This article was downloaded by: [Monash University Library]On: 06 October 2014, At: 18:09Publisher: Taylor & FrancisInforma Ltd Registered in England and Wales Registered Number: 1072954 Registered office: Mortimer House,37-41 Mortimer Street, London W1T 3JH, UK

Transactions of the American Fisheries SocietyPublication details, including instructions for authors and subscription information:http://www.tandfonline.com/loi/utaf20

Predicting Invasions of North American Basses in JapanUsing Native Range Data and a Genetic AlgorithmKei'ichiro Iguchi a , Keiichi Matsuura b , Kristina M. McNyset c , A. Townsend Peterson c ,Ricardo Scachetti-Pereira d , Katherine A. Powers c , Dave A. Vieglais c , E. O. Wiley c & TaigaYodo ea National Research Institute of Fisheries Science , Ueda, Nagano, 386-0031, Japanb National Science Museum , Shinjuku, Tokyo, 169-0073, Japanc Biodiversity Research Center University of Kansas, Department of Ecology and EvolutionaryBiology , Lawrence, Kansas, 66045, USAd Centro de Referência em Informação Ambiental , Avenida Romeu Tórtima 388, 13084-520,Campinas, São Paulo, Brazile Japan Science and Technology Corporation , Ueda, Nagano, 386-0031, JapanPublished online: 09 Jan 2011.

To cite this article: Kei'ichiro Iguchi , Keiichi Matsuura , Kristina M. McNyset , A. Townsend Peterson , Ricardo Scachetti-Pereira , Katherine A. Powers , Dave A. Vieglais , E. O. Wiley & Taiga Yodo (2004) Predicting Invasions of North AmericanBasses in Japan Using Native Range Data and a Genetic Algorithm, Transactions of the American Fisheries Society, 133:4,845-854, DOI: 10.1577/T03-172.1

To link to this article: http://dx.doi.org/10.1577/T03-172.1

PLEASE SCROLL DOWN FOR ARTICLE

Taylor & Francis makes every effort to ensure the accuracy of all the information (the “Content”) containedin the publications on our platform. However, Taylor & Francis, our agents, and our licensors make norepresentations or warranties whatsoever as to the accuracy, completeness, or suitability for any purpose of theContent. Any opinions and views expressed in this publication are the opinions and views of the authors, andare not the views of or endorsed by Taylor & Francis. The accuracy of the Content should not be relied upon andshould be independently verified with primary sources of information. Taylor and Francis shall not be liable forany losses, actions, claims, proceedings, demands, costs, expenses, damages, and other liabilities whatsoeveror howsoever caused arising directly or indirectly in connection with, in relation to or arising out of the use ofthe Content.

This article may be used for research, teaching, and private study purposes. Any substantial or systematicreproduction, redistribution, reselling, loan, sub-licensing, systematic supply, or distribution in anyform to anyone is expressly forbidden. Terms & Conditions of access and use can be found at http://www.tandfonline.com/page/terms-and-conditions

845

Transactions of the American Fisheries Society 133:845–854, 2004q Copyright by the American Fisheries Society 2004

Predicting Invasions of North American Basses in Japan UsingNative Range Data and a Genetic Algorithm

KEI’ICHIRO IGUCHI1

National Research Institute of Fisheries Science,Ueda, Nagano 386-0031, Japan

KEIICHI MATSUURA

National Science Museum, Shinjuku, Tokyo 169-0073, Japan

KRISTINA M. MCNYSET AND A. TOWNSEND PETERSON

Biodiversity Research Center, University of Kansas,Department of Ecology and Evolutionary Biology, Lawrence, Kansas 66045, USA

RICARDO SCACHETTI-PEREIRA2

Centro de Referencia em Informacao Ambiental,Avenida Romeu Tortima 388, 13084-520 Campinas, Sao Paulo, Brazil

KATHERINE A. POWERS, DAVE A. VIEGLAIS, AND E. O. WILEY*Biodiversity Research Center, University of Kansas,

Department of Ecology and Evolutionary Biology, Lawrence, Kansas 66045, USA

TAIGA YODO3

Japan Science and Technology Corporation, Ueda, Nagano 386-0031, Japan

Abstract.—Largemouth bass Micropterus salmoides and smallmouth bass M. dolomieu have beenintroduced into freshwater habitats in Japan, with potentially serious consequences for native fishpopulations. In this paper we apply the technique of ecological niche modeling using the geneticalgorithm for rule-set prediction (GARP) to predict the potential distributions of these two speciesin Japan. This algorithm constructs a niche model based on point occurrence records and ecologicalcoverages. The model can be visualized in geographic space, yielding a prediction of potentialgeographic range. The model can then be tested by determining how well independent pointoccurrence data are predicted according to the criteria of sensitivity and specificity provided byreceiver–operator curve analysis. We ground-truthed GARP’s ability to forecast the geographicoccurrence of each species in its native range. The predictions were statistically significant forboth species (P , 0.001). We projected the niche models onto the Japanese landscape to visualizethe potential geographic ranges of both species in Japan. We tested these predictions using knownoccurrences from introduced populations of largemouth bass, both in the aggregate and by habitattype. All analyses robustly predicted known Japanese occurrences (P , 0.001). The number ofsmallmouth bass in Japan was too small for statistical tests, but the 10 known occurrences werepredicted by the majority of models.

Assessing the threat of species invasions and thepossible spread of such species is a global chal-lenge that requires a global perspective (Carlton

* Corresponding author: ewiley@ku.edu.1 Order of authorship is alphabetical, reflecting the

collaborative nature of this paper.2 Present address: Biodiversity Research Center, Uni-

versity of Kansas, Lawrence, Kansas 66045, USA.3 Present address: Department of Life Science, Fac-

ulty of Bioresources, Mie University, 1515 Kamihama,Tsu, Mie 514-8507, Japan.

Received October 9, 2003; accepted December 3, 2003

1996; Enserink 1999). To achieve this perspective,we can apply analytical tools and use informationon ecological landscapes that can be gathered ona global scale as well as specimen records to seekthe sets of factors that are useful in forecastingplaces where the establishment of species is pos-sible. The point is not necessarily to learn moreabout the ecology of the species—this would re-quire detailed study of local landscapes. Rather,the aim is to build models of the ecological nicherequirements of a particular species within its na-tive range with respect to factors that are availableglobally, to test whether these models can predict

Dow

nloa

ded

by [

Mon

ash

Uni

vers

ity L

ibra

ry]

at 1

8:09

06

Oct

ober

201

4

846 IGUCHI ET AL.

known occurrences of the species within its nativerange, and to use the models to forecast where onnew landscapes conditions might be conducive tothe establishment of populations. The point is notto model community interactions but to forecastinvasion threats under the null model that no bioticfactors prevent establishment of the species andgiven that community interactions can only bestudied after the introduction and when it is usuallytoo late to prevent establishment. In fact, the goalis not to understand the relationship between thebiology of the species and the environmental dataused to generate the models, although the modelsmay clarify the relationship directly or indirectlyand that is a matter for further study.

In this paper, we introduce such an approach tothe problem of predicting invasive species in fresh-water ecosystems. The analytical tool that we em-ploy is the genetic algorithm for rule-set prediction(GARP); the ecological and environmental infor-mation comes from climatic and topographic datasets available worldwide; and the specimen rec-ords for native distribution are those readily avail-able over the Internet, coupled with unpublishedrecords of the species where they have alreadyinvaded. Herein we report applications of this ap-proach, demonstrating accurate a posteriori pre-diction of a historical invasion and predicting pos-sible consequences of a more recent invasion. Inboth cases, we emphasize the need to take actioneither to stop or to mitigate these invasions. Thespecies in question are largemouth bass Microp-terus salmoides and smallmouth bass M. dolomieu,and the landscapes are North America and the Jap-anese archipelago.

Approximately 300 species belonging to 145genera and 53 families of freshwater fish are foundin the Japanese archipelago. Japanese freshwaterfishes include many native species in a wide va-riety of environments; indeed, although Japan is asmall country, it has at least 52 endemic speciesand subspecies (Kawanabe and Mizuno 1989) andthe latest Red List named 88 endangered speciesof freshwater fish (Japanese Ministry of the En-vironment 1999). More generally, island popula-tions appear to be more prone to extinction thanrelated continental populations (Smith et al. 1993;Frankham 1997).

Human activities such as introductions of non-native species and pollutants are important causesof species extinction on islands (Williamson1996). In fact, at least 26 species of nonnativefreshwater fish have successfully invaded Japanthrough introductions both intentional and unin-

tentional (Kawanabe and Mizuno 1989). Interna-tional agreements have been struck to reduce theintroduction and expansion of nonnative species(Bohn and Amundsen 2001).

One of the most damaging alien species in Japanis the largemouth bass, which has had impacts onpopulations of native aquatic organisms becauseof its role as a top predator (Shinya and Watanabe1990; Maehata 1992). This North American fishwas initially introduced into Lake Ashinoko in1925 as a fisheries resource and was strictly con-fined to the lake until the 1960s, when sportfishingwith lures became popular (Hosoya 1997). Illegalreleases by anglers spread the species across Ja-pan, and large populations are now established(Kawanabe and Mizuno 1989).

Genetic analyses of largemouth bass in Japandetected only two haplotypes in the ND1 regionof mitochondrial DNA (Kitagawa et al. 2000), sug-gesting that the fish were derived from a smallfounder pool. Introduced bass feed on a wide va-riety of aquatic organisms (e.g., fish, crustaceans,and insects), but individuals tend to shift to pis-civory as they grow (Yoshizawa et al. 1980; Shinyaand Watanabe 1990; Maehata 1992; Azuma andMotomura 1998; Yodo and Kimura 1998). Al-though growth rates vary among localities (Yodoand Kimura 1996), they fall within the ranges ob-served in North American populations (Bennet1937; McCaig and Mullan 1960; Smagula andAdelman 1983). In invaded areas, largemouth basshave had serious impacts on local communities,causing a decrease in total fish biomass (Takahashiet al. 2001). Recently, the smallmouth bass hasalso been introduced into Japan, and successfulreproduction has been confirmed in natural waterssince the mid-1990s (Iguchi et al. 2001). Althoughlittle is known about the ecological aspects ofsmallmouth bass in nonnative habitats like Japan,the threat of this species to native populations ofaquatic organisms is clear from their impact on anumber of aquatic vertebrates in North and MiddleAmerica.

The largemouth bass ranks among the most im-portant game fishes in North America. It is thelargest bass (up to 11 kg) and prefers clear, quietwaters with abundant aquatic vegetation (Scott andCrossman 1973; Carlander 1977; Lee 1980b). Al-though introductions have largely masked their na-tive range, largemouth bass seem originally tohave been distributed from northeastern Mexiconorth to southern Quebec and Ontario in the Mis-sissippi River drainage, east to Florida, and alongthe Atlantic coast to central South Carolina (Hubbs

Dow

nloa

ded

by [

Mon

ash

Uni

vers

ity L

ibra

ry]

at 1

8:09

06

Oct

ober

201

4

847PREDICTING BASS INVASIONS IN JAPAN

and Lagler 1964; Carlander 1977; Lee 1980b; Pageand Burr 1991). They are found in a variety ofnative habitats, do very well in artificial impound-ments, and can tolerate higher turbidity than con-geners. The largemouth bass is one of the mostwidely introduced fishes in the world, having beenintroduced into at least 60 countries worldwide(McDowall 1968; Robbins and MacCrimmon1974; Carlander 1977) as well as throughout NorthAmerica wherever suitable reservoirs are found(Lee 1980b). Carlander (1977) presented extensivelife history data on the species, including data onintroduced populations.

In contrast to largemouth bass, smallmouth bassprefer clear, fast-flowing streams and flowingpools with gravel bottoms (Lee 1980a) and wereoriginally distributed from Minnesota and south-ern Quebec south to the Tennessee River drainageof Alabama and west to eastern Oklahoma (Hubbsand Lagler 1964; Carlander 1977). In terms ofstream habitat, the smallmouth bass behaves as thenorthern equivalent of the spotted bass M. punc-tulatus and inhabits intermediate habitats when allthree species are found sympatrically (i.e., certainstreams in Ohio: Trautman 1981). Although small-er than the largemouth bass, its fighting qualitiesare appreciated by anglers. It has been introducedinto 14 countries as well as into suitable habitatthroughout nonnative portions of North Americabeginning in the 1850s (Carlander 1977). Its abil-ity to extend its range without direct human intro-duction is demonstrated by its colonization of theHudson River drainage after the opening of theErie Canal in the 1820s (Hubbs and Bailey 1938).

As large, piscivorous predators, basses pose se-rious threats to native fish faunas (U.S. GeologicalSurvey 2003). Both largemouth and smallmouthbass were introduced into California as early as1874 (Emig 1966a, 1966b). In the western UnitedStates, introduced basses have been implicated inthe decline of several species of minnow (tui chubGila bicolor, roundtail chub G. robusta, Gila chubG. intermedia, White River spinedace Lepidomedaalbivallis, relict dace Relictus solitarius, and speck-led dace Rhinichthys osculus), pupfishes (Owenspupfish Cyprinodon radiosus and other Cyprinodonspp.), and one goodeid (White River spring fishCrenichthys baileyi) (Miller and Pister 1971;Minckley 1973). U.S. Fish and Wildlife Servicerecovery plans detail the impacts of these basseson other species (U.S. Fish and Wildlife Service1985, 1994). Introduction of largemouth bass toLake Atitlan, Guatemala, is implicated in the elim-ination of several native fishes, reduction of total

fish biomass, and predation on and competitionwith the now-extinct Atitlan grebe Podilymbus gi-gas (LaBastille 1974). Introductions of small-mouth bass decreased populations of brown troutSalmo trutta and introduced bass tapeworms tospecies such as yellow perch Perca flavescens inLake Opeongo, Ontario (Martin and Fry 1973).Smallmouth bass prey on juvenile Pacific salmo-nids and smolts in the Columbia River (Dentler1993; Tabor et al. 1993) and are the most signif-icant predator of Chinook salmon Oncorhynchustshawytscha and wild steelhead O. mykiss subyear-lings in reservoirs on the lower Snake River (Ben-nett 1998). Bass introductions have also been im-plicated in the declines of populations of the Chir-icahua leopard frog Rana chiricahuensis and theCalifornia tiger salamander Ambystoma califor-niense in California and Arizona (Hayes and Jen-nings 1986; Rosen et al. 1995; Dill and Cordone1997). Jenkins and Burkhead (1994) have specu-lated that bass introductions caused the extinctionof an isolated population of trout-perch Percopsisomiscomaycus in Virginia.

Recent advances in modeling species’ invasions(Peterson and Vieglais 2001) are now being ap-plied to aquatic organisms. Bass invasions of Ja-pan from North America provide an ideal test case,as distributional data are available for both nativeand invaded ranges, permitting quantitative as-sessment of model success or failure. Moreover,although largemouth bass are now well establishedin the southern islands of Japan, smallmouth bassare only beginning to establish populations there,leaving open the possibility that eradication is stillfeasible. Accurate distributional predictions couldhelp to focus eradication efforts. Further, neitherspecies is established in Hokkaido, and assessingthe threat of both species to introduction into Hok-kaido will provide decision makers with a threatassessment for that island. To that end, we developecological niche models for both species based onknown native occurrences in North America, pro-ject them to Japan, and test the accuracy of thepredictions using known occurrences in Japan.

Methods

Occurrence data for the two bass species inNorth America were obtained primarily fromFishNet, a distributed biodiversity data resourceproviding access to natural history museum da-tabases through the Species Analyst (http://speciesanalyst.net); the institutions providing datainclude the University of Florida, Tulane Univer-sity, the Museum of Comparative Zoology of Har-

Dow

nloa

ded

by [

Mon

ash

Uni

vers

ity L

ibra

ry]

at 1

8:09

06

Oct

ober

201

4

848 IGUCHI ET AL.

vard University, the University of Kansas NaturalHistory Museum, the Illinois Natural History Sur-vey, and the University of Michigan Museum ofZoology. We obtained 2,046 unique occurrencepoints for largemouth bass and 903 unique occur-rence points for smallmouth bass within their pre-sumed native ranges (summarized by Lee 1980a,1980b). To reduce the bias caused by uneven pointdispersal, we subsampled states with more than100 unique records for largemouth bass by se-lecting 100 records at random, yielding a total of1,085 points. From this pool, we selected 100points at random and set them aside for testing theGARP models within the native range (extrinsictest data) and used 985 points in the modelingprocess (intrinsic training data). Smallmouth bassrecords were processed in a similar fashion exceptthat only 50 points were included from each statethat had more than 50 records, 50 records werereserved for testing as the extrinsic test data, and607 points were used as intrinsic training data inthe modeling process. For Japanese localities, oc-currence points (634 for largemouth bass, 10 forsmallmouth bass) were accumulated from annualreports on inland sportfishing (Nakama 1996; Oka-be 2000) and were georeferenced using Zenrin(1998).

We used 10 ecological–environmental geo-graphic data sets (‘‘coverages’’) as the dimensionsof the ecological niches to be modeled. Topograph-ic and hydrologic data (elevation, slope, aspect,topographic index, flow accumulation, and flowdirection) were obtained from the Hydro-1K dataset (Land Processes DAAC 2003; 1-km spatial res-olution raster coverages). ArcAtlas (ESRI 1996)supplied vector coverages summarizing averageannual temperature and precipitation. Land use,land cover, and percent tree cover were obtainedfrom the University of Maryland Global LandCover Facility (2003). All coverages were clippedto the regions 23.5–518S, 65–1278W (North Amer-ica) and 26–478N, 127–1548E (Japan) and griddedto a resolution of 0.018 for analysis.

Ecological niches and potential geographic dis-tributions were modeled with GARP (Stockwelland Noble 1992; Stockwell 1999; Stockwell andPeters 1999). In general, the procedure focuses onmodeling ecological niches (i.e., the conjunctionof ecological conditions within which a species isable to maintain populations: Grinnell 1917). Spe-cifically, GARP relates the ecological character-istics of known occurrence points to those ofpoints randomly sampled from the rest of the studyregion, seeking to develop a series of decisionrules that best summarize the factors associatedwith the species’ presence.

Initially, GARP divides the intrinsic trainingdata mentioned above (e.g., the 1,085 points forlargemouth bass) into two intrinsic data sets at aratio specified by the investigator. One intrinsic setis used to generate rules (rule-generating data); theother is used to test the rules within each iteration(rule-testing data). It is important to note that these‘‘intrinsic rule-testing data’’ are different from theextrinsic test data (e.g., the 100 points for large-mouth bass) that are used to evaluate the overalloutcome of the modeling process. The investigatorcan determine the ratio of intrinsic rule-generatingand intrinsic rule-testing data; in these analyses,we used a ratio of 80:20 to maximize the numberof points used for training while reserving an ad-equate number of points for testing.

The GARP procedure includes several distinctalgorithms for niche modeling in an evolutionary-computing environment. It works in an iterativeprocess of rule selection, evaluation, testing, andincorporation or rejection. A rule is chosen froma set of possibilities (e.g., logistic regression andbioclimatic rules), applied to the intrinsic rule-generating data, and allowed to evolve. Rules mayevolve by a number of means that mimic DNAevolution: point mutations, deletions, crossingover, and so forth. Predictive accuracy is then eval-uated on the basis of 1,250 points that are resam-pled (with replacement) from the pool of intrinsicrule-testing data and 1,250 points that are sampledrandomly from the study region as a whole. Thechange in predictive accuracy from one iterationto the next is used to evaluate whether a particularrule should be incorporated into the model; thealgorithm runs until it has either performed 1,000iterations or achieved convergence (convergenceis determined by a criterion specified by the in-vestigator; we selected a default value of 0.01).

All of the modeling in this study was carriedout via a desktop implementation of GARP(Scachetti-Pereira 2001). The algorithm’s random-walk process can produce many models that mustthen be evaluated to choose a ‘‘best-model’’ set(Anderson et al. 2003). The procedure is based onthe observations that (1) models vary in quality,(2) variation among models involves an inverserelationship between errors of omission (leavingout the true distributional area) and commission(including areas not actually inhabited), and (3)the best models (as judged by experts who areblind to the error statistics) cluster in a region ofminimum omission of independent test points andmoderate area predicted (an axis related directlyto commission error). Hence, among the 1,000 rep-

Dow

nloa

ded

by [

Mon

ash

Uni

vers

ity L

ibra

ry]

at 1

8:09

06

Oct

ober

201

4

849PREDICTING BASS INVASIONS IN JAPAN

TABLE 1.—Description of data used in a general algorithm for rule-set prediction model analysis, including the numberof training points used to generate rule-set predictions in the native range (Train), the number of testing points used totest the models (Test), the number of testing points predicted by the intersection of all 10 models, 6–9 models, and 1–5 models, and the area under the curve (AUC), standard error (SE), and Z-score associated with receiver–operator curveanalysis. Native range refers to the presumed native range of each species in North America; all habitats refers to theaggregation of Japanese records from specific habitats (lake, reservoir, etc.); P , 0.001.*

Analysis Train Test

Model subsets

10 6–9 1–5 AUC SE Z

Largemouth bass

Native range 985 100 97 2 1 0.840 0.03 16.91*

Japan

All habitatsLakeReservoirPondCreekRiver–stream

63443

20324740

101

57732

1832253899

5611192222

101000

0.8370.8010.8350.8370.8450.851

0.010.040.020.020.040.02

43.40*10.38*24.41*27.06*10.80*24.41*

Smallmouth bass

Native rangeJapan

607 5010

506

03

01

0.967a

0.02 12.05*

* P , 0.001.a Too few points for statistical analysis.

licate models, we (1) eliminated all of the modelswith omission errors greater than zero (based onthe internal test points), (2) calculated the medianarea in which species were predicted to be present(as a percentage of the area analyzed) among thesezero-omission points, and (3) identified the 10models closest to the overall median distributionalarea as the basis for further analysis. Ten modelsis the minimum number needed for further statis-tical tests.

The 10 best models were then tested using theoriginal extrinsic testing data set (i.e., the 100 datapoints for largemouth bass and 50 data points forsmallmouth bass) segregated from the native rangeoccurrence points of each species. Each of themodels was imported into ArcView (ESRI 1996),and the sum of the models visualized in geographicspace as a mosaic of model intersections (Figure1). That is, for any one area within North America,a particular pixel might carry the prediction that10 models predict presence, that 5 models predictpresence, or that no models predict presence. Theextrinsic test data are overlain on the mosaic ofmodels and evaluated through receiver–operatorcharacteristic (ROC) analysis (Hanley and McNeil1982; Centor 1991; Vida 1993). This analysis teststhe sensitivity (a measure of omission error) andspecificity (a measure of commission error) of theprojected mosaic of models relative to their abilityto successfully predict the presence of the extrinsictest data points (e.g., those 100 points withheld

entirely from the modeling process for largemouthbass). In short, ROC scores are maximized for anyparticular set of models when all of the extrinsicdata points fall into areas for which all 10 modelshave predicted species presence and are minimizedwhen the extrinsic data points are as likely to befound where the models predict species absenceas where they predict presence. Thus, if the modelsdo no better than randomly predict the occurrenceof the extrinsic test data, they will receive a scoreof 0.50; if the models do a perfect job of predictingthe extrinsic test data points, they will receive ascore that approaches 1.0 (mediated by the size ofthe landscape and the area of the landscape forwhich presence is predicted). The departure fromrandomness can be easily tested using the Z-statistic, yielding a test of model set accuracy.

To enable visualization of ecological niche mod-els in geographic dimensions (i.e., as a map),GARP includes a module that projects rule-setsonto landscapes. Although the Web-based version(BIODI 2003) simply projects the rule-set onto thesame coverages from which the model was de-veloped, the desktop version permits projection toalternative landscapes as well. This step is key topredicting species’ invasions, as multiple coveragesets can be included for visualization of areas fit-ting the model conditions (Peterson and Vieglais2001). We projected the models onto the nativedistributional area to permit quantitative tests of

Dow

nloa

ded

by [

Mon

ash

Uni

vers

ity L

ibra

ry]

at 1

8:09

06

Oct

ober

201

4

850 IGUCHI ET AL.

FIGURE 1.—Geographic predictions for largemouth bass in (a) North America and (b) Japan. White circlesrepresent the intrinsic training data used to model the native distribution in North America. On the North Americanlandscape, yellow triangles represent the extrinsic data used to test the native-range model; on the Japaneselandscape, yellow triangles represent the known occurrences used to test the predicted geographic range. On bothlandscapes, pink to dark brown areas represent the intersection of various model sets, dark shades representing theintersection of all 10 models and light shades the intersection of fewer models. The maps are not to scale relativeto each other.

model quality as well as onto Japan to predict spe-cies’ potential ranges.

Results

Largemouth Bass

Visualization of the 10-best model set from thenative range is shown in Figure 1a. The results

were highly significant, 97 of 100 test points fall-ing within the 10-model intersection (Table 1).Projecting the niche model onto Japan (Figure 1b)reveals that the species might be present over muchof the landscape, including Hokkaido. Of the 634test point occurrences from Japan, 577 were pre-dicted by all 10 models, yielding a highly signif-

Dow

nloa

ded

by [

Mon

ash

Uni

vers

ity L

ibra

ry]

at 1

8:09

06

Oct

ober

201

4

851PREDICTING BASS INVASIONS IN JAPAN

FIGURE 2.—Geographic predictions for smallmouth bass in (a) North America and (b) Japan. See the caption toFigure 1 for additional details.

icant prediction (Table 1) that was not significantlydifferent from the ROC score received for test datadrawn from the native range (P . 0.2149). Themodels also accurately predicted the Japanesepoints when they were parsed among the varietyof habitats into which largemouth bass have beenintroduced (Table 1).

Smallmouth Bass

Visualization of the 10-best model set from thenative range is shown in Figure 2a. All 50 extrinsictest data points were predicted by all 10 models,

yielding an ROC area-under-the-curve score of0.9667 (SE 5 0.0178; Z 5 112.052; P , 0.001).Projecting the niche model onto Japan (Figure 1b)reveals that the species would be present overmuch of the landscape, including Hokkaido. Ofthe 10 Japanese point occurrences that have beenrecorded, 6 were predicted by all 10 models whileall of them were predicted by 5 or more models.We did not attempt an ROC analysis due to thelow number of introduced occurrence points. AllJapanese smallmouth bass test data points weredrawn from lake samples.

Dow

nloa

ded

by [

Mon

ash

Uni

vers

ity L

ibra

ry]

at 1

8:09

06

Oct

ober

201

4

852 IGUCHI ET AL.

Discussion

The distributional areas predicted for large-mouth bass in Japan showed excellent correspon-dence with known occurrences, and the statisticalsignificance of the results indicates that our eco-logical niche model based on native-range NorthAmerican occurrence points has excellent predic-tive ability. One might wonder why we modeledboth species only in their native ranges rather thanthroughout their present North American range.There are two reasons. First, modeling only withinthe native range provides the most conservativeassessment of the threat. Second, many potentialinvasive species are known only from their nativeranges. To assess the utility of ecological model-ing, it is important to demonstrate that modelingwithin a species’ native range will yield accuratepredictions of its ability to invade new areas. Thisstudy takes advantage of the introduction of twoinvasive species that affords the possibility of test-ing such predictions quantitatively. Unfortunately,largemouth bass have already occupied many po-tential habitats in the southern islands of Japanand are now widely distributed in the country.However, this species has not yet been introducedinto the island of Hokkaido. Our results indicatethat it is a potential threat there, and serious at-tention should be given to blocking its introduc-tion.

Very few large, carnivorous fishes are native toJapanese freshwater habitats. A cyprinid, Opsa-riichthys uncirostris uncirostris, that is endemic toLake Biwa feeds diurnally on fish (Tanaka 1964),and a catfish, Siurus asotus, and an eel Anguillajaponica, both widely distributed over the Japa-nese archipelago, forage on fish at night (Kobay-akawa 1989; Tabeta 1989). Except for Opsari-ichthys, large, diurnal, carnivorous freshwater fish-es are absent, providing an apparent ecologicalvacuum that is now filled by invasive largemouthbass.

Considering the predictive power of our modelsfor largemouth bass, for which detailed tests wereboth possible and highly significant, our confi-dence in our prediction of the potential distribu-tional areas for smallmouth bass in Japan is high.The present distribution of this species is limitedto a few lakes and is much less extensive than thepredicted area overall. This limited range probablyresults from its comparatively recent invasion.

The wide distribution of largemouth bass in Ja-pan is apparently explicable by the numerous re-leases by sport anglers. Smallmouth bass com-

monly occur in running water, which may indicatemuch better capacity for dispersal in Japan. Inter-estingly, largemouth bass that are introduced intolakes often decrease in numbers after the releaseof smallmouth bass (Iguchi and Yodo, unpublisheddata). Displacement of congener alien species maysupport the idea that lack of competitors had fa-cilitated the invasion of alien species. However,this observation does not argue for the introductionof new alien species to compete with others; onthe contrary, such an introduction would make thematter worse. Of particular interest is that the pre-dictions for both basses include Hokkaido, yet nei-ther has invaded there. Native salmonids are dom-inant there, however, and may either prevent alieninvasions or be vulnerable to them, as they are inthe Pacific Northwest of the United States (Taboret al. 1993; Bennett 1998).

An important means of conserving native bio-diversity is to prevent invasion by alien species.Once established, alien species are generally im-possible to eliminate completely. Early detectionof smallmouth bass invaders has made possibletheir eradication, as in Lakes Chuzenji and Mo-tosu, where removal of individual fish was suc-cessful (Hideki Ohama, Yamanashi PrefecturalFisheries Technology Center, personal communi-cation). Broad monitoring is needed to provide theearly detections that make eradication of alien fish-es possible. The present predictions are quite use-ful in this regard because they provide a strongbasis on which to focus monitoring efforts.

Acknowledgments

Funding for this study was provided by the U.S.National Science Foundation (DEB-9985737), theOffice of Naval Research (N0014-00-1-0887)through the Census of Marine Life/Ocean Biogeo-graphic Information System Program, and the U.S.Geological Survey Caribbean Science Center. TheBiodiversity Research Center, University of Kan-sas, provided the biodiversity informatics infra-structure that made this study possible.

References

Anderson, R. P., D. Lew, and A. T. Peterson. 2003. Eval-uating predictive models of species’ distributions:criteria for selecting best models. Ecological Mod-eling 162:211–232.

Azuma, M., and Y. Motomura. 1998. Feeding habits oflargemouth bass in a non-native environment: thecase of a small lake with bluegill in Japan. Envi-ronmental Biology of Fishes 52:379–389.

Bennett, D. H. 1998. So many predatory resident fishes:what needs to be done? Pages 197–202 in ODFW

Dow

nloa

ded

by [

Mon

ash

Uni

vers

ity L

ibra

ry]

at 1

8:09

06

Oct

ober

201

4

853PREDICTING BASS INVASIONS IN JAPAN

(Oregon Department of Fish and Wildlife) andNMFS (National Marine Fisheries Service). Man-agement implications of co-occurring native andintroduced fishes. ODFW and NMFS, Portland,Oregon.

Bennet, G. W. 1937. The growth of the large mouthedblack bass, Huro salmoides (Lacepede), in the wa-ters of Wisconsin. Copeia 1937:104–118.

BIODI (Biological Diversity). 2003. Available: http://biodi.sdsc.edu.

Bohn, T., and P.-A. Amundsen. 2001. The competitiveedge of an invading specialist. Ecology 82:2150–2163.

Carlander, K. D. 1977. Handbook of freshwater fisherybiology, volume 2. Iowa State University Press,Ames.

Carlton, J. T. 1996. Pattern, process, and prediction inmarine invasion ecology. Biological Conservation78:97–106.

Centor, R. M. 1991. Signal detectability: the use of ROCcurves and their analyses. Medical Decision Making11:102–106.

Dentler, J. L. 1993. Noah’s farce: the regulation andcontrol of exotic fish and wildlife. University ofPuget Sound Law Review 17:191–242.

Dill, W. A., and A. J. Cordone. 1997. History and statusof introduced fishes in California, 1871–1996. Cal-ifornia Department of Fish and Game, Fish Bulletin178.

Emig, J. W. 1966a. Largemouth bass. Pages 332–353in A. Calhoun, editor. Inland fisheries management.California Department of Fish and Game, Sacra-mento.

Emig, J. W. 1966b. Smallmouth bass. Pages 354–366in A. Calhoun, editor. Inland fisheries management.California Department of Fish and Game, Sacra-mento.

Enserink, M. 1999. Biological invaders sweep in. Sci-ence 285:1834–1836.

ESRI (Environmental Systems Research Institute, Inc.).1996. ArcAtlas. ESRI, Redlands, California.

Frankham, R. 1997. Do island populations have lessgenetic variation than mainland populations? He-redity 78:311–327.

Grinnell, J. 1917. Field tests of theories concerning dis-tributional control. American Naturalist 51:115–128.

Hanley, J. A., and V. J. McNeil. 1982. The meaning anduse of the area under a receiver operating charac-teristic (ROC) curve. Radiology 143:29–36.

Hayes, M. P., and M. R. Jennings. 1986. Decline of ranidfrog species in western North America: are bullfrogs(Rana catesbeiana) responsible? Journal of Herpe-tology 20:490–509.

Hosoya, K. 1997. [Introduction of exotic species andbiohazard.] Bulletin of the National Research In-stitute of Aquaculture Supplement 3:155–156. (InJapanese.)

Hubbs, C. L., and R. M. Bailey. 1938. The smallmouthbass. Cranbrook Institute of Science Bulletin 10:1–92.

Hubbs, C. L., and K. F. Lagler. 1964. Fishes of the Great

Lakes region, with a new preface. University ofMichigan Press, Ann Arbor.

Iguchi, K., T. Yodo, and N. Matsubara. 2001. [Spawninghabit of introduced smallmouth bass, Micropterusdolomieu.] Suisanzosyoku 49:157–160. (In Japa-nese with English summary.)

Japanese Ministry of the Environment. 1999. Red listof brackish and freshwater fishes in Japan. JapaneseMinistry of the Environment, Tokyo.

Jenkins, R. E., and N. M. Burkhead. 1994. Freshwaterfishes of Virginia. American Fisheries Society, Be-thesda, Maryland.

Kawanabe, H., and N. Mizuno, editors. 1989. [Fresh-water fishes of Japan.] YAMA-KEI Publishers, To-kyo. (In Japanese.)

Kitagawa, T., T. Okita, Y. Banno, S. Sugiyama, T. Oka-zaki, M. Yoshioka, and S. Kashiwagi. 2000. [Mi-tochondrial DNA of the Florida subspecies of large-mouth bass Micropterus salmoides floridanus de-tected in Ikehara Reservoir, Nara Prefecture, Japan.]Nippon Suisan Gakkaishi 66:805–811. (In Japanesewith English summary.)

Kobayakawa, M. 1989. Catfish. Pages 412–415 in H.Kawanabe and N. Mizuno, editors. [Freshwater fish-es of Japan.] YAMA-KEI Publishers, Tokyo. (InJapanese.)

LaBastille, A. 1974. Ecology and management of theAtitlan Grebe, Lake Atitlan, Guatemala. WildlifeMonographs 37:1–66.

Lee, D. S. 1980a. Micropterus dolomieu. Page 608 inD. S. Lee et al., editors. Atlas of North Americanfreshwater fishes. North Carolina State Museum ofNatural History, Raleigh.

Lee, D. S. 1980b. Micropterus salmoides. Page 605 inD. S. Lee et al., editors. Atlas of North Americanfreshwater fishes. North Carolina State Museum ofNatural History, Raleigh.

Maehata, M. 1992. [Food habits of largemouth bass inthe southern basin of Lake Biwa.] Annual Reportof Biwako Bunkakan 10:1–8. (In Japanese.)

Martin, N. V., and F. E. J. Fry. 1973. Lake Opeongo:the ecology of the fish community and of man’seffects on it. Great Lakes Fisheries CommissionTechnical Report 24:1–34.

McCaig, R. S., and J. W. Mullan. 1960. Growth of eightspecies of fishes in Quabbin Reservoir, Massachu-setts, in relation to age of reservoir and introductionof smelt. Transactions of the American FisheriesSociety 89:27–31.

McDowall, R. M. 1968. The proposed introduction ofthe largemouth black bass, Micropterus salmoides(Lacepede) into New Zealand. New Zealand Journalof Marine and Freshwater Research 2:149–161.

Miller, R. R., and E. P. Pister. 1971. Management of theOwens pupfish, Cyprinodon radiosus, in MonoCounty, California. Transactions of the AmericanFisheries Society 100:502–509.

Minckley, W. L. 1973. Fishes of Arizona. Sims PrintingCompany, Inc., Phoenix, Arizona.

Nakama, Y., editor. 1996. [Fishing places of bass andsnakehead fish in 1996.] Gakken, Tokyo. (In Jap-anese.)

Dow

nloa

ded

by [

Mon

ash

Uni

vers

ity L

ibra

ry]

at 1

8:09

06

Oct

ober

201

4

854 IGUCHI ET AL.

Okabe, M., editor. 2000. [Fishing places of bass andsnakehead fish in 2000.] Gakken, Tokyo. (In Jap-anese.)

Page, L. M., and B. M. Burr. 1991. A field guide tofreshwater fishes. Houghton Mifflin, Boston.

Peterson, A. T., and D. A. Vieglais. 2001. Predictingspecies invasions using ecological niche modeling.BioScience 51:363–371.

Robbins, W. H., and H. R. MacCrimmon. 1974. Theblack bass in America and overseas. Biomanage-ment and Research Enterprise, Sault Ste. Marie, On-tario.

Rosen, P. C., C. R. Schwalbe, D. A. Parizek, Jr., P. A.Holm, and C. H. Lowe. 1995. Introduced aquaticvertebrates in the Chiricahua region: effects on de-clining native ranid frogs. Biodiversity and man-agement of the Madrean Archipelago: the sky islandof the southwestern United States and northwesternMexico. U.S. Forest Service General Technical Re-port RM-GTR-264:251–261.

Scachetti-Pereira, R., 2001. Desktop GARP. Available:http://www.lifemapper.org/desktopgarp/. (October2002.)

Scott, W. B., and E. J. Crossman. 1973. Freshwater fish-es of Canada. Fisheries Research Board of CanadaBulletin 184:1–866.

Shinya, K., and S. Watanabe. 1990. [Feeding habit ofadult largemouth bass, Micropterus salmoides, in theUshiku-numa Lake in the central Honsyu.] Suis-anzosyoku 38:245–252. (In Japanese with Englishsummary.)

Smagula, C. M., and I. R. Adelman. 1983. Growth ina natural population of largemouth bass, Microp-terus salmoides Lacepede, as determined by phys-ical measurements and [14C]-glycine uptake byscales. Journal of Fish Biology 22:695–703.

Smith, F. D. M., R. M. May, R. Pellew, G. J. A. Marshall,and P. R. Brown. 1993. How much do we knowabout the current extinction rate? Trends in Ecologyand Evolution 8:375–378.

Stockwell, D. R. B. 1999. Genetic algorithms II. Pages123–144 in A. H. Fielding, editor. Machine learningmethods for ecological applications. Kluwer, Bos-ton.

Stockwell, D. R. B., and I. R. Noble. 1992. Inductionof sets of rules from animal distribution data: a ro-bust and informative method of analysis. Mathe-matics and Computers in Simulation 33:385–390.

Stockwell, D. R. B., and D. P. Peters. 1999. The GARPmodeling system: problems and solutions to auto-mated spatial prediction. International Journal ofGeographic Information Systems 13:143–158.

Tabeta, O. 1989. Eel. Pages 47–49 in H. Kawanabe and

N. Mizuno, editors. [Freshwater fishes of Japan.]YAMA-KEI Publishers, Tokyo. (In Japanese.)

Tabor, R. A., R. S. Shivley, and T. P. Poe. 1993. Pre-dation on juvenile salmonids by smallmouth bassand northern squawfish in the Columbia River nearRichland, Washington. North American Journal ofFisheries Management 13:831–838.

Takahashi, K., T. Onodera, and A. Kumagai. 2001. [Ap-pearance of largemouth bass and changes in speciescomposition of fish caught by set net at Izunumaand Uchnuma.] Miyagi Prefectural Reports of Fish-eries Science 1:111–118. (In Japanese.)

Tanaka, S. 1964. The food and growth of ‘‘hasu’’ Op-sariichthys unicirostris (T. & S.) in the Lake Biwa.Physiology and Ecology Japan 12:106–114.

Trautman, M. B. 1981. The fishes of Ohio. Ohio StateUniversity Press, Columbus.

University of Maryland Global Land Cover Facility.2003. Available: http://glcf.umiacs.umd.edu/index.shtml.

U.S. Fish and Wildlife Service. 1985. Recovery planfor the Pahranagat roundtail chub, Gila robusta jor-dani. U.S. Fish and Wildlife Service, Portland,Oregon.

U.S. Fish and Wildlife Service. 1994. White Riverspinedace, Lepidomeda albivallis, recovery plan.U.S. Fish and Wildlife Service, Portland, Oregon.

U.S. Geological Survey. 2003. Nonindigenous Fish dis-tributions information. Available: http://nas.er.usgs.gov/fishes/.

Vida, S. A. 1993. A computer program for nonpara-metric receiver operating characteristic analysis.Comparative Methods and Programs in Biomedi-cine 40:95–101.

Williamson, M. 1996. Biological invasions. Chapmanand Hall, London.

Yodo, T., and S. Kimura. 1996. Age and growth of thelargemouth bass Micropterus salmoides in LakesShorenji and Nishinoko, central Japan. FisheriesScience (Tokyo) 62:524–528.

Yodo, T., and S. Kimura. 1998. [Feeding habits of large-mouth bass Micropterus salmoides in Lakes Shorenjiand Nishinoko, central Japan.] Nippon Suisan Gak-kaishi 64:26–38. (In Japanese with English sum-mary.)

Yoshizawa, K., M. Takayanagi, S. Mogi, S. Kobayashi,K. Nobusawa, A. Sato, T. Ikeda, and M. Murata.1980. [Resource ecological studies on a warm waterpiscivore, IV. Feeding ecology of largemouth bassin Lake Haruna.] Bulletin of Gumma PrefecturalFisheries Experimental Station 29:32–45. (In Jap-anese.)

Zenrin. 1998. [A software package for the Atlas of Ja-pan.] Zenrin Co., Ltd., Tokyo. (In Japanese.)

Dow

nloa

ded

by [

Mon

ash

Uni

vers

ity L

ibra

ry]

at 1

8:09

06

Oct

ober

201

4