Drivers of Bornean orangutan distribution across a multiple-use tropical landscape of Bornean orangutan distribution across a multiple-use tropical landscape. Remote Sensing.

: Logging and conversion of tropical forests in Southeast Asia have resulted in the expansion of landscapes containing a mosaic of habitats that may vary in their ability to sustain local biodiversity. However, the complexity of these landscapes makes it difficult to assess abundance and distribution of some species using ground-based surveys alone. Here we deployed a combination of ground-transects and aerial surveys to determine drivers of the Critically Endangered Bornean Orangutan ( Pongo pygmaeus ) distribution across a large multiple-use landscape in Sabah, Malaysian Borneo. Ground-transects and aerial surveys using drones were conducted for orangutan nests and strangler fig trees (an important food resource) in 48 survey areas across 76 km 2 , within a study landscape of 261 km 2 . Orangutan nest count data were fitted to models accounting for variation in land use, above-ground carbon density (ACD; a surrogate for forest quality), strangler fig density, and elevation (between 117 and 675 m). Orangutan nest counts were significantly higher in all land uses possessing natural forest cover, regardless of degradation status, than in monoculture plantations. Within these natural forests, nest counts increased with higher ACD and strangler fig density, but not with elevation. In logged forest (ACD 14 – 150 Mg ha -1 ), strangler fig density had a significant, positive relationship with orangutan nest counts, but this relationship disappeared in forest with higher carbon content (ACD 150- 209 Mg ha -1 ). Based on an area-to-area comparison, orangutan nest counts from ground transects were higher than from counts derived from aerial surveys, but this did not constitute a statistically significant difference. Although the difference in nest counts was not significantly different, this analysis indicates that both methods under-sample the total number of nests present within a given area. Aerial surveys are therefore a useful method for assessing orangutan habitat use over large areas, however the under-estimation of nest counts by both methods suggests that a small number of ground surveys should be retained in future surveys using this technique, particularly in areas with dense understory vegetation. This study shows that even highly degraded forests may be suitable orangutan habitat as long as strangler fig trees remain intact after areas of forest are logged. Enrichment planting of strangler figs may therefore be a valuable tool for orangutan conservation in these landscapes.

were significantly higher in all land uses possessing natural forest cover, regardless of degradation 23 status, than in monoculture plantations. Within these natural forests, nest counts increased with on an area-to-area comparison, orangutan nest counts from ground transects were higher than from 28 counts derived from aerial surveys, but this did not constitute a statistically significant difference. 29 Although the difference in nest counts was not significantly different, this analysis indicates that 30 both methods under-sample the total number of nests present within a given area. Aerial surveys 31 are therefore a useful method for assessing orangutan habitat use over large areas, however the 32 under-estimation of nest counts by both methods suggests that a small number of ground surveys 33 should be retained in future surveys using this technique, particularly in areas with dense 34 understory vegetation. This study shows that even highly degraded forests may be suitable  Enrichment planting of strangler figs may therefore be a valuable tool for orangutan conservation 37 in these landscapes.

Introduction 43
Tropical forests are home to two thirds of the world's biodiversity, but are being lost or 44 degraded due to the expansion of agriculture and logging [1]. Since 2000, the area of 45 intact forest has been reduced by 7.2% globally, and South East-Asian forests specifically 46 have shrunk by 13.9% [2]. As intact forest declines, species are forced to adapt to more 47 degraded habitat conditions and to mosaics of anthropogenic land use types.
48 Understanding how species respond to human modified forests can inform land use 49 decisions and species-specific management strategies. 50 51 Bornean orangutans (Pongo pygmaeus morio) are critically endangered due to hunting [3, 52 4], habitat loss arising from logging and conversion of forest to industrial oil palm 53 plantations and other forms of agriculture [3]. It is estimated that habitat destruction, 54 fragmentation and hunting drove a decline of approximately 100'000 Bornean 55 orangutans between 1999 and 2015 [3] and that 78% of Bornean orangutan range lies 56 outside protected areas, within logging concessions and partially forested oil palm and 57 timber plantations [5]. This suggests that the capacity of orangutans to survive in 58 human-modified habitats and across a mosaic of land use types is critical to their future 59 persistence. 60 61 Orangutans construct a nest in the branches of trees on an almost daily basis, for resting 62 overnight and sometimes during the day [6]. The traditional approach to surveying 63 orangutan density is to make observations of their nests along ground-transects within 64 discrete areas of homogenous habitat [7,4] ]. However, unless multiple surveys can be 65 conducted across a large area, information collected from ground-transects is based on 66 orangutan activity within a narrow band of habitat, limited by the horizontal distance at 67 which an observer can identify a nest under forest cover [6]. Moreover, in human-68 modified landscapes, the small size of forest fragments and presence of multiple land 69 use types can result in a complex mosaic of habitats that are difficult to survey using a 70 ground-transect approach. 71 72 An alternative method to overcoming the small-scale habitat complexity and large-scale 73 sampling effort is to implement aerial surveys using helicopters or drones and to 74 quantify the number of canopy-visible nests. Information gained from aerial surveys can 75 capture data from a rapidly changing landscape and provides more extensive coverage 76 at lower cost than ground-based surveys [8]. Helicopter surveys have been used to 77 assess orangutan population densities for several years; however, helicopter flights are 78 significantly more expensive than aerial surveys by drones and can be prohibitively 79 expensive for small NGOs [9]. Helicopters are also in high demand and can therefore be 80 difficult to secure for surveying purposes. Additionally, helicopter surveys do not 81 generally collect precise information on nest locations, which is required for research on 82 the fine-scale drivers of orangutan habitat choice. A comparison of these methods across 83 a relatively small study area (5 km 2 ) in Sumatra found that orangutan nest counts were 84 significantly lower in aerial surveys by drone than from ground-transects [8]. The aerial 85 survey reported by Wich et al., (2016), was conducted at 150 m above ground-level with 86 a 12 MP camera [29], whereas a similar study of chimpanzee nest detection by drone 87 survey found that the nest detectability increased with image resolution [8]. Image 88 resolution is therefore expected to have a strong effect on nest detection and therefore on 89 the difference in nest encounter rates between aerial and ground-transects for nests. In 90 this study we compare nest counts from aerial surveys and ground-transects over a 91 much larger and more complex landscape to fully understand the strengths and 92 weaknesses of each approach to sampling orangutan populations and to assess the 93 conditions and resources associated with estimating orangutan population density. 94

95
Environmental variables known to affect orangutan nest distribution and habitat 96 preference were mapped in order to determine the drivers of orangutan nest 97 distribution within this landscape. It is well known that forest quality is a strong 98 predictor of orangutan habitat suitability [10]. Forest degradation due to logging and 99 agricultural conversion generally results in lower food resource availability and higher 100 energetic costs associated with dispersal [11]. However, this relationship may not be 101 linear, as low-intensity disturbance to forests can result in higher availability of fruit-102 producing tree species, providing greater foraging opportunities [12,13]. Additionally, 103 the highest recorded orangutan abundances in Borneo occur in selectively logged forests 104 in Kalimantan and Sabah, and old growth forest in Sarawak [3]. However, high 105 orangutan densities in degraded forest may also be the result of refugee crowding, as 106 individuals flee from areas of active logging into neighbouring intact forest [14]. The 107 relationship between forest quality and orangutan nest density in regions with multiple 108 land uses is therefore worthy of further study. In this study, above-ground carbon 109 density (ACD) was used as a surrogate for forest quality across the study landscape, 110 which is justified by the sensitivity of ACD to logging intensity across our study region 111 [15]. 112

113
The highest orangutan densities occur within lowland habitats, and they are generally 114 rare or absent at elevations over 500 m [14]. This elevational decline may be driven by 115 changes in the abundance and phenology of important food sources such as strangler fig 116 trees and fruit-producing lianas [16]. Strangler fig (Ficus spp.) trees are considered a 117 keystone food resource for multiple frugivores in Bornean forests, including orangutans 118 [17], providing a rich source of sugars, protein, carbohydrates, and calcium [18]. Bornean 119 forests possess a distinct episodic reproductive phenology, characterised by irregular 120 synchronous masting of canopy trees on cycles of 7-10 years [19]. Thus, it has been 121 suggested that the carrying-capacity of orangutans in lowland dipterocarp forest is 122 largely dependent on the amount of fall-back food resources available outside masting 123 events, including leaves, bark, pith, and insects [20].  [15]. (c) Map of above-ground carbon 147 (ACD) Mg ha -1 , derived from LiDAR survey across the study area [15]. (d) The grid 148 system used to organise the distribution of aerial plots within the study area and provide 149 a reference for the spatial random effect used in the model. Each blue square represents 1 150 km 2 , with each black square representing the location of survey areas. 151 The multiple-use forest landscape was defined by the Sabah State government in 2012 to 152 bring the management of protected areas and commercial land use types under a 153 common management umbrella ( Figure A1). Heavy historical timber extraction from 154 forests in Sabah has resulted in a recent decline in logging revenue, and efforts are being 155 made to create revenue from production forests by embedding short (8-15 yr) rotation 156 plantations within existing logging concessions, referred to hereafter as Integrated 157 Mosaic Planting (IMP) areas which cover 12.8% of the study area (33,512 ha).
158 Approximately 56.7% of the study area (148,357 ha) is composed of protected Class 1 159 Forest Reserves, which contain a mix of logged and unlogged forest where logging and 160 hunting are banned. A further 9.0% of the study area (23,977 ha) consists of unmanaged 161 rubber (Hevea brasiliensis) and acacia (Acacia mangium) plantations. Approximately 9.0% 162 of the study area (23,847 ha) is proposed for conversion to oil palm plantations, of which 163 a quarter had been cleared and terraced by the mid-point of our sampling in 2017. Five 164 separate forest fragments, amounting to 7,311 ha, or 2.8% of the total study area, are 165 protected as 'Virgin Jungle Reserves', consisting mainly of unlogged primary forest on 166 steep topography. 167 168 For the purposes of this analysis, five land uses were recognised: (i) Class 1 protected 169 forest, (30 survey areas, 59.31 km 2 ), (ii) oil palm plantations (3 survey areas, 1.15 km 2 ), 170 (iii) silvicultural plantations of rubber (3 survey areas, 3.23 km 2 ) or Acacia mangium (2 171 survey areas, 1.37 km 2 total) labelled 'silviculture' from hereon, (iv) integrated mosaic 172 plantations (5 survey areas consisting of 1-5 hectare patches of timber trees, interspersed 173 with remnant forest patches, 7.25 km 2 total) and small 'agroforestry' areas (2 survey 174 areas, 2.98 km 2 total) labelled 'IMP areas' from hereon, and (v) natural riparian forest of 175 roughly 100 m width embedded within oil palm plantations (3 survey areas, 1.1 km 2 176 total). 177 178

Sampling design and survey methods 179
Orangutan nests and large strangler fig trees (Ficus spp.) were surveyed across 48 areas. 180 These survey areas were determined at random to sample at least three survey areas 181 within all land use types (after combining Acacia and rubber plantations, due to similar 182 land-cover characteristics) and subject to the constraint that surveys had to be accessible 183 to sampling on foot and by drone (i.e., < 2.5 km from a road). Furthermore, land uses 184 that covered larger areas were sampled more comprehensively based on their relative 185 representation within the study landscape. On average, the 48 aerial surveys covered 186 149 ha (range 38 to 252 ha, SEM 0.083), for a total area of 76.39 km 2 , or approximately 187 28% of the study landscape. Forty-four areas were surveyed using both aerial and 188 ground-transect methods (Figure 1 [24]. The bearing was 234 then adjusted to account for the difference between the direction of the drone and true 235 north. The distance between pixels on the ground was calculated using the ground-236 surface distance formula [25] and Vincenty's Formula [26]  Europe Ltd, United Kingdom) were used as the length measurement for calculating the 248 actual distance covered during each transect. It is estimated that this model has an 249 average positioning accuracy of 4.5 m [28]. Transect width was calculated using the 250 Effective Strip Width (ESW) function of the 'Distance' package [29] in R version 2. 15 For question two, we assessed the effects of land uses (continuous forest, integrated 294 mosaic plantation areas, oil palm plantations, oil palm riparian strips and silviculture 295 areas) on nest counts, fig counts, and ACD within each aerial survey area. We used a 296 generalised linear mixed model with a Poisson error structure for the count data and a 297 linear mixed effects model for ACD, using the 'lmer' function in the 'lme4' package in R 298 [30]. The location of samples within grids was included as a random effect to account for 299 spatial autocorrelation as above. The log transformed area of each polygon used in this 300 analysis was included using the 'offset' function, to account for differing polygon sizes. 301 For question three, we investigated how forest degradation affects orangutan nest 302 density, estimating the influence of ACD, elevation, and strangler fig density on 303 orangutan nest counts derived from aerial surveys, within the subset of polygons 304 containing forest along a disturbance gradient. Survey areas covering monocultures and 305 IMP were excluded, but those with riparian forest within oil palm plantations were 306 included. This set of samples encapsulated an ACD range from 31 to 209 Mg ha -1 that is 307 assumed to reflect a gradient of forest quality, as tree species diversity is known to 308 increase with aboveground carbon density in human modified landscapes [31]. Data 309 were fitted to generalised additive models (GAM) using the 'mgcv' package in R,

Effects of survey method on orangutan nest counts 334
Based on an area-to-area comparison of nest counts derived from each method, mean (± 335 SEM) orangutan nest count derived from aerial surveys (0.402 ± 0.020 nests km -2 ) was 336 not significantly different (F1,80 = 1.007, P= 0.773, Figure 2    4.1 Comparison of survey methods 396 Mean orangutan nest count density did not differ between aerial surveys and 397 ground-transects across our study area in Southeast Sabah. This result contrasts with 398 previous research in Sumatra showing that orangutan nest counts were significantly 399 lower in aerial surveys by fixed-wing drone than in ground-transects that sampled the 400 same habitat [8]]. However, the aerial surveys in the Sumatran study were made from 401 approximately 50 m higher than that adopted in our study, and using a 12 MP camera[8], 402 which is significantly lower resolution than the 20 MP camera used for 96% of the surveys 403 in this study. Therefore, it remains a possibility that the lower nest count density in the 404 aerial surveys of the Sumatran study is a methodological artefact, resulting from the 405 higher altitude surveys and use of a lower resolution camera. 406 407 Despite the absence of a difference in nest counts between the two survey methods, 408 it is likely that both methods under-estimate the true density of Orangutan nests. This is 409 because nests constructed on top of tree crowns, which are most visible in aerial surveys, 410 are difficult to detect by an observer from the ground, and conversely, nests below the 411 tree crown may be invisible in drone surveys. The under-estimation of nest counts in 412 ground-transects may be particularly acute in the dense second vegetation typical of 413 highly degraded forest, while aerial surveys might be expected to under-estimate nest 414 counts in high quality forest with a more heterogeneous canopy structure [8]. However, 415 the absence of a significant interaction between survey method and ACD in our study 416 suggests that the relative success of the two survey methods does not vary in response to 417 forest quality. In order to estimate the extent to which each survey method under-418 estimates true nest density, future studies should record precise coordinates of each nest 419 and then overlay maps of nest locations to determine those that had been missed in each 420 case. This would allow researchers to compute a local conversion factor for scaling nest 421 counts from aerial surveys to total counts in each setting. In order to compute these 422 conversion factors, ground transects are still required to complement aerial survey 423 techniques in orang-utan nest surveys. 424 425 4.2 Effect of land use on orangutan nest counts, strangler fig counts and above-426 ground carbon density 427 Conversion of logged forest to create single-species plantations of oil palm, acacia, or 428 rubber resulted in a reduction in orangutan nest counts, even when these plantations 429 retained small patches of remnant forest. Only one nest was observed in 3.2 km 2 of rubber 430 plantations surveyed, and none were observed in 1.5 km 2 of oil palm plantations, 0.21 km 2 431 of oil palm riparian strips or 1.4 km 2 of acacia plantations. Integrated mosaic plantation 432 areas had higher median orangutan nest counts and fig density than monoculture 433 plantations, but values were still substantially lower than in areas with a continuous cover 434 forest, except where that forest was very heavily degraded. These data suggest that loss 435 of forest cover reduces habitat quality for orangutans, even when natural forest cover is 436 replaced by tree plantations equivalent in height and ACD to some natural forests. The 437 factor that unites all the non-forest land uses compared here is the clearance of land prior 438 to planting, and the creation of a woody vegetation with a much more homogeneous 439 structure and species composition. Orangutans have been documented feeding on oil 440 palm fruits within plantations, however agricultural monocultures are infrequently used 441 by orangutans for nesting purposes [13] so nest construction in these land use types is 442 unlikely even if orangutans are present. This study confirms this finding and extends it 443 by recording limited use by orangutans of rubber and acacia plantations. 444 445 The low abundance of orangutan nests in silviculture plantations may arise for 446 multiple reasons, including an inappropriate forest structure for nesting or arboreal 447 dispersal[33], increased likelihood of disturbance or mortality of orangutans due to 448 contact with humans and domestic animals [3] or an absence of food resources [34]. . It is possible that the limited sampling 459 of these areas coupled with unique characteristics of this study site explains the low 460 number of nests recorded. In our study area, a major road passes between the single estate 461 surveyed and neighbouring natural forest, therefore the riparian strips sampled are only 462 connected to one fragment of continuous forest and they would not be able to function as 463 uninterrupted dispersal corridors. Ficus spp have been observed growing in higher 464 densities in riparian forest in Thailand [37], which may explain the high numbers 465 observed in our study, despite the small area sampled. These observations suggest that 466 the relationship between orangutan occupancy of a habitat and the availability of figs may 467 be decoupled by the spatial structure of the habitat, as a lack of connectivity between these 468 riparian strips and larger forest fragments makes dispersing for this food resource a less 469 viable feeding strategy.  [38, which showed that the correlation between nest 479 density and ACD was weak and non-significant. However, this may be because the LKWS 480 covers a smaller range of land use types, comprising primarily disturbed forest that 481 possesses a narrower range of ACD values (0 -150 Mg ha -1 ), than those included in the 482 multiple-use forest landscape we examined [38]. 483 484 Higher nest counts in less degraded forest may arise because of orangutan 485 preferences for specific forest structural characteristics that are modified by logging, 486 combined with changes in food resource availability linked to logging disturbance. Tall 487 and stable trees with a complex branching structure are preferred for nest building, 488 possibly because they create a stable platform for nests in wind and rain and provide a 489 useful vantage point over the forest [ [39,18]. Figs are also a reliable and consistent food resource, because 505 different species fruit asynchronously and the intervals between fruiting events are short 506 [40]. Consequently, they are highly sought after, and trees possessing large fruit crops can 507 result in aggregations of orangutans and other frugivores [41]. 508 509 Changes in food availability in response to logging may also be a significant driver 510 of orangutan nest abundance. Mean fruit availability is a strong predictor of orangutan 511 density [42] and disturbed forests are known to have lower food availability for 512 orangutans [14]. This is reflected by the findings of this study, as nest counts generally 513 increased with higher ACD. On the other hand, the five highest nest counts observed in 514 this study were located in more disturbed forest (ACD < 150 Mg ha -1 ). This partial 515 decoupling may occur for several reasons. First, Bornean orangutans display considerable 516 dietary flexibility, which allows them to extend their range into more disturbed 517 environments when foraging for alternative food sources [12]. The fruits and leaves of 518 pioneer species such as Macaranga pearsonii and Neolamarckia cadamba that are abundant in 519 degraded forests across the study area are potentially important alternative food sources 520 [12], while tree bark and insects also provide a reliable source of nutrients 530 531 Contrary to expectations, there was no evidence of a decline in orangutan nest counts 532 across the range of elevations surveyed in this study (117 to 675 m). A possible explanation 533 for this lack of effect of elevation is that our entire study area was above the threshold 534 elevation of 100 m that makes a difference for orangutan abundance. For example, a 535 previous study of Bornean orangutan populations in Kalimantan showed that densities 536 declined beyond 100 m asl. [4]. That interpretation may also explain the generally low 537 population densities of orangutans across our study area in Sabah (nest densities in the 538 range 0 -93.6 km -2 in forested habitats) compared to populations examined in forests at 539 lower altitudes (10-20 m asl) where nest densities are in the range 87.5-1149.9 km -2 in 540 forested habitats [45]. 541 542 5. Conclusions 543 This study highlights the drivers of orangutan distribution in a multiple-us 544 landscape, based on the observation of nest counts across multiple survey areas within 545 this landscape. Orangutan nest counts declined significantly in response to increasing 546 intensity of land use (Fig 3 a), in conjunction with decreasing ACD (Fig 4 b). These results 547 emphasize the importance of remnant forest, with low rates of human disturbance as 548 important orangutan habitat in multiple-use forest landscapes. Strangler fig density was 549 also shown to be a significant driver of orangutan nest density, with high nest counts 550 observed in forest with a higher densities of strangler fig trees (Fig 4 b). The importance 551 of strangler fig trees as food sources for orangutans in logged and degraded forests, which 552 is supported by our study as well as others [18, 22,], justifies specific management 553 interventions that might enhance the conservation of orangutans in these habitats. For 554 example, enrichment planting of strangler fig trees might be an effective technique for 555 increasing food availability and habitat quality in degraded secondary forests, especially 556 when combined with other measures for restoring forest structure and species 557 composition [46]. In addition, restrictions on cutting lianas with fleshy fruits consumed by 558 orangutans would limit the reduction in strangler fig trees and fruit-producing lianas that 559 occurs when generic climber cutting practices are used to aid regrowth of mature trees in 560 logged forest [47]. In multiple-use landscapes, forest patches may be small and isolated, 561 but they often possess sub-populations of orangutans that are vital to sustaining the 562 viability of the regional metapopulation, distributed across a heterogeneous landscape 563 [14]. The ability to conduct rapid surveys of forest fragments in their entirety across these 564 landscapes may be a vital tool for monitoring the status of orangutan populations in the 565 future. Our work demonstrates that drone surveys have the potential to play an important 566 role in that effort. 567 568 Despite the under-estimation of orangutan nest density by both aerial surveys and 569 ground-transects, the larger area sampled by drones than ground surveys for an 570 equivalent effort expands the scope and accuracy of inferences about the drivers of 571 orangutan abundance and distribution, particularly when sampling heavily disturbed 572 environments or populations with low individual density. When coupled with an 573 effective correction factor for under-sampling of nests, and high throughput image 574 analysis, drone surveying could serve as an effective rapid assessment tool for monitoring 575 orangutan populations [8]. However, the process of sorting through aerial images 576 individually was time-consuming and prone to human error. Adopting a machine 577 learning approach for identifying orangutan nests in aerial images may save time and 578 improve standardisation in future surveys [48].