Orangutans (Pongo pygmaeus) play an essential role in maintaining forest composition and biodiversity by contributing to seed dispersal (Tarszisz et al., 2018a,b). However, they rely on forest architecture and tree connectivity to perform movements in order to find nesting and food trees (Thorpe et al., 2007; Manduell, 2012). This species is considered critically endangered, with their populations in sharp decline due to destruction, degradation, and fragmentation of their habitats, in particular, forest fires (Ancrenaz et al., 2016). Although orangutans can adapt and survive in a fragmented forest, their abundance is significantly lower compared to those in less disturbed and regenerating forests (Seaman et al., 2019). The growing attention directed toward orangutan rehabilitation and reintroduction (Russon, 2009; Sherman et al., 2020; Preuschoft et al., 2021) reflects the high number of displaced orangutan individuals, as well as the importance of this problem for this species' survival.

Understanding and predicting the impacts of forest disturbance on orangutan behavior, health, and population viability is, therefore, of great importance. To this end, extensive field studies have been conducted to understand the interrelationships between orangutans and their habitat. Long-term observation of orangutans' daily behavior in their natural habitat revealed, e.g., that fruit availability changes influence orangutans' activity budgets, diet, and energy balance (Knott, 1998; Morrogh-Bernard et al., 2009; Harrison et al., 2010). Moreover, although orangutans have been observed nesting in burned habitats, their abundance is far lower than in neighboring unburned forests (Rachmawati et al., 2022); peat-fire threatens this species by limiting access to food sources (especially forest fruit), thus presenting challenges for maintaining a healthy energy balance (Ashbury et al., 2022). Orangutans are recognized as arboreal, spending most of their time in the canopy, traveling among the trees using a variety of different techniques (Manduell et al., 2011, 2012). Habitat deterioration could, therefore, impact orangutan populations through increasing terrestrial movement (Ancrenaz et al., 2014), causing emigration to neighboring less disturbed forests (Husson et al., 2015), or—if dispersal is not possible—even to the death of the orangutans.

Among computational modeling approaches, agent-based models (ABMs) are predisposed to assess the interrelations between animal behavior, and the impact of disturbances on habitat quality. The popularity of ABMs owes to their ability to explicitly describe individual organisms, their behavior, interactions, responses to environmental conditions, and many other features influencing the particular fate of organisms in a changing world (McLane et al., 2011; Semeniuk et al., 2011). Furthermore, ABMs are very effective in combining field research with simulation experiments since model descriptions start on the level of individual animals, thus corresponding with field observations (Railsback and Grimm, 2019), and allowing explicit representation of individual features within an animal movement ecology framework (Nathan et al., 2008; Tang and Bennett, 2010).

The use of ABMs to describe animal movement and behavior are manifold. ABMs have been used to explore the impact of habitat fragmentation and disturbances on different species including marine mammal populations (Nabe-Nielsen et al., 2018), tiger populations (Imron et al., 2011), and plant-pollinator networks (Newton et al., 2018). ABMs have also been applied to understand and predict the importance of animal behavior for selected ecosystem functions such as seed dispersal (Bialozyt et al., 2014). In the context of primates, ABMs have been developed to explore the ecology and behavior including emotional dimensions (Evers et al., 2014), as well as orangutans' dispersal and survival chances in a disturbed landscape (Seaman et al., 2021). Seaman et al. (2021) explored orangutan population viability in the function of land use scenarios, including the conversion of natural forests into palm oil plantations with and without forest retention measures. Seaman et al. (2021), however, described animals' movement in a phenomenological way in terms of probabilities to leave, reach or settle in different landscape plots depending on their particular environmental conditions including forest type, disturbance level, or population density.

To the best of our knowledge, an explicit mechanistic description of the orangutan movement decision is still missing. Mechanistic models have the following advantages: (1) they provide us with the ability to understand the behavior of orangutans in environments that are different from the one for which data are available (e.g., forest burnt with different patterns), (2) they allow the incorporation of available expert knowledge about the system and supplementing missing information using data, and (3) they enable the prediction of the impacts of future forest disturbances on orangutans to support population viability and risk analyses, as well as aiding the development of conservation measures. Therefore, the development and deployment of such models could provide important advances in understanding the impact of forest structure, tree connectivity, habitat disturbance, energetics, and mid-term survival in threatened primates such as the orangutan (Thorpe and Crompton, 2006; Thorpe et al., 2007; Manduell et al., 2011; Van Casteren et al., 2012; van Casteren et al., 2013; Morrogh-Bernard et al., 2014).

To address these questions, we propose the BORNEO (arBOReal aNimal movEment mOdel): an agent-based model that simulates orangutan movement behavior through a network of trees within a fine spatio-temporal scale (1-m and 1-s granularity) with a corresponding scale of field observational data (10-cm and 5-min). In this model, the orangutans' choice of locomotion modes (sway, brachiation, climb, descent, and walk) strongly depends on tree connectivity, energy expenditure, food availability, as well as the availability of suitable nesting trees. We exploit field data including (i) habitat surveys to reconstruct an adequate forest structure and (ii) long-term orangutan observational data to parameterize and test our model using pattern-oriented modeling techniques (Grimm and Railsback, 2012). As model parameterization requires extensive computational resources, we took advantage of high-performance computing clusters through the OpenMOLE software platform developed for exploring the behavior of complex computer-based models (Reuillon et al., 2013). Our study, thus, introduces an integrative approach combining field ecology, agent-based modeling, and advanced calibration methods.

Field data were collected in Sebangau National Park, Central Kalimantan, Indonesia (Figure 1). The research site had experienced historic human-induced disturbances involving selective logging by concessions and illegal loggers before the establishment of the national park in 2004 and the control performed by locally-led community patrol teams (Husson et al., 2018). Forest alterations, including the establishment of timber extraction canals by illegal loggers, have led to a legacy of imposed peat drainage and heightened dry-season fire risk (Wösten et al., 2008; Harrison et al., 2009). As such, the area experienced multiple fire episodes in 1992, 1994, 1997, and 2002 (Balai Taman Nasional Sebangau, 2015). During the dry season in 2006, 2015, and 2019 forest fires re-occurred in the area (Hayasaka et al., 2020).

BORNEO is the first ABM to incorporate complex movement decisions of primates (in particular, orangutans) involving their habitat structure and locomotion behavior. The model (partially) reflects the complexity of the orangutan's habitat by representing forest structure as a tree-network that influences orangutan movement. Such detailed representation of habitat structure has not been considered in similar primate ABMs. Evers et al. (2014) presented a model of primate emotional-social behavior, which described the habitat as simplified grid cells. Meanwhile, Gluckman and Bryson (2011) did not include explicit spatial representation for their primate social structure model. The ABM primates models more similar to ours (e.g., Bonnell et al., 2013; Bialozyt et al., 2014) did not consider the complexity of the forest structure as an influence on movement decisions. While Bialozyt et al. (2014) considered two types of movement toward fruiting trees (i.e., directional and semi-directional types), orangutans in BORNEO actively search for the most efficient way to move to fruit trees using various movement types, depending on the forest structure conditions. Bonnell et al. (2013) included spatial memory, while the current version of BORNEO has not explicitly included spatial memory.

The use of OpenMOLE (Reuillon et al., 2013) as a comprehensive suite to support the simulation experiments has enabled a robust, efficient, and reproducible application of Pattern Oriented Modeling (Railsback and Grimm, 2019). While some agent-based models use systematic, uniform sampling of the parameter space and manual comparison to the desired pattern (e.g., Pietzsch et al., 2021; Widyastuti et al., 2021), the NSGA2 algorithm (Deb et al., 2002) implemented in OpenMOLE allows exhaustive exploration of the model's parameter space. This study, thus, has employed a more recent practice in simulation experiments, which could be beneficial for similar agent-based models.

The findings from our BORNEO model complement findings from Erb et al. (2018), which highlighted the behavioral and energetic response of wild orangutans during fire incidence in a neighboring, ecologically similar habitat to our study site (Ashbury et al., 2022). Changes in forest structure increased the frequency of terrestrial movement in BORNEO. This behavior was also observed in real observations of orangutans in disturbed habitats derived from camera trap observations (Ancrenaz et al., 2014).

Our simulation results clearly show how orangutans are challenged by even a medium-level disturbance: their daily behavior is rapidly disturbed at relatively low fire intensity and is overwhelmed by high fire intensity. Our model showed that males are more sensitive to fires than females, probably due to their greater body mass, and the consequently greater energy intake that especially flanged males have to maintain. At a high burning intensity, orangutans significantly increased the proportion of terrestrial walking for both flanged males (AFM and H) and adult females (AF and I). The simulated orangutans in BORNEO maintained a high proportion of arboreal movement during no and low fire disturbance and increased their proportion of walking when the disturbance was at medium and high. Our findings also suggest that male energy balance is more sensitive to disturbance than that of females: a consequence of the aforementioned increase in travel distance and decreased feeding at lower burn intensity for males.

Since we used a small spatial representation (1 ha), we assumed that the orangutan has complete knowledge of the most efficient route to fruiting trees, though we recognize that in practice this assumption is unlikely to be fully met for wild orangutans in a constantly changing forest-food landscape. Orangutan travel might, therefore, be less efficient than indicated by our models. While the BORNEO model provides a fine scale movement simulation, a future challenge of the model in providing ecological insight for this endangered species would be to incorporate a larger area covering a full orangutan home range size and maintaining important details of habitat features for movement. Such implementations, however, are challenging due to data availability on detailed forest structures at larger scales, as well as simulation computation costs.

Incorporating more orangutans and different species in the model could also provide a platform to investigate competition and resource exploitation among individuals and between different species, in particular individuals living together; i.e., adult females with (semi-)dependent offspring (Goossens et al., 2006; van Noordwijk et al., 2012). Incorporating other individuals in the model could also provide the opportunity to test whether orangutans can perform spatial synchrony (Liebhold et al., 2004) of home-range arrangement (Wang and Grimm, 2007), an important feature in mammals. Furthermore, since fire can occur at large scales beyond a protected area (e.g., Imron et al., 2022), understanding their impacts across landscape levels while maintaining relevant details is a further challenge for BORNEO development. As orangutan movement decisions are also a result of interactions with other individuals, future work could focus on movement decisions (a) in small groups (e.g., females with their infant), (b) due to resource competition, or (c) socio-sexual reasons (i.e., males moving away from each other to avoid fights, or toward each other to fight; males moving toward females to attempt courtship, and females moving away from or toward males depending on mating triggers). Exploring such questions could provide valuable insights into movement ecology.

BORNEO is based on the premise that orangutans are omniscient about the environment. Hence, exploration, learning, and memory of the environment as well as learning processes are important in such environments. Moreover, the body weight is a constant in the present model, which lacks a proper metabolic/energetic submodel. Further model improvements could include incorporating the interaction between learning and metabolism aspects to consider feedback between these (Vincenot et al., 2015). Different search strategies of orangutans, explicit consideration of the Dynamic Energy Budget approach, and forest dynamics would also improve the model.

Although the impact of fire on plant diversity was not specifically modeled, a reduction in the number of trees of particular fruiting trees may have important influences on orangutans' movement target and direction. Incorporating into the BORNEO model other models such as peatlands fire simulations considering incidence (Widyastuti et al., 2021) and burn intensity to simulate movement decisions of orangutans during fires might further improve our understanding of how this species could survive through catastrophic events. In addition, the use of the BORNEO for Population Viability Analysis at the landscape level can provide insights for conservation guidance for this endangered primate as has been demonstrated for other species (Imron et al., 2011). Expansion of BORNEO for mitigating human-orangutan conflicts (Meijaard et al., 2011) would also provide opportunities to further understanding new emerging diseases as part of the one-health issue (Chappell and Thorpe, 2022).

The datasets presented in this study can be found in online repositories. The names of the repository/repositories and accession number(s) can be found in the article/Supplementary material.

KW designed and developed the model, collected tree survey data, performed simulation experiments, analyzed the simulation result, and wrote the manuscript. RR supervised the simulation experiments execution, analyzed the model simulation result, proposed model improvements, and reviewed the manuscript. PC analyzed the model simulation result and reviewed the manuscript. WA reviewed the model design, proposed model improvements, and reviewed the manuscript. DN, MH, and HM-B supported the field work and data collection and reviewed the manuscript. MI and UB supervised the research project and reviewed the manuscript. All authors contributed to the article and approved the submitted version.

This research is part of a project funded by UK Research and Innovation (UKRI) through the Global Challenges Research Fund (GCRF), grant number NE/T010401/1. For the purpose of open access, the authors have applied a Creative Commons Attribution (CC BY) license to any Author Accepted Manuscript version arising. KW was supported by DAAD doctoral research grant.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.