Riparian buffers act as microclimatic refugia in oil palm landscapes

1School of Biological and Chemical Sciences, Queen Mary University of London, London, UK; 2Asian School of the Environment, Nanyang Technological University, Singapore City, Singapore; 3Durrell Institute of Conservation and Ecology (DICE), School of Anthropology and Conservation, University of Kent, Canterbury, UK; 4Department of Zoology, University of Cambridge, Cambridge, UK; 5Department of Plant Sciences, University of Cambridge Conservation Research Institute, Cambridge, UK; 6Forest Research Centre, Sabah Forestry Department, Sandakan, Sabah, Malaysia; 7Institute for Tropical Biology and Conservation, Universiti Malaysia Sabah, Kota Kinabalu, Sabah, Malaysia; 8School of Biological Sciences, University of Bristol, Bristol, UK and 9Department of Zoology, University of Oxford, Oxford, UK

Riparian buffers, or riparian reserves, are areas of non-production habitat (often forest) retained around rivers in agricultural landscapes, primarily as a means of protecting water quality by reducing run-off (Tabacchi et al., 2000). In many tropical nations, riparian buffers of a designated width are required by law, often based on the size of the river in question (Luke, Slade, et al., 2019). In addition, policies on riparian buffer width have been adopted by groups such as the Roundtable on Sustainable Palm Oil (RSPO) as part of their certification criteria for mitigating the detrimental impacts of oil palm development on the environment and local communities (Luke, Slade, et al., 2019). In addition to their primary role in protecting water quality, riparian buffers provide a range of other ecosystem services, such as carbon storage (Mitchell et al., 2018), flood protection (Tabacchi et al., 2000) and subsidising water courses with terrestrially derived organic matter (Allan, 2004). Buffers can also provide co-benefits for a variety of terrestrial (Barlow et al., 2010;Keir et al., 2015;Zimbres et al., 2017) and aquatic taxa (Cunha & Juen, 2017;Giam et al., 2015). There are also examples of these habitat remnants serving as corridors between other forest areas, promoting connectivity for various taxa Keuroghlian & Eaton, 2008).
In common with all habitat fragments, the efficacy of riparian buffers for safeguarding biodiversity and promoting connectivity will depend on habitat area and quality, the level of contrast with the surrounding matrix, and the biology of the taxa in question (Lees & Peres, 2008). In general, the attributes of riparian buffers that support terrestrial biodiversity remain poorly understood. Recent studies have demonstrated the role of buffer width for birds (Keir et al., 2015;Lees & Peres, 2008;Mitchell et al., 2018), mammals (Zimbres et al., 2017) and dung beetles (Barlow et al., 2010;Gray et al., 2017), with several subsequently linking observed biodiversity patterns to habitat quality (Lees & Peres, 2008;Mitchell et al., 2018).
However, studies investigating riparian buffer microclimates and the features that shape them are scarce (e.g. Nagy et al., 2015).
Insights into the effects of microclimate on tropical biodiversity are often limited by issues of scale and accuracy Schulze et al., 2001), with most studies relying on coarse-resolution mapping databases such as WorldClim (Fick & Hijmans, 2017).
Advances in technologies such as Light Detection And Ranging (LiDAR) make it possible to map landscapes and vegetation with unprecedented levels of accuracy and precision (Zellweger et al., 2019), allowing studies to better quantify and link physical habitat structure to microclimate . The decreasing costs of microclimatic dataloggers have also catalysed an increase in research investigating fine-scale microclimatic conditions (e.g. Hardwick et al., 2015;Law et al., 2019).
Here we combine information from airborne LiDAR with fieldbased microclimatic measurements to investigate the efficacy of forested riparian buffers of different widths and habitat composition for providing microrefugia within oil palm plantations. We deployed dataloggers across three riparian habitats: oil palm, riparian buffers and continuous logged-forest, in Sabah, Malaysian Borneo. First, we examine if riparian buffers in otherwise microclimatically extreme plantations maintain conditions similar to those found in continuous riparian forest. We then demonstrate how vegetation conditions and topography shape the microclimate within riparian buffers, and across human-modified landscapes as a whole, before evaluating how edge effects influence buffer microclimate and the implications this has for policy pertaining to buffer width. Finally, to assess the capacity of tropical riparian buffers to act as microrefugia for a key invertebrate indicator group (Scarabaeinae), we couple the microclimatic data with dung beetle community data across the modified landscape.

| Study site
Fieldwork was conducted in and around the Stability of Altered Forest Ecosystems project (www.safep roject.net; 4°81′N 117°25′E -4°43′N 117°64′E, plot elevation ranged from 125 to 450 m a.s.l) in Sabah, Malaysia (Northern Borneo, Figure 1a). This region is characterised by a tropical climate, with annual rainfall ~2,700 mm and a mean annual temperature of 26.7°C (Walsh & Newbery, 1999), although a recent study shows the region has become hotter and drier in recent years (Chapman et al., 2020). The area was formerly continuous lowland dipterocarp forest with much of the remaining forest having been selectively logged in the 1970s and 2000s, and subsequently salvage logged in 2013 and 2015 in preparation for oil palm (Struebig et al., 2013). At the time of fieldwork, this forest was highly fragmented, bounded to the north in the landscape: in oil palm plantations (OP: n = 2 rivers, 6 transects), riparian forest buffers within oil palm plantations (RB: n = 16 rivers, 48 transects) and continuous logged forest (CF: n = 2 rivers, 6 transects).
Distances between adjacent transects on a river were 342-591 m, with rivers varying from 0.5 to 48 km apart (median 18 km). In oil palm and continuous forest, dataloggers were deployed in transects perpendicular to the riverbank at distances of 5, 15, 25, 35 and 45 m. In riparian buffer transects, dataloggers were deployed at 5 and 15 m from the riverbank and at ~5 m from the buffer-oil-palm edge, and at 5 and 25 m into the oil palm Figure 1d). A range of buffer widths (0-324 m) were investigated to investigate the effect of proximity-to-edge on microclimate. All units were suspended at 3 m above the ground with a polystyrene plate rain-cover and left to record temperature (T, °C) and relative humidity (RH, %) for 3-7 weeks at intervals of 30 min.

| Microclimate data
Datalogger data were collated to calculate maximum (T max ) and mean (T mean ) daily temperatures for each sampling day. RH was used to calculate vapour pressure deficit (VPD, hPa) -the difference in the partial pressure of water vapour in the air compared with saturated air at a given temperature (T): Vapour pressure deficit represents the evaporative demand of the air, pulling water up through the soil-root-stem-leaf continuum. Thus, it is a critical determinant of plant ecology, strongly influencing potential evapotranspiration and the ability of plants to supply their leaves with sufficient water during the driest parts of the day, and thereby regulating seedling growth and mortality (Williams et al., 2013). Maximum (VPD max ) and mean (VPD mean ) daily VPD were generated for each sampling day.

| Vegetation quality, topography and distance from buffer-oil-palm edge
To understand how vegetation quality and topography influence microclimate in riparian habitats we used airborne LiDAR data collected over part of the landscape (see Jucker et al., 2018). For the 35 transects coinciding with the 2014 LiDAR information, we extracted a set of vegetation and topographic metrics from the canopy height, digital terrain and plant area index model rasters, using a 12.5-m radius extraction. LiDAR-derived metrics were mean plant area index (PAI, log-transformed), maximum canopy height (H max ), topographic position index (TPI), elevation, aspect and slope. PAI was calculated empirically from the raw LiDAR data as an integrated measure of canopy density (m 2 /m −2 ; Holst et al., 2004). H max (m) was calculated as the maximum canopy height value (after ground-normalising the LiDAR point cloud) within 12.5 m of the logger. Four topographic covariates were calculated using the terrain function in the raster package (Hijmans, 2016) in O D R effects of manipulating buffer width on microclimate. For dataloggers associated with riparian buffers outside the LiDAR area (n = 22 transects), distances from the buffer-oil-palm edge were measured manually for buffer edge sites. River width was subtracted from total buffer width (as calculated from Google Earth imagery) and halved to give an estimate of buffer width for each transect. Distance from river was then subtracted from buffer width to give estimates of distance from edge for buffer core sites.

| Dung beetle diversity sampling
To understand how microclimate impacts the efficacy of riparian buffers as a means of supporting biodiversity, we carried out two dung beetle (Scarabaeinae) sampling campaigns, in January 2015 and from September 2017 to March 2018. The climate in our study landscape is relatively aseasonal (Walsh & Newbery, 1999), although sampling dates broadly correspond to the marginally wetter season (Marsh & Greer, 1992), where dung beetle activity is highest. Dung beetles are a useful indicator group due to their sensitivity to disturbance, high diversity, well-established taxonomy, ease of sampling and importance for a range of ecosystem functions (Nichols & Gardner, 2013). Like most tropical ectotherms, dung beetles are thought to be operating close to their thermal maxima, putting them at a greater risk of extinction due to climatic shifts (Deutsch et al., 2008). Dung beetle assemblages were sampled using human-dung-baited pitfall traps (following Slade et al., 2011). For each datalogger transect in a buffer (n = 48), one trap was deployed for two trapping nights ~10 m from the river.
The minimum distance between transects with traps was 381 m.
Beetles were collected into 90% ethanol and identified to species or morpho-species using reference collections housed at the Universiti Malaysia Sabah.

| Riparian buffers as microclimatic refugia
To examine whether riparian buffers and plantations maintain microclimatic conditions similar to those found in continuous riparian forest, we ran a mixed-effects model of each of our four microclimatic response variables (T max , T mean , VPD max and VPD mean ) against a fixed effect of four habitat types: continuous forest, riparian buffer core (buffer interior > 10 m from the buffer-oil-palm edge, hereafter referred to as buffer core), riparian buffer edge (buffer interior ≤ 10 m of the buffer-oil-palm edge, hereafter referred to as buffer edge) and oil palm. Sampling transect was fitted as a random effect, and models were run in the lme4 package in r (Bates et al., 2015) with a Gaussian error distribution. Habitat-type models were compared against the null model (only containing the random-effect) by comparing AIC, where a difference of −4 supports one nested model over another (Bolker, 2008).

| Effects of vegetation quality and topography on microclimate
To analyse the effects of vegetation quality and topography on microclimate, we took a subset of our data from 36 transects that coincided with LiDAR information. We defined separate maximal linear mixed-effects models for each of our four microclimatic response variables (T max , T mean , VPD max and VPD mean ) with all of our seven explanatory variables (PAI, H max , TPI, elevation, aspect, slope and habitat) fitted as fixed effects, and with one interaction term (H max : aspect) following Jucker et al. (2018), with sampling transect as a random effect and a Gaussian error distribution. By sequential removal of terms, every possible subset of each maximal model was generated (159 for each response variable) and ranked by AIC weight in the bbmle package in r (Bolker & R Development Core Team, 2017).
Models were then subsetted to retain the fewest possible models that cumulatively accounted for 0.95 or more of the total AIC weight.
The AIC weighted proportion of explanatory variable retention in the final models is reported.

| Edge effects on buffer microclimate
We examined the impact of edge effects on microclimatic conditions using distance from buffer-oil-palm edge. We analysed a subset of the full data that only included dataloggers deployed within riparian buffers (both buffer edge and core habitat types). Similar to the aforementioned habitat type analyses, each of the four microclimatic response variables (T max , T mean , VPD max and VPD mean ) were entered into mixed-effects models with distance into buffer from edge fitted as a fixed effect, sampling transect as a random effect and a Gaussian error distribution. Models were then compared to respective null models using AIC.

| Buffer microclimate impacts on dung beetle diversity
To analyse how riparian buffer microclimate impacts biodiversity, we matched our dung beetle assemblage samples to buffer core dataloggers. Microclimate data from sites 5 m from the river were used, unless data were only available from points 15 m from the river. Dung beetle diversity, calculated as Shannon diversity in the vegan package in r (Oksanen et al., 2010), was fitted as the response variable in four maximal linear models (for each of T max , T mean , VPD max and VPD mean ) with a Gaussian error distribution.
Each maximal model had distance from buffer edge (log-transformed), the microclimate variable of interest and an interaction term between the two, as explanatory variables. For each microclimatic explanatory variable, all possible combinations of explanatory variables were compared to the maximal model using dAIC. Similar analyses were conducted for species richness (see Appendix S1).

| RE SULTS
Of the 300 dataloggers deployed in riparian transects, 198 were recovered fully functioning, resulting in 5,438 days of microclimatic recordings. Of the 198 units, 110 were recovered within the LiDAR area, while 79 were located in riparian buffer core or edge and had width data available (see Table S1). All microclimatic variables (T max , T mean , VPD max and VPD mean ) were strongly correlated (Pearson's r > 0.6, see Table S2).

| Riparian buffers as microclimatic refugia
We found strong support for the impact of habitat type on T max ,    Figure 3; Table 2). TPI was a strong positive predictor of all four microclimatic variables (Figure 3). Elevation was a weak predictor of T max and T mean , with an increase of 100 m elevation resulting in a mean drop of 0.27°C (Table 2). Aspect was a weak predictor of T mean and VPD mean , with east-facing slopes being hotter and drier than west-facing slopes (Table 2). Slope and the interaction term between H max and aspect were not frequently retained in best-fitting models ( Table 2). Habitat type, the only non-LiDAR derived variable, was retained in the best-fitting models for T max , T mean and VPD mean (

| Edge effects on buffer microclimate
Linear mixed-effects models of distance into the buffer from the buffer-oil-palm edge (log-transformed) were strongly sup-  (Figure 4). All models had Cook's distances < 0.5 (see Figure S2).

TA B L E 2
Weighted proportions of the retention of each fixed effect across best-fitting models for T max , T mean , VPD max and VPD mean . AIC weights were generated for all models, before models were subsetted to include only those that cumulatively made up 0.95 of the total weight. AIC w. prop. is the proportion of the 0.95 cumulative weight constituted by models containing the fixed effect of interest. Values given in bold fell above an arbitrary threshold value of 0.5. Effect sizes of models with only a single explanatory variable are given, with the exception of habitat, as it is a categorical variable (see Table 1

| Buffer microclimate impacts on dung beetle diversity
Of the 48 transects associated with riparian buffers, 31 had functioning dataloggers in the buffer core, with associated data on both dung beetle diversity and distance from edge into buffer. Dung beetle diversity was driven by an interaction between distance from buffer edge and both T max and T mean ( Table 3). As distance from the edge decreased, the relationship between temperature (T max and T mean ) and dung beetle diversity became more negative, whereas at 80 m it is relatively flat (Figure 5). Responses for the interaction between VPD and distance from edge were similar to those of temperature but with lower AIC weight, particularly for VPD mean (Table 3). Further, lower dung beetle diversity was associated with higher T mean , VPD max and VPD mean , (Tables 3 and 4). Note, the VPD max model lacking an interaction term failed our leverage tests and must be regarded with caution (see Figure S3). Species richness analyses showed similar responses to Shannon diversity (see Appendix S1).

| D ISCUSS I ON
Our results demonstrate the capacity of riparian buffers to provide microclimatic refugia in human-modified tropical landscapes.
All four measures of temperature and vapour pressure deficit (T max , T mean , VPD max and VPD mean ) were lower in the core area of riparian buffers than in the surrounding oil palm, although these values were TA B L E 3 AIC weights of all models of Shannon Diversity for each microclimatic variable (T max , T mean , VPD max and VPD mean ), where 'interaction' denotes models containing the interaction term between buffer width and microclimate, 'additive' denotes a model containing microclimate and buffer width, 'buffer width' and 'microclimate' denote models containing only that term, and 'null' is the null model. Values given in bold make up the best-fitting models, as calculated by a cumulative ranked weight >0.95

F I G U R E 5
Visualisations of the interaction between the effects of (a)T max and (b)T mean , and distance from edge on dung beetle diversity. Solid lines give the effect of temperature on diversity, given two set distances (20 and 80 m) from the edge of the buffer, as predicted using estimates from linear models (see Table 3), with dashed lines denoting 95% confidence intervals. Grey-scaling on points and lines gives the magnitude of distance from buffer edge (grey > black)

TA B L E 4
Outputs of best-fitting linear models predicting the Shannon diversity of dung beetles (H′) sampled in riparian buffers.
In the model column '~' means 'as a function of' and '*' means the two terms individually and the interaction between the two are included still higher than those in continuous riparian forest. We reveal that buffer edge effects mediate microclimate, with the interior of the buffer being substantially cooler and more humid than edges and plantation. We subsequently demonstrate the key roles that greater vegetation complexity and topographic sheltering play in increasing the microclimatic buffering capacity of these set-asides. Finally, we elucidate the link between buffer width and microclimate, and dung beetle communities, revealing that proximity-to-edge and temperature can synergistically decrease local diversity.
Consistent with our results, Nagy et al. (2015) found that microclimates in riparian buffer cores in the southern Amazon were comparable to those of continuous riparian forests. Cooler and wetter conditions here were more strongly associated with wide buffers, particularly those 80 m or more in width (Table 5).
Although our buffer core sites were generally cooler than oil palm, edge habitat was characterised by more extreme conditions than adjacent plantation. Oil palm is a perennial crop with a peak yield occurring at an age of 9-18 years (Alam et al., 2015), with older plantations forming tall canopies with cooler microclimates (Luskin & Potts, 2011). We postulate that high T max and VPD max in buffer edges is due to gaps in vegetation associated with riparian buffer edges (JW personal observation). Such gaps are dominated by bare ground, grasses or low-lying vines, and may be due to clearing and spillover of herbicides from the plantation. The gaps could also elevate T and VPD for short periods of the day. The temperature and humidity extremes in buffer edges are consistent with welldocumented microclimatic changes seen in other edge habitats, which are typically attributed to increased solar radiation and wind (see Williams-Linera, 1990).
We reveal the link between several vegetation and topographic features, and microclimate, across a human-modified tropical landscape. In particular, PAI, a measure of vegetation quality, had a strong influence on microclimate. Increased PAI is associated with more complex vegetation (Holst et al., 2004), causing decreased wind and light exposure to give cool, humid conditions (Hardwick et al., 2015).
H max (maximum canopy height) was also strongly associated with T mean , a relationship driven by increased shading by tall trees . Like other edge habitats, riparian buffers are characterised by factors impacting vegetation structure, with reduced seedling abundance, tree basal area, canopy height and woody plant diversity compared with continuous riparian forest (Keir et al., 2015;Lees & Peres, 2007;Nagy et al., 2015). Topography was also a key predictor of our microclimatic variables, with TPI having strong positive correlations with temperature and VPD. Such results are indicative of the relative exposure of ridges (high TPI) and depressions (low TPI) to light and wind (Dobrowski, 2011). Aspect had a small positive effect on T mean and VPD mean , where east-facing slopes tended to be hotter and drier, likely due to daily solar radiation and wind patterns in the region (Smith, 1977). Elevation negatively predicted T max and T mean , with a 100-m increase resulting in a mean drop of 0.27°C, a lower impact than we might expect given the literature  and likely due to a limited range of elevations in our study. Our results highlight the importance of understanding how heterogeneous vegetation and topography must be taken into account when defining the extent of riparian buffers in environmental policies, as well as predicting landscape-or regional-level diversity responses under climate change scenarios (Elsen et al., 2020). In addition to microclimate, we also found that distance from edge was associated with higher local diversity, supporting a pool of literature demonstrating the positive impact of increased buffer width on terrestrial biodiversity (Gray et al., 2017;Keir et al., 2015;Zimbres et al., 2018 (Table 1) (80 m for dung beetles and forest-specialising birds; Gray et al., 2017;Mitchell et al., 2018) are in the region where some of our proximity-to-edge microclimatic response curves intersect with the 95% confidence interval for continuous forest and where edge effects generally tail off in many systems (e.g. Didham & Lawton, 1999;Laurance et al., 2002). The consequences of edge effects for biodiversity may be more pronounced in tropical landscapes with sparse open habitat where species have not experienced long-term selection pressures for avoiding edges (Betts et al., 2020). Ideally, edge effects would be investigated in the same set of models as vegetation quality; however, due to strong correlations between proximity-to-edge and our LiDAR-derived variables, this was not possible.

| Policy implications
Our results are important to riparian buffer policies in humanmodified tropical landscapes, supporting suggestions that mandatory riparian buffer widths in the tropics should be wider than they currently are, that more attention should be given to buffer habitat quality (Luke, Slade, et al., 2019), and that topography should also be considered when planning networks of buffers across landscapes. We show that buffers begin to reach microclimatic conditions comparable to those of continuous riparian forest at approximately 80 m and above, on each side of the river, a width previously suggested as adequate for maintaining representative levels of species diversity (Gray et al., 2017;Mitchell et al., 2018).
At this buffer width, the negative impacts of temperature on biodiversity are far less pronounced than at 20 m, the width typically required by law in Sabah, Malaysia. These recommendations are emphasised by the finding that buffer edges, and thus narrow buffers (<10 m), may be more microclimatically extreme than no buffer at all. In addition, many tropical countries do not consider vegetation complexity in riparian management policies (Luke, Slade, et al., 2019), but doing so could help contribute to improved microclimate conditions and long-term sustainability of waterways in agricultural areas. We therefore advocate efforts to extend buffer widths, prevent further degradation and restore riparian buffers (Luke, Advento, et al., 2019). In addition, by determining the vegetation and topographic features that drive microclimate in tropical riparian buffers, we hope to inform the future planning of buffer locations and networks. Taken together, our results suggest that safeguarding riparian buffer microclimate may help to limit the local extinction of species by providing microrefugia. This finding is likely to become increasingly important in the face of anthropogenic climate change (Hampe & Jump, 2011), particularly if demand for agricultural land near waterbodies increases with drier climates.