Supporting Information Mapping perceptions of species’ threats and population trends to inform conservation efforts: the Bornean orangutan case study Nicola K. Abram; Erik Meijaard; Jessie A. Wells; Marc Ancrenaz; Anne-Sophie Pellier; Rebecca K. Runting; David Gaveau; Serge Wich; Nardiyono; Albertus Tjiu, Anton Nurcahyo; Kerrie Mengersen. Corresponding author: Nicola K. Abram ([email protected]) Appendix S1: Spatial data framework Orangutan habitat The known geographic distribution of breeding orangutan populations were derived from (Wich et al., 2012). Using the orangutan distribution layer we generated two layers: (1) Euclidian distance from known orangutan breeding populations; and (2) the summed values of neighbouring cells (at 1 km cell size) within a 10 km radius (using the focal statistics tool in ArcGIS 10.0). Land-use and land cover Eight land cover layers were used (see Table S1) including: mangrove; intact natural forest; logged forest; severely degraded logged forest; agro-forests/forest re-growth; industrial timber plantations; oil palm plantations; and 'other' land cover types. Details of the generation of these layers can be found in (Gaveau et al., 2014). We included a protected area network layer, for details see (Wich et al., 2012). For these nine layers we calculated Euclidian distance to that feature (in meters). We developed an above ground carbon stock layer from LiDAR (Light Detection and Ranging) data from (Baccini et al., 2012) and converted values to Mg of CO2 per hectare at 30 arc-second resolution (approx 1 km2). For these ten variables we calculated the summed values of neighbouring cells within a 10 km radius using the focal statistics tool in ArcGIS 10.0. The 10 km threshold was determined under the assumptions that this was a reasonable distance for a person travelling on foot and a reasonable area for defining the local environment of the village surroundings. Table S1 Classes of Land cover types and degree of forest degradation for forested regions; along with their descriptions and processing steps. Class Description Processing steps 1. Intact natural forest Medium to tall old-growth natural forests that have never been logged by the timber industry. Open to closed canopy: closure is probably higher than 30%. Includes Lowland and montane dipterocarp forests, riverine forests, heath forests on plateaux, tall closed-canopy peat forests and open-canopy pole peat forests. Note, our intact forest class may include areas where the forest has been degraded slightly by small-scale logging, which we could not detect using Landsat imagery. Step 1: Generated a forest cover map for year 2010 using SarVision 2010 (unpublished) land cover map. Merged classes: 3 (pole peat forest), 6 (closed forest), 11 (riparian forest) and 17 (peat-swamp forest close canopy). Step 2: transformed the 1990-2000-2010 logging road network (indicating mechanized logging) into a road density index (km/km2) of 1x1km grid cell (search radius 5km). Step 3: overlaid the road density map onto the forest cover map generated in Step 1 and recoded forest into intact in areas where road density = 0. Areas where road density > 0 were coded as class 9 (see 1 below). 2. Mangroves Closed canopy Medium Forest with closed canopy of 10 % to 30% occurring in tidal affected zones. This class has been taken directly from SarVision 2010 land cover map. 3. Agroforests / forest regrowth Medium to tall agro-forests and forest regrowth. Open to closed canopy: closure is probably equal or higher than 30%. Includes traditional rubber agroforests, fruit gardens, and land under fallow, where forests is regenerating. Step 1: Generated a broad vegetation cover map including agroforests/forest regrowth/very degraded logged forests using SarVision 2007 (Hoekman et al., 2010) and 2010 land cover map. Merged class 8 (forest mosaic, fragmented or degraded forest) from SarVision’s 2007 land cover map with class 2 (woodland) and class 13 (open forest) from SarVision’s 2010 land cover map. Step 2: Overlaid the logging road network onto this broad vegetation map and recoded this class into agroforests/forest regrowth in areas where there were no logging roads. 4. Non-forest Includes: 1) Low vegetation of grasses or shrubs occurring on drained soils, occasionally flooded; 2) dry rice cultivation; 3) Low herbaceous vegetation with including tall grasslands and ferns; 4) can include agricultural cropland areas; 5) dry to occasionally flooded terrain; 6) areas of herbaceous vegetation, 7) shrub lands and young forest regrowth in fallow lands. This class has been generated by merging classes 4, 5, 7, 8, 9, 10, 15 from SarVision’s 2010 land cover map. 5. Water bodies Large lakes and large rivers. As identified up by SarVision’s 2010 land cover maps 6. Oil palm plantations in 2010 Planted or recently cleared industrial scale oil palm plantations as of year 2010. Industrial Oil palm plantations in 1990-, 2000-, and 2010-eras were manually digitized in ArcGIS 10 by visual inspection of >150 Landsat satellite images downloaded from the Global Land Survey database (http://earthexplorer.usgs.gov/). Industrial-scale plantations were readily identified as large geometrically-shaped areas with distinctive homogeneous spectral signatures characteristic of monoculture stands. We digitized any area planted with or being cleared for oil palm. Imagery acquired at earlier dates from the main key dates were often required to verify clearing and planting because newly cleared plantations (<1yr since planting) is usually easiest to detect using Landsat imagery. 7. Industrial timber plantations Planted or recently cleared industrial scale timber plantations as of year 2010. Industrial Oil palm plantations in 1990-, 2000-, and 2010-eras were manually digitized in ArcGIS 10 by visual inspection of >150 Landsat satellite images downloaded from the Global Land Survey database (http://earthexplorer.usgs.gov/). Industrial-scale plantations were readily identified as large geometrically-shaped areas with distinctive, homogeneous spectral signatures characteristic of monoculture stands. We digitized any area planted with or being cleared for rubber or Acacia mangium). Imagery acquired at earlier dates from the main key dates were often required to verify clearing and planting because newly cleared plantations (<1yr since planting) is usually easiest to detect using Landsat imagery. 8. severely degraded logged forests This class includes natural old-growth forests that have become so severely degraded that they no longer resemble the spectral signatures of forests in class 1 or 9. These forests are primarily found in east Kalimantan, and elsewhere only occur in Step 1: Similar to the process for generating class 4, we first generated a broad vegetation cover map including agroforests/forest regrowth/very degraded logged forests using the SarVision 2007 and 2010 landcover map. Merged class 8 (forest mosaic, 2 9. Logged forests small areas in Sabah, Sarawak and south Kalimantan. In east Kalimantan these forests have been burnt severely twice in March-April 1983 and March-April 1998 (i.e. during the two most intense El-Niño fire pulses on record, also declared national disaster in Indonesia). This forest class shows little sign of regenerating towards tall forest, probably because of invasion by flammable grasses. fragmented or degraded forest) from SarVision’s 2007 land cover map with class 2 (woodland) and class 13 (open forest) from SarVision’s 2010 landcover map. Medium to tall old-growth natural forests that have been logged by the timber industry using heavy machinery and networks of logging trails. Open to closed canopy: Includes Lowland and montane dipterocarp forests and tall closedcanopy peat forests. Step 1: Generated a forest cover map for year 2010 using SarVision 2010 landcover map. Merged classes: 3 (pole peat forest), 6 (closed forest), 11 (riparian forest) and 17 (peat-swamp forest close canopy). Step 2: Overlaid the logging road network onto this broad vegetation map and recoded this class into severely degraded logged forests in areas where there were logging roads. Step 2: transformed the 1990-2000-2010 logging road network (indicating mechanized logging) into a road density index (km/km2) of 1x1km grid cell (search radius 5km). Step 3: overlaid the road density map onto the forest cover map generated in Step 1 and recoded forest into logged forests in areas where road density > 0. (If road density = 0, see class 1). Climate and topographical variables Four least correlated climatic variables were used: temperature seasonality; temperature annual range; annual precipitation; and precipitation seasonality. These variables, as well as elevation were downloaded (at 30 arc-second resolution) from WorldClim, ver. 1.4 dataset (http://www.worldclim.org). A ruggosity layer was generated from the elevation data using the surface area and ratio tool from Jenness (2012). Two river layers were generated: Firstly, river density was calculated using kernel density tool in ArcGIS 10 with data sourced from HydroSHEDS (http://hydrosheds.cr.usgs.gov/index.php) (Lehner et al., 2006). Secondly, a major river vector file digitised from landsat images was used to calculated the euclidian distance from a major river. Accessibility Two accessibility layers were calculated using least-cost functions (path distance in Arc GIS 10.0); using population data as the source, ‘time distance’ data as the cost/impedance (based on time given to cross 1km via boat/road or walking) and elevation information as the surface (as this varies time to cross 1km). The first accessibility layer was calculated with a single threshold population density (10 or more people). The second layer used weighted sums, i.e. a 'weighted' accessibility layer calculating and summing path distance for cells with 1 or more person, 2 or more people, 3 or more people and so on up to 6,000; meaning cells that were accessible more easily by more people get a higher score. Socio-economic and Cultural factors Socio-economic and infrastructure data layers consisted of: constructed impervious surface 2010 density layer (Elvidge et al., 2009); poverty index layer (Elvidge et al., 2009); and human population 2007 density layer (estimated number of people per 1km2) from LandScan 2007TM (Bright et al., 2008). From the LandScan 2007 TM, we calculated settlement density for cells with 10 or more people per grid (using kernel density function). We generated a road density index from digisited 1999 to 2002 road data (using a line density function) see (Wich et al., 2012) for more details. 3 Cultural factors consisted of ethnic group identity, and representation of major religions as a proportion of the population. To incorporate cultural variation we used a digitized broad ethnic group map for Borneo (Sellato, 1989). Groups included: central-northern groups; Dusun and north-eastern groups; Iban and Ibanic groups; Kayan and Kenyah groups; land Dayak and western groups; Malay groups; Ngaju and Barito groups; Nomadic groups and an unknown category. Finally, as religion was a dominant factor influencing forest use and perceptions in a concurrent study (Meijaard et al., 2013) we used percentage data of population at provincial- or district-level resisted as Christian and Muslim. We obtained provincial- or district-level proportions of the population that were registered as Christian and Muslim, from Government Statistical departments from online sources (e.g., http://kalteng.bps.go.id/GIS.html) and hard documents (BPS-KalSel, 2009; BPS-KalBar, 2011; BPS-KalTim, 2011) which were then imported into ArcGIS 10.0. Appendix S2: Observed orangutan killing localities across Kalimantan Fig. S1 Location of the 512 villages surveyed within Kalimantan, Indonesia (in 2008-2012), indicating those with one or more reported orangutan killings around village (red dot; n = 116) and those with no reported cases (blue dot; n = 396), shown on a base map of 2010 Landover classes and protected areas (cross hatch). 4 Appendix S3: Subset data for conflict and killing models Fig. S2 Location of the 245 sampled villages (in 2008-2012) in the subset data in Kalimantan, Indonesia whose respondents had seen an orangutan around the village in the year prior to being surveyed, and those villages that had reported orangutan killings (red dot). Subset data were extracted for those villages that had reported orangutan sightings around their village in the year prior to the survey. Of this subset, eighty (33%) villages had reported incidences of human-orangutan conflict. The human-orangutan conflict model with the subset data (n=245 villages; Fig. S2) performed very well between the predicted and observed responses (Fig. S3a); performing slightly higher than that of the same model based on the full (principal) dataset (Fig. 2b). High conflict likelihood was predicted for villages that: were nearer to severely logged forest; that had very low or high road densities in regions and with greater temperature seasonality (Fig. S4a). These variables were the top three most important for this model (Table S2; Fig. S4a) and for the full dataset model (Table 2; Fig. 3b). Settlement density was seemingly important for the subset data model (Table S2; Fig. S4) with lower densities contributing towards higher conflict likelihood (Fig. S4a). Similar to the full dataset model, the mapped output of the subset data model predicted higher likelihoods of conflict in regions 1, 6 and 4; but with moderate conflict risk probability in and around region 5, rather than high risk as per predicted for the full model predicted (Fig. S5a). 5 (b) 0.50 0.6 0.5 0.1 0.0 0.2 0.5 0.3 0.4 predicted 1.5 1.0 predicted 2.0 0.7 2.5 0.8 (a) 0.98 0.0 0.5 1.0 1.5 2.0 2.5 3.0 0.0 0.5 1.0 1.5 2.0 2.5 3.0 observed observed Fig. S3 Correlations of goodness of fit and plots between observed and predicted responses for the two Boosted Regression Tree models that used the subset data of 245 villages for: (a) likelihood of human-orangutan conflict risk; and, (b) likelihood of orangutan killings occurring within one year prior to the surveys, by anyone in the village. Of the subset data (n=245 villages), 87 (36%) of the villages had reports of someone killing orangutan around their village within the year prior to the survey. The model for orangutan killings for the subset data had moderate performance, similar to the full dataset model (Fig. S3b; Fig. 2c). Paralleling the full dataset model, killing occurrence likelihood increased with distance from village to oil palm plantations (at around 6 km or more from plantations), which was by far the most important variable, along with other land cover types (at around 2 km or more) and in areas with higher surrounding orangutan habitat (Table S2; Fig. S4b). The mapped output of this model infers extensive high killing likelihood in region 5 (Fig. S5b). High killing likelihood was however, less extensive in this model than in that of the model of the full dataset (Fig. 4c) with many areas (e.g. parts of regions 1, 4, 5 and 6) predicted as moderate instead of high likelihood of killings around villages within these areas. Table S2 Boosted Regression Tree models top ten most influential spatial variables (for models based on the 245 village subset data), their percentage contribution (%) and the summed total percentage of these top ten variables. Likelihood of humanorangutan conflict % svlogged_m 12 road_d 11 temp_season 9 settlemt_d 7 indtim_m 5 pa_m 3 oilpalm_m 3 access_sum 3 elevation 3 intact_s 3 Total 59 Likelihood of orangutan killings around villages % oilpalm_m 52 ou_s 14 otherlc_m 13 elevation 4 logged_s 3 prec_annra 2 pa_m 2 access_sum 1 christian 1 svlogged_m 1 Total 93 For explanation of codes see Table 1. 6 (a) Likelihood of human-orangutan conflict (b) Likelihood of orangutan killings around the villages Fig. S4 Plots for the top four most influential spatial predictor variables within the two subset data Boosted Regression Tree (BRT) models that relate to perceptions of human-orangutan interactions. Plots show the effect of spatial predictors on the respondent’s response variable with relative importance values for each variable shown in parentheses on the x-axis. Fig. S5 Mapped outputs from the two subset data Boosted Regression Tree models (based on interviews from 245 villages) overlaid with protected areas (hatched) and provinces (black line). Figure shows the predicted likelihood of a given response mapped as tertiles for the likelihood of: (a) human-orangutan conflict risk; and, (b) orangutan killings occurring in the year prior to the surveys by anyone in the village. Table S3 Correlation matrix between the observed responses of the 6 boosted regression tree models for the full dataset (n=512 villages) Orangutan sighting Human-orangutan conflict Orangutan killing around village Orangutan killing by respondent Orangutan population change over last ten years Population change in next ten years Orangutan sighting 1 Humanorangutan conflict 0.371 1 Orangutan killing around village 0.456 0.150 1 Orangutan killing by respondent 0.111 0.148 0.232 1 Population change over last ten years -0.205 -0.112 -0.131 -0.025 1 Population change in next ten years -0.148 -0.064 -0.076 -0.050 0.793 1 7 Appendix S4: Correlations between the spatial predictor variables To understand the relationships between variables, correlation coefficients were calculated for each pair of spatial variables using Pearson’s correlation coefficient (see Table S4). Several of the spatial variables were strongly correlated including a very strong negative correlation between the proportion of the population that is Christian and those who are Muslim. Strong positive relationships occurred between: the 'weighted' accessibility layer with summed cover of intact forest, distance to mangrove, and distance to oil palm; ruggedness and summed carbon values suggesting as ruggedness increases so does carbon stock; and, between distance to mangrove with distance to oil palm. A strong negative relationship was found between summed carbon values and summed other land cover types, suggesting that with increasing carbon stock there is a decrease in the extent of other land cover types. Sixty moderate correlations were found among other variables and can be seen in Table S4. Table S4: Correlation matrix between the 39 spatial predictor (explanatory) variables 8 REFERENCES Baccini, A., Goetz, S.J., Walker, W.S., Laporte, N.T., Sun, M., Sulla-Menashe, D., Hackler, J., Beck, P.S.A., Dubayah, R., Friedl, M.A., Samanta, S. & Houghton, R.A. (2012) Estimated carbon dioxide emissions from tropical deforestation improved by carbondensity maps. Nature Climate Change, 2, 182-185. BPS-KalBar (2011) Kalimantan Barat Dalam Angka. Kalimantan Barat in Figures 2011. Badan Pusat Statistik Provinsi Kalimantan Barat, Pontianak, Indonesia. BPS-KalSel (2009) Kalimantan Selatan Dalam Angka. Kalimantan Selatan in Figures 2009. Badan Pusat Statistik Provinsi Kalimantan Selatan,Banjermasin , Indonesia. BPS-KalTim (2011) Kalimantan Timur Dalam Angka. Kalimantan Timur in Figures 2011. Badan Pusat Statistik Provinsi Kalimantan Timur, Samarinda, Indonesia. Bright, E.A., Coleman, P.R., King, A.L. & Rose, A.N. (2008) LandScan 2007. Oak Ridge National Laboratory, Oak Ridge, TN. Elvidge, C.D., Sutton, P.C., Ghosh, T., Tuttle, B.T., Baugh, K.E., Bhaduri, B. & Bright, E. (2009) A global poverty map derived from satellite data. Computers & Geosciences, 35, 1652-1660. Gaveau, D.L.A., Sloan, S., Molidena, E., Yaen, H., Sheil, D., Abram, N.K., Ancrenaz, M., Nasi, R., Quinones, M., Wielaard, N. & Meijaard, E. (2014) Four Decades of Forest Persistence, Clearance and Logging on Borneo. PLoS ONE, 9, e101654. Hoekman, D.H., Vissers, M.A.M. & Wielaard, N. (2010) PALSAR Wide-Area Mapping of Borneo: Methodology and Map Validation. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 3, 605-617. Jenness, J. (2012) DEM Surface tools. Jenness Enterprises. Flagstaff, AZ,USA. http://www.jennessent.com/arcgis/surface_area.htm>. Lehner, B., Verdin, K. & Jarvis, A. (2006) HydroSHEDS. World Wildlife Fund US, Washington, DC. Meijaard, E., Abram, N.K., Wells, J.A., Pellier, A.-S., Ancrenaz, M., Gaveau, D.L.A., Runting, R.K. & Mengersen, K. (2013) People’s Perceptions about the Importance of Forests on Borneo. PLoS ONE, 8, e73008. Sellato, B. (1989) Hornbill and Dragon, arts and culture of Borneo, Ragabooks, Palo Alto, CA, USA. Wich, S.A., Gaveau, D.L.A., Abram, N.K., Ancrenaz, M., Baccini, A., Brend, S., Curran, L., Delgado, R.A., Erman, A., Fredriksson, G.M., Goossens, B., Husson, S.J., Lackman, I., Marshall, A.J., Naomi, A., Molidena, E., Nardiyono, Nurcahyo, A., Odom, K., Panda, A., Purnomo, Rafiastanto, A., Ratnasari, D., Santana, A.H., Sapari, I., van Schaik, C.P., Sihite, J., Spehar, S., Santoso, E., Suyoko, A., Tiju, A., Usher, G., Atmoko, S.S.U., Willems, E.P. & Meijaard, E. (2012) Understanding the Impacts of Land-Use Policies on a Threatened Species: Is There a Future for the Bornean Orangutan? PLoS ONE, 7, e49142. 9
© Copyright 2026 Paperzz