Landscaping the behavioural ecology of primate stone tool use

Ecology is fundamental to the development, transmission, and perpetuity of primate technology. Previous studies on tool site selection have addressed the relevance of targeted resources and raw materials for tools, but few have considered the broader foraging landscape. In this first landscape-scale study of the ecological contexts of wild chimpanzee (Pan troglodytes verus) tool-use, we investigate the conditions required for nut-cracking to occur and persist over time at discrete locations in Bossou (Guinea). We examine this at three levels: selection, frequency of use, and inactivity. We find that, further to the presence of a nut tree and availability of raw materials, abundance of food-providing trees as well as proximity to nest sites were significant predictors of nut-cracking occurrence. This suggests that the spatial distribution of nut-cracking sites is mediated by the broader behavioural landscape and is influenced by non-extractive foraging of predictable resources, as well as non-foraging activities. Additionally, the number of functional tools was greater at sites with higher frequency of nut-cracking and was negatively correlated with site inactivity. Our findings indicate that the technological landscape of the Bossou chimpanzees shares affinities with the ‘favoured places’ model of hominin site formation and provides new insights for reconstructing ancient patterns of landscape use. Résumé L’écologie est fondamentale pour le développement, la transmission et la pérennité de la technologie des primates. Des études antérieures ont identifié la disponibilité des ressources cibles ainsi que les matières premières pour les outils comme des facteurs influents dans la sélection des emplacements pour les activités technologiques. Cependant, il y a peu d’études qui abordent cette recherche à l’échelle du paysage et du comportement fourrager. Dans cette première étude paysagère sur l’utilisation d’outils par le chimpanzé sauvage (Pan troglodytes verus), nous recherchons les conditions écologiques qui influencent la sélection, l’utilisation et l’inactivité des emplacements utilisés pour le cassage des noix en Bossou, Guinée. Nos résultats montrent qu’en plus de la présence d’un noyer et de la disponibilité des matières premières, l’abondance d’arbres nourriciers ainsi que la proximité des sites de nidification étaient des prédicteurs significatifs de l’occurrence du cassage des noix. Cela suggère que la distribution spatiale des sites de cassage de noix est influencée par le paysage comportemental et est influencée par le fourrage non-extractive de ressources prévisibles, ainsi que par des activités non-fourragers. Nos résultats indiquent que le paysage technologique des chimpanzés de Bossou partage des affinités avec le modèle des « lieux favoris » de la formation des sites hominidés et fournit de nouvelles perspectives pour reconstruire les modes d’utilisation du paysage anciens.

technology. Previous studies on tool site selection have addressed the relevance of targeted 27 resources and raw materials for tools, but few have considered the broader foraging 28 landscape. In this first landscape-scale study of the ecological contexts of wild chimpanzee 29 (Pan troglodytes verus) tool-use, we investigate the conditions required for nut-cracking to 30 occur and persist over time at discrete locations in Bossou (Guinea). We examine this at three 31 levels: selection, frequency of use, and inactivity. We find that, further to the presence of a nut 32 tree and availability of raw materials, abundance of food-providing trees as well as proximity 33 to nest sites were significant predictors of nut-cracking occurrence. This suggests that the 34 spatial distribution of nut-cracking sites is mediated by the broader behavioural landscape and 35 is influenced by non-extractive foraging of predictable resources, as well as non-foraging 36 activities. Additionally, the number of functional tools was greater at sites with higher frequency 37 of nut-cracking and was negatively correlated with site inactivity. Our findings indicate that the 38 technological landscape of the Bossou Reed, 1997), but 176 few studies have addressed empirically how these changes may have affected patterns of 177 landscape use and the distribution of early hominin tool sites (e.g. Rogers et al., 1994). Thus,178 investigating the conditions that influence the cessation of activity at chimpanzee tool sites, 179 can provide important clues as to the factors that lead to their temporary or long-term 180 abandonment. 181 Throughout these steps we assess the effect of ecological parameters that have been 182 found to correlate with nut-cracking activities (nut availability; abundance of raw materials; 183 distance to water) as well as variables that encapsulate two key aspects of chimpanzee activity 184 patterns: non-extractive foraging (abundance of food resources: wild food trees; wild fruit Bossou (7° 39' N, 8° 30' W) is located in the southeast of the Republic of Guinea (West 189 Africa), 6 km from the foothills of Mount Nimba Strict Nature Reserve (Figure 1) (Humle,190 2011a; Yamakoshi and Sugiyama, 1995). The chimpanzee community has been studied 191 continuously in both natural (since 1976) and experimental (outdoor laboratory since 1988) 192 settings (Tetsuro Matsuzawa et al., 2011; Sugiyama and Koman, 1979). Between 1976 and 193 2003 the population size ranged from 18 to 23 individuals (Sugiyama, 2004), but has since 194 declined largely due to a catastrophic flu-like epidemic from which it never recovered (Humle,195  November to February, and a long rainy season extending from March to October (Humle,201 2011a; Yamakoshi, 1998). Nut-cracking occurs year-round, but is most prevalent during peak 202 wet season (June -August) and at the start of the dry season (November -December) when 203 fruit is less abundant (Yamakoshi, 1998). 204 The Bossou forest has an estimated area of 16 km 2 and is intersected by roads ( Figure  205 2; Hockings, Anderson and Matsuzawa, 2006). Within this, the chimpanzees range a core 206  seasons. For each nut-cracking site we established 1-km transects intersecting the site datum 220 at 500 metres. Nut-cracking sites within 100 metres of a pre-established transect were either 221 assigned to that transect or became the mid-point of a new perpendicular transect, to ensure 222 even forest coverage. All transects were oriented N-S or E-W, except for two that were 223 oriented NE-SW and NW-SE due to access difficulties ( Figure 2). 5-metre radius survey 224 quadrats were established at every 100-metres along the transects starting from the midpoint 225 where the nut-cracking site was located. At each quadrat nut-cracking specific and general 226 ecological and vegetation data were collected (further details below). 227 We employed a fully digital method of data collection. Quadrat datums and all data 228 entries (food-providing vegetation, tools, raw materials) were georeferenced using an Arrow 229 application (MGISS, 2019) on an android device where further data could be entered through 232 custom made forms. 233 Oil palms 234 For each oil palm encountered during quadrat surveys we documented diameter at 235 breast height (DBH), and number of fruit bunches (total and ripe). For 25 of the nut trees 236 associated to nut-cracking sites we also collected information on nut availability and new 237 traces of nut-cracking on a weekly basis during the first field season (22JAN18-03MAY18) and 238 once at the beginning and end of the second season (weeks of 29OCT18; 10DEC18). 239 Additional data was collected by Henry Camara during the weeks of 30SEP19, 27APR20, and 240 25MAY20. As per Koops et al. (2013), we scored presence of edible nuts on the ground within 241 a 2-metre radius of the nut tree: (0) nuts absent; (1) 1-50 nuts; (2) 51-100 nuts; (3) > 100 nuts. 242 With aid from field guides, nut suitability was determined by checking a sample of randomly 243 collected nuts for whether the nuts contained an edible kernel or were rotten (following Koops 244 et al., 2013). Nuts were not opened so as not to affect future availability, but the local people 245 also crack oil palm nuts and are able to identify whether or not they are edible (Humle and 246 Matsuzawa, 2004). 247 Tools and raw materials 248 All lithic material was recorded for size, raw material type, and portability (whether loose 249 or imbedded in the ground). Adapted from Koops et al. (2013), size was scored into six 250 categories: (1) 1-2 cm; (2) 3-5 cm; (3) 6-10 cm; (4) 11-20 cm; (5) 21-30 cm; (6) >30 cm. Tools 251 and bi-products of nut-cracking were defined as stones that showed at least one of the 252 following: a) traces of wear from nut-cracking; b) nutshell remains on or around them; c) could 253 be refitted with another stone with evidence of a) or b). For this study, the variable Tools 254 included all lithic materials used for nut-cracking excluding fragments that no longer or could 255 no longer be used for nut-cracking.

12
The collective tool assemblage was also scored for status of nut-cracking activity: 257 (Active) New signs of nut-cracking activity were recorded during the fieldwork period. Nut 258 powder or cracked nut kernels were visible on top of or around tools, and there was at least 259 one hammer and anvil pair with impact points that had not rusted over; (Inactive) There were 260 no signs of recent nut-cracking activity during the entire fieldwork period. Cracked nut kernels 261 were either absent or present but showed clear signs of decay. Iron oxide or moss developing 262 on tool impact points. 263 investigate the effect of five main predictors: raw materials, wild food trees, wild food THV, 304 distance to nearest nest cluster, distance to nearest river, on the presence (1) versus absence (0) of a tool site in a given quadrat. We also analysed three sub-models to determine whether 306 more restricted variables yielded a better model fit (Appendix, Table A2). The first sub-model 307 replaced raw materials with a subset of raw materials of size class corresponding to the three 308 most common tool size classes (95% of tools. Size class: 3, 4, 5; Appendix, Table A1 Tool site use 315 We investigated whether the hypothesized ecological variables (i.e., nut availability, raw 316 materials, food trees, distance to nearest nest cluster, and distance to nearest river) influenced 317 the frequency a nut-cracking site was used. From a total of 361 monitoring observations, only 318 35 cases of recent nut-cracking events were identified for 17 out of the 25 monitored nut-319 cracking sites, where frequency of recent activity ranged between 1 and 4. Because of single 320 (N = 1) sample sizes for 2 and 4 events, frequency of activity was recoded as "Low" (≤ 2 321 events; N = 10) and "High" (> 2 events; N = 7). The small sample size (N = 17) was deemed 322 too small to justify a GL(M)M, therefore we only discuss descriptive statistics for this question 323 using two-sample t-tests (or Man-Whitney U tests when assumptions of normality were not 324 met). 325 Tool site inactivity 326 We used a binomial GLM with 'logit' link to investigate the effect of mean nut availability, 327 raw materials, and food trees, on tool-site inactivity. The response variable included nut-328 cracking sites that were classified as active (response = 0; N = 24) with those classified as 329 inactive (response = 1; N = 16). The final dataset included 40 tool sites. Akin the model for tool site selection, we also investigated four sub-models with raw materials of size class 3-5, 331 tools, and fruit tree subsets (Appendix, Table A6 assess the significance of the full models and sub-models, we ran likelihood ratio tests (LRT) 341 using the 'anova' function which compared each model to a corresponding null model from 342 which all fixed effects were excluded (Dobson, 2002). We tested the significance of main 343 effects for each model by systematically dropping them one at a time and comparing the 344 resulting model with the full model using the 'drop1' function (Dobson, 2002). P-values for the 345 individual effects were based on the LRT results from the 'drop1' function. The AICc for model 346 selection was calculated using the 'MuMIn' package (Bartón, 2020

369
Tool site selection 370 The sub-model where raw materials were replaced by a subset of size class 3-5 was the 371 best fitted model according to the AICc (Appendix , Table A3), and had a clear effect on the 372 probability of a nut-cracking site occurring in a location where at least one oil palm was present 373 (full-null model comparison, LRT: df = 5, deviance = 56.52, p < 0.001). Raw materials had a 374 significant positive effect on tool site prediction, as did food trees, while distance to nest cluster 375 had a significant negative effect (Table 1; Figure 3). All other fixed effects were non-significant 376 (Table 1). The sub-model replacing wild food trees with the fruit trees subset yielded the worst 377 model fit in which fruit trees were not a significant predictor (Appendix, Table A3, Table A4). 378  during the monitoring period (high frequency). In general, mean nut availability was 386 significantly higher at nut-cracking sites that registered a higher frequency of nut-cracking 387 activity (T-test: p = 0.03; Figure 4; Table A5). Furthermore, distance to nearest nest cluster 388 revealed a negative trend, whereby high frequency sites tended to be nearer to nesting 389 locations (Wilcoxon rank-sum test: p = 0.07). For all other variables of interest (raw materials, 390 mean fruit bunches, wild food trees and distance to nearest river) there were no significant 391 differences between the groups (p > 0.15; Figure 4; Table A5). 392 393 Figure 4 -Frequency of tool site use in relative to: a) Raw materials that have been used as tools; b) Distance to nearest nest cluster (km); c) Raw materials of size class 3 to 5; d) Wild trees that are sourced by chimpanzees for food; e) Wild trees that are sourced by chimpanzees for fruit; f) Distance to nearest river (km). Grey circles represent individual points, and means indicated by diamonds. ** p < 0.05; * p < 0.07.

** *
Out of the sub-models, the tool subset model yielded the best fit, although the AICc for 395 the tool and fruit trees model was only marginally higher and produced comparable results 396 (Appendix , Table A7, Table A8). Comparison of the tool subset model with the null model was 397 significant (LRT: df = 3, deviance = 13.20, p < 0.01). Overall, we found that lower values of 398 mean nut availability and a lower number of tools were both significant predictors of tool site 399 inactivity, while wild food trees had no effect ( Figure 6; Table 3). However, the data 400 distributions shown in Figure 6 suggest that the model is not very robust.  From the initial inspection of the data, it is evident that a minimum of one oil palm, 409 specifically an oil palm in close proximity (within 10 metres), is required for nut-cracking to 410 occur in a given location. Further to this, our results show that the abundance of raw materials 411 and food trees as well as proximity to the nearest nest cluster are also important predictors for 412 whether a tool site is established at an oil palm location. This suggests that, in addition to the 413 ecological pre-requisites of nut-cracking, i.e., a producing oil palm and raw materials for tools, 414 other predictable resources that form part of the chimpanzee diet (wild food-providing trees), 415 as well as non-food related activities (sleep sites), are influential in the spatial distribution of 416 nut-cracking locations. In contrast, THV had a negative but non-significant effect on whether 417 tool-sites occurred. While THV is a frequently consumed food item by the chimpanzees of 418 While water was not a significant factor for Bossou, we predict that it could be a major 461 ecological driver regarding the spatial distribution and reuse of tool sites by savannah-living 462 chimpanzees. The Fongoli chimpanzees do not crack nuts, but they engage in termite-fishing, 463 which is also a spatially discrete technological activity tethered to the location of termite 464 mounds Pruetz, 2008, 2011), much like nut-cracking. 465 Nevertheless, climatic or hydrological differences cannot explain the differences 466 between Bossou and Diecké. Given the proximity of both field sites (approx. 50 km) and similar 467 climates it is unlikely that this is due to differences in aridity or water availability. However, the 468 chimpanzees of Diecké crack different nut species, Panda oleosa and Coula edulis, which are 469 absent in Bossou and may be more water dependent than the oil palm. Thus, this discrepancy Distance to nearest nest cluster showed a weak yet noteworthy difference, whereby the 492 frequency of nut-cracking events was marginally greater at tool sites that were closer to nest 493 locations. These results mirror those found for tool-site selection and offer further tentative 494 support that active tool sites and their frequency of use is influenced by their distribution 495 relative to current activity hotspots. 496 The number of wild food and fruit trees was largely the same for all active nut-cracking 497 sites. This suggest that, while food providing trees are good indicators of tool-site selection, 498 they may not good predictors of site use because the data collected did not capture temporal 499 changes in food availability or frequency of foraging activity. On the other hand, nests are 500 temporary features that rarely preserve for longer than six months in non-savannah 501 environments (Ihobe, 2005;Kamgang et al., 2020;Zamma and Makelele, 2012). Therefore, 502 they are a better spatial proxy for recent ranging patterns and possibly explains why 503 differences were found for nests, but not for vegetation.
Some consideration needs to be given as to the low number of weekly traces of nut-505 cracking events recorded per tool site during the 15-weeks of monitoring. This is partially due 506 to the fact that not all active tool sites were monitored, with a further seven traces were found 507 through indirect observations at non-monitored sites. Our data indicates that a minimum of 40 508 nut-cracking events took place at natural nut-cracking sites during the 15 week monitoring 509 period, averaging approximately three events per week, which may be sufficient for the 510 existing chimpanzee population. 511 conditions required for nut-cracking to occur and persist over time in a particular location, and 529 the factors that might lead to their abandonment. Our data suggests that mean nut availability, used as a proxy for tree productivity, and a high abundance of tools are important in 531 maintaining the active status of a nut-cracking site. However, there are clear exceptions that 532 appear to not quite fit the model (Figure 6), suggesting that other factors that were not 533 considered in the analysis may also be at play. 534 The Bossou forest suffers from a great deal of human activity, particularly slash-and-535 burn agriculture, which leads to frequent and rapid changes in the spatial distribution of 536 resources and localized vegetation composition (Hockings, 2011). While oil palms are not cut 537 down during this process and are highly resistant to fire (Yamakoshi, 2011), the changes in 538 the surrounding landscape and the increase in human presence may deter chimpanzees from 539 visiting those areas, especially if they are near the forest boundary. Conversely, cultivated 540 land that contains desirable food items (e.g. banana, mango, papaya) can often attract 541 chimpanzees (Hockings, 2011), and perhaps, under these conditions, the chimpanzees 542 prioritize the prized fruit over nuts that can be found almost anywhere. Site inactivity could 543 also be an artefact of population decline, whereby fewer resources are sufficient to sustain the 544 entire population. Previous literature has suggested that the Bossou forest has a carrying 545 capacity for around 20 chimpanzees (Sugiyama and Fujita, 2011), so it is possible that the 546 current population may no longer need to depend as highly on nuts to supplement their diets. 547 A future longitudinal comparison drawing from historical and contemporary data will help 548 investigate and test this further. 549

550
Our results indicate that proximity to a nut tree, an abundance of raw materials and 551 predictable resources, as well as proximity to a nesting site are important ecological 552 parameters for the establishment of a nut-cracking site in a given location. Distance to nearest 553 nest cluster was also correlated with frequency of nut-cracking, which could potentially indicate 554 that nesting sites are important anchors for ranging and activity patterns. Similarly, tool 555 availability was significantly correlated with tool site use, as well as tool site inactivity, suggests 556 that familiarity of materials used for tools or the visual cues of tool use could be important in 557 the persistence of nut-cracking activities once a site has been established. While there was 558 no significant difference in nut availability among oil palms at active sites, the odds of tool site 559 inactivity were greater when mean nut availability was low, potentially indicating that a decline 560 in oil palm productivity at nut-cracking sites is driver of sites disuse. Together, these results 561 postulate that nut-cracking in Bossou is not only tethered to locations that provide the 562 necessary resources for this activity but is also intimately connected to a broader foraging and 563 behavioural landscape that is mediated by the spatio-temporal availability of primary target 564 resources, such as predictable food-providing trees, as well as the distribution of frequently 565