\\','stCnl North An",ric,u, 'i.,tmal"t 71(2i, © 201 I, l'P. IbJ·,17.~
IMPROVING I\'ATIOKAL-SCALE INVASION MAPS: TAMARISK IN THE \VESTERK UNITED STATES
Catherine'S, Inrnevich 1.,1, Paul Evanl!elista2. Thom..].> J. Stohkren I. and Ielfery \1ori,elte 1
ABSTRWJ:~'iew imasiol)s. hcttn lipId oata, ,H"lno\'"l ,p,ltia]'l11odd",g tc<'lm',!ll'" oftel) drih' the n",'d to l't"'lSLt prc"jous map' and modcb of lJw""i"l' spec,el. ~ul'h is the case with t}lC <It lea,! 10 .\pede, oj' l'wwri.l. which ,In'
ilwilcling- riparian ."i)~tcrns in the w .....,tnrn lJnih.'t.I State::. alld pxpanthng their range throughtull N'm·th :\tn('rit~,. rn 20(l().
we dc..:'\.'dollCd L' Nntion(tl'l~lnlads~ :\-1a.p hy u:-.ing a tompiJation ell PI'I.'Sl'llCC' LUlll (ll).;;('"(:(' lo{.'aUon~ Wlt)1 n.·nmtt'ly ,')cnSl,'d d"I<. and slati.,tlc<.1 !liodd,"g lechl);q1Jl·.,. SiliC'l' tl1t' pllh"catlOll of I"al work, Ol)r d"t.,I,;I><' of n'1I1wi:r d,stnhnl1011S "dS
gr<W'OH ~igllitkant}y,
F~in:.{ tlw upd,ttcd dataha.....p of specic.... ()(;I'll1Tenf,'e, III.'\\' [In.-dido)' \'<.l1"i<lbk~1 and the DM....dmlllH entropy (\laxCJlt)
l1lo<ld. l\.t' b<IVL' revisoo ollr potent.itt} 'UllIlflri.Y di.o:;;tribution map fOl the l"e~tcn\ United St~tt('s. Dist<tHt't'oto-'\vatel wa.l. tht' strongc,... t pn:{hctuT III the lIlodd {58 l~), VI·hik· Illcau tCl1lLX'ratnH:.' of the ".... an'l('$( (LlIiUll'l' \....w·,o the <;l'(:OU<.J hest pn'di(;{tlr (l~ ..I<Yci. \lod,'1 ",.hd.tion, "vcmgcd ii'om 25 modd iteration.', indit"tcd that our anal)',i, hao str(ln~ prcrl!di\"c p<Tfor, ",anec V\VC '" 0.!l3\ ,.nd lhat the t'Xlcnt of '1(/111(11';" d;,tnllllti(IIlS is Jlluch !;'l'catcr lhan prC'do\lsly thought. The .<o"th· """ten) United State, had thc ~rc'tcst ,uitabk hal"tat, and thi' re;llit o,ffcn·d fmm the 200G modd Our work hh;h· Hght." the utility nf H<:rath·(: ll\ocll·ling fell invash<.' ~pr~('ie~ hahJlttt lHnddm~ .,s new inl()rmtttion her-omes d\~Ii1.L1Jk.
Ht M ,-([·;'\:.,-A llwnHdo las nlll·v~l.'1 i)l\''cl'iionl~S, rllejiJre:-. uatos d'l' campo r t(·cnit..a~ llovedc}!'>.ls de model.H}O t.:spack~1
irnpul.'HAO la cu.:'tlluli7...K'itlll dl~ 10;, m:Jpa~ y dl' lo~ Illoddus ('"bit(,lll~~~ dt~ ('species ill\':.I.SOT<.Ut, Estc <.'~ el C;.lj,(l u(':11 nl<:'no" JO f·· ...ppdl· .. de Ihmarix, Ja.l. ('udle ... ('stan ll)\'ad((~nd() )r)~ "iistl'nKl~ \'lb(!rcfln~ en d oe:\le clf~ to:. EE.UU ~-' e~ll'ndh'll(lo ~Il
d.isml>uci<'J.n pOl' tod.L Nortearl\C1ic.l, 1~1l ~006> <.J("sarrolldmos lIn lU.lL}cllHlClon,i.! dlll t.a1\l~lri..'l(·u (Nal:ion.u T.lJIl(uisk "rap'. IItllizcmdo una l·onlpiJad':'n (k~ sitj(J~ (h: pn·~cnci.;t y ::tu.seol'i<t con llclto~ de '\cn~on.':, rCllloto:::. y t.i:enil'~t... de nH.)(h·I~~LMm
cstmlfstk:a. J)<:~de ta pul,lu:al'ion de t'~te hOHhoil,JO, fll.1C~tl'il bu"l\.' de d(lto~ ~ohrt' }i.' dl~lrjblld6n de 1f!lfW";of ha 1:loCt'H}U
e<Jlls"\cr,,blcml'llt<'. llhlizando I" haw de dahl' act1l"liz.;"la tle (ll't"{'neia de c"p"cic;, nUl""'" variable, pr,'dict'lnls .'" d moddo de
muxinlil l'ntrupf,1 (\La.xentJ. 11l<HlO:'\ morlificado lil1c:"·;tro m,lpa UC 1.1 di.,;tl'ihuci6n potencial de 'lrww:ri.r p.:1_l'll cl O('ste de los E£ Ill'. EI !,Il.'didl'" ,.,(,; hl{"ie ell <;i mndd, Ji", I" distane;a al a!!ua (.'is, 100. ~ I" ICl)lperatura promcr!io ol,1 trinlt's!".' nuts e<.ilido Jil(' d s('~und() l1H ..f%). L"i \-alidacl6n d{~ rJJl..)dl'Jo. L'(-llculac.la c:OlllO d prollHxlio de 2.5 itf;.-ndonc.__ del llH"xl(·!oo indica CJl1e nll"stro an:ilis;; I'll"" una alta clpacidad prcdictiva (ABC = 0,93). Y CjllC I" ,Jistrih.,c;6" de TlII1lOlr,. l'S mUdlO11\., "'xt,,,nsa dl' I" que .Ie pe,,,"],a. CJ snfOestc de los EI~.LJ u. [llVO I" mayor r'anlidad de h;,hit"t "dCCll'ldo 1""." I" "'I'cu", y cst<' ,."",It.do din..i6 d,,1 "".,t.Ido <I" 200(i, 1\11"'1.1" tral""io enfah/<l I" lItiI.d"d ,J,-I modr·hl(lo ltcrat"" para lllodchn eI )I;,bitat dc la~ {"')peek's iU\a.sOTa.; ti m\.~dld.... (jill' ~c dispollga de 011(.'\':.1 ~ni(Jnllad6n.
H.ipmian ecosystems throughout the south processes (Christensen 1962, Robinson W6.). western United States 11dve beeu invaded by II anis 196f)), It ,>,;as estimated that tam;lrisk li:nrwl'ix specit's, collecti"dy kno\\>11 HS tarna occupied appl'Oximately 4000 ha in the 1920s tisk or saltcedar, Introduced Ii'om EUrHsia in aIld had grown to more tllan 500,000 ha by the the early 1800, lo control erosion, create wind· 11lid-1960s (Hobimon 19(5). Initially; lm.n;lJisk breah, and act ns ornamentals, tamarisk was intt~,.,t<lt.ioll was primarily conhned to regions of acclaimed for its nbility to wit11stand drought. the southwestern llnited States (i.t',. Colorado, hent. and diverse soil conditions (Carletml H1l4, Arizona, ~ew l>.'1exieo, Utah. and 'lbas) bulnow Di'f<lmaso 1998). [JOW('Vt'l: by the mid-1900s, occurs throughout the ILorthern and northresource managers hJ.d ".... itnc·ssed lmnarbk's western Great Plains (Lesica and !\Jucs 2001. rernarkablt, ability to spread and modify ceo· Sexton et al. 2006. Kems d aI, 20(9) ;Uld, to a system proce"t.'s, As a result, the spedes have lesser e:<tE'nt. in several ca,tern stntes (BallIn had dramatic, and often negative. effects on 1~07, Pearce and Smith 2007, USDA N HCS native nora. wildlife Iwbitat, and hvdrologic 20(9). Althou~h tamarisk may be spreading at
Jr'orl {ollif'.~ ~l'll'I1l'{' (~(>nt(1.· l.~ ("'('Ih~'t(..il ~lll'"'l~ l)."')() C,.. nll\· -\'.'(-" KuAhJll! C, h:)I'1 <:\~ll!l" co ~n')1f.,-" ~ III ?'UIUI \! k,'\dllr('l' f·l'(.!I'~\ I.....l!:,rl.lhrr). {'q]vl ~d,> ~Iah' l rl1WT\1t\. h~n Coll\n_~ co ill.l52:)-~,lg9
~1':-ln"lJl J,U111'\ k·lll·(r/ll$l~.l:!("
164
2011] hIPROV[,\'C !t\VAS10\; Iv1.~ps 16.5
<l S]OWt'l' rate than in recent decade~, there is no evidence that the invasion has reaC'lK,d its potential extent. Outside ol'thc United SIales, tamarisk can now Iw founcl in Mexico (Scott el al. 2009), Canada (LinclgL'('n t't a!. 2010). and ~Ol1th AmeriC'l (]\atale et aJ. 2008), \vell heyond the latitndinallitnits of its native rangt' (Gaskin and 5cb<l<ll 2002, Caskin and Shafroth 200:5), Anthropo)l:enie influences Ol\ natural bydrologienl processes (Everitt )9S0, 19%), and the specics' abilitr to hybridize and adapt to new Cllvil'OllInCllls (\Vhikraft et a!. 2007) are conct'rnjn~ to reSOlll'ce managers: further invasions are likely to continue,
Tbere are a nmnher of statistical and geospatial models that l1ave been developed rpeent.J}' to predict species habitat and potential disttibutions (Elith et a!. 2006), The~t· modds are increasingly llsf:'d by landscilpe ecologists <inc! resource managers to map disllibutiollS and foreea~t LWW inva~ions of nOlma!lV{" species (Elith et n1. 2006. Evangelista pt >11. 20(8). Most models rely on known OCCUtTcncc locations (i.e .. presence points) to identify ecosystem chm'aclpristics that define the species' habitat parameters and predict potential distribution across a hndscnpe (Stockwell and Pdcrs 1999, Phillips el al. 2(06), Predictive models are of particular imporlance to resource managers because they identify the potential extpnt of infestations fmd highligh t habitats that may be vulnerable 10 new invasion (National Inv~ivc
Specips Council 2008). Model resulb can also oller insight into ecosyslC'llJ eharactelisties that are either conducivc or prohibitive for a par· ticular species, while proViding critical information Oll how an invader mil}' respond to new habitats (Evangelista et aL 2(08).
In 200f). \loriSt'tle et al. reported the first detailed national-seale model of suitable habitat for lamarisk in the United States. Since then. a sil!:nificant nnmbf:'l' of new tamarisk infestations havt" been reported: SOUl(' occurred in regions thilt had not previuu~J y bc<,n infesh.d (Kprns et al. 20(9), wllilp uthers were simply absent from the original data set. NC:'w predictor v<lliabk>s have al~o become flvailable in geospatial fonnats at national scalps. including ali expanded suite of climate data that may also lead to improvements iii model pcrformwlCe. In addition, new lllodeling tedmiqnes, such as maximum entropy modeling (\laXCllt), arc beller suited for pn'diding lamal1sk distributinns with the types of data that are accessihle
(Evangeli.stA et a!. 20G8), Maxent h a newer technique <lpplied to species distlibution modeling that is especially suited for inv(lding ~pf'cies because it compares presence locations to tllC' availilhle environment rather than t.l)'ing to distinguish betweelJ presence awl absence locations. Ahsenee locations can be problematic when modeling invasive species because 01' the dynamic state of invading species' distrihutions (e,g.. lug limes),
Specillc<t1ly. mallY Hew lIlodeling techniqnes <Ire designed to he fit with presence-only data and provide huilt-in featun's th,lt Pvahl£1.te model performanec and predietive contributions of the tcstc~d \·m1ilbles. \IOlisPltt' et at (lOOCi) used logistic regrpssion, \vhich requires presence and ahsence datn il)r model development. AlthulJgh a proVI'll statistiCitI techni(llle in ecological modclin~, log;islie regression may nol acctlJ'iltely represent absence data V..JlCll the rcwessioIl is used for in\'asiv(~ species. Obtaining tl1.le "absence" data .It a scale C'onllnonlv used for national-level models (e.g" 1 klll:2) is not simple for any species, much !pss {(Ir an invasive species that is continuing to e",panel its range. 1\·1odels that can be Ht with presence-only data are not cOllstrainpd by a,ss1.llncd absem'es and thus lllay be hettcr suited for predicting invasive species where the species may not occupy all suitable habitilt (J-linel et a1. 20(H, Brotons et al. 2004. 1\:111I1ar et al. 2009).
[n this papel: we revisit our National Tamalisk Map (Morisette et a1. 20(6) ll.sing revised data sets. nl'W predictor variables, and a n0W modeling teehniqlle. Our ohjectives were to (1) improve our predictions of hftbitilt suitability in light of new information for tamarisk at a nation-al scale and (2) compare the methods and rcsults (if the 2 modeling approaches.
\'11::'1'1101):>
Data AC<]lIisiti(ln
The tamarisk database used in thiS amJysis was founded on the s,nne occurrence information used iu the l'\ationul Tamarisk M<ip (rv!orisettc ct al. 2006), bnt it included additiomu data and data from a \vider geographic extent (Fig. I). Following the publication of MOlisl'tlt' et al. (2006), OUI' team was c:ontaded by a number of reSIHlrcp nrall<li!ers infonning us of lleW' infestations and large-scale gaps in the data lJsecll,e.g., North Dakota, South Dakota). Mau~ of these eontacts provided us with
166 \YESTEH1\ NURTII A\IEmc.-\/\ :-.I ~TU~ALlST [Volume 71
Post 2005 data
\
500 1,000 Kilometers I
o I
FI~ 1. POInt location' ",,:cl ill clnd()pin~ tht' hal,iu,t ",it"hility n",del. \\;,h dil~l lIsed in the '''h,riwttc ('I a!. (2006) tam,ni,k hahitat <lIilahiliiy map ".j hln<l ,nH) 11(.:" data in ~ray: Proj('clior" l;SA AI1'cI' E'I"al Art'a Conic.
nt'w OC'clIlTenCC inU>mlation to upd.ate our database. \Ve also incorporated new tamarisk data that ~,ere reported to the- r\ahonal Institulc of Invasive Spt'cit's Science (\JISS) \Veb~ites
(i.e., httpJ/www.lamarisklllap.org and http:// ",'\\'w.niiss.org; Grakun et a12007). \Ve further ~llppll'rn(mt(>d Ollf data set b~ contacting state weed c<xlt'dinators and other re~ear('hers monitoring tamarisk di~tributions (e.g., htlp:!lwwlv .11unariskCoalitioll.org) and through literature searches whert' actual occurrence coordinates wel'e l'eportt'd or wht'l'e GIS layen \\('re made available. Thesc data included pr('sence locations for tanHl.risk, hut other information. ~uch
as abundance, was gent'rally uU<lvaihbk:. The GIS data we collected included point,
line, and polygon formats. The point data did not require any manipulation beyond modifying geographic projection; howevcr, the line and polygon data required processing to generate.\' and !f (:()ordinales. Line data wel'e converted 10 l-km:! ~l"ids comparable to 0111' predictor lavers. Ttl extract presence points for our analysb, we randomly seleeted 10% of the grid cens that represenled linear t<'imarlsk fea
tures. l~v .selecting only 1o'if of the grid cells. W(' were able to capture the preWDCt' of the species in the related location but not artificially i.ncre~L~e the sample size or crt'ate a hi!1;hly clustered group of presencc points. Polygon data had a relatively small scale, comp.tred to the spatial re.,;olution of predictor laytol"s. Presence points wert' generated by cRkulatitl.!1; the centroid of each polygon leature and adding the cl)ordilHltes to our tamarisk databa.se. In tlltal, wt' compiled ;2.') dispm'att' data sets (including those frow 2006: Appendix), ~"hjch
contained ::U5,56:) \'ecof(l\ of tamarisk occurrence. AJI occurrence (hta were converted to a GIS ra,tel' format with a pixel size of I krn2.
Whl'll more tluHl one preSt'nce point fell within a singh' pixel, that pixel repres('llted ,I single OCCIllTence {llr the pHl'puses of our analysis. This proces, reduced our data set to 11,601 occurrence p<linh.
\Ve as~embled 29 climatic, topographic. and geographic layers to lIse as prt'clictor, in the model {Table Ji. Climatic data consisted of 19 biodimatie variables c1es(:Iibing the alJnual and seasonal variation jn temperature and
20111 167
I'r('d ido,. "(lriahl" SOUf{'t'
Ml'an knlpemtun.' oftht'" \'r''HlTIiust I..jU1.lrtt'1'
Pr('~ipiti-lti(m of the ,":uttest month Size "I' pn'd[l11:Ltioll ('ven I Ceo!obn,.
'Ihllpel~~ltnl,(' l\{"'asonality Pn;clplt<.ltllln of 1,1)(' 'wanllC'sl qllm·tf~]
rvle'lIl t('!llp{T... tUH: of the w('tl,,'sl qllarler
Prc<:lpit<ltion .~t:aS()lldllty
'ICinpcn.tul"(· IUn}",,' n:lO~(~
£~k'alo...n
Ilad",li'lli ~~\lJlridit\
PTedpit"tior, oftl", ('oldest qll,"'tc~'
Ltlng" ITl FYI PT(:<:iplbtirm nftlh' dru.\,t month (::Jothl:nlMI,t'i :\ll'<ln kmpcr,ltlln' of Ihe Ldd",t 'Luart"r f."n'(]lIt'lH:\: of rl'(~..:ipitation e\'ents
~kan !empera!un' oflile' d"e,t (jllltr!'T ~I()p(' (d(~gn'cs)
:Vfean of EVI :"dl'an diurnal nmg('
4-U (42.)--4;5.-!1
2J.--I (2D 1-2~.7)
j a'5 \-8.:."1 4.1 (3.1-501) .) f; ,3.\-1.1'1
;) " (2 7·,1.11) :j (I nS·,1.(j) 2.7 (2.3,104)
21 (l ."~-1.D)
2 f (J.7 ··2.7'; 1.1 \0.99·· 1.5)
1.2 (D.S-- 1.5) 0.78 '~O.5-LO}
0.7iO..5-·1.1'1 0--1 (0 2-10)
D.~ iO 2-1.3'1 0.3 (0.2-0.5)
D3 WI-0.51 0.3 (O.l---D.5) 0.2 IO.l-H.:'i) 0.2 (0.1-0.4)
{} I :,0.1-0.4) a I (0.1 -11.2i
J)erivec1lrolll N,.II"n,,1 I\tliL' Hj,:t'rs dlld Strl'am.s
I)erived frOIll DA'r ,,1[-:"1' Ikri""d from DAn I!C.)'
DAY\1E:f USGS prlldHd (httpj/p"!,,
Wig~ gov/dcb./dth.ll,:) I)"rived frolll 1).\) \1 E'i" IknvC'lL li(nn l).\Y~-1I:T
1)1·";w'd from D·\\" ~ n:T Deri,,('d li'om D,\n·l ET I)('I';V('(/ (rom 1).\) \IET
.""lion,,1 EIe,"tion 1);11.".,el. (Ill I'll .//1 }Nl.lbgs. gO\' /~
Dt\Y"'1ET I)A\'\1£T
I)('ri"ec\ from DAH·I Cl" DL'riv{'(/ frOIll \IODIS
r)co'''''c,l from D.... n·1 [1" l.),·riv"d from DA'rM ET
l)(TI""d from DAY"'·IIC'l' DAY\I£T Derived from DAHI 1:.:"1' Nati(Jn ..~1 EIl'Vilth1n J)~lt:L~('t
ihltp:l/ncd,u,'lgs l!O\/)
Ikrl\ed from \10DIS I)"";""d frolll D....Y1,,1 L·:'i"
IlITcipitaUoJ) (Hijuwns 2006), including 8 DAY!vIET vlU'iables (i.e., number of ll"ost and growing degree day" humidity. radiation. minimum and maximum tpmperature, and precipitation ~ize, fnequeney. and annual '\\'erage; Thnmloll et aL Hl97). Tbpographie (lat" indmled elevatioIl. slope, and a~pt:'ct. \Ve abo used a :3year mean and range of the Enhanced Vcgda(jon Index (EVI) lhat was delivcd n'OIn the Moderate Hesolutioll SpectroraJiolTIeter (MODIS), geoID!.,,)' unimarily p,m~nt materi<J), and EucliJcan distancc-to-waler gClleratcd ll.~ing toobets available in ArcGIS software (ESfil 2006). \Ve examined <111 variable~ for cross-correlations in Systat 12 software (SYSTAT software, San Jose. CA), and removed ~I1lY highly cOlTcIakd variables (I" > +O.!'\ or r < -0.8), leaving 2.3 predictor variables for our Jnt)(lel. Although the modeling technique we deso<ilw beJo\\' is nol sensitive to issues of JIlult)('ollinearity relat('d to variable corre]atiom, indmion of highly correlated ",uiabJes can nffecl vmiable contribution ,U!t! interpretation of results. \Ye were interest('J in ('xamining thes<' dYeet,.
To validate Ule model we compiled 2 Jif fercnt (lata sets. We gathered the absencc data used in the 2006 paper as om' validation data set a.nd obtained an independent data sl'l estimating aer<:-agp of UlllHtrisk by quarter qnadrangle (one quarter of it 7.5-minute llSGS quadrangle) in the wc"stem United Stc1tes. These data were compiled by the \Ve,tcrn \Vcc:d Coordinating: ComrniUc:e: county wecJ coordinators reported an estimateJ acreage for each quarter quadl'anglc ill their jurisdiction liJr 16 "vestprtJ states (i.e., our study ~ll'pa without l:e'(Hs). While these data are based on expprt knowledge (rather thm, actual field data) and arc likely to have ebtu gaps. the information is of value' for testing needs.
Spatial Modeling ilnd Statistical AnalYSiS
For our anal~!sis, \V(' used the ro..'la.\Eml moclel version ;3.2.1 (http://w\Vw.cs.princeton.edu! -schapire/maxt'nt) (Phillips et al. 2(06), which is iJ)ercasillgly being lIH'd for predicting spedes distributions and has perlill1ned well based on evaluation metrics slIeh ilS ll)()se vve describe below when appliN] to predicting anJ
WESTER" N(JHTll A\IEHH..'\I-- ?\ HliRALlST [Volume 71
mapping taJl1ari,~k (Evangdist,1 et a!. 20nR. 2009). Using the principle of maximum entropy. the' tvlaxent TIlodei identilks l,elaliol)ships betw~'en spedes OCClllTt'UCC' <tild the pr~'dictor \'ariahles in order to rlevelop a habitat suitability map. \'Iaxent requires presence-onlv data and extracts random hackground p[)illt~
from the training extent of the model (Phillips et a1. 2(06). Because Maxont is sensitive to s~ullpling bias. background point sell:'c[iol1 was limited to counties for which we had smnples, and the reSlllting moue! was projected to the eul"ire westt'l'll United ~lates.
'Vc conduded 2~ separate model iterations, withholding ;J480 presence points (30% of 0111'
presence points) randomly selected prior to each analysis. Onr analysis prol!u(;cd 3 maps, a habitM suitabilitv map for the westem United States that aVt'ragecl tll" 25 model iterations, a map of standard deviation between the 25 iterations. and an average elaIllping map acTOSS
the 25 iterations. The damping lllap highlights locations ill the model projection (the entire western U.S.) thilt h'lve environmental conditions outSide th(:" range found within the training region kounties with dat<l points), 'I) prpdictions at illeSt' location;; may ha\e higher uncertainty 'l.~sociated with tItem.
Our model results wen' validated using several dilJ'erent staUstical tests. First, !\Iaxcnt calculates the area under the receiver operating characteristic curve (AUq, which is a llH:'asure of how well thl' modd discrirninates between pre~cnl'e ]oC'atiom and background locations (Fielding and Bell 1997). \Im,ent also calculates the JO percentile training presence threshold that can be used to devC'lop a binary map of suitable and unsuitable h<thitHt. Additionally, we calculated the AUC the correct classification rate, sensitiVity. specjficit\'. and kappa (Fielding and Bell 1997) using R slatistieal software (v 2.8.1; httpJicran.r-projecl.org), These tests ,",'ere conducted using tll<-' absence data froll) !vlorisettt' et al. (2006) and the <]uartt'l' <julldmug1e data sci. For each quartel' (]Il<ldranglr, we t()ok the maximum suitability ...·alue /i'om any pixel ...\ithin the quarter quadrangle 10 calc lllat t' an AUe. Quadrangles with any l'epOlled acn'agc were dassified as pre.~enee.
and all others were classified as ahsenl'e. Morisette et al. (200m rcported the area of
'Hitable b<lbittt for t~ull<trisk in each of the 4& continental states and the Distriel of Columbia. \Ve recalculated that dala for onl~,' those
we~tern states included in this analysis and also cakulaled new ...'alues {i'om th~ model rt'sl1lts of thiS ,study. As prt"viouslv dcscribed, we trained tIlt' modcl by limiting tllt' extent tll locations where we had fidd data, \VI,en this model ",as projected onto the entire United States, almost all of tJw eastern llnited States had verv high damping vallies (dat.a not shO\vn). Tbus, we limited our analyscs to the western United States. Thi.~ analysis \vas conducted by examining the continuum from unsuitable to suitable hahitat and sl:'leeted areas in t.he 99th (lUantile and the 90th qllantil~' of suitability (referred to as high and In()uerate suitability respectively) <lnd totaling tlle Humber of 1-km2
pixels in each of these categ'oties bv state.
RE6liLTS
The 1\·1axcn t models and validation tests sbowed strong peJt{)nnanccs in predicting suitable hahitat for tamarisk. The avcnlge AUC value calculated by JVlaxent for the 25 model it:el'atio!lS was 0.930 for the training data and 0.$)26 for the test data. The di,tanL'e-to-water variable had the most predictive contribution in all models, averaging 58.1 % (Tthlc 1). I !abi&'lt suitahility increased with decreasing distance to water. Mean temperature of the \,armest qUiU'ler averaged 18..,l.%, and precipitation of the wettcst month meraged :3.8(';f of the predictive contribution to model results. The relationship with the \vanl1l'st quarter is a lo~istic
curvc where suitability is low at cooler temperatures, increases qUickly at in tennediate temp(;rahlrt'S, and is greatest at high temper,ltures. Habitat suitahility w,lS greatest, with relativel" lowel' predpitatioll values in tbe wettest months. An other variables contributcd <3':'(, to th~· model.
Independent \'aliuation tests, ming pl'esenc(' data (.'30% of prcsence poin(s com hilled with (llJarter quadrangles tklta ~et) and absence data (from our 2006 study) aho demonstrated stron~
mouel per!l:lnnances. The Ave usiltg all presence 10catiOlls and abscnee data from the 2006 c1<tla set was 0.965. Otber metTies induded sensili\'ity (0.938), specificity (0.936), correct classification rate (0.!:J37), and kappa (0.872). Using; the <[lJ<lrtt'r (juadrang1e data, metrics included all Aue or 0,735, sensitivity or 0.780, spedikity of 0,6HO, correct classification rat{, of 0.700, and kappa vallie of O.(j35. '111cse vallle.~
an", lower, but the semitivily value, "hielt \\'as
2011] 169
a)
b)
c)
fig 2. \o1axcnt rnodr-l n~",u\l, nCI'o~" 2~ rfUldc'] iteration.. wlthhulding<l dil1('n'lIt mmloOi 3<Wr 01 the data III ~;",h
h''>t nm: (..I) .1VCf\=\!!(' hc\IHtat sUH.uhIHtv, {b) standard d(;\,l..l
tlOO LHllong the ln~x)d itt'I'atjon~. and .(c) '\\'l'n.\~C (;h.\mp~ng (d,·)~,.(;,· 0/ d<:paltnn' from the cnv;mllllll,·,t <'aphm:d h) (hl' Irninin!( lo(·"tion<) I'",jed;ull. I"SA A!ll<'rs Eqll,,1 A"·,, Conl(
o 250 500 Kilomete rs 1 I
"'g 3. Biliary map of ~\\j(ahk <llId unslI;!a),!<- h:th'l:lt I'm t.:lJl\.i:lrisk U:'lIl1:! th(~ 10 pen:eoHle tnllnjn~! J,W(;Sl'1)(,(.: threshold (0.,,2;. Plllj,'<:tion: USA AlIl<:l's E'1<ml Area Conic.
still relatively good. shuulc..l be gi\·eu lIlore \"eight due to the nature of the quarter (Iuadrnnglc oata set. \Ve t'cln be relatively eerluill lhat quarter <luaorangles with reportt:cl acreage arc aCCU1'[lte as (ar a~ presence is eOlJcernecL hut we cannot say the same ror absel\ce locations. as it would be difficult to senTeh overv quarter quadrangle e;-.hanstiveJ\' far a simd~ tamarisk s(;C'oling or plant.
SuitHble habitat noticeably aligns with water bodies (Fig. 2a), as expected hy" the high contribution of the distance-to-\vater variable. Disthlcl patche~ OCCllt' in the soutbwestern United Stntes, particutarly Utah and Arizon<l. A high standarc..l deviation among model runs notaf;ly occur.' along th(~ Pacit}(' ~re.,>t, spots of !vlontana and Idaho, and along the eastern (-'dge of our study area \Fig. 2h). \iinirnal damping o('clJrred (Fig. 2c) and was in locations 111imarily distinct from the areas wi th high sllitable hal!itat, suggesting thal predictions are Ilot lowered b) lHicertainty related to instability (i.e., hi~h ;;('anoard tkviation) or
l'.\LrapoL{tion to l~ovcl envirtlnrncnts (i.e., clamping). Classil)'ing our suitability map into the binary catl'~ories of suitable and unsuit able halitat hj~h1ights associations between tam:uisk arlO the water boclies llotl·d with Figure 2a (Fig. 3).
170 \V£<;lTR1\ NORJ1I .;\.~lERlc.\.r- N"TlIR,\L!ST [Volume 71
0 20000 40000 60000 80000 100000 120000
titan ,:. 45628
Arizona '0' ~
54550lexas
','::'
39635 Colorado
,8617
35817 Caliromla
Nevada
New MexICO . , .. ,. 86354
Oklahoma
Kansas
Soulh DaI<a1a
Washll1glo.,
Montana
ldar.o
Nebraska
Wyom,ng
Oregon
North Dakola J El2009 moderate area 0 2006 moderate area
fig <I'L Prl'dictl'd .m:a (km 2) or ,,,ita],l,, I,abit"l h~ stat" Il)T Ollr cum'nl 1I1l),1<-1 anti t<:>1' till' \Inri'dtc l't •.J. (20(0) n,odd lUI' ll)'x]''''atc h.,[,it;'t \\Iit"l.>ility.
Utah ranked higJll'\t b~ area and by percent model, where CalHClmia, \Vashingtoll, Texas. of area hoth Jar rnoclemtely and hig;blv suitahle and Ari7,()lHt eaeh mnked highest in one of the habitats (fig. 4. ;\ppendi~). Arizona and Colo 4 cate.gories, rado exchanged the second and the third rankjng~ am()n~ the 4 categories. Kansas ranked D rsel' ssw\' last Ic)r highly sLJitablt, habitat, and North Dakota and Oregon were ranked the least moder Existing models for tamarisk clisttihntions ately suitable habitat by area and percent or in the LIJited States are few and arc ~cncmU~
~rea. respectively. Overall, patterns 0(' OUI' constructed at small scales for local mana~emodel result\ wer'c similar to the distribution lIlenl of infest.\liOll'. EV;ln(~elista et al. (2008) of fidd data <mel patterm exhibited in Figures Jllodeled tamarisk dhtributiOll for the Crand 2 and 3. These results diHer frum the 2006 Staircase-Escalante NaliolJal :\:lollunlt'nt in
2011] IMPRoY[r\G jl\V;\.SI0' \hps 17J
0 500 1000 1500 2000 2500 3000 3500 4000 4500
2376
2259
ZO$
Idaho
Nebraska
Wash>nglon
New Mexico
NeYada
Oregon
426$ Utah
2823 Colorado
Arizona
Call(ornia
1729 Texas
1868
1293 North Dakota
124
639 Oklahoma ..;.
577.'.
777 Montana
,1416
!
i
I ........... ", .. , J
lIll2009 high area 02006 high area'
Fig ·~b. }'rl'didccl .tr(~\ (kill:?) of;;uitahk hilhitat lw sti.lt.e {c)t" ollr <.:nrr(.'nf H1odell\lld (or tht~ .\lwisuttc l't at. (2<XXi) model l<)r hiJ!h Id,it"t 'lIil,,"i!;ly.
Utah, primarily using a suitt> of topographic variables to test Illultiplp modeling tecbni(l!ll's. Mo~t of the moclds thpy tpsted could rl:asnnnhl~.' predicl ""hpre tamarisk would oe('m~ ""ith ov!:"rland di,tancp-to-water bein(l; U]{, most important pre-dieto!'. Wbile tilt' spatial extent of their ~tudv area was much smaJler (i,e .. apprm:ilJ1atell' 2 million acres), their study corrohoratt'~ t!;r importallce of water as ,1 predictor of suitable tamarisk habitat at all scales.
Kerns el al. (2009) modded tamarisk in tlw nOlthwestprn United Shltes. but ,tt thi,~ ~cale.
distancp-to-water was only marginally import,tnt ilnd was olltweighed hy climatic factors. t\'fa.:l:imul1J teml1er,ltnre was the most important fae-tor, and was corrdated with our second most important faetol~ mean tempcr:lture of the w,lrn1t'~t quarter. Friedman el aL (200-5') found a slrong relationship lwtweell mean annual minirnUI11 temperatuH> and tamarisk presence in the western UIliled Stales, They did nut
172 \VESTFR, "UHTII A~'IERICAf\ NATI1RALISl [Volume 7J
examine the relationship between tamarisk presenc-'€ and mean aJlnu,ll maximum tempenlLure, but maximum tClHl1erature of thc \vannest month and minimum tcmpcratnn: of the coldest month wt're llighly currelatf"d (r = 0.79),
),-IOliselle el al. (2006) used remole seming data and presence-absence data to lTf"ate a National Tamarisk J\Iap for the lJnited States, Sitllilar to our results, their stud, l(JUlld that the sOlltl"v(-stern staks hall tIll' highest CO\}
centration of suitable habitat; hov..·ever, our new JlIudd shows llUllHrisk having a much greater pntenti,tl distrihutioTl aenJSS the entin·' \vestern United States. \Ve should note that despite tht> hick uf data for the \}orthwestern states, the 2006 model was able to predict occurrenccs in some arf"as, However, the nf"W model adds new areas tu tIlt' potential habitat and captures more of the currently known uistriblltiOlJ than the 2006 modd dill, Additional agTeement fill' sOlltbwe~tern states having the most tamarisk arises frum il. higher pet'~(:ntnge of sampled stream reaches in Arizona (5~V!+) than ill northf"rn states «5%) (Hingold pt aI, 2(08). Some uiHerPllces in the 2 nlodels can be attributeu to difl~'rentll1cthouologics. The 2006 paper was focused on lltili?.ing 1'<>motc-sensing (bta and did nut inclune uislancp-to-water as a predictor variahle. Our' model was dlivcn bv thLs variable. The models were also ll.m for ~lifJ(.~rent .spatial extents, The lOO(, nlodel inclllueu loc,\tions in the eastern United States, where tamarisk is lIot considen'd illvaslve, hut the updated model focust>d on regions of known inV'ksion. These diJfen'\Ices, along with the additiollal data points, (:()nlriIHlted to (1)(: diHerences bdwceu thl' models. As this paper was not meant to be a modd comjl,\riSOll paper hut rHther an update to lhe current knowlnlge on tIll' distribution of tamarisk, we h<\vf' left an in-c!ppth comparison bf"{\H'en ttl£:' :2 studies for future work.
Ca\'eats
There are s('veral cav(~ab assochltl:-U with our modeling that need recognition. First. the field data used to fit our models werl' compiled from disparate data sds of different cullection designs and methous, induding opportunistic sampl ing. As a result, the data are t::.pected to be bias/"d to Varyillg degrees. Like other species distribution modeling technique.s, Maxent is seusiti\'c to thcsc biast's (Phillips et aJ. 200l:Jj.
and they likely influt"nced our mo(lel results, \Ve beli/"ve that at least some of these prol>lellls were alleviated by lillliting the area [rom which hackground points could he drawn to the are,l.' with field data samples (i,e" connties). but furtlwr investigation o[ tIl(' effects lU'e warranteu, fidd uata used [or training and t<'Sting \t·('r("· not indqwudent (i,e" thev"were drawn from tlw salllP clata S/"t) bllt w/"rp likely spatially autocorrelated (j.e., the uata cuvaried g(>ographicaIly), which has been demonstratcd to inflate '\l.iC values (Segurado et aI. ~006).
1~1StJy~ we suspect that the coarse resolu tion or the points extmcted from qw.rtcr quadrangle data ,sets may cause inflated evaluation metrics Uu(' to this miSlnatch in rl'solution of the predictor variables (1-km2 cells). Finer-resolution OC<:llfl'{'nl'(> datu paired with finer-resolution preuiclor variables would improv{' tIl(' mOflel.
Maxent is a eun'datiw moue\. \Ve can use the variabll' importance measure to uevelop hypotheses abuut what \,ve think ma~' bf" uri\'ing tamarisk dishibution, but w(' l'NIIlllt uelermine C,H1S,llit\ with these mcthods.
This uwdel desc\ihe, potential suitahle hahitat for Tawarisk, HowP\/t'r, moving toward <l
lIlodel of abuIluarwe wllulu be more useful. I:.:vangelista et al. (z007) l:t'eated a IrJcHkJ for tamarisk biomas~ baseo on estimates of height ami eover in southwestern Colorado, hut thiS has not yet heen extended to a greater geographical art'a, Creating a moelel of presence locations of tamarisk can bc som('what misleading in that all areas of suitable habitat lllay not he locations where the tmnarisk has iI high impact, Obtaining these types of information across large sp,ltial (''{ten t.s Illny be uifficult, as the data are geueral!> not i1Hlibble; however. the inronn<ltion wtluld help lnanai!;er, prioritize locations for management activities. Adllitional!y, thi:; mode! was developed f(u' the genus 7iJwarix mther than indiddual speeif"s or hybrids, because most avaihlb]p location data is not accurate to the species levt'!, Creating species-specific models may lend greater imight into the factors <lssociated with tamarisk invasion,
Neither lhe 2006 model nor tIl{' current lllOdcl completely captured the lldl (',tPllt of tamarisk invasion in the Unitt'd ~tates. \lodeJ.s are hypothe~e~ that (:,lll he useful for guiuing management activities, but tlll'y call never perfectly represent truth. Howpvtc'r, we helic\'e the current model proviues a more accurate
173 lOll]
representation of tamarisk (',tent lwcausc it incorpofiltes improved distlihlltioJl dM<\.. a different suile of predictor variables, and a diller('nt modeling technique,
Conclusions
The tlitlereltces between the 2006 model of tamarisk habitat suitahility and our curr('nt model clearly highlight the utility of'iterative JIlodding 11)1' mappilJg invasive species distributions, as ,uggested bv Stohlgrcll mId Schna;.t' (2006), Hedsiting distrihution models of'inva.,ive species is of particular importance to reSOllrce managers \v ho require updated, accurate information to address lOanagement conCCI11S and kJrlTlulate strategic planuing, 'Vith this example, the initial map (Morisplte et al. 200()) intlmned tll0 wider commuH.itv of such modeling dTurts ,wd led oIlier> to'contribute new OCC'lllTence infonn,ltion to our database, Fbrthermort" in the time hct\\·een this study and the 2006 study, there have bet'n advancenwnts in mode!ilJ~ algorithms; and newly available predictor layers, updated ()cC\lITl'nee (bta. and improved nwdeling: l'xperlhe have becoml' available vVe exped these trends to wnti Ilue, We belie\-"(' mOst habitat suitabilHy maps could be improved by tIH'sl' adV<1l1eeS, which l1<lvP strengllwned our ability to predic:t suitable habitat for tamarisk and forecast new invasions. Finallv, like the 200G tamarisk model, our new mo~lel for the western United States can also guide fntnrt' data collection dIe)rts (to areas with hig:h uncertainty' ur to high probability areas not yet searched), serve as an early warning of tanl,u'i,k's pOI'ential spread into Ill"" area." and pnwide an estimate of current tamarisk infestation and its impacts to ecosystem processes,
'Ve thank entities and persons with tamarisk data sets who were willing to shart' their data for this work (Appenchx). We thank Amy Randell. Sunil Kumar. and ;\Iyda Crall for reviewing the manuscript. Logisti(' support w,tS provided by the ll~GS Fort Collins Science Center and the Natural RC80urce Ecology LahoratOl;" at Colorado State Uni· ~'ersitv. Any use or trade, product, or firm names is for descriptive pn1l)oses only and does not imply t>nt!orsement by the United States ~overlllnent.
LrTEJ\ATl'HE CITED
Ihl \1. II H. 19fi7. lnlrotlucl'u ,mel natural,,(·,l 'amari.\ks in Ih" L:nit"d Stat"s anu Comad.1 ['h,mari".lcc.\{·]. fl,.';]'"a 15·J9·-25.
Hu, H.F, ;'1) .0\ ,-\. S'II.·It. 200S. SUl":c;, of ddi\"l' rC"C~l'tatioll .lftvr "/riJ/wri, j"t:.'ltlonli in rip<:ll"i~i.n e{;(I:'iY~k~ll:-' or fliP sn!lthwe ... tf·rn United St~ltc:,: ..l Ililanl:itatj\.~,
as~e~~nlL'nt or pa~t re~t()nlti(>li pro,i<.'cb Hestnri.Jlw,.. F"ola!(\' 16: lJ 1-·121i.
1!f:U'D'ls, L.., \\' 1'111 Il.IU', \-1.1l. AH\IJO, ,~I)A H. J-lllrn,L 2004. Prl·"ieT1(.:e-ab~cnl.."e \·er~\l~ prc~('lll"(:-oT11y mouellio~ 1"'1<'lhod" for [Jrcdid.i/l~ 1Iinl h"hi~JI 5\'ital ..litl E..o~r"pll\ 27:,-137-4,18
C-\HI.I·:TO'. ~·I A. 191·4. Auaptation of the talil"ri,k for d,y land,. SC'('JK" ·19f;82....(i!H.
ClIIU'TI",E', [.\1. W(;2 "1111' rat('ofnat,".dilalion Of'nIHLIJ
rix in l ",ll American \lidl"nd Natllralht h8:51"';;7. CO: 011·\1)0 DCp\l;rIlLY" ,w TI\.""I-;U-:·L,I'!O'. 2002 Colo,
r.ldo Ikparhnl'nt of ll-allS[XlItahol' C<"i!""plllc Data. C"j'mldn J)q,,,rlmcnt of'Ir.""1Jortatil>n.
1),\"1;1", T 2006 A predict;\'(' mn<Td hlm"mk hahitat III
C(lliron~id dJH.l Co}onldo ;\:(ast<"t· ... lhui\\_ CoIOr~HI{)
St"te UOl\·<.nit\·, Vorl Collin, CO DITo\-l\v), .1.\1. 199b. Imp'R'!. hiolng,·, alld ,','olo(!y 01"
sa]tecx)"r !"lilllwri.k ,pp.l in t.h,' southwestcl1l V nited St.rtc.I. \\l,,·d 'I""lmo!o!--'V 12-:326-:3:36.
}o;, I'll, J., C. H Grl<" ~ II. H.P A' PUISO' \1. D, r),~, S FI"HKIi':K, A G\ J,A\, H..). H'J\'~\j" V H"."·IT\'~.\','H Lr·:AI"JJ\\lC~. A L':Il\·IA'f\. ;:1" .'>1" 2U06. !\o\'c! ·rnd:hlll.ls improve pn'di<.:hoJl of 'l-PC(')("s' distrihutioo.\ fi'ollJ OCC·'"Tl.·1\CC data. E(:ograph" 29.U9-l5J.
ll~SHlj E'I'II,)\vrr·,r.u Srsn,.>Os HF"':AfH:11 1\~"'Tnin.
:2006.•\rc\1'l» q.:2 L':SIU, lkdhlld.<, CA. E\'~"(oEI..I.\T~. P S K\ '''H, Tj. S IlJllLCr,F', -\.\1' CHAI J.•
,"" C.I· Nl""'\, ".2007 M"lding "hO\D1'C0llml bionla.~.~ nf TwJtlO7.r rmJUM'1SstHUl ill tn(' ,'\rli.lllsa.~ nJ'{~1
h",;n of .\Ol\t]w",I"TT1 Colo,·,\d". L SA. Wl'-,t"m North "'Olerit"n \"l"llnJist f;7:.51H-50!J.
I':\"'CEI ISH. P.' S. "l.'IAn, 'fJ. ~Tnl·II.C"J", C.S ),\re\l.",
\Ier,- ,\.W. ell.ILL,) B. NmJ\J." Ill. -\\0 n B.-,ll' \LT" 200"3 MorJl'lling inv;J,;oJl lor ,\ h,.hltal ;.(clln"lii<J( (Uld.1 ..pedilhst plant ~[)l:<.:jl.:'\. Di,,('ro,lty ~lntl Di~
trilH,twn5 J,l.iiOij..,';17. E";\(;r,.,J~n, PH .. T). STD'I!.(.III:', j.'f \-l<l':J..\t·:rl'l:, A,n
S ,,\ 'l~H. 2<XJ9 '.1,.ppmg inv",;v(' ta""mll: C"liJllkJli.r). a l:Ulllplut~m of ~in,!..dl'-,)""l:rw i.lnd tanH'~:'j'rLe.~ imJI:~'
ses of !"elHotch sensed ela('a. fklllotc Sensing of Ell, i~
rnl1\1}('llt l:5 (9--513. En·:oun. B,L. WHO. E('ol,)gy of saltenJ"r--i' plea i(,r
res(-'areh I~~n\'ironnwl]td Gcolog} I: Bl.'r1in I .J:7i'·""L ___. 1995 Chronology of thl.· ,pH',,,l of t"man,k 11\ tlK'
c<'Ilh'al Hi" G,',u'u,·. \\d],md5 l"·f~-);;....rj(j1i
r,l'1 Ol.\e. A.H . ·"D IT Bu.!.. H197 A rt\;p\\, n! r"l'lh'l<ls t<lT the ds~c~sm~'nt of pn.'i.hetion L'rror~ Il) <:on....<:n il~ tlt)n prc":l'·IH..'C/dhs(·nce model." E nVinmmt'nt~tl Con".",,,ti,,n 2·j·1'.1..·49.
FErEU''''', ,I ..\J, C T AnHY, PH. Sn"'KOTII, \1.1.. SlX"'J,
\I.F \fl·.,ueLlA"} \"1.1) l',uxnu,c, .\.,r; E K. Git'I-"r1'-.l.2:000_ J)ominU1H'"(' of non-lli.Hive riP'::lrl;"l1"l tn:t'\
in ,vesteDl US;\._ Hio!o!.!l<:al hwa:;lOns 774';-751(;.\S)(I,\, j.E, -\M' ll.\ Sell 1\1. ZO(!;2 Hl brirJ -j;WI<lriX
witk'spr('ad in U.·t.;. invi1~i(}1) i.md undd('ckd 1n native Asi;'\11 fungo, ProL"()('din:;!:> ()[ tlH' j'Lltiol1<\) ;,\{;,Idt'my
of Sd"ll<"(;' 01 til<' Untll',l :'t"u,., "I" :\nwrk,a 99: 11256... 112S9.
174 \\'Esn:R\ '\OIlTll AMERICAI\ NATUHALIST [VO!l1lllO 71
C,ls~r~. J E. -I'd) I'll, SII,\fHOTlI, 2005. HybJi<!i,oltiOll of 1(I1n.o.,-ix rlJlIlosiSfJ'i,tUl. ~I.nd r. ('hin~'i.~;.S (saltl:(~<.Jar~)
\I ilh T (ll'h'lIl(l !uthl'i) n:lI"ane<\<:I·.:iI·i in II,e ""'Ih, "",,,Iern liSA udel1ninetl Irom DNA "'qlll'IK'(' tlata. Vfad rono .;2:1-10
GBAJIHI. J., G. NEII'\l,<,\, C .I11(,\1·:I)ClI. H SWIll), l~fJ
T I STOI-ILcr,,:, 200;. A gloh"l OI'g'III"m ddcdl())) and H1(}nit()rin~ .'iy~t<:m r()~ nnn-Iltltrv<: "'[Ic("il..'~ J-:l:()~ lOgJcallntill1natic\ 2:1;7-1.~.3.
IhHKl.>. D.l{ 191.16 Recc'nl plant in\'asion.1 in tl", "'id a",1 «'mi-,ui" .,o"II""c,t or Ih,· Unit,·cJ ~Iat"s Annal, of Ih<: Assodation of American G(X)j!r'lphcl'\ 56:'lO8-122.
1-IIJ\t-I1'<', It J. ZOO!';. \'lkIlC\'al'\ !\\1 L. V,,(',ion 2.3. A"<lllahle (rom: hllp.l/worlddi,n.0l·f(,1,iodilll.htm
H I liZ': I " A.H., V 111':1 i I·.H, \~I) r \11' !'HII,. ilOOI. A",·,.'in~
h"hjtat-slli(a"ilit~ Illock-II '\ith iI virlual 'pedl". E<:o~ logil'al \Imldlio;! 145.111-12].
};.J:H,\s, Il.K .. H.j. N,lYI.t>Ii. \1. Ill;()'\(ll'.\'\!.. e.G. I'.ln~,
~"D B. [{O<:L:ns. 2009. \)oflcling tanl<lmk (1'lIfV;rix Spp.i hah'!'aL and dim"t" d'<I1l>(C dll'ds in the nOlil,"""Inn U"ikd 51"tn. 111\;lsi\,[, Plant ~('icnc<: ,<rid \lan"~"lIlenl 2.200-·21.5.
KL \I.\K. S., S.A. SI'U 1.111'<:, 'l:J )TIlIIICKl',\. " HrHM~'\'\, T S[ II '" 1l'I, ..1'\1) L 1l.'"I' ~009. POLl'"I;al h.•hit«t ui.stribllt.on for th" li'Cshwatt-r di"tolll [>idym'J.il,hcnil' ~l}tJI:ini1ll.' jn t~w c.:ontilH.'llt.Jl U~ F'l'onti['\~ III Et~)101.'" alld the Env,rOIHllcnl 7:'1I5--.J20.
Li'SJ(;'I, I~, '.;,,) S :\IILES. :2001. 'HlInamk growLh "I Ih,' norlh:.:nl nHu'(....rin (}ril~ o';ltllralized ran~(' in \Jonl~IW.i,
US.~. Wl'Ii;."d, 21:2<10-2<16. LI'\I)(;III,:', C.. C. h:Af:('I:, A'\ I) 1\.. '\1: 1'0'\ 2010. The'
!Jio]o~'y of invasive nlien pl~~nts in Ccln,ula. 11. 1(WUi
ri.1' IY/I/l().si"ima LeddJ.. r <:liitU'IlSis Lillir. and hvhrids. ClI""lian JOllrnal or Plant Sl'ient·c' l)O,! 11-12·1
\'](l):,"I:'rn:, .IT. C.S . .I",K,\EI·IL'II, A. ULI,M), W,l. C", J.A. Pun:r,n, J.E. CF";TI.I'" T.J. S'nJlILt:llU" ·1'\'1) J.L. SCJI'\'t-:. 2006. A. lam"n.,k hahitat sllliability mdp rnr the ('un tinl'ntal LJTIlted Stal<:' Frontkr; in E,'O]0b'" aocl th(' Enl'Ll'Onn1C'llt 1:1l-J7
N"']ILE, E.S" J. G ISKI'. S.:"1. Z~I.IH, ~'1. CI-:HA: [j'" '\"'1)
H E. HI:r\OSU. 2·00R. JfUfl,aJ7X spl'de,., {'liUHdric;,u:I'<:e) jnvaflio~ natunu "md i<..'mill~ltUl'al h...\lit.J.l"s in ,\rg(~olin:\. !loklill dl' I" So('i('cbd f\TgcJltina de lIotani<:a ~.'l: lJi-l-iS.
Nxi'!o'Ol. 1'\""'"1' 51'1-.(',11'., CoV'CII. 2008. 2OCI'}-2012 Natu)Oa) 111\ ClSIV(' Spt,x;h~~ ~r;ln'.'lgcnH·l)t Ptlil.
/\,\1'1(>",,1 P.... fit\ Scn, ICE ;2·0Cy) t\lltHl"ct,} He.o.,ollr<.:(' fnl()nlla
tion Portal [online I. 1\ ,'L'IOI1,t1 ['ark SeC'i,·C'. ,\""Iab](' i"om: hUp.'l.!/llrinf-i).Hlb.gov:'rdl'rcrleo.rnvc:lr<'(<.'r('nce/
l'i'.IHt! .. C'.\1., \\1) D.C.. S'IIJ·II. ~OUi. 11I\,,,i\'(' ",IIL-(,(I,11' r.J(P1wri.rL It$ !>pread trOln th(~ ..<\·Inen{·~ln SOllt.hwl'"t to lh," llorth"nl Gr('"t Plain,. Phy.<i"al c."o;(l'aphy 21>::'",(1,-530.
1'1111,111'.,.5.1.. R p. A'DEJ\SO~, ';''\1> H E S( IJAI'lllf.. 20()(), ~·Ii-l:'\in;HIll t'ntl'()p~' I'Hmk·J1I1~ of !'p('t'k~ ~l'o~l"aphil' cl!llrilliltion,. E('olo)!icul \10,klling 1l)O:2:31-~"O
1'1111 1.11''', S.]., \1. 1)\ I.lL~, J. ~'l.Irtr, C.H GMIJ,\~I••-\.
1,111'1\", ]. LI UIIII'ICK, \'\Il S. Fl.RKln. 2U09. Safl~pk .sclcClitlO 'liil~ and prl'~l"T1C:l'wol'lty dJ~tribli~ tion modcb: 111lph<.:ation:lo ~~lr hm.:kb'1"l)Ulid .md psctHlo"b.lcnec UUll. 1::\'oILl!!ie,J A[lpkaliOJlI W.l,~l-Hli,
KI'\C(H I), 1'1..., TK. \1.1(;1-:1-:, I~I) D.\: FI-:\.~. ~(lm; 'I,,,,,"e In\"IL1'livl,.' planl b.lX;.1 til t:S '"\'vst.... rn npi\rian l'L'OS)':-'
ll'!ll!'. ..lnurn:JI nf dl(.~ \;orth Amcrit'.JH Ikllihologkal Suu('[v 27.9~9-9W.
Ikl[\I'\~IJ". ·i'. 1%.5 IlIlrodud,oll. ,pmad ami m,.,,1 extent t)r salt(:('dar UfUIl(i.rh.) iTl 'ht" '\!l..'~tnTl United Stat('~.
r.~. G,·"lo>!;('al S,I1.... l"· I'ro!essiLlnall'apcr 491,,-\. Sn>TI; \1 L. I'L: N·,el.f';. 'EY GLF", C. Y"l.o,:,,·C"'H·
I.\S. J..~. EK~'·:II. E IV. H,·."OI.D" I~B. ':trI\I'BOTrt. E. GOlflV.-LI\IO'\, .\~J) c.L. Jll;'/':> 2008. A.I.lc»in>( the t:.~;\\('n t aod {hn"Tslty Ilf ripl.lrian 'I,~c.·t)"ystt'nh in Sonora,
;l.k.<I<:o. lllOdiV('rsil~ "lid CllnseT\,atioJl 18.247-269, S!'(:U\AI)O j~. \-1.B.•-\Il~IJl), ",,,) \-IZE. K',~I~, 2006 COIl
SC(l1J(,n~(;s of ~patia! alltllconcl"tioo li,r niche-ha,~,d
1Ilo,\"ls JOlllll,J of App!t"d E~(1I()~y ·J3.··I:):J-I·H. SnTO'l. J.I~, A. S \!,A. 'I~D K. \'-Il KH.\\. 2(1()6 Ck<:llll·(·nC(;.
pt:I- ... bterlt·(l. and c"xpan,s;nf! nf' saltl'l'd\\r t]'lJ./rwri.'r 'P)l ~ populatiOn> in Ib" C'u,t Ph"m of \Ioot"n" \Ibtnll l\orth Amel;"a" Ndhlr"lbt 61j.1-1l.
Sr<~ .. ~WEI.I., n H.ll.,"''') D. I'r:TI·.K~. 19\)9. I'h" GAl{1' 11loJ(,.,liol! ,;;, ... tl'ln pT()hlt'lll~ ;lnd ,;olutiofls to ~llitO·
matecl SLl<lti~1 prnl:clioH. 1n teHldtion,,1 Jnl1l'1l"l of C"ogl'<iphjl' Infonllatlon Slic'ncc L3:j.j~-1;;1l
$TO(II.<.:lIf." TJ., ·I~I) J.L.. SClI'\",E. 2·liO!i. HlSk '"1aly,i, tOT lJiological hal..ards: what we n("(,xl to J..:lll)\\- ,1!XHlt
IlwaSlVC ;p"cies Ihs~ An,tll'\;; :W.1&}-'Ji3. TIIO,IIAS. K ..\'\ f) I' GU:r.Tl:-". Z007 <'oulhwest 'lJon-n;lti\'e
InvasivL' Plaot l)alahasc (S\\iE\1 1'07;. l;.S. G<:ologl('a! Sur\'n, SOl1thwest 1l1OIo~ie,,1 Scienl'" Cel1ter (USGS-SIN:). .
TJliJI>'\TO', 1', S III '''\)(', ''\II \J. 1\'1111'1 .. HJ97. G..',,,'I'· (\tlri~ ~IJTfal'c;) of (l1Jl~ lIll'teOlu!ogy \'al'~<Lhk~ 0\'('1" laTge rc~ioll~ of (,:llmpkx tCIT(lin. JOUTlMl oj l-h<lrology 190:21-1-251.
l:I)\'pl.", :\.. D. I1I'\KLD. ,''Ill KC. AI}~", ~OO5. Plaol diver'lll\' in r"ipnl"hlll forest." in Jlortll\\-(",( Colorado t."n(:<.:t~ of titUl,.' ~,nd riv("r regulatIon. Forl'"t Eeoiogy and \I;\ll:tgc:rll~ot218'I07~11 J.
USDA Nnc:S. ~(l(l9. Th" F'L-\S'l'~ f),lt"b".\('. N,ttion,,1 PI""t Dllt" Ccntcl; 1l,'Lon n(HI!!l', L-\ [dtl'd :n ,\pnl 200Y]..~\'aila"lc I'·om. http://plant•.U,lcJ,U~("
WIIIT<..MrI; CIL. I.H..1. '('.,I.II:Y. J.A. CIl(JO~" J. BOI.\ .... U,
,\\'1) J C.b""!'. ~(J07. hl"(l"iinn of GtIH,lflS-k rnlIH{/ri..r sPP) ill a ~ollthl'ro Cald()I,d~1 S.::IJt nhu·-:;h. Bmlo~i("al
I", ""iom 9:·g",-S79.
Rccl'iCl'c! 14 Jt/ly 2010 Accepted 2 .Hm'c!, ZO] 1
See Appendix 011 page 175.
20111 17.';;
API'!"\:)i'. Data .set~ ,~ilt!lcnx:l that mdnd(1 dd1:a f;)T tarrK\ri"k. 1J(lI<1 \\-t'1"(' downloaded fi"Om wW\\'.rliiss.'lr~ on 2 ,lldy 200S.
B,\\" ,md '>1 ll'T ,:20()fj) lh-:"lsh,l\\ il' Ilpllbli"hed data 2006; Colol'"do l)"Ix;rhllvnt of'r","sportation (2002J Colorado projcl't (v.'\.\-"\\.niiss.or~j
{.:nlm"ado St"t(~ PiU-l::S IlI11ppilllj (tl.ta iUnpnhlished <lill'" 200,,»)
I)",,'rn {20(6) FLng(;rpnnting biOd" <'Tsit> (CSli ,n"i L','>G,'> fie'ld data, W,'{w,Ilihs,OT~1
Se"tun <,t "I. 1.2(06) , GI'.lnd St"irt'a,c E";alallt,, \',.tiona] .\Jo""n)~nt rE..-allgGlist,. et ill. 2tXlS'1 HnhlJani L.!k(' [\\ww,niis5 org', Uin",l... ,.( a 1. \2005\ . NiltHlO,,1 )';irk '>el'\';e" (200.» GIS dala N",tional \Vildlif,· H"fll~" PTO,wet {l.:np"l,li,lwd <latai S,'ng\;pt" et al. Nevada (J"'ppin~ ,hit" l,lJnpuhli;hed dat,IiO(5) NJ I55 Citikn Scient''.' Wells it!' )'r"jed, (w,"'\,~itfd,,,rg)
Nortl, lJakota lkpartnwnt ofA&~ieulhll'('ll'npuhlisl",ci data 20031 Otero Counly, Cobrado (L'npuh'i"hed dali. 20(3) Hollimon t)9Wi'j l:,~ Bun:auo!' I..md \Ianagl'nwnl. H"vaJ Gor~l' weed d.,la
(U npnhlishcd ,llta 2oo:'\,i .. SOllth J)akot;, 1J,'partTIIl'nt of "i;ril'lllt\lH' iU npublished d.\tu 200fi) Southwc;;l" Exotic PldJlt \h,ppin~ Prugnllll {Thoma~ <Illd G\lerthi ZOOI, 'famnmk Co,,)itiol1 (ll npuhlished dala 2,008\ C,!orado i\atnr,d He";t.,!() !~'ogr"lll (lillpuhlbh<'d dala 2008i L.S, Iln)'e"u Ilf L'nd ~lanaJ!;l'TlIl'nt, llt"h I)mCl' noxIOUS wced lhlta
(U npuhl i,hed dat" ;200()) Kern, d al (2000)
79poinh 2931 POlllL
18 pol\'gons .5:,3 point., ]bpoinl' 5 pO!Jg(l1'IS
6391.)()j]lt, [.1.3 pojnts 20 point,
1881 points 10 pO]ygOllS II P0lJ)!S
1291 p(,inb,
2 f.lol~ gll!" J5~ pomb \00 pOint,
26'[8 pOlOls }(12.:2: p()int~
143poinb H) point,
16 pol..-goo. 899 point,
22B7 pol\'~O"~ J f points
Z.171~)inf'"
10-14 points
)b Ye... No
)c~/N(i
)c,
1\0 Yc~/~o
Yes Y~,
'C~ '1"" No Yes No