Rapa Nui Journal Vol. 25 (2) October 2011

The tempo of change in the leeward Kohala eld system,

Hawai‘i Island

Thomas S. Dye

Abstract

Reanalysis of radiocarbon dates that pre-date features of

the leeward Kohala ﬁeld system on Hawai‘i Island was

carried out within a Bayesian statistical framework. Results

of the analysis indicate that features of the ﬁeld system

were developed late in traditional Hawaiian times. Many of

the features appear to have been constructed subsequent to

Cook’s visit in AD 1779. These results do not support the

hypothesis that agricultural intensiﬁcation began in the early

seventeenth century, linked to a rise in the authority of chiefs.

Introduction

Archaeology’s radiocarbon revolution has been a blessing

and a curse for archaeologists working in Hawai‘i. When

the method was ﬁrst applied in the early 1950s it appeared

to offer a scientiﬁc way to measure time that would be an

improvement over relative dating methods that had yielded

poorly in Hawai‘i (Dye 2010). In practice, however, many

of the

C age determinations returned by dating laboratories

proved difficult to interpret sensibly. Over the years,

archaeologists have responded with a variety of interpretive

schemes, all of them ad hoc in the sense that they are not

based on an explicit chronological model. Ad hoc interpretive

schemes are certainly capable of yielding good results, but

the history of their application in Hawai‘i is symptomatic of

an unscientiﬁc method. This is perhaps easiest to see in the

case of Polynesian colonization, a question that has been at

the forefront of archaeological research in Hawai‘i since the

dawn of the radiocarbon revolution. In science, a properly

formulated solution yields increasingly accurate and precise

results as the number of relevant observations grows. In

contrast, the ad hoc interpretive approaches have disdained

precision, proposing colonization date estimates without

corresponding error terms, and have failed to converge on

a solution (Dye 2011). Ad hoc estimates of the colonization

event proposed over the last two decades range over an

eye-opening 1,200 years.

This paper argues that the failure of ad hoc interpretive

methods is systemic. Statements about what happened in

old Hawai‘i based on ad hoc interpretations often reﬂect

failures of the method more than they do events in the

past. An example is a general statement about sequences of

agricultural development across the archipelago.

“… the chronological development of the Kohala,

Kona, Waimea, Kahikinui, and Kalaupapa ﬁeld systems,

spanning three islands, is remarkably congruent. While

there was some low intensity land use in Kohala and Kona

prior to AD 1400, in all cases the onset of major dryland

cultivation began around AD 1400. Following about two

centuries of development, a ﬁnal phase of intensiﬁcation,

typically marked by highly formalized garden plots and

territorial boundaries, commenced about AD 1600 to

1650, and continued until the early post-contact period.

Unlike the irrigation systems, many of which have

continued in use throughout the nineteenth and twentieth

centuries, the dryland field systems were all rapidly

abandoned within a few decades following European

contact” (Kirch 2010: 153).

This statement is part of a larger argument about the

chronology of changes in ali‘i authority (Kirch 2010: 77 ff.),

which has its basis in interpretations of traditions that were

transmitted by the ruling ali‘i, and which served to legitimate

their rule. The period boundaries for agricultural development

deﬁne evenly spaced, approximately two-century intervals

that link features of the contact-era political situation with

origination points identiﬁed by interpretation of the traditions.

The “onset of major dryland cultivation” in AD 1400 is when

some scholars believe the traditions become historically

accurate, in the western sense of that term (Kirch 2010: 81). In

this interpretation of the traditions, ali‘i history begins around

AD 1400. The “ﬁnal phase of intensiﬁcation” around AD 1600

marks the ﬁrst Gregorian century in which the traditions are

interpreted to indicate that Hawai‘i and Maui Islands were both

ruled by paramount chiefs. The Hawai‘i Island ali‘i, ‘Umi a

Līloa, whose reign was later used to legitimate Kamehameha

the Great’s usurpation of the Hawai‘i Island paramountcy on

his way to uniting the islands, ruled at about this time. Thus,

ﬁeld system developments are seen as congruent among

themselves and also with a particular interpretation of the

development of political authority in traditional Hawai‘i.

Thomas S. Dye | T.S. Dye & Colleagues, Archaeologists, 735 Bishop St, Suite 315, Honolulu, HI 96813 USA. tsd@tsdye.com

Rapa Nui Journal Vol. 25 (2) October 2011

The tempo of change in the leeward Kohala ﬁeld system, Hawai‘i Island

program that speciﬁcally excavated beneath agricultural

walls and under curbstones of trails to identify termini

post quem for wall and curb construction events. The

detailed dating record they have produced offers analytic

opportunities that are unmatched in Hawaiian archaeology.

It is the only dating record from the leeward Kohala ﬁeld

system capable of yielding the analytic precision required

to evaluate the proposed temporal congruence.

This rich set of data is analyzed here with Bayesian

methods (Buck et al. 1996), which build a detailed

chronological model of ﬁeld system development from the

stratigraphic relations of the agricultural walls and trails, and

then ﬁx this model in time with the

C age determinations.

The

C age determinations from the field system are

generally quite young and their combination with the detailed

chronological model yields results that are more precise than

typically achieved in Hawai‘i. The precision of the results

adds strength to the observation that they are not congruent

with the chronology of political development yielded by the

interpretation of ali‘i traditions. Instead of the steady march

of change implied by the ad hoc interpretation, the Bayesian

analysis indicates that the tempo of change varied over time.

Much of what we recognize today as the ﬁeld system—most of

the walls and many of the trails—was built during a brief pulse

of intensiﬁcation at the end of the sequence. In fact, much

of the construction appears to have taken place within the

historic period, which suggests that contingent events might

have played a larger role in agricultural development than

the interpretation of ali‘i traditions would lead one to expect.

Figure 1. Periodization of ﬁeld system features by building phase (after Ladefoged and Graves 2008: Figure 7).

It is argued here that congruencies such as these are,

in part, artifacts of the ad hoc methods used to interpret

the dating evidence. Ad hoc methods spawn two kinds of

errors, both of which bolster the appearance of congruence.

First, their disdain for uncertainty conceals the fact that age

estimates for some key events are very imprecise. In these

cases, linking the archaeological record to a precise time

doesn’t constitute archaeological support for a particular

hypothesis. Rather, it reﬂects an assumption of the hypothesis

to shore up weaknesses in the archaeological results. Second,

the ad hoc methods operate outside a coherent statistical

framework and are typically wasteful of chronological

information. They yield relatively weak results. Precise

results with the potential to distinguish one chronology from

another are thus kept out of reach, leaving the impression of

congruence intact.

This general argument is made by way of a speciﬁc

example, a model-based calibration and re-interpretation of

the developmental chronology of a portion of the leeward

Kohala ﬁeld system on Hawai‘i Island. The leeward Kohala

ﬁeld system offers a unique opportunity in this regard.

As Rosendahl (1972) pointed out many years ago, the

fabric-like structure of the ﬁeld system—trails that connect

the ﬁeld system to the coast provide the warp for the weft of

agricultural walls that divide ﬁelds from one another—yields

an opportunity to establish relative ages of features at every

intersection of a wall with a trail. Within this rich mesh of

chronologically ordered construction events, Ladefoged

and Graves (2008) have carried out a sophisticated dating

Rapa Nui Journal Vol. 25 (2) October 2011

Figure 3. Chronological model of ﬁeld system features. Features are labeled as in Figure 2.

Thomas S. Dye

Figure 2. Diagram of the detailed study area. Uphill is toward the top of the diagram. Trails are

indicated by capital letters in boxes and walls by lower case letters in ovals. The labels of dated

features are shaded gray.

Rapa Nui Journal Vol. 25 (2) October 2011

The tempo of change in the leeward Kohala ﬁeld system, Hawai‘i Island

Relative Chronology of Field System

Development

The relative chronology of ﬁeld system development in the

detailed study area has been addressed in two publications

(Figure 1) (Ladefoged et al. 2003; Ladefoged & Graves

2008). In these cases, the ﬁeld system features were assigned

to building phases or temporal units; two similar analytic

constructs that group features based on stratigraphic relations

and a set of propositions or assumptions independent of the

stratigraphic relations. The chronological model used in the

Bayesian calibration is based solely on stratigraphic relations,

so it can’t be based on the published building phases or

temporal units (Figure 2).

Accordingly, a Harris matrix (Harris 1989) of the

dated ﬁeld system features was constructed (Figure 3). In

addition to the dated ﬁeld system features, also included in

the Harris matrix are undated features that show the relative

stratigraphic relations of dated features that don’t intersect

one another, but whose relationship can be determined with

the map evidence. Figure 3 thus represents the components

of the full Harris matrix for the detailed study area needed

to construct a chronological model for the Bayesian

calibration. This ﬁgure represents the chronological model

that distinguishes Bayesian from ad hoc interpretations.

Figure 3 is a directed acyclic graph, also known in the

literature as a DAG or an acyclic digraph. The properties

of DAGs are well known and graph theory has developed

terminology that makes it possible to talk about them in a

precise way (Hage & Harary 1983: 65 ff.). This can be a

tremendous advantage when trying to conceptualize and

work with a structure as large and complex as the detailed

study area (Figure 2). It would appear to be essential to any

attempt to deal with larger sections of the ﬁeld system or to

comprehend the ﬁeld system whole.

The properties of a DAG make it an ideal graph structure

to represent a Harris matrix. A DAG consists of a ﬁnite set

of points and a collection of ordered pairs of points, known

as arcs (Hage & Harary 1983: 68). The directed property of

the DAG refers to the fact that each arc consists of an ordered

pair of points, or a direction that runs from the ﬁrst point

to the second point. In the context of a Harris matrix, the

direction of an arc encodes the stratigraphic relation “older

than/younger than.” In Figure 3, the arrows used for the arcs

of the graph point from an older feature to a younger feature;

each arrow represents an observed stratigraphic relationship

at the intersection of a trail and a wall. The acyclic property

of the DAG means that there is no sequence of points and

arcs, where the points of each arc are in order, that starts

and ends at the same point. The lack of cycles in the graph

ensures that no feature can be either older or younger than

itself, which is a requirement of the stratigraphic model.

Figure 3 is laid out with the arrows pointing down, so

older features are at the top of the graph and younger features

are at the bottom. The structure of the graph, with alternating

rows of walls and trails, reﬂects the nature of the evidence;

none of the walls cross another wall, and none of the trails

cross another trail. Two features are related chronologically

if and only if one is reachable from the other; two points in a

digraph are reachable if it is possible to move from one to the

other in the direction of the arcs. Walls g and d, for instance,

are reachable from the same set of features, which includes

walls c, i, j, k, and h and trails B and C. They are not, however,

reachable from one another. Thus, although the stratigraphic

relations indicate that both walls are younger than trail C, it

is not possible to tell on the basis of the relative stratigraphic

information which of the two was built before the other.

The graph of Figure 3 is weakly connected because

it contains pairs of points that are not reachable from one

another. This occurs fairly frequently in situations like the

one discussed above with walls d and g, where the walls are

physically close to one another and share similar stratigraphic

relations to neighboring trails. It also occurs frequently

with walls on opposite sides of a trail. For example, walls

d and e are both younger than trail B, but it is not possible

to determine on stratigraphic grounds which of the walls

is older than the other. It is true that wall e is older than

trail A and that wall d is younger than trail C, but there is

no stratigraphic information on the relative ages of trails A

and C, so this information does not yield a temporal order

for the two walls. The fact that the periodization of Figure 1

assigns relative ages to these two walls, and to others that

share similar stratigraphic relations, is an indication that the

building phases it proposes are not strictly chronological.

These two examples of weak connections are both local

in scope. However, weak connections also occur at points

that distinguish larger sections of the ﬁeld system, and these

might provide clues to the history of development. The prime

example of this in the detailed study area is wall b. None

of the points that reach wall b from the left hand side of

Figure 3 is reachable from any of the points that reach wall b

from the right hand side of the ﬁgure. Thus, the stratigraphic

structure of the detailed study area is broken between trails

C and D in Kahua 1.

C Dating of Field System Features

Table 1 lists 21 of the 25

C age determinations associated

with agricultural features in the leeward Kohala ﬁeld system

published by Ladefoged and Graves (2008: Table 1). It

includes all 17

C age determinations from the detailed

study area at Pāhinahina and Kahua 1, along with four

of the eight

C age determinations from features outside

the detailed study area. All of the age determinations in

the table are on short-lived materials. The four excluded

C age determinations are on materials identiﬁed as dicot

wood. They were excluded because of the potential in-built

age carried by this material. The

C age determinations all

derive from archaeological contexts that “date activities that

occurred before the construction of the agricultural walls”

Rapa Nui Journal Vol. 25 (2) October 2011

or that “pre-date the construction of the trails” (Ladefoged

& Graves 2008: 778).

Table 1 provides the label assigned to the age

determination by Ladefoged and Graves (2008) in the

last column; the label assigned by the dating laboratory

in column 4; the wall or trail feature with which the age

determination is associated, keyed to Figure 2, in column

2; and the calibration group to which the determination has

been assigned in column 3. The values in the ﬁrst column,

labeled θ, identify the age determinations in the Bayesian

analysis. Technically, in the Bayesian model each θ represents

the true calendar age of the sample, which is estimated by the

corresponding

C age determination. The values in the table

start with θ

and run through θ

. This is because the ﬁeld

system calibration is carried out in the context of an estimate

of when the islands were initially colonized by Polynesians,

which requires seven age determinations assigned to θ

1…7

(Dye 2011). The column labeled “Outlier” is an analytic

result, discussed below.

A striking feature of Table 1 is that most of the

C age

determinations are relatively young. This is the case even for

C age determinations associated with the oldest features in

the detailed study area. Two of the

C age determinations

associated with Group 1 walls are less than 300

C years old,

and the youngest of these, associated with wall c, dates to 200

± 40 BP. The sample collected from beneath the curbstone of

the oldest trail, trail B, dates to 130 ± 30 BP. Keeping in mind

that these

C age determinations pre-date construction of the

associated features, and that the ﬁeld system was abandoned

“within a few decades following European contact” (Kirch

2010: 153), or about 100 BP, it would appear that most of

the features in the detailed study area were built within the

span of about 100

C years.

Because a 100

C year span seems too brief for

construction of the field system facilities, an analysis

was performed to identify outliers among the

C age

determinations (Christen 1994). The expectation was that

the young age determinations associated with the oldest

Thomas S. Dye

Table 1.

C age determinations.

See http://www.tsdye.com/research/tempo.html.

† See Figure 2.

‡ See Figure 1 and http://www.tsdye.com/research/tempo.html.

§ Conventional

C age (Stuiver & Polach 1977).

Source: Ladefoged & Graves (2008).

Rapa Nui Journal Vol. 25 (2) October 2011

The tempo of change in the leeward Kohala ﬁeld system, Hawai‘i Island

features would be identiﬁed as outliers and could be removed

from the calibration. The results of the outlier analysis are

presented in column 6 of Table 1 as the difference between

an uninformative prior probability assigned to each

C age

determination and the posterior probability returned by the

analysis. Negative numbers indicate

C age determinations

that are less likely to be outliers than was estimated by the

prior probability and positive numbers indicate

C age

determinations that are more likely to be outliers. The outlier

identiﬁcation procedure doesn’t establish a metric for how

big this difference must be for a

C age determination to be

considered an outlier. In practice, the analyst uses the results

to draw attention to particular

C age determinations and

these are scrutinized as necessary before a decision is made

either to keep them in the analysis or discard them as outliers.

The results of the outlier analysis indicate that there

is no reason to question the integrity of most of the age

determinations. The young age determination from under the

curbstone of trail B and the age determination associated with

wall i in Group 1 are not outliers. The only age determination

possibly indicated by the analysis as an outlier is the age

determination associated with wall c. Ladefoged and Graves

(2008: 779) don’t discuss this particular age determination

and it appears not to have played a role in their interpretation

of the dating results. However, there are several reasons why

this age determination should not be treated as an outlier:

(i) the dating model typically has few age determinations

per group and this makes outlier determination less reliable

than it would be with more samples; (ii) the result returned

by the outlier analysis is not particularly strong—the prior

probability of 0.1 increased to 0.3, about a quarter of the

possible maximum; (iii) the

C age determination is only

C years younger than the next youngest sample from

beneath a Group 1 wall; (iv) the

C age determination

associated with the feature immediately younger than it,

trail B, is stratigraphically correct and about 70

C years

younger than it; and (v) charcoal from the later swidden

activities might be relatively rare if, as appears to be the

case, secondary growth were consistently used as a source

of mulch, or if burned secondary growth consisted mostly

of grasses (Kirch 2010: 53). On balance, then, there appears

to be no compelling reason to discard this age determination

as an outlier. However, this is an issue that might repay

identiﬁcation and dating of additional samples from beneath

Group 1 walls.

Developmental Periods and Their Boundaries

The history of the leeward Kohala ﬁeld system is typically

described according to a theory of agricultural development

that distinguishes processes of expansion and intensiﬁcation

(Kirch 2010; Ladefoged & Graves 2008, 2010). The process

of expansion involves “conversion of previously unused

areas to cultivation” (Ladefoged & Graves 2010: 95). It is

recognized archaeologically beneath the oldest ﬁeld system

walls in units of stratiﬁcation that “show clear signs of

clearing or cultivation, such as digging stick holes, churned

sediments, and charcoal lenses or ﬂecking” (Ladefoged &

Graves 2008: 778). The process of intensiﬁcation increases

“the amount of labor in a ﬁxed area of land to increase

production” (Ladefoged & Graves 2010: 95). It is recognized

archaeologically by construction of the ﬁeld system walls.

In use, the walls were typically planted with sugar cane that

helped them serve as windbreaks, which increased yields by

protecting crops from the famous Kohala winds and reducing

evapotranspiration (Ladefoged & Graves 2010: 94).

The periods of expansion and intensiﬁcation can be

augmented with two additional periods that set the leeward

Kohala ﬁeld system within the framework of a ﬁrst-order

cultural sequence for Hawai‘i. The ﬁrst of these embraces

the time between Polynesian colonization and the onset of

agricultural expansion. The land that would later become

the leeward Kohala ﬁeld system lay undeveloped and was

either unused or used so lightly that archaeologists are unable

to detect it. At the other end of the sequence is the time

since the ﬁeld system was abandoned in the mid-nineteenth

century. Historically, use of the area during this period was

for cattle ranching, but other commercial activities have been

attempted, all of them made possible by the introduction of

certain property rights and the alienability of land during

the Māhele (Banner 2005; Chinen 1958, 2002; Moffat &

Fitzpatrick 1995). For ease of reference, the periods are

here labeled Colonization, Expansion, Intensiﬁcation, and

Alienation. The model was calibrated with the BCal software

package (Buck et al. 1999).

Estimates of the period boundaries yielded by the

Bayesian calibration are shown in Figure 4. The colonization

event is based on model (3) of Dye (2011), which includes a

C age determination on rat bone from the ‘Ewa Plain that

did not control for the possibility of a marine component

in the rat’s diet that would make the bone appear too old.

Model (3) was used because it yields a relatively precise

estimate of the colonization event, but one which maintains

the central tendency of the less precise estimate without the

rat bone date (Dye 2011). Still, the 67% highest posterior

density (HPD) region of the estimate, analogous to the one

standard deviation error term of frequentist statistics, covers

almost two centuries. The 95% HPD region, analogous to

two standard deviations, spans more than three centuries.

The distribution is centered around AD 980 and is relatively

symmetrical.

The estimate for the beginning of the Expansion period

is slightly more precise than the estimate of the Colonization

period. The 67% HPD covers about 120 years and the 95%

HPD about 280 years. The central tendency of the distribution

is clearly within the fourteenth century; probabilities drop

off quickly after AD 1400, and the long, low early tail takes

in the eleventh through thirteenth centuries.

The precision of the estimate improves markedly in

the Intensiﬁcation period, due primarily to the constraints

Rapa Nui Journal Vol. 25 (2) October 2011

Figure 4. Period boundary estimates. The 67% highest posterior density regions are: top left, AD

860–1029; top right, AD 1290–1409; bottom left, AD 1640–1729; bottom right, AD 1850–1869.

Figure 5. Chronology of dated features in the leeward Kohala ﬁeld system detailed study area. See

Table 2 for estimates of precision and Figure 3 for the deﬁnition of groups.

Thomas S. Dye

Rapa Nui Journal Vol. 25 (2) October 2011

The tempo of change in the leeward Kohala ﬁeld system, Hawai‘i Island

imposed by chronological relations of the ﬁeld system

features (Figure 3). Given the model and current evidence,

the 67% HPD covers 90 years and the 95% HPD covers 210

years. The distribution has a marked peak around AD 1680

that falls rapidly in the eighteenth century but has a long,

low early tail that extends through the sixteenth century.

The estimate for the Alienation period is included on

Figure 6 for the sake of completeness. This period boundary

is a ﬂoating parameter in the model that was modeled as a

normal curve with a ten year standard deviation centered at

AD 1850. Land records from the Māhele appear to indicate

that the ﬁeld system was abandoned by the middle of the

nineteenth century. In any event, archaeological excavations

in the field system did not yield information on the

abandonment event, so the estimate yielded by the Bayesian

calibration is mainly a reﬂection of the prior probability.

Estimates for the construction of facilities within the

detailed study area are shown in Figure 4 and the precisions

of the estimates are listed in Table 2. The high precision of

these estimates is due to the many constraints yielded by

the stratigraphic relations of the trails and walls (Figure 3)

and to the apparent brevity of the Intensiﬁcation period.

The estimate for Group 1 is also the estimate for the onset

of Intensiﬁcation and was discussed earlier. Group 2 dates

the construction of the curb along trail B, which marks

the boundary between Pāhinahina and Kahua 1. This trail

appears to have been built early in the eighteenth century.

The distribution of the estimate is centered on AD 1720,

with a 67% HPD region that spans 60 years. The Pāhinahina

agricultural walls e and f, in Group 3, are estimated to

have been constructed around the middle of the eighteenth

century. The distribution of the estimate is centered on

AD 1760. The 67% HPD region spans 70 years. Trail C,

in Kahua 1, but structurally associated with features in

Pāhinahina , appears to have been built around the turn of

the nineteenth century. The 67% HPD region for this event

spans 60 years. Finally, the two Pāhinahina walls in Group

5a and the two Kahua 1 walls in Group 5b are estimated

to be penecontemporaneous. The estimates for these two

groups both peak around AD 1840 and both have 67% HPD

regions that span 50 years.

Tempo of Change

An alternative view of the calibration results takes the focus

away from estimates of period boundaries and puts it instead

on estimates of period duration. Figure 6 shows duration

estimates for the Colonization, Expansion, Intensiﬁcation,

and Alienation periods.

The Colonization and Expansion periods are both

relatively long, on the order of three to ﬁve centuries, and

imprecisely estimated, with 67% HPD regions between 160

and 260 years. In contrast, the Intensiﬁcation and Alienation

periods are relatively short. Most of the difference in their

durations is due to a convention of

C dating that deﬁnes

Present as AD 1950. Adding an extra 60 years to the

length of the Alienation period would shift its distribution

to the right and bring it almost precisely in line with the

Intensiﬁcation period. Duration estimates for both periods

are relatively precise, although, as noted above, uncertainty

in the duration of the Alienation period is mostly an artifact

of the model’s assumptions.

Discussion

The extended quote in the introduction of this paper

(Kirch 2010: 153) is structured as an origin narrative. Like

other origin narratives, it has two goals—to establish the

plausibility of the events and processes it projects onto the

past, and to claim authority by locating them at particular

times (Moore 1995). This particular origin narrative identiﬁes

the processes of agricultural expansion and intensiﬁcation

and ﬁxes their origins at AD 1400 and 1600-1650, two times

that an interpretation of tradition ﬁnds important in the rise

of ali‘i authority. The regularity of the process identiﬁed

in the narrative—200 years of expansion followed by 200

years of intensiﬁcation into the early historic period—gives

it an aura of inevitability, as if the present were predicted by

the origin events in its past. Bayesian calibration yields the

precise dating results with which to evaluate these claims

about agricultural development, at least in a portion of the

leeward Kohala ﬁeld system.

The expansion process, whose origin is described as

“the onset of major dryland cultivation” is hypothesized to

have originated about AD 1400. This is a time when land

was cleared for cultivation of sweet potato, a crop plant

that originated in America and was introduced to Eastern

Polynesia by voyagers who made the return trip to the coast

of South America (Storey et al. 2007). On present evidence,

it was introduced to Hawai‘i some three to six centuries after

the islands were colonized (Dye 2011: Table 2). Excavations

in the leeward Kohala ﬁeld system collected a charred tuber

tentatively identiﬁed as sweet potato that represents the

earliest dated occurrence of the plant in Hawai‘i (Ladefoged

et al. 2005). The

C age determination for this probable

sweet potato tuber, Beta-208143, is the oldest from the ﬁeld

system (Table 1), and thus marks the onset of the Expansion

Table 2. Precision of estimates for facility construction.

Rapa Nui Journal Vol. 25 (2) October 2011

period. Ladefoged and Graves (2008: 779) interpreted this

information as placing the start of the Expansion period “as

early as AD 1290 but certainly by AD 1430.” The Bayesian

calibration relies on the same evidence for its estimate and

gives a similar result; stratigraphic relations that might

constrain the calibrated age of this sample are absent. The

date of AD 1400 for the expansion process singled out by

the origin narrative falls at the late end of this range. It is a

plausible estimate for the onset of the Expansion period, but it

is only one of very many plausible estimates. The calibration

results from the detailed study area are equally “congruent”

with an origin of the Expansion period anytime in the

fourteenth century or even a bit earlier. The archaeological

information is less certain than the origin narrative implies.

In this case, the origin narrative is imposing its structure on

the archaeological data rather than the other way around.

The second process identiﬁed in the origin narrative is

“a ﬁnal phase of intensiﬁcation” that “commenced about

AD 1600 to 1650.” This range of dates is at odds with the

interpretation put forward by Ladefoged and Graves (2010),

who believed the intensiﬁcation started earlier. They assign

early construction dates to walls j and k in Group 1 based on

the presence of relatively old charcoal beneath them. In their

view, this puts the start of the Intensiﬁcation period “as early

as AD 1410 but possibly not until AD 1630” (Ladefoged &

Graves 2010: 779). This inference appears to be based on a

logical error, however. It is only possible to know that the

charcoal collected under a wall is older than the wall; it is not

possible to know, in the absence of other information, how

much older it is. The Bayesian calibration corrects this logical

error and yields a much later estimate. According to it, the

intensiﬁcation process got underway in AD 1640–1729, about

a half century later than the range hypothesized by the origin

narrative. This disparity grows when the pace of intensiﬁcation

is considered. At least three analyses have indicated that most

of the wall construction effort in the leeward Kohala ﬁeld

system was concentrated in the later phases of wall building

(Ladefoged & Graves 2000, 2008; Ladefoged et al. 2003). This

trend can be seen clearly in the detailed study area in the walls

related stratigraphically to trail B. There are 28 of these; four

belong to the early Group 1 walls and the rest belong to Group

3, which dates to AD 1730–1799, and Group 5, which dates

to the early nineteenth century. Thus, the Bayesian calibration

indicates that the main thrust of ﬁeld system intensiﬁcation

can be dated to the eighteenth and early nineteenth centuries.

Much of it seems to be a post-contact phenomenon.

This disparity between the hypothesized rise of ali‘i

authority, as intrepreted from ali‘i traditions, and ﬁeld system

intensiﬁcation is supported by evidence for development of

the spatial structure of the ﬁeld system. Application of graph

theoretic principles to the detailed study area indicates a

structural break between trails C and D within Kahua 1 and not

at the boundary of Pāhinahina and Kahua 1 as implied by an

earlier analysis (Figure 1). This structural break was not closed

until sometime after the curb for trail C was constructed,

which the Bayesian calibration estimates at AD 1770–1829.

Thomas S. Dye

Figure 6. Tempo of change in the leeward Kohala ﬁeld system. The ﬁgure is in row major order with the

oldest period in the upper left. The 67% HPD intervals are: top left, 270–489 years; top right, 260–419

years; bottom left, 100–189 years; bottom right, 100–139 years. Note that the Alienation period is

compressed somewhat by the use of AD 1950 as Present, a convention in

C dating.

Rapa Nui Journal Vol. 25 (2) October 2011

The tempo of change in the leeward Kohala ﬁeld system, Hawai‘i Island

The implication of this ﬁnding is that construction projects

were carried out in sub-regions of the ﬁeld system whose

boundaries were not coincident with ahupua‘a boundaries

until relatively late in traditional Hawaiian times and quite

possibly into the post-contact era. To the extent that ali‘i

authority was projected into the ﬁeld system within ahupua‘a

land units, this result suggests that ali‘i authority played a late,

largely post-contact, role in construction of the ﬁeld system.

A consideration of the tempo of change indicated by

the Bayesian calibration contraindicates the impression of

regularity and inevitability left by the chronology of the

origin narrative. Instead, the expansion of agriculture into the

region made possible by the late introduction of sweet potato

was a fairly long, drawn out affair that is imprecisely dated

with current evidence. This is a period during which expert

agriculturalists experimented with a new crop plant in areas

that had previously seen little, if any, use. Presumably, it was at

this time that the limits of rain-fed cultivation of sweet potato

were discovered—the arid boundary of the lowland ﬁelds and

the nutrient deﬁcient boundary in the wet uplands (Vitousek

et al. 2004). Some experimentation with agricultural walls

in the late seventeenth century indicate efforts, presumably

successful, to control soil moisture against the combined

effects of strong winds and variability in precipitation.

This long period of expansion and initial experimentation

was punctuated, probably early in the historic period, by a

period of intensive wall construction and ﬁeld subdivision

that ended less than a century later when the ﬁeld system

was abandoned. The irregular tempo of change revealed

by the Bayesian calibration, with a late burst of investment

in the ﬁeld system infrastructure followed soon after by its

abandonment, suggests the importance of contingency in the

history of agricultural development and raises the possibility

that the response to contingent events, which disrupted several

hundred years of apparently successful agricultural and social

development, was not in the end sustainable.

Acknowledgements

Thanks to Rob Hommon, Jim Bayman, Eric Komori, Thegn

Ladefoged, and an anonymous reviewer for comments on

drafts of the paper. Peter Langfelder offered help eliminating

transitive relations in the calculations that produced

Figure 3. Eric Shulte designed most of the framework for the

reproducible research version of the paper and kindly tested

and improved the project make ﬁle. Krickette Murabayashi

prepared the digital version of the paper to the journal’s

speciﬁcations. The author is responsible for any errors that

might remain.

References

Banner, S. 2005. Preparing to be colonized: land tenure and legal

strategy in nineteenth-century Hawaii. Law & Society Review

39(2): 273-314.

Buck, C.E., W.G. Cavanagh & C.D. Litton. 1996. Bayesian

approach to interpreting archaeological data. Statistics in

practice. Chichester: John Wiley & Sons.

Buck, C.E., J.A. Christen & G. James. 1999. BCal: an on-line

Bayesian radiocarbon calibration tool. http://bcal.shefﬁeld.ac.uk

Chinen, J.J. 1958. The Great Mahele: Hawaii’s land division of

1848. Honolulu: University Press of Hawai‘i.

——2002. They cried for help: the Hawaiian land revolution of the

1840s and 1850s. Honolulu: Jon J. Chinen/Xlibris.

Christen, J.A. 1994. Summarizing a set of radiocarbon determinations:

a robust approach. Applied Statistics 43(3): 489-503.

Dye, T.S. 2010. Traditional Hawaiian surface architecture: absolute

and relative dating. In Research designs for Hawaiian

archaeology: agriculture, architecture, methodology. Special

Publication 3. T.S. Dye (ed.): 93-155. Honolulu: Society for

Hawaiian Archaeology.

——2011. A model-based age estimate for Polynesian colonization

of Hawai‘i. Archaeology in Oceania 46(3): 130-138.

Hage, P. & F. Harary. 1983. Structural models in anthropology.

Cambridge: Cambridge University Press.

Harris, E.C. 1989. Principles of archaeological stratigraphy.

Second Edition. London: Academic Press.

Kirch, P.V. 2010. How chiefs became kings: divine kingship and the

rise of archaic states in ancient Hawai‘i. Berkeley: University

of California Press.

Ladefoged, T.N. & M.W. Graves. 2000. Evolutionary theory and

the historical development of dry-land agriculture in North

Kohala, Hawai‘i. American Antiquity 65(3): 423-448.

——2008. Variable development of dryland agriculture in Hawai‘i:

A ﬁne-grained chronology from the Kohala ﬁeld system,

Hawai‘i Island. Current Anthropology 49(5): 771-802.

——2010. The leeward Kohala ﬁeld system. In Roots of conﬂict:

soils, agriculture, and sociopolitical complexity in ancient

Hawai‘i. School for Advanced Research Advanced Seminar

Series. P.V. Kirch (ed.): 89-110. Santa Fe: SAR Press.

Ladefoged, T.N., M.W. Graves & M. McCoy. 2003. Archaeological

evidence for agricultural development in Kohala, island of

Hawaii. Journal of Archaeological Science 30: 923-940.

Ladefoged, T.N., M.W. Graves & J.H. Coil. 2005. The introduction

of sweet potato in Polynesia: early remains in Hawai‘i. Journal

of the Polynesian Society 114: 359-373.

Moffat, R.M. & G.L. Fitzpatrick. 1995. Surveying the Mahele:

mapping the Hawaiian land revolution. Volume 2.

Palapala‘āina. Honolulu: Editions Limited.

Moore, H. 1995. The problems of origins: poststructuralism and

beyond. In Interpreting archaeology: ﬁnding meaning in the

past. I. Hodder, M. Shanks, A. Alexandri, V. Buchli, J Carman,

J. Last and G. Lucas (eds.): 51-53. London: Routledge.

Rosendahl, P.H. 1972. Aboriginal agriculture and residence patterns

in upland Lapakahi, island of Hawaii. Unpublished PhD

Dissertation. University of Hawai‘i.

Storey, A.A., J.M. Ramirez, D. Quiroz, D.V. Burley, D.J. Addison,

R. Walter, A.J. Anderson, T.L. Hunt, J.S. Athens, L. Huynen &

E.A. Matisoo-Smith. 2007. Radiocarbon and DNA evidence

for a pre-Columbian introduction of Polynesian chickens

to Chile. Proceedings of the National Academy of Sciences

104(25): 10335-10339.

Stuiver, M. & H.A. Polach. 1977. Discussion: reporting of

C data.

Radiocarbon 19: 355-363.

Vitousek, P.M., T.N. Ladefoged, P.V. Kirch, A.S. Hartshorn, M.W.

Graves, S.C. Hotchkiss, S. Taljapurkar & O.A. Chadwick.

2004. Soils, agriculture, and society in precontact Hawai‘i.

Science 304: 1665-1669.

This article has been peer-reviewed. Received 27 July 2011,

accepted 1 September 2011.