be sampled,the frequency of sampling, the size and number of sam-pling units, and the duration of the program all haveconsequences on the ability of a BACI, or any otherdesign, to detect impacts. Among the uncertainties thatmust be considered are the unknown amount of natural

1814 JOHN R. STEINBECK ET AL. Ecological ApplicationsVol. 15, No. 5

FIG. 1. Location of study area and intertidal sampling stations in central California, USA.

variation in the system over the future time period thatthe sampling will be done, the spatial extent of theimpact effects, the numerous species that could pos-sibly be affected, the potential differences in the mag-nitude of species’ responses, and the consequent effectof these on the statistical power to detect change.

Despite the potential power and usefulness of BACIdesigns there are few published examples for any eco-system, because of the need for repetitive samplingover a long period of time using the same samplingdesign, which is expensive and difficult to achieve ina commercial and regulatory framework. In one marineapplication, a BACI design was used in southern Cal-ifornia to detect the effects in a sand-dominated kelphabitat of a thermal discharge released subtidallythrough diffusers from a power plant (Schroeter et al.1993). Regular sampling over five years detectedchanges in 10 species. The macrophyte and macroin-vertebrate assemblages were not particularly speciose,so there were relatively few target species for sampling.Nevertheless, it was instructive that considerable sam-pling effort (18 surveys) was required to detect statis-tically significant changes in a few species, even foreffect sizes of ;75%.

This study of the thermal discharge from the DiabloCanyon nuclear power station (DCPP) on intertidal ma-rine communities was required by the U.S. Environ-mental Protection Agency under the National PollutantDischarge Elimination System Permit for the powerstation issued by the State of California (Central CoastRegional Water Quality Control Board). The purposeof the study was to determine the magnitude and spatialextent of the effects of the thermal discharge on themarine environment. The information from the studywould then be used by the regulatory board to deter-

mine if the thermal discharge limits set in the dischargepermit for the plant were protective of the marine en-vironment. The study was started in 1976 but delaysin construction resulted in the plant not starting fulloperation until 1986. As a result of these delays, aswell as regulatory decisions related to how monitoringresults were reported, a rigorous analysis of the datadid not occur until 1995. These events, therefore, pro-vided almost 10 years of data during operation, a periodthat included two ‘‘100-year’’ storms and several ElNino Southern Oscillation (ENSO) warming events(Dayton et al. 1999). Although the study was designedprior to the development of robust statistical methodsfor impact analysis (Green 1979, 1993, Stewart-Oatenet al. 1986, Underwood 1992, Wiens and Parker 1995),the problems of assessing power plants’ impacts wererecognized (Thomas et al. 1978), and fortunately sam-pling occurred in control and impact areas before andduring plant operation that allowed us to adapt the de-sign to a BACI analysis.

A critical feature in setting up the original samplingdesign was predictions about the spatial extent and be-havior of the thermal discharge plume. These predic-tions were made using a mathematical model and a1:75-scale physical model of Diablo Cove. The modelspredicted that the plume would mostly affect the south-ern half of Diablo Cove and exit the cove betweenDiablo Rock and the south point of the cove (Fig. 1).The models and preliminary sampling and studies onthe thermal tolerances of algae in the area (e.g., Abbottand North 1972) were used to predict the potentialbiological effects of the discharge, and the monitoringwas designed to test these predictions. Due to uncer-tainties about the accuracy of the model and to detecta possible gradient of change within the cove, sampling

October 2005 1815DETECTING LONG-TERM CHANGE

PLATE 1. Transect and quadrats at the 10.9 m MLLW tidal level at North Control Station NC-1 approximately 2.6 kmnorthwest of the Diablo Canyon Power Plant discharge. Photo credit: J. Steinbeck.

stations were set up throughout Diablo Cove and inareas to the north and south (Fig. 1). This proved tobe fortunate and crucial to detecting changes becausethe models did not predict that the plume would affectthe entire shoreline of Diablo Cove.

This paper has three primary objectives:1) Present the process used to develop a set of de-

cision rules used for determining the appropriate set ofstations, surveys, and species for BACI analysis. Thelong-term nature of the study combined with naturaldisturbances affected what data could be used in theBACI analyses. Although these types of decisions arenormally a part of the design of any biological impactsassessment (Clarke and Green 1988), they were uniquehere because they were applied to an existing studythat included a larger set of stations and surveys thatended up not being included in the analysis.

2) Show how these decision rules were used for ro-bust analyses to detect community-wide changes dueto the power plant, which discharges into the intertidalzone and produces a thermal plume that affects ;2 kmof coastline. The community-wide changes throughtime show the numerous indirect effects that can occurwith a prolonged press disturbance (Bender et al. 1984,

Schmitz 1997). The long-term nature of the study alsoallowed us to determine the effects of major naturaldisturbances, such as ENSO events and storms, on theability to detect change. These types of events can af-fect the analysis and interpretation of results, and, aswe discuss, need to be considered in the initial studydesign.

3) Use the results of the study to provide severalimportant lessons that may be useful to the design andanalysis of impact assessment in complex communi-ties.

METHODS

Site description

The DCPP is a nuclear-fuelled steam-turbine powerplant with an output of 2200 MW, located in centralCalifornia (358129 N, 1208529 W) (Fig. 1). Except forthe power plant, the coastline is largely uninhabitedand undeveloped (see Plate 1). The power plant hastwo generating units with separate condenser coolingsystems that are each served by two seawater pumps.The pumps for both units withdraw seawater from ashoreline intake structure ;70 m across with openings


that extend from approximately 21.5 m to a depth of28.7 m. After passing through the cooling system,;9.5 3 109 L/d of heated seawater is discharged intoDiablo Cove, which has a surface area of ;15 ha. Thetransfer of heat from the condensers of both units re-sults in a temperature at the point of discharge that is118C warmer than the seawater temperature at the in-take structure. The discharge structure is ;27 m highand 19 m wide at its base, the bottom of which is at atidal elevation of about 21.5 m MLLW. The first warm-water discharges occurred intermittently in late 1984with the initial testing of Unit 1. Nearly continuousthermal discharges began with the commercial opera-tion of Units 1 and 2 in May 1985 and March 1986,respectively.

The thermal plume from the DCPP discharge hasmomentum from the pumps and the 27-m drop in el-evation from the generating units to the shore, andbuoyancy from its increased temperature. The result isa surface layer of warm, buoyant water that turbulentlymixes with the cooler water in Diablo Cove as it flowsoffshore. The immediate receiving-water area is shal-low, ;3–4 m deep over a distance of 130 m from theoutfall. Thermal dilution from mixing with ambientwater in the immediate area of the discharge is affectedby tidal height, waves, current and wind conditions,and the volume and velocity of the discharge flow intoDiablo Cove. During low tides, subtidal ridges deflectthe thermal plume towards Diablo Rock and the northentrance to the cove (Fig. 1), whereas during high tidethe plume is not affected by the ridges and is more inline with the discharge structure and the south entranceto the cove. This movement results in the warm waterfrom the plume contacting a large portion of the DiabloCove shoreline. During periods when the plant is op-erating at full power the temperature is 28–48C abovethe average control station temperature at the southopening of Diablo Cove, and 58–68C above the averagecontrol station temperature at the north opening.

Sampling methods

Permanent sampling stations were established in1976 at 19 locations inside and outside of Diablo Coveto provide wide spatial coverage. Stations in the in-tertidal zone were on the mixed bedrock and bouldersubstratum that dominates the shoreline. Each stationconsisted of horizontal 30 m long permanent transects,one each at the 10.3-m and the 10.9-m MLLW tidallevels (see Plate 1). Ten 1-m2 quadrats were randomlyassigned along each transect at the start of the study,permanently marked, and then sampled in subsequentsurveys throughout the study (Greig-Smith 1964). Sta-tions were sampled at two-month intervals from April1976 through December 1987. After that, sampling wasreduced to two summer and two winter surveys through1995, which helped assure sampling during annualmaximum and minimum species’ abundances. Withina survey, sampling of all stations was usually com-

pleted within a 30-day period. The same core group ofbiologists did the sampling over the entire period, en-suring that the methods and taxonomic identificationswere consistent.

The sampling quadrat was a 1-m2 frame subdividedinto 16 equal subquadrats. Coverage of each macroal-gal species, and of bare substratum (rock, cobble, orsand not covered with macro-organisms) was recordedas the number of 1/16-m2 subquadrats covered plus thenumber of subunits (of the total of nine) of the sub-quadrat additionally covered (determined by visuallyseparating a 1/16-m2 subquadrat into nine equal parts).Species found in less than one subunit (i.e., 1/144 m2)in a quadrat were recorded as present. Overstory spe-cies were estimated first and then moved aside to allowfor estimates of understory and crustose species. Thesedata were later converted to percentages for analysis.Total algal cover per quadrat often exceeded 100% dueto the overlayering of multiple species.

Invertebrates were counted using two methods. In 5of the 10 quadrats all species were recorded as eitherpresent or absent, and individuals .2.5 cm in theirgreatest dimension were counted. In the other fivequadrats the same method was used, except that somespecies (e.g., predatory gastropods, anemones, turbansnails, and limpets) were counted regardless of size. Inall 10 quadrats black abalone were counted and thepercentage cover of encrusting invertebrates, such assponges and tunicates, was estimated as for algae.

Species that could not be identified in the field werecollected from outside the quadrats and identified inthe laboratory. Some species that were more difficultto identify in the field were combined into larger groupsfor analysis. Algae were identified according to Abbottand Hollenberg (1976), Scagel et al. (1989), and Silva(personal communication). Invertebrates were identi-fied according to Smith and Carlton (1975), McLean(1978), Morris et al. (1980), and Behrens (1991).

Temperature monitoring

Temperature recorders at the 10.6-m MLLW tidallevel at nine intertidal stations, were analyzed to de-termine the extent of shoreline contacted by the thermalplume. The recorders synchronously logged tempera-tures at 20-min intervals with an accuracy of 60.28C.Air temperatures were removed from the intertidal datausing information on tidal elevation collected fromNOAA data sources and from subtidal depth recorderslocated in Diablo Cove and the intake cove. Water tem-peratures recorded at stations NC-2 (North Control),SC-1 (located in Patton Cove), and at SC-3 (located 5km south of the plant; Fig. 1) were used in calculatingdelta T 8. Delta T 8 is the difference between the meancontrol station (ambient seawater temperature) and im-pact station seawater temperatures.

Analytical methods

A process for selecting the data for BACI analysiswas developed because of several complications that


arose over the 20 years of sampling that included.700 000 data entries (Fig. 2). These complicationsinvolved changes in the program that affected the num-ber of stations sampled, loss of stations due to stormimpacts, changes in some sampling procedures earlyin the program, realization of the actual extent of thethermal plume, and the inability to test some speciesdue to their low abundance and frequency of occur-rence. This process is presented in the Results, but herewe describe the procedures for the assumption testingand BACI analysis, which occurred after data selection(Fig. 2).

The data from the 10.3-m and the 10.9-m transectsat each station were analyzed separately because therewere differences in species composition and abundancebetween elevations during both periods. Combiningdata from both transect elevations in the analysis wouldhave affected the statistical analyses by introducing alarge number of very small or zero abundances for themany species that were not found at both elevations.

The BACI procedure itself analyzed the differences(deltas) between the mean of the control stations (Stew-art-Oaten and Bence 2001) and each of the impact sta-tions for the stations eventually selected for analysis.The deltas were defined as the meancontrol 2 meanimpact,so that the differences between the period mean deltasaccurately reflected the direction of change in abun-dances at the impact stations. This assisted in the in-terpretation of the large number of results and did notaffect the magnitude of the delta values.

The deltas for the stations were tested for additivity,linear trend, serial correlation, and homogeneity of var-iances. BACI and other ANOVA methods require thatdata conform to certain statistical assumptions (Stew-art-Oaten et al. 1986, Schroeter et al. 1993, Stewart-Oaten 1996). These tests were run on raw and trans-formed data using either log(x), 1/(x), or Ï(x) forcounts, or arcsine(x) for percentage cover. The trans-formations of the count data required that a constantbe added to account for observations with values ofzero. The value of the constant used can potentiallybias the analysis of the data (Clarke and Green 1988,Schroeter et al. 1993). Therefore, assumption tests wererun with constants of 1.0, 0.1, and .01 and the bestcombination of constants and transformations was cho-sen according to the rationale described below. All as-sumption testing was done at the 95% level of signif-icance.

Significant serial correlation, which violates the sta-tistical assumption of independence of errors, is a com-mon problem in sampling designs that include datacollected over time from the same sites. Serial corre-lation was tested using the ratio of mean square suc-cessive differences to the variance (von Neumann1942). The Tukey one degree of freedom test was usedto test the assumption of additivity (Tukey 1949), andthe assumption of homogeneous variances was testedusing Levene’s test (Milliken and Johnson 1984). A

regression of the preoperational period deltas againsttime was used to detect significant trends in the data.Due to the large sample size, it was expected that alarge number of the regressions would be significant.Therefore, the coefficient of determination (R2) wasalso calculated to determine the proportion of the var-iability explained by the regression. A data set wasanalyzed only if it met the assumption of additivity anddid not have a significant trend in the preoperationalperiod with an R2 of .0.5.

Data that passed assumption testing were analyzedusing the Proc Mixed procedure in SAS (SAS Institute1999). This program can analyze mixed-model ANO-VAs that include both random and fixed factors. Theprogram provides options for modeling the covariancestructure of random factors using several heteroge-neous variance models, including autoregressive mod-els (Littell et al. 1996). If significant serial correlationwas detected, an autoregressive term was added to theANOVA model in order to model the autocorrelatederror in the data. The data were run with a single au-toregressive error term for the entire data set, and alsowith separate autoregressive terms for each period, toaccount for potential differences in autocorrelation be-tween periods.

The Proc Mixed procedure also provides an optionfor using Satterthwaite’s adjusted degrees of freedomwhen results of the Levene’s test (Milliken and Johnson1984) indicated that the assumption of homogeneity ofvariances was violated. If both serial correlation andheterogeneity of variances were present in a data set,only the autoregressive term for modeling the hetero-geneous variance structure was added to the model.This followed from the assumption that the heteroge-neity was a result of the autocorrelated errors in thedata.

A BACI model, similar in design to models describedby Eberhardt and Thomas (1991) and Schroeter et al.(1993), was used to test the hypothesis that impact sta-tions were unaffected by the power plant discharge:

X 5 m 1 S 1 P 1 SP 1 Tijk i j ij k(j)

where Xijk 5 the delta value (after transformation, asappropriate) for impact Station i, Period j (preoperationor operation), and survey Time k (within Period j); m5 the mean difference across all impact Station (S),Period (P), and survey Time effects (T); Si 5 the effectof the ith impact Station; Pj 5 the effect of the jthPeriod; SPij 5 the Station 3 Period interaction; andTk(j) 5 the effect of the kth survey Time within the jthPeriod. The Station, Period, and Station 3 Period termswere defined as fixed effects in the SAS Proc Mixedprocedure, and survey Time within Period was definedas random. Proc Mixed uses a generalized least squaresapproach to estimate treatment effects for the model,in contrast to ordinary least squares used by most sta-tistical software (Littell et al. 1996). The estimates andstandard errors computed by the procedure incorporate


FIG. 2. Decision process for selecting stations, surveys, and taxa for analysis.


random effects in the model, such as the variationamong survey Times within Periods (Tk(j)) that couldaffect the estimated differences between periods.

The factor for station was included in the modelbecause it was highly likely that the effects of the dis-charge would vary among stations. While the majorityof effects from the discharge were expected to occurin Diablo Cove, stations from Field’s Cove and SouthDiablo Point were also included as impact stations inthe analysis. This increased the likelihood that the in-teraction between station and period effects would besignificant, because the temperature increases from thedischarge varied among the three impact areas. To al-low testing for discharge effects in the different areas,a set of a priori planned comparisons was used in theanalyses. The testing proceeded by first determining ifthere was a significant Station 3 Period interaction. Ifthere was no significant interaction among stations be-tween periods, the overall Period effect was used totest the hypothesis that there was no significant dif-ference in mean deltas between periods. If the inter-action was significant, comparisons between periodswere made for the three thermally homogeneous areas:Diablo Cove, South Diablo Point, and Field’s Cove.Comparisons between periods for the stations withinthe three impact areas were examined to determine thenature of the interaction within each area. A dischargeeffect would be indicated if the comparisons showedthat all of the stations within an area were affected, butto varying amounts.

A probability level of 90% was used in all analysesof discharge effects. This was chosen over the morecommonly used 95% to increase the statistical powerof the tests, thereby decreasing the probability of mak-ing a Type II error (Winer et al. 1991). This lowerprobability level increased the likelihood of finding sig-nificant changes where none may have occurred (TypeI error), but in ecological impact analyses it is impor-tant to balance this error against the potentially moreserious error of not recognizing a significant changewhen one has occurred (Type II error) (Mapstone1995). The power of a test is a measure of the prob-ability of correctly concluding that no change occurred(Winer et al. 1991). In these analyses, power was cal-culated as the ability of the test to detect a theoretical50% change in the parameter being tested (Schroeteret al. 1993). This was done by using the preoperationalperiod (Before) data and the (Before data)/2 for thesimulated operational (After) data. The error termsfrom the original data were used in computing the es-timates of power. This provided an indication of wheth-er nonsignificant results correctly indicated littlechange in the data, or merely reflected low detectionpower.

The combination of transformations, constants, andaddition of autoregressive error terms resulted in alarge number of potential analyses for each data set.The results of the assumption tests for each data set

were examined to determine if a particular set of trans-formations could be tested by BACI. A data set wasnot analyzed if the assumptions of additivity or trendwere violated and if none of the combinations of trans-formations passed the additivity or trend tests. The re-sults from all the assumption tests and BACI analysesfor a data set were ranked, based on the power of theanalyses to detect a 50% change in the data. If thepower was similar for two or more analyses for a singletaxon (,0.10 difference), the result of the Tukey testfor additivity was used to identify the transformationthat best met the additive assumption of the model. Ifthe choice of analysis was still not clear, transforma-tions that did not introduce autoregressive error termsor Satterthwaite adjustments to the data were chosenover those that did, and log transformations for countdata, and arcsine transformation for percentage coverdata, were chosen over other transformations.

In addition to BACI analysis, it was crucial to iden-tify the magnitude of change both for individual speciesand the community as a whole. We did this in two ways:by graphing abundances in the preoperational and op-erational periods for individual taxa, and by multivar-iate analysis of the community.

Multivariate analysis.—Correspondence analysis, amultivariate ordination technique for summarizingcommunity changes (Digby and Kempton 1987), wasused to contrast changes in Diablo Cove and controlareas. Data for the analyses were compiled by firstcalculating a mean abundance for individual algal andinvertebrate taxa from the quadrats at the upper andlower transect for each survey at the four Diablo Coveand three control stations used in the BACI analyses.The means at each station from both tidal levels werethen averaged into a mean for each survey for the Dia-blo Cove and control areas. The algal and invertebratedata were analyzed separately. Correspondence anal-ysis is sensitive to the influence of rare species (Min-chin 1987). To reduce this influence an analysis wasdone to determine the suite of species accounting forup to 99% of the total abundance, and then only thespecies from that list occurring in .20% of the surveysin each area were included in the analyses. For the algaldata, only foliose, noncrustose algae were analyzed.Invertebrates that were too small to be counted (,2.5cm) and always recorded as only present were not an-alyzed. Algae and invertebrates that were analyzedwere assigned values of 1.0 if recorded as present. Datafor invertebrates were also log-transformed to help ac-count for scale differences between species enumeratedusing counts and coverage estimates (Greenacre 1984).The first dimension survey scores were plotted overtime to contrast community changes during the pre-operational and operational periods.

RESULTS

The process for selecting the station, surveys, andspecies for analysis is described in the following sec-tions and shown in Fig. 2.


Stations.—A BACI analysis requires concurrentsampling of both control and impact stations. There-fore, although a large number of intertidal stations weresampled during the study, only the subset of the stationsthat were sampled consistently before and during plantoperation could be analyzed. Of the original 19 inter-tidal stations in 1976, only 11 could be analyzed at theend of the study in 1995. Severe storm waves in March1983 eroded cliffs along the shoreline and completelyburied two stations in south Diablo Cove. Seven of theoriginal 19 stations were sampled early in the study,but were sampled irregularly after the power plant wentinto operation and had insufficient data for analysis.Because of the extended period before plant operationwe were able to add a new station (South Diablo PointSDP-1) in 1981 that could be analyzed.

A critical decision in any BACI analysis is the al-location of control and impact sites. The largest effectsof the discharge were predicted to occur in the southernportion of Diablo Cove. However, after plant start-up,the largest temperature increases were measured innorth Diablo Cove. Temperature increases were alsomeasured in Field’s Cove, an area included in the orig-inal study as a control. This station, where increasedtemperatures were not predicted but subsequently oc-curred, was included with the other Diablo Cove impactstations. Using these criteria, three stations (NC-1, NC-2, SC-1) were designated as control stations, and eightstations (FC-3, NDC-1, NDC-2, NDC-3, SDC-2, SDP-1, SDP-2, SDP-3) were designated as impact stations(Fig. 1). The South Diablo Point stations (SDP-1, SDP-2, SDP-3) had greater wave exposure than the otherstations. Due to the difficulty of regularly sampling atthe lower tide level at South Diablo Point, only a singletransect was located at the 10.9-m tidal level. Thespecies composition at these 10.9-m tidal-level tran-sects was more similar to the 10.3-m transects fromother areas due to the greater wave exposure. There-fore, the deltas used in BACI analyses for the 10.9-mtransects at South Diablo Point were computed usingthe 10.3-m transect data from the control stations, andthe data were analyzed with the other 10.3-m data sets.

Surveys.—Although surveys began in April 1976,those done through 1977 were not analyzed because ofseveral alterations to the program, including changesin invertebrate sampling methods and personnel thataffected data consistency due to varying levels of tax-onomic discrimination. The remaining surveys (i.e.,1978–1995) were partitioned into preoperational (Be-fore) and operational (After) periods. The preopera-tional period was defined as 1 January 1978 to 31 De-cember 1984 (41 surveys), while the operational periodwas defined as 1 January 1987 to 30 June 1995 (36surveys). The preoperational and operational periodsexcluded the two years encompassing most of the start-up testing and initial operation of Units 1 and 2 (cf.Schroeter et al. 1993).

Surveys were also not included in the BACI analysisif all three control stations used in the analysis werenot sampled during a survey. Using these criteria, 34of the 42 surveys at the 10.9-m and 31 of the 42 atthe 10.3-m levels in the preoperational period and 29of the 36 surveys at both levels in the operational periodwere analyzed.

Taxa.—Nearly 400 intertidal taxa (invertebrates andalgae) were sampled, but most could not be statisticallyanalyzed for changes because they were encounteredtoo infrequently.

Algal taxa were selected for analysis by compilingseparate lists for the 10.3-m and 10.9-m transects, andpreoperation and operation periods. At each of the tran-sect elevations, the two lists of taxa comprising 99%of the noncrustose average algal cover from each period(grand mean percentage cover across all impact stationsand surveys) were combined into a single list for anal-ysis. Combining the taxa from the two periods, ratherthan compiling a single list computed from grand meanabundances across all surveys and stations, ensuredthat taxa that were rare in one period but abundant inthe other were accounted for in the analysis. This pro-cess resulted in 33 algal taxa at 10.3 m and 24 at 10.9m. Furthermore, we analyzed percentage cover of com-bined groupings of coralline and noncoralline crustosealgae and total noncrustose algal cover, percentage cov-er of bare rock, and species richness.

The list of invertebrates was developed similarly, butmean abundances were computed only from the fivequadrats per transect in which invertebrates werecounted regardless of size. Invertebrates that were enu-merated as percentage cover (e.g., sponges) were com-piled separately using data from all 10 quadrats. Theinvertebrates for analysis included taxa comprising thetop 99% of the cumulative abundance at 10.3-m and10.9-m levels. The process resulted in a list of 35invertebrate taxa at 10.3 m and 22 at 10.9 m. Speciesrichness was also analyzed.

Power plant discharge temperatures

Temperature increases above ambient within DiabloCove were first detected in 1985 when the power plantbegan start-up testing and commercial operation (Ap-pendix A). Disruptions to the volume of water in thedischarge flow occurred frequently in 1985–1986 dur-ing the initial phases of operation and testing, causingfluctuations in the discharge temperature. However, byautumn 1987 the plant was fully operational. The deltaT 8 between intake and discharge temperatures was;118C, although absolute discharge temperatures fluc-tuated seasonally with changes in ambient seawatertemperatures. Seawater temperature increases at the in-tertidal stations were far lower than at the point ofdischarge due to the warm-water plume mixing withambient seawater in Diablo Cove.

Ambient seawater temperatures within the area fluc-tuated seasonally and interannually. Highest ambient


seawater temperatures were recorded during ENSO pe-riods in 1983–1984, 1987, 1990, and 1993. Ambientseawater temperatures recorded at the three control sta-tions differed by an average of ,0.48C. The dispersionof the plume into Diablo Cove and adjoining areasresulted in varying levels of plume contact and tem-perature increase at the impact sites. During operation,the average delta T 8 at the intertidal stations insideDiablo Cove was 3.58C (Appendix A: Part A) and atSouth Diablo Point was ;1.48C (Appendix A: Part B).Farther north at Field’s Cove, the average delta T 8 was0.98C (Appendix A: Part C). Prior to operation (1978–1984) there were differences in monthly average sea-water temperatures between control and impact sites,but the differences were typically less than during op-eration.

Algae

Thirty-three algal taxa at 10.3 m and 24 taxa at 10.9m (not including crustose algal groups analyzed byBACI) comprised 99% of the mean algal cover in bothperiods. Some of these taxa represented species com-plexes (e.g., the geniculate corallines Calliarthron/Bos-siella) because of the difficulty of distinguishing someorganisms to species or genera. A striking feature ofthe data was the symmetry in abundances between pe-riods for the controls and the asymmetry between pre-operational and operational periods for the impact sta-tions (Fig. 3). Another general feature was that therewere very few algae with average abundances of $5%for either period.

At the control sites at both tidal elevations there weresmall proportional differences between periods for in-dividual taxa (Fig. 3A, B). One of the larger changeswas in Endocladia muricata at 10.9 m, but most chang-es in other species were ,1% in absolute cover. Themeans between periods were remarkably similar fortotal algal cover, algal diversity, and algal richness.

At the impact sites most foliose algae decreased inabundance between periods, especially at the upper tid-al level (Fig. 3B). The changes in foliose algal coverwere accompanied by increases in cover of bare rock,and in several ephemeral algal taxa including diatoms,filamentous red algal turf, and the green algal complexUlva/Enteromorpha. These green algae are usually as-sociated with disturbed habitats and early successionalstages in intertidal communities (Foster et al. 1988).Overall, there were large percentage changes in mosttaxa relative to control areas. These figures depict long-term averages over the preoperational and operationalperiods and include data from all of the impact stations.Therefore the figures do not show the full magnitudeof change in species such as Mazzaella flaccida, whichvirtually disappeared from the impact stations in DiabloCove, although not in Field’s Cove, by the end of thestudy.

Of the 67 algal data sets (including crustose algae,total cover, species richness, and bare rock) from both

tidal levels that were analyzed using a variety of trans-formations and constants, 20 (30%) did not pass thenecessary assumptions for BACI analysis (AppendixB), mostly because of failure to meet the assumptionof additivity among station deltas in the preoperationalperiod. Significant serial correlation in the preopera-tional deltas was detected in almost half (49%) of thedata sets. While significant serial correlation might beexpected for cover of bare rock, which tends to remainrelatively constant over time, it was not detected forthe two crustose algal groups, which also tend tochange slowly. There was no apparent pattern in thetaxa with significant serial correlations; it was detectedin foliose taxa with seasonal changes in abundance, aswell as in ephemeral taxa that typically experience con-siderable variation in abundance among years. Hetero-geneity of variances was detected in most data sets butwas corrected in the analyses using adjusted degreesof freedom.

In the BACI analyses a significant Period 3 Stationinteraction was detected for all of the data sets, andtherefore impacts were analyzed separately using threegroups of stations (Diablo Cove, South Diablo Point,and Field’s Cove) (Table 1A, B). At the lower tidallevel, significant differences between periods were de-tected in 69% (20 of 29) of the data sets from DiabloCove, 71% (17 of 24) from South Diablo Point, and59% (17 of 29) from Field’s Cove (Table 1A). SouthDiablo Point data were not included in five of the datasets because they did not pass the necessary assump-tions when those stations were included (see Fig. 2).Field’s Cove, the station with the least thermal impact,had the fewest significant changes. In addition, severalof the significant differences between periods detectedfor Field’s Cove represented relative increases or de-creases that were opposite to the effects detected inDiablo Cove. For example, Egregia menziesii andCorallina officinalis declined between periods in Dia-blo Cove, while they increased in Field’s Cove. Thereduced effects between Diablo and Field’s coves sug-gest a gradient of effects along the shoreline as thethermal plume dissipated. In contrast to Field’s Cove,the percentage of significant effects for South DiabloPoint was similar to Diablo Cove.

At the upper tidal level there were significant dif-ferences between periods in 83% (15 of 18) of the datasets for Diablo Cove and 67% (12 of 18) for Field’sCove (Table 1B), with a higher percentage of changesthan in the lower zone. Almost all abundances de-creased between periods in Diablo Cove except forGelidium pusillium and bare rock. Again, the fewersignificant differences found in Field’s Cove are con-sistent with reduced effects of the discharge at thismore distant site.

Invertebrates

Thirty-five invertebrate taxa at 10.3 m and 22 taxaat 10.9 m comprised 99% of the total abundance in


FIG. 3. Average algal abundances at the control and impact stations for the preoperation and operation periods at the (A)10.3-m MLLW and (B) 10.9-m MLLW transects. Period means for the control area were calculated from stations NC-1,


TABLE 1. Results of BACI ANOVA analyses for algae at the (A) 10.3-m MLLW tidal level, and (B) 10.9-m level transects.

Taxon Transformation

Power todetect 50%

change

P value

PeriodDiabloCove

SouthDiablo Pt.

Field’sCove

A) 10.3-m tidal elevation transectsAcrosiphonia spp. none 0.39 0.69 (0.19) 0.13 0.15Calliarthron/Bossiella spp. arcsine .0.99 ,0.01 (,0.01) ··· 0.45Callithamnion/Pleonosporium spp. arcsine .0.99 0.07 0.06 (0.07) 0.14Chondracanthus canaliculatus arcsine .0.99 0.01 (,0.01) ··· (0.44)Juvenile geniculate coralline algae arcsine 0.98 ,0.01 0.63 ,0.01 0.20Corallina officinalis arcsine .0.99 ,0.01 (,0.01) (,0.01) 0.03Cryptopleura violacea arcsine .0.99 0.19 (0.24) (0.22) (0.12)Egregia menziesii arcsine .0.99 ,0.01 (,0.01) (,0.01) ,0.01Endocladia muricata arcsine .0.99 ,0.01 (,0.01) (,0.01) (0.12)Filamentous red algae arcsine .0.99 ,0.01 0.14 (,0.01) (0.03)Gelidium coulteri none .0.99 0.10 0.26 0.01 (0.08)Mastocarpus jardinii arcsine .0.99 ,0.01 (,0.01) (0.57) (0.01)Mastocarpus papillatus none .0.99 0.24 0.93 (0.02) (0.07)Mazzaella affinis none 0.93 0.01 (,0.01) (0.18) (0.09)Mazzaella flaccida arcsine .0.99 ,0.01 (,0.01) (,0.01) (,0.01)Mazzaella phyllocarpa arcsine .0.99 ,0.01 (,0.01) (,0.01) (,0.01)Mazzaella leptorhynchos arcsine 0.68 0.08 (0.10) (0.18) (0.03)Mazzaella lilacina arcsine 0.97 ,0.01 (,0.01) ··· 0.25Microcladia coulteri arcsine 0.68 0.46 (0.03) 0.26 (0.71)Neorhodomela larix arcsine 0.68 0.02 0.54 ,0.01 ,0.01Osmundea spp. arcsine .0.99 0.34 (0.82) (0.07) (0.44)Porphyra spp. arcsine 0.78 0.51 (,0.01) 0.04 0.08Prionitis spp. none .0.99 ,0.01 ,0.01 ··· (0.01)Ulva/Enteromorpha spp. arcsine 0.98 0.62 0.09 (0.13) 0.08Coralline algae (nongeniculate) none .0.99 0.07 0.11 ··· 0.05Noncoralline algae (crustose) arcsine .0.99 ,0.01 ,0.01 0.02 0.04Total foliose algal cover none .0.99 ,0.01 (,0.01) (,0.01) (0.22)Species richness 1/(x 1 0.01) .0.99 ,0.01 (,0.01) (,0.01) (0.91)Bare rock substrate arcsine .0.99 ,0.01 ,0.01 ,0.01 0.01

B) 10.9-m tidal elevation transectsCalliarthron/Bossiella spp. arcsine .0.99 ,0.01 (,0.01) ··· (,0.01)Chondracanthus canaliculatus arcsine .0.99 ,0.01 (,0.01) ··· (0.04)Endocladia muricata none .0.99 ,0.01 (,0.01) ··· (0.02)Gastroclonium subarticulatum arcsine 0.96 ,0.01 (,0.01) ··· (0.02)Gelidium coulteri arcsine .0.99 ,0.01 (,0.01) ··· (,0.01)Gelidium pusillum none .0.99 0.02 0.01 ··· 0.46Mastocarpus papillatus arcsine .0.99 ,0.01 (,0.01) ··· (0.52)Mazzaella affinis arcsine .0.99 ,0.01 (,0.01) ··· (0.01)Mazzaella flaccida none .0.99 ,0.01 (,0.01) ··· (0.06)Mazzaella phyllocarpa arcsine .0.99 ,0.01 (,0.01) ··· (,0.01)Mazzaella leptorhynchos arcsine 0.72 0.29 0.45 ··· 0.04Pelvetia compressa none 0.96 0.01 (,0.01) ··· 0.29Filamentous red algae arcsine 0.18 0.37 0.30 ··· 0.77Coralline algae (nongeniculate) arcsine .0.99 ,0.01 (,0.01) ··· 0.76Noncoralline algae (crustose) arcsine .0.99 0.47 0.85 ··· 0.10Total foliose algal cover none .0.99 ,0.01 (,0.01) ··· (0.03)Species richness Ï(x 1 1.00) .0.99 ,0.01 (,0.01) ··· 0.85Bare rock substrate arcsine .0.99 ,0.01 ,0.01 ··· ,0.01

Notes: P values ,0.10 are in boldface to indicate significance. Paired comparisons of periods for South Diablo Point thatare marked with ellipses (···) were not included to better meet the assumptions of the analysis. For abundance that decreasedbetween periods, P values are in parentheses. All data sets had significant (P # 0.10) Station 3 Period interactions.

←

NC-2, and SC-1, and for the impact area from stations FC-3, NDC-1, NDC-2, NDC-3, and SDC-2 (stations SDP-1, SDP-2,and SDP-3 were included for 10.3-m level). The list includes foliose algae comprising 99% of the total abundance in thepreoperation and operation periods (not including crustose taxa and cover of bare rock that are also shown). The periodmeans exclude data from 1976–1977 and 1985–1986. Values ,10 are printed on the figure. Total average foliose algal coveris printed on the figures due to differences in scale with the individual taxa.


both periods. As for the algal data there was symmetryin abundances between periods for the controls andasymmetry between preoperational and operational pe-riods for the impact sites (Fig. 4). These figures depictlong-term averages over the preoperational and oper-ational periods and include data from all of the impactstations including Field’s Cove. The assemblage wasdominated by a few taxa. For example, at both tidallevels in control and impact areas the turban snail Te-gula funebralis was by far the most abundant species.Other abundant species were the anemone Anthopleuraelegantissima/sola, the limpet Macclintockia scabra,hermit crabs Pagurus spp., and the barnacle Tetraclitarubescens.

At the control sites at both levels there were gen-erally small differences between periods for individualtaxa (Fig. 4A, B). Exceptions were decreases in M.scabra and A. elegantissima at 10.9 m. However, spe-cies richness was remarkably similar between periods.In the impact area most taxa increased between periods(Fig. 4A, B) including T. funebralis, M. scabra, and T.rubescens at 10.3 m, and T. funebralis, M. scabra, andT. rubescens at 10.9 m. Two notable changes occurredin larger invertebrates. At the lower level purple seaurchins, Strongylocentrotus purpuratus, increased from4 to 32 urchins/m2 and at both elevations, and at bothcontrol and impact sites, there was a decline in blackabalone (Haliotis cracherodii) caused by ‘‘witheringsyndrome’’ disease, first observed in Diablo Cove in1988 (Steinbeck et al. 1992).

Of the 59 invertebrate data sets (including speciesrichness) from both tidal levels that were analyzed us-ing a variety of transformations and constants, nine(15%) did not meet the assumption of additivity (Ap-pendix C). As with the algal data sets, there was noapparent pattern to the taxa with significant serial cor-relations, which were detected in taxa that did not ex-perience large fluctuations in abundance such as theseastar Pisaster ochraceus and the chiton Nuttalina cal-ifornica, and in highly variable species such as hermitcrabs and turban snails. As for the algal data sets, serialcorrelations were accounted for in the analyses, andheterogeneity of variances were corrected using ad-justed degrees of freedom.

Impacts were analyzed using the same three groupsof stations as for the algae because a significant Period3 Station interaction was detected for almost all datasets (Table 2A, B). Of the 30 invertebrate data setsanalyzed at 10.3 m, significant differences betweenperiods were detected in 70% (21 of 30) of the datasets from Diablo Cove, 52% (15 of 29) from SouthDiablo Point, and 20% (6 of 30) from Field’s Cove(Table 2A). Invertebrate species richness was analyzedwithout the data from South Diablo Point because itdid not pass the necessary assumptions for analysiswhen those data were included (see Fig. 2). Again, thestation in Field’s Cove, with the least thermal exposure,had far fewer significant changes than in Diablo Cove.

The number of significant changes at South DiabloPoint was intermediate between Diablo Cove andField’s Cove, consistent with the thermal exposure eacharea received.

At 10.9 m there were significant differences betweenperiods in 65% (13 of 20) of the data sets for DiabloCove and 60% (12 of 20) for Field’s Cove (Table 2B).Although the number of significant changes was similarfor the two areas, there were several notable differencesin the species that changed. For example, the barnacleCthamalus fissus, the volcano limpet Fissurella vol-cano, M. scabra, and T. rubescens all increased in Dia-blo Cove, while no changes were detected in Field’sCove. However, some changes were detected in Field’sCove for invertebrates that were not detected in DiabloCove, such as Ocenebra and Lottia pelta. Therefore thesimilar number of changes detected for the two areasmay not indicate similar levels of impacts.

Statistical power

The power to detect a theoretical 50% change in thepreoperational mean abundances for the main periodeffect was .80% for 78 of the 97 data sets analyzed(Tables 1 and 2). There were two major reasons forlower statistical power for some of the taxa. One wasan artifact of the analytical methods. In some taxa withsignificant serial correlations, the autoregressive errorterm added to the model had the effect of increasingthe variance, thereby effectively lowering the power.Examples of this are Mazzaella leptorhynchus, fila-mentous red algae, and Nuttalina californica (Tables 1and 2, and Appendices B and C). Another cause of lowpower was the small difference in abundances betweenperiods for some taxa for the power calculations, whichdid not reflect the final, much larger, magnitude ofchange. Examples are Chthamalus fissus and Stron-gylocentrotus purpuratus. The high power for mosttaxa provided good evidence to conclude that failuresto detect a significant change were due to the absenceof change and not the inability to detect it (i.e., a TypeII error).

Finer-scale temporal effects

The interpretation of impact and recovery was great-ly complicated by the fact that the power plant becameoperational immediately after one of the strongestENSO events of the century, which affected controland impact sites almost equally, and by the high sea-sonal variability displayed by many taxa (Fig. 5). Theuse of deltas in BACI helped dampen the effects of theENSO and seasonal variability in the analyses, but ab-solute abundances reflected differences in recovery ofthe intertidal communities at control and impact sta-tions. For example, the total cover of foliose algae atimpact and control sites, except Field’s Cove, declinedduring the winter of 1982–1983 due to ENSO-relatedstorms (Fig. 5). In Diablo Cove, the algal cover con-tinued to fluctuate seasonally, but declined during the


FIG. 4. Average invertebrate abundances at the control and impact stations for the preoperation and operation periods atthe (A) 10.3-m MLLW and (B) 10.9-m MLLW transects. Period means for the control area were calculated from stationsNC-1, NC-2, and SC-1, and for the impact area from stations FC-3, NDC-1, NDC-2, NDC-3, and SDC-2 (stations SDP-1,SDP-2, and SDP-3 included for 10.3-m level). The list includes the invertebrates comprising 99% of the total abundance inthe preoperation and operation periods (separate lists compiled for count and cover invertebrates). The period means excludedata from 1976–1977 and 1985–1986. Values ,10 are printed on the figure. Values for Tegula funebralis are printed on thefigures due to differences in scale with the other taxa.


TABLE 2. Results of BACI ANOVA analyses for invertebrates at the (A) 10.3-m MLLW tidal level and (B) 10.9-m leveltransects.

Taxon Transformation

Power todetect 50%

change

P value

PeriodDiabloCove

SouthDiablo Pt.

Field’sCove

A) 10.3-m tidal elevation transectsAnthopleura elegantissima/sola log(x 1 1.00) .0.99 0.02 ,0.01 0.02 0.98Anthopleura xanthogrammica log(x 1 0.01) 0.93 0.03 0.06 0.04 0.01Balanus spp.† log(x 1 0.10) 0.91 0.92 0.86 0.85 (0.66)Encrusting bryozoans† arcsine 0.22 0.10 (0.07) (0.14) (0.90)Chthamalus fissus arcsine 0.30 0.18 ,0.01 (0.23) (0.24)Epiactis prolifera 1/(x 1 1.00) .0.99 0.04 (,0.01) (0.78) (0.58)Fissurella volcano 1/(x 1 1.00) .0.99 ,0.01 ,0.01 ,0.01 (0.99)Haliotis cracherodii Ï(x 1 0.10) .0.99 0.03 (,0.01) (0.49) (0.15)Leptasterias spp. log(x 1 0.10) .0.99 ,0.01 (,0.01) (0.02) (0.46)Lottia digitalis 1/(x 1 0.01) 0.10 0.96 0.13 (0.10) (0.73)Lottia limatula log(x 1 0.10) 0.98 0.01 ,0.01 (0.92) 0.09Lottia pelta log(x 1 0.01) 0.99 ,0.01 ,0.01 ,0.01 (0.02)Macclintockia scabra log(x 1 0.10) 0.93 0.03 ,0.01 0.52 (0.62)Mopalia spp. 1/(x 1 1.00) 0.97 0.38 (,0.01) 0.07 (0.68)Nemertea unidentified Ï(x 1 0.01) .0.99 0.11 (,0.01) 0.35 0.56Nuttallina californica 1/(x 1 1.00) .0.99 0.24 0.99 0.01 0.99Ocenebra spp. log(x 1 0.10) 0.99 ,0.01 (,0.01) (0.42) 0.45Pachygrapsus crassipes log(x 1 0.10) 0.60 ,0.01 ,0.01 0.01 ,0.01Pagurus spp. Ï(x 1 1.00) .0.99 0.04 0.58 (,0.01) (0.11)Pisaster ochraceus log(x 1 0.10) 0.96 0.02 0.01 0.01 (0.81)Pugettia spp.† log(x 1 0.10) 0.99 0.02 (0.07) (0.01) (0.15)Serpulorbis squamigerus 1/(x 1 0.01) 0.10 0.60 (0.92) (0.17) 0.48Spirorbidae† none 0.11 0.02 (0.05) (0.05) (0.04)Strongylocentrotus purpuratus 1/(x 1 0.10) 0.21 ,0.01 ,0.01 (0.03) ,0.01Tectura scutum Ï(x 1 1.00) .0.99 0.45 0.70 0.31 (0.65)Tegula brunnea log(x 1 1.00) .0.99 0.59 (0.03) 0.34 0.38Tegula funebralis log(x 1 1.00) .0.99 0.60 0.20 (0.75) (0.94)Tetraclita rubescens log(x 1 1.00) 0.99 0.92 0.57 0.25 0.63Colonial/social tunicates arcsine .0.99 ,0.01 (,0.01) (0.01) (0.62)Species richness 1/(x 1 0.01) .0.99 0.71 (0.43) ··· 0.29

B) 10.9-m tidal elevation transectsAcanthina spp. log(x 1 0.01) 0.75 0.71 0.86 ··· (0.42)Anthopleura elegantissima/sola Ï(x 1 1.00) 0.99 0.40 0.40 ··· 0.50Chthamalus fissus none 0.10 0.13 0.05 ··· (0.88)Cyanoplax spp. log(x 1 0.10) 0.93 ,0.01 ,0.01 ··· ,0.01Fissurella volcano log(x 1 0.10) 0.91 0.01 ,0.01 ··· (0.52)Haliotis cracherodii 1/(x 1 0.10) 0.94 0.97 (0.44) ··· ,0.01Lottia digitalis log(x 1 0.10) 0.82 ,0.01 ,0.01 ··· 0.01Lottia limatula Ï(x 1 0.01) .0.99 ,0.01 ,0.01 ··· 0.01Lottia pelta log(x 1 0.10) .0.99 0.08 0.16 ··· 0.01Macclintockia scabra log(x 1 0.10) .0.99 ,0.01 ,0.01 ··· 0.58Nuttallina californica log(x 1 0.01) 0.73 ,0.01 (,0.01) ··· (0.03)Ocenebra spp. log(x 1 1.00) .0.99 0.28 0.69 ··· (0.01)Pachygrapsus crassipes none .0.99 ,0.01 ,0.01 ··· 0.05Pagurus spp. log(x 1 0.10) .0.99 0.60 0.70 ··· 0.39Phragmatopoma californica arcsine 0.32 ,0.01 ,0.01 ··· ,0.01Tectura scutum† log(x 1 0.01) .0.99 0.44 0.84 ··· (0.06)Tegula brunnea log(x 1 0.10) .0.99 ,0.01 (,0.01) ··· 0.65Tegula funebralis Ï(x 1 0.01) .0.99 ,0.01 ,0.01 ··· 0.06Tetraclita rubescens log(x 1 1.00) 0.75 0.16 0.06 ··· (0.54)Species richness log(x 1 0.01) .0.99 ,0.01 ,0.01 ··· ,0.01

Notes: P values ,0.10 are in boldface to indicate significance. Paired comparisons of periods for South Diablo Point thatare marked with ellipses (···) were not included to meet the assumptions of the analysis. For abundance that decreased betweenperiods, P values are in parentheses. All data sets had significant (P # 0.10) Station 3 Period interactions except those taxamarked with a dagger (†) symbol.

† Period 3 Station interaction not significant at 0.10 level of probability.

operational period (Fig. 5A). At control stations algalcover fully recovered by 1987 to pre-ENSO levels (Fig.5B). The South Diablo Point stations had more seasonalvariation due to greater exposure to large winter waves(Fig. 5C). While recovery occurred to pre-ENSO levels

by 1988, there was a steady decline in average coverthrough 1995. This was interpreted as a delayed effectdue to the reduced exposure to the thermal plume inthat area, relative to Diablo Cove. It is noteworthy thatField’s Cove, which had the least exposure of the im-


FIG. 5. Average foliose algal abundances from both tidal levels for (A) Diablo Cove stations, (B) control area stations,(C) South Diablo Point stations, and (D) the Field’s Cove station, 1978–1995. The vertical line in each figure is the datewhen the power plant became fully operational.

pact sites to the thermal plume, had a significant re-duction in algal cover at 10.9 m, but for both tidallevels combined showed very little change (Fig. 5D).

Community effects

The first-dimension results from the correspondenceanalysis show that the primary sources of variationwere different for the control and impact areas and thealgal and invertebrate data sets. The variation amongfirst-dimension survey scores for the algal assemblageat the control sites was primarily attributable to sea-sonality (Fig. 6A). The pattern of seasonality was dis-rupted by the 1983 ENSO, but then returned to itspreoperational pattern by 1986. The first dimension forthe invertebrate assemblage in the control areas wasprimarily attributable to a response to the ENSO, asseen in the reduction in scores during 1983–1984 (Fig.7A). Similar to the algae, the invertebrates at the con-trol sites recovered and assumed the preoperational pat-tern by late 1986. The variation among the first-di-mension survey scores for Diablo Cove algal (Fig. 6B)and invertebrate (Fig. 7B) assemblages was mostly at-tributed to changes during the operational period, incontrast to the controls. This is seen for both the algaland invertebrate assemblages in Diablo Cove by strongdirectional trends throughout the operational period.Biologically, there were cascading effects within thecommunities following the loss of perennial foliosealgae. These included recruitment of grazers, bloomsof ephemeral algae, and settlement of barnacles in open

space (Schiel et al. 2004). The directional changes overtime are a clear indication that both the algal and in-vertebrate assemblages were still changing 10 yearsafter plant operation began.

DISCUSSION AND CONCLUSIONS

Establishing a causal argument for effects detectedin an observational study relies on the consistency ofthe patterns of change among species and on plausiblebiological mechanisms for them. Observational studiesof this type are also subject to biases, largely due tothe absence of any randomization in treatment assign-ment (Eberhardt and Thomas 1991). Where few specieschange or the direct signals detected in the analysesare not strong, it is difficult to attribute the changes toparticular causative factors or to eliminate alternativehypotheses for potential causes (Schroeter et al. 1993).In our study, however, there was clear evidence ofchange in most algae and invertebrates, with the af-fected communities in the impact areas changing frommultilayered algal domination to domination by graz-ing invertebrates.

Were these changes a result of the thermal discharge?The evidence seems unequivocal within the impact areaaround Diablo Cove, where statistically significantchanges after plant operation began were detected in;70% of the data sets analyzed. The multiple impactstations used in the study allowed us to identify thevariation in effects among stations, and to determineif the pattern was consistent with an expected gradient


FIG. 6. Correspondence analysis first-dimension surveyscores from analysis of average foliose algal abundances fromboth tidal levels for (A) control stations and (B) Diablo Covestations.

FIG. 7. Correspondence analysis first-dimension surveyscores from analysis of average invertebrate abundances fromboth tidal levels for (A) control stations and (B) Diablo Covestations.

of effects with increasing distance from the point ofdischarge. The reduced number and magnitude ofchanges detected in Field’s Cove and at South DiabloPoint are consistent with this pattern. Most majorchanges in all three impact areas occurred after thepower plant began operation, and stood in marked con-trast to the relative constancy of species abundances atthe control areas throughout most of the study. Al-though there were no experimental studies to determinethe mechanisms responsible for the changes (e.g., di-rect effects of temperature on adult mortality, or in-direct facilitative, interference, or grazing effects), thetiming and spatial pattern provide strong evidence forboth acute and chronic effects of the thermal dischargeoperating through a variety of direct and indirect path-ways.

These effects were comprehensive compared to thoseseen in a similar study by Schroeter et al. (1993) inthe subtidal zone of southern California. Although theywere able to detect statistically significant changes inabundance for 10 of 18 species over a five-year study,there were numerous species that could not be analyzeddue to low abundances and various violations of sta-tistical assumptions. They concluded that the changeswere most likely due to indirect physical effects of thedischarge diffuser, particularly increases in sedimen-tation and turbidity that may have affected larval set-tlement and reproduction, and caused mortality of ses-sile prey. In our study, there were also clear physicaleffects of the discharge on the intertidal zone that we

did not consider here. For example, there was consid-erable scouring and turbidity in the immediate vicinityof the discharge. While these undoubtedly affected theimmediate environment around the discharge, such im-pacts were anticipated and there were no stations lo-cated in these areas.

A complicating factor in our study was the inter-action between large-scale natural impacts to the entirecoastline during the 1983–1984 ENSO, mainly due tostorms, and the thermal elevation as the power plantstarted operation toward the end of the ENSO. Thishad several ramifications for the treatment of data andanalyses as well as for the ecological interpretation ofcommunity-wide effects. Most algae declined and in-vertebrates increased during the ENSO, which effec-tively lowered or raised the preoperational means. Thisaffected the assumption of additivity for a few specieswhere the responses to the ENSO were not consistentamong stations. The temperature increase due to thepower plant in 1985 resulted in additional thermalstress to populations already affected by the ENSO.However, the BACI analysis was remarkably robust tothis dramatic coast-wide natural event because the var-iable analyzed was the delta between the impact andcontrols. However, ecologically, what we looked forwas a return to the pre-ENSO condition for populationsand the communities. This occurred within two yearsat all the control sites, but not at the impact sites. Per-haps fortuitously, we were able to detect these impor-


tant trends only because of the long-term nature of thestudy. For example, if the study had begun two yearsprior to plant start-up, which is typical of most envi-ronmental impact studies, there would have been nopre-ENSO data allowing us to determine when, or if,recovery had occurred. In effect, the highly disturbedENSO period would have been the preoperational con-dition. This may not have affected the BACI analysisbecause of the use of the deltas, but it would have madean ecological interpretation of the results quite difficult.

What lessons does this study provide in designingand implementing environmental monitoring pro-grams? There are several: include redundant sites andstations; assume something unpredictable will happen;assume that you will not be able to analyze all dataand therefore be prepared to adapt your analytical ap-proach; and sample as long as possible to account fornatural variation. In addition, all studies, to some ex-tent, are based on the predicted or anticipated types,magnitude, and spatial extent of impacts. If these areincorrect, an inadequate monitoring program may re-sult.

In our study, key predictions proved not to be par-ticularly useful. For example, an early physical modelof Diablo Cove, on which the original allocation ofstations was predicated, proved to be inaccurate andunderestimated the potential spatial extent of the ther-mal plume. It was not anticipated that the plume wouldgreatly affect the northern portions of Diablo Cove orthat it would extend northwards into Field’s Cove. Thepredicted effects of the power plant discharge studiedby Schroeter et al. (1993) also varied considerably fromthe actual effects detected in their study. Ambrose etal. (1996) reviewed the predicted and actual impactsfrom the Schroeter et al. and other studies. They con-cluded that in most cases actual effects were usuallyvery difficult to predict, partially due to the complexityof ecosystem-level interactions.

Predicted species responses that were based on lab-oratory testing of potential thermal effects on somespecies at Diablo Canyon also generally underesti-mated the actual magnitude of change. For example,the foliose red alga Mazzaella flaccida, which domi-nated much of the midintertidal zone, was predicted tobe affected only in Diablo Cove during the warmesttimes of the year. However, by midway through theoperational period it had virtually disappeared from thecove, and the loss of this species and other foliose algaeprobably triggered cascading effects through the com-munity. Underwood and Peterson (1988) point out theproblems of using laboratory experiments to evaluateor predict the effects of pollutants. Recent predictionsof responses of intertidal taxa to ocean warming basedon their biogeographical distributions (Barry et al.1995, Sagarin et al. 1999) also were not useful in ex-plaining the number and magnitude of changes (Schielet al. 2004). Most of the species that changed signifi-cantly after plant operation were cosmopolitan in dis-

tribution, rather than near their northern or southernlimits.

Given the unanticipated effects it is perhaps sur-prising that our study was so robust at detecting chang-es to the intertidal communities. This was because theoriginal design did not rely solely on the predictionsof plume behavior, but also on the intuition of the orig-inal investigators for the potential spatial extent of im-pacts along a heterogeneous coastline. One of the orig-inal control stations (Field’s Cove) eventually provedto be impacted by the discharge, albeit substantiallyless than the stations in Diablo Cove. In other cases,stations were abandoned due to storm damage andcould not be used in the final analysis. Altogether 40%of the original stations could not be used in the finalanalysis. If the design had not included redundant con-trol and impact stations, the ability to detect the effectsof the discharge would have been compromised.

In practice it would only be possible to lose or elim-inate sites in most BACI studies; it is usually difficultto add sites in the Before period given the relativelyshort time of most studies, and no sites can be addedin the After period. Redundancy allows the necessaryflexibility that will almost always be required as a pro-gram proceeds. By flexibility we mean the ability toadapt the program to changing circumstances relatedto sites. This includes loss of stations; it also includesincreases in the spatial extent of impacts due to inter-action of the induced impact with natural variation,such as ENSO events, which may exacerbate impacts.In all cases it is obviously advisable to have multipleimpact and control sites (Underwood 1992, Stewart-Oaten and Bence 2001), not only because of the in-creased redundancy and spatial coverage, but also be-cause it allows for the possibility of eliminating sitesto meet statistical assumptions, and increases the abilityto estimate the effect as a function of distance fromthe plant or development, or of other spatial features,thus providing clearer evidence of cause (Schroeter etal. 1993). Having multiple control and impact sites atthe end of a study is probably achievable only throughhaving more sites than seems strictly necessary at thebeginning of a study.

Using this approach in other studies will probablyrequire following a process similar to the one devel-oped here for determining the appropriate stations, sur-veys, and species for analysis. The decision rules allowfor flexibility in determining the stations that are des-ignated as impacts and controls if the spatial extent ofeffects is larger than predicted. Other decisions involvethe appropriate alpha level for the statistical analyses.Our decision to use an alpha level of 0.10 was done apriori to maximize the detection of effects, even thoughthis theoretically doubled the Type I error rate. How-ever, in our study the decision to use the 0.10 level ofsignificance made little difference, because only 4 ofthe 97 data sets analyzed had P values between 0.05and 0.10. These and other decisions in the process need


to be clearly presented for regulatory agencies and re-source managers. In any study, the process is going toresult in the analysis of many fewer species than weresampled because of the need to meet statistical as-sumptions of the analysis.

The BACI analyses in this study, relative to otherimpact assessments (e.g., Schroeter et al. 1993), ben-efited from analytical improvements that allowed us toanalyze a large number of species. We were able toanalyze data that did not meet certain assumptions ofthe BACI model due to recent advances in statisticalsoftware for analyzing complex linear models with het-erogeneous error structures. For example, serial cor-relation was detected in the data for many species, butthese were analyzed by incorporating an autoregressiveerror structure into the ANOVA model using the SASProc Mixed procedure (Littell et al. 1996). Heteroge-neous variances were also accounted for by using Sat-terthwaite adjusted degrees of freedom in the analysis(Littell et al. 1996). The only species that were notanalyzed by the BACI model were those that violatedthe assumption of additivity among the preoperationalperiod deltas, and those having a significant trend dur-ing the preoperational period. Fortunately, these vio-lations occurred in only a few of the many speciesanalyzed.

A major strength of the study was the statistical pow-er to detect discharge effects. The average power todetect a theoretical 50% change in the preoperationaldeltas exceeded 80% in the great majority of species.This is far higher than the power in similar studies,such as Schroeter et al. (1993), where the power wasgenerally ,30% to detect a 50% change. The differencewas primarily due to the length of the study. Repli-cation in BACI analyses is derived from the numberof paired control–impact samples through time, andtherefore, the length of a study is critical to the powerof the analysis to detect impacts.

What does this study tell us about the length of timethat should be allocated to impact studies, since moststudies will not continue as long as ours? Within DiabloCove, most of the changes both to the algal and in-vertebrate communities were acute and occurred withintwo to three years after the plant became fully opera-tional. This included a shift from a multilayered algalassemblage to one dominated by grazers and other in-vertebrates (Schiel et al. 2004). The abundances of thedominant species at the impact stations were more var-iable than the pre operational community and continuedto change relative to the controls, as evidenced by themultivariate trajectories through time. However, out-side of Diablo Cove, particularly in Field’s Cove,changes were not acute and in fact took several yearsto occur. They may have been caused by lag effectsdue to reduced plume contact, or by the interaction ofthe plume with natural ocean warming due to severalENSO events during the operational period. These

more subtle effects would not have been detected in ashorter study.

Overall, our study highlights several cogent pointsfor programs of this nature. In summary, be flexible,plan for unpredictable events, include redundancies,and allow for longer study periods to account for nat-ural variation that may affect the nature of the impact.Based on prior knowledge of the species, community,or ecosystem under study, there may be good reasonto predict acute effects and their spatial extent. How-ever, our study showed the difficulty in predicting theeffects of anthropogenic disturbances on biologicalcommunities, and demonstrated that intuition and re-dundancy in the design of impact assessment studiesare also required. Detecting chronic and cumulativechanges will always be problematical. They may besubtle, involve relatively uncommon species, or resultfrom longer term lags in community processes or lifehistory events, such as a decline in reproduction orrecruitment (Ives and Gilchrist 1993, Reed et al. 1996).It is highly unlikely that these types of changes willbe predictable either in magnitude or timing, and short-er term studies may miss them altogether (Ambrose etal. 1996). How long is long enough? There is no de-finitive answer. In our study there were obvious acuteeffects due to the thermal discharge soon after opera-tion. This may not always be the case; studies withonly chronic or indirect impacts may take much longerto manifest. We believe that because of the uncertaintyof the responses, the only effective way to deal withboth acute and chronic changes is to use an approachthat samples a large number of species, and uses fre-quent analysis of data during the operational period sothat any trends of change will be detected early. Thisapproach would allow for decision making regardingthe need for additional field studies or, as Schroeter etal. (1993) point out, experiments that may provide astronger linkage between the impact and the biologicalresponses. Had we known that the study would con-tinue for such a long period of time, field experimentscould have been implemented to provide this linkageand the mechanisms for the changes, but laboratory orfield experiments generally will not substitute for awell-designed field assessment, since they will neverbe able to duplicate the complexity of responses to aspecific impact (Osenberg and Schmitt 1996).

The need for increased scope and length of studysuggested here provide several problems for regulatorsand resource managers who are involved in the finaldecisions on impact assessments. While it is importantto ensure that the study period is long enough to de-termine the actual magnitude and spatial extent of im-pacts, this approach will also result in increased sta-tistical power to detect increasingly smaller changesthat may not be ecologically significant. Unfortunately,the detection of these smaller changes is normally nota problem, since most impact assessments are con-ducted for much shorter periods of time, and are usually


only able to detect large-magnitude changes. Schroeteret al. (1993) suggest that a solution to the problem thatmost studies may not be able to detect effects on theorder of 50%, is to balance the costs of the Type I andType II error rates in a study (Fairweather 1991), butthis requires predictions regarding the magnitude of theimpacts. Our study shows the difficulty in making ac-curate predictions regarding the magnitude and espe-cially the spatial extent of impacts, which are moredifficult to detect since they will decrease with increas-ing distance from the source of the disturbance. Whilefew other studies will be able to provide the necessarysampling to detect the small changes found in ourstudy, this level of information has been extremelyvaluable to decision-makers, since most of the ambi-guities regarding the nature of the impact from thethermal discharge have been resolved. This has shiftedthe focus away from discussions regarding the mag-nitude and spatial extent of the impacts, and whether,in fact, they occurred as a result of the discharge, tothe legal interpretations of the regulations. ‘‘Science-based’’ decision-making works in this way, but it isstill up to the regulatory agency to determine whetherthe thermal limits for the discharge are ‘‘protective ofthe marine environment,’’ since regulatory goals thatinclude terms such as ‘‘protective,’’ ‘‘health,’’ ‘‘ben-eficial uses,’’ etc., are subjective and include valuejudgments that cannot be resolved by science. In thecase of the Diablo Canyon Power Plant, discussionsregarding a settlement for the environmental effects ofthe thermal discharge (and a decision on whether thethermal discharge limits are protective) are ongoing fora variety of reasons. Fortunately, these reasons do notinclude arguments regarding the nature of the impacts.

ACKNOWLEDGMENTS

Many people were involved in this study, but we especiallythank J. Blecha, J. Carroll, C. Ehrler, and S. Kimura, whowere involved in the data collection and all other aspects ofthe study from its inception in 1976, Susan Helberg and KimKubasek for assistance with data management, and PacificGas and Electric Company and the Central California Re-gional Water Quality Control Board for full access and useof the data set. We also thank Jay Carroll, Scott Kimura, andespecially Allan Stewart-Oaten for their helpful comments onearly drafts of this paper, and to Scott Kimura for the ‘‘tor-nado’’ graphs. The final content and organization of the paperalso benefited from comments by Joy Zedler and two anon-ymous reviewers. Finally, credit should be given to DaveMayer and others whose original study turned out to be re-silient to numerous impacts to its design.

LITERATURE CITED

Abbott, I. A., and G. J. Hollenberg. 1976. Marine algae ofCalifornia. Stanford University Press, Stanford, California,USA.

Abbott, I. A., and W. J. North. 1972. Temperature influenceson floral composition in California coastal waters. Pages72–79 in K. Nisizawa, editor. Proceedings of the SeventhInternational Seaweed Symposium. Wiley Interscience,New York, New York, USA.

Ambrose, R. F., R. J. Schmitt, and C. W. Osenberg. 1996.Predicted and observed environmental impacts: can we

foretell ecological change? Pages 345–369 in R. J. Schmittand C. W. Osenberg, editors. Detecting ecological impacts.Academic Press, New York, New York, USA.

Barry, J. P., C. H. Baxter, R. D. Sagarin, and S. E. Gilman.1995. Climate-related, long-term faunal changes in a Cal-ifornia rocky intertidal community. Science 267:672–675.

Behrens, D. W. 1991. Pacific coast nudibranchs: a guide tothe opisthobranchs from Alaska to Baja California. SeaChallengers, Monterey, California, USA.

Bender, E. A., T. J. Case, and M. E. Gilpin. 1984. Perturbationexperiments in community ecology: theory and practice.Ecology 65:1–13.

Clarke, K. R., and R. H. Green. 1988. Statistical design andanalysis for a ‘biological effects’ study. Marine EcologyProgress Series 46:213–226.

Dayton, P. K., M. J. Tegner, P. B. Edwards, and K. L. Riser.1999. Temporal and spatial scales of kelp demography: therole of oceanographic climate. Ecological Monographs 69:219–250.

Digby, P. G. N., and R. A. Kempton. 1987. Multivariateanalysis of ecological communities. Chapman and Hall,London, UK.

Eberhardt, L. L., and J. M. Thomas. 1991. Designing envi-ronmental field studies. Ecological Monographs 61:53–73.

Fairweather, P. G. 1991. Statistical power and design re-quirements for environmental monitoring. Australian Jour-nal of Marine and Freshwater Research 42:555–567.

Foster, M. S., A. P. DeVogelaere, C. Harrold, J. S. Pearse,and A. B. Thum. 1988. Causes of spatial and temporalpatterns in rocky intertidal communities of central andnorthern California. Memoirs of the California Academyof Science 9:1–45.

Green, R. H. 1979. Sampling design and statistical methodsfor environmental biologists. J. Wiley and Sons, New York,New York, USA.

Green, R. H. 1993. Application of repeated measures designsin environmental impact and monitoring studies. AustralianJournal of Ecology 18:81–98.

Greenacre, M. J. 1984. Theory and applications of corre-spondence analysis. Academic Press, London, UK.

Greig-Smith, P. 1964. Quantitative plant ecology. Second edi-tion. Plenum Press, New York, New York, USA.

Hurlbert, S. 1984. Pseudoreplication and the design of ecologicalfield experiments. Ecological Monographs 84:187–211.

Ives, A. R., and G. Gilchrist. 1993. Climate change and eco-logical interactions. Pages 120–146 in P. M. Kareiva, J. G.Kingsolver, and R. B. Huey, editors. Biotic interactions andclimate change. Sinauer Associates, Sunderland, Massa-chusetts, USA.

Littell, R. C., G. A. Milliken, W. W. Stroup, and R. D. Wol-finger. 1996. SAS system for mixed models. SAS Institute,Cary, North Carolina, USA.

Mapstone, B. D. 1995. Scalable decision rules for environ-mental impact studies: effect size, Type I and Type II errors.Ecological Applications 5:401–410.

McLean, J. H. 1978. Marine shells of southern California.Natural History Museum of Los Angeles County, Scienceseries 24, (revised edition), Los Angeles, California, USA.

Milliken, G. A., and D. E. Johnson. 1984. Analysis of messydata. Van Nostrand Reinhold, New York, New York, USA.

Minchin, P. R. 1987. An evaluation of the relative robustnessof techniques for ecological ordination. Vegetatio 69:89–107.

Morris, R. H., D. P. Abbott, and E. C. Haderlie. 1980. In-tertidal invertebrates of California. Stanford UniversityPress, Stanford, California, USA.

Osenberg, C. W., and R. J. Schmitt. 1996. Detecting ecolog-ical impacts caused by human activities. Pages 3–16 in R.J. Schmitt and C. W. Osenberg, editors. Detecting ecolog-ical impacts. Academic Press, New York, New York, USA.


Reed, D. C., A. W. Ebeling, T. W. Anderson, and M. Anghera.1996. Differential reproductive responses to fluctuating re-sources in two seaweeds with different reproductive strat-egies. Ecology 77:300–316.

Sagarin, R. D., J. P. Barry, S. E. Gilman, and C. H. Baxter.1999. Climate-related change in an intertidal communityover short and long time scales. Ecological Monographs69:465–490.

SAS Institute. 1999. SAS/STAT online user’s guide. Version8. SAS Institute, Cary, North Carolina, USA.

Scagel, R. F., P. W. Gabrielson, D. J. Garbary, L. Golden, M.W. Hawkes, S. C. Lindstrom, J. C. Oliveira, and T. B.Widdowson. 1989. A synopsis of the benthic marine algaeof British Columbia, Southeast Alaska, Washington andOregon. University of British Columbia, Phycological Con-tribution Number 3. Vancouver, British Columbia, Canada.

Schiel, D. R., J. R. Steinbeck, and M. S. Foster. 2004. Tenyears of induced ocean warming causes comprehensivechanges in marine benthic communities. Ecology 85:1833–1839.

Schmitz, O. J. 1997. Press perturbations and the predictabilityof ecological interactions in a food web. Ecology 78:55–69.

Schroeter, S. C., J. D. Dixon, J. K. Kastendiek, R. O. Smith,and J. R. Bence. 1993. Detecting ecological effects of en-vironmental impacts: a case study of kelp forest inverte-brates. Ecological Applications 3:331–350.

Smith, R. I., and J. T. Carlton, editors. 1975. Light’s manual:intertidal invertebrates of the central California coast. Thirdedition. University of California Press, Berkeley, Califor-nia, USA.

Steinbeck, J. R., J. M. Groff, C. S. Friedman, T. McDowell,and R. P. Hedrick. 1992. Investigations into a mortalityamong populations of the central California black abalone.

Pages 203–213 in S. A. Shepherd, M. J. Tegner, and S. A.Guzman del Proo, editors. Abalone of the world: biology,fisheries and culture. Blackwell Scientific, Oxford, UK.

Stewart-Oaten, A. 1996. Problems in the analysis of envi-ronmental monitoring data. Pages 109–132 in R. J. Schmittand C. W. Osenberg, editors. Detecting ecological impacts.Academic Press, New York, New York, USA.

Stewart-Oaten, A., and J. R. Bence. 2001. Temporal and spa-tial variation in environmental impact assessment. Ecolog-ical Monographs 71:305–339.

Stewart-Oaten, A., W. M. Murdoch, and K. R. Parker. 1986.Environmental impact assessment: ‘‘pseudoreplication’’ intime? Ecology 67:929–940.

Thomas, J. M., J. A. Mahaffey, K. L. Gore, and D. G. Watson.1978. Statistical methods used to assess biological impactat nuclear power plants. Journal of Environmental Man-agement 7:269–290.

Tukey, J. W. 1949. One degree of freedom for non-additivity.Biometrics 5:232–242.

Underwood, A. J. 1992. Beyond BACI: the detection of en-vironmental impacts on populations in the real, but vari-able, world. Journal of Experimental Marine Biology andEcology 161:145–178.

Underwood, A. J., and C. H. Peterson. 1988. Towards anecological framework for investigating pollution. MarineEcology Progress Series 46:227–234.

von Neumann, J. 1942. Distribution of the ratio of the meansuccessive difference to the variance. Annals of Mathe-matical Statistics 12:367–395.

Wiens, J. A., and K. R. Parker. 1995. Analyzing the effectsof accidental environmental impacts: approaches and as-sumptions. Ecological Applications 5:1069–1083.

Winer, B. J., D. R. Brown, and K. M. Michels. 1991. Sta-tistical principles in experimental design. McGraw-Hill,New York, New York, USA.

APPENDIX A

A figure showing monthly mean and delta T 8 intertidal (10.2 m MLLW) seawater temperatures at Diablo Cove, SouthDiablo Point, and Field’s Cove is available in ESA’s Electronic Data Archive: Ecological Archives A015-054-A1.

APPENDIX B

A table showing results of assumption testing for BACI ANOVA for foliose algae comprising 99% of the total abundancein the preoperation and operation periods at the 10.3-m MLLW tidal level and 10.9-m level transects is available in ESA’sElectronic Data Archive: Ecological Archives A015-054-A2.

APPENDIX C

A table showing results of assumption testing for BACI ANOVA for counted invertebrates comprising 99% of the totalabundance and five invertebrates with the highest percentage coverage in the preoperation and operation periods at the 10.3-m MLLW tidal level and 10.9-m level transects is available in ESA’s Electronic Data Archive: Ecological Archives A015-054-A3.

detecting long-term change in complex communities: a case study from the rocky intertidal zone

Documents