Examining Prehistoric Migration Patterns in the Palauan Archipelago: A Computer Simulated Analysis ofDrift Voyaging RICHARD CALLAGHAN AND SCOTT M. FITZPATRICK INTRODUCTION ARCHAEOLOGISTS WORKING IN THE PACIFIC have long been concerned with how islands in Remote Oceania were colonized prehistorically. To determine patterns of oceanic human dispersal to these islands, researchers have typically relied upon the distribution ofstylistically or geochemically unique artifacts (e.g., Descantes et al. 2001; Dickinson and Shutler 2000; Irwin 1978; Pavlish et al. 1986; Weisler 1990, 1998; White 1996; see also Rolett 2002; Rolett et al. 2002), while linguistics and genetic studies have also played an important role in estab lishing where settlers may have originated (e.g., Bellwood 1997; Blust 2000; Lum 1998; Lum and Cann 2000; Reid 1998; Starosta 1995). One useful means for investigating voyaging and prehistoric colonization pat terns is the recording of traditional seafaring techniques and the experimental construction and use of watercraft based on ancient technologies (see Bechol 1972; Doran 1978; Finney 1988; Lewis 1978; Ling 1970; also Bednarik 1998; Irwin 1998). To complement these studies, computer simulations of seafaring have provided additional data about how peoples may have traveled over time through the Pacific (Avis et al. 2007; Irwin 1992; Levison et al. 1973). However, these and other simulations are not explicitly interested in, nor can they hope to provide great insight, into the timing ofhuman arrival, which must be ascertained instead through careful consideration ofradiocarbon chronologies from stratified archaeological and paleoenvironmental deposits. Nonetheless, these experiments can help to refine theories on prehistoric migration routes and direct future re search objectives to areas that may be understudied. Archaeologists have postulated that human populations in the Palauan archi pelago (see Fig. 1), may have migrated from Taiwan, Indonesia, New Guinea, or the Philippines based on archaeological, historic, linguistic, and genetic studies (see Bellwood 1997; Irwin 1992; Reid 1998; Semper 1982 [1873]:17-18). To Richard Callaghan is associate Professor, Department ofArchaeology, University ofCalgary, Al berta, Canada; Scott M. Fitzpatrick is assistant Professor, Department ofSociology and Anthropol ogy, North Carolina State University, Raleigh. AsiatlPcrspt'Ctil'cs,Vol.47,No.1 ©2008bytheUniversityofHawai'jPress. CALLAGHAN AND FITZPATRICK . PREHISTORIC MIGRATION PATTERNS 29 8'15' CHINA ~ (Jraiwan Kayangel ':'.NorthemMarianas :'Saipan oOGuam .... Kosnie Pohnpei 7'45' PhilippineSea ...-.... CoralSea "·000 ..... '. 7'30' AUSTRALIA ~ .. ~ ..•. , , ..",.:/:R.:, "(.' 7'15' Ulon,g,- .. ...../ ~geruktabel Jfr~eCherChar "';$,' ),,~'" ~~ereChOng , Ngemelis ':-' V~ .'" Tasmania ~~~l.eliU 7' A Oo-C-E=A5N0I0Akm 134'pAngaur B PoALAU10km Fig. 1. Maps (leftto right) ofthe western Pacific (A) and the main Palauan archipelago (B). further examine possible routes ofmigration to western Micronesian islands, we investigated computer simulated drift voyages to the Palauan archipelago. These simulations take advantage ofdetailed oceanographic, anemological, and climato logical data to determine how watercraft will move or react to ocean conditions during a given time ofthe year. This type ofapproach to studying ancient seafar ing has been increasingly used worldwide to investigate colonization, migration, and culture contact in the Pacific Islands (Avis et al. 2007; Irwin 1992; Levison et al. 1973), the Caribbean (Callaghan 2001, 2003a), between Ecuador and Mexico (Callaghan 2003b), Costa Rica and Colombia (Callaghan and Bray 2007), from Japan to North America (Callaghan 2003c), and other regions (Montenegro et al. 2006). The results ofthese simulations can then be coupled with other data to de velop more robust hypotheses regarding the human colonization ofislands world wide. There are a number ofvoyaging strategies that can be investigated using com puter simulations. The two basic divisions are between purposeful or intentional voyages and drift voyages. In this analysis we have chosen to investigate drift voyages. For questions ofinitial discovery, using a strategy ofdrift voyages rather 3° ASIAN PERSPECTIVES . 47(1) . SPRING 2008 than purposeful voyages is preferable because it is best to make the fewest assumptions possible in the analysis. In all ofthe simulations it was assumed that the crew was lost at sea. Dening (1963:138-153) notes that the limited empirical evidence ofknown drift voyages in Polynesia suggests a common pattern of behavior in which sailors conclude they are lost early in the voyage and respond by allowing the vessel to drift before the wind with no further attempt to navigate in a particular direction. This strat egy allows close to the maximum distance to be covered in a given time when there is no clear indication ofrelative location. The Palauan archipelago serves as an interesting case study for conducting computer simulated drift voyages because regional currents and winds are known to have high velocity and volatility which also allowed it to remain virtually iso lated from direct European contact until 1783 (Callaghan and Fitzpatrick 2007; Hezel 1972, 1983). For the simulations we focused on testing hypothetical routes from Taiwan, the Philippines (Samar and Mindanao), New Guinea, the Bis marcks, Halmahera, and Guam, all ofwhich have been mentioned in the litera ture as possible origin points for early Palauans, or in the case ofthe latter two, had some historical connections with Palau. Historical accounts also testify to the frequency of drift voyages from the Carolines to the Philippines (Hezel 1972, 1983). In 1664, Jesuit missionaries recorded some 30 Carolinian canoes that had accidentally drifted to the Philippines with no less than nine different landings be tween 1664 and 1669 (Hezel 1972:28). In December 1696, Father Paul Klein, after meeting 30 Carolinians on Samar who had blown offcourse while sailing from Lamotrek to Fais (Hezel 1972:27), described the events in a letter to the Jesuit General in Rome that spurred a keen interest in conducting exploratory voyages into the Western Carolines. We first discuss briefly what is presently known about the prehistoric coloniza tion ofPalau. Next we describe the local geographic, oceanographic, and anemo logical conditions which serve to contextualize how these factors may have af fected the passage ofwatercraft. Based on these and other historic data, we then report on the simulations we conducted ofdrift voyaging to Palau. Results sug gest that the southern Philippines would have been the most likely origin for early Palauan settlers, supporting the most widely accepted hypothesis ofcoloni zation. PREHISTORIC COLONIZATION OF WESTERN MICRONESIA Using linguistical evidence, Tsang (1992) and Starosta (1995) proposed that the Marianas (e.g., Guam, Saipan, Rota, Tinian) were colonized by peoples from Taiwan. Others such as Reid (1998) have argued that Chamaru (Chamorro) speakers have more immediate links to the Philippines. Recent genetic evidence indicates that the Marianas and Yap were probably settled independently from Island Southeast Asia and that gene flow may have also occurred from central eastern Micronesia (Lum and Cann 2000:165). Palau appears to have genetic contributions from Southeast Asia, Central-Eastern Micronesia, and New Guinea (Lum and Cann 2000:166). Based on archaeological investigations, the Marianas appear to have been settled by as early as 3500 B.P. (Butler 1994; Kurashina and Clayshulte 1983; Kur- CALLAGHAN AND FITZPATRICK . PREHISTORIC MIGRATION PATTERNS 3I ashina et al. 1984), Yap by at least 2000 B.P. (Gifford and Gifford 1959:159; Intoh 1997; Takayama 1982:91), and Palau sometime in the period 3000-3300 B.P. (Clark 2004, 2005; Fitzpatrick 2002, 2003; Liston 2005). Douglas Osborne (1979) first proposed that the islands were settled in a "stepping-stone" fashion northward beginning around 4500 B.P. based on C14 ages from pottery sherds, an assertion later disputed by Masse (1990) who compiled an additional suite ofac ceptable radiocarbon dates, none of which was earlier than 2000 B.P. Recent paleoenvironmental studies on all ofthese island groups (Athens and Ward 2002, 2004; Dodson and Intoh 1999), as well as paleoshoreline evidence in Palau (Dickinson and Athens 2007), however, suggests that colonization may have occurred even earlier around 4000-4500 B.P. Some researchers dispute this proxy evidence without confirmation of associated cultural remains which, although 1000-1500 years earlier than the generally accepted dates from stratified archaeo logical deposits, seems to be a widespread phenomenon in the region (for discus sion ofthese issues see Athens and Ward 2002,2004; Welch 1998; Wickler 2001, 2004). Despite the complex settlement pattern that has emerged, involving genetically distinct populations from various regions over time, it now appears likely that prehistoric settlement of the region was relatively contemporaneous, whether one accepts the short (archaeological) or long (paleoenvironmental) sequence of occupation. Unfortunately, a general paucity ofarchaeological research has pre vented the establishment ofclearer linkages between possible staging areas in Is land Southeast Asia or Melanesia and western Micronesian islands. ENVIRONMENTAL BACKGROUND Palau consists ofseveral hundred islands-the largest ofthese is Babeldaob which is 330 km2 and the second largest island in Micronesia next to Guam. The main archipelago is roughly 800 km equidistant from the Philippines to the west and Irian Jaya to the south. Palau, situated at 7°30' north of the equator and sur rounded by a complex barrier reefsystem, is over 160 km long, 25 km across at its widest point, and oriented in a northeast-southwest direction. The main climatic feature in western Micronesia is the equatorial trough of low pressure that induces an airflow known as the Northeast Trades (see Siedler and Church 1951; see also Hazell and Fitzpatrick 2006). These occur throughout the winter and spring from October to May (Morris 1988) and are fairly consis tent winds with a mean speed of around 10 knots (18 kmh). With decreasing northeast trade winds in the month of April, the east winds are more frequent and become dominant in the region. On average, the mean wind velocity de creases from 10.5 km per hour inJanuary to 5.8 km per hour inJune, primarily due to the development of the Australian anticyclone that shifts the prevailing winds (Corwin et al. 1956). FromJuly through October, winds are comparatively light with the mean monthly velocity ranging between 0.5 and 1.5 km per hour. The Northeast Trades offer a comparatively consistent velocity and direction. This has two advantages for sailing-a constant wind direction that aids naviga tion on the open sea and a constant southeast or northeast ocean swell (depend ing on the month). Outside ofthe Northeast Trades, the highest number ofdays for winds originating from any given direction occurs in September. These 4-7 ASIAN PERSPECTIVES . 47(1) . SPRING 2008 knot (7.2-12.6 kmh) winds come from the southwest for about seven days (Morris 1988; United States Navy 1995). Calms (or low wind velocity) are reported for a similar number ofdays. Palau is located at the junction of three major ocean currents-the North Equatorial Current (NEC), Equatorial Counter Current (ECC), and the South Equatorial Current (SEC). The formation ofthese can produce considerable sea sonal current pattern changes in the region (Gatty 1944; Irwin 1992; Jenkins 1973; Morris 1988; Rapaport 1999; United States Navy 1995). Between Yap and Palau, for example, ocean currents are fairly consistent for most ofthe year, with only minor local variations in and around the various islands and reef systems (Heyen 1972; see also Hazell and Fitzpatrick 2006). The currents flow east to west at around 0.75 to 1.25 knots (2.5 km/h), except from August to October when there is a variable, weaker northeast current (although a rate of3 knots has been reported) (Morris 1988). During September, surface currents can have a northwest to southwest set. The variable winds and surface currents also make this and othermonths with similar conditions difficult periods for sailing. Changes in the basic features ofthe waves such as shape, frequency, and size are known to alert navigators to the proximity ofland (Gatty 1944; Golson 1972; Heyen 1972). Based on these data collected in greater detail as part ofthe U.S. Navy's (1995) Marine ClimaticAtlas ofthe World, we can simulate voyages under a host ofvarying conditions to hypothesize from where Palauan colonists may have originated. COMPUTER SIMULATIONS OF VOYAGING The problem of predicting from where prehistoric peoples were most likely to have discovered Palau is addressed here using computer simulations based on the known patterns ofwinds and currents throughout the year. When running simu lated drift voyages, four main variables are considered: (1) current patterns; (2) wind patterns; (3) vessel type; and (4) propulsion. The structure ofthe simulation is the actual mechanics of the program; for example, how data are selected and success as a percentage ofall voyages calculated. Parameters include such factors as the starting position of the vessel, drift or directed voyage, duration at sea, and what constitutes a successful voyage. Currents will affect any objects caught in then'! in a 1:1 ratio. That is, the ob ject will have the same speed and set as the current unless other forces are operat ing. Any object floating with an appreciable part above the water will be more af fected by wind than by currents unless the current is an exceptionally strong one. The effect ofwind on objects floating high in the water was used by traditional navigators in Kiribati to determine the direction ofland when recent winds dif fered from the flow of the current (Lewis 1972:212). The source ofwind and current data is the CD-ROM version ofthe U.S. Navy's Marine Climatic Atlas of the World (Version 1.1, 1995). The type ofvessel and how it is propelled make up the final variables ofthe problem, as the shape ofan object both above and below the waterline will dic tate wind effects. However, in this problem, we do not know what types ofves sels were used prehistorically, but some form ofsailing canoe is likely (see Doran 1981). Speeds at which the vessels used in this simulation can travel under various wind conditions are taken from figures provided by Levison et al. (1973). The program itselfis based on the U.S. Navy's Marine Climatic Atlas (1995) and CALLAGHAN AND FITZPATRICK . PREHISTORIC MIGRATION PATTERNS 33 it has been expanded to include all ofthe world's seas and oceans with the excep tion ofArctic waters. The data are organized in a resolution ofone degree Mars den squares (one degree oflongitude by one degree oflatitude). In particular, this resolution allows the effects ofsmaller and more variable currents to be accurately reflected in the outcomes. The program randomly selects wind and current data that are frequency-weighted according to the compiled observations ofthe Marine Climatic Atlas. These forces are then allowed to operate on vessels for a twenty four-hour period before a new selection is made (see Levison et al. 1973 for ajus tification ofthe period length). The actual distance and direction traveled are based on the wind and current data combined with the speeds from Levison et al. (1973:19) and parameters selected by the program operator. Examples ofparameters here include the use of sails, sea anchors to keep avessel oriented into the wind during storms, or drogues to slow the vessel and prevent following seas from swamping the vessel. It is also possible to include here changes ofbearing when under sail. This last feature is important when assessing the level of navigational skill required to reach a selected target during directed voyages. The program automatically shifts to the database for the following month after the month originally selected has expired. This feature better reflects the reality of changing wind and current conditions over long voyages. The result ofthe simulations is expressed as the percentage of successes for a particular vessel type from selected points. Parameters of the simulation are choices made by the program operator in order to set up the simulation to answer a particular question. This includes the following information: (1) point of origin and destination; (2) crew strategy; (3) performance characteristics; (4) duration of voyages; (5) time of year; and (6) number ofsimulations. Seven points oforigin were used with Palau as the desti nation. These were southern Taiwan; east ofSamar (central Philippines); southeast of Mindanao (southern Philippines); north of Teba on New Guinea; north of Halmahera; north ofthe Bismarck Archipelago. The second parameter is the strategy used by the crew. In all ofthe simulations it was assumed that the crew was lost at sea. As noted above, Dening (1963:138 153) states that the limited empirical evidence of known drift voyages in Poly nesia suggests a common pattern ofbehavior in which sailors conclude they are lost early in the voyage and respond by allowing the vessel to drift before the wind with no further attempt to navigate in a particular direction. This strategy allows close to the maximum distance to be covered in a given time when there is no clear indication ofrelative location. In this case, the performance character istics are simply the speed for small vessels sailing before the wind (provided by Levison et al. 1973:19). For this type ofquestion, using a strategy ofdrift voyages rather than directed voyages is preferable. Looking at initial discovery, it is best to make the fewest assumptions as possible in the analysis. A drift strategy limits decisions on the part ofthe operator. In virtually all cases, ifit is easy to discover point B from point A in a drift voyage, it will easier to navigate to point B once its location is known. Directed voyages are better used when evaluating the navigation skill needed when traveling between two points after initial discovery. The duration ofthe drift voyages was set to 75 days after which it was assumed that the crew perished. Seventy-five days would by no means be the longest drift voyage in an open boat due to shipwreck or other misfortunes documented in the TABLE 1. SUCCESS RATES FOR DRIFT VOYAGES TO PALAU AND DURATIONS IN DAYS JANUARY FEBRUARY MARCH APRIL MAY JUNE JULY AUGUST SEPTEMBER OCTOBER NOVEMBER DECEMBER Taiwan 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% Samar 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% Mindanao 0% 0% 0% 0% 0% 1% (42) 2% (11,19) 2% (11, 13) 11%(10-30) 10% (13-56) 1% (22) 0% Halmahera 0% 0% 0% 0% 0% 10%(15-32) 17%(8-17) 18%(9-28) 35% (9-34) 5% (39-75) 0% 0% Teba NG 0% 0% 0% 0% 0% 0% 1% (15) 0% 0% 0% 0% 0% Bismarck 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% Archipelago CALLAGHAN AND FITZPATRICK . PREHISTORIC MIGRATION PATTERNS 35 Pacific (Howay 1944; Levison et al. 1973:20-21). The maximum recorded drift seems to be on the order ofseven to eight months and several ofthese have been documented in the last few decades. A number ofthese recorded voyages covered distances ofaround 5500 km over a period ofsix to ten weeks, and even more traveled slightly shorter distances. For the purposes ofthis study, 100 drift voyages were simulated for all 12 months ofthe year from each ofthe points oforigin. RESULTS OF SIMULATIONS Table 1 gives the success rates for drift voyages to Palau and the durations in days. No voyages were successful from Taiwan, Samar, or the Bismarck Archipelago at any time ofthe year. From the area ofTeba on New Guinea, only one voyage in July conditions was successful, lasting for 15 days. Figure 2 shows the drift voyages for July. From Mindanao, drift voyages were successful under June to November conditions (Fig. 3). Success rates are from 1 to 11 percent with dura tions of 10 to 56 days. The average number ofdays successful voyages took was 23. Voyages starting in September had the highest rate of success (11 percent). Voyages originating offHalmahera had by far the greatest chance ofsuccess from June to October. Rates ofsuccess were from 5 to 35 percent. Voyages beginning under September conditions (Fig. 4) had the highest success rate. Durations ranged from 8 to 75 days with an average of 26 days. It should be noted that under October conditions, the average durations were 62 days which somewhat biases the overall average. Simulations are also shown from the Bismarck Archi pelago (Fig. 5) to show drift voyages from a point that were completely unsuc cessful in reaching Palau. DISCUSSION AND CONCLUSIONS The computer simulations we conducted suggest that voyagers drifting from off western New Guinea would have had only a minimal chance of reaching the Palauan archipelago while Taiwan, the Bismarcks, and Samar were unlikely ori gins. Peoples sailing before the wind from Mindanao and Halmahera to Palau would have had the most success oflandfall. The high rate ofsuccess at finding Palau from Halmahera (35 percent) is explained by the higher than normal winds coming from the southwest during the month ofSeptember (Morris 1988). In ad dition, sailors departing from the southern Philippines, even ifthey were unable to reach Palau, would have had a good chance ofreturning home. This also helps to explain the significant number of Carolinian canoes that ended up drifting there (Hezel 1972:27-28). These computer-simulated drift voyages support an initial discovery and migra tion from the southern end ofthe Philippines based on the high rate ofsuccess in finding Palau. IfPalau was in fact first settled by Austronesian (western Malayo Polynesian) speaking groups from the Philippines (see Bellwood 1997:105-107; Craib 1999; Wickler 2001), it appears more likely that they would have come from the most southerly ofthe islands. It should be noted, however, that we are not suggesting that peoples could not have voyaged to Palau from other locations either purposefully or accidentally. Our simulations do suggest chance occur rences ofdrifting from New Guinea to Palau, but this was very low (1 percent). Based on mtDNA lineage analyses of Palauan samples, Lum and Cann ( \ :--... t N .~~ L.-...J 150nm ~Guam ) jJ 4_~, ~ /:::-- -/J(Y-C-'..-.-<1>- ~- ('> ,,0 ~ s~s:' Ha~lmahera /~ ~.-~-.- ~7 A.r~~i~'> ' w __,,7.\~\ ~~Ne'"WGlJl~ea:.,_Bismal<;;1('.• 'f .' ..... ;;.'. I .,", I"J'. '-. c . .~ , Q L 'o'q; r5r1 . . .. .... ~~ c:::> ~~. """","'\. ~~.:::::::"~0"c1D~ --'" ~ £~...-.-./.--.--- "=::::, ~ ~~ J~: \_~ ~. dJ ' .' .r J. ",. . o t", ''",' r-' ).--.,.) ,( , <. Fig. 3. Simulations from Mindanao to PalauinSeptember.