The main Hawaiian Islands are situated ~20° north latitude within the subtropics. The main high volcanic islands typically have a wet windward side where moisture-laden northeast trade winds deposit most of the orographic rainfall during November to March, that over millennia have sculpted deeply incised, amphitheater-headed valleys where rich alluvial soils fostered intensive aroid cultivation, primarily kalo (Colocasia esculenta), the foundation of Hawaiian society from earliest times, nearly a thousand years ago (Weisler et al., 2023a). The western, leeward regions are often positioned in a rain shadow and, consequently, the landscapes are typified by rolling hills and broad slopes covered with mesic and dry-adapted sparse trees and shrubs with natural or anthropogenic savannas. The typical windward-leeward divide is characteristic of the elongate island of Moloka‘i (61 × 16 km; 676 km²), built of two broad shield volcanoes that overlap in a central saddle. The eastern volcano rises 1,515 m above sea level, while the western and oldest volcano at Mauna Loa is just 421 m elevation (Figure 1). About a dozen cinder and spatter cones lying primarily along the southwest and northwest rift zones of Mauna Loa volcano are Pleistocene-age late-stage eruptions, many of which have fine-grained basalt used for adze manufacture (Weisler, 2011). Indeed, there are more documented adze quarries and sources here than on any other island in the archipelago; thus appropriately, the traditional land unit (ahupua‘a) is named Kaluako‘i—literally, the adze pit.

The underlying geology dictates the nature of the shorelines, coastal habitats, and the slope and depth of offshore waters (Figure 2). For example, on Moloka‘i, sitting atop the shallow, flat offshore substrate is a 1 km wide barrier reef that fronts most of the south shore (Figure 2E). In contrast, the north coast descends precipitously resulting in deep nearshore waters; this is seen in Figures 2A, B where dark blue, deep water is close to shore. The implication here is that deep water benthic fishing could be conducted near the north and northwest shores targeting snappers (Lutjanidae) and large trevally (Carangidae), while the south coast is ideal habitat for seine netting surgeonfish (Acanthuridae) and spearing parrotfish (Scaridae) amongst others. In the broader arena of marine foraging, the contrasts in the north and south coasts are seen in the distribution of more sessile foods such as helmet urchins (Colobocentrous atratus) and limpets (Cellana spp.) that are far more common on the rocky, wave-splashed intertidal zone along the north and northwest shores (Figure 2D; Rogers and Weisler, 2020, 2021, 2022; Weisler et al., 2020). These broad patterns bely the fishing knowledge acquired over dozens of human generations where marine foragers developed an intimate understanding of the daily, monthly and seasonal movements, aggregations, and spawning events of routinely captured fish species—knowledge that was recorded by early Hawaiian scholars (e.g., Kāhā‘ulelio, 2006; Malo, 2020; Manu and Kawaharada, 2006). Today, much of this knowledge continues to be passed through example such as by participating in fishing sorties, and from word of mouth with many of the island's elders, such as Uncle Mac Poepoe, who has incredible insights into marine lore and traditional management practices specific to the north coast of the study area. Indeed, Poepoe produced a “Pono Fishing Calendar” in 2011 with monthly information on fish spawning, no-take (kapu) periods, how to distinguish male and female fish to determine spawning and foster sustainability practices (“don't just eat your favorite fish”), and seasonal changes in habitats (see Poepoe et al., 2007). On this latter point, most local fishers know that the sandy shoreline along Kalani (Figure 1) is well-known for threadfin (moi, Polydactylus sexfilis) which, as will be demonstrated below, is common in archaeological sites within the immediate area. However, only with multi-year observations was it recognized that the large-scale movement of sand, especially in Kawa‘aloa bay, represents seasonable shifts and therefore changes in local habitats tied, in some cases, to spawning periods and the presence of threadfin. Additionally, large trevally (ulua, Caranx ignobilis, C. lugubris) are routinely caught today on lines tossed from the rocky north coast into nearby deep water, a modern technique called “slide-bait” where a weight and line, attached to a pole, are cast offshore and fixed to the bottom, then short, baited leader lines are slid down the line, resting above the bottom. A similar traditional technique called kau lā‘au is still used today on Hawai‘i Island. There is evidence, for a related, or perhaps identical, fishing technique evidenced by the bones of 20 kg ulua recovered from coastal archaeological sites.

Continuing west along the coast from Kawa‘aloa bay and the sandy shoreline at Kalani, the sea cliffs rise over 100 m making the narrow, rocky coast only accessible by boat, except for a couple of steep trails descending, in one instance, from an isolated habitation complex ~1 km east from ‘Īlio (Weisler and Rogers, 2020: Figure 1). Rounding ‘Īlio, the seas are often rough as windward-generated swells wrap around the point and collide with oncoming currents. The safest canoe landing from here is a rocky embayment just north of one of the study sites, rock shelter Mo5 (Figure 2D). Further south, the west coast has several shallow sandy embayments punctuating the rocky coast including a 3 km long white sand beach at Pāpōhaku and a 1.3 km stretch of beach at Kamāka‘ipō before reaching Lā‘au at the southwest point. East of Lā‘au there is a near continuous “soft shore” of terrigenous silt descending from upland slopes exacerbated from historically introduced ungulates (mostly cattle and deer) and coralline marine sands interrupted by rocky points (Figures 2E, F; Rogers and Weisler, 2024).

With the above environmental context as the template, the general characteristics of the archaeological landscape are framed. Surrounding the summit of Mauna Loa volcano are the only likely permanently occupied residential complexes, many associated with extensive agricultural fields and the occasional temple (heiau). Marine shellfish and fish bones found in many of the habitation sites point to either direct access to the coastal areas, no closer than 5 km distant, and/or a mauka-makai (mountain-sea) interaction network for the transfer of goods—a system recorded for Ka‘u, Hawai‘i Island nearly a century ago (Handy and Pukui, 1972) and reasoned to have greater antiquity (Weisler et al., 2023b). Descending seaward from the summit is a broad ring surrounding the volcano devoid of archaeological sites, essentially a barren zone, with only the occasional, temporarily occupied small stone constructed shelter. It is not until reaching the coast, on all three shorelines, where extensive habitation complexes are encountered. Situated primarily around bays and other parts of the coast with easy access to the shoreline, are residential complexes consisting of low, dry-laid stone walls in the form of rectangular, square, C-shape and L-shape structures surrounding level soil areas and stone filled free-standing platforms (Weisler et al., 2006: Figure 1). Extensive coastal middens, common world-wide, are relatively rare. Sited on promontories with clear views along the coast and out to sea are fishing shrines (ko‘a) where all manner of seafood were deposited for first fruits rituals for ensuring plentiful harvests and marking offshore fishing grounds (Hiroa, 1964, p. 529; Handy, 1927, p. 296–311; Valeri, 1985, p. 75; 183; 184; Weisler and Rogers, 2020; Weisler et al., 2022). With this settlement landscape as context, in 1952, Bonk (1954) located six rock shelters, one house site and two middens most within 100 m of the shoreline; these sites are described below.

As part of his master's thesis research supervised by Kenneth P. Emory of the Bishop Museum, William Bonk and a fellow student from the University of Hawai‘i, conducted an extensive month-long archaeological reconnaissance survey along the north, west and south shores of west Moloka‘i recording “numerous shelters and house sites, mapping surface structures and collecting artifacts” (Bonk, 1954, p. 4). Nine sites were chosen for extensive excavation and selected site characteristics are presented in Table 1. Adhering here to the archaeological site numbering scheme used by Bonk (1954), all site numbers are prefaced with Mo for Moloka‘i, then the unique site number; for example, Mo1.

Imagine having limited archaeological field experience, typical of a first-year master's student, yet you co-direct the excavation of nine pre-contact habitation sites with relatively inexperienced assistants totalling a staggering 53.4 m³–in just 2 months! Bonk's fieldwork took place three years after the publication of the influential A Guide to Archaeological Field Methods in 1949 edited by Robert Heizer then of the University of California at Berkeley. It seems likely that Bonk consulted this volume as many of his field methods are identical to some of those described in the book. Bonk (1954, p. 5–6) mapped each site prior to excavation, either with a transit or carpenter's level and compass and steel tape. A site datum was established, and a grid system was defined using wooden stakes at the intersection of 3-foot excavation squares (Heizer, 1958, p. 34–36; Figures 6, 7D). Within units, the sediment was excavated using primarily short-handled spades—as seen at Mo1, atop unit E7 in Figure 4B—in “six inch levels [as was] customary” (Heizer, 1958, p. 53), a methodology referred to as “the American practice of metrical stratigraphy” (Heizer, 1958, p. 42). (Unfortunately, Bonk did not report if all the site sediments were sieved, the screen sizes used, or the systematic collection of midden and artifacts.)

Bonk excavated >50% samples from most sites (Table 1). Like what Heizer (1958, p. 35) describes, once the stratigraphic section was exposed, adjacent units were excavated toward the center of the site. A single unit, usually near the thickest portion of the cultural deposit at each site, was reserved for “quantitative collecting” (Bonk, 1954, p. 5, 119; Figure 4B) where the weight of shellfish, fish bones, urchin, charcoal, crustacean, candlenut (Aleurites sp.), “animal bones” (presumably, non-fish), and vegetable material and the count of “basalt chips” were reported (Bonk, 1954: Table VIII). The presence (ubiquity) of midden constituents were noted at all sites with major classes, using Bonk's terms (Bonk, 1954: Table IV), including “vegetable” (charcoal, gourd, candlenut, pandanus, wood, and other), “animal” (bird, crustacean, deer, dog, echinoidea, fish, fowl, man [sic], molluscs, pig, rat, turtle animal bone) and mineral (basalt chips, cooking stones, hematite, obsidian (actually volcanic glass), pumice, quartz and red ochre). This sampling procedure, designed for comparing midden constituents between sites, was so influential that a similar strategy was used 25 years later to quantitatively compare the weight of midden components between six cave sites on Hawai‘i Island (Kirch, 1979, p. 121–140). Most site excavation bags were labeled with the site, grid unit, depth and date of excavation.

Bonk's field techniques in the early 1950s were reasonable for the time, especially considering the state of archaeological field research in Oceania, yet it is difficult to apply contemporary analytical techniques or use statistical applications without knowing more precisely how his material was collected. Weisler interviewed William J. Bonk and two crew members (Catherine “Cappy” Summers and Mary Judd née Stacy), in 1989, to record their memories regarding precise field techniques used during the excavation and collection of artifacts and faunal materials as explicit collection methods were not described in Bonk's master's thesis (Bonk, 1954). It is only fair to reiterate that these interviews were conducted almost 40 years after the field work. I concluded from these discussions that much of the cultural deposits were sieved using one or more of 3.2 mm, 6.4 mm, or 12 mm mesh screens. In fact, Cappy Summers showed me a 3.2 mm sieve kept under her Moloka‘i house that was used during field work; one such screen is seen in Figure 7D. All highly shaped artifacts were collected, while most basalt flakes were discarded—this would have included most retouched and used flakes. “Type samples of all extraneous matter” (Bonk, 1954, p. 119) were retained including volcanic glass flakes, chalcedony nodules and flakes, quartz, whole and well-preserved marine shellfish, “diagnostic” fish bone (including a large otolith from the bone fish, Albula sp.) and the larger mammal and bird bones. From the review of the sites, units and levels written during field work on midden bags, only some units were excavated in six-inch (15 cm) arbitrary levels while, for others, artifacts and faunal materials were assigned only to a grid square (see Section 5 below).

Bonk (1954, p. 5–6) used excavation level record forms for each grid square recording the location of artifacts, faunal and floral remains, type of soil (sediments), stratification and other features. Additional information was recorded in a field notebook, a field catalog listed all artifacts, and a photographic record was made during the excavations (after Heizer, 1958). During attempts over many years, every reasonable effort was made to locate these records and missing archaeological collections. Weisler checked multiple times with the senior archaeologist Peter Mills at the University of Hawai‘i at Hilo where Bonk worked for decades until his retirement, the Archaeology Lab Manager Jo Lynn Guinness at the Department of Anthropology, University of Hawai‘i at Mānoa—the campus where Bonk completed his master's thesis, Betty Lou Kam at the Bishop Museum Archives (Honolulu, Hawai‘i) and Stephanie Lambert (Archaeology Collections Manager) also at Bishop Museum. Weisler was able to locate Bonk's son-in-law and asked him multiple times over many years to see if the missing records and archaeological materials may have turned up at Bonk's house, long after he passed away on 25 November 2008. To date, the missing records and archaeological collections have not been located and are presumed lost or destroyed.

The University of Queensland archaeological field school was directed by Weisler for the primary purposes of collecting dating material for determining site age and duration of occupation and obtaining a faunal record to compare to the 1952 excavations. Mo1 was chosen for excavation as surface indications suggested that intact archaeological deposits remained, it was reasoned to be one of the earliest sites in the region, and fauna was abundant and well-preserved considering what Bonk had collected in 1952. Three 1m² units were gridded out on a low remnant hill outside the dripline near the center of the presumed unexcavated area of the rock shelter (Bonk, 1954: Figure 2; Figure 3A). It was not possible to locate the precise location of any of Bonk's units due to erosion over the past 67 years and the fact that almost the entire cultural deposit within the rock shelter was removed. A 0.50 × 1.0 m unit was positioned well-outside the dripline in a soil area surrounded by very large sandstone blocks (Figures 3A, B). Over the course of 2 weeks in late June and early July 2019, the field school staff and students conducted six full days of excavation and 5 days of in-field lab work with 16 people.

Excavation proceeded in 5 cm thick spits (arbitrary levels) within cultural layers. Sediment characteristics (color, texture, consistency, etc), disturbances, and features (combustion features, artifact and faunal concentrations) were recorded for each spit. Photographs were taken of all features and stratigraphic sections. All sediments were sieved through 6.4- and 3.2-mm screens. The material from the 6.4 mm screen was gross sorted on-site, while the 3.2 mm size class was water-sieved in the field lab, dried and gross sorted into midden constituents. All residue remaining in the screens after field sorting was checked by Weisler prior to discard. Shellfish and urchins were preliminarily identified at the field lab, while fauna was identified to lowest taxon at the University of Queensland using extensive reference collections. All cultural material was retained. Additionally, a column sample was taken at unit 2 and floated in the field to capture organics and all cultural material retained for future analysis.

Identification of fish bones and otoliths were made to the lowest taxonomic level constrained by the completeness of the reference collection for each family. Weisler's reference collection includes 332 specimens consisting of 36 families, 80 genera and 153 species. Here, we agree with Gobalet (2001, p. 377) that it is essential to describe the reference collection used for the identification of archaeological fauna (Cannon et al., 2018: Supplementary Table S1; Weisler, 2001: Appendix 3) and the facilities where the assemblages are curated; in this case, at the Bernice Pauahi Bishop Museum, Honolulu. Photos of archaeological fish bones are included to illustrate examples of elements used for identification to taxon, the general state of bone preservation and as a photographic record of the some of the identifications (Figure 8). A tag on the bone in Figure 8L is one of the earliest examples of archaeological fish bone identification in Hawai‘i. Genus and species identifications in the current study were made when the reference collection had most or all genera for a family or most species for a genus. Monotypic taxa, such as threadfin (moi) Polynemidae Polydactylus sexfilis, was routinely identified to species because there is only one genus and species in the Hawaiian Islands. Fish otoliths were identified using the fish reference collection and scanning electron images of Indo-Pacific fishes (Rivaton and Bourret, 1999; also, Smale et al., 1995). Images of selected archaeological otoliths were captured with a desktop scanning electron microscope (SEM; Hitachi TM3030plus) and a Documents camera DC110. Taxonomic fish names are after (Mundy, 2005). The provenance of each fish bone was recorded, along with the identification, element, side, condition (whole or fragment), presence of burning, reference specimen used for identification, weight of bone element, estimated live weight when possible and notes. Live weight was estimated by comparing relatively complete archaeological bones to reference specimens of known weight (see, e.g., Bouffandeau et al., 2018; Goto, 1986, p. 412–422; Rurua, 2020; Weisler et al., 2024), then assigning reconstructed sizes to a particular weight bin, for example, 0–25 g (usually very young threadfins, Polydactylus sexfilis), 200–300 g, 500–1,000 g or as heavy as 20 kg (e.g., large trevally, Caranx lugubris). The concentration index for identified fish bone weight per site was used to infer the relative intensity of fish capture, consumption and site function. For Bonk's 1952 excavations, all bones were considered for identification to lowest taxon. For the Mo1 unit 2 (2019) fish bones, all elements were considered for identification to taxon that were retained in the 6.4 mm sieve, while all bones, not including most vertebrae, were identified from the 3.2 mm sieve. Any relatively whole archaeological fish bones that couldn't be identified further than “fish” were noted as a check on the completeness of the fish reference collection. There were only nine nearly whole elements that could have been identified with an expanded reference collection. These included one each of a basypterygium, quadrate, parasphenoid, cleithrum, dentary and ultimate vertebra, with two each of an opercle, pharyngeal and supracleithrum. This suggests that the reference collection was fit for purpose.

Fish bone quantification is routinely reported as the number of identified specimens (NISP) and the minimum number of individuals (MNI) listing the number and percentage of each taxon by layer or site. Both measures are complimentary and have their strengths and weakness that have been discussed at length over the years (Grayson, 1984; Lyman, 2008, 2018; Reitz and Wing, 1999). However, Albarella (2016, p. 314) opines that “… quantification, and the measures on which it relies (recording and counting), remain some of the most misunderstood and problematic areas in zooarchaeology.” Considering the site contextual information and how little we know about collection techniques used for the legacy collections analyzed here, we use NISP for quantification. While the degree of fragmentation influences NISP, MNI is highly correlated to the units of aggregation in that for the same assemblage, smaller units of aggregation will predictably yield a higher MNI (Grayson, 1984, p. 29). Ideally, we would include MNI for each site using the stratigraphic context. However, Bonk assigned fish bones to arbitrary 6-inch (15 cm) levels for some fish bones, but to entire units for others. Therefore, the only consistent unit of aggregation is each entire site, which would yield the lowest possible MNI. While it may seem reasonable to quantify MNI by site, this level of aggregation would not be very meaningful across sites with excavation volumes ranging from 0.4 to 24.3 m³ (5.93 ± 7.56). Consequently, we use NISP for comparing all site assemblages.

The lowest depth for each unit that had fauna was determined, then the volume of each unit calculated. For those units that were excavated without levels, the maximum depth of cultural deposits was estimated by examining the stratigraphic profiles that were closest to each of these units. Using this procedure, Mo1 had 18.19 m³ of cultural deposits with fauna that was used in the present analysis. Because the fauna from Mo8 and 9 was only provenanced to the site and not individual units, the volume for the site was used as the “analytical volume.” However, for Mo7, Bonk (1954, p. 50) stated that “the maximum depth of refuse deposition was found to be 9 inches” (~23 cm) so that value was used for all excavated units.

Diversity indices were calculated in PAST v4.15 (Hammer et al., 2001) using raw fish NISP counts at the family level for each site. For each assemblage, species richness (NTAXA), Shannon diversity (H′), Simpson diversity (1–D), Fisher's α, and exponential Shannon evenness (e^H/S) were calculated to assess taxonomic diversity and evenness independent of sample size (Colwell, 2009, p. 260; Heip et al., 1998, p. 63, 80; Legendre and Gallagher, 2001; Morris et al., 2014, p. 3515; Rogers and Weisler, 2024). These metrics quantify both the number of taxa and their relative abundances, allowing comparison of assemblage heterogeneity across sites. To evaluate compositional similarity among sites, chord distances were calculated on the same abundance data. Chord distance is a Euclidean metric, which measures dissimilarity in relative composition while minimizing the influence of sample size (Hammer et al., 2001, p. 172; Hayek and Buzas, 2010; Magurran, 1988, p. 35–37). Pairwise chord distances were used to identify assemblages with closely related taxonomic structures and to assess overall variability in fish assemblage composition across the Moloka‘i sites.

Compositional relationships between sites and fish taxa were also explored using Correspondence Analysis (CA) in PAST v4.15 (Hammer et al., 2001). The analysis was performed on fish NISP counts at the family level to identify patterns of taxonomic association and variation among sites. CA is a multivariate ordination technique that applies a x² distance metric to abundance data, producing a biplot in which samples (sites) and variables (taxa) are displayed in a common ordination space (Greenacre, 1984; Hammer et al., 2001). Axes represent the principal gradients in taxonomic composition, with the percentage of inertia on each axis reflecting the proportion of total variance explained. The resulting biplot was used to visualize which taxa contribute most strongly to compositional differences and to identify clusters of sites sharing similar fish assemblage structures.

Paramount to better understanding the collection techniques used in the 1952 excavations, bone weight provides insights into potential selection bias (were larger fish bones preferentially collected?), while fish bone weight per site in relation to excavated volume and number of identified bones can be related to consistency in fish bone collection and intensity of fish bone deposition. We explore these issues here.

For comparisons between the 1952 and 2019 studies for Mo1, all scales, otoliths and teeth were removed from unit 2 and only the weight for fish bones was calculated. Mo1 (Bonk) included 997 fish bones with a mean weight of 0.54 ± 0.77 g, range 0.01–11.44 g and total weight of 533.2 g. Unit 2 had 1,482 fish bones with a mean weight of 0.16 ± 0.42, range 0.001–6.03 g, and a total weight of 229.5 g. In other words, 67% more identified fish bone was recovered from ~1.2 m³ (unit 2), than 30.9 m² (18.2m³) of excavation during the 1952 excavations; the volume of cultural layers excavated in 1952 was 15.2 times more than in 2019 (Note that only units with fish bones were included as not all the 1952 excavated units were available for study.) Fish bones collected by Bonk and colleagues were 3.4 times heavier per bone than those retained in 2019 suggesting that the screeners at Mo1 from the 1952 field work were preferentially selecting larger fish bones from their sieves which interestingly yielded much less overall bone by weight perhaps because not all large bones were collected.

But were screeners collecting bones consistently across all nine excavated sites? On the one hand, there really is no way to know for sure. However, with a r² of 0.988 (p = < 0.001) for the linear relationship of excavated volume to the weight of fish bones, and a r² of 0.985 (p ≤ 0.001) for excavated volume and fish bones identified to at least the family level, this suggests that fish bones from all nine sites were collected in a similar manner.

Table 2 lists the elements used for identification to taxon, by rank order abundance, for the 1952 and 2019 assemblages for Mo1. Some 40 unique elements were identified in the 2019 assemblage for unit 2, while 31 elements were identified in the 1952 assemblage for the entire Mo1 site. The pterygiophore category also includes first and second dorsal and ventral pterygiophores and anal spines of Holocentridae. Elements routinely identified in the 2019 assemblage but not in the 1952 material includes Balistoidea scales (not included in quantification measures) used here for the identification of triggerfishes, leatherjackets or filefishes taxa, otoliths, the surgeonfish (Naso sp.) horn, post coracoid process, ceratobranchial and mesocoracoid of threadfin (Polydactylus sexfilis), dermal spine of porcupinefish (Diodontidae), carapace plate of boxfishes (Ostraciidae), and vomer of a trevally (Caranx sp.). Some fragmentary elements were so small that they were only identified to premaxilla or dentary. Figure 8 illustrates examples of elements used for identification, some species identified, and the excellent state of preservation of the bones.

All vertebrae were identified from Bonk's assemblages and the fish bones retained in the 6.4 mm sieve for unit 2. The vertebrae from the 3.2 mm size class at unit 2 were not systematically identified, but they were checked for species not identified in the 6.4 mm size class.

It is no surprise that only one large otolith was recovered in 1952 as otoliths are rarely reported today from Pacific Island assemblages, while 40 were retained, mostly in the 3.2 mm sieve, in 2019. Otoliths are especially useful for genus and species identifications and sometimes their identification contributes new taxa identified from archaeological sites in the Pacific Islands such as in the Hawaiian Islands (Weisler, 1993 and herein), New Zealand (Weisler et al., 1999) and New Caledonia (Weisler, 2002). These identifications are discussed below. Examples of otoliths identified from Mo1, unit 2 are illustrated in Figure 9.

The collection practices in 1952 focussed on retaining relatively larger fish bones during screening which resulted in a heavier reconstructed individual fish weight for most taxa compared to the Mo1, unit 2 excavations, where smaller sieves were used, and collection of all fish bones resulted in a broader range of fish sizes per taxon. Figure 10 illustrates the range of reconstructed weights for threadfins, surgeonfish and wrasses showing that consistently larger fish, with fewer smaller fish, were collected in 1952. This has implications for interpreting fishing strategies, habitat targeting and sustainability practices which are discussed below.

Table 3 lists the number of identified fish bones (NISP) and percent NISP for Mo1-9 including one subclass, one order, one suborder, 29 families, 35 genera and 16 species. Unusual for any assemblage of fish bones from the Pacific Islands, threadfin (moi, Polynemidae Polydactylus sexfilis) is the highest ranked family at 20.42% for the combined sites. Only three reef-associated families comprise 58.26% of total NISP: surgeonfish (Acanthuridae), wrasses (Labridae) and threadfins. Just from visual inspection of Table 3, species richness is clearly related to site sample size. These identifications are discussed below.

Diversity indices reveal marked variation in fish taxonomic richness, evenness, and overall assemblage structure among the nine Moloka‘i sites (Table 4). The 2019 excavation of Mo1 (unit 2) yielded the highest diversity across all metrics (NTAXA: 25; H′: 2.18; 1–D: 0.84; Fisher's α: 4.20), slightly exceeding the 1952 Mo1 assemblage (NTAXA: 23; H′: 2.14; 1–D: 0.83; Fisher's α: 4.19). This difference likely reflects the adoption of finer screens used during excavation and recovery methods in 2019, yielding a broader spectrum of small-bodied and reef-associated taxa.

Other sites display markedly lower richness (NTAXA: 5–12) and diversity (H′: 1.46–1.57), which could be explained through coarser recovery, reduced occupational intensity, relatively limited fishing focus, or specialized fishing targeting particular taxa. For example, Mo2 has moderate richness (NTAXA: 12; H′: 1.46; 1–D: 0.65), dominated by a small number of taxa. The low evenness (E: 0.36) suggests that one or two families contributed disproportionately to the assemblage, likely driven by the high abundance of Polynemidae and, to a lesser extent, Kyphosidae.

Mo3 and Mo4 are compositionally very similar (chord: 9.38) and both are characterized by low richness (Mo.3 NTAXA: 5; Mo.4 NTAXA: 7) but moderate to high evenness (Mo3 E: 0.96; Mo4 E: 0.66). Despite low richness, the balanced representation of taxa indicates consistent and uniform exploitation of a few key taxa, with Mo4 diversity being influenced by some dominance from Pomacentridae (1–D: 0.691). Mo5 and Mo.6 display intermediate richness (Mo5 NTAXA: 14; Mo6 NTAXA:15) and diversity (Mo5 H′: 2.06; Mo6 H′: 1.82), reflecting a moderately varied fish catch. Evenness values suggest that while several taxa are represented, a few are more dominant. The data suggest a relatively mixed fishery drawing from multiple nearshore habitats. Mo7, 8, and 9 all show moderate richness and evenness, with similar representation of common reef taxa.

Chord distance values also indicate clear similarities in fish-family composition between the Moloka‘i sites (Table 5). The lowest distances occur between Mo9 and Mo6 (0.05) and Mo9 and Mo7 (0.14), indicating that these assemblages are compositionally very similar and likely reflect exploitation of similar reef or nearshore habitats or similar capture strategies. Mo2 shows the greatest difference to all other assemblages, likely due to the abundance of Polynemidae. Both Mo1 assemblages (1952 and unit 2, 2019) are markedly different from all other assemblages, exhibiting higher chord distance values that signal distinct assemblage composition. As both the 2019 and 1952 assemblages display this difference, it is likely that the Mo1 dissimilarity represents a more taxonomically diverse fish assemblage influenced by local environmental variability, people's fishing decisions and sample size, rather than purely the influence of recovery techniques. Overall, chord distance results show that most Moloka‘i assemblages share a common reef-dominated signature, while the fish assemblages from Mo1 and Mo2 are more distinct.

Correspondence Analysis (CA) revealed clear patterns of variation in fish-family composition among the Moloka‘i fish assemblages and supported the results of the diversity indices (Figure 11). The first two CA axes account for 45.24 and 24.32% of total inertia, respectively, summarizing almost 70% of the variance in taxonomic composition (Supplementary material S1). The 1952 and 2019 Mo1 assemblages plot distinctly from all other assemblages and are positioned near Acanthuridae, Balistidae, and Labridae families, reflecting broad and diverse fish exploitation. The Mo2 assemblage is compositionally distinct from all the other sites and is positioned toward Polynemidae, indicating the high abundance of threadfins in the Mo2 assemblages compared to the other sites. The remaining sites (Mo3–9) cluster closely near the center of the ordination, representing more compositionally uniform reef or nearshore assemblages with fewer contributing taxa. Overall, the CA shows that while most Moloka‘i sites share a common reef-based signature, Mo1 (1952, 2019) and Mo2 display distinct compositional patterns.

Aside from the taxonomic identification of fish bones, reconstructed fish size is one of the most important characteristics for making inferences about the habitats targeted by ancient foragers, capture techniques used and sustainability practices. In a common fish life cycle, juvenile fish often congregate in shallow inshore habitats somewhat protected from larger predatory fish (Figure 12). Depending on the species, individuals, later in life, establish on the broad reefs, seaward facing slopes or in deep waters. There are movements on a daily basis where, for example, ‘iao (Hawaiian silverside, Atherinomorus insularum) form schools inshore in bays (such as in Kawa‘aloa), then disperse at night to feed (Randall, 2007, p. 136) However, sexually mature fish participate in spawning events that are dictated by seasons where spawning, in general, occurs during periods of increased current strength or reversals of direction that are beneficial to dispersing spawn away from predators (Johannes, 1978, p. 68; Sadovy, 1996, p. 46). Spawning is also in sync with moon phases (May et al., 1979). Only fish of a certain size, unique to individual species, participate in periodic spawning movements, whether dictated by monthly (moon phases) or seasonal changes. Ancient fishers were aware of the timing and locations of spawning events and could target these aggregations at specific habitats (Johannes, 1981; Kāhā‘ulelio, 2006). Indeed, threadfins spawn during the northern summer months (June to August) close to shore in Kawa‘aloa bay when the sand stops moving toward shore and before it infills holes in the reef (Poepoe et al., 2007, p. 334; also, May et al., 1979, p. 900). Because threadfin reach sexual maturity at 20–29 cm for males and 30–40 cm fork lengths for females (Callan et al., 2012, p. 3), individuals of these size ranges were likely captured along the shoreline at Kalani and inside Kawa‘aloa bay close to the sandy beach (as Weisler has done with seine net), especially during spawning periods (Lowell, 1971). Seine nets are a productive technique to maximize capture of nearshore aggregations.

With preferential collecting of the larger fish bones in 1952, most bones represent the heavier size of fish caught as bones from smaller fish are under-represented in the assemblage. Considering the distribution of fish size for threadfin in Figure 10, ~90% of the bones collected in 1952 represent sexually mature threadfins; that is, heavier than about 400–500 g which is sexually mature for males and females (Callan et al., 2012, p. 3; Chu et al., 2011). The 1952 data, then, suggest a foraging practice of harvesting mostly older individuals (only 13% are below reproductive size) ideally after spawning events. However, when the 2019 data are added to the mix, a much different picture emerges as 75.5% of bones represent fish of non-reproductive age. These combined datasets argue for unsustainable capture or “recruitment-overfishing” (Froese, 2004, p. 86) of this species as declines in modern catches is thought to be related to the capture of moili‘i or juvenile moi, less than ~18 cm (Rao, 1977, p. 22).

With the addition of the 2019 bones of juvenile threadfins to the combined assemblage—37.7% had reconstructed weights less than 100 g or ~110 days old (Chambers et al., 2001, p. 5)—it is likely that habitats targeted by ancient foragers included tidal pools at the base of the cliffs immediately west of Kalani (Figure 12; see also Hosaka, 1973, p. 92; Rizzuto, 1978, p. 216) and along the sandy shores (Malo, 2020, p. 2; 118) where juvenile threadfins could have been captured using fine mesh (nae) scoop nets (Hiroa, 1964, p. 301, Figure 207a). Targeting juvenile fish may have contributed to an unsustainable fishery.

Because the threadfins had a relatively high NISP (they comprised the dominant taxon when combining all sites) and were identified to the species level it was possible to use ecological, spawning, habitat and capture data specific to the species. Species level identifications are ideal as many families are species rich, such as the wrasses, surgeonfish and carangids, with fish sizes and habitat preferences varying greatly within each family thus limiting the use of species-specific ethnohistoric and ecological information.

Fish weight can also inform on spawning seasons when the presence of sexually mature individuals of certain species is determined by the reconstructed weight represented by specific fish bones. For example, relatively large individuals of carangids (Caranx lugubris) spawn during March to June, threadfins (Polydactylus sexfilis) June to August, several species of parrotfish (Chlorurus perspicillatus, C. spilurus) spawn during March to April and September to November (Scarus rubroviolaceus). Sites with sexually mature fish of all these species were occupied at least during March to November. This makes sense since the seas are rough and the waves much bigger during the winter months (December–February) when it is riskier to forage along the rocky shoreline or go to sea in a canoe.

The density of fish bone per site provides an approximate indication of the intensity of fish capture, consumption and deposition (e.g., Cannon et al., 2018). Table 6 lists the rank-order concentration indices (CI) of fish bone for Mo sites 1–9. CIs range from 3.72 g m³ at Mo4 to 191.25 g m³ at Mo1 (unit 2). Assuming that collection techniques were similar at the sites excavated by Bonk (as we have discussed previously), there are some interesting differences between sites. For example, Mo4 contained the most bird bones of the endemic Hawaiian goose (nēnē, Branta sandvicensis; Weisler and Gargett, 1993: Table 1), was the farthest from the coast (at 600 m) of any site and at the highest elevation (Table 1, Figure 1). With the lowest CI of fish bones, the site may have been a bird hunting camp with little emphasis on fish exploitation. Likewise, the small rock shelter Mo3 was the closest to the large basalt adze quarry nearby as exemplified by the densest concentrations of adze materials. At 500 m from the coast, a CI of 4.13 g of fish bones clearly shows that fish consumption was minimal. Despite a small, excavated sample of a site with unknown dimensions, the dune midden (Mo9), only 100 m from the shoreline, was sparsely used considering not only fish bone deposition, but the low quantities of marine shellfish and other midden (Bonk, 1954). This low CI is in marked contrast to the midden accumulation immediately north that is associated with the house foundation (Mo8) close to the water's edge. With a CI of 179.35 g, the house site, although set amongst a sparse, sprawling coastal midden, seemingly acted to focus the midden (fish bone) deposition. Mo2 (CI = 48.71) and Mo6 (CI = 56.92) contain dense middens with many ash lenses and ovens and were staging locales for exploiting the nearby coasts with an emphasis on fish and shellfish. Mo1 was the largest excavation which resulted in a broad array of fishing gear but a relatively low CI at 29.31 considering the 1952 assemblage; this is 6.5 × less than the unit 2 excavation in 2019.

By comparing the 1952 and 2019 reconstructed fish weights for Polydactylus sexfilis, Naso and Bodianus from Mo1, it was determined that larger fish bones, and therefore fish of reconstructed larger sizes, were preferentially collected from the earlier excavations (Figure 10). Despite this collection bias, the reconstructed weights of the larger fish inform on targeted marine habitats, inferred capture techniques and possible over-fishing of some species. Spawning times of sexually mature fish, whose bones are found in archaeological sites, can also inform on season(s) of site occupation.

Large carangids, known in Hawai‘i as ulua, are highly sought after and it is common to see tails of large individuals nailed to the sides of garages and sheds on Moloka‘i. There are biannual tournaments, attracting several thousand dollars in prize money, for catching the largest fish. Recently the largest two ulua caught were ~23 and 21 kg (personal communication, Walter Mendes). Serious fishers have their secret spots around the island. Ulua were the most common large fish whose bones were recovered from Mo1, 2, 6, 7, and 8; that is, from the north, west and south coasts. Eight bones were identified from the giant trevally (ulua aukea, Caranx ignobilis), known locally as white ulua. These larger fish are primarily bottom feeders (Rizzuto, 1978, p. 50) that can be caught on baited incurved (rotating) hooks (see examples in Emory et al., 1968) from land where deep water is close to shore as along the north coast of the island. The archaeological bones were from fish ranging from 5 to 20 kg (13.2 ± 4.9 kg). A heavy stone plummet line sinker (pōhākialoa, lit, long stone; Hiroa, 1964: Figure 235d), used for deep water fishing, was surface collected near the house site at Mo8 in 1952. Here, three C. ignobilis bones which inhabit depths to 350 m (Mundy, 2005, p. 367) and can weigh up to 87 kg (Randall, 2007, p. 229), had reconstructed fish weights of 20 kg, and one caudal vertebra was the largest the senior author had ever seen, weighing 10.1 g. Eight bones of C. lugubris or black trevally (ulua lā‘uli) from Mo1, 2, and 6, ranged in reconstructed fish weight from 2 to 10 kg. Eleven bones, from sites Mo1, 2, 7, and 8, could only be assigned to Caranx but included individuals with reconstructed weights from approximately 2–10 kg. Ulua lā‘uli reaches nearly 18 kg and 80 cm long and it spawns from March to June when it comes close to shore in large numbers (Hosaka, 1973, p. 113). It is interesting to note that the species is “not common in the Hawaiian Islands and appears to be restricted to the Northwestern Hawaiian Islands” (Randall, 2007, p. 229). This was probably not the case several centuries ago.

Snappers (Lutjanidae) are commonly taken on baited hooks in deep water, often greater than 100 m (Tinker, 1978, p. 221), while handlining from a canoe. The only snapper identified to species from the 1952 assemblages is the uku or green jobfish (Aprion virescens) that inhabits the rocky bottom off the reef or deep places near shore (Hosaka, 1973, p. 123) where it is more abundant in June (Rizzuto, 1978, p. 107). The Hawaii state record is 17 kg (Randall, 2007, p. 248), while the reconstructed fish weight of 15 archaeological bones ranges from 2.75 to 6.0 kg (4,250 ± 1,053 g). All but one of the bones was from Mo1 where deep water is close to shore along the north coast.

The unicornfish (kala, Naso unicornis) is one of the largest surgeonfish (Acanthuridae) and inhabits inshore reefs, often in large numbers, to graze on seaweed (limu), its principal food. The association of a particular reef plant with unicornfish is so strong that the seaweed Sargassum echinocarpum is named limu kala where it is found along the low rocky intertidal zones with appreciable wave action generated from northern winter storms (Huisman et al., 2007, p. 236), such as immediately west of Mo1. Entering surprisingly shallow water in its quest to eat leafy seaweed (Randall, 2007, p. 439), large kala can be easily surrounded by seine net or speared (Hosaka, 1973, p. 142). The fish is oily and considered excellent eating when dried. When roasted, the large intestine, filled with edible seaweed, is eaten today and called “limu sausage” by locals. It was likely important in traditional Hawaiian diets as limu contributed vitamins and essential minerals not found in other staple foods such as fish and taro (McDermid and Stuercke, 2003, p. 513). Indeed, it is a unique nutritive contribution of Naso unicornis. The bones of kala, with reconstructed weights of 2.5–3 kg—considered large individuals (Titcomb, 1972, p. 85)—were recovered from Mo1, 2, 3 (only one bone), and 6 where ideal habitat was just seaward of the sites.

Of the four species of Bodianus wrasses in the Hawaiian Islands, the large fish in Bonk's sites are most certainly the endemic B. albotaeniatus (‘a‘awa, Hawaiian hogfish) or B. bathycapros (Hawaiian pigfish) as the other two taxa are much smaller. B. bathycapros does not appear to have a specific Hawaiian name (other than hinālea, a general name referring to all wrasses) and is restricted to deep water, greater than 165 m (Parenti and Randall, 2011, p. 31) which likely makes it uncommon in archaeological sites, while B. albotaeniatus is more accessible living at 3–200 m (Mundy, 2005, p. 432) where it can be caught from a line cast from shore (Hosaka, 1973, p. 147) or a baited hook and line dropped from a canoe in deeper water. Nine bones representing individuals weighing 2.0–5.0 kg, were recovered mostly from Mo1, aside from one bone from Mo5. These weights exceed some of the average values reported (e.g., Hosaka, 1973, p. 147) which might be because adults are no longer common in the main Hawaiian Islands (Randall, 2007, p. 321). This may be a case of post-contact hunting pressure.

A relatively small family, parrotfish (Scaridae) often make up a large portion of archaeological fish bone assemblages from tropical Pacific Islands (Bouffandeau et al., 2018; Butler, 1994; Ono and Clark, 2012; Weisler and Green, 2013; Weisler et al., 2024) where they form a dominant part of the herbivorous fish community (Bellwood, 1994, p. 3). The heaviest reconstructed weights from fish bones at sites Mo1, 2, 5, 7, and 8 ranged from 2.0 to 5.0 kg consisting of two genera. One genus is Chlorurus cf. perspicillatus (Spectacled parrotfish, uhu ‘ahu‘ula or uhu uliuli depending on sex and growth stage) and C. cf. spilurus (Pacific bullethead parrotfish, uhu). Since C. perspicillatus is the heaviest and most common of all Hawaiian Islands parrotfish at about 6.5 kg (Gosline and Brock, 1976, p. 238; Randall, 2007, p. 361), some of the archaeological bones are likely from this species. Parrotfish are closely associated with coral reefs and are commonly found in shallow water where they can be caught with a net or speared along the outer reef slope down to 10+ m (Hosaka, 1973, p. 152–153). It is seldom taken with a hook baited with ‘a‘ama crab, shellfish or seaweed, the latter observed by Weisler in Kona, Hawai‘i Island in the 1970s. The other genus is Scarus where some archaeological bones might be from S. rubroviolaceus (Ember parrotfish, uhu pālukaluka or uhu ‘ele‘ele depending on sex and growth stage) due to reconstructed weight of 5.0 kg close to the upper size range of 5.7 kg (Randall, 2007, p. 365). Although different species of parrotfish reach sexual maturity at different sizes (DeMartini and Howard, 2016), fish from these two genera were of reproductive age with C. perspicillatus and S. spilurus reaching peak spawning March to April and S. rubroviolaceus during September to November (DeMartini and Howard, 2016, p. 6). Of the 100 scarids with reconstructed weight data from all sites excavated by Bonk, 45 were of fish that exceeded 2 kg and likely of sexual maturity regardless of species.

Sea chubs or rudderfishes (Kyphosidae) consist of four genera and about 10 species with Kyphosus sandwicensis (nenue) the most common species in the Hawaiian Islands (Randall, 2007, p. 271). Attaining 5.5 kg, these primarily herbivorous fish can form large schools while grazing on benthic algae (such as Sargassum, also the common food of Naso) over rocky substrates found along Kalani where they can be surrounded by seine net; occasionally they can be caught with ‘a‘ama crab baited hooks (Hosaka, 1973, p. 126). Of the 83 bones identified at Mo1, 2, 5, 6, and 8, 61 had reconstructed weights between 250 g and 3.0 kg—86.9% were from Mo2 with reconstructed weights from 2.0 to 3.0 kg—the heaviest of any site. The occupants of Mo2 were targeting the oldest (heaviest) fish, but in a study of the reproductive status of K. cinerascens (similar in size, form and habits of K. sandwicensis) and other reef fish, the smaller fish (not the largest) are responsible for most of the egg production (Longenecker et al., 2017, p. 107). Therefore, the Kyphosidae fishery may well have been sustainable.

While it is not always possible to excavate a site where substantial legacy collections were obtained, the renewed excavations at Mo1 in 2019 provided key insights into the 1952 Mo1 excavations. These included: (1) the collection of bones from smaller fish to enable a fuller understanding of fishing sustainability, especially for moi (Polydactylus sexfilis); (2) the identification of ‘iao (Hawaiian silverside, Atherinomorus insularum)—a new species record for the Pacific Islands; (3) markedly increased species richness for some taxa, such as the filefish (‘o‘ili uwi‘uwi, Monacanthidae Pervagor spilosoma); (4) documentation of a greater diversity of fish families (21–24); (5) significant increase in richness within families (Kuhliidae, Kyphosidae, Labridae, Monacanthidae, Polynemidae, Exocoetidae, and Pomacentridae) and sharks and rays (Elasmobranchii); and (6) identification of more taxa to genus (11–19) and to species (5–15) levels. Not to mention, of course, the opportunity to redraw the stratigraphic sections and collect dating samples of identified short-lived materials, amongst others.

It is no surprise that the marine environments adjacent to archaeological sites constrain the kinds of fish represented in pre-contact habitations. Although there is a broad range of reef associated taxa that inhabit the inshore waters of coastal west Moloka‘i, nearly 71% of the 3,764 identified fish bones and otoliths (scales not included here) are from four families: threadfins (Polynemidae, 21.33%), surgeonfishes (Acanthuridae, 20.75%), wrasses (Labridae, 18.78%) and filefishes (Monacanthidae, 18.78%; Figure 13). To local residents, the shoreline between Kawa‘aloa bay and along Kalani to Mo2 are “moi grounds” where threadfins are netted and moili‘i (small moi) were, based on archaeological evidence, scooped from tide pools (Figure 12)—a harvesting pattern that goes back perhaps 600 years. The herbivorous surgeonfishes are netted and speared and the southern fringing reef seaward of Mo6 is ideal habitat, especially for kala (Naso spp.). The rocky point immediately west of Kawa‘aloa bay and the adjacent flats are covered with limu kala, a seaweed that attracts unicornfish. The third most common family, wrasses comprise more than 15% NISP at sites Mo1, 5, and 8 which are near rocky shorelines where baited hooks would be an effective technique for capture. The fourth ranked family is the filefishes and the high percentage of these fish bones in Mo1 (mostly retained in the 3.2 mm sieves during the 2019 renewed excavations) is interesting when considering the life cycle and ethnohistoric record for the omnivorous filefish (‘o‘ili uwi‘uwi, Monacanthidae Pervagor spilosoma). Inhabiting the sea floor at depths of 6–138 m (Mundy, 2005, p. 530), Pervagor is about 10 cm long with rough skin and a distinctive dorsal spine flanked by needle-like projections (Figure 8B). These fish contain little flesh that is said to taste bitter (Tinker, 1944, p. 357) but they were sometimes eaten. Perhaps associated with unusually cooler ocean temperatures in a particular year (Randall, 2007, p. 481) during the northern winter months (Tinker, 1978, p. 480), the fish is occasionally blown ashore dead in the millions (hence, ‘o‘ili, lit. to appear) and gathered for fuel (Titcomb, 1972, p. 119), yet none of the archaeological bones were burnt. It is likely that at these rare times, ‘o‘ili uwi‘uwi were collected at the beach above the high tide line at Kawa‘aloa. Some 97% of filefish, considering all sites, were from Mo1, a stone's throw from the beach.

Aside from these four highest-ranked fish families, the numbers of the jacks or trevally (Carangidae, NISP = 153, 4.1%) bely its importance to pre-contact subsistence as bones representing fish with reconstructed weights greater than 10 kg were found at sites Mo1, 2, 6, 7, and 8 adding appreciably to the overall assemblage live fish weight at these sites. Today, large ulua are often shared amongst extended families and neighbors which may reflect an age-old practice. Bones of the deeper water snapper, uku (Aprion virescens) were recovered at Mo1 and 6 which points to hand lining with baited hooks past the reef slope on the north and south shores.

With coral reefs near all sites, there were surprisingly few parrotfish (uhu, Scaridae) since scarid is the dominant family at most sites in tropical Polynesia. On west Moloka‘i, only 4.3% of all NISP comprised parrotfish.

As the statistical analysis has demonstrated, Mo1 (including unit 2) was the most species rich and diverse site which accounts for the presence of eight families found at no other sites. These included the flagtails (Kuhliidae), flyingfish (Exocoetidae), coronetfishes (Fistulariidae), butterflyfishes (Chaetodontidae), silversides (Atherinidae) only identified from an otolith (Figure 9A), goatfishes (Mullidae), boxfishes or cowfishes (Ostraciidae) and bigeyes (Priacanthidae). Flyingfish are often captured well offshore with multiple canoes (Kāhā‘ulelio, 2006, p. 89) implying planned and well-coordinated fishing sorties. Significantly, 12 of the 15 elements identified to this taxon were otoliths (Figure 9H).

Considering all sites, net fishing in predominantly shallow water was the common strategy for capturing a range of inshore fish species (and deep offshore waters when foraging for flyingfish). Handlining with baited hooks dropped from a canoe in deep water is indicated for snappers (Lutjanidae) and for probably some of the heavier carangids even though they can be fished from shore where deep water is close to land. Very small fish, especially moili‘i (threadfins) could be collected using fine mesh scoop-type nets in the tidal pools west of Kalani and just seaward of Mo2 (Figure 12).