please scroll down for article - el pilar af modeling set patterns lcm...modeling settlement...

PLEASE SCROLL DOWN FOR ARTICLE

This article was downloaded by: [CDL Journals Account]On: 29 May 2009Access details: Access Details: [subscription number 785022369]Publisher RoutledgeInforma Ltd Registered in England and Wales Registered Number: 1072954 Registered office: Mortimer House,37-41 Mortimer Street, London W1T 3JH, UK

Annals of the Association of American GeographersPublication details, including instructions for authors and subscription information:http://www.informaworld.com/smpp/title~content=t788352614

Modeling Settlement Patterns of the Late Classic Maya Civilization withBayesian Methods and Geographic Information SystemsAnabel Ford a; Keith C. Clarke b; Gary Raines c

a ISBER/MesoAmerican Research Center, University of California, Santa Barbara b Department ofGeography, University of California, Santa Barbara c Reno, Nevada

First Published on: 29 May 2009

To cite this Article Ford, Anabel, Clarke, Keith C. and Raines, Gary(2009)'Modeling Settlement Patterns of the Late Classic MayaCivilization with Bayesian Methods and Geographic Information Systems',Annals of the Association of AmericanGeographers,99999:1,

To link to this Article: DOI: 10.1080/00045600902931785

URL: http://dx.doi.org/10.1080/00045600902931785

Full terms and conditions of use: http://www.informaworld.com/terms-and-conditions-of-access.pdf

This article may be used for research, teaching and private study purposes. Any substantial orsystematic reproduction, re-distribution, re-selling, loan or sub-licensing, systematic supply ordistribution in any form to anyone is expressly forbidden.

The publisher does not give any warranty express or implied or make any representation that the contentswill be complete or accurate or up to date. The accuracy of any instructions, formulae and drug dosesshould be independently verified with primary sources. The publisher shall not be liable for any loss,actions, claims, proceedings, demand or costs or damages whatsoever or howsoever caused arising directlyor indirectly in connection with or arising out of the use of this material.

http://www.informaworld.com/smpp/title~content=t788352614

http://dx.doi.org/10.1080/00045600902931785

http://www.informaworld.com/terms-and-conditions-of-access.pdf

Modeling Settlement Patterns of the Late ClassicMaya Civilization with Bayesian Methods and

Geographic Information SystemsAnabel Ford,∗ Keith C. Clarke,† and Gary Raines‡

*ISBER/MesoAmerican Research Center, University of California, Santa Barbara†Department of Geography, University of California, Santa Barbara

‡Reno, Nevada

The ancient Maya occupied tropical lowland Mesoamerica and farmed successfully to support an elaboratesettlement pattern that developed over many centuries. There has been debate as to the foundation of thesettlement patterns. We show that Maya settlement locations were strongly influenced by environmental factors,primarily topographic slope, soil fertility, and soil drainage properties. Maps of these characteristics were createdat the local scale and combined using Bayesian weights-of-evidence methods to develop probabilistic maps ofsettlement distributions based on the known, but incomplete, distribution of Maya archaeological sites, bothdomestic and monumental. The predictive model was validated with independently collected point-sampledfield data for both presence and absence, predicting 82 percent of undiscovered Maya sites and 94 percent of siteabsence. This information should be of use in conservation planning for the region, which is under threat fromcontemporary agricultural expansion. Key Words: Maya settlement, predictive modeling, weights-of-evidence.

Los antiguos mayas ocuparon las tierras bajas tropicales de Mesoamerica y produjeron una agricultura tan exitosacomo para sostener un elaborado patron de poblamiento, desarrollado a lo largo de varios siglos. Se ha presentadodebate en lo que concierne a la fundacion de los patrones de aquel asentamiento. Nosotros mostramos que las loca-ciones de los asentamientos maya fueron fuertemente influidas por factores ambientales, en especial la pendientetopografica, la fertilidad del suelo y las propiedades del drenaje de los suelos. Para representar estas caracterısticas,se crearon mapas a escala local, combinandolos con el uso de metodos bayesianos de peso-de-evidencia a fin dedesarrollar mapas probabilısticos de la distribucion de los asentamientos. Se tomo en cuenta la conocida aunqueincompleta distribucion de los sitios arqueologicos maya, tanto domesticos como monumentales. El modelo pre-dictivo fue validado con datos de campo independientemente obtenidos en puntos de muestreo para presencia yausencia, pronosticando el 82 por ciento de sitios maya sin descubrir y el 94 por ciento de ausencia de sitio. Estainformacion deberıa ser utilizada para planificar la conservacion de la region, que se halla bajo la amenaza con-temporanea de la expansion agrıcola. Palabras clave: asentamientos mayas, modelacion predictiva, peso-de-evidencia.

When the ancient Maya civilization was atits height in the Late Classic, from aboutAD 600–900, a vast and complex system of

centers developed that included a tiered urban hierar-chy of major and minor centers, household farms, andremote buildings spread across the landscape. Popula-tions of the southern Yucatan Peninsula—the greaterPeten surrounding northeastern Guatemala typifying

the Maya forest (The Nature Conservancy 2007)—have been estimated to be as high as nine times that oftoday. Currently, only a small fraction of the Maya set-tlements have been discovered, mapped, and analyzed.Given this incomplete record, our research appliedmethods that use the integrative power of geographic in-formation systems (GIS) and Bayesian statistical mod-eling to predict the probable spatial distribution and

Annals of the Association of American Geographers, 99(3) 2009, pp. 1–25 C© 2009 by Association of American GeographersInitial submission, February 2006; revised submissions, March 2007 and April 2008; final acceptance, June 2008

Published by Taylor & Francis, LLC.

Downloaded By: [CDL Journals Account] At: 19:45 29 May 2009

2 Ford, Clarke, and Raines

number of Maya settlements from the subset of knownsites. The purposes of the modeling were to (1) assess theinfluence of environmental constraints on ancient LateClassic period Maya settlement and agriculture, (2) per-mit tests of the validity of the Maya settlement model inthe field, (3) allow the existing estimates of Maya pop-ulation size and density to be assessed, and (4) proposea map showing those lands with both undisturbed for-est and numerous Maya settlements that have the mostto gain by conservation. Importantly, we offer the firststrong quantitative and spatial method for linking an-cient Maya settlement patterns to ancient agriculturalpatterns.

As an agrarian society, it is well documented thatthe development of the ancient Maya transformed theregion known to conservationists as the Maya forest1

into a sustained and sustainable human landscape formillennia. Although the development of the ancientMaya is well researched, there remains debate overthe complexity of their subsistence adaptation and pat-terns of settlement distribution. The largest centers,such as Tikal, arose in areas where no agro-engineeringfeats have been identified, implying a constellation ofcomplex labor-intensive land use strategies that haveleft no trace (cf. Nations and Nigh 1980; Nigh 2008).The dense forest environment encountered today hasshrouded the archaeological view of the Maya land-scape, one of the few ancient civilizations that emergedin the now densely vegetated tropics (Bacus and Lucero1999). The incomplete data on settlement distribution,civic center locations, and land use stimulate debateon ancient Maya population (e.g., Rice 1978; Turner1990; Ford 1991b). In most cases, scholarship has ig-nored the quantitative relationship between the spatialdistribution of Maya settlements and its patterned in-teraction with the environment that controlled aspectsof everyday life. In this study, the significance of geo-graphic resources to the Maya settlement and land usepatterns are considered for the culminating period ofthe Late Classic when the ancient Maya civilizationwas at its apex and nearly all habitation sites were oc-cupied (Culbert and Rice 1990). We use the data fusioncapabilities of GIS with data from the recently devel-oped University of California at Santa Barbara (UCSB)Maya Forest GIS (ADL 2004). Using these data, we ap-plied weights-of-evidence Bayesian statistical techniquesto explore the unknown components of the settlementpatterns as a supplement to those known to archaeology.

There is a substantial body of research and data onMaya settlement and interpretations of patterns basedon surveys in sample areas of the region (e.g., Bullard

1960; Ashmore 1981; Fedick and Ford 1990). Thesestudies show that environmental factors played a rolein Maya site location, but the spatial implications ofthese arguments have not been fully pursued. It is gen-erally acknowledged that settlement densities were highand that distributions should be related to agriculture(Fedick and Ford 1990), suggesting a wide diversity inagricultural pursuits (see Whitmore and Turner 2001).These settlement data are used as a proxy for population(Culbert and Rice 1990) and, by extension, land use.

Population estimates for the ancient Maya are mostoften a starting point for discussions of land use and agri-culture (e.g., Dahlin et al. 2005), with ranges from 100to 1,000 persons per square kilometer or more, muchgreater than existing densities in the same area. Majorcenters, such as Calakmul, Tikal, and Caracol, haveproposed densities of 1,000 to 2,000 people per squarekilometer (Culbert and Rice 1990), equivalent to con-temporary Sri Lanka. These estimated population fig-ures are incompatible with other aspects related tothe steady expansion and development of the Maya inprehistory (Turner 1978). Yet without a spatially dis-tributed context for the Maya population, it is difficultto validate the estimates (Turner 1990, 178–211). Abetter understanding of the dimensions of settlementpatterns is required to understand settlement distribu-tion and land use. This is the objective of our research.

The archaeological record is, of course, incomplete.There are a number of reasons for this, but foremostamong them in the Maya context are (1) the difficultyof exploring the mostly forested and isolated parts ofthe region; (2) archaeology’s bias toward research onthe largest settlements; (3) the incomplete record of theentire settlement context from peripheral farmstead,household, and village to urban settings; (4) the vari-able nature of settlement data collection in the region;(5) the lesser likelihood of survival of smaller settle-ments, and (6) the difficulty in surveying a geographi-cally significant bounded unit such as a drainage basin.Given these constraints, how can we interpolate the un-known Maya sites, both residential and civic, based onthe recorded partial record? One way to accomplish thisis with probabilistic spatial modeling. We chose to testa predictive model of the patterns of Maya settlementusing the GIS at a local geographic scale of 1:50,000.At this scale, we focus on field data collection in the ElPilar area in the vicinity of Belize River (Figure 1). Theweights-of-evidence method was used to convert mapsfor the area into probabilities of settlement using theknown point distributions of archeological sites fromthe El Pilar area (Ford 1990, 1991a, 1992, 1993; Ford


Modeling Settlement Patterns of the Late Classic Maya Civilization with Bayesian Methods and GIS 3

Figure 1. Mesoamerica and the study area with application andvalidation zone indicated.

and Fedick 1992). New field data were collected to val-idate the model. The results impact our understandingof Maya land use in the past and resource managementfor the future.

The Geography of the Maya Forest

The Maya forest region, as defined by conserva-tionists (The Nature Conservancy 2007), includesthe tropical lowland setting of the greater YucatanPeninsula including parts of eastern Mexico, north-ern Guatemala, and Belize (Figure 1). The region wasdensely occupied in prehistory. Today, it is character-ized by rolling limestone hills and ridges (Turner 1978),most covered by a deciduous hardwood forest. This ver-dant growth thrives on an annual rainfall average of1,000 to 3,000 mm that falls mainly from June to Jan-uary. A dry season runs from January to June. Activitiestoday are seasonal, impacted by this wet–dry delugeand drought sequence, as they were in the Maya prehis-tory (Haberland 1983; Ford 1986, 1996; Scarborough1998; Lucero 2002; Peterson and Haug 2005; see alsoEwell and Sands 1987). The Maya forest, a biodiver-sity hotspot, consists of large protected areas (Nations2006), but population expansion and associated farmingas well as commercial agricultural and cattle pasturageare pushing the development frontier into the contigu-ous forest areas. Until recently, 85 percent (30,000 km2)of the Peten of Guatemala was covered with semidecid-

uous subtropical moist forest. Less than 50 percent nowremains (Sader et al. 2004).

Land clearing often reveals the location of archeo-logical sites and leads to their destruction by erosion,looting, and plowing. With the long-standing interestin the Maya, the impact of tourism has been slight,but larger sites with monumental architecture are of-ten developed for tourism and some rank among thegreatest existing windows into the ancient Maya world.Mapping of smaller ancient sites, although incomplete,varies by country. Current trends to move to digitalcoverage have produced a surprising amount of GISdata that can be gathered from existing sources andmapping agencies in Belize, Guatemala, and Mexico.Despite the spotty nature of the accessible settlementdata in the Maya area, it is clear that, at the height ofthe ancient Maya during the Late Classic, the numberof inhabitants of the region as a whole was significantlygreater than that of today[,] with little difference ingeographic resources. The general climate regime hasretained its wet–dry annual cycle within the context ofa general drying trend over the past 4,000 years (Deeveyet al. 1979; Hodell, Curtis, and Brenner 1995; Curtis,Hodell, and Brenner 1996; Haug et al. 2001).2 It is alsoevident that, despite the blanket population densities,settlement and land use patterns were far from uniformover the landscape (e.g., Turner 1978; Fedick and Ford1990).

The ancient settlement distribution in the Mayaforest relates to a complex mosaic of regional landresources (Fedick 1996; Ford 1986). Settlement den-sities are recorded as the highest in the well-drainedhills of the region (Bullard 1960; Puleston 1973; Rice1976; Rice and Puleston 1981; Fedick 1989; Fedick andFord 1990). For the El Pilar Study area, 98 percentof the residential sites tested dated to the Late Clas-sic period (Ford 1992; Ford and Fedick 1992). Thisis the period when the land use was at its greatestand the overwhelming majority of sites were occupied(Culbert and Rice 1990). The Late Classic is the periodthat we considered in our modeling of ancient Mayasettlement.

The limestone ridge lands are concentrated in thecentral areas of the Maya forest (Turner 1978). Thearea is characterized by shallow and fertile Molisols(also known as Rendzinas; Food and AgricultureOrganization of the United Nations [FAO] 2008) ofexcellent quality (Fedick 1988, 1989), representingbetween 1 and 2 percent of the world’s tropics, yetnearly 50 percent of the soil around the ancient Mayacity of Tikal (Fedick and Ford 1990; see also Kellman



and Tackaberry 1997, 82). Soil on these ridges is en-riched by surface organics derived principally from leaflitter (Kellman and Tackaberry 1997). The thin soilcovers broken limestone bedrock terrain and is supe-rior for the hand cultivation methods that character-ize traditional Mesoamerican farming (Fedick 1989;Wilken 1987; Whitmore and Turner 2001), but it isinappropriate for contemporary commercial farmingmethods that focus on arable (i.e., plowable) lands(Jenkin et al. 1976; Birchall and Jenkin 1979), afact of relevance to conservation risks in the regiontoday.

There is a distinct relationship between the availabil-ity of well-drained ridges, settlement density, and the re-gional Maya settlement hierarchy (see Fedick and Ford1990). This is evident in the local settlements around ElPilar, where extensive fieldwork of surveys and excava-tions has taken place (Fedick 1988; Ford 1990, 1991a;Ford and Fedick 1992). Although it has been proposedthat this dispersed pattern can be explained by culturaldimensions of political or economic factors (e.g., Freidel1981), it has also been argued that the uneven distribu-tion of land resources has influenced Maya settlementpatterns (Fedick 1989). Only a spatial examination ofresources can resolve this divergence.

The ancient Maya occupation of the central low-lands encompassing an area of at least a 100-km radiusaround Tikal can be traced back 5,000 years (see Ley-den 1987). Settlement data show an expansion overthe area, which appears to have followed rivers, thenlakes, and, ultimately, spread across the entire interior(Puleston and Puleston 1972; Rice 1976; Ford 1986).Evidence indicates that the interior Peten area aroundTikal dominated the study region in the Late Classic pe-riod between AD 600 and 900 (Marcus 1983, 1993; Ford1986; Culbert, Levi, and Cruz 1990; Fedick and Ford1990). Maya settlement and civic construction were at amaximum at this time. The process of centralization wassustained through AD 900, when the Maya civilization“collapsed” (Webster 2002). Tikal and other major ad-ministrative and political centers witnessed abrupt haltsin public projects, although some continued throughthe Terminal Classic Period (AD 900–1000). Settle-ment surveys between Tikal and Yaxha (Ford 1986,1991b), in the El Pilar area (Ford 1985, 1990, 1991a),as well as in northern Belize (Pendergast 1981; D. E.Z. Chase and Chase 1988; Hammond et al. 1998) at-test to persistent occupation. In some areas, such as ElPilar, monument building continued, only to cease inthe Postclassic (Ford and Orrego 1996; Ford et al. 1996;Ford et al. 1998; Ford 2004).

Scholars have focused on the dramatic Classic Mayacollapse (e.g., Culbert 1973, 1988; Redman 1999;Diamond 2005) and the search for the causes of the col-lapse has generated a substantial literature (see Webster2002). Whatever the cause of the abandonment of thegreat Classic Maya cities, the development of Maya so-cial complexity was based on a gradual rise in populationand concomitant intensification of land use, includingfarming of increasingly more difficult and marginal land(Fedick 1992).

Early investments and development in the Maya for-est landscape endured over time (see Whitmore andTurner 2001, 111) and the ancient Maya civilizationintegrated populations of the region over a span ofmore than 2,000 years. Environmental dimensions of-fered attractions as well as constraints for the early an-cestral Maya. Among these attractions is soil fertilityconstrained by the critical problem of surface water(Ford 1996). Subsistence strategies and cultural devel-opments mediated constraints and agricultural diversitysustained the steady growth of the Maya civilization(Whitmore and Turner 2001, 100–107). Given the finescale evidence for precipitation variation over this longperiod (Haug et al. 2001) coupled with the extensivestudies of the Peten lakes (Leyden et al. 1993; Curtis,Hodell, and Brenner 1996; Islebe et al. 1996; Curtiset al. 1998; Hodell, Brenner, and Curtis 2000; Brenneret al. 2002; Rosenmeier et al. 2002), this endurancereflects a flexible adaptation and robust strategies thatbridged periods of disturbance and extremes, such ashurricanes and drought.

Despite the evidence that the Maya forest environ-ment did support high settlement densities in the pastbased on a sustained land use strategy, little effort hasbeen made to identify the balance that was attainedby the ancient Maya land use pattern. Instead, mod-ern industrial strategies are proposed for developmentthat bear no resemblance to the past (for the BelizeRiver area, see Birchall and Jenkin 1979) and canbe seen as a destructive short-term trajectory in pro-cess (Turner 1990; Sever 1999; Sever and Irwin 2003;Nations 2006). By identifying the spatial land use pri-orities of the ancient Maya, we can assess the contrastof development planning used in the region. The suc-cess of the past strategies, the needs of contemporarylocal populations, and the conservation demands of thetwenty-first century underscore the imperative to un-derstand the basis of the past prosperity and sustain-ability of Maya agriculture.

To what extent does the fundamental geographicsetting of the ancient Maya civilization account for its



adaptation? Our research focuses on geographic data asa means to isolate the subsistence economy from thepolitical and social factors. We use geographic variablesthat directly influence agriculture, past and present(Jenkin et al. 1976; Birchall and Jenkin 1979), and in-clude field improved digitized 1:50,000 maps of the soils(based on Fedick 1988, 1989, 1994). We focus on set-tlement patterns of the Late Classic period, the apogeeof the Maya civilization, representing the accumulationof two millennia of Maya accomplishment.

Weights-of-Evidence Analysis

The premise of our model is that an actual map ofancient Maya sites and settlement distribution consistsof a set of all locations, including the subsets of lo-cations that remain unknown and locations of knownarcheological settlements. The actual distribution can-not be known but can be modeled probabilistically, andby focusing on strictly geographic variables we avoidbasic weaknesses of proxies identified in earlier predic-tive models (see Kohler and Parker 1986). The settle-ment distribution is assumed to spatially reflect mappeddistributions of major geographic and environmentalfactors. Among these are topographic slope, soil fer-tility, soil drainage, and proximity to water sources.These factors each contribute independently to the lo-cation of all settlements, but in different degrees. Theseamounts, or weights, are the basis for the use of weights-of-evidence (Monts-Homkey 2000; Ford and Clarke2001; Raines, Bonham-Carter, and Kemp 2000; Sirjean2003; Monthus 2004). Weights-of-evidence origins arein mining geology (Bonham-Carter 1999). The essen-tial tools were integrated into a GIS software package,ESRI’s ArcView 3.2 and later ArcGIS (Sawatzky et al.2004), and extended into a broader analysis packagecalled ArcSDM.

Weights-of-evidence methods are Bayesian meth-ods. They involve a given point distribution, assumedto be a sample or incomplete representation of a spa-tial distribution that is the consequence of the com-bined influence of multiple influential factors or themes,represented as binary or categorical maps. The the-matic maps are usually reduced category sets, or clas-sified fields, processed by buffers or other map-basedtransformations. The influences can be positive or neg-ative (i.e., a map class is prohibitive to the samplerather than causative) but must be independent of eachother. These thematic maps, when overlain with thepoint sample, result in an expected statistical densityfor each class. The explanatory significance is compared

to random variation; that is, does a particular class in-crease or decrease the likelihood of finding a positivesample by a significant amount. Each map contributesexplanatory power to the point distribution proportion-ally (the weight). By assembling the sum of the weightedcontributing factors as a map overlay, a probabilistic sur-face representing the likely complete distribution den-sity can be generated and used in modeling, simulation,and forecasting.

Weights-of-evidence analysis follows six steps:

1. Select known points of features to be modeled,such as Maya sites.

2. Select thematic maps that are suspected tocontribute to the explanation of the pointdistribution.

3. Using the correlation analysis tools of weights-of-evidence, convert selected map layers to binary orcategorical form in the most predictive manner.

4. Test for conditional independence comparingprior and posterior probabilities by class combi-nations, eliminating those maps that do not con-tribute explanatory power.

5. Create a set of weights to use for each layer usingBayesian methods.

6. Develop posterior probability and the associ-ated uncertainty maps using the weighted layers(Bonham-Carter 1994; Raines 1999).

Any model is, of course, only as good as its ability(1) to be calibrated and validated to reflect past and ex-isting data; (2) to summarize and simplify a real systemto allow experiments and test hypotheses; (3) to al-low aggregate investigations that would not be possibleby testing parts in isolation; and (4) to be transportableacross scales or regions. The weights-of-evidence modelallowed us to forecast where to expect and not expectMaya settlements, promoting an independent means ofvalidation. The model also allows us to see the relativeimportance of environmental factors in Maya settle-ment location choices. The probabilities can then beused as environmental weights in a settlement simu-lation. Although the actual application is not directlytransportable, because the weights are derived for eachcase empirically, the approach is applicable universallyand the posterior probability layer can be input intofurther models.

Each input map is converted into two or more cate-gories based on the data. Of the detailed 1:50,000 soilclasses on the UK Overseas Development Administra-tion (ODA) survey of the Belize Valley (Jenkin et al.1976), we condensed characteristics of soil drainage and



soil fertility into four categories: good, moderate, poor,and very poor. The number of categories for a con-tinuous variable, such as percentage topographic slope,was determined using the software and some simplemeasures. The weights-of-evidence method then com-bines layers and categories spatially to determine whichpoints in the evidential theme, in our case, archeologi-cal sites, fell inside or outside of the regions on the map.The weights were computed both as layer/class contrib-utors to a distribution and their opposite; that is, they“repel” the evidential theme.

The computation of positive weights (W+) reflectsthe possibility of the presence of sites if the theme ispresent. If the positive weight itself is positive (W+ >

0) in the presence of the class or theme, we will findmore sites than chance would determine. If the pos-itive weight is negative (W+ < 0), then presence ofthe theme means fewer sites than what chance woulddetermine. If W+ is equal to zero, the class or themedoes not influence the location of sites. In this neutralcase, the computation shows that it is not an eviden-tial theme and so should be excluded from the model.Negative weights (W−) are the complement. If the neg-ative weight is negative (W− < 0), the class or themeis a good negative indicator, whereas if (W− > 0),we would expect that in the absence of the class ortheme on the map, we will find more sites than whatchance would suggest. Contrast is the difference be-tween the positive and the negative weights (C = W+–W−) and is a measurement of the correlation between aparticular class or theme and the training points, whichin our case are the known archeological sites. If contrastfor a theme is high, then that theme in general is a goodpredictor. Weights are log ratios of conditional proba-bilities and expressed in continuous decimal fractions.

The ArcSDM weights-of-evidence software providesfor the computation of a table of class weights, the con-trast, and an estimate of the variance of the contrast.This allows statistical tests of significance of the evi-dential themes, allows the computation of uncertainty,and permits trial-and-error testing of class and layercombinations to build strong models. A final output isthe sum of the posterior probabilities for all contribut-ing classes and layers in the form of a map showing theirspatial distribution. The result is a map of the spatialpredictive model of the dependent variable.

UCSB Maya Forest GIS

Research began with the collection, integration,harmonization, preservation, and distribution of the

scattered environmental and settlement data from thegreater Maya forest region. The result was collectedinto the UCSB Maya Forest GIS, a data repository forthe greater Maya forest region of Belize, Guatemala,and Mexico. Data contributed by agencies and scien-tists have been combined into a regional GIS (Ford andClarke 2000) and encoded with Federal GeographicData Committee standard metadata for archiving andredistribution via the Alexandria Digital Library (ADL;Smith and Frew 1995; ADL 2006). The Maya ForestGIS data are available through the online library ofADL.

Our original compilation on CD (Ford and Clarke2000) was based on the GIS developed at the re-gional scale by the Paseo Pantera Consortium forthe U.S. Agency for International Development. In-tegrated with this are digitized maps of topographyand soils, Sader’s (1999) land use data for the Peten,geo-referenced Landsat MSS satellite images, the lo-cal GIS database developed by Fedick (1989) on soilsbased on the UK ODA map for the entire Belize Rivervalley at the resolution of 1:50,000 (topography, soil,geology), as well as the Late Classic period settlementdata from Ford’s Belize River Archaeological Settle-ment Survey (BRASS) project at the resolution of1:2,000 (1983–2003) and a 1998 1:6,000 photomosaicof the El Pilar Archaeological Reserve for Maya Floraand Fauna (Girardin 1999). This first version (Fordand Clarke 2000) has been shared with collaborators inBelize and Guatemala and published on the Internet bythe SCS Belize Country Conference (Ford and Clarke2002). A key recent addition for this research has been90-m-resolution digital topography from the NationalAeronautics and Space Administration Shuttle RadarTopographic Mapping program (Sifuentes 2005) thatprovides seamless slope coverage for the entire Mayaregion. Many of these data themes provided the inputweighting layers for the weights-of-evidence modeling.The specific characteristics of the input data are listedin Table 1.

Our study area includes the diverse characteristicsof the greater Maya region: well-drained ridge lands,poorly drained lowlands, wetlands, as well as the rarealluvium. The majority of the El Pilar study area is com-posed of well-drained ridges (Fedick 1989; Fedick andFord 1990), featuring the characteristic limestone ridgeswith the shallow and fertile Molisols that surround thenearby Maya centers of Tikal, Yaxha, and Naranjo (20–50 km distant). Like Tikal, the limestone ridges aroundthe major center of El Pilar are noted for their absenceof surface water and compose nearly 35 percent of the



Table 1. Input data for the weights-of-evidence analysis

ThemeScale

and extent Map source Resolution Classification Transform

Soil fertility El Pilar area Digitized paper map(Jenkin et al. 1976)

1:50,000 vector 4 classes 4 classes

Soil drainage capability El Pilar area Digitized paper map(Jenkins et al. 1976)

1:50,000 4 classes 4 classes

Topography El Pilar area Paper official topo 40 m 1:50,000 vector 40 m contours Slope percent (percent),below and above 16percent

Belize dataRivers and streams El Pilar area Paper official topo 40 m 1:50,000 vector Perennial and

intermittentstreams

Distance buffers

Belize dataLakes and water bodies El Pilar area Paper official topo 40 m 1:50,000 vector All Distance buffersArchaeological sites Site local Belize River

ArchaeologicalSettlement Survey/ElPilar Institute ofArchaeology

1:20001:50,000

Major or minor civiccenter/residentialsitesundifferentiated

1 class

area (Ford 1992). The ridge lands are the concentrationof the settlements and the location of the largest Mayacenter in the area, El Pilar, situated 10 km from theBelize River (Ford and Fedick 1992, 45). The El Pilarstudy area includes only a minor presence of closed-depression wetlands, generally a significant componentof the landscape of the entire Maya area (Fedick andFord 1990; Dunning et al. 2002).

The landscape of the El Pilar area features the BelizeRiver, one of the few rivers that run east from the cen-tral Peten. Rivers, and for that matter any surface water-courses, are scarce in the Maya region (Scarborough andGallopin 1991; Scarborough 1994, 1998; Ford 1996;Lucero 2002). This makes the examination of the ElPilar area important. The alluvium of the El Pilar area,including the settlement of Barton Ramie (Willey et al.1965), is notably fertile, yet makes up less than 5 per-cent of the study area (Fedick 1988, 1989; Fedick andFord 1990; Ford 1992). Although likely a critical com-ponent of the earliest agricultural adaptations (Powiset al. 1999; Lohse, Awe, and Griffith 2006), the im-posed growth limits are evident given the later periodsof occupation that concentrate in the abundant ridgelands north and into the interior. The valley providesa very restricted area for expansion and development;consequently, it does not constitute a major focus ofLate Classic Maya occupation (Ford 2006).

The importance of water cannot be overemphasized(Scarborough 1994; Ford 1996). Unchecked, surface

water is quickly absorbed into the limestone bedrockand follows invisible and inaccessible aqueducts withinthe limestone. Our examination of the El Pilar area willprovide a means of identifying the priority of water-courses for Maya settlement patterns.

Regional examinations of major centers have sug-gested a number of underlying factors contributing tolocation. These include local and long-distance trade,political competition, and access to critical resources(Marcus 1973, 1993; Fry and Cox 1974; Rice 1981;Freidel 1986; Sabloff 1986; Ashmore 1991; Demarest1992; Martin and Grube 1995, 2000; Fox et al. 1996;Foias and Bishop 1997). Patterns of spacing have beenconsidered along with center size and site sustainingarea (Marcus 1976). Our research examines geographicvariables most useful from the agricultural perspective.Early agricultural occupations have been recognizedas later settlement concentrations and ultimately thefocus of Maya civic centers (Ford 1986; Fedick andFord 1990). Thus, our predictive modeling targetedarchaeological patterns as a whole, residential and civic,reflecting the agrarian foundations of the Maya. Weconsidered all archaeological signatures, complex andsimple residential units as well as major and minor mon-umental centers together, as seen in Table 1.

From the numerous data contained in the Maya For-est GIS (ADL 2006), a subset of our environmentaldata layers defined by the northern drainage catchmentof the Belize River were extracted and clipped to the



Figure 2. Location map for the ap-plication zone and validation zoneshowing the training samples from theBelize River Archaeological Settle-ment Survey and Global PositioningSystem field data site locations.

extent of our study area (Figures 1 and 2). The subsetand input layers were selected at the local scale andextent as listed in Table 1 and are shown by the outlinepolygon in the southwest of Figure 2. The study area islocated between the Belize–Guatemala border and theBelize capital of Belmopan and focused on the northside of the river. The area extends from the westernlimestone ridges and hills to the eastern lowlands andwetlands. All data themes of the GIS were consistentlyregistered to WGS84 using the Universal TransverseMercator coordinate system in zone 16 North.

The weights-of-evidence module used here was anextension to ArcView 3.3 as an Avenue script utilitythat is introduced into ArcView as a regular extensionand becomes an additional menu on the main ArcView

software toolbar. The extension has been rewritten andextended and is now available as a more general an-alytical package and for other versions of the ESRIGIS software (ArcSDM; Sawatzky et al. 2004). Us-ing the tool involves following the weights-of-evidencesteps outlined earlier. The tool includes a means toexperiment with layer combinations, layer indepen-dence, category aggregation, and statistical fitting of themodel. After sorting and preprocessing statistical testsof different class breakdowns, we reduced the combi-nations to a workable set of four principal classes andthemes: soil fertility, soil drainage, slope, and distanceto watercourses.

The soil layers were based on the four described soilsuites and subsuites in Birchall and Jenkin (1979) and



Table 2. Soil fertility ranking

Fertilityclass

Cation exchangecapacity Base saturation Phosphate pH

1 High High up 100% Low 5.8–8.22 High High Low 4.1–8.23 Moderate to high Moderate to high Low 4.2–8.34 Variable Variable Low 3.5–7.6

are detailed in Fedick (1988, 179–202). The suites arebased on landforms and the subsuites based on loca-tion. The locations include alluvial parent material inriver valleys with young, mature, and poorly drainedsoil; limestone on the ridges, escarpments, and plainsthat vary based on parent location; Pleistocene coastaldeposits primarily of sand and clay; and wetland areasflooded in the rainy season based on limestone or acidrocks (Birchall and Jenkin 1979, 13).

According to the field results of Birchall and Jenkin(1979) and Fedick (1988), we ranked soil fertilityfrom 1 (fertile) to 4 (infertile). Soil fertility rankis based on cation exchange capacities (CEC), basesaturations (Ca2+, Mg2+, K+, and Na+), along withphosphate and pH levels as identified in Birchalland Jenkin (1979) and employed by Fedick (1988,1989). We summarize the fertility classification inTable 2.

Based on the analysis of Birchall and Jenkin (1979)coupled with the field survey of Fedick (1988, 1989),we were able to classify drainage characteristics of thesoil. We ranked drainage from 1 (good) to 4 (poor).Soil permeability, infiltration, and clay content formthe basis of our ranking, summarized in Table 3.

The themes of slope and watercourses were definedbased on the 1:50,000 scale maps digitized as GIS layersas well as the Shuttle Radar Topographic Mapping(SRTM) data. We defined slope in percentages. In con-sidering access to water, based on statistical tests, weexplicitly eliminated as explanatory variables distanceto the major river, distance to lakes, and distance struc-

Table 3. Soil drainage ranking

Drainageclass

Permeabilityrange (m/d)

Infiltrationrate (m/d)

Claycontent

1 ∼0.34–0.85 2.9–5.0 30–40%2 ∼0.2–0.7 0.8 20–70%3 ∼0.06–0.35 0.12–0.22 40–85%4 ∼0.1–0.3 0.18 up to 90%

tured by Horton stream order. Watercourses were thusdefined as any perennial stream or river.

Our set of four principal classes and themes of soil fer-tility, soil drainage, slope, and distance to watercourseswere included in this study as the evidential layer inputsfor investigating the geographic conditions of Maya set-tlement (Figure 3). They are as follows:

� Theme 1: Soil fertility. Classes: good, moderate,poor, very poor.

� Theme 2: Soil drainage. Classes: good, moderate,poor, very poor.

� Theme 3: Slope (percentage). Classes: inte-ger percent, reduced to above and below16 percent.

� Theme 4: Distance to watercourses (meters).Classes: integer meters, tested in 100 m bands,reduced to 0 to 500 m, 500 to 1,000 m, and greaterthan 1,000 m.

The choice of weights-of-evidence model weightswas based on the statistics, especially the number ofpoints explained by the weighted combination. Thegrid size for the model was one hectare, with an areaof application of 83.59 km2 in the local case. Thearchaeological sites layer included 262 sites in theevidential theme. The analysis was repeated, first withthe application zone surrounding El Pilar and next withthe validation zone for the entire El Pilar/Belize Riverarea of 1,290.11 km2, constituting a major portion ofwestern Belize. The relationships between the two areasare shown in Figure 2 and 3. This formed the basis of ouranalyses.

Analyses and Results: Predicting AncientMaya Settlement

The predictive modeling of ancient Maya settlementpatterns focused on the archaeological survey data of theBRASS/El Pilar surveys, the source of the training sitesfor the predictive model (Ford 1991b, 1992). The loca-tions of the training sites were based on 1:2,000 scalecomposite maps of archaeological sites, residential andcivic (Ford and Fedick 1992). The modeling exercisewas divided into the two phases. First, we used exist-ing data from the surveys to define the application zone(83.59 km2) and developed the initial model. Second,we expanded the initial model from the applicationzone to the validation zone (1,290.11 km2) for the fieldcorroboration. This phased examination provided thebasis of our model development and the field tests.



Figure 3. El Pilar application andvalidation zone input maps for theweights-of-evidence model. A: slope;B: distance to streams; C: soil drainage;D: soil fertility.

The application zone for the El Pilar area was grid-ded by the software at 100 m resolution (one-hectarecells). For the 262 surveyed archaeological training sites(predominantly residential and including civic centers;see Table 1 and Figure 2), the expected a priori proba-bility of encountering a site in a cell was calculated as0.0313. As defined in Table 4, contrast indicates thatsoil fertility has the strongest spatial correlation withtraining sites, followed by proximity to watercourses.Soil drainage and slope are approximately equal factors,both of less importance than soil fertility and proximityto watercourses.

Table 5 shows the criteria used to select the evidencelayers and the thresholds defined by the weights anal-ysis. Soil fertility and drainage are ordered data, eachwith four ranks. In both cases the highest ranks were

Table 4. Weights table for the application zone

Evidence W1 W2 Contrast Confidence

Soil fertility 0.6813 −0.9796 1.6609 11.2877Watercourse proximity 0.6314 −0.4575 1.0889 8.7517Soil drainage 0.6015 −0.2391 0.8407 6.5745Slope 0.0193 −0.8032 0.8225 1.6316

defined as the optimal predictor of training sites. Slope,an ordered measure of slope angle, was analyzed usingthe cumulative ascending method to define a 16 per-cent slope as the optimal maximum slope preferred bythe Maya, as indicated by the training sites. Slope hasthe lowest confidence, where a value of 1.6316 wouldindicate an approximate level of confidence of 95 per-cent. This significance test is approximate, as it is basedon all of the data rather than a random sample.

In the first test of our model in the application zone,197 sites were explained with the weighted evidencemodel. Sixty-five of the 262 points had a posterior

Table 5. Criteria and thresholds for the chosenweighting model

Evidence Criteria Threshold

Soil fertility Preferredhigh-fertility soils

Class = Good soilfertility

Watercourse proximity Near to the river butnot too close

500–1,000 m fromriver

Soil drainage Preferredwell-drained soils

Class = Good soildrainage

Slope Preferred low slopes Less than or equalto 16◦



Table 6. Categorical weights table for proximity tothe rivers

Distance (m) Contrast Confidence STUD CNT

100 −1.1785 0.2199 −5.3599200 −0.9705 0.2159 −4.4953300 −0.4319 0.1935 −2.2317400 0.0055 0.1892 0.0291500 0.5389 0.1838 2.9316600 0.8825 0.1976 4.4652700 0.6473 0.258 2.5089800 1.5585 0.216 7.2158900 1.5633 0.2425 6.44791,000 1.27 0.2994 4.24261,100 −0.1194 0.589 −0.20271,200 0.2465 0.5925 0.416

probability less than the prior probability and thereforewere not explained by the weighted evidence model,resulting in a 75.2 percent overall explanatory level.Of these sixty-five unexplained points, forty fell in thefirst slope class (0–9 percent) and grouped in clusters atthe south of the BRASS survey transect near the mi-nor center Yaxox, in the preferred watercourses Class2; that is, within 100 to 500 m of a watercourse.

The analysis of the proximity to watercourses is par-ticularly interesting, as the data show that the Mayapreferred to live close to, but not too near, watercourses(Table 6). This might have been to avoid areas of flood-ing along the rivers and streams, as is supported by fieldobservations, and with a consistent avoidance of wet-lands. The actual buffer interval was defined by usinga categorical weights table. Note that the weights arewell defined because the distance intervals are large.In analyses with narrower distance intervals, the se-lected interval was mostly of positive contrast, but therewere occasional negative contrasts inside the interval(Sirjean 2003). This is a consequence of the samplesize of the training set. This particular evidential layerand its analysis represent a unique contribution of thisresearch project and are potentially useful for othersworking with weights-of-evidence. Rarely is a singleinfluential theme both a positive and negative contrib-utor to the explanation of a distribution (Table 6), withdistance forming the key variable.

All evidence was generalized within the context ofthe weights-of-evidence to optimize prediction into bi-nary classes (i.e., predictive or nonpredictive), by select-ing the maximum contrast that had a confidence levelof 90 percent or greater based on an approximate Stu-dent’s t test. The soil fertility, soil drainage, and slopewere all analyzed by the appropriate cumulative weights

measurement to define the threshold for good and badevidence as summarized in Table 5. Slope weights anal-ysis was done using the cumulative ascending methodbecause the question asked of the data was this: Whatwas the maximum low slope preferred by the Maya farm-ing settlements? Soil fertility and drainage were appro-priately analyzed by cumulative ascending because theordered ranks were numbered with one as the highestrank. This ordered numbering sequence dictates usingcumulative ascending weights analysis to answer thequestion of how good of a soil would the Maya farmeruse. The answer for both fertility and drainage was Class1, or good. The proximity to watercourses weights anal-ysis was done using a unique categorical weights analysissupported by testing the reclassification of the evidencewith categorical weights (results are shown in Table 6).

Conditional independence is always an issue whenusing a Bayesian method such as weights-of-evidence.The rule is to accept a conditional independencevalue greater than 0.85. In our test, the overall con-ditional independence test reported a value of 0.97.This overall test is considered conservative, and thusthere is no conditional independence problem in ourmodel. The chi-square pairwise test indicates a weakconditional dependence between soil fertility and soildrainage and between soil fertility and proximity to wa-tercourses. The overall test, however, indicates no sig-nificant problem of conditional independence. Theseconditional dependencies are reasonable because fertil-ity and drainage and the favorable linear buffered regionaround the watercourses are probably moderately associ-ated. These considerations, however, do not change themodel because the intended use of the model is to rankareas.

Our results in the application area are supportive.We were able to predict 75 percent of the sites basedon the combined environmental inputs. The next stepwas to perform the field tests of the model.

Validation of the Predictive Modelof Maya Sites

Field validation tests expanded the results of the ElPilar application zone to the entire Belize River val-idation area (Figure 4). The continuous weights wereclassified into four categories (very high, high, low, verylow) and applied to the validation area. The exerciseto confirm the predictive model of ancient Maya sitesfocused on specific zones of interest based on the topog-raphy and the model results themselves. Three zones



Figure 4. Weights-of-evidence posterior probabilities for the El Pilar application and validation zone.

were defined for the field validations. Zone One repre-sents areas comparable to those of the BRASS/El Pilarsurveys with a similar geography. Zone Two is locatedsouth of the BRASS/El Pilar surveys in the area be-tween the Mopan and Macal River branches of theBelize River. The last zone, Zone Three, was focusedon the valley north of Barton Ramie, where the predic-tive model shows serpentine shapes of contrasting poorand moderate prediction for Late Classic archaeologicalsites (see Figure 4), similar to the location of the areaof unexplained sites of our model in the applicationzone.

Each validation zone was visited, canvassed, and sur-veyed in the field. We located areas within each zonefor the validation tests by vehicle. Once oriented, theselected area was covered by foot. Sites were recordedwith ESRI’s ArcPad running on a Compaq iPAQ witha NAVMAN Global Positioning System (GPS) sleeve,while displaying the weights layer in the GIS. Thesedata were then compared with the model and evalu-ated in terms of reliability.

The first validation zone was focused on the north-western area of the model, north of El Pilar around Ca-dena Creek. This zone is very similar to the geographyaround the major center of El Pilar. Today, Zone Oneis largely in plough and pasture with good access by all-weather roads maintained by the Mennonite commu-

nity of Spanish Lookout. Concentrating on the placeswith the strongest predictability for sites, we plottedthe unmapped Maya residential sites and minor centersencountered in the many cleared fields. Predictably, ahigh density of sites characterized the locales with highprobability weights. Most ancient Maya residential siteswere found within 20 to 30 m of each other. Several mi-nor centers were identified as well. This density of sitesin this zone is typical of the densely settled areas ofthe Maya lowlands and leads one to imagine an intenseoccupation of these high-probability areas. Other prob-ability areas were canvassed as well, and all sites wererecorded.

In total, we identified 225 sites in Zone One, allpreviously unmapped. Of the total, 122 of the sites,or 54.2 percent, were recorded in the areas of veryhigh probability and a further 55 or 24.4 percent inthe high-probability area. Sites were also located inthe lower probability areas: 35 or 15.5 percent in thelow-probability areas and 13 or 5.7 percent in verylow-probability areas. Most of the sites in these low-probability areas are near areas of high probability,suggesting some imprecision or generalization of edgelocations in the evidential themes. Considering to-gether areas of very high and high probability, recordedsites account for 76.8 percent of all sites of the valida-tion test. The model proved reliable in this zone.



The second validation zone we examined was thearea south of the El Pilar area between the Mopan andMacal Rivers. Zone Two is marked by difficult brokenterrain and steep hills that are not well reflected in thedigital elevation model (DEM) for the region. Never-theless, eighty-one new sites were located, of whichseventy-four were in the high-probability areas, ac-counting for 91.3 percent of the documented sites in thiszone. The remaining seven sites, or 8.6 percent, were lo-cated in the low-probability areas. The model predictedcorrectly all the locations of newly identified sites. Inaddition, the high-probability areas correspond to thosemapped in the area of the Chan site (Robin 2002). Inthis area no sites fell in the very low-probability area.The model predicted well the occurrence of sites byprobability.

The third validation zone we considered was east ofthe El Pilar area bordered by the Belize River where thefirst archaeological surveys of the area were conductedaround Barton Ramie (Willey et al. 1965). This is wherethe model shows serpentine areas of low probabilitynext to areas of high probability based on watercourses.This model pattern comes from the relationship of sitesto watercourses, where the majority of site locationsfall between 500 and 1,000 m from a watercourse (seeTable 6). Yet in the case of Zone Three, both areas havepoor soils, generally a poor predictor of site locations.Only three sites were found in the vast expanses of ZoneThree. Of these sites, one was in the high-probabilityarea (33.3 percent), one was in the low-probability area(33.3 percent), and one was in the very low-probabilityarea (33.3 percent). None were located in very high-probability areas.

Although there are very few sites in Zone Three,they are evenly divided among the high-, low-, andvery low-probability classes. Only one site, or 33.3 per-cent, was located in a high probability area, and two,or 66.6 percent, were located in low-probability areas,suggesting some uncertainty in the predictability in this

very lightly settled zone. All the sites were situatedon the margins of the defined zone. The one site inthe high-probability area was located 13 km north ofthe river near limestone ridges know as Yalbac Hills.The other two were located close to the Belize River.These results suggest that work on the precise natureof the influence of drainage and Maya site location stillneeds to be addressed. Nevertheless, barely 1 percent ofthe all the newly recorded validation sites were identi-fied in the large expanse of Zone Three.

In conclusion, the predictive model of ancient Mayasites held up well to the field validation tests. We lo-cated and recorded 315 sites in the validation testsand 82 percent of these sites fell in the very high andhigh-probability areas as mapped by the weights-of-evidence model. These newly recorded sites were di-vided relatively evenly between probability areas: veryhigh with 39 percent and high with 43 percent (Table7). The very high-probability areas are densely settled;there were 122 recorded sites in only 5.6 percent of thestudy’s areal extent. The high-probability areas weremore spread out and included 136 sites recorded in 19percent of the study area (Table 7). Together these258 new sites, accounting for four fifths of all valida-tion sites, were predicted for only 24.6 percent of thestudy extent. The remaining 18 percent of the siteswere distributed in areas of low and very low probabil-ity. Only forty-three sites (14 percent) were recorded inthe area of low probability, and only fourteen sites (5percent) were located in the very low-probability areas.Sites in the low-probability areas were distributed across32 percent of the study, yet the few sites in very low-probability areas cover nearly 43 percent of the study ex-tent (see Table 7). Thus, our model could exclude 42.7percent of the study area based on geographic predic-tors with 96 percent confidence that no sites would befound.

It must be borne in mind that these figures reflectthe uncertainty of the data collection and assimilation,

Table 7. El Pilar/Belize River area settlement distribution by probability class

Posterior probability

Percentmasked area

in class

Number of validationpoints (newly mapped

sites) in region

Percentpredicted

sites

Class 1: 0.1039 to 0.225 black 5.55 122 39Class 2: 0.0314 to 0.1039 dark gray 19.09 136 43Class 3: 0.0087 to 0.00313 light gray 32.63 43 14Class 4: 0.0 to 0.0086 white 42.74 14 4Total 100 315 100



limitations of data resolution, and assumptions of theBayesian model. Alternative interpretations are possi-ble. For example, the negative watercourse proximityresults could be the consequence of the destruction ofMaya settlements by flooding and thus not caused byMaya settlement preferences. Still, the strength of thepreference for other factors, especially soil fertility, can-not be easily discounted. Thus, we see that our weights-of-evidence model proves very reliable for predictingancient Maya sites.

These results support the validity of the model inpredicting the major proportion of Late Classic Mayaarchaeological sites, a time when at least 98 percentof residential units and all of the civic centers in thestudy area were occupied (Ford 1990, 1992; Ford andFedick 1992; see also Culbert and Rice 1990). In ageneral review of the residential sites found in thehigh-probability areas, we note that they are relativelylarge, complex, numerous, and dense. Sites that occurin the lower probability areas are small, widely dis-persed, and often close to a sharp probability boundarywith the higher probability areas. Thus, based on thefield validation test, we conclude that our four-factorenvironmental model explained the location of themajority of Late Classic Maya sites. Taking modelprobabilities of greater than 0.0314 as “high” andincluding 24 percent of the area (see Table 7), thenthe model predicted the occurrence of 82 percent or252 of the 315 newly identified validation sites. Inaddition, we have demonstrated that the exclusionof probabilities less than 0.0086 making up nearly 43percent of the area would still recover the majority of96 percent of the sites. These are strong results.

Implications

The combined geographic themes were selected fortheir reflection of agricultural factors in the choice ofsite locations. In our research, we confirmed the con-ditional independence of the geographic variables withregards to the training sites, critical to the overall tests.The geographic themes were each contributors to themodel and the combined themes of the predictive modelcan account for the majority of sites based on the val-idation field tests. The strong influence of soil fertilityamong the four themes is particularly relevant, sup-porting the view that agricultural strategies were at thefoundation of Maya settlement patterns and develop-ment. Watercourses were relatively important, but moresignificant was that distances were found to be neither

very close nor too far. Drainage and slope also playedroles, ranked below fertility.

Our investigation of the predictive model for ancientMaya sites provides a means to evaluate land use pat-terns of the Late Classic Maya. Consistently, the resultsof our model implicate geographic motives for site lo-cation. These choices match contemporary smallholderfarming choices for residence locations and underscorethe importance of farmer choice in the selection of res-idential site locations. These locations would be thenatural focus for the development of civic centers thatwould have responded to established settlement areas.

In this examination, sites have been predicted basedon geographic variables and highlight the value ofgeography in explaining site locations. Although theagricultural basis of Maya settlement patterns has beenappreciated, the spatial relationships had not been rig-orously explored until now. Of particular importancehere is the power of GIS to synthesize and coregistermaps, supporting the analytical results and their val-idation in the field. Consequently, this examinationbrings to the fore the value of the weights-of-evidencemethod of evaluating multiple variables across space,demonstrating the combined contribution of geographyand the value of agricultural choices to the settlementpatterns of the ancient Maya.

With the weights-of-evidence method, we have iso-lated features of the geography in the Maya area thataccount for the majority of the archaeological sites, bothresidential and monumental. The areas of strongestprobability of sites include only 24 percent of the entirevalidation area (see Table 7). As can be seen in themaps, these areas occur in scattered patches throughoutthe El Pilar area (Figures 4 and 5). In other words,82 percent of the sites discovered in the validationtests (252) are found in dispersed high-probability ar-eas without concentrating in any one particular area,such as adjacent to the Belize River or near the majorcenter of El Pilar. This is noteworthy as the ear-liest surveys surmised that the Belize River exertedsignificant influence on Maya settlement (Willey et al.1965).

The geographic preference for settlement in the lime-stone ridges is significant in that this geographic land-form is a dominant characteristic of the entire Mayaregion from the El Pilar area west across the greaterPeten of Guatemala and north to the modern state ofthe Yucatan, Mexico. The exploration of this model-ing strategy for other study areas as well as the regionas a whole might prove promising in identifying large-scale regional patterns of Maya land use. How much the



Figure 5. Examples of sites in high-probability areas from the application area, A: Laton, B: Alta Vista, and C: Bacab Na.

weights of the contributing factors vary with scale andregion is a topic for future research.

The areas that evince the very lowest probability ofsites include more than 42 percent of the El Pilar area(see Table 7) and yielded very few sites (fourteen, or5 percent) in the field validation tests. Interestingly,these zones found in the El Pilar application area showboth low density and small sites. Within the El Pi-lar surveys of the BRASS project, examples of such

sites underscore the varied land use objectives of theMaya.

Examples of small sites in very low-probability ar-eas around El Pilar were identified and tested in theBRASS survey (Ford 1990, 1991a, 1992; Ford andFedick 1992). The sites are uniformly diminutive andexcavations suggest that they were specialized in nona-gricultural activities. Artifacts collected from excava-tions, for example, on the Yaxox transect (Figure 6),

Figure 6. Examples of sites in very low-probability areas from the application area: Near the major center of El Pillar (left image) and northof the minor center Yaxox (right image).



indicate some of these were small-scale stone tool pro-duction sites that expanded in the Late Classic intothe poorly drained lowlands adjacent to a chert quarry(Ford and Olson 1989); these areas were quickly aban-doned in the Terminal Classic (Fedick 1989). A similarexpansion pattern is noted for the very low-probabilityarea adjacent to El Pilar (see Figure 6) where sites arerecorded after the initial Preclassic, clearly related tothe presence of El Pilar with its early construction dat-ing to the Middle Preclassic (Ford 2004) and its growthand expansion thereafter. Thus, occupation in some ofthe lowest probability areas reflects other factors thatmight not relate to agriculture. Whereas sites around ElPilar may simply be infilling, the case of the Yaxox sitesmight be related to entrepreneurial economic special-izations where resources, such as chert for stone toolsor clay for ceramics, are found (cf. Brumfiel and Earle1987).

Critically, our model demonstrates that ancientMaya populations that were living in the Late Classicperiod were not homogeneously distributed. There werespecific areas that were preferred and general areas thatwere avoided for Maya settlement. The majority of siteswere concentrated in the areas favorable to smallholderagriculture where fertility, watercourses, drainage, andslope supported farming. These features of the localgeography are scattered, primarily based on parent ge-ology, and occurred in less than one quarter of the studyarea (24 percent). An overwhelming majority of sites,82 percent, were identified in these high-probabilityareas. This result has significant implications for thedevelopment of population estimates. If, as in the ElPilar case, the majority of sites that are used as proxyfor population are concentrated in a small percentageof the land area, then projections of population need tobe adjusted significantly downward and should excludethe majority of the landscape.

Although there are scholars who have clearly recog-nized the geographic preferences of the ancient Maya forthe well-drained ridges (Turner 1978; Dunning 1996;Dunning et al. 1998), there are Maya scholars whoproject high population densities evenly across a re-gion for the Late Classic Maya (Haviland 1969; A.F. Chase and Chase 1987; Dahlin et al. 2005; Healyet al. 2007; except Webster 2005). Our results, consis-tent with Turner and others (Turner 1978; Whitmoreand Turner 2001), support the understanding of landuse in developing population estimates (see also Turner1990). In other words, part of the problem of under-standing Maya land use comes from the way in whichpopulations are estimated. Population estimates might

be high in areas of settlement, but these are limited bygeographic factors related largely to farming choices,as shown in our spatial model. The predictive modelfor the Maya sites of the El Pilar area provides a clearspatial view of the settlement and by extension popula-tion distribution. Not only should population densitiesbe focused on areas that have a strong probability foroccupation, but because these populations are largelyagrarian in nature, these will tend to be the areas whereagriculture would be practiced. Our predictive modelprovides the essential spatial criteria for making suchassessments.

Translating our map results for the ancient Mayainto land use and agricultural practice is an importantstep. To undertake this exercise, we first compare ourpredictive model to that of the Western model of con-temporary development priorities. Given the prosperityapparent for the ancient Maya, how do the priorities ofthe predictive model for ancient Maya settlement com-pare with the contemporary development model? Theanswer to this is facilitated by the fine-scale and detailedsoils research and agricultural undertaken in the BelizeRiver area published in the 1970s (Jenkin et al. 1976;Birchall and Jenkin 1979). This agricultural assessmentwas undertaken within the context of a developmentscheme promoted by the United Kingdom and consis-tent with the United States for emerging economiessince the last half of the twentieth century (Boserup1965; Bray 1994; see also Fukuoka 1978; Mollison 1988;Nash 2007).

Following Fedick’s digitized classifications of thesedata and our field update and verification of the soilpolygons, we evaluated the priorities of the UK ODArecommendations based on “suitability for arable use”(Birchall and Jenkin 1979, 5; Jenkin et al. 1976,215), explicitly defined as geographic characteristicsamenable to plowing (Jenkin et al. 1976, 215–16).With these data, we were able to compare the re-sults of our model of ancient Maya settlement pri-orities based on Prehispanic Maya hand cultivationdependent on stone tools (Denevan 1992) and rain-fed systems (see Whitmore and Turner 2001) to thatof the contemporary development model based onmechanical cultivation dependent on technologicalsystems.

To evaluate the relationship of the two models, weemployed the attribute in the table of the 1:50,000 soilmaps that ranked land capability class (Jenkin et al.1976, 215–23). For the development model, we simpli-fied the nine classes of the ODA study to four classes(1–4, high to low development priority) to compare



Table 8. El Pilar/Belize River area development class

Developmentclass

Percentarea

in class

Mean posteriorprobability,

Set 1

Mean posteriorprobability,

Set 2

Class 1: Black 5.8 0.0546 0.0578Class 2: Dark gray 51.5 0.0014 0.0014Class 3: Light gray 16.5 0.0418 0.0459Class 4: White 23.2 0.0653 0.0671

with the Maya model (Table 8), based on the actualweights-of-evidence posterior probabilities (see Table7). Two sets of random points were generated to pop-ulate each of the four development classes with onehundred points using Hawth’s tools (Spatial Ecology2008). The total 400 points per set were intersectedbetween class in the development model and poste-rior probability for the Maya model. The probabilityof finding a Maya settlement within each set of thefour development classes was calculated based on theweights-of-evidence statistic and plotted for both sets ofrandom numbers. The results of this test are presented inFigure 7 and Table 8.

Overall, the contemporary development classes showlittle in common with the Maya model. Excludingthe top 6 percent of the potential development area,there is an inverse relationship with respect to theMaya model. The most dramatic distinctions come fromthe comparison of the lowest development class as-signed for the limestone hills preferred by the Maya

and the high development class assigned to the poorlydrained lowlands virtually avoided by the Maya (seeFigure 7).

Although the highest development priority areas,Class 1, proved a relatively good predictor of Maya sites(but hardly comparable to the weights-of-evidence) andshares some qualities in common with the Maya model,this class represents less than 6 percent of the study areaand includes principally the restricted well-drained al-luvial areas that offer value for hand and mechanicalagriculture (Figure 8). Development Class 2 accountsfor over 50 percent of the landscape of the study areawith an assessed high potential for contemporary devel-opment, yet this high development class is the poorestpredictor of Maya sites with mean weights-of-evidenceprobability in the very low-priority areas for the Maya(Table 7). Conversely, the lowest development priority(Class 4) is the best predictor of Maya sites, incorporat-ing 23 percent of the study area largely located in thehilly terrain with thin soil (Figure 8). Thus, the con-temporary development model that prioritizes based onarable land does not explain the ancient Maya land usestrategies.

In considering alternatives to explain the LateClassic period Maya settlement model, Maya ethno-graphic and ethnohistorical data are important (VillaRojas 1945; Redfield and Villa Rojas 1962; HernandezXolocotzi, Bello Baltazar, and Levy Tacher 1995;Zetina-Gutierrez and Faust 2006). These cases showthat our predictive model shares much in commonwith the time-honored traditional practices of farmers

Figure 7. Comparison of the Mayamodel probabilities with the develop-ment model priorities.



Figure 8. The development model(left) with an example of Class 1 de-velopment priority that overlaps withhigh Maya settlement probability at Ba-cab Na (right above) and Class 4 de-velopment priority and very low Mayasettlement probability at Laton (rightbelow).

in the Maya region today. The choices of soil, drainage,and slope characterize traditional land use techniques(Nations and Nigh 1980; Fedick 1995; Beach, Far-rell, and Luzzadder-Beach 1998; Levasseur and Olivier2000; Gomez Pompa et al. 2003; Levy Tacher, Rivera,and Rogelio 2005). Even the contemporary forest itselfshows the signatures of ancient impacts with a plethoraof economic plants (see Balick and Cox 1996; Balick,Nee, and Atha 2000; Campbell et al. 2006; Ford 2008;see also Balee 1994). Moreover, these high settlementdensity areas identified in the predictive Maya modelexhibit considerable homogeneity of species indicatedby high beta diversity (Campbell et al. 2006), reflect-ing a strong human impact of centuries of cultivationof tree-dominated plots despite current abandonment(see also Peters 2000).

Smallholders today who practice traditional farm-ing strategies emphasize trees in their “forest gardens”and demonstrate the conservation of soil productivity,water retention, promotion of biodiversity, and sup-port for birds and other animals (Gomez Pompa, Flores,and Sosa 1987; Gomez Pompa and Kaus 1990, 1999;Hernandez Xolocotzi, Bello Baltazar, and Levy Tacher1995; Griffith 2000; Ferguson et al. 2003; Ford 2008; seealso Wilken 1987). In this way, our predictive modelprovides a link from ancient Maya settlement patternsto contemporary ecological, botanical, and agroforestrystudies.

The traditional farming techniques provide a basisfor interpreting the ancient Maya site selection pat-terns (cf. Teran, Rasmussen, and Cauich 1998). Theland use priorities identified in our predictive modelmaximize access to a combination of agricultural ar-eas and vary the intensity of land use based on agri-cultural potentials across a wide sustaining area. Thedifferent land use strategies would be adapted to lo-cal productive capacities (Hernandez Xolocotzi, BelloBaltazar, and Levy Tacher 1995), from polycultivatedhigh-performance milpas (Wilken 1987) to shadedtree orchards with understory cover crops and pro-ductive palms as found in home gardens and out-fields today (Gomez-Pompa and Vazquez Yanez 1981;Corzo Marquez and Schwartz 2008). These differinglevels of land use intensity would be reflected in thesettlement size, composition, and density. Permanentresidential units would be focused in the high pro-ductive areas, with seasonal and intermittent use oflow productive areas, while avoiding the least produc-tive areas for agricultural pursuits (see Zetina-Gutierrezand Faust 2006; Zetina-Gutierrez 2007). This is theinfield–outfield model espoused by Netting (1977) andothers.

In summary, we see the results of the predictivemodel of Late Classic Maya settlement as a critical ba-sis for reexamining current themes of ancient Mayadevelopment and demise, for considering more diverse



subsistence adaptations, and for a reassessment of pop-ulation levels. This first test at modeling ancient Mayasettlement patterns has resulted in patterns that spreadsettlements and populations out into the mosaic of lim-ited preferred zones of dense occupation of green com-munities (Voorhies 1982; Graham 1998), large areaswith light occupation, and a proportion of the area withlittle or no evidence of occupation (see Fedick 1988).With these results we can begin to model land use andpopulation levels that would best explain the patternsof the predictive model.

Further, the predictive model provides a foundationfor examining the value of geographic variables at largerand smaller scales. Extending the model to the regionallevel for the whole lowland Maya area at, for example,1:250,000, could provide insights into the variationsof land use across the Yucatan Peninsula. We haverecently developed a soil coverage theme for fertilityand drainage of the entire Maya area (Sifuentes 2005).These new themes provide an opportunity to scrutinizethe preference for farming sites in the region and toconsider changes over time.

The results of our predictive model are promising,demonstrating that areas with Late Classic Maya settle-ment focused in environmental settings that support in-tensification of hand cultivation, avoiding areas that aredifficult for these labor-based agricultural systems. Theexample from the El Pilar area reveals that the agricul-tural landscape is dispersed and fragmented, spreadingthe sites and populations out into large and small areaspreferred for settlement and agriculture with no pat-tern of focus on rivers. Because the whole society wasagrarian and dependent on the management of foodproduction, our predictive model suggests that farmingchoices outweighed other factors in the developmentof the overall landscape. Wider testing of this modelwill help to better understand the diversity across theancient Maya landscape.

Conclusion and Discussion

Late Classic Maya settlement distributions were ar-rived at over a long period of time. The patterns wehave discovered are likely the product of adjustmentsto the overall characteristics of the landscape. From thisstudy’s perspective, however, it is not relevant whetherthe patterned results were deliberate or based on cen-turies of trial and error. Overall, what our results show isthat four geographic and environmental factors predict82 percent of the Maya sites in the high-probability ar-

eas. Furthermore, fully 96 percent of the settlement, avast majority, can be predicted for less than 60 percentof the El Pilar study area.

Our conclusions are moderated by the uncertain-ties embedded in our data and in the assumptions ofthe Bayesian model. Given that the model is proba-bilistic, it might be difficult to use it for populationdensity forecasts. A critical value is how many settle-ments are missing from the record. Exogenous estimatesfrom cultural records or other sources could, never-theless, be used with the model to simulate possiblepopulation distributions. Replication of this model inother local areas of the Maya lowlands would be im-portant to build a broader database in support of ourmodel. In addition, testing its validity at other scales,such as the greater region-wide scale, would provideanother basis for interpreting land use and populationdistribution.

We propose that the probabilities computed for themodel here can be used as surrogates or proxies forland use intensity and population distribution. Sim-ple aggregation and integration would be sufficient totest the contemporary hypotheses about Maya popula-tion sizes and distributions. Thus, the foundation modeldeveloped here for the El Pilar area could shed lighton Maya farming sustainability as well as the infamousMaya “collapse.”

Predictive models will become increasingly useful asmore of the Maya archaeological record is destroyed.Although the likelihood of finding new major centersis not high, there must remain thousands of unmappedand unknown lesser centers, and certainly numerousfarm and house sites that could be located with thistool, as our validation efforts have shown. Modelinggeographic preferences could be a means of initial ar-chaeological surveys for areas that have had little atten-tion, such as the Lacandon forest of Chiapas, Mexico.

Given the broader implications, we believe that theuse of GIS data with Bayesian methods offers value topredictive mapping of archaeology in the Maya forest.Additional factors, perhaps religious and political, notto mention economic, are likely to explain variationfrom the expected patterns, and better environmentaldata would improve the current example. Finally, wesee a great potential of wider application of the model,perhaps to the whole Yucatan region, as well as a nar-rower one at a major center such as El Pilar, amongthe next steps. Of most interest for such future tests isthat the factor weights of the new models will showhow environmental influences vary by geographicalscale.



Acknowledgments

This research owes its inception to the generousResearch Across Disciplines program sponsored byUCSB’s Office of Research under Vice ChancellorFrance Cordova. The project benefited from the earlyinvestments from UCSB students Erica Gardener andAric Monts-Homkey. Engineering interns from EcoleSuperieure des Geometres et Topographes (Le Mans,France), including Anne Girardin, Elise Sirjean, andFlorent Monthus, improved this work with their dispas-sionate fieldwork and GIS knowledge while discoveringthe patterns of the ancient Maya landscape. We alsothank Belize’s Institute of Archaeology and particularlyJohn Morris and Jaime Awe for their continuity of sup-port of the research at El Pilar. We are equally gratefulto the communities around El Pilar for their generosityand assistance throughout our field endeavors. We, ofcourse, alone are responsible for our works.

Notes1. The term Maya forest refers to the moist forest ecosys-

tems of the lowland tropical parts of Maya culture arealocated between 15.45◦ N and 19◦ N latitude and 88◦ Wand 91◦ W longitude at elevations below approximately300 m (see also Whitmore and Turner 2001, 24, Table2.1).

2. There has been considerable controversy on the sub-ject of climatic change and the Maya area. Nonetheless,the general climatic regime is characterized as tropicalwith a wet–dry annual sequence that has continued overthe recent Holocene period (Curtis et al. 1998; Hodell,Brenner, and Curtis 2000; Brenner et al. 2002; Rosen-meier et al. 2002; Neff et al. 2006). This period hadincluded significant punctuated variations of deluge anddrought, particularly in short-term precipitation changesthat would have impacted the Maya (e.g., Haug et al.2001; Peterson and Haug 2005).

References

Alexandrian Digital Library (ADL). 2004. Version 1of the Digital Geographic Database for the MayaForest Region of Mexico, Guatemala and Belize.www.alexandria.ucsb.edu/adl (last accessed 26 April2009).

———. 2006. UCSB Maya Forest GIS 2000 www.alexandria.ucsb.edu/adl (last accessed 26 April 2009).

ArcGIS. Version 9.2. Redmond, WA: ESRI. www.ige.unicamp.br/sdm/ArcSDM31/default e.htm (last accessed27 April 2009).

ArcView. Version 3.2. Redmond, WA: ESRI. www.ige.unicamp.br/sdm/ArcSDM31/default e.htm (last accessed27 April 2009).

Ashmore, W., ed. 1981. Lowland Maya settlement patterns.Albuquerque: University of New Mexico Press.

———. 1991. Site planning principles and concepts of di-rectionality among the ancient Maya. Latin AmericanAntiquity 2 (3): 199–226.

Bacus, E. A., and L. J. Lucero, eds. 1999. Complex politiesin the ancient tropical world. Arlington, VA: AmericanAnthropological Association.

Balee, W. L. 1994. Footprints of the forest: Ka’aporethnobotany—The historical ecology of plant utilization byan Amazonian people. New York: Columbia UniversityPress.

Balick, M. J., and P. A. Cox. 1996. Plants, people, and culture:The science of ethnobotany. New York: Scientific Ameri-can Library.

Balick, M. J., M. H. Nee, and D. E. Atha. 2000. Checklist ofthe vascular plants of Belize with common names and uses.Bronx, NY: The New York Botanical Garden Press.

Beach, T., P. Farrell, and S. Luzzadder-Beach. 1998. Pre-hispanic agrosystems and adaptive regions in theMaya lowlands. Culture and Agriculture 20 (2–3): 87–101.

Birchall, C. J., and R. N. Jenkin. 1979. The soils of the BelizeValley, Belize Volume 1 & 2, 90. Surbiton, UK: LandResources Development Centre.

Bonham-Carter, G. 1994. Geographic information systems forgeoscientists: Modeling with GIS. Oxford, UK: Pergamon.

———. 1999. Integration of geological datasets for gold ex-ploration in Nova Scotia. Photogrammetric Engineeringand Remote Sensing 54:1585–92.

Boserup, E. 1965. The conditions of agricultural growth: The eco-nomics of agrarian change under population pressure. NewYork: Aldine.

Bray, F. 1994. Agriculture for developing nations. ScientificAmerican 271 (1): 18–25.

Brenner, M., M. F. Rosenmeier, D. A. Hodell, and J. H.Curtis. 2002. Paleolimnology of the Maya lowlands:Long-term perspectives on interactions among climate,environment, and humans. Ancient Mesoamerica 13:141–57.

Brumfiel, E. M., and T. K. Earle, eds. 1987. Specialization, ex-change, and complex societies. Cambridge, UK: CambridgeUniversity Press.

Bullard, W. R., Jr. 1960. Maya settlement pattern in north-eastern Peten, Guatemala. American Antiquity 25 (3):355–72.

Campbell, D. G., A. Ford, K. Lowell, J. Walker, J. K. Lake, C.Ocampo-Raeder, A. Townesmith, and M. Balick. 2006.The feral forests of the eastern Peten. In Time and com-plexity in the neotropical lowlands, ed. C. Erickson and W.Balee, 21–55. New York: Columbia University Press.

Chase, A. F., and D. Z. Chase. 1987. Investigations at the classicMaya city of Caracol, Belize: 1985–1987. Monograph 3.San Francisco: Pre-Columbian Art Research Institute.

Chase, D. E. Z., and A. F. Chase. 1988. A postclassic perspec-tive: Excavations at the Maya site of Santa Rita Corozal,Belize. San Francisco: Pre-Columbian Art ResearchInstitute.

Corzo Marquez, A. R., and N. B. Schwartz. 2008. Traditionalhome gardens of Peten, Guatemala: Resource manage-ment, food security, and conservation. Journal of Ethno-biology 28:305–17.



Culbert, T. P., ed. 1973. The Classic Maya collapse. Albur-querque: University of New Mexico Press.

———. 1988. The collapse of Classic Maya civilization. InThe collapse of ancient states and civilizations, ed. N. Yof-fee and G. L. Cowgill, 69–101. Tucson: University ofArizona Press.

Culbert, T. P., L. J. Levi, and L. Cruz. 1990. Lowland Mayawetland agriculture: The Rio Azul agronomy program.In Vision and revision in Maya studies, ed. F. S. Clancyand P. D. Harrison, 115–24. Albuquerque: University ofNew Mexico Press.

Culbert, T. P., and D. S. Rice, eds. 1990. Precolumbian popula-tion history in the Maya lowlands. Albuquerque: Universityof New Mexico Press.

Curtis, J. H., M. Brenner, D. A. Hodell, R. A. Balser, G.A. Islebe, and H. Hooghiemstra. 1998. A multi-proxystudy of Holocene environmental change in the Mayalowlands of Peten, Guatemala. Journal of Paleoliminology19:139–59.

Curtis, J. H., D. A. Hodell, and M. Brenner. 1996. Climatevariability on the Yucatan Peninsula (Mexico) duringthe past 3500 years, and implications for Maya culturalevolution. Quaternary Research 46 (1): 37–47.

Dahlin, B. H., T. Beach, S. Luzzadder-Beach, D. R. Hix-son, S. R. Hutson, A. Magnoni, E. Mansell, andD. E. Mazeau. 2005. Reconstructing agricultural self-sufficiency at Chunchucmil, Yucatan, Mexico. AncientMesoamerica 16 (2): 229–48.

Deevey, E. S., D. S. Rice, P. M. Rice, H. H. Vaughan, M.Brenner, and M. S. Flannery. 1979. Mayan urbanism—Impact on a tropical karst environment. Science 206(4416): 298–306.

Demarest, A. A. 1992. Ideology in ancient Maya culturalevolution: The dynamics of galactic polities. In Ideol-ogy and pre-Columbian civilizations, ed. A. A. Demarestand G. W. Conrad, 135–58. Santa Fe, NM: School ofAmerican Research Press.

Denevan, W. M. 1992. Stone vs metal axes: The ambigu-ity of shifting cultivation in prehistoric Amazonia. Jour-nal of the Steward Anthropological Society 20 (1–2): 153–65.

Diamond, J. 2005. Collapse: How societies choose to fail orsucceed. New York: The Penguin Group.

Dunning, N. P. 1996. Reexamination of regional variabilityin the pre-Hispanic agricultural landscape. In Managedmosaic: Ancient Maya agriculture and resource use, ed. S. L.Fedick, 53–68. Salt Lake City: University of Utah Press.

Dunning, N., T. Beach, P. Farrell, and S. Luzzadder-Beach.1998. Prehispanic agrosystems and adaptive regions inthe Maya lowlands. Culture and Agriculture 20 (4): 87–101.

Dunning, N. P., S. Luzzadder-Beach, T. Beach, J. G. Jones,V. Scarborough, and T. P. Culbert. 2002. Arising fromthe bajos: The evolution of a neotropical landscape andthe rise of Maya civilization. Annals of the Association ofAmerican Geographers 92:267–83.

Ewell, P. T., and M. Sands. 1987. Milpa in the Yucatan: Along fallow maize system and its alternatives in the Mayapeasant economy. In Comparative farming systems, ed. B.L. Turner and S. B. Brush, 95–129. New York: Guilford.

Food and Agriculture Organization of the United Na-tions (FAO). 2008. Key to the FAO soil units in

the FAO/Unesco soil map of the world. http://www.fao.org/ag/AGL/agll/key2soil.stm (last accessed 27 April2009).

Fedick, S. L. 1988. Prehistoric Maya settlement and land usepatterns in the Upper Belize River area, Belize, CentralAmerica. PhD dissertation, Arizona State University.

———. 1989. The economics of agricultural land use andsettlement in the Upper Belize River Valley. In Prehis-toric Maya economies of Belize, ed. P. A. McAnany andB. L. Isaac, 215–54. Greenwich, CT: JAI.

———. 1992. An agricultural perspective on prehistoricMaya household location and settlement density. InMemorias del primer Congreso Internacional de Mayistas;Mesas Redondas, Arquelogıa, Epigrafıa, ed. Centro de Es-tudios Maya UNAM, 87–108. Mexico City: UniversidadNacional Autonoma de Mexico.

———. 1994. Ancient Maya agricultural terracing in theUpper Belize area: Computer-aided modeling and theresults of initial investigations. Ancient Mesoamerica 5(1): 107–27.

———. 1995. Land evaluation and ancient Maya land usein the Upper Belize River area, Belize, Central America.Latin American Antiquity 6 (1): 16–34.

———, ed. 1996. The managed mosaic: Ancient Maya agricul-ture and resource use. Salt Lake City: University of UtahPress.

Fedick, S. L., and A. Ford. 1990. The prehistoric agriculturallandscape of the central Maya lowlands: An examinationof local variability in a regional context. World Archae-ology 22 (1): 18–33.

Ferguson, B., J. Vandeermeer, H. Morales, and D. Grif-fith. 2003. Post-agricultural succession in El Peten,Guatemala. Conservation Biology 17:818–28.

Foias, A. E., and R. L. Bishop. 1997. Changing ceramicproduction and exchange in the Petexbatun region,Guatemala: Reconsidering the classic Maya collapse.Ancient Mesoamerica 8:275–91.

Ford, A. 1985. Maya settlement pattern chronology in theBelize River area and the implications for the develop-ment of the central Maya lowlands. Belcast Journal ofBelizean Affairs 2:13–32.

———. 1986. Population growth and social complexity: An ex-amination of settlement and environment in the central Mayalowlands. Tempe: Arizona State University.

———. 1990. Maya settlement in the Belize River area:Variations in residence patterns of the central Maya low-lands. In Prehistoric population history in the Maya lowlands,ed. T. P. Culbert and D. S. Rice, 167–81. Albuquerque:University of New Mexico Press.

———. 1991a. Economic variation of ancient Maya residen-tial settlement in the Upper Belize River area. AncientMesoamerica 2 (1): 35–45.

———. 1991b. Problems with the evaluation of populationfrom settlement data: Examination of ancient Maya resi-dence patterns in the Tikal-Yaxha intersite area. Estudiosde Cultura Maya 18:157–86.

———. 1992. The ancient Maya domestic economy: An ex-amination of settlement in the Upper Belize River area.Paper presented at the Primer Congreso Internacionalde Mayistas, San Cristobal, Mexico.

———. 1993. Variaciones regionales de antiguos asen-tamientos Maya e implicaciones econonomicas para el



area superior del Rıo Belice [Regional variations ofancient Maya settlements and economic implicationsfor the Upper Belize River area]. Mesoamerica 25:39–61.

———. 1996. Critical resource control and the rise of theClassic period Maya. In The managed mosaic: AncientMaya agriculture and resource use, ed. S. L. Fedick, 297–303. Salt Lake City: University of Utah Press.

———. 2004. Integration among communities, centers, andregions: The case from El Pilar. In The ancient Maya ofthe Belize valley: Half a century of archaeological research,ed. J. Garber, 238–56. Gainesville: University Press ofFlorida.

———. 2006. Adaptive management and the community ofEl Pilar: A philosophy of resilience for the Maya forest. InOut of the past, for the future: Integrating archaeology andconservation, ed. N. Agnew and J. Bridgland, 105–12.Washington, DC: The Getty Institute.

———. 2008. Dominant plants of the Maya Forest and Gar-dens of El Pilar: Implications for paleoenvironmentalreconstructions. Journal of Ethnobiology 28:179–99.

Ford, A., and K. Clarke. 2000. UCSB Maya Forest GIS 2000.Santa Barbara: MesoAmerican Research Center, Uni-versity of California at Santa Barbara.

———. 2001. Mapping land use intensity using GIS. Washing-ton, DC: American Anthropological Association.

———. 2002. Out of the past, in the present, andfor the future of the Maya forest. Paper presentedat the Beyond Walls: Multi-Disciplinary PerspectivesConference, Belize City, Belize. http://www.uwichill.edu.bb/bnccde/belize/conference/paperdex.html (lastaccessed 27 April 2009).

Ford, A., and S. L. Fedick. 1992. Prehistoric Maya settlementpatterns in the Upper Belize River area: Initial resultsof the Belize River Archaeological Settlement Survey.Journal of Field Archaeology 19 (1): 35–49.

Ford, A., and K. Olson. 1989. Aspects of ancient Maya house-hold economy: Variation in chipped stone productionand consumption. In Prehistoric Maya economies of Belize,ed. P. A. McAnany and B. L. Isaac, 185–211. Greenwich,CT: JAI.

Ford, A., and M. Orrego. 1996. Investigaciones arqueologicaen la estructura EP7 El Pilar: Una interpretacion pre-liminar [Archaeological investigations of structure EP7El Pilar: A preliminary interpretation]. Santa Barbara:MesoAmerican Research Center, University of Califor-nia at Santa Barbara.

Ford, A., D. C. Wernecke, V. Bunton, W. C. Poe, N. Ger-ardo, and K. Veach. 1998. A new way of examining thepast: The El Pilar interdisciplinary model. 1998 Field Re-port, 33. Santa Barbara: MesoAmerican Research Cen-ter, University of California at Santa Barbara.

Ford, A., D. C. Wernecke, M. Grzybowski, and C. Ocampo.1996. The first steps towards revitalizing El Pilar. 1996Field Report, 59. Santa Barbara: MesoAmerican Re-search Center, University of California at Santa Barbara.

Fox, J. W., G. W. Cook, A. F. Chase, and D. Z. Chase.1996. Questions of political and economic integration:Segmentary versus centralized among the ancient Maya.Current Anthropology 37:795–801.

Freidel, D. A. 1981. The political economics of residen-tial dispersion among the lowland Maya. In Lowland

Maya settlement patterns, ed. W. Ashmore, 371–82. Al-buquerque: University of New Mexico Press.

———. 1986. Maya warfare: An example of peer polity inter-action. In Peer polity interaction and socio-political change,ed. C. Renfrew and J. F. Cherry, 93–108. Cambridge,UK: Cambridge University Press.

Fry, R. E., and S. C. Cox. 1974. The structure of ceramicexchange at Tikal, Guatemala. World Archaeology 6:209–25.

Fukuoka, M. 1978. The one-straw revolution. New York: Ban-tam Books.

Girardin, A. 1999. The El Pilar archaeological mappingproject: A geographic information systems objective.MA thesis, Ecole Superieure des Geometres et To-pographes, Le Mans, France.

Gomez-Pompa, A., M. F. Allen, S. L. Fedick, and J. J. Jimenez-Osornio, eds. 2003. The lowland Maya area: Three mil-lennia at the human–wildland interface. New York: FoodProducts Press.

Gomez-Pompa, A., J. S. Flores, and V. Sosa. 1987. The “petkot”: A man-made tropical forest of the Maya. Intercien-cia 12 (1): 10–15.

Gomez-Pompa, A., and A. Kaus. 1990. Traditional manage-ment of tropical forests in Mexico. In Alternatives to de-forestation: Steps toward sustainable use of the Amazon rainforest, ed. A. B. Anderson, 45–64. New York: ColumbiaUniversity Press.

———. 1999. From pre-Hispanic to future conservation al-ternatives: Lessons from Mexico. Paper presented at theNational Academy of Sciences Colloquium on Plantsand Population: Is There Time? Irvine, CA.

Gomez Pompa, A., and C. Vazquez Yanez. 1981. Successionalstudies of a rain forest in Mexico. In Forest succession:Concepts and application, ed. D. C. West, H. H. Shugart,and D. Botkin, 246–66. New York: Springer-Verlag.

Graham, E. 1998. Metaphor and metamorphism: Somethoughts on environmental metahistory. In Advancesin historical ecology, ed. W. Balee, 119–40. New York:Columbia University Press.

Griffith, D. M. 2000. Agroforestry: A refuge for tropical bio-diversity after fire. Conservation Biology 14 (1): 325–26.

Haberland, W. 1983. To quench the thirst: Water and set-tlement in Central America and beyond. In Prehistoricsettlement patterns: In honor of Gordon R. Willey, ed. E. Z.Vogt and R. M. Leventhal, 79–88. Albuquerque: Uni-versity of New Mexico Press.

Hammond, N., G. Tourtellot, S. Donaghey, and A. Clarke.1998. No slow dusk: Maya urban development and de-cline at La Milpa. Antiquity 72:831–37.

Haug, G. H., K. A. Hughen, D. M. Sigman, L. C. Peter-son, and U. Rohl. 2001. Southward migration of theintertropical convergence zone through the Holocene.Science 293 (5533): 1304–308.

Haviland, W. A. 1969. Tikal, Guatemala and Mesoamericanurbanism. World Archaeology 2:186–96.

Healy, P. F., C. G. B. Helmke, J. J. Awe, and K. S. Sunahara.2007. Survey, settlement, and population history at theancient Maya site of Pacbitun, Belize. Journal of FieldArchaeology 32 (1): 17–39.

Hernandez Xolocotzi, E., E. Bello Baltazar, and S. I. LevyTacher. 1995. La Milpa en Yucatan: Un sistema de



produccion agrıcola tradicional [The Milpa of Yucatan: Atraditional system of agricultural production], Vol. 1 & 2.Montecillo, Mexico: Colegio de Postgraduados.

Hodell, D. A., M. Brenner, and J. H. Curtis. 2000. Climatechange in the northern American tropics and subtrop-ics since the last ice age: Implications for environmentand culture. In Precolumbian New World ecosystems, ed.D. L. Lentz, 13–38. New York: Columbia UniversityPress.

Hodell, D. A., J. H. Curtis, and M. Brenner. 1995. Possiblerole of climate in the collapse of classic Maya civilization.Nature 375 (6530): 391–94.

Islebe, G. A., H. Hoogheimistra, M. Brenner, J. H. Curtis,and D. A. Hodell. 1996. A Holocene vegetation historyfrom lowland Guatemala. The Holocene 6:265–71.

Jenkin, R. N., R. Rose Innes, J. R. Dunsmore, S. H. Walker,C. J. Birchall, and J. S. Briggs. 1976. The agriculturaldevelopment potential of the Belize Valley, Belize. Sur-rey, UK: Land Resources Division, Ministry of OverseasDevelopment.

Kellman, M., and R. Tackaberry. 1997. Tropical environments:The functioning and management of tropical ecosystems.New York and London: Routledge.

Kohler, T. A., and S. C. Parker. 1986. Predictive modelsfor archaeological resource location. In Advances in ar-chaeological method and theory Vol. 9, ed. M. B. Schiffer,397–452. New York: Academic.

Levasseur, V., and A. Olivier. 2000. The farming system andtraditional agroforestry systems in the Maya communityof San Jose, Belize. Agroforestry Systems 49 (3): 275–88.

Levy Tacher, S. I., A. Rivera, and J. Rogelio. 2005. Suc-cessional pathways derived from different vegetation usepatterns by Lacandon Maya. Journal of Sustainable Agri-culture 26 (1): 49–82.

Leyden, B. W. 1987. Man and climate in the Maya lowlands.Quaternary Research 28:407–17.

Leyden, B. W., M. Brenner, D. A. Hodell, and J. H. Curtis.1993. Late Pleistocene climate in the Central Ameri-can lowlands. In Climate in continental isotopic records,ed. P. K. Swart, xiii, 374. Washington, DC: AmericanGeophysical Union.

Lohse, J. C., J. Awe, and C. Griffith. 2006. Preceramic oc-cupations in Belize: Updating paleoindian and archaicperiod. Latin American Antiquity 17:209–26.

Lucero, L. J. 2002. The collapse of the Classic Maya: A casefor the role of water control. American Anthropologist 104:814–26.

Marcus, J. 1973. Territorial organization of the lowland clas-sic Maya. Science 180 (4089): 911–16.

———. 1976. Emblem and state in the Classic Maya low-lands: An epigraphic approach to territorial organization.Washington, DC: Dumbarton Oaks Trustees for HarvardUniversity.

———. 1983. Lowland Maya archaeology at the crossroads.American Antiquity 48:454–88.

———. 1993. Ancient Maya political organization. Paperpresented at Lowland Maya Civilization in the EighthCentury A.D.: A Symposium at Dumbarton Oaks, Wash-ington, DC.

Martin, S., and N. Grube. 1995. Maya superstates: How a fewpowerful kingdoms vied for control of the Maya lowlands

during the Classic period (A.D. 300–900). Archaeology48 (6): 41–46.

———. 2000. Chronicle of the Maya kings and queens. London:Thames and Hudson.

Mollison, B. C. 1988. Permaculture: A designer’s manual. Sis-ters Creek, Tasmania, Australia: Tagari.

Monthus, F. 2004. Weights of evidence and Maya settle-ment. MA thesis, Ecole Superieure des Geometres etTopographes Le Mans, France.

Monts-Homkey, A. 2000. GIS applications in archaeology.Santa Barbara: MesoAmerican Research Center, Uni-versity of California at Santa Barbara.

Nash, J. 2007. Development to unite us: Autonomy and mul-ticultural coexistence in Chiapas and Guatemala. Jour-nal of Marxism and Interdisciplinary Inquiry 1 (1): 14–39.

Nations, J. D. 2006. The Maya tropical forest: People, parks,and ancient cities. Austin: University of Texas Press.

Nations, J. D., and R. Nigh. 1980. The evolutionary po-tential of Lacandon Maya sustained-yield tropical forestagriculture. Journal of Anthropological Research 36 (1): 1–30.

Neff, H., D. Persall, J. G. Jones, B. A. D. Pieters, and D. E.Freidel. 2006. Climate change and population history inthe Pacific lowlands of Southern Mesoamerica. Quater-nary Research 65:390–400.

Netting, R. M. 1977. Maya subsistence: Mythologies, analo-gies, possibilities. In The origins of Maya civilization, ed.R. E. W. Adams, 299–333. Albuquerque: University ofNew Mexico Press.

Nigh, R. 2008. Trees, fire and farmers: Making woods and soilin the Maya forest. Journal of Ethnobiology 28:231–43.

Pendergast, D. M. 1981. Lamanai, Belize: Summary of exca-vation results, 1974–1980. Journal of Field Archaeology 8(1): 29–53.

Peters, C. M. 2000. Precolumbian silviculture and indigenousmanagement of neotropical forests. In Imperfect balance:Landscape transformations in the Precolumbian Americas,ed. D. L. Lentz, 203–23. New York: Columbia UniversityPress.

Peterson, L. C., and G. H. Haug. 2005. Climate and the col-lapse of Maya civilization: A series of multi-year droughtshelped to doom an ancient culture. American Scientist93:322–29.

Powis, T. G., N. Stanchly, C. D. White, P. F. Healy, J. J.Awe, and F. Longstaffe. 1999. A reconstruction of Mid-dle Preclassic Maya subsistence economy at Cahal Pech,Belize. Antiquity 73 (280): 364–76.

Puleston, D. E. 1973. Ancient Maya settlement patterns and en-vironment at Tikal, Guatemala: Implications for subsistencemodels. Ann Arbor, MI: University Microfilms.

Puleston, D. E., and O. Puleston. 1972. A processual modelfor the rise of Classic Maya civilization in the southernlowlands. Paper presented at the International Congressof Americanists, Genova, Italy.

Raines, G. L. 1999. Evaluation of weights of evidence topredict epithermal-gold deposits in the Great Basin ofthe Western United States. Natural Resources Research 8(4): 257–76.

Raines, G. L., G. F. Bonham-Carter, and L. D. Kemp. 2000.Predictive probabilistic modeling using Arcview GIS.ArcUser 3 (2): 45–48.



Redfield, R., and A. Villa Rojas. 1962. Chan Kom: A Mayavillage. Chicago: University of Chicago Press.

Redman, C. L. 1999. Human impact on ancient environments.Tucson: University of Arizona Press.

Rice, D. S. 1976. The historical ecology of Lakes Yaxha andSacnab, El Peten, Guatemala. PhD dissertation, Penn-sylvania State University, University Park, PA.

———. 1978. Population growth and subsistence alter-natives in a tropical lacustrine environment. In Pre-Hispanic Maya agriculture, ed. P. D. Harrison and B. L.Turner, 35–61. Albuquerque: University of New MexicoPress.

Rice, D. S., and D. E. Puleston. 1981. Ancient Maya set-tlement patterns in the Peten, Guatemala. In LowlandMaya settlement patterns, ed. W. Ashmore, 121–56. Al-buquerque: University of New Mexico Press.

Rice, P. M. 1981. Evolution of specialized pottery produc-tion: A trial model. Current Anthropology 22 (3): 219–40.

Robin, C. 2002. Outside of houses: The practices of everydaylife At Chan Noohol, Belize. Journal of Social Archaeology2:245–68.

Rosenmeier, M. F., D. A. Hodell, M. Brenner, J. H. Cur-tis, and T. P. Guilderson. 2002. A 4000-year lacustrinerecord of environmental change in the southern Mayalowlands, Peten, Guatemala. Quaternary Research 57 (2):183–90.

Sabloff, J. A. 1986. Interaction among Classic Maya poli-ties: A preliminary examination. In Peer polity interac-tion and socio-political change, ed. C. Renfrew and J. F.Cherry, 109–16. Cambridge, UK: Cambridge UniversityPress.

Sader, S. 1999. Deforestation trends in northern Guatemala:A view from space. In 13 ways of looking at a tropicalforest: Guatemala’s biosphere reserve. Washington, DC:Conservation International.

Sader, S. A., R. R. Chowdhury, L. C. Schneider, and B. L.Turner. 2004. Forest change and human driving forcesin Central America. In Land change science: Observ-ing, monitoring, and understanding trajectories of changeon the Earth’s surface, ed. A. J. G. Gutman, C. Jus-tice, E. Moran, J. Mustard, R. Rindfuss, D. Skole, andB. L. Turner, II, 379–94. Dordrecht, The Netherlands:Kluwer.

Sawatzky, D. L., G. L. Raines, G. F. Bonham-Carter, andC. G. Looney. 2004. ARCSDM3.1: ArcMAP extensionfor spatial data modelling using weights of evidence, lo-gistic regression, fuzzy logic and neural network anal-ysis. http://www.ige.unicamp.br/sdm/default e.htm (lastaccessed 10 January 2007).

Scarborough, V. L. 1994. Ancient water management in thesouthern Maya lowlands. Research & Exploration 10 (2):184–200.

———. 1998. Ecology and ritual: Water management andthe Maya. Latin American Antiquity 9 (2): 135–59.

Scarborough, V. L., and G. G. Gallopin. 1991. A water stor-age adaptation in the Maya lowlands. Science 251 (4994):658.

Sever, T. L. 1999. The ancient Maya landscape from space.In 13 ways of looking at a tropical forest: Guatemala’s Mayabiosphere reserve, ed. J. D. Nations, 20–25. Washington,DC: Conservation International.

Sever, T. L., and D. E. Irwin. 2003. Landscape archaeol-ogy: Remote-sensing investigation of the ancient Mayain the Peten rainforest of northern Guatemala. AncientMesoamerica 14 (1): 113–22.

Sifuentes, J. M. 2005. Geographical information systemsdata implementation for analyses of settlement pat-terns of an early agricultural society. PhD dissertation,University of California, Santa Barbara, Santa Barbara,CA.

Sirjean, E. 2003. Predictive modeling and the Maya of El Pilar.Santa Barbara: MesoAmerican Research Center, Uni-versity of California at Santa Barbara.

Smith, T. R., and J. Frew. 1995. Alexandria Digital Library.Communications of the ACM 38 (4): 61–62.

Spatial Ecology. 2008. Hawth’s tools. www.spatialecology.com/htools 2008 (last accessed 28 April 2009).

Teran, S., C. H. Rasmussen, and O. M. Cauich. 1998. Lasplantas de la Milpa entre los Mayas: Etnobotanica de lasplantas cultivadas por los campesinos mayas en las milpas delnoroeste de Yucatan, Mexico [The plants of the Milapa of theMaya: Ethnobotany of cultivated plants of Maya Milapas ofNE Yucatan Mexico]. Yucatan, Mexico: Fundacion TunBen Kin, A.C.

The Nature Conservancy. 2007. The Maya forest. http://www.nature.org/wherewework/northamerica/mexico/work/art8622.html (last accessed 6 October 2007).

Turner, B. L. 1978. Ancient agricultural land use in the cen-tral Maya lowlands. In Pre-Hispanic Maya agriculture, ed.P. D. Harrison and B. L. Turner, 163–83. Albuquerque:University of New Mexico Press.

———. 1990. Population reconstruction of the central Mayalowlands: 1000 B.C. to A.D. 1500. In Precolumbian pop-ulation history in the Maya lowlands, ed. T. P. Culbert andD. S. Rice, 301–24. Albuquerque: University of NewMexico Press.

Villa Rojas, A. 1945. The Maya of east central Quintana Roo.Washington, DC: Carnegie Institution.

Voorhies, B. 1982. An ecological model of the early Mayaof the central lowlands. In Maya subsistence: Studies inmemory of Dennis Edward Puleston, ed. K. V. Flannery,65–95. New York: Academic.

Webster, D. 2002. The fall of the ancient Maya: Solving themystery of the Maya collapse. New York: Thames &Hudson.

———. 2005. Political ecology, political economy, and theculture history of resource management at Copan. InCopan: The history of an ancient Maya kingdom, ed. E. W.Andrews and W. L. Fash, 33–72. Albuquerque: Schoolof American Research.

Whitmore, T. M., and B. L. Turner, II. 2001. Cultivated land-scapes of Middle America on the eve of conquest. Oxford,UK: Oxford University Press.

Wilken, G. C. 1987. Good farmers: Traditional agricultural re-source management in Mexico and Central America. Berke-ley: University of California Press.

Willey, G. R., W. R. Bullard, J. B. Glass, and J. C. Gifford.1965. Prehistoric Maya settlements in the Belize Val-ley. In Papers of the Peabody Museum of Archaeology andEthnology, ed. G. R. Willey, xxix, 589. Cambridge, MA:Peabody Museum.

Zetina Gutierrez, M. d. G. 2007. Ecologıa humana de lasrancherıas de Pich, Campeche: Un analisis diacronic



[Human ecology of the rancherias of Pich Campeche:A diachronic analysis]. Masters thesis, Centro de Inves-tigacion y Estudios Avanzados del Instituto PolitecnicoNaciona, Merida.

Zetina-Gutierrez, M., and B. Faust. 2006. The shift from no-madic agriculture to tractors in Pich, Campeche, Mexico.Paper presented at the annual meetings of the AmericanAnthropological Association, San Jose, CA.

Correspondence: ISBER/MesoAmerican Research Center, University of California, Santa Barbara, CA 93106–2150, e-mail: [email protected] (Ford); Department of Geography, University of California, Santa Barbara, CA 93106–4060, e-mail: [email protected] (Clarke);940 Sumack Reno, NV 89509, e-mail: [email protected] (Raines).


please scroll down for article - el pilar af modeling set patterns lcm...modeling settlement...

Documents