Complex Trophic Interactions in Deserts: an Empirical Critique of Food-Web Theory Author(S): Gary A

The University of Chicago

Complex Trophic Interactions in Deserts: An Empirical Critique of Food-Web Theory Author(s): Gary A. Polis Reviewed work(s): Source: The American Naturalist, Vol. 138, No. 1 (Jul., 1991), pp. 123-155 Published by: The University of Chicago Press for The American Society of Naturalists Stable URL: http://www.jstor.org/stable/2462536 . Accessed: 13/09/2012 15:42

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at . http://www.jstor.org/page/info/about/policies/terms.jsp

. JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship. For more information about JSTOR, please contact [email protected].

The University of Chicago Press, The American Society of Naturalists, The University of Chicago are collaborating with JSTOR to digitize, preserve and extend access to The American Naturalist.

http://www.jstor.org Vol. 138, No. 1 The American Naturalist July1991

COMPLEX TROPHIC INTERACTIONS IN DESERTS: AN EMPIRICAL CRITIQUE OF FOOD-WEB THEORY

GARY A. POLIS Departmentof Biology, VanderbiltUniversity, Box 93 Station B, Nashville, Tennessee 37235

SlubmittedDecember 28, 1988; Revised Jauillat-y26, 1990; Accepted Marcli 23, 1990

Abstr-act.-Food webs in the real worldare muchmore complex thanfood-web literature would have us believe. This is illustratedby the web of the sand communityin the Coachella Valley desert. The biota include 174 species of vascular plants, 138 species of vertebrates,more than 55 species of arachnids, and an unknown (but great) number of microorganisms,insects (2,000-3,000 estimated species), acari, and nematodes. Trophic relations are presented in a series of nested subwebs and delineationsof the community.Complexity arises fromthe large numberof interactivespecies, the frequencyof omnivory,age structure,looping, the lack of compartmentalization,and the complexityof the arthropodand soil faunas. Web featuresfound in the Coachella also characterizeother communities and should produce equivalentlycomplex webs. If anything,diversity and complexityin most nondeserthabitats are greaterthan those in deserts. Patterns from the Coachella web are compared with theoreticalpredictions and "empiricalgeneralizations" derivedfrom catalogs of publishedwebs. The Coachella web differs greatly: chains are longer, omnivoryand loops are not rare, connectivityis greater(species interactwith many more predatorsand prey),top predatorsare rareor nonexistent,and prey-to- predatorratios are greaterthan 1.0. The evidence argues thatactual communityfood webs are extraordinarilymore complex than those webs cataloged by theorists.I argue that most cataloged webs are oversimplifiedcaricatures of actual communities.That cataloged webs depict so few species, absurdlylow ratios of predatorson preyand preyeaten by predators,so few links, so littleomnivory, a veritableabsence of looping, and such a highproportion of top predators argues stronglythat they poorly represent real biological communities.Consequently, the practice of abstractingempirical regularities from such catalogs yields an inaccurateand artifactual view of trophicinteractions within communities. Contrary to strongassertions by many theorists,patterns from food webs of real communitiesgenerally do not supportpredictions arising fromdynamic and graphicmodels of food-webstructure.

Feeding relationshipsin communitiesare delineatedin threeways. The firstis the classic food web, a schematicdescription of trophicconnections. The second quantifiesenergy or mass flow.Finally, interaction or functionalwebs experimen- tallyidentify strong links (Paine 1980; Menge and Sutherland1987). Superficially, littlework is needed to constructfood webs; consequently,they most frequently representcommunities. A rough, qualitative knowledge of "who eats whom" is all that is necessary to produce a simple food web, whereas experimental manipulationsor quantitativemeasurements are necessary to constructwebs of interactionor energyflow. Several approaches analyze food webs (DeAngelis et al. 1983; May 1986; Law- ton 1989; Schoener 1989). One uses models based on stabilityanalysis. The re-

Am. Nat. 1991. Vol. 138, pp. 123-155. ? 1991 by The Universityof Chicago. 0003-0147/91/3801-0008$02.00.All rightsreserved. 124 THE AMERICAN NATURALIST sults are complex and beyond the scope of this article. However, theybasically show that model systems decrease in stabilitywith more species, more links (connectance), or greaterlinkage strength.The dynamic constraintsneeded to maintain stabilityare hypothesized as importantin shaping the propertiesof webs. Stable webs are relativelysimple, short (with few trophic levels), and compartmentalizedand exhibitlittle omnivory or looping (Pimm 1982). A second approach analyzes real food webs to determineregularities in their properties.Analyzed webs were compiled by Cohen (1978), Cohen et al. (1986, 1990), Briand (1983), and Schoenly et al. (1991). Cohen et al. (1986) published a catalog of 113 webs, and Schoenly et al. compiled95 insect-orientedwebs. Theo- rists (Pimm and Cohen) argue that empiricallyderived patternsare consistent with and validate predictionsof the dynamic models above (dynamic models: Pimm 1982, Pimm and Rice 1987; cascade model: Cohen et al. 1990). "Indeed, there is a close tie between the theoreticaland observationalstudies: real food webs have a statisticalpredominance of those featuresthat, in models, increase the chance thatthose models will be stable. The firstis thattrophic interactions, thoughhighly complex, are reasonablypatterned-they demonstratea large cata- logue of assembly rules" (Pimm and Rice 1987, p. 304). Some empiricalpatterns and assembly rules are presentedin Appendix A. In this article, the food web of a desert communityis analyzed explicitlyto evaluate the patternsin Appendix A. Observed patternsare quite differentfrom those assembled from published webs. I argue that most cataloged webs are overly simplifiedand poorly representactual communities.Consequently, the practice of abstractingempirical regularities yields an inaccurateand artifactual view of trophicinteractions within communities.

GENERAL PROBLEMS IN THE ANALYSIS OF EMPIRICAL FOOD WEBS Four substantialproblems beset the catalogs of webs and make them totally inadequate for the types of analyses thathave been conducted (also see Glasser 1983; May 1983a; Taylor 1984; Paine 1988; Sprules and Bowerman 1988; Lawton 1989; Winemiller1990). 1. Inadequate representationof species diversity.-The major problemis that the numbersof species in cataloged communitiesare far less than those in real communities.Most authorsof these webs simplyignored unfamiliar species, con- centratedon taxa in theirexpertise, and/or aggregated or "lumped" unfamiliar species intohigher categories. Lumping is a severe problem.Cohen (1978) labeled lumped categories "kinds of organisms." "'Kinds' are equivalent classes with respect to trophicrelations" (Cohen 1978, p. 7). Briand (1983, p. 253) clarifies and expands: "A 'kind of organism' (interchangeablehenceforth with the term 'species') may be an individualspecies, or a stage in the lifecycle of a size class withina single species, or it may be a collectionof functionallyor taxonomically related species." "Kinds" are also called "trophic species" (Briand and Cohen 1984) and "species" (Cohen and Newman 1985). Kinds include "basic food," "benthos," "other carnivores" (matrix1 in Briand 1983); "algae," "plankton," "birds" (matrix9); "zooplankton," "ice invertebrates,""fish" (matrix21); and DESERT FOOD WEBS 125

"trees and bushes," "insects," "spiders," "soil insects and mites," and "parasites" (matrix27). Only 28.7% of the total kinds in all Briand's webs are real species; nine matriceshave no real species. The "kinds" simplificationwas cIiti- cized by Glasser (1983), May (1983a), Taylor (1984), Paine (1988), Lawton (1989), Lockwood et al. (1990), Winemiller(1990), and Cohen (1978) himselfin a self- critique(but see Sugihara et al. 1989). Lumping is not uniform:plants, arthropods,parasites, and organismsthat live in the soil or benthos are most frequentlygrouped. Invertebratesare analyzed in muchless detail than vertebrates(Pimm 1982),thus obscuring food-web complexity (Taylor 1984; Paine 1988). The incompletepresentation of these taxa is a serious flaw. In particular,arthropods are central to the structureof terrestrial communities.The -800,000 identifiedspecies of insects represent-89% of all animal species (5-50 millioninsect species are estimatedto exist; May 1988). Soil organismsare usually ignored or lumped in spite of theirimportance as major pathways of energy flow in terrestrialcommunities (Cousins 1980; Odum and Biever 1984; Rich 1984). The tactics of ignoringand lumpingspecies produce the depauperate webs compiled by Cohen and Briand. This is obvious froman inspection of Cohen et al.'s (1990) 113-web catalog. The number of "'kinds" rangedfrom 3 to 48 withthe average web "community"having 16.9 kinds. Real communitieshave more species. This is illustratedby enumeratingthe species fromthe sandy deserts of the Coachella Valley (hereafterCV; RiversideCounty, Calif.): 174 species of vascular plants, 138 species of vertebrates,more than55 species of arachnids,and a large but unknownnumber of lower plants,nematodes, acari, and insects(Polis 1991a). Insects are estimatedat 2,000 to more than 3,000 species; I have identified123 families.A still-incompletesurvey in the adjacent Deep Canyon Desert Preserve identified24 orders, 308 families,and more than 2,540 species (Frommer 1986). 2. Inadequate dietaryiniformation.-Published analyses of diets or listsof ene- mies (predators,parasites, and/orparasitoids) suggestthat most species eat and are eaten by from10' to 103other species (see below). The inadequate incorpora- tion of these trophic links is another major weakness of cataloged webs. The numberof prey items recorded is usually a functionof the amount of time and effortdevoted to observation. A "yield/effort"curve (Cohen 1978) is illustrated by analyzingthe diet of the scorpionParliroctotntus mesaensis (fig.1). The number of prey species continues to increase with observation time. The 100th prey species was recordedon the 181stsurvey night; an asymptotewas never reached in 5 yr and more than 2,000 person hours of fieldtime. This suggests that the amount of effortand time needed to determinethe complete diet of just the numericallydominant species is astronomical.It is unlikelythat such an effort was made for most species in the cataloged webs. Thus a food web containing all species stillwould be an inadequate descriptionof communitytrophic relations unless diets were known withmore confidence. Such inadequacy is manifestedin cataloged webs. For example, they show a highproportion (28.5%, Briand and Cohen 1984; 46.5%, Schoenly et al. 1991) of top predators(consumers without predators). It is unlikelythat even 1%, let alone almost one-half,of all animals do not sufferpredators sometime during their lives 126 THE AMERICAN NATURALIST

100- s a) a 80

z 40-

0 1 00 200 Survey Number

FIG. 1. -Yield effortcurve forfirst 100 species of preycaptured by the scorpionParutrocto- nuls tnesaensis in the Coachella Valley.

(see below). These highfigures partially result from grossly incomplete data: for example, in Cohen et al.'s (1986) catalog, 57 chains were of lengthone, that is, 57 herbivoreswere with no recorded predators!Such top predatorsinclude spiders, mites, midges, mosquitos, bees, weevils, fishlarvae, blackbirds,shrews, and moles. 3. Age structure.-Age-related changes in food and predators are not well incorporatedinto web analysis. Populations are composed of age/size classes, each exhibitingsignificant differences in resourceuse, predators,and competitors (Polis 1984, 1988a; Wernerand Gilliam 1984; Polis and McCormick 1986a). Age classes often eat differentfoods, therebyexpanding diet ("life-historyomnivory"; Pimm and Rice 1987). "Ontogeneticdiet shifts"characterize species that undergometamorphosis. The juveniles of at least 27 familiesof CV holometabolic insects eat radically differentfoods (live arthropods)from adults (plants). For species thatgrow slowly througha "wide size range," diet changes more gradually as prey size increases with predator size (e.g., arachnids, reptiles; Polis 1988a). In fact,differences in body sizes and resourceuse among age classes are oftenequivalent to or greaterthan differencesamong most "biological species" (Polis 1984). This magnitudeof change is typical of wide-size-rangepredators (mostinvertebrates, hemimetabolic insects, larva of holometabolicinsects, arachnids, fish,and reptiles; Polis 1984). Predatorsalso change duringgrowth. Juveniles are eaten by species too small to capture adults. Such developmental"escapes" are common to all communities. For example, snakes eat eggs and newborn(but not adults) of carnivorous birds and desert tortoises.Alternately, adults are eaten by predatorsthat do not eat smalljuveniles. Thus some predators(e.g., owls, kitfoxes) in the CV eat only adult scorpions (Polis et al. 1981). In summary,age/size differencesin predators,prey, and competitorsare the normin terrestrialas well as aquatic habitats(contrary to Pimmand Rice's [1987] assertions) and may be major determinantsof populationdynamics and communitystructure. Unfortunately, the richnessthat age structurecontributes has been largelyignored (but see Pimmand Rice 1987). Usually only adults are considered or the diets of all age classes are combined. Age is recognizedin only 22 of 875 DESERT FOOD WEBS 127 kinds in Cohen's (1978) catalog and in 3 of 422 in Briand's (1983). Age structure is oftendifficult to incorporateinto studies; neverthelessit is paramountto com- munitydynamics. 4. Looping.-Looping is a feedinginteraction whereby A eats B and B eats C but eitherB (in mutualpredation) or C (in a three-speciesloop) eats A. Cannibal- ism is a "self-loop" (A eats A; Gallopin 1972). Food-web theoristsdismiss loops as "unreasonable structures"(Pimm 1982, p. 70; see also Gallopin 1972; Cohen 1978; May 1983a; Cohen et al. 1990). Pimmsummarizes from the catalog of webs: "I know of no cases, in the real world,with loops" (p. 67). He has modifiedthis view to include loops in aquatic age-structuredspecies (Pimmand Rice 1987) but stillmaintains that loops are rare in terrestrialsystems. Looping is widespread. Cannibalismis reportedin morethan 1,300 species and is a key factorin the dynamicsof many populations(Polis 1981; see also Elgar and Crespi, in press). Cannibalisticloops are frequentin the CV. Ontogenetic reversal of predationamong age-structuredspecies is the most common formof mutualpredation: A juveniles are eaten by B but A adults eat B (and/orB juveniles) (Polis et al. 1989). This occurs among CV insects, spiders, scorpions, and solpugidsand among predaceous lizards, snakes, and birds. For example, gopher snakes (Pituophis)eat eggs and youngof burrowingowls whereas adultburrowing owls eat younggopher snakes. Mutual predationcan occur independentlyof age structure(Polis et al. 1989; Schoener 1989; Winemiller1990). Two examples fromthe CV are black widow spiders and CV ants. Black widow spiders (Latrodectus hesperus) catch three species of scorpions by using web silk to pull them up off the ground; black widows travelingon the ground are captured by these same scorpions (Polis and McCormick 1986b). Second, CV ants (Messor pergandei, Pogonomyrmex californicus,Myrmecocystus flaviceps) regularlyeat each other (Ryti and Case 1988). Killingand predationof wingedreproductives (after swarming) and work- ers (duringterritorial battles) are a regularinteraction among manysocial insects (see Polis et al. 1989).

THE COACHELLA VALLEY The Coachella Valley is located in Riverside County, California(166?37'W, 33?54'N; area = -780 km2). Wintersare mild; summers,dry and hot. It is a low-elevationrain shadow desertwith average annual rainfallat Deep Canyon of 116 mm. The sand dune/intergradingsand flathabitat was chosen for analysis. This communityis well studied because of the presence of the Universityof CaliforniaDeep Canyon Desert Preserve. Beginningin 1969,Mayhew (1983) sur- veyed the vertebratesfor the Deep Canyon TransectStudy. This workestablished a list of 138 reptiles,birds, and mammalson sandy soils. Some species are not included in the web; for example, mountainlions and badgers are now absent. Further,only the 56 birds (out of 97 residents)that actually nest in the area are considered. Additional data on CV vertebrateswere obtained from Weathers (1983) and Ryan (1968). Invertebratesare less well known. Much of the knowledge of themresults from my long-term research (since 1973) over 17 yrand more 128 THE AMERICAN NATURALIST than 4,000 h of fieldworkincluding more than 4,300 trap days for arthropods. Taxonomic lists are obtainedfrom this fieldwork and fromcatalogs of CV arachnids (Polis and McCormick 1986b) and insects (Frommer1986). Plants were sur- veyed by Zabrinskie (1979). Food and predatorswere determinedfrom the literatureand my work. Since 1979, I have read -820 papers to assemble data forthe CV web. Data fromthe CV were used withpriority; however, I was forcedto supplementthis information with thatfrom other regions, some quite near (e.g., the Palen Dunes) and some muchfarther away (e.g., the ChihuahuanDesert). I could not findthe diets (from desert areas) of 18 birds and one rodent. Finally,interviews with scientistscon- ducting research in the CV provided taxonomic and dietarydata. Because of space limitations,I include only the most importantreferences. Additional references and the identitiesand diets of vertebratespecies and arthropodfamilies are given in Polis (1991a). A series of representativesubwebs depicts trophicrelations in the CV sand community.Subwebs proceed fromplants and detritusto various secondarycon- sumers. Subwebs are connected so that organismsin one web consume (or are consumed by) organismsin the next. Webs are incompletebecause thereare far too many species to include and adequate diet data are unavailable for many species. I thus concentrateon well-studied,focal species, the trophicinteractions of which are relativelywell known. Webs include only species that live in the CV and only interactionsfor which evidence exists. Similar complexityis expected for other, less known species. Thus, these webs understateactual complexity. Consumers are classified in termsof resource specialization (i.e., numberof species eaten withinone group, e.g., plants) or trophicspecialization (i.e., number of differenttypes eaten, e.g., plants, detritus,arthropods) (Levine 1980). Species varyfrom resource specialiststhat eat a few species of the same resource type to trophicgeneralists (omnivores) that feed on several food types. Closed- loop omnivoryis a special case of omnivoryin which A eats both B and C but B also eats C (Sprules and Bowerman 1988).

RESULTS

Plant-HerbivoreTrophic Relations Herbivory describes feeding interactionsinvolving several plant products: leaves, seeds, fruit,wood, sap, nectar, roots, and tubers. Desert herbivoresin- clude microbes, nematodes,arthropods, and vertebrates.I cannot detail the herbivores of the hundredsof CV plants. Rather, I discuss broad groups with the hope of conveyingthe complexityin the plant-herbivorelink. A wide varietyof arthropodseat desertplants (Orians et al. 1977; Powell and Hogue 1979; Crawford 1981; Wisdom 1991). At least 74 familiesof CV arthropodsare herbivoroussome- time in theirlives. Plants that grow in the CV are attacked by many species of insect: more than 60 eat creosote (Larrea tridentata;Schultz et al. 1977), more than 200 on mesquite (Prosopis glandulosa), and more than 89 on ragweed (Am- DESERT FOOD WEBS 129 brosia dlumosa) (Wisdom 1991). Both resource specialists and generalistslive in the CV. Specialists include the grasshopperBootettix punctatus (on creosote; Mispagel 1978). Insects on cactus usually specialize, and generalistsdo not attack cactus (Mann 1969). (Damage caused by feeding on cactus facilitates several specialistand generalistfungi.) Resource generalistsare more common in deserts than are specialists (Orians et al. 1977; Crawford1981, 1986). Generalistsin the CV include the harvesterant Messor pergandei (97 species of CV seed; Gordon 1978) and the camel cricketMacrobaenetes valgumn(16 plantspecies; G. A. Polis, unpublisheddata). Most herbivorousarthropods are trophicspecialists on plants all theirlives: forexanmple, hemimetabolic (Orthoptera, Hemiptera, Homoptera, Thysanoptera) and some holometabolic insects (e.g., curculionid,chrysomelid, scarabid, and buprestidbeetles). However, many holometabolic insects have larvae that are parasiticor predaceous on arthropodsbut whose adults feed on plants (Ferguson 1962; Andrews et al. 1979; Powell and Hogue 1979; Wasbauer and Kimsey 1985). Trophicallyflexible generalists include CV harvesterants (more than 40 categories of foods includingseeds, flowers,stems, spiders, and insects fromat least six orders includingfour ant species; Ryti and Case 1988) and camel crickets (15% plant detritus,41% animal detritus,and <1% conspecifics). Most CV mammals (16 of 18 species) eat plant tissue (the two bats did not). Plants (fruit)formed 0.2%-4. 1% of the diet of the largest mammal,the coyote (Johnsonand Hansen 1979). This is the smallestplant componentfor any of the 16 mammals.Over 50% of the scats of the desertkit fox containedplant material. The two rabbits and the gopher are the only trophicspecialists; however, they are resource generalistseating many species. Omnivorousantelope groundsquirrels fed on a seasonally changingdiet (10%-60% foliage,20%-50% seeds, 62%- 95% total plants by volume; W. Bradley 1968). Rodents (Dipodomnys,Peromys- clIs, Perognathus) fed on seeds and plant parts (and arthropodprey). In total, 15 mammals eat seeds (only bats and the gopherdo not). Nine regularlyconsumed more than 50% seeds in their diet. No CV mammals specialize on particular plants. For example, pocket mice Perognathusformosusfeed on 27 plant species; antelope ground squirrels,24 species. With the exception of the gopher and the two rabbits,all plant-eatingmammals are trophicgeneralists that include arthropods in theirdiet, for example, 1%-17% forDipodomnys merriami and 2%-35% for antelope ground squirrels(this species also eats vertebrates;see below). Many (34 of 56 species) birdsin the sand communityfeed on plantparts (seeds, nectar, flowers,and fruit).Frugivorous birds are common (e.g., cactus wren, phainopepla, verdin, and doves). Some birds eat fruitas a minorcomponent of an omnivorous diet (e.g., roadrunner,Scott's oriole, western tanager, western bluebird,warbling vireo, Bewick's wren). Granivoryis also common: 22 of 56 CV birds were reportedto eat seeds (13 are primarilygranivorous). Many insec- tivorous desert birds eat significantquantities of seed when insects are scarce (Brown et al. 1979; Brown 1986; Wiens 1991). No herbivorousbirds are resource specialists. In fact, trophicspecialists are rare; of 34 plant-eatingbirds, only five are not recorded to eat arthropods. Two species of CV reptilesare primarilyherbivorous: desert tortoise and desert 130 THE AMERICAN NATURALIST

iguana. Both are resource generalistseating a wide varietyof plants (17-40 species forthe tortoise)and plant parts. Only the tortoiseis a trophicspecialist. The diet of the desert iguana contains 1%-5% arthropods.Five of the nine other lizards (none of 10 snakes) consume a minorportion of plants. Detritus,Soil Biota, and BelowgroundHerbivory Detritusis a broad termapplied to nonlivingorganic matter from living organisms. It is a universalcomponent of food webs simplybecause all organismsdie, plant parts senesce, and animals defecate. It may originatefrom plants (e.g., wood, leaves, seeds, flowers, and roots [rhizodeposition])or animals (feces, urine, secretions,molted skin or fur,and dead animals). Most primaryproductivity flows directly or indirectlythrough the detritalcom- ponent of food webs. Herbivores process 1%-50% of net primaryproductivity; the restenters the detritalsystem (Macfayden 1963; Odum and Biever 1984). This is particularlyso in deserts, where the main energyflow often proceeds directly fromautotrophs to detritivores(Seely and Louw 1980; Wallwork1982; Crawford 1991; Freckman and Zak 1991). The plant-herbivore-carnivorelink forms12%- 33% of the fate of plant productionin deserts; the remaindergoes throughthe soil/detrituschain. Nevertheless,Cousins (1980) is one of the few to incorporate detritusexplicitly into food-web analysis (see also Odum and Biever 1984). He disputes placing autotrophsalone at the basal positionof webs; rather,herbivory and detritivoryshould be consideredequally importantlinks in a "trophiccontinuum." Energy,produced by autotrophsand consumed duringsecondary produc- tion, is recycled and made available to otherconsumers by detritivores. A diverse biota lives withindesert soils (Crawford1981, 1991; Wallwork 1982; Freckmanand Zak 1991). Microbes (fungi,yeast, bacteria, protozoa), nematodes, mites, termites,some ants, Collembola, Thysanura, cockroaches, tenebrionid larvae, millipedes,and isopods are some of the morecommon of the manydetriti- vores that live withinCV soils. Althoughspecies in these taxa degrade organic material,many include facultativeor obligate herbivoreson belowgroundplant parts (Crawford 1981, 1986). Over 50% of net primaryproduction is commonly allocated to belowgroundplant parts (Andersen 1987). For example, in Russian deserts, 65% of the plant biomass is belowground(Rodin and Bazilevich 1964). Species fromseven orders of insects, mites,nematodes, and some rodentshave adopted belowgroundherbivory as theirprimary feeding mode (Andersen 1987). Soil organismsare quite abundantin deserts. Nematode biomass of 1-20 g/m2 normallyoccurs (Freckman and Zak 1991). Detritivoresform 37%-93% (mean = 73%) of all individualmacroarthropods in fourdeserts analyzed by Crawford (1991). Termites are particularlyabundant; their biomass is often an order of magnitudehigher than that of any otherdesert animal (MacKay 1991; Polis and Yamashita 1991). Wallwork (1982) emphasized theirimportance in desert webs: termitesfix nitrogen,eat large quantities of detritus,recycle nutrientswithin theircolonies via trophallaxisand cannibalism,and ultimatelyrelease nutrients to predators. A rich web based on detritusand undergroundplant parts exists withindesert soils (see, e.g., Whitford1986). Nematodes occupy several trophicroles (Freck- DESERT FOOD WEBS 131 man and Zak 1991): herbivores,plant parasites, microbialfeeders, fungivores, omnivores,omnivore predators, parasites. In the Mojave, thereare fourtrophic groups of Prostigmatasoil mites: phytophages(1 family),fungivores and detritivores (3), parasites (1), and predators(8) (Franco et al. 1979; see also Santos et al. 1981). Predatorymites in litterare as common as nonpredatorsin nearby Joshua Tree Monument (Wallwork 1982). These mites eat nematodes, Collem- bola, and other mites. The large numberand diversityof predatorymites led Edney et al. (1974) to conclude thattwo or more predatortrophic levels exist in the decomposer web. Wallwork (1982) and Santos et al. (1981) suggested that decomposer pathwaysin desert soils were regulatedby mitesthat prey on nematodes thatfeed on microorganisms. Soil interactionsbecome even morecomplex with the inclusionof macroarthropods. Most detritivorousarthropods not only eat detritus,but also feed on microorganisms(bacteria, fungus,protozoa) feedingwithin the detritus(Janzen 1977). For example, in the CV, the burrowingcockroach Arenievaga feeds below ground on livingand decaying roots and ensheathedmycorrhizae. Further, most detritivores (e.g., cockroaches, tenebrionids,millipedes) host cellulolyticmicrobes that degrade plantdetritus (Crawford 1991). Althoughsuch symbiontsform a separate energetic"trophic level," theyusually are not includedin web analysis. Finally, several desert insects consume detritus directly or are predaceous on microarthropodsand nematodes (e.g., in the CV, larvae of asilid, bombyliid,and theriviidflies, and staphylinidand clerid beetles; Edney et al. 1974; Powell and Hogue 1979). Soil interactionsare not separate fromthe rest of the community.Surface and subsurfaceherbivores are involvedin competitionand facilitation(Seastedt et al. 1988). Most important,energy flows in bothdirections via the trophiccontinuum. Many surface dwellers in deserts either spend part of theirlives in the soil (as larvae, e.g., tenebrionids)or feed on arthropodsthat live permanentlyor tempo- rarilybelow ground (see Ghilarov 1964). For example, 46% of Parulroctonuis rnesaensis prey live in soils as larvae. Arenivaga form23% by weight of this scorpion's diet. Termites(10 CV species) and tenebrionids(16 species) are importantconduits of energyflow from below groundwhen theyare eaten by a diverse groupof arthropodand vertebratepredators (Wallwork 1982; 35 species of known surfacepredators in the CV). Such predationby surfacedwellers exports much of the energy recycled by detritivoresand links the soil subweb to that above ground. Thus, even if herbivoresand detritivoresoperate in distinctmicrohabi- tats, energyflowing further into the communitymerges into the bodies of predators common to both consumers(Odum and Biever 1984). No studyhas analyzed trophicinteractions within detritus and soil in the CV. The studiesabove were conductedin deserts(Mojave, Chihuahua)geographically adjacent to the CV. I combined informationfrom these studies (esp. Whitford 1986; Freckman and Zak 1991) withmy data to constructa soil/detritussubweb (fig.2) using CV taxa. Note the complexity(even withextensive lumping), loops, frequentomnivory (often at nonadjacentlevels), closed-loop omnivory,chains of 4-5 links, and links between belowgroundconsumers and aboveground predators. 132 THE AMERICAN NATURALIST

Further into the food web: Arthropodiuorous insects, arachnids and uertebrates on the surface

I Predaceous insect laruae (e.g., asilids, bombyliids, staphylinids, clerids)

Predaceous/ i Nematophagous mites _(Gamas,na) (tyjdeids, parat jdeids it-

Predaceous Hicrophagous ematodes a. M (cephalobid iacroarthropods nematodes Fungiuorous (e.g.,tenebrionids, jW sj ~Mites firenivago, Thysanura,

15 ^ . , pyrsoemotd)j isopods millipedes

Protoz oa Fung ivoroDUS,f 2/ nematodes// lPAroti/foa - t e 1Microarthropods Bacteria Orabatid IO species) (e.g.,Collembola)

Fungi Lt? Detritus \ Below ground liuing t \ ~~~~~planttissue

Dead plant and animal tissue Surface detritiuores

FIG. 2.-Trophic interactionswithin sandy soils in the Coachella Valley. The identitiesof a few of the importantdetritivorous species are as follows: Isopoda (Venizillo arizonicus), Collembola (Entomobryididae),Thysanura (Leucolepisa arenaria, Mirolepismadeserticola), Gryllacrididae(Macrobaenetes valgurn),Blattidae (Arenivaga investigata), Isoptera (five Amitermesspp.; five other species), Tenebrionidae (16 species). Note that not all trophic links are represented(e.g., for tenebrionidsand termites).An arrow returningto a taxon indicates cannibalism.

Carrion Feeders The decompositionof carrionresults from the cumulativeaction of microorganisms, necrophagousinsects, and some vertebrates.A richcarrion fauna occurs in deserts (McKinnerney 1978; Schoenly and Reid 1983; Crawford1991). Complex interactionsinvolve from28 to more than 500 species. McKinnerney's analysis of carrionfrom two rabbitsthat occur in the CV identifies63 arthropodand four vertebrateconsumers. Some specialize on particulartissue; others do not. Tro- phic specialists and generalistsoccur. Species compositionand diversitychange throughtime. Necrophagous insects, generalistand specialistpredators, and omnivores are common. Vertebratesnot only eat carrion but eggs and larvae of insects. Carrionfeeders also consume microorganismswithin the carrion(Janzen 1977). Further,many carrion species are well-knowncannibals (Polis 1981). Inter- actions within carrion do not constitutea distinctcompartment: much of the energyfrom carrion is exported to the rest of the community.Most organisms DESERT FOOD WEBS 133

Further into the food web ad

Clerid beetles iguetia - spiders (e.g., Phyllobaneus)

Facultatiue hyperparasitoids < _ (2 Species)

1I7 sp ecies of /Primary parasitoids Predators parasitoids (7 species) (2 species) Bileys parasitoid

Cecidomyid gall formers Inquilines in gall i(8 species) (7 species) Other herbiuorous arthropods

Larres and 4 Rtripleu salt bush other species | of shrubs l

Larres (creosote) food web (>60 species of insects and 26 species of spiders)

FIG. 3.-Trophic interactionswithin galls on saltbush Atriplex canescens within the Coachella Valley. In total, 67 species interactwithin galls formedby Cecidomyidae (midge) larvae. Many of these species are also involvedin the subweb centeringon creosote (Larrea divaricata; see Schultz et al. 1977). Note thatthe interactionsof some species are not fully represented(e.g., Diguetia, Phyllobaneuls). Modifiedfrom Hawkins and Goeden (1984). associated with carrion are opportunistic:spiders, solpugids, Opiliones, ants, asilids, staphylinidbeetles, reduviids, and the vertebratesnot only eat insects associated withcarrion (or carrionitself) but also preyon otherspecies (McKin- nerney1978). In turn,all these species are eaten by otherpredators. ArthropodParasitoids Parasitoids from several families of flies, wasps, and beetles are a diverse component of webs representingmore than 10% of all animal species (Askew 1971). They lay eggs in or on arthropodhosts; larvae feed on and cause the death of the host (in contrast to parasites in general). Adults almost always feed on other foods (usually of plant origin). Parasitoids' trophicrelations are generally quite complex (Askew 1971; Price 1975; Pimm 1982; Hawkins and Goeden 1984; Polis and Yamashita 1991). Hawkins and Lawton (1987) estimatethat each species of insect herbivoreis host to 5-10 species of parasitoids. A few studies detail parasitoid-hostrelationships in the CV. Hawkins and Goeden (1984) studied insects associated withsaltbush (Atriplex) galls. The system is complex with67 species (40 commonones), at least fivetrophic links, and 134 THE AMERICAN NATURALIST

O Rrthropodouores further into the food web Tastiotenis hyjperp arasi toi d P. mesaensis Cho isI6n scorpions parasitoid wasp spiders and parasitoid wasp copon ~~~~thersolpugids -.0 Mime tus Stestoda (>50 species) spider -spid,ers \*< N A4 -1 / rLatrodectus spiders Pteromalid IO other spiders _ IA parasitoid inuading egg sac Di ti L w asp I _ v O~~igue tia _A adults Salticid s ers Phyllobaneus (>6 species) < \, / n \ \ beetle (>6 species) igu ticlerid

FPredaceous au Partsitoi ects(e.g., asilids, li insects (e.g., eggsdelnsader mutillidar and tiphiid antlions, mantidsi.4 wasps; Photopsis) (>48 families) n Rnts \ (bombyliids, >75

FIG. 4.-Trophic interactionsabove the soil surface involvinga few of the predaceous arthropodsliving withinthe Coachella Valley. This subweb is focused around the spiders Diguxetiamezohlalvea and Lat6odectus hespecis. Note thatno vertebratesare represented. extensive omnivory.Gall-forming midges (3 species), parasitoid Hymenoptera (26), predators(4), and inquilines(7) interactwithin galls (fig. 3). Midge larvae are either resource specialists on Atriplexor generalistson other plants. Most parasitoidsare primary,attacking only midges or inquilines; seven also feed on gall tissue. Two facultativehyperparasitoids feed on gall tissue, midges, and primaryand hyperparasitoids.A clerid beetle (Phyllobaenus) is in 10% of galls and feeds on at least 17 species fromall trophicgroups (and spiders; fig.4). The trophicrelations of Photopsis (an abundantmutillid wasp in CV sands) are diagrammedin figure5 (fromFerguson 1962 and my data). Females oviposit into larval cells, and Photopsis larvae consume the entirehost. They parasitize several species of hymenopteranlarva and are hyperparasitoidon parasitoids of these larva (i.e., other Hymenoptera; stylopid,meloiid, and rhipiphoridbeetles; and bombyliidflies). Some hyperparasitizedwasps (e.g., sphecids) also may parasitize spiders (this is likely but not established). Up to 37% are destroyedwhen Pho- topsis larvae themselves are parasitized by some of the same parasitoids (e.g., sphecids) that fall host to Photopsis. This is an example of looping via mutual parasitismand tertiaryparasitism. Some Photopsis larvae are also host to bombyliid(e.g., Bembix) and stylopidparasitoids. Further,Photopsis larvae also are DESERT FOOD WEBS 135

Rrthropodouores further into the food web

Scorpions (e.g., Paruroctonus messensis)

Hyperparasitoid Spiders (e.g., Biguetia Hymenoptera and Latrodectus)

Photopsis adults

Photopsis laruae

Coleoptera and Diptera parasitoids

Hymenoptera laruae

Partitioning into larual cell by adult Spiders

Insects

Pollen

FIG. 5.-Trophic interactionsinvolving parasitoid mutillid wasps in the genus Photopsis withinthe Coachella Valley. Note thatthe interactionsof some species are not fullyrepre- sented (e.g., scorpions, spiders). A double-headed arrow indicates looping via mutualpredation. cannibalized (and self-regulated?[Ferguson 1962]). Adults eat not only nectar, pollen, and flowersbut also ground-nestingHymenoptera. Adults are frequent prey to many arthropods(e.g., scorpions and spiders; see figs.4 and 6). These parasitoidsubwebs are characterizedby long chains and frequentomnivory,closed-loop omnivory,and looping(via cannibalism,mutual predation, and three-speciesloops). Further,key species exportenergy from this subweb when theyare predatorsor prey in the rest of the community. Overall, more than 20 familiesof insects are parasitoidsof CV insects. Many preferhost species; others, however, are more generalized (see figs. 3 and 5). They sometimescause highmortality (Mispagel 1978). Several dipteran(tachinid, bombyliid,sarcophagid) and hymenopteran(tiphiid, ichneumonid, mutillid, sphecid, and chalcidoid wasps) parasitoidsare commonin the CV. These develop on eggs and immaturestages of Orthoptera,Neuroptera, Coleoptera, Lepidoptera, Hymenoptera,and Diptera. Velvet mite larvae (Dinothrombiumpandorae) are ectoparasitesof CV grasshoppers(Tevis and Newell 1962). Spiders host many parasitoids (pompilid, sphecid, and ichneumonidwasps; many Diptera). Adult wasps partitionnests with captured spiders; developing 136 THE AMERICAN NATURALIST

Further into the food web (no carniuory depicted here) Kit fox X ON Coyote -4

BurrowJing -101o a d iu....e. and 3 4 other owJls I D Grasshopper B fintelope < _ o mouse ground squirrel

_ K~~gPredaceous Loggerhead_G No lizards shrike (1O0 species) Pallid bet 11fn n d rthropod-s

Pi eating snakes Large s ow resources sRadult Hadrurus

Abundant Apors hirsutsan th lesst uommon Pscorpions

L a trodes i es)aen srpies p

9ptostichus ^< \ / Spiders > -Juuenile P. / F|

Partesaensasite sadrurus scorpions -Other spiders and solpugids and species-

clants and/or detritus tions fomaphis and ectarfomdmor tharuotnu1slnpce. oeC opld FoG.6.aTrophic interactionscentered around the preyand predatorsof the scorpion ParuroB4ctonuxsinesaensis. Note thatthe interactionsof some species are notfully represented (e.g., "spiders" and "insects") and carnivory on vertebrates is not depicted (see figs. S and captued7). pry(G.a.d Pis,esona obevto)

larvae eat the moribundspiders. Adults usually eat nectar. Wasps vary from resource specialistson particularfamilies to generalistson several families.Many pompilids(more than I11species) occur in the CV (Wasbauer and Kimsey 1985). AbundantAporus h1irsutuisand the less common Psorthaspis planata feed trap- door spiders (Aptostichus,Ctenizidae) to theirlarvae; adults drinksugar secre- tionsfrom aphids and nectarfrom more than 10 plantspecies. Some CV pompilids are hyperparasitoids(e.g., Evagetes mnohave).Spiders are also beset by a diversityof egg parasitoids/predators(Polis and Yamashita 1991). Further,kleptopara- sitic insects (e.g., Drosophiloidea flies)and spidersparasitize spidersby robbing captured prey (G. A. Polis, personal observation). Parasites Parasites are a diverse and species-richgroup that feed on almostall taxa. They can influencegreatly population dynamics and communitystructure (May 1983b) and formanother step in the flowof energy(some ectoparasitesform yet another DESERT FOOD WEBS 137

"gratis" level when theythemselves host parasites). Many feed on several hosts duringtheir complex lifecycle. However, parasites are neitherwell repr-esented in food webs nor well studied in naturalcommunities. Of animals living in the CV, only the parasites of the coyote, lizards,and scorpionswere examined in any detail. Other-wise,few data exist on the hundredsof CV endo- and ectoparasites. Telford (1970) identifiedparasites in 10 CV lizards. Lizards were infectedby several protozoans (mean = 7.8 species; range = 3-10) includingflagellates, ciliates, amoebas, sporozoans, and gregar-ines.Helminth parasites (mean = 1.4; range = 0-5) included nematodes, cestodes, and Acanthocephala. Each lizard also was infestedwith an unknownnumber of mite species. Coyote par-asitesinclude mange mites, ticks (seasonal), lice (rare), and Pillex fleas (on all individuals) (Gier et al. 1978). Adult fleas feed on blood; flea larvae feed on organic debris. Endopar-asites,specifically the tapeworm Tae;lia pisti- formes,occur in 60%-95% of all coyotes. The primeintermediate host of Talen1ia is the cottontailrabbit. It is likelythat most (all?) of the fi-ee-livinganimals in the CV harbor-parasites. For example, 23.4% of 1,525 birds (112 species) fromdeserts and otherareas in the southwesternUnited States had blood parasites (Woods and Herman 1943; Welty [1962] lists the diver-separasite fauna of birds). Inspection of CV spiders and insects usually reveals mite infestation.Many genera of CV spiders were reportedwith nematode parasites. Coachella Valley scorpionssupport nematodes and eight(pterygosomid) mite species (G. A. Polis, unpublisheddata). Asthiiopod Pr-edators Arthropodsare one of the most importantconduits of energyflow in deser-t webs. Most consumed primaryproductivity in deserts is utilized by arthr-opods ratherthan by vertebrates(Seely and Louw 1980). These arthropods,in turn,are eaten by a host of predators,the vast majorityof which are other arthropods (Crawford1981). Many predaceous arthropodsare dense and may play important communityroles (Polis and Yamashita 1991); in the CV these include species of scorpions, solpugids, spiders, mantids,ant lions, robberflies,small car-abidbee- tles, and facultativelypredaceous ants. A great varietyof arthropodsare predaceous sometimeduring their lives. In the CV, mites (8 families),arachnids (more than 23), and insects (more than 21) are predators as juveniles and adults. The complex life cycle of holometabolic insectsoften results in differentfeeding habits between stages. At least 27 families of Diptera (e.g., Tachinidae), Hymenoptera (e.g., Tiphiidae), and Coleoptera (e.g., Cleridae, Meloidae) are trophicgeneralists: they are predaceous as larvae and herbivorousas adults. For example, Pherocera (Therevidae) flylarvae are predatorson beetle, fly,and mothlarvae (and are cannibals) in sandy CV soils; adults drinknectar. Some parasiticHymenoptera (e.g., sphecids,pompilids) func- tion as predators: adults eat pollen and nectar but capture prey to feed larvae. Finally, some taxa (e.g., ants, camel crickets)are occasionally but regularlypre- daceous. For example, afterheavy rains in the CV, the diet of M. perganldei included 80% Amitermeswvheeleri termites; normally, arthropods form 2%-10% of this ant's diet (Gordon 1978). 138 THE AMERICAN NATURALIST

TABLE 1 DIET CLASSIFICATIONOF SOMEREPRESENTATIVE PREDATORS ON ARTHROPODS

PERCENTAGEOF ARTHROPODTAXA IN DIET

TAXON Predaceous and Parasitoid Herbivorousand Detritivorous n

Arachnids:* Hadturuisarizonensis 53 47 15 Paruroctonus luteoluls 60 40 10 Parulroctonulsmesaensis 47 53 126 Vaejovis confulsus 50 50 12 Diguetia mohavea 45 55 71 Latrodectulshesperus 54 46 35 Arachnidaverage + SD 51.5 ? 5.4 48.5 ? 5.4 47.3 Lizards: Callisauruzlsdraconoides 45 55 22 Cnernidophorustigris 46 54 15 Gainbelia wislizenii 33 67 15 Phrynosomaplatyrhinos 28 72 18 Uta stansburiana 39 61 28 Birds: Blue-graygnatcatcher 36 64 69 Burrowingowl 36 64 14 Loggerhead shrike 35 65 17 Roadrunner 35 65 23 Mammals: Ammospermophilusleucurus 71 29 7 Antrozuspalliduis 31 69 16 Onycloinys torriduis 40 60 15 Pipistrelllushesperis 32 68 22 Viulpesinacrotus 67 33 6 Vertebrateaverage ? SD 41 ? 12.9 59 ? 12.9 20.5

NOTE.-A taxon is the designatedunit in whichthe diet was classifiedby the author.It varies from species to familiesand orders. * The firstfour arachnids are scorpions; the last two are spiders.

Most are resource generalists (Polis and Yamashita 1991); for example, P. mesaensis is recordedto eat more than 125 prey species, Digueta mohavea more than 70 species, and Latrodectus hesperus, 35 species. In fact, some scorpions and spiders are neithertrue trophic specialists nor obligatepredators: both scav- enge dead arthropodsand some spiderlingsare aerial planktonfeeders, eating pollen and fungalspores trappedby theirweb (Polis and Yamashita 1991). Facul- tative predatorsare trophicgeneralists eating plants, detritus,dead arthropods, and live prey. A few specialize. Adult velvet mites, D. pandorae, feed almost exclusivelyon termites(Tevis and Newell 1962). Mimetusspiders prey primarily on other spiders. Trophic interactionsare complex. Generalized diets are established by size relationships:predators catch what theycan subdue. Consequently,smaller and/ or youngerarthropods are potentialprey and predatorseat fromall trophiclevels. For example, the diet of six CV arthropodsaverages 51.5% other predaceous arthropods(table 1). Predator-predatorfeedings are particularlycommon in deserts because predatorsform a highproportion of all arthropods(Crawford 1991; Polis and Yamashita 1991). Clearly,a web representingall CV predaceous arthro- DESERT FOOD WEBS 139 pods would be difficultto depict. Thus, I presentsubwebs (figs.4 and 6) centered on three common species that I have studied extensively:the scorpionP. mnes- aensis (see earlier references) and the spiders D. inohavea and L. hesperius (Nuessly and Goeden 1984; Polis and McCormick 1986a, 1986b; G. A. Polis and K. H. Sculteure, unpublishedmanuscript). SpiderlingD. mohavea develop in sacs protectedby the motheruntil her death. Eggs and spiderlingsare then eaten by invaders (spiders [9 families,more than 14 species], solpugids,mites, mantispids, chrysopids, and theclerid beetle Phyllo- baenus fromfig. 3). These stages of most spidersare attackedby similarinvaders (Polis and Yamashita 1991). Siblingcannibalism is also frequent(Polis 1981) and occurs among D. mohavea (and L. hesperues)spiderlings. At least threetrophic levels occur in the D. inohavea retreat:the clerid likelyeats otheregg predators and is itselfhost to a pteromalidwasp parasitoid. Adult Diglietia eat more than 70 species, including14 familiesof predatoryinsects (e.g., Photopsis and Phyllo- baen,is), and eightspider species, includingthe same species of invadingHabro- natiis salticids and the araneophagous Mimetus. One-thirdof all D. ;nohavea webs include salticid prey. Adult D. mohavea are fed upon by Mi;net,is, P. mesaensis, birds, and a parasitoidpompilid. Diguetia is involved in at least four cases of mutualpredation. Predators form54% of L. hesperus's diet (table 1). Latrodectus hesperus is prey of at least eightpredators including other L. hesperuesand three predators thatit eats (mutualpredators = Steatoda grossa, Steatodaftulva, P. mesaensis). The sphecid (Chalybian californicuin)specializes on Latrodectiusand otherthe- ridiidspiders (e.g., Steatoda) (Wasbauer and Kimsey 1985). Tastioteniafestiva (Pompilidae) is a hyperparasitoideating both cached theridiidspiders and developing wasps. The biomass (g/ha) of P. mnesaensisis the greatestof any CV predator(including vertebrates).Diet shiftsduring growth partially explain trophicinteractions with more than 125 prey species, including47% other predators(table 1), and mutual predation with at least 10 species (three scorpions, five solpugids, two spiders-young P. mnesaensisare eaten by the same species eaten by adults). Note the complexityof these webs: loopingvia mutualpredation and cannibalism is frequent;(closed-loop) omnivoryis the norm;omnivorous predators feed on herbivores,detritivores, predators, and predatorsof predators.Consequently, chain lengths are long even excluding parasitoids and loops (e.g., detritus- termite-Messor ants-ant lion-Latrodectus-Steatoda-Mimetus-P. mesaensis- Eremobatid solpugids-Hadrurus scorpion-[vertebratesubweb]). I stronglysus- pect that the depicted interactionsare representativeof the hundredsof other arthropodpredators in the CV. Omnivory(due to age structure,opportunism, and generallycatholic diets) combinedwith a highdiversity of insectand arachnid predatorsnecessarily creates trophiccomplexity. Complexity increases even fur- therwhen we consider vertebratepredators of these arthropods. ArthropodivorousVertebrates Arthropodivoryis the consumptionof arthropods.A less familiarword than insectivory,it conveys thatpredators eat all types of arthropods(insects, arach- 140 THE AMERICAN NATURALIST

TABLE 2

FEEDING CATEGORIES OF THE VERTEBRATES RESIDENT IN THE COACHELLA VALLEY

VERTEBRATE CLASS

Reptiles Birds Mammals All Vertebrates FEEDING CATEGORY (n = 21) (n = 56) (n = 18) (n = 95)

Granivory: Primary 0 (0) 14 (25) 8 (44) 22 (23) Secondary 0 (0) 0 (0) 0 (0) 0 (0) Total 0 (0) 14 (25) 8 (44) 22 (23) Herbivory: Primary 2 (10) 5 (9) 5 (28) 11 (12) Secondary 5 (24) 15 (27) 1 (6) 21 (22) Total 7 (33) 20 (34) 6 (33) 32 (34) Arthropodivory: Primary 11(52) 34 (61) 9 (50) 55 (58) Secondary 4 (19) 15 (27) 5 (28) 24 (25) Total 15 (71) 49 (88) 14 (78) 79 (83) Carnivory: Primary 9 (43) 12 (21) 3 (17) 24 (25) Secondary 2 (10) 2 (4) 1 (6) 5 (5) Total 11 (52) 14 (25) 4 (22) 29 (31)

NOTE.-Species are classified as belongingprimarily (food is a major componentof the diet) or secondarily(food < 10% of total diet) to a feedingcategory. Some omnivorousspecies (2 reptiles,9 birds,7 mammals)belong to two (each > 33% of diet) or three(each >20% of diet) primarycategories. The numbersin each column indicate the numberand percentage(in parentheses)of species in the primary(e.g., 11 herbivorousspecies) or secondary (e.g., 21) feedingcategory and the total in this category(e.g., 32). nids, myriapods,and terrestrialCrustacea). Most vertebrates(83% of 95 species) in the CV eat arthropods(table 2). Over half (58%) primarilyeat arthropods, including52% of the reptiles,61% of the birds, and 50% of the mammals. Most (80%) of the 25 primarycarnivores also feed on arthropods.Two-thirds (24 of 36) of the primarilyherbivorous/granivorous vertebrates eat arthropods,at times in large quantities(e.g., 88%-97% of the seasonal diet of the sage sparrow in the Great Basin Desert). In total, 71% of all reptilespecies, 88% of birds, and 78% of mammalsprimarily or secondarilyeat insectsor arachnids(only 17 species are not reportedto eat arthropods). These vertebratesare resource generalistseating all trophiccategories of arthropods (herbivores, detritivores,predators, parasitoids). For example, of 36 arthropodivorousbirds whose diet is detailed sufficiently,58% eat spiders in additionto insects. Seven of 10 lizards eat spidersand foureat scorpions. Spiders are eaten by three of 14 arthropod-eatingmammals; scorpions, by five. Preda- ceous arthropodsaverage 41% of the diet of the vertebratesin table 1. Further, these vertebratestend to be trophicgeneralists. Most (28 of 55 species = 51%) primaryarthropodivores eat plants (59% of 79 vertebratesthat eat arthropods also eat plants). Of these 79, 32% are also carnivorous.A few arthropodivorous vertebratestend to specialize. Ants form89% by frequencyof the prey of the hornedlizard Phrynosomaplatyrhinos. However, these lizards eat 17 othercate- gories of prey (includingspiders and solpugids)and 20%-50% of the diet of some DESERT FOOD WEBS 141

_i- Great horned s rGolden eid s Coyote j ow*i%2

noryehbt b h i X vl oe Ktingsnake P e iadewnis erat Gopher snake h

_ l-Burrow ing Roadrunner /

Plants~~~~~~~~

r-oCunnsquirorres V Arthropod-s rieopard eeraseL (R eating snak se 2 izard 5

are p ayArthropor eatingou _ - \ \ ~~~~lizardsA bOther birds= 2 a and rodents Arachnids

I nsects

Plants

FIG. 7. Trophic interactionsinvolving a few of 96 vertebratesresident in the Coachella Valley. Note that the bottomof this subweb is simplifiedfrom the procedingfive subwebs (figs.Carnivores2-6). kill ada vetbrts.Ote95vrtbats,2 repimrl individuals is beetles. The worm snake (Leptotyphlopshumilis) mainly eats termitesand ants. No otherCV vertebratesspecialize on certainarthropods. The omnivoryexhibited by arthropodivorousvertebrates is illustratedin the web focused on predators of P. mesaensis (fig. 6). All 16 reptiles, birds, and mammals recorded to eat P. mesaensis eat (predaceous, detritivorous,herbivo- rous)insects; 44% eat plantmaterial. Many (81%H) are also carnivores(see fig.7; threeof II arthropodseating P. mesaensis also eat vertebrates). Car-nivor-oiisVer-tebt-ates Carnivores kill and eat vertebrates.Of the 95 vertebrates,24 are primarily carnivorous and five others eat vertebratesoccasionally (table 2). Reptiles are the most carnivorous(9 primary,2 secondaryof 21 species = 52%; 8 of 10 snakes are primarilycarnivores). The proportionof carnivores is about the same for birds (12, 2 of 56 species = 25%) and mammals(3, 1 of 18 species = 22%). All carnivoresare resourcegeneralists preying on manyvertebrates. Most (79% =19 of 24 primarycarnivores) are trophicgeneralists that eat arthropods(71%) 142 THE AMERICAN NATURALIST

and/orplants (33%). For example, coyotes eat mammals(12 of 19 species in the CV: rabbits,rodents, gophers, antelope groundsquirrels, and even kitfoxes and other coyotes), birds (including eggs and nestlings;e.g., roadrunners,doves, quails), snakes (e.g., gopher and kingsnakes),lizards (e.g., horned lizards), and young tortoises as well as scorpions, insects, and fruit(see fig.7; Johnsonand Hansen 1979; G. A. Polis, unpublisheddata). Coyotes activelyforage on nonver- tebrate prey; for example, they excavate entireant nests in the CV (Ryti and Case 1986). Great horned owls consistentlyeat arthropods(14.7% of diet; e.g., spiders,centipedes, Orthoptera) in additionto rodents,squirrels, lizards, snakes, and otherhorned owls (Ohlendorf1971; Johnegard1988). Many carnivores(33%) feed on carrion in addition to live prey: sidewinderrattlesnake, raven, golden eagle, hornedowl, red-tailedhawk, coyote, kitfox, and antelope groundsquirrel. Such scavengingmeans that these primarycarnivores also include in theirdiets bothmicroorganisms (Janzen 1977)and the richarthropod fauna that use decaying carcasses (e.g., the hornedowl). Figure 7 is a subweb focusingon some carnivores in the CV. Note the fre- quency of omnivory,closed-loop omnivory,cannibalism (10 of 16 species in fig. 7), and mutualpredation. In each of the threecases of mutualpredation A adults eat B, but the eggs or nestling-stageindividuals of A are eaten by B. For example, many snakes are nest predators(e.g., in the CV, whip snake, sidewinder,rosy boa, gopher snake). These snakes eat eggs and nestlingsof species (e.g., bur- rowing owls) whose adults are predators upon the same snakes (e.g., gopher snakes, sidewinders). Finally, note that only the golden eagle approaches the status of top predator (i.e., a species withoutpredators). However, even the golden eagle may not be a truetop predator:at otherlocations, gopher snakes eat golden eagle eggs, and parasiticthrichomoniasis and siblingfratricide/cannibalism cause nestlingmortality (30% and 8%, respectively;Olendoroff 1976; Palmer 1988). Thus, here is a communityand food web withpotentially no top predators.

DISCUSSION Sandy areas in the Coachella Valley are the habitatfor about 175 species of vascular plants, 100 vertebrates,and thousandsof arthropods,parasites, and soil organisms.These species forma communityconnected trophicallyinto a single food web. Althoughonly a few componentsof the CV web were presented,it is possible to summarizeseveral general trends. 1. Each subweb is complex because of the large numberof interactivespecies, age structure,and highomnivory. A web representingall species would increase complexityeven more. 2. Age structureis central to complexity.Growth necessitates and/orallows changes in diet (life-historyomnivory), either gradually or radically(e.g., predaceous larvae, herbivorousadults). Predatorsalso change with size. 3. Differentmicrohabitats and timesare connectedtrophically. At firstinspec- tion, distinctcompartments appear (e.g., diurnalvs. nocturnal,surface vs. sub- surface). However, extensive crossover exists. Foragingtimes change fromday to nightas a functionof temperature;nocturnal species eat diurnalprey (e.g., DESERT FOOD WEBS 143 scorpions on robberflies).Plants and detritusare eaten during all periods by species thatlive above and below the surface.Arthropods feeding below ground export energy when they become surface-dwellingadults (e.g., tenebrionids). "Different-channelomnivores" linkcompartments by feedingfrom different subwebs or "energy channels" (Moore et al. 1988). Consumerseat food types rather than specialize on trophiclevels (Polis et al. 1989). Such connectionsdecompart- mentalizewebs and increase complexity. The CV is not unique in its diversityor propertiesof its food web. Otherdeserts are also characterized by thousands of species (Polis 1991b). The same suite of plants, herbivores,detritivores, arthropodivores, parasitoids, parasites, and carnivores is present in all deserts. In particular,the universalexistence of diverse assemblages of predaceous arthropods(Polis and Yamashita 1991) and soil organisms (Freckman and Zak 1991) must contributeto a trophic complexity similarto thatof the CV. Further,omnivory is normalamong desert consumers (Noy-Meir 1974; Orians et al. 1977; Seely and Louw 1980; Brown 1986). R. Bradley (1983) also illustratesthe complexityin deserts witha source web focused on predatorsof camel cricketsin sandy areas of the Chihuahuandesert. These predators (scorpions, solpugids, burrowingowls, grasshoppermice, and pallid bats) are quite omnivorous(only 2 of 27 species pairs are noninteractive, i.e., not linked as predator or prey). Closed-loop omnivoryoccurs for every predator.Looping is common: six cases of mutualpredation occur; six of eight species are cannibalistic.Finally, at least 18 three-speciesloops exist. Comparison withPutblished "Empirical Generalizationis'' Theorists have produced a series of generalizationsderived fromcatalogs of publishedwebs (see App. A). These generalizationsare enteringaccepted ecological literature(see May 1986, 1988; Lawton and Warren 1988; Lawton 1989; Schoener 1989). The food web of the CV offerslittle support for these patterns. Web patternsfrom the CV are now compared withthose frompublished catalogs of webs. An argumentcould be made thatCV web statisticsshould not be calculated. Even with its complexity,this web is vastly incompleteand literally scores of other subwebs could be presented.Statistics would representonly the level of complexityI arbitrarilypresent rather than true web parameters.How- ever, to facilitatecomparisons, I present a highlysimplified web/matrix of the CV community(App. B). The thousands of species are heavily lumped into 30 "kinds of organisms" forming22,220 chains. Table 3 compares 19 web statistics fromthe CV with those frompublished catalogs of webs (Pimm 1982; Briand 1983; Cohen et al. 1986; Schoener 1989; Schoenly et al. 1991). In the discussion below, I use the statisticsfrom the simplifiedweb and those from the more complete web depictionpresented throughout the text. 1. The numberof interactivespecies in the CV web is two ordersof magnitude greaterthan the average number(17.8, Briand 1983; 16.7, Cohen et al. 1986; 24.3, Schoenly et al. 1991) from webs analyzed by theorists. In fact, Briand and Cohen's most species-richweb containsonly 48 species, less diverse thanany of the followingCV taxa: plants, nematodes, mites,arachnids, bees, beetles, bombyliidflies, and birds. 144 THE AMERICAN NATURALIST

TABLE 3

COMPARISONOF STATISTICSFROM THE COACHELLAVALLEY FOOD WEB WITH MEANS OF THE STATISTICSFROM CATALOGED WEBS

Cohen et al. Schoenly Coachella* and Briand et al.

Total numberof "kinds" or species (S) 16.7 24.3 30 Total numberof links per web (L) 31 43.1 289 Number of links per species (L/S) 1.99 2.2 9.6 Number of prey per predator 2.5 2.35 10.7 Number of predatorsper prey 3.2 2.88 9.6 Total prey taxa/totalpredator taxa .88 .64 1.11 Minimumchain length 2.22 1 3 Maximumchain length 5.19 7 12t(18) Mean chain length 2.71, 2.86 2.89 7.34t Connectance (C = L/S[S - 1]/2) .25 .49t Species x connectance (SC) 2-6 4.3 14.7t Basal species (%) 19 16 10 Intermediatespecies (%) 52.5 38 90 Top predators(%) 28.5 46.5 0 Primarilyor secondarilyherbivorous (%) 14.6 60 Primarilyor secondarilysaprovorous (%) 21 35.5 37 Omnivorous(%) 27 22 78 Consumers with 'self-loop" (%) <1.0 <1.0 74 Consumer-swith mutualpredation loop (%) << 1.0 << 1.0 53

SOURCES.-Briand 1983; Cohen et al. 1986. See also Schoener 1989; Schoenly et al. 1991. * The Coachella statisticsare derived fromthe extremelysimplified and highlylumped food-web matrixpresented in App. B. t Mutual links (loops) were not used to calculate the statistic.

2. Coachella Valley chain lengthsaverage more linksthan 2.86 (Briand 1983), 2.71 (Cohen et al. 1986), or 2.89 (Schoenly et al. 1991). Even excludinglooping and parasites, lengthsof 6 to more than 11 are common. Both J. E. Cohen (personal communication)and I interpretshort chain lengthas an artifactof totally inadequate descriptionsof real communities.Short average lengthsare derived fromcatalogs biased towardless complex, vertebrate-centeredwebs. Chains are lengthenedin the CV primarilyduring trophic interactions among the soil biota, the arthropods,and intermediatelevel predators. Webs includinginvertebrates are typicallymore complex than those centeringon vertebrates(see 10 below). Shorterchains exist (e.g., plant-rabbit-eagle),but these are much less frequent simplybecause vertebratesform a relativelysmall proportionof all species when arthropodsand soil biota are not ignored.For example, chains containingplants, insect herbivores,and insect parasitoidsare estimatedto containover halfof all known metazoans (Hawkins and Lawton 1987). The average lengthof all chains in the simplifiedCV web is 7.3; its maximumchain lengthis 18 (12 withno loops). 3. Omnivoryis frequentin the CV web but statistically"rare" in cataloged webs (Pimm 1982; Yodzis 1984). In cataloged webs, 22% (Schoenly's catalog) to 27% (Briand and Cohen's catalog) of all "kinds" are omnivorous; 78% in the simplifiedCV web are omnivorous.Adequate diet data, not lumpingarthropods, and the inclusion of age structurepartially explain the ubiquityof omnivoryin the CV. It is possible that long chains in the CV exist because energyto top DESERT FOOD WEBS 145

consumerscomes frommany (lower) trophiclevels in additionto adjacent upper levels (see also Sprules and Bowerman 1988). 4. Loops are "(unreasonable" to modelers and purportedto be ""veryrare in terrestrial"webs (Pimm 1982; Pimm and Rice 1987; Cohen et al. 1990). Looping also violates the assumptionthat body-size relationsarrange species along a cascade or hierarchy,such that a species preys on only those species below it and is preyed on only by those above it (""uppertriangular web structure";Warren and Lawton 1987; Lawton 1989; Cohen et al. 1990). However, looping is neither rare nor abnormal in the CV. Most frequently,loops are a productof age structure: cannibalismand mutualpredation usually resultwhen large individualseat smalleror youngerindividuals. Other factors also produce mutualpredation (e.g., grouppredation by ants). In the simplifiedcommunity matrix, 74% of the consumers show self-loops;44% are involved in loops withanother ""kind." 5. Coachella Valley species interactwith many more species than those in cataloged webs. Cohen (1978) and Schoenly et al. (1991) calculated the number of predatorson each prey (mean = 3.2 and 2.88, respectively)and the number of prey per predator (2.5 and 2.35). Overall, cataloged species interactdirectly with3.2-4.6 other species (Cohen et al. 1985). Parametersfrom the CV are one to two orders of magnitudegreater. Higher values exist firstbecause most CV consumers eat many species (e.g., see table 1; diets range from15 to more than 125 items; a few arthropodherbivores and some parasites are exceptions). Sec- ond, most individual species are eaten by tens to hundreds(e.g., mesquite) of species. Such discrepancies occur because cataloged webs lump several species into ""kinds" and diet data are grossly inadequate. However, even the highly lumped CV web shows thateach ""kind"is eaten by about 10 predatorsand each consumereats about 10 prey (table 3). 6. Top predatorsare rare or nonexistentin the CV. Coyotes, kitfoxes, horned owls, and golden eagles (the largestpredators in the CV) sufferthe fewestpreda- tors,but each is the reportedprey of otherspecies. This findingstands in marked contrastwith cataloged webs: 28.5% (Briand and Cohen 1984; Cohen et al. 1990) to 46.5% (Schoenly et al. 1991) of kinds were top predators.This great discrep- ancy is undoubtedlydue to the inadequacy of diet informationin cataloged webs and/orto the fact thatthese webs only focus on a limitedsubset of a trophically linked community. 7. Coachella Valley data pose great difficultyto the observation that prey/ predatorratios are greaterthan 1.0 (0.88, Briand and Cohen 1984; 0.64, Schoenly et al. 1991); thatis, the numberof organismsheading rows (prey) in web matrices is less than the numberheading columns (predators).It is easy to show that the ratio in the CV and other real communitiesshould be greaterthan 1.0. Because all heterotrophsmust obtain food, every animal should head a column. Rows (prey) include plants, detritus,and all animals except those with no predators (i.e., top predators). Let x be the numberof animal species thatare intermediate predators (i.e., both predatorand prey). Then the total numberof prey is x + the numberof plant species (174 in the CV) + the numberof categoriesof detritus and carrion; the numberof predatorsis x + the numberof top predators(0 or 1 in the CV). If more species of plants exist than top predators,then the prey/ 146 THE AMERICAN NATURALIST

predatorratio will always be greaterthan 1.0. Few real (no?) communitieshave more top predatorsthan autotrophs.The appearance thattop predatorsare more frequentis an artifact(see 6 above). It appears thatlumping obliterates the actual ratioprimarily because more species of plantsare lumpedthan (easily recognized) animals thatare top predators.The CV communityweb exhibitsa prey/predator ratio of 1.11. Glasser (1983), Paine (1988), and Lockwood et al. (1990) also criti- cize prey/predatorratios. 8. Factors 1-7 make the CV web much more complex than cataloged webs. For example, the numberof trophiclinks varies from31 (Cohen et al. 1986) to 43 (Schoenly et al. 1991); only 2 of Cohen et al.'s (1986) 113 webs had more than 100 links. The average CV subweb (figs.2-7) has 54.7 links; the carnivoresubweb alone, 107; and the simplifiedcommunity web, 289. 9. The CV web questions the utilityof the concept of "trophic level." A trophic level is definedas a set of organismswith a common numberof chain links between themand primaryproducers. The nearlyuniversal presence of omnivory and age structuremakes this concept nonoperational.What trophic level are consumers that ontogenetically,seasonally, or opportunisticallyeat all trophic levels of arthropodsin additionto plantmaterial and (forcarnivores) vertebrates? Looping furtherblurs the concept. If A eats B but B eats A, is B on the first, third,or (afteranother loop) fifthtrophic level (ad infinitum)?I am not alone in criticismof this concept (see Gallopin 1972; Cousins 1980, 1987; Levine 1980; Lawton 1989). 10. Patterns 4 and 5 in Appendix A are confirmedby the CV web. First, separate compartmentsdid not exist within one habitat. Second, arthropod- dominatedsystems are morecomplex thanthose dominatedby vertebrates.How- ever, few communitiesare not dominated(in numberof individualsor species) by arthropods(Hawkins and Lawton 1987; May 1988). So not lumpingarthropods, the most species-richtaxon on thisplanet, should increase the complexity of any web, includingthose cataloged by theorists. Overall, a general lack of agreementexists between patternsfrom the CV and those fromcatalogs of published webs. Is the CV web unique or are cataloged webs so simplifiedthat they have lost realism? That cataloged webs depict so few species, such low ratios of predatorson prey and prey eaten by predators, so few links, so littleomnivory, a veritableabsence of looping, and such a high proportionof top predatorsargues stronglythat they do not adequately describe real biological communities.Taylor (1984), Paine (1988), Lawton (1989), and Winemiller(1990) reach a similarconclusion. This conclusion carries importantimplications. First, controversies over such issues as the causes of shortchain lengthand omnivory,complexity versus stabil- ity of ecosystems, and the role of dynamics versus energeticsin shaping web patterns(see DeAngelis et al. 1983; Yodzis 1984; Lawton 1989) are based on patternsabstracted from cataloged webs. If catalogs are an inadequate data base, these controversiesmay be nonissues and theoristsare tryingto explain phenom- ena that do not exist. This is a real possibility(see Lawton and Warren 1988; Paine 1988; Lawton 1989; Winemiller1990). Second, characteristicsof the CV web (complexity,omnivory, long chain lengths,looping, absence of compart- DESERT FOOD WEBS 147 ments)considerably reduce stabilityin dynamicmodels of food webs (see Pimm 1982; Pimmand Rice 1987). These simplifiedmodels apparentlytell us littleabout the structureof food webs in nature.

Self-Critiquie and Prospectuis Several issues need to be addressed. 1. Should all trophiclinks be included,or are some too weak or unusual to list (May 1983a, 1986; Paine 1988; Lawton 1989; Schoener 1989)? I includedall links in the CV web. This decision was based on fourfactors. First, it is arbitraryand impossible to evaluate which links are and which are not important.Most CV consumers eat 20 to more than 50 items. Which should be included, which ex- cluded? Some consumers (esp. in deserts) likely exist because they are suffi- cientlyflexible to include a numberof infrequentlinks that sum intoan important source of energy,at least duringsome periods. For example, the seasonal switch by granivorousbirds to arthropodsprovides proteinand water for nestlingsand may allow permanentresidence forsome desertbirds (Welty 1962; Brown 1986). Second, dynamics and trophic linkage are not necessarily correlated (Paine 1988; Lawton 1989). A 1% frequencyof a rare species in the diet of a common species may produce high mortality;conversely, a rare species eating only one common species may scarcely affectdynamics. Further, short but intensivepre- dation events may contributelittle to the diet of a predatorbut may be centralto prey dynamics(e.g., Wilburet al. 1983; Polis and McCormick 1987). Third,each linkenriches the descriptionof the community(the originalpurpose of food webs). We must not blur the distinctionbetween food webs and interaction/functionalwebs. Food webs (such as the CV web) are based on observations of trophicrelationships while interactionwebs summarizethe subset of all interactions that are "strong." This insightargues that all trophiclinks should be included because this is a food web, not an interactionweb. No justification exists forexcluding links from a food web. Fourth,the exclusion of certainlinks produces systematicbias against complexity. 2. Are desert food webs (a)typical of other webs? Deserts may differin two main ways fromother systems. First, deserts are relativelysimple ecosystems characterized by low productivityand low species richness (Noy-Meir 1974; Seely and Louw 1980; Louw and Seely 1982; Whitford1986; Polis 1991b). Should such depauperate communitiestranslate into relativelysimple webs? If so, the complexityof the CV web is much less than thatof more species-richsystems. Second, is omnivorymore common in deserts than in other habitats? It is impossible to answer this question with rigor.I can only indicate that features thatpromote omnivory (and web complexity)in the CV are presentin all systems: age structure,life-history omnivory, predators that ignore the feedinghistory of prey (different-channelomnivory), opportunistic feeding, and consumersthat eat food in which other consumers live (e.g., scavengers, frugivores,granivores, detritivores;Polis et al. 1989). High omnivoryis not restrictedto deserts. Price (1975) and Cousins (1987) argue thatit is normalthroughout the animal kingdom. Omnivorycharacterizes feeding in aquatic and marinehabitats (Menge and Suth- erland 1987). These authors and Walter (1987) maintainthat omnivoryis also 148 THE AMERICAN NATURALIST commonin terrestrialcommunities. Walter and Moore et al. (1988) provide strong evidence that omnivoryis one of the most frequentand dynamicallyimportant trophiclinks among soil and detritalwebs. Sprules and Bowerman (1988) analyzed zooplankton in 515 lakes and concluded thatomnivory is common. (They also note long chains, looping, cannibalism,and mutualpredation.) Regardless of whetherdeserts are unique, desertfood webs are stillof general importance.Deserts occupy at least one-quarterof theearth's land surface(Craw- ford 1981; Polis 1991b), and patternsobserved in the CV thus should describe a good fractionof the terrestrialcommunities on this planet. 3. Not all diet data for Coachella Valley species came fromthe Coachella. How this affectsthe web is uncertain.However, the overall conclusion of great complexityshould not be influencedunduly by errorsarising from use of these studies. 4. The sections of this articlethat critique "food-web theory" tend to be nihil- istic. A more satisfyingarticle would furthersuggest how we can replace or revise these theories. Because of space limitations,I cannot focus on futuredirections other than to offerthe general caveat that we cannot overlook complexityand realityin our attemptsto comprehendnature. However, thisarticle suggests that a host of importantissues need to be resolved. Foremost, we need criteriato recognize which links are "sufficientlyimportant" to include in descriptionsand analyses of communities.We mustestablish more uniformly the spatial limitsand species membershipof a communityand not analyze webs as differentas those foundin catalogs (Arctic seas to toad carrion;Briand 1983; Sugihara et al. 1989). We need to assess what easily studied subwebs (e.g., phytotelmata,plant galls, carrion)tell us about the structureof communitywebs and how isolated or connected these subwebs are with respect to the rest of the community.We must determinehow omnivoryaffects the recognitionof importantlinks and how the ubiquity of omnivoryaffects web structure.Finally, we must incorporateage structure,changes over time and space, and recyclingvia the detritalloop into web analyses.

CONCLUSIONS It appears that much "food-web theory" is not very descriptiveor predictive of nature. The catalogs of webs used to abstractempirical generalizations were derivedfrom grossly incomplete representations of communitiesin termsof both diversityand trophicconnections. Consequently,theorists have constructedan oversimplifiedand invalidview of communitystructure. The inherentcomplexity of natural communitiesmakes web constructionby empiricistsand analysis by theoristsdifficult. So what good are food-webrepresentations and analyses? At a minimum,they are descriptorsof communitiesand trophicinteractions. As a futuregoal, they may tell us how communitiesfunction and are assembled. In an ideal world, hypothesesand generalizationsmade by theoristscould be viewed as a stage in the evolution of understandingfood webs; they are definitelynot :finishedprod- ucts (J. E. Cohen, and S. L. Pimm,personal communication). Theory is designed DESERT FOOD WEBS 149 to provoke concept-focusedfieldwork that hopefully promotes the next iteration of descriptivemodels and generalizationsthat will, in turn,encourage more em- piricalwork, all the while approachingmore accurate predictionsand descriptions of reality. However, a strongimplication of my analysis is that many existing hypothesesand generalizationsare simplywrong. For example, muchtheory (see Pimm 1982; DeAngelis et al. 1983; May 1983a; Lawton and Warren 1988) would hold that the Coachella Valley food web should be completelyunstable. To ad- vance our understanding,an adequate data base of communityfood webs must be assembled, experimentsmust be conducted, and new questions asked. This will take time. Only then will a useful, heuristictheoretical framework be con- structed.Meanwhile, we should reevaluate where we now stand: it appears that much food-web theoryis in criticalneed of revisionand new direction.

ACKNOWLEDGMENTS Many people contributedideas, data, and energyduring the 10-yrgestation of this paper. S. McCormick Carter was central to its development. S. Frommer, W. Icenogle, W. Mayhew, J. Pinto, K. Sculteure, R. Ryti, A. Muth, and F. Andrews provided data on the Coachella. The manuscriptbenefited greatly from suggestionsby J. Cohen, C. Crawford,R. Holt, J. Lawton, B. Menge, J. Moore, C. Myers, S. Pimm, K. Schoenly, T. Schoener, and the anonymous reviewers. A special thanks to K. Schoenly for calculating many of the web statisticsin table 4. Fieldworkwas financedpartially by the National Science Foundationand the Natural Science Committeeand Research Council of VanderbiltUniversity. APPENDIX A

A PARTIAL SUMMARY OF 'FEATURES OBSERVED IN REAL FOOD WEBS" These "empiricalgeneralizations" were derived by food-web theorists from the catalog of publishedwebs assembledby Cohen(1978), Briand (1983), and Schoenlyet al. (1991; see text).All quoted phrasesare fromchapter 10 in Pimm1982 (see esp. table 10.1). Romannumerals in parenthesesrefer to patternnumber in Pimm'stable 10.1. 1. Chainlengths are limitedto "typicallythree or four"(2.86-3.71) trophic levels (iii) (Pimm1982; Briand 1983; Cohen et al. 1986;Schoener 1989; Schoenly et al. 1991). 2. Omnivoresare statistically"rare" (iv) (Pimm1982; Yodzis 1984). 3. Omnivoresfeed on adjacenttrophic levels (v) (Pimm1982). 4. Insectsand theirparasitoids are exceptionsto patterns2 and 3 (vi) (Pimm1982; Hawkins and Lawton1987). 5. Webs are usuallycompartmentalized between but not within habitats (viii) (Pimm1982). 6. Loops are rareor nonexistentand do notconform to "biologicalreality" (i) (Gallopin1972; Cohen 1978;Pimm 1982; Pimm and Rice 1987;Lawton and Warren 1988;Cohen et al. 1990). 7. The ratioof preyspecies to predatorspecies is greaterthan 1.0 (0.64-0.88)(x) (Cohen 1978;Briand and Cohen 1984;Schoenly et al. 1991). 8. The proportionof speciesof top predatorsto all speciesin a communityaverages 0.29-0.46 (Briandand Cohen 1984;Schoenly et al. 1991). 9. Species interactdirectly (as predatoror prey)with only 2-6 otherspecies (Cohen 1978;Cohen et al. 1985;Schoenly et al. 1991). APPENDIX B TABLE Bi

SUMMARY FOOD-WEB MATRIX FOR THE COACHELLA VALLEY SAND COMMUNITY

4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30

2 ~ ~

5/ /

6/11 / I/ I / 11111 / I/ I/ II

10 / / I / / I I / /

12 / / I//I I /

13 1 I /

14 / / / I I/I 15 / / I / / I I / / 16 I / / I I I I / /

17/ / / / // /

18/ / / / // /

19/ / / / // /

20// / / /

21 1 22II I I I 23II/

24 I I / I I I I

25 I I / I I I I I

26 I / I I I I I I DESERT FOOD WEBS 151

TABLE B1 (Continuled)

4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30

29 1 1

NOTE.-The thousands of species were lumpedextensively to 30 "kinds of organisms" in order to facilitatecomparison with webs containedin the various catalogs. Entriesalong the diagonal indicate a "self-loop" withinthat kind of organism.Entries below the diagonal are loops caused by mutual predation.Some statisticsfrom this matrixare summarizedin table 3. The kinds of organismsare as follows: 1, plants/plantproducts; 2, detritus;3, carrion;4, soil microbes;5, soil microarthropodsand nematodes; 6, soil micropredators;7, soil macroarthropods;8, soil macroarthropodpredators; 9, surface arthropoddetritivores; 10, surfacearthropod herbivores; 11, small arthropodpredators; 12, mediumarthropod predators; 13, large arthropodpredators; 14, facultativearthropod predators; 15, life-historyarthropod omnivore; 16, spider parasitoids; 17, primaryparasitoids; 18, hyperparasitoids; 19, facultativehyperparasitoids; 20, herbivorousmammals and reptiles; 21, primarilyherbivorous mammalsand birds; 22, small omnivorousmammals and birds; 23, predaceous mammalsand birds; 24, arthropodivoroussnakes; 25, primarilyarthropodivorous lizards; 26, primarilycarnivorous lizards; 27, primarilycarnivorous snakes; 28, large, primarilypredaceous birds; 29, large, primarilypredaceous mammals; and 30, golden eagle.

LITERATURE CITED

Andersen, D. 1987. Below groundherbivory in naturalcommunities: a review emphasizingfossorial animals. QuarterlyReview of Biology 62:261-286. Andrews, F., A. Hardy, and D. Giuliani. 1979. The coleopterous fauna of selected Californiasand dunes. CaliforniaDepartment of Food and AgricultureReport, Sacramento, Calif. Askew, R. 1971. Parasitic insects. Elsevier, New York. Bradley, R. 1983. Complex food webs and manipulativeexperiments in ecology. Oikos 41:150-152. Bradley, W. 1968. Food habits of the antelope groundsquirrel in southernNevada. Journalof Mam- malogy49:14-21. Briand, F. 1983. Environmentalcontrol of food web structure.Ecology 64:253-263. Briand, F., and J. Cohen. 1984. Communityfood webs have invariant-scalestructure. Nature (Lon- don) 5948:264-267. Brown,J. 1986. The role of vertebratesin desertecosystems. Pages 51-71 in W. Whitford,ed. Pattern and process in desert ecosystems. Universityof New Mexico Press, Albuquerque. Brown, J., 0. J. Reichman, and D. Davidson. 1979. Granivoryin desertecosystems. Annual Review of Ecology and Systematics 10:201-227. Cohen, J. E. 1978. Food webs and niche space. Monographs in Population Biology 11. Princeton UniversityPress, Princeton,N.J. Cohen, J. E., and C. Newman. 1985. A stochastictheory of communityfood webs. I. Models and aggregated data. Proceedings of the Royal Society of London B, Biological Sciences 224:421-448. Cohen, J. E., C. Newman, and F. Briand. 1985. A stochastic theoryof communityfood webs. II. Individual webs. Proceedings of the Royal Society of London B, Biological Sciences 224:449-461. Cohen, J. E., F. Briand, and C. Newman. 1986. A stochastictheory of communityfood webs. III. Predictedand observed lengthsof food chains. Proceedingsof the Royal Society of London B, Biological Sciences 228:317-353. 1990. Communityfood webs: data and theory.Springer, Berlin. Cousins, S. 1980. A trophic continuumderived from plant structure,animal size and a detritus cascade. Journalof Theoretical Biology 82:607-618. 152 THE AMERICAN NATURALIST

. 1987. The decline of the trophiclevel concept. Trends in Ecology & Evolution 2:312-316. Crawford,C. S. 1981. Biology of desert invertebrates.Springer, New York. 1986. The role of invertebratesin desertecosystems. Pages 73-92 in W. Whitford,ed. Pattern and process in desert ecosystems. Universityof New Mexico Press, Albuquerque. 1991. Macroarthropoddetritivores. Pages 89-112 in G. A. Polis, ed. Ecology of desertcommunities. Universityof Arizona Press, Tucson. DeAngelis, D., W. M. Post, and G. Sugihara. 1983. Currenttrends in food web theory:report on a food web workshop. Technical Memo 5983. Oak Ridge National Laboratory, Oak Ridge, Tenn. Edney, E., J. McBrayer, P. Franco, and A. Phillips. 1974. Distributionof soil arthropodsin Rock Valley, Nevada. InternationalBiological Program,Desert Biome Research Memorandum 74-32:53-58. Elgar, M. A., and B. J. Crespi. In press. The ecology and evolutionof cannibalism.Oxford University Press, Oxford. Ferguson, W. 1962. Biological characteristicsof the mutellidsubgenus Photopsis Blake and their systematicvalue. Universityof CaliforniaPublications in Entomology27:1-82. Franco, P., E. Edney, and J. McBrayer. 1979. The distributionand abundance of soil arthropodsin the northernMojave Desert. Journalof Arid Environments2:137-149. Freckman, D., and J. Zak. 1991. Desert soil communities.Pages 55-88 in G. A. Polis, ed. Ecology of desert communities.University of Arizona Press, Tucson. Frommer,S. I. 1986. A hierarchiclisting of the arthropodsknown to occur withinthe Deep Canyon Desert Transect. Deep Canyon, Riverside,Calif. Gallopin, G. 1972. Structuralproperties of food webs. Pages 241-282 in B. Patton, ed. Systems analysis and simulationsin ecology. Vol. 2. Academic Press, New York. Ghilarov,M. 1964. Connectionof insectswith soil in differentclimatic zones. Pedobiologia 4:310-315. Gier, H., S. Kruckenberg,and R. Marler. 1978. Parasites and diseases of coyotes. Pages 37-69 in M. Bekoff,ed. Coyotes: biology,behavior and management.Academic Press, New York. Glasser, J. 1983. Variationin niche breadthand trophicposition: the disparitybetween expected and observed species packing. AmericanNaturalist 122:542-548. Gordon, S. 1978. Food and foragingecology of a desertharvester ant, Veromessorpergandei (Mayr). Ph.D. diss. Universityof California,Berkeley. Hawkins. B., and R. Goeden. 1984. Organizationof a parasitoidcommunity associated witha complex of galls on Attriplexspp. in southernCalifornia. Ecological Entomology9:271-292. Hawkins, B., and J. Lawton. 1987. Species richnessof parasitoidsof Britishphytophagous insects. Nature (London) 326:788-790. Janzen, D. 1977. Why fruitsrot, seeds mold, and meat spoils. AmericanNaturalist 111:691-713. Johnegard,P. 1988. NorthAmerican owls: biology,and naturalhistory. Smithsonian Institute Press, Washington,D.C. Johnson,M., and R. Hansen. 1979. Coyote food habitson the Idaho National EngineeringLaboratory. Journalof WildlifeManagement 43:951-956. Lawton, J. 1989. Food webs. Pages 43-78 in J. Cherrett,ed. Ecological concepts. Blackwell Scientific, Oxford. Lawton, J., and P. Warren. 1988. Static and dynamicexplanation of patternsin food webs. Trends in Ecology & Evolution 3:242-245. Levine, S. 1980. Several measures of trophicstructure applicable to complex food webs. Journalof Theoretical Biology 83:195-207. Lockwood, J., T. Christiansen,and D. Legg. 1990. Arthropodprey-predator ratios in a sagebrush habitat: methodologicaland ecological implications.Ecology 71:996-1005. Louw, G., and M. Seely. 1982. Ecology of desert organisms.Longmans, New York. Macfayden, A. 1963. The contributionof the fauna to the total soil metabolism.Pages 3-17 in J. Doeksen and J. van der Drift,eds. Soil organisms.North-Holland, Amsterdam. MacKay, W. 1991. Ants and termites.Pages 113-150 in G. A. Polis, ed. Ecology of desertcommunities. Universityof Arizona Press, Tucson. Mann, J. 1969. Cactus feedinginsects and mites. SmithsonianInstitution of the United States, Na- tional Museum Bulletin256:1-158. DESERT FOOD WEBS 153

May, R. 1983a. The structureof food webs. Nature (London) 301:566-568. . 1983b. Parasitic infectionsas regulatorsof animal populations. AmericanScientist 71:36-45. . 1986. The search for patterns in the balance of nature: advances and retreats. Ecology 67:1115-1126. . 1988. How many species are there on earth'?Science (Washington,D.C.) 241:1441-1448. Mayhew, W. W. 1983. Vertebratesand theirhabitats on the Deep Canyon Transect. Deep Canyon, Riverside, Calif. McKinnerney,M. 1978. Carrioncommunities in the northernChihuahuan desert. Southwester-nNatu- ralist23:563-576. Menge, B., and J. Sutherland. 1987. Communityregulation: vairiation in disturbance,competition, and predation in relation to environimentalstress and recruitment.American Naturalist 130:730-757. Mispagel, M. 1978. The ecology and bioenergeticsof the acridid grasshopper,Bootettix pulnctatis, on creosotebush, Larrea tridenitata-,in the northernMojave Desert. Ecology 59:779-788. Moore, J., D. Walter,and H. Hunt. 1988. Arthropodregulation of microbiotaand mesobiota in below grounddetrital webs. Annual Review of Entomology33:419-439. Noy-Meir,I. 1974. Desert ecosystems: highertrophic levels. Annual Review of Ecology and System- atics 5:195-213. Nuessly, G., and R. Goeden. 1984. Aspects of the biology and ecology of Digiletia lflo/(loe(l Gertsch (Araneae, Diguetidae). Journalof Arachnology12:75-85. Odum, E., and L. Biever. 1984. Resource quality,mutualism, and energypartitioning in food chains. American Naturalist 124:360-376. Ohlendorf,H. 1971. Arthropoddiet of westernhorned owl. SouthwesternNaturalist 16:124-125. Olendoroff,R. 1976. The food habits of NorthAmerican golden eagles. AmericanMidland Naturalist 95:231-236. Orians, G., R. Cates, M. Mares, A. Moldenke, J. Neff,D. Rhoades, M. Rosenzweig, B. Simpson, J. Schultz, and C. Tomoff. 1977. Resource utilizationsystems. Pages 164-224 in G. Orians and 0. Solbrig, eds. Convergentevolution in warm deserts. Dowden, Hutchinson& Ross, Stroudsburg,Pa. Paine, R. T. 1980. Food webs: linkage,interaction strength and communityinfrastructure. Journal of Animal Ecology 49:667-685. 1988. On food webs: road maps of interactionor gristfor theoretical development? Ecology 69:1648-1654. Palmer, R. 1988. Handbook of North Americanbirds. Vol 5. Diurnal raptors.Pt. 2. Yale University Press, New Haven, Conn. Pimm, S. L. 1982. Food webs. Chapman & Hall, New York. Pimm,S. L., and J. Rice. 1987. The dynamicsof multispecies,multi-life-stage models of aquatic food webs. Theoretical Population Biology 32:303-325. Polis, G. A. 1981. The evolutionand dynamicsof intraspecificpredation. Annual Review of Ecology and Systematics 12:225-251. 1984. Age structurecomponent of niche widthand intraspecificresource partitioning:can age groups functionas ecological species? American Naturalist123:541-564. 1988. Exploitation competitionand the evolutionof interference,cannibalism and intraguild predationin age/size structuredpopulations. Pages 185-202 inlL. Perrsonand B. Ebenmann, eds. Size structuredpopulations: ecology and evolution. Springer,New York. 1991a. Food webs in desert communities: complexity, diversityand omnivory. Pages 383-438 in G. A. Polis, ed. Ecology of desert communities.University of Arizona Press, Tucson. 1991b. Desert communities:an overview of patternsand processes. Pages 1-26 inl G. A. Polis, ed. Ecology of desert communities.University of Arizona Press, Tucson. Polis, G. A., and S. McCormick. 1986a. Patternsof resource use and age structureamong species of desert scorpion. Journalof Animal Ecology 55:59-73. 1986b. Scorpions, spiders and solpugids: predationand competitionamong distantlyrelated taxa. Oecologia (Berlin) 71:111-116. 1987. Intraguildpredation and competitionamong desert scorpions. Ecology 68:332-343. 154 THE AMERICAN NATURALIST

Polis, G. A., and T. Yamashita. 1991. Arthropodpredators. Pages 180-222 in G. A. Polis, ed. Ecology of desert communities.University of Arizona Press, Tucson. Polis, G. A., W. D. Sissom, and S. McCormick. 1981. Predatorsof scorpions:field data and a review. Journalof Arid Environments4:309-327. Polis, G. A., C. A. Myers, and R. Holt. 1989. The evolution and ecology of intraguildpredation: competitorsthat eat each other. Annual Review of Ecology and Systematics20:297-330. Powell, J., and C. Hogue. 1979. Californiainsects. Universityof CaliforniaPress, Berkeleyand Los Angeles. Price, P. 1975. Insect ecology. Wiley, New York. Rich, P. 1984. Trophic-detritalinteractions: vestiges of ecosystem evolution. American Naturalist 123:20-29. Rodin, L., and N. Bazilevich. 1964. Doklady Akademii Nauk SSSR 157:215-218 (fromE. J. Kor- mandy, 1969, Concepts of ecology, Prentice-Hall,New York). Ryan, M. 1968. Mammals of Deep Canyon. Desert Museum, Palm Springs,Calif. Ryti,R., and T. Case. 1986. Overdispersionof ant colonies: a test of hypotheses.Oecologia (Berlin) 69:446-453. 1988. Field experimentson desert ants: testingfor competitionbetween colonies. Ecology 69:1993-2003. Santos, P., J. Phillips, and W. Whitford.1981. The role of mites and nematodes in early stages of litterdecomposition in the desert. Ecology 62:664-669. Schoener, T. W. 1989. Food webs from the small to the large: probes and hypotheses. Ecology 70:1559-1589. Schoenly, K., and W. Reid. 1983. Communitystructure in carrion arthropodsin the Chihuahuan desert. Journalof Arid Environments6:253-263. Schoenly, K., R. Beaver, and T. Heumier. 1991. On the trophicrelations of insects: a food web approach. American Naturalist137:597-638. Schultz, J., D. Otte, and F. Enders. 1977. Larrea as a habitatcomponent for desert arthropods. Pages 176-208 in T. Mabry,J. Hunziker,and D. DiFeo, eds. Creosote bush: biologyand chemistry of Larrea in new world deserts. US/IBP (InternationalBiological Program)Synthesis Ser. 6. Dowden, Hutchinson& Ross, Stroudsburg,Pa. Seastedt, T., R. Ramundo, and D. Hayes. 1988. Maximizationof densitiesof soil animals by foliage herbivory:empirical evidence, graphicaland conceptual models. Oikos 51:243-248. Seely, M., and G. Louw. 1980. First approximationof the effectsof rainfallon the ecology and energeticsof a Namib Desert dune ecosystem. Journalof Arid Environments3:25-54. Sprules, W., and J. Bowerman. 1988. Omnivoryand food chain lengthsin zooplanktonfood webs. Ecology 69:418-426. Sugihara, G., K. Schoenly, and A. Trombla. 1989. Scale invariancein food web properties.Science (Washington,D.C.) 245:48-52. Taylor, J. 1984. A partialfood web involvingpredatory gastropods on a Pacificfringing reef. Journal of ExperimentalMarine Biology and Ecology 74:273-290. Telford,S. 1970. A comparativestudy of endoparasitismamong some Californializard populations. American Midland Naturalist83:516-554. Tevis, L., and I. Newell. 1962. Studies on the biologyand seasonal cycle of the giantred velvet mite, Dinothrornbiumpandorae. Ecology 43:497-505. Wallwork,J. 1982. Desert soil fauna. Praeger,New York. Walter,D. 1987. Trophic behavior of "mycophagous" microarthropods.Ecology 68:226-229. Warren,P., and J. Lawton. 1987. Invertebratepredator-prey relationships: an explanationfor upper triangularfood webs and patternsin food web structure?Oecologia (Berlin) 74:231-235. Wasbauer, M. S., and L. Kimsey. 1985. Californiaspider wasps of the sub-familyPompilinae. Bulletin of the CaliforniaInsect Survey 26:1-130. Weathers, W. 1983. Birds of Southern California's Deep Canyon. Universityof California Press, Berkeley and Los Angeles. Welty,J. 1962. The life of birds. Saunders, Philadelphia. Werner,E., and J. Gilliam. 1984. The ontogeneticniche and species interactionsin size-structured populations. Annual Review of Ecology and Systematics15:393-426. DESERT FOOD WEBS 155

Whitford,W. G. 1986. Decomposition and nutrientcycling in deserts. Pages 93-118 in W. G. Whit- ford,ed. Patternand process in desertecosystems. Universityof New Mexico Press, Albu- querque. Wiens, J. A. 1991. Birds. Pages 278-310 in G. A. Polis, ed. Ecology of desertcommunities. University of Arizona Press, Tucson. Wilbur,H., P. Morin, and R. Harris. 1983. Salamander predationand the structureof experimental communities:anuran responses. Ecology 64:1423-1429. Winemiller,K. 0. 1990. Spatial and temporalvariation in tropicalfish trophic networks. Ecological Monographs 60:331-367. Wisdom, C. 1991. Herbivorousinsects. Pages 151-179 in G. A. Polis, ed. Ecology of desertcommunities. Universityof Arizona Press, Tucson. Woods, A., and C. Herman. 1943. The occurrence of blood parasites in birds fromsouthwestern United States. Journalof Parasitology29:187-196. Yodzis, P. 1984. How rare is omnivory?Ecology 65:321-323. Zabriskie,J. 1979. Plants of Deep Canyon and the centralCoachella Valley, California.Deep Canyon, Riverside, Calif.