Building ecological networks with local ecological knowledge in hyper-diverse and logistically challenging ecosystems
Handling Editor: Robert B. O'Hara
Abstract
- Collecting interaction data to build frugivory or seed dispersal networks is logistically challenging in ecosystems that have very high plant and animal diversity and/or where fieldwork is difficult or dangerous. Consequently, the majority of available networks are from ecosystems with low species diversity or they represent a subset of the community.
- Here, we propose an approach applying local ecological knowledge (LEK) of indigenous communities to build quantitative interaction databases and networks that would otherwise be difficult to achieve with direct observations. Indigenous communities live in many hyper-diverse ecosystems and the people within these communities often have detailed knowledge of ecological processes.
- Working in a Sundaland biodiversity hotspot—Royal Belum State Park, Peninsular Malaysia—we used visually oriented interviews with indigenous people (Orang Asli, in the Jahai and Temiar ethnic subgroups), field data and published records to collate interactions, and their estimated frequency of occurrence, of animal fruit consumption and seed dispersal.
- We documented 2,063 fruit consumption and 1,360 seed dispersal interactions among 164 plant species and 34 animal taxa, the latter representing groups of closely related species or individual species. The majority of the interactions (97%) were identified by the LEK interviews, with the additional methods (field data and published records) used to support and marginally expand the interview data. The metrics for the networks we built reflect those of networks structured by biological mechanisms, supporting the validity of our novel approach.
- LEK is highly relevant for building detailed databases for ecological interactions in hyper-diverse and/or challenging ecosystems. Such ecosystems are among the most vulnerable on earth, harbouring ecological interactions that are often poorly documented at a community level. We show how LEK can broaden our knowledge of such sensitive ecosystems, but our approach is useful for any ecosystem in which people hold rich LEK.
1 INTRODUCTION
Ecological interaction networks are an invaluable tool for understanding community structure, evolutionary processes and vulnerabilities within ecosystems (Guimarães et al., 2011; Thébault & Fontaine, 2010). Mutualistic networks depicting interactions between plants and their seed dispersers have been conducted across diverse ecosystems (Donatti et al., 2011; Heleno et al., 2013; Timóteo et al., 2018), to, for example, identify species that are important in maintaining communities (Cagua et al., 2019), investigate how defaunation or invasion influences network structure (Fricke et al., 2018) and even to predict the seed dispersal roles of prehistoric animal assemblages (Pires et al., 2018). A major limitation for compiling networks is the requirement for broad and detailed datasets. Consequently, a majority of studies depicting mutualistic seed dispersal interactions are from relatively simple systems, such as islands where species diversity tends to be low (González-Castro et al., 2012; Heleno et al., 2013), or for a portion of the broader community (Mello et al., 2011). In hyper-diverse regions, building comprehensive network datasets can take many years (Donatti et al., 2011), and even then, certain animal species or even broad taxa might be missed, which could bias our interpretations of network structure and lead to important species and interactions being overlooked (Vidal et al., 2013).
Network studies focused on frugivory and seed dispersal commonly use frugivory censuses, camera trapping and faecal contents to build the interaction database (Donatti et al., 2011; Timóteo et al., 2018). These, generally achievable, methods can become infeasible without years of study at high costs, particularly for networks that encompass diverse frugivore guilds that require multiple sampling strategies. In these systems, many plants and animals occur at low density (Wright, 2002), requiring vast spatial scales and a massive effort to document interactions, let alone to document whether these interactions translate to effective function (Howe, 2016). The task becomes even more immense if the plants have supra-annual phenologies. Ecosystems with safety risks also constrain sampling. Frugivory censuses and nocturnal observations cannot always be conducted safely in places where megaherbivores and large predators are common. Loss of cameras to elephants and humans is another major problem, even when locks and heavy-duty protective equipment are used to protect the cameras (Meek et al., 2018).
The ecosystems that are difficult to study using standard ecological methods for building interaction databases can be among the most vulnerable and complex on earth; hence, their study is a high priority. Some indigenous communities are well-versed in the natural history of the hyper-diverse ecosystems they inhabit, having a deep knowledge of plants, animals and their interactions (e.g. Cámara-Leret et al., 2019). This knowledge has been accumulated through personal observations required for hunting and as life skills, or from ‘diachronic observations’ that are accumulated over generations from the elders and broader community (Gadgil et al., 1993). Scientists have gained insights from traditional knowledge for a wide range of biological disciplines such as systematics, ecology and community biology (Drew, 2005). This local ecological knowledge (LEK) can also be of great value to understand interaction networks.
Here, we show how LEK of indigenous communities can be applied, in conjunction with other methods, to build an interaction database in ecosystems where the use of ecological methods alone is not feasible, or the cost or time required are prohibitive. To do this, we collected data on frugivory and seed dispersal interactions for canopy and sub-canopy plants in a hyper-diverse Sundaic forest in Peninsular Malaysia.
2 MATERIALS AND METHODS
2.1 Study area
We conducted our study in Royal Belum State Park (hereafter Belum), part of the Belum-Temengor Forest Complex (BTFC), in northern Peninsular Malaysia (Appendix S1). BTFC (5°30′N, 101°20′E) is a Sundaic rainforest of c. 3,500 km2 that forms a transboundary forest complex with Southern Thailand. Belum occupies 1,175 km2 of primary and protected forest with an altitude range of 130–2,160 m a.s.l., a mean daily temperature of 24.3°C (range: 20.8–33.5°C), and humidity between 70% and 98%. Belum's vegetation is mainly hill dipterocarp (72%) mixed with lowland dipterocarp, upper dipterocarp and montane forests. Trees from the family Dipterocarpaceae dominate the forests, producing fruit in masts of 2 to 10-year intervals (Ashton et al., 1988). Many other plant species fruit supra-annually, either independently or in conjunction with the community mast. Belum is rich in wildlife, with over 300 bird species and a complex assemblage of terrestrial mammals, including elephants Elephas maximus, tigers Panthera tigris, gaurs Bos gaurus, tapirs Tapirus indicus and sun bears Helarctos malayanus. Sundaic forests have low densities of frugivores, a consequence of the few food resources provided by dipterocarps and unpredictable fruit availability, compared with other tropical rainforests (van Schaik et al., 1993). BTFC has a large man-made lake (Appendix S1) that flooded 180 km2 of lowland forest in the 1970s. For more details on the study area see Lim et al., 2019.
2.2 Indigenous communities in BTFC
The Orang Asli (meaning ‘first’ or ‘original’ people) are the indigenous people and oldest inhabitants of Peninsular Malaysia, arriving in the region at least 50,000 years ago (Lim et al., 2019). Approximately 6,800 Orang Asli people live in BTFC, mostly from the Jahai (sub-ethnic of Negrito) and Temiar (sub-ethnic of Senoi) groups. Traditionally, the Jahai are hunter-gatherers and the Temiar are swidden farmers (Lim et al., 2019). At present, many Jahai people live within Belum's forests, where they fish, hunt and gather for food (Loke et al., 2020). Most Temiar people are small-scale farmers who live in Temengor, the Southern part of BTFC where forests are selectively logged. The Temiar occasionally hunt, and gather products such as honey from the forest.
2.3 Overview of methods
We used a three-step approach to collect plant–animal interactions data (Figure 1). In summary, we first conducted field surveys to catalogue locally available fruit and record interactions with their consumers. Second, we conducted a series of interviews with Orang Asli from different villages, using a fruit photographic catalogue and other visual aids associated with potential animal consumers, to build the interaction database (summary and sample questions shown in Appendix S2). Third, we used published literature and other means to support and expand the interactions recorded in the interviews (Appendices S3 & S4).
Our ultimate goal was to generate two interaction matrices between plant taxa that fruited during our study period and the animals that (a) consume their fruit—a consumption matrix, and (b) disperse their seeds—a seed dispersal matrix. The consumption matrix includes animals that consume any part of the fruit, including seed predators. The seed dispersal matrix includes animals that defecate, regurgitate, spit or hoard seeds, provided this can sometimes result in successful dispersal.
2.3.1 Step 1: Compiling the fruit photographic catalogue and recording interactions directly
First, we compiled fruit collected from Belum to form the base of the interaction database. We systematically recorded fruit that fell on transects at monthly intervals for 16 months (August 2016 to November 2017), including a mast-fruiting season. We excluded plant species known to occur in Belum but whose fruit were not encountered on the transects. We focused on canopy and sub-canopy plants because the understorey of dipterocarp forests has few fruiting plants (Corlett, 2007) and these plants were less likely to drop fruit on the transects. The six transects had a mean length of 1.1 ± 0.2 km (SD), and spanned a total distance of 6.7 km (Appendix S1). Transects were 1.5 m wide in total, the maximum width we could efficiently sample. Trail preparation was done in July 2016 by removing old fruit. The transects were positioned to run inland from the lake to avoid edge effects; steep slopes, where fruit could easily roll off, were also avoided. The spread of transects was designed as a compromise between sampling efficiency and being able to capture fruiting at a scale appropriate to elephants, the largest animal in the network.
When we found fruit along the transect, we searched for the fruiting plant (up to 15 m from the transect); trees (diameter at breast height >10 cm) and lianas (stem size >2 cm dbh) were marked with a tag. We recorded fruit of both zoochoric and non-zoochoric plants because the latter are food to some seed predators. Temiar authors and field team members (PbP, CMTbT, HSD and RbA) provided Orang Asli (Temiar) names for all species collected. We could not locate and identify 11 individual plants (from a total of 175 species and 613 individuals) and these were excluded from further analyses. We later identified most plants at genus level, allowing us to match interactions using published literature (step 3); Temiar names were retained for 21 unidentified plants (13% of 164 species in total; Appendix S5).
Fruit collected from these transects were collated in a photographic catalogue (Appendix S6). When possible (n = 164 species), we photographed different plant features (trunk, crown, leaves, intact fruit and seeds, cross-sections of fruit and seeds) and collected leaves for preparation as herbarium samples to aid identification. We recorded fruit and seed measurements (weight, length, width, breadth and seed number), fruit type (four categories: indehiscent dry, indehiscent fleshy, dehiscent dry and dehiscent fleshy fruit) and other fruit traits (e.g. hardness of fruit and seed). These data are not shown in this paper.
Fruit–animal interaction data collected in the field
During the field transects, we also collated fruit–animal interactions from limited observations of fruit feeding signs, camera traps and faecal contents, to complement the interview data (step 2).
For each fruiting source, we recorded fruit feeding signs within quadrats positioned along the transect and under the parent plant. Quadrats were usually 1 m × 1 m, but larger fruit (>4 cm diameter) were sampled in 1 m × 2–4 m sized quadrats, depending on fruit size. Feeding on fruit and seeds was identified by signs left from the teeth impressions according to LEK. Teeth marks identified were assigned as monkeys (sometimes langurs or macaques could be specified), gibbons, squirrels, porcupines, rats, bats and deer (Appendix S7).
Camera traps (model: Trophy Cam HD Bushnell) were set monthly under a selection of fruiting plants, favouring larger fruit potentially important to terrestrial consumers, which rarely left feeding signs (see above). We arranged fruit (usually 10) in a grid pattern within the camera's range, which was set to record 60s videos and left in place for 1 month. We had 20 camera units available of which 10 were removed (most likely by poachers; Clements et al., 2010), and two destroyed by elephants over the study period (despite being secured with a cable lock and a heavy-duty metal box). Altogether, we retrieved video data from 61 camera trap set-ups on 35 plant species.
Finally, we searched for seeds and fruit remains in animal faeces along the transects, but few data were obtained by this method. Only identified seeds were included in the interaction dataset (mostly from elephant dung). The unidentified—and excluded—seeds were mainly from rarely found faeces of civets, gaur, tapir, sun bear and wild boar Sus scrofa.
2.3.2 Step 2: LEK interviews to build the interaction database
We identified (a) fruit or seed consumers for our plant collection and (b) seed handling by each consumer, through interviews with Orang Asli from six villages (N = 30 participants). Using wildlife guidebooks, we finalised a list of 34 frugivorous animal taxa; these include single species or groups of related species with published evidence of a shared dispersal niche (e.g. hornbills), and/or animals difficult to distinguish at the species level (e.g. rats, bats; Appendices S8 & S9).
In total, our interviews included: (a) 12 pairs of participants from six villages and (b) three single-person interviews with individuals particularly familiar with plant–animal interactions (Appendix S10), two of whom were paraecologists in our team. Paired interviews were animal targeted, whereby we asked participants to identify the fruit consumed (and seed handling) by each of the 34 animal consumer taxa. We encouraged discussions between participants to reach decisions. The single-person interviews were meticulously conducted from the plant perspective (i.e. participants were asked to identify animal consumers for each of the 164 fruit species), and therefore were more inclusive of rare interactions.
Interviews were visually oriented and did not require lengthy explanations. Each interview was conducted by two interviewers in the Malay and Temiar languages. We presented an image of the focal animal(s) and requested the participants to select plants eaten by the animal(s) from the photographic catalogue (Appendix S2). The catalogue consisted of photographs of fruit and seeds—ordered according to fruit size—as well as other available plant parts of the 164 species in our dataset (Appendices S5 & S6).
We collected information on how animals handled the seeds of each consumed plant using a diagram representing seven forms of seed handling (Figure 2): seeds found (a) intact, (b) partially intact, or (c) fragmented in faeces, (d) seeds regurgitated (by ruminants), (e) seeds (not pulp) eaten and hoarded, (f) seeds eaten but not hoarded and (g) pulp eaten and seeds carried away from the parent plant (spat or regurgitated without entering the animal's stomach). Importantly, we requested participants not to make guesses.
We forfeited the answers of one pair of participants, as one respondent gave hasty replies and assigned many more interactions than other participants raising doubts with the accuracy of this answer set. Only seven interviews provided comprehensive information on interactions involving less common birds and elusive mammals; to avoid their under-representation in the network we restricted the interviews used for all taxa to include only the most detailed seven interviews specific to each taxon.
2.3.3 Step 3: Building an interaction database based on interview data
We adopted a conservative strategy by curating our dataset to buffer inaccuracies that could arise from problems with communication due to cultural differences or possible misidentification of interactions. The process of curation involved a number of criteria (Table 1; Appendix S3) that plant–animal interactions could meet in order to be included in our more conservative curated networks. First, we accepted interactions that were identified in at least two LEK interviews. Second, we accepted interactions identified in one interview and an additional data source with no clear seed dispersal information (i.e. teeth marks on fruit, consumption recorded by camera traps or publication record). Third, we accepted interactions where no LEK is available but is supported with clear evidence of seed dispersal from the non-LEK data sources. Interactions identified from the published record were collated from empirical studies (Appendix S4) on frugivory, seed predation or seed dispersal of plant species within the same genera as those in our fruit sample and the animal taxa in our network.
| Method | Source of primary data | Supporting evidence | No. plant species | No. interactions |
|---|---|---|---|---|
| 1 |
LEK interviews (≥2 groups) |
None required | 162 | 1,755 |
| 2i |
LEK interview (1 group) |
Field survey—feeding signs | 27 | 29 |
| 2ii | Field survey—cameras | 6 | 7 | |
| 2iii | Publications | 78 | 213 | |
| 3i |
Without LEK interview |
Field survey—feeding signs | 7 | 8 |
| 3ii | Field survey—cameras | 2 | 3 | |
| 3iii | Publications | 34 | 53 |
For the seed dispersal network, we required seed handling information specific to each plant–animal interaction, since a frugivore can handle seeds in different ways. Such details can be difficult to observe, and may be easily mistaken in short-term observations (e.g. fruit consumed assumed to be dispersed). The information to build this dataset came from LEK, field signs and published records. The capacity of frugivores to ‘defecate’, ‘regurgitate’, ‘spit’ and ‘carry’ seeds is delimited by seed size (e.g. Albert et al., 2013; Kitamura, 2011), and we used published seed size limits to delineate various seed dispersal interactions (12 endozoochory and 12 synzoochory limits; Appendix S3). When unavailable, we used the limits identified by LEK (largest drupe fruit or largest seed indicated as swallowed by the animal by ‘at least two LEK groups’). We prioritised empirical, published size limits and records over LEK accounts because published limits are not constrained by plants within our sample, and may serve to better represent an upper limit for dispersal.
We adopted a ‘hierarchy of quality of evidence’, whereby we considered LEK interviews as ‘expert opinions’ and published studies as ‘case reports’, a similar but slightly stronger form of evidence (e.g. Sprague et al., 2008). In case of conflict, we gave preference to information from the published literature. On one occasion, we consulted academic experts, primatologists, to assign dispersal modes to specific interactions.
2.4 Weighting the network
Network weights are important to identify modularity, the strength of interactions and the importance of highly interactive species (Newman & Girvan, 2004). We used the number of interviews in which an interaction was reported (from seven in total) as weights for consumption interactions. For example, if four interviews reported an interaction, we assigned it a weight of 4. Seed dispersal interactions were assigned factors representing seed handling quality (Appendix S3): seeds dispersed through hoarding (0.25), seeds spat by macaques or carried by bats (0.5), seeds carried and regurgitated by birds and deer (1) and seeds swallowed and defecated by all animals (1). The final seed dispersal weights assigned reflect values of seed dispersal effectiveness, measured as consumption frequency × seed handling quality (Schupp et al., 2010).
2.5 Data analysis
We carried out network analysis with the bipartite package (Dormann et al., 2016), using R statistical environment 3.5.3 (R Core Team, 2019). We computed modules using the Newman and Girvan (2004) modularity measure. As the assignment of network modules is based on an optimisation process, we computed them 50 times (Donatti et al., 2011), and the composition with the highest modularity was selected. We present indices that are commonly used to present a network overview: complementary specialisation (H2′), weighted connectance, and weighted NODF (a metric of nestedness; e.g. Donatti et al., 2011; Heleno et al., 2013; Mello et al., 2011). These indices were compared to the Patefield null model with randomised interactions (Dormann et al., 2016), to assess the likelihood of biological mechanisms determining network structure (Bascompte & Jordano, 2014).
To evaluate the detail lost through the curation process (e.g. by excluding interactions identified by just one LEK interview), we compared the properties of our curated network to those of a ‘liberal’ network, which included all interactions identified by LEK, hereafter, referred as the ‘liberal networks’.
2.6 Ethics statement
This study complied with the ethics requirements of the Science & Engineering Research Ethics Committee, University of Nottingham Malaysia (permits #LO081016 and #LO200218). Research permits were kindly granted by the Malaysian Department of Orang Asli Development (JAKOA; permits #JAKOA/PP.30.052JId13(10)), Peninsular Malaysia's Department of Wildlife and National Parks (DWNP; permit #JPHL%TN(IP): 80-4/2) and Perak's State Parks Corporation (one permit for each trip). Prior and informed consent was obtained from village heads and all participants in this study, after full disclosure of the goals and planned use of the research data. The local Orang Asli community were not involved in the design of the study goals, although some members of the community are part of the research team and contributed significantly to the design of the study methodology (see details in Author contributions).
3 RESULTS
We recorded a total of 164 plant morphospecies fruiting in the 6.7 km of transects over 16 months. We sampled a total of 613 individual plants, including 526 trees, 83 lianas and 4 palms. They represent fruit from at least 40 families and 95 genera, of which, 21 species were unidentifiable (Appendix S5). Of the 164 plant species, we identified 163 (99%) as having fruit consumed by animals, and 160 had seeds dispersed by animals (Table 1).
Between the 34 animal taxa and 164 plant species, we identified 2,063 consumption interactions, of which 1,360 (66%) led to seed dispersal. Most of these interactions (97% out of 2,063) were identified by the interview participants, with 85% (out of 2,063) based on two or more LEK interviews and the remainder 12% based on one interview and an additional evidence source. Interactions identified only by field data or a publication were few (3%; Table 1). For the 1,755 interactions identified by LEK (≥2 interviews), only 3% (49) were identified in all seven interviews; most were identified by fewer interviews—two interviews: 728 (41%), three: 390 (22%), four: 271 (15%), five: 175 (10%) and six: 142 (8%).
From the field surveys (feeding signs, camera traps and faecal inspection), we obtained evidence of 268 (13% of 2,063) interactions. A total of 983 interactions were collected from published records, of which 61% (595 interactions) provided seed dispersal information, and 39% (388 interactions) provided only frugivory information. Of these, 16% (77 seed dispersal and 78 frugivory) provided interactions not recorded by LEK (Appendix S11). The availability of published data was highly variable, with a range of 0–59 interactions for the plant genera. Several bird and mammal species had no matching consumption records, while a few animals had matching consumption records for more than 40 plant species (gibbons, macaques, squirrels; Appendix S12). Seed dispersal information was more limited, with a range of 0–27 interactions for the plant genera (Appendix S12).
The most dominant frugivores were langurs (147 plant species consumed; sum of weight (SW) = 489), squirrels (156; SW = 470), macaques (130; SW = 370), gibbons (110; SW = 369), rats (132; SW = 361), wild boar (86; SW = 274), binturong (Arctictis binturong; 69; SW = 238) and Asian elephant (82; SW = 217) (Figure 3; Appendices S8 & S13). The most prominent seed dispersers were gibbons (87 plant species dispersed; SW = 296), elephant (81; SW = 215), binturong (61; SW = 204), macaques (69; SW = 190.5), civets (52; SW = 176), sun bear (51; SW = 169), hornbills (64; SW = 147) and Asian fairy-bluebird (42; SW = 137).
In the curated dataset, we excluded 1,050 interactions represented by one LEK interview. The distribution of animals' sum of weights in the curated and the liberal seed dispersal networks did not differ much (Appendix S13).
The frugivory network was made up of three modules (modularity of 0.294; Figure 3) and the seed dispersal network also comprised four modules (modularity of 0.297; Figure 4). In both cases, the results were significantly more nested, more connected and less specialised than randomised interactions. Despite having ~30% fewer interactions in the curated network, both types of network had the same number of modules and similar module compositions, showing that much of the resolution was preserved after the curation (Table 2; Appendix S14).
| Metric | Estimate | Null model (N = 1,000) | ||||
|---|---|---|---|---|---|---|
| Mean | Upper and lower CI | |||||
| Curated | Liberal | Curated | Liberal | Curated | Liberal | |
| Consumption network | ||||||
| Number of modules | 3 (100%) | 3 (100%) | ||||
| Weighted NODF | 39.1 | 38.3 | 14.9 | 20.2 | 13.8–16.3 | 18.8–21.5 |
| Specialisation (H2) | 0.144 | 0.0956 | 0.294 | 0.202 | 0.287–0.3 | 0.198–0.205 |
| Weighted connectance (C) | 0.229 | 0.282 | 0.166 | 0.234 | 0.165–0.168 | 0.233–0.235 |
| Seed dispersal network | ||||||
| Number of modules | 4 (74%), 5 (24%), 3 (2%) | 4 (100%) | ||||
| Weighted NODF | 24.4 | 26.3 | 11.3 | 14.5 | 10.3–12.2 | 13.6–15.7 |
| Specialisation (H2) | 0.238 | 0.198 | 0.444 | 0.372 | 0.432–0.448 | 0.365–0.379 |
| Weighted connectance (C) | 0.149 | 0.178 | 0.103 | 0.132 | 0.101–0.104 | 0.130–0.133 |
4 DISCUSSION
Using a novel approach we identified 2,063 unique interactions between fruit (164 plant species) and their animal consumers (34 taxa) in a diverse rainforest within a 16-month field study. Although this rainforest biodiversity hotspot presented many challenges for using standard ecological methods of data collection, we were able to envision the structure of a hyper-diverse plant and frugivore community, rather than just a subset, as is generally achievable in similarly diverse ecosystems (Carreira et al., 2020; Mello et al., 2011). We were also able to distinguish between mere consumption (frugivory) and seed dispersal interactions. In our study system, LEK was the most informative and data-rich source of information about ecological networks, at least given the technology and timeframe available for our data collection, contributing to describe the highly structured organisation of both networks. Both the frugivory and seed dispersal networks for Belum showed modularity akin to other networks with a high number (>150) of interacting species (Olesen et al., 2007), and were more nested, connected and less specialised than the null models, as expected for networks structured by biological mechanisms (Bascompte & Jordano, 2014).
The challenges presented by our study system are common to many hyper-diverse ecosystems, particularly where fieldwork is challenging or dangerous. The large diversity of plant species available, often at low densities, required intensive sampling—both to ensure sufficient replication and to include enough species to represent the community (Appendix S13). In Southeast Asia's dipterocarp forests, the required sampling effort is increased further because of supra-annual phenologies (Ashton et al., 1988) and low animal densities (van Schaik et al., 1993). Exacerbating these issues is the disruptions to sampling caused by megaherbivores, megacarnivores and poachers. We lost 60% (12 out of 20) of our camera trap units due to elephants, theft and camera malfunctions. Furthermore, nocturnal and even early morning observations were not feasible on a large scale due to the presence of tigers and elephants, which pose a significant threat to people.
Indigenous people inhabit many of the world's biodiversity-rich ecosystems. Their invaluable knowledge can further our academic understanding of plant–animal interactions, potentially enhancing databases achieved through ecological research methods. The knowledge of a single person is a combination of personal and community observations, and interaction observations can occur in multiple forms and over many years (Gadgil et al., 1993). For example, indigenous communities might observe interactions while hunting, or find seeds when preparing the gut of animals for consumption. In comparison, limitations with ecological methods might miss interactions. For example, camera failures are common (Meek et al., 2018); frugivory censuses are very labour intensive, unsuitable for animals adapted to hunters, likely to miss cryptic, irregular and nocturnal feeding, and could result in misperceptions of frugivores as seed dispersers (Howe, 2016); and some animals, such as Asian elephants, feed so infrequently on some plant species (Ong et al., 2019) that multiple years are needed for adequate sampling. LEK can also provide systematic quantitative information of extinct interactions, which cannot be gathered using ecological methods. During the first interviews we conducted, we included the recently extinct (<20 years; Clements et al., 2010) Sumatran rhinoceros Dicerorhinus sumatrensis, but the number of participants able to report on rhino interactions was too small to include in the dataset. Consequently, while our novel approach was conceived for hyper-diverse ecosystems, it could also be useful in other contexts where the local inhabitants have good knowledge of the forest, and of recently extinct fauna (e.g. Hainan Island, Turvey et al., 2019).
We identified several caveats with using local ecological knowledge to generate interaction data. First, differences in communication (e.g. interpretation), culture and personality traits (e.g. confidence level) between people could have introduced biases to the dataset. We omitted one interview dataset because we interpreted the participants as overconfident; they provided many more interactions than other participants, and we could not be certain of the origin of the interactions (i.e. observations or assumptions). It is possible, however, that the participants had actually provided accurate information. Second, knowledge might be biased towards certain animal groups, such as those they use for food (Loke et al., 2020). Some of the Orang Asli people we interviewed only reported interactions for the more common animals, and we consequently limited the interview data to reduce their over-representation. Third, many indigenous people are leaving their traditional lifestyles, and with this change, their traditional ecological knowledge is also being lost. Fourth, to account for some of these limitations, we used a conservative data curation. This decision may have led to the exclusion of some rare interactions which can be important for plants (Carlo & Morales, 2016). Finally, our networks comprised groups of closely related animal species (e.g. hornbills were pulled together in a single group), in many cases, to reduce the length and complexity of the interviews. Species within these groups likely vary in their interactions and relative importance. Here, they are represented by a composite value that could over-represent their individual frugivory and seed dispersal roles, and the final network structure. Overall, we can improve the method of incorporating LEK into scientific methodology by increasing the number of interviews, exploring more literature sources and tapping into reliable unpublished data from different fields.
Most network studies use weighted networks and the data required for this cannot be conventionally generated through interviews. We used the frequency of reported interactions as a substitute for interaction frequency, but we do not know how well this represents actual foraging patterns as interactions might be more conspicuous for other reasons. Nevertheless, a non-weighted network could overestimate the strength of weak interactions, and uniform weights among interactions are not real-world patterns. The importance of using a weighted approach was supported by the consistency in the sum of weights of different animals in both liberal and curated networks (Appendix S13). The ideal weight of an interaction represents the per-capita effect of one species on its partner (Bascompte & Jordano, 2014). Our applied weights might not be fully representative, since they did not account for the quantity of a crop consumed or for animal abundance. These would be partially reflected in the number of interviews an interaction was identified in, but animals rarely encountered by people (such as non-food sources, or cryptic, nocturnal animals) might be under-represented. While some of these limitations are countered by the broad spectrum of observations included (i.e. feeding signs, finding seeds in guts and faecal material), the weighting could be improved in future studies by collecting and integrating information such as animal, fruit or seedling densities to the network (Gleditsch et al. 2017; Howe, 2016), although these would require more intensive field sampling if the information is not readily available.
The novel approach we describe here would need to be modified to fit the specific requirements of other study systems, but some of our techniques are important to retain. Working with people who understand the culture of the indigenous community is important to ensure interviews are conducted sensitively and to enhance the quality of the information collected. For example, we interviewed the Orang Asli in pairs because discussion was important for them to consolidate their observations. Furthermore, visually oriented interviews are more engaging for the participants, particularly when many interactions are being asked about. It is important that the fruit species list is an accurate representation of the plant community. We only included fruit that fell on the transects during the study period, but approaches using a broader range of fruit might be valid, depending on the study question. Although we used a conservative approach, the similarities in structure between our curated and liberal networks suggest that the liberal approach of including all interactions identified by LEK interviews could also be acceptable. To answer questions where rare interactions are important, sampling could be increased to ensure better coverage of rare interactions. For seed dispersal studies, we think it is advisable to retain or refine the delimitation of seed dispersal by seed size for different dispersers (Appendix S3).
Indigenous communities have inhabited a wide range of ecosystems for thousands of years. While indigenous people manage a substantial portion of the world's ecologically important landscapes and represent more than 80% of the cultural diversity, they constitute less than 5% of the world's population (Toledo, 2001). By collating indigenous knowledge on interactions, we can broaden our understanding of these hyper-diverse ecosystems and document their immense knowledge before it might erode (Aswani et al., 2018). Our LEK-based approach could also be used to compile information on other types of interactions and in less diverse habitats. Such methods will allow the collection of network data in often vulnerable ecosystems, enhancing our understanding of community organisation and the underlying mechanisms of networks. Our investigation of networks at a community level in Belum would have been impossible without LEK and the use of the Temiar language. Professionally engaging indigenous people as paraecologists presents a good opportunity to strengthen the identity of native speakers. While working with indigenous communities and their LEK, we need to consider the fair benefits and long-term beneficence of the community, and issues associated with intellectual property rights (Riddell et al., 2017).
The interaction network we present here should not be considered as the final product. We rather see it as the first stage in describing hyper-diverse networks. This basic network can continue to be polished by additional direct observation data. This approach has allowed us to produce the first comprehensive frugivory and seed dispersal networks in a hyper-diverse and megafauna-rich Sundaic forest and is replicable in other highly diverse and sensitive ecosystems.
ACKNOWLEDGEMENTS
Field activities and LO's Ph.D. scholarship were funded by Yayasan Sime Darby (grant M0005.54.04) and the University of Nottingham Malaysia, and subsequent work by the Southeast Asia Biodiversity Research Institute (SEABRI; grant #Y4ZK111B01). We thank Susan Lappan and Lye Tuck-Po for their advice, Praveena Chackrapani and Loo Yen Yi for assistance and feedback, and Apok bin Alok, Adeline Hii Yen Mei, Hii Ning, Violet LokePei Xing, and Lisa Davenport for their time. We are especially grateful to the Orang Asli of Belum-Temengor, from the villages of Kampung (Kg.) Tiang, Kg. Semelor, Kg. Sungai Kejar, Kg. Pulau Tujuh, Kg. Desa Damai, and Kg. Banun for sharing their knowledge and shedding light on the complex plant–animal interactions in Belum-Temengor.
CONFLICT OF INTEREST
The authors declare no conflict of interest.
AUTHORS' CONTRIBUTIONS
L.O., K.R.M. and A.C.-A. conceived the research questions, developed and refined the methodology. Methods from steps 1 and 2 were fine-tuned with the contribution of LEK from P.b.P., C.M.T.b.T., H.S.A/L.D. and R.b.A. (Temiar authors), including the set-up of transects, the identification of animal feeding signs, the search and identification of fruit trees in field surveys, and as participants for multiple preliminary interviews. L.O., V.P.W.L., W.H.T., A.S.-M. and K.R.M. also carried out field surveys. L.O. processed samples and herbariums for the pictorial library. L.O., N.A.b.A., A.S.-M., P.b.P., C.M.T.b.T. and H.S.A/L.D. carried out the LEK interviews. L.O., N.A.b.A. and O.L. collected plant information and searched for published records. L.O. analysed the data. K.R.M., L.O. and A.C.-A. led the writing and corrections of the manuscript.
Open Research
PEER REVIEW
The peer review history for this article is available at https://publons.com/publon/10.1111/2041-210X.13685.
DATA AVAILABILITY STATEMENT
The interaction dataset for fruit consumption is available on Dryad Digital Repository https://doi.org/10.5061/dryad.jm63xsjbh (Ong et al., 2021).
