Fortified castles on Okinawa Island during the Gusuku Period, AD 1200-1600.

LADEFOGED, THEGN N. ; PEARSON, RICHARD

Introduction

Fortified castles on the island of Okinawa, southwestern Japan, date from approximately AD 1200 to 1600. These castles are termed gusuku in Okinawan dialect. The larger of these multifunctional sites usually contained a number of different activity areas (Pearson 1992; 1997; 1999). Katsuren Castle, located on the east coast of Okinawa, is one example (FIGURE 1). Extensive excavation and restoration of the castle (Asato 1990) suggest that it was approximately 12,000 sq. m in size, and included four enclosures. The smallest, highest and most protected enclosure was the sacred precinct. The second enclosure contained the residential palace, and the third was an area in front of the residential palace used for assemblies. The fourth, lowest enclosure included storage, a work place for artisans and a well. The largest castles ranged from 38,000 to 43,000 sq. m (Pearson 1962: 268).

[Figure 1 ILLUSTRATION OMITTED]

Okinawan histories, such as the Chuzan Seikan (Haneiji 1983) written in the 17th century, describe competition for power throughout the island in the form of various confederacies until the emergence of three polities in the 14th century. These polities were located in the north, central region and south of the island, and they were unified by the central polity, the Chuzan Kingdom, in AD 1429. Okinawan prefectural archaeologists have located 192 gusuku sites (Okinawa Ken 1983), shown as circles, squares and triangles in FIGURE 2. All of these castles were built within the period c. AD 1200-1554. By 1429 Shuri Castle was the centre of the Chuzan Kingdom, and by 1554 only four specialized castles throughout the island were maintained (Pearson 1997: 126). In 1609 the Chuzan Kingdom was invaded by the Satsuma fiefdom of southern Japan, which controlled it until 1879. It became known as the Ryukyu Kingdom.

[Figure 2 ILLUSTRATION OMITTED]

Asato (1990: 128-37; 1998) published a discussion of site groupings and their distribution, and their development into 17th- and 18th-century political divisions (magiri) and cultural subgroups. Based on a non-statistical analysis, he traced the development of prehistoric shell-mound site groups, to subgroups sharing certain kinds of pottery temper, to castle groups and finally to administrative sub units at the time of the Ryukyu Kingdom (AD 1609-1879). Our study parallels and extends his historical interpretation by using a geographical information system to analyse the distribution of fortified castles in relation to each other and the soils of the island.

Soils and subsistence activities

During the first half of the 2nd millennium, subsistence cultivation on Okinawa was based on a combination of dry-field wheat (Triticum aestivum), barley (Hordeum vulgare), foxtail millet (Setaria italica), broomcorn millet (Panicum miliaceum) and irrigated rice (Oryza sativa) (Sasaki 1973; Asato 1990). A preponderance of wheat, barley, foxtail millet and broomcorn millet in flotation samples analysed by Takamiya (1997) may indicate that dry-field farming was predominant; however, only two sites have been analysed and the samples are relatively small. In addition, taro was grown extensively in areas with adequate moisture (Sasaki 1973). Takamiya (1997) suggests that rice was not a primary crop on the island because it matures during the typhoon season and could therefore be easily ruined.

Chinzei et al. (1967) grouped the soils of Okinawa into five basic types (podzolics, forest, alluvium, regosols and lithosols). We have made a further subdivision within lithosols, and group the Okinawan soils into six types (see FIGURE 2). The reddish podzolics of central and northern Okinawa are derived from metamorphosed sediments of Palaeozoic Age, principally phyllites and feldspathic sandstone. The brown forest soils are found primarily in the central region of the island, and are derived from raised limestone coral. The alluvial soils, including gley and immature gravelly and sandy soils, are found in narrow inland valleys and coastal flats throughout the island. The regosols of southern Okinawa are azonal soils (without A and B horizons) that developed from deep unconsolidated deposits of marl-like rocks or soft minerals with no genetic horizons. The lithosols consist of recently weathered rock fragments on steep slopes and are found in both the north and south of the island. We distinguish the northern lithosols which are acidic and are derived from metamorphosed Palaeozoic sediments, and the southern lithosols which are neutral or alkaline soils derived from raised limestone.

In general the brown forest soils and deeper regosols would have been suitable for growing wheat and barley. Millet would have been grown in these deeper soils but also in the shallower southern lithosols. The moisture-retentive alluvium would have been suitable for growing rice or dryland taro. Traditionally, slash-and-burn cultivation would have been practised on the relatively poor podzolics and northern lithosols, but production would have been quite low (Sasaki 1973).

Analysis

Our analysis suggests that the 192 castles are not distributed randomly across the island. A chi-square statistic indicates that they are differentially distributed with respect to the six different soil types (FIGURE 3). Given the relative proportions of the soils, there are significantly fewer castles then expected on the podzolics and northern lithosols, and far more than expected on the forest, regosols, alluvium and southern lithosols. Sites are comparatively fewer in the central and northern regions of Okinawa and their distribution is distinct from the rest of the island. They are usually found near pockets of alluvium, with low densities of sites found in the large areas of northern lithosols and podzolics.

[Figure 3 ILLUSTRATION OMITTED]

Spatial clustering

The pattern concerning the distribution of castles is clarified by considering the distances between the sites. The majority of the castles are spaced at a relatively short distance from each other, at approximately the same distance from their nearest neighbours. The average distance between castles is 1.282 km with a standard deviation of 1.692. Despite this, there are a number of castles that are relatively far from some of their nearest neighbours, and thus a clustering effect occurs. This is confirmed by a Clark-Evans nearest neighbour coefficient of -7.389, suggesting that there is a significant level of clustering in the distribution of the castles. A K-means statistic can be used to determine the number of clusters formed by the Okinawan castles. Kintigh & Ammerman (1982) and Blankholm (1990: 65) note that the optimum number of clusters in a geographic distribution is indicated by the inflection point or points in a graph of the log percent of the sum of the squared distances from each point within a cluster to the centroid of the cluster. This measure is referred to as the log% SSE value. When the rate of reduction of log%SSE values begins to decline, thereby creating an inflection point, it is at this point that the optimum number of clusters occurs. FIGURE 4 includes a graph of the log%SSE values for the cluster configurations of all the castles on Okinawa. There is an inflection point on the graph at three clusters. The distribution of the castles assigned to each of the three clusters is shown in FIGURE 2. The castles form a northern group (n=34), a central group (n=55), and a southern group (n-103). This clustering is to some extent the result of the linear shape of the island and the presence of peninsulas in the northern, central, and southern areas. In addition, the avoidance of lithosols in the north and podzolics in the middle of the island, creates natural buffers between the three clusters and a relatively barren region for agriculture in the far north of the island.

[Figure 4 ILLUSTRATION OMITTED]

The three clusters generally correspond to the three competing historically known 14th-century polities of Sanhoku in the north, Chuzan in the central region and Sannan in the south. There is, however, an interesting discrepancy between the clustering and the historically known distribution of the three polities. The statistical boundary between the southern and central cluster is located too far south. The castles of Urasoe and Shuri are included in the southern cluster, but historically these two sites were part of the central Chuzan polity. Historically, the Chuzan polity had strong links to the southeastern part of the island.

Soil catchments

Each of the three clusters displays differences in the composition of soils surrounding the castles. The catchments or areas immediately surrounding the castles in each of the three polities are diverse. This diversity probably represents a degree of variation in the subsistence activities of the people in each area. In the north, castles are disproportionately located on or adjacent to tracts of alluvium. These alluvial pockets occur near the coast and therefore all of the northern sites have coastal locations. In the central and southern regions inland sites are common. Significantly, the castles in the north are often not located in the middle of alluvial tracts, rather they are located close to the boundary of alluvial tracts and other types of soils. Despite the disproportionately low number of northern castles on northern lithosols, the catchments of the castles often include considerable areas of northern lithosols. This pattern is reflected in the average percentages of the soils contained within a 500-m radius catchment surrounding the 34 northern castles (FIGURE 5). A distance of 500 m was selected for the radius of the catchments as this is just under half the average distance (1.282 km) between castles. Despite there being a disproportionate number of castles located on alluvium with an avoidance of castles being actually situated on northern lithosols, access to a number of different soils was apparently important. The mean number of different types of soil contained within the 500-m radius catchments of the northern castles is 1.76, with a standard deviation of 0.61. It would appear that it was important for the northern castles to be situated so that their inhabitants would have had access to alluvium for taro or rice, and either northern lithosols, podzolics, and/or forest soils for dryland agriculture.

[Figure 5 ILLUSTRATION OMITTED]

In the central region there is a disproportionate avoidance of locating castles on podzolics, with an over representation of castles on forest soils. In a manner similar to the northern castles, however, the central castles are also situated close to the boundary of different types of soil. The mean number of soils within the catchments of the central castles is 1.95, with a standard deviation of 0.76. The average percentage of the different types of soils found in the central catchments is shown in FIGURE 5. The catchments of the central castles focused on soils suitable for dryland agriculture and de-emphasized alluvium to a greater extent than the catchments of the northern castles.

The southern castles contain the highest number of different types of soils in their catchments (mean-2.16, s.d.=0.70). Although the southern castles are generally not located on the regosols of the interior of the southern peninsula, many of the castles are located close to this zone and include a large proportion of regosols in their 500-m radius catchment (see FIGURE 5).

There is a tendency for all of the castles, regardless of their geographic location, to be situated in locations that had access to more than one soil type. Apparently situating castles in ecotones between soils was preferred to locating castles in the middle of one particular type of soil. This preference, however, appears to be strongest in the south. Although the zones of regosols and southern lithosols are quite large, sites occur on or near their edges. An analysis of variance (ANOVA) of the mean number of soil types in the catchments of the three clusters indicates that the southern castles had a significantly larger number of different soils within their catchments than the other two regions (F=4.498, sig.=0.012).

Sub-clusters

There is further patterning in the distribution of castles within each of the three clusters. A Clark-Evans nearest neighbour coefficient of -0.204 for the 103 castles in the southern cluster suggests that the castles are close to being distributed in a clustered manner but are, in statistical terms, randomly distributed. A plot of the log%SSE values from a K-means statistic, however, indicates that the castles in the southern cluster can themselves be grouped into a number of different sub-cluster configurations. FIGURE 4 includes a graph of the log%SSE values for the castles in the southern cluster, inflection points on the graph suggest that the 103 castles in the southern cluster can themselves be grouped into 4 sub-clusters and 14 sub-clusters.

The four sub-cluster configuration of the southern castles is shown in FIGURE 6. The castles in the south of the island form sub-clusters on the small western peninsula (n=21), the southern peninsula (n=40), the eastern peninsula (n-23), and the northwest portion (n=19). The four sub-cluster solution reflects sub-groupings in the Sannan polity. The northern sub-cluster includes Urasoe and Shuri castles and is the area where the political centre of the Chuzan Kingdom formed from the 13th century onwards (see FIGURE 2 for locations). The other three sub-clusters of the southern cluster correspond to powerful groups within the Sannan polity. The location of the capital of Sannan is debated, with proponents divided between the sites of Shimazoe Ozato Gusuku in Ozato Village, (southeast), and Shimajiri Ozato in Itoman City (southwest) (see FIGURE 2 for locations). In the Chuzan Seikan, Ozato Gusuku in Ozato Village is the residence of the King of Sannan. However in Ming Dynasty historical documents, Ozato Gusuku appears to refer to the Itoman location. Takemoto & Asato (1993: 72) and Kin et al. (1988: 85) state that the Sannan capital is Shimajiri Ozato in Itoman City. Regardless of whether the capital was in the southeast or southwest, it is clear that both were important centres of power forming the nucleus of each of two of the sub-clusters. A fourth subcluster of sites is centred around Tomigusuku, one of the larger castles in the region.

[Figure 6 ILLUSTRATION OMITTED]

The 14 sub-cluster configuration of the southern cluster includes a degree of variation in the number of castles assigned to each cluster, the maximum number being 18, the minimum being 2, with a mean of 7.3 and a standard deviation of 4-6 (FIGURE 7). Five of the sub-clusters parallel groupings identified by Asato (1990: 132): sub-clusters 3 (Chinen), 4, (Haebaru), 7 (Ozato), 13 (Gushichan) and 14 (Makishi). These groups are historically identified as 17th-century administrative divisions (magiri). Four other sub-clusters show partial overlap with the historical groups identified by Asato (1990) and five sub-clusters show little or no isometry.

[Figure 7 ILLUSTRATION OMITTED]

The 55 castles assigned to the central cluster have a Clark-Evans nearest neighbour coefficient of-2.869, indicating that they themselves are clustered. A plot of the log%SSE values from a K-means statistic shows a slight inflection point at 5 clusters (see FIGURE 4). The castles assigned to the 5 sub-clusters are shown in FIGURE 6. The clustering is formed to some extent by the avoidance of podzolics and a preference for coastal locations. Four of the five sub-clusters coincide with historically known power centers. Sub-cluster 1 is dispersed, centering on Iha. However, sub-cluster 2 is associated with a group of castles on the islands of Henza-Miyagi, Hamahiga, and Ikei; sub-cluster 3 is associated with the Katsuren Peninsula and Tsuken Island; sub-cluster 4 includes Chatan and Nakagusuku; and sub-cluster 5 is associated with the 15th-century Zakimi and its antecedents in the Yomitan area.

A Clark-Evans nearest neighbour coefficient of 0.464 for the 34 castles assigned to the northern cluster indicates that they are distributed randomly. However, the plot of the log%SSE values for the northern castles shows a marked inflection point at three clusters (see FIGURE 4), and the northern castles assigned to these three clusters are shown in FIGURE 6. Again, the avoidance of lithosols and the preference for coastal locations are apparent. The three clusters fall generally into a Motobu group, dominated by Nakijin (sub-cluster 1); a group in the Haneiji lowland area (sub-cluster 2); and a small cluster in the northern region around Nejana castle site (sub-cluster 3). Sites in the northern region of the island appear to be situated on lithosols, or on or very near alluvial pockets, which served as their agricultural base. Haneiji, for instance, is a well known center for irrigated rice cultivation, from the Gusuku Period with a dominant castle and satellite settlements (Asato 1998: 181).

Conclusion

Historical evidence suggests that three polities had formed on Okinawa by the 14th century. The distribution of the fortified castles provides clear archaeological evidence of the three polities. Statistically the castles cluster into three groups that coincide with the historically known political units. There is one area where the statistical clusters and the political units do not coincide. The inclusion of the central castles of Shuri and Urasoe in the southern group shows old historical links between the southern and central clusters. In the early 14th century, Sho Shisho, founder of the First Sho Dynasty of the Chuzan Kingdom, with the support of his sons and the lords of Sashiki in the southeast, defeated the lord of Urasoe, Bunei, and became king of Chuzan. The catchments surrounding the castles in each cluster differ. The northern castles appear to stress access to alluvium to a greater extent than the central and southern castles. In general, however, most of the castles are located close to ecotones between different types of soil. The southern castles in particular seem to be situated to take advantage of access to a number of different types of soils. The castles within each of the three political units form sub-groupings that again often coincide with historically known political factions. This would suggest that at one time the three polities each consisted of several smaller individual political factions, which go back into the Shellmound Period (100-1100 AD) as proposed by Asato (1990: 192-3). Through time, these smaller units were integrated into one of the three polities, with the final integration of the island into a single island wide polity in the 15th century.

Acknowledgements. We thank Sarina Pearson and Peter Sheppard for their comments on the manuscript, and Martin Bale, Warren Hill and Susan Matson for the preparation of base maps of site locations and soil zones. Research on Okinawan castle sites by Richard Pearson was supported by grants from the Humanities & Social Sciences Research Council of Canada and the Humanities & Social Science Research Grants Committee of the University of British Columbia.

References

ASATO, S. 1990. Kokogaku kara Mita Ryukyu Shi, Jo (Ryukyu history as seen from archaeology 1). Naha: Okinawa Bunko.

1998. Gusuku; Kyodotai; Mura: Okinawa Rekishi Kokogaku Josetsu (Castle, co-operative group, village: an introduction to Okinawan historical archaeology). Ginowan: Yoju Shorin.

BLANKHOLM, H.P. 1990. Intrasite spatial analysis in theory and practice. Aarhus: Aarhus University Press.

CHINZEI, T., K. OYA & Z. KOJA. 1967. Soils and land use in the Ryukyu Islands. Shuri: University of the Ryukyus, College of Agriculture.

FUJII, H. 1995. Chusei no Shiro to Kassen (Castles and battlefields of the Middle Ages). Tokyo: Asahi Shimbunsha.

HANEIJI, C. 1983. Chuzan Seikan (Mirror of the ages of Chuzan Naha: Okinawa Ken Kyoiku Iinkai. (Newly edited version by Okinawa Ken Kyoiku Cho Bunka Ka. Original published in 1650.)

KIN, S., M. TANA, I. CHINEN & S. TOMA. 1988. Gusuku Rodo Okinawa no Gusuku Monogatari. Naha: Mugisha.

KINTIGH, K.W. & A.J. AMMERMAN. 1982. Heuristic approaches to spatial analysis in archaeology, American Antiquity 47: 31-63.

OKINAWA KEN, KYOIKU IINKAI, 1983 Gusuku: Gusuku Bunpu Chosa Hokoku (1): Okinawa Honto Oyobi Shuhen Rito (Castle Sites: Report on Distribution (1): Okinawa Main Island and Surrounding Offshore Islands). Naha: Okinawa Ken Kyoiku linkai.

PEARSON, R. 1992. Trade and the rise of the Okinawan State, Bulletin of the Indo Pacific Prehistory Association 10: 263-81.

1997. The Chuzan kingdom, Okinawa, as a city state, in T. Charlton & D. Nichols (ed.), The archaeology of city states: cross cultural approaches: 119-34. Washington (DC): Smithsonian Institution Press.

1999. Ancient Okinawa: the rise of the state. Manuscript on file, University of British Colombia.

SASAKI, K. 1973. Okinawa honto ni okeru dentoteki hatasaku noko gijutsu (Traditional upland farming techniques on the Main Island of Okinawa), Jinrui Kagaku 25: 79-107.

TAKAMIYA, H. 1997, Subsistence adaptation processes in the prehistory of Okinawa. Ph.D dissertation in Anthropology, University of California, Los Angeles.

TAKEMOTO, S. & S. ASATO. 1993. Okinawa: Nihon Kodai Iseki 47 (Okinawa: Ancient Sites of Japan Series 47). Osaka: Hoikusha.

THEGN N. LADEFOGED & RICHARD PEARSON(*)

(*) Ladefoged, Department of Anthropology, University of Auckland, Auckland, New Zealand. t.ladefoged@auckland.ac.nz Pearson, Department of Anthropology/Sociology, University of British Columbia, 6303 Northwest Marine Drive, Vancouver BC, Canada V6T 1Z1. rpearson@unixg.ubc.ca

Received 21 April 1999, accepted 14 July 1999, revised 6 January 2000.