Beyond Climatic Variation: Human Disturbances Alter the Effectiveness of a Protected Area to Reduce Fires in Tropical Peatlands of Sumatra, Indonesia

The occurrence of fires has frequently been used to highlight environmental hazards at regional and global scale, and as a proxy for the effectiveness of protected areas. In contrast, the mechanism behind wildfire dynamics in tropical peat land protected areas had been poorly addressed thus far. Our study provides a novel application of assessing fire patterns from a tropical peatland protected area and surrounding landscape. We investigated the importance of both climatic factors (top-down mechanism) and human interventions (bottom-up mechanism) on fire occurrences through analyzing 15-year (2001 - 2015) LANDSAT and MODIS images of the Padang Sugihan Wildlife Reserve (PSWR). Fire density along side road and canal construction were analyzed jointly together with the monthly and annual precipitation, and evidences of climatic anomalies. The reserve was effective in limiting fire occurrences from surrounding landscapes only in wet years. We revealed that peat fire patterns in the protected area and the landscape matrix emerged beyond climatic factors, and the distance from canal system could explain the fire occurrences. Our results show that it is essential to address processes at a landscape level, particularly at the surroundings of the reserve, in order to increase the effectiveness of fire protection, including the development of fire-prone classes maps.

Fire has a long historical relationship with humans [1]. Although less significant than 2 other causal factors, fires are frequently used as an important indicator to evaluate the 3 effectiveness management of protected areas [2,3]. A common approach to manage fire 4 in protected areas is applying active fire management or prescribed burning [4][5][6][7] which 5 aims to reduce fuel availability for preventing and controlling wildfire [8]. Nowadays, a 6 paradigm shift resulted in managers purposely burning grassland and forests to 7 maintain the ecological mechanisms which drive ecosystem dynamics and diversity [7,9]. 8 Anthropogenic and natural factors lead to different patterns of fire occurrences in 9 protected areas across various ecosystems. Fire density was found to be two times 10 higher in non-protected areas than within protected areas in Myanmar [10] and 11

PLOS
1/15 Amazonian regions [2]. It has been shown that human intervention influences a 12 protected areas susceptibility to disturbance. At a global scale, forest loss rates in 13 protected areas is associated with high proportions of agricultural land in the country 14 [11]. Managing anthropogenic factors which include fewer road construction, less human 15 impact mechanisms [2] as well as fire-free land management [12] in the reserves has 16 shown their effectiveness in reducing fire-driven deforestation. In contrast to fire 17 occurrences in tropical areas, natural mechanisms caused higher fire density in 18 protected areas than in non-protected areas in West and Central Africa [13]. More fires 19 occurred in 59 percent of the area where deforestation rates dropped between 2000 and 20 2007, since more fuel was available for ignition [12]. Here, controlled ignition and active 21 fire management are required to mitigate fuel availability. 22 Fire is a critical attribute to peatland but rarely occurred in the remote forest until 23 the last 3000 years as anthropogenic factors started affecting peatland [14]. Peatlands 24 in South-east Asia holds 26 million ha which represents 69% of Tropical Peatlands 25 globally [15][16][17]. In Indonesia, peatlands are vital ecosystems, covering 18-20 million 26 hectares, or 10% of terrestrial area which significantly contribute to primary sources of 27 wood and livelihood of local people [18]. 28 In the last two decades, Indonesia has been experiencing a drastic shift in peat-land 29 dynamics, from being frequently, from being frequently inundated and moist-ecosystem 30 into human-made drained-ecosystem. This ecosystem change brings consequences for 31 the frequent occurrences of peat-fires and subsequent health, environmental and 32 biodiversity problems [17]. The recent fires of 2015 in Indonesia were the highest since 33 the 1997 megafires and almost the whole Sumatra island was engulfed by smoke, while 34 South Sumatra province holds the second highest number of hotspots amongst the 35 Indonesian provinces during 2015 [19]. Particularly in Sumatra island, the combination 36 of deforestation from adjacent areas [20], the rapid expansion of palm oil plantation, 37 and the use of fire for land preparation [21] significantly increase threats in protected 38 areas as well as biodiversity conservation of selected species [22]. Studies on the 39 effectiveness of protected areas in a peatland to reduce wildfire are still limited. Fire in 40 Indonesian peatland has been intensively studied through remote sensing data to 41 predict fire effects [23][24][25], peat hydrology [15], fire database management [26], effect 42 of fire on bio-physical attributes attributes [24,27,28], and peat restoration planning 43 [29]. Nevertheless, studies with emphasis on protected areas are still rare. 44 Fire management in Indonesia has different characteristics from the majority 45 paradigm. While the use of fire is encouraged in savannas [7,9], efforts on fire 46 prevention are still relevant to address ecological issues in Indonesia, particularly in 47 peatlands. In both cases, understanding causal mechanism of fire occurrence is 48 important. The patterns and causal factors of fire in a protected area might help reveal 49 more information to guide effective fire management. In addition, current studies often 50 overlook the importance of peatland protected areas on reducing fires in the landscape. 51 We aim to gain insight into pattern and causal factors of fires through the 52 comparison of fire occurrences within and surrounding the Padang Sugihan Wildlife 53 Reserve (PSWR), a protected area in South Sumatra which is dedicated for biodiversity 54 conservation. We observed a 15-year period of fire occurrences data to provide a basis 55 for evaluating the effectiveness of peatland protected area to reduce fire intensity from 56 the surrounding landscape. river serves as the border to the east. To the south of the reserve, lies Butung river, and 66 a canal is built to the north (Fig 1). Administrative boundary maps of South Sumatra 67 (http://www.bakosurtanal.go.id/peta-rupabumi/) and Padang Sugihan Wildlife Reserve 68 maps were used in this study and obtained from BKSDA. 69 Fire detection and Environmental Studies 70 Fire detection was processed using Active Fire data provided by the Moderate

71
Resolution Imaging Spectrometer (MODIS). It provides the geographical position of fires 72 within 24 hours of monitoring [13,30] as well as 1-2 day return interval which is useful 73 to monitor the change of fire detection at any given location [30]. We selected MODIS 74 active fire data in the period between 2001 and 2015 with a confidence level of 80%. 75 Since Indonesia is influenced by El Nino / Southern Oscillation (ENSO), we categorized 76 the fire data to wet and dry periods according to NOAA Climate Prediction Centre [31]. 77 To compare fire occurrences within and surrounding the reserve, we calculated fire 78 frequency within the reserve and within 10 km surrounding its border. A buffer zone of 79 10 km was chosen because it is equal to the longest distance from the border to the 80 reserve center point (Fig 1) and is commonly used to assess protected area 81 effectiveness [32]. The buffer size was 1,781.68 km 2 , combined with PSWR (881.48 km 2 ) 82 it occupies a total area of 2,663.16 km 2 . We divided our study site into 1 x 1 km 2 grid 83 cells to identify whether fires can occur in the same area. Recurrent incidences of fire in 84 certain cells over the observed period are classified as repeated fire. 85 We plotted fire occurrences and daily precipitation to extract relational patterns 86 between the two factors. To understand the impact of reserve border in reducing fire 87 density, we compared fire frequency within the reserve and 1 km surrounding it. We Kernel Density Estimation using ArcGIS 10.

91
To understand the relation between fire occurrence and precipitation, we obtained   The role of precipitation dynamics on fire emergence was clearly shown with a 114 functional delay of monthly precipitation decrease in July followed by increasing 115 numbers of wildfires from August to September (Fig 2). Additionally, the average 116 monthly precipitation and number of rainy days shows associations with number of 117 hotspots during dry years, wet years as well as for all years during 2001-2015 (Table 1). 118 The fire occurrence in PSWR and its surrounding area was most frequent during the 119 dry season from July to November. Only a few anomalies of fire occurrence were 120 observed during May and December. Fires (n = 180) were able to emerge when no rain 121 occurred or up to 37 consecutive dry days happened prior to the ignition (median = 8 122 days). Whereas, the duration of fire (n=180) to maintain burning was between 1 -14 123 days with a median of 1 day.

124
During the observed period, most fire incidences, within and surrounding PSWR, 125 occurred only once. We found fewer grid cells in which fire occured twice or more, and 126 only few were burnt more than seven times. It was clearly shown that fire frequencies 127 was significantly lower within the reserve. In the 15 year period, a very small portion 128 from the surrounding of PSWR was burnt annually while within the reserve, only 4 129 repeated fire occurred at most (Fig 3).  In the wet years, fire density was higher surrounding PSWR, whereas in dry years 131 fire occurs proportionally within and surrounding the reserve. Fire density decreased 132 from the reserve border to the core during the wet years, whereas an unclear pattern 133 was shown in dry years. During wet years the density of fire within 1 km from the 134 reserve border was low, and at a gradually further distance fire occurrences were found 135 to be increasingly rare. In addition, fire density outside the reserve tended to increase 136 with further distance -up to 6 km from the border (Fig 4). Conversely, in dry years we 137 could not see the gradual reduction of fire occurrences from the border to both within 138 and surrounding the reserve. In dry years, the irregular pattern of fire density increment 139 within PSWR was similar to which surrounding the reserve. However, a relatively lower 140 density of fires was identified near the reserve border.

141
During dry years, distance to canals, distance to roads, NDVI, and EVI have a 142 profound effect on fire presence probability, whereas in wet years the most critical factor 143 affecting fire occurrence is the distance to roads and distance to the canal ( Table 2). decrease of fire incidents as the distance from canals and roads increase. Meanwhile, the 150 distance from reserve border has a less clear impact on fire frequency (Fig 4).

151
Our binomial generalized linear model clearly shows a sudden shift of reduced size of 152 low fire-prone probability class within the PSWR from wet years (36%) to dry years 153 (1%). In contrast, a sharp inclination was shown for the middle fire-prone probability 154 class from wet years (62%) to dry years (95%) within the PSWR (Fig 5). The 155 proportion of each class was not changed significantly for the surrounding PSWR from 156 wet to dry years. However, during the dry season, none of the surrounding areas of 157 PSWR had the low fire-prone probability class. Our findings highlight that surrounding 158 of the PSWR is continuously under the threat of fire, while within PSWR can protect 159 some areas during wet years but not during dry years. Predicted map of fire-prone probability class and its proportion for wet years (left) and dry years (right) from the binomial generalized linear model. The probability of fire occurrence is represented by different colors ranging from red (high) to blue (low). The area size of each fire-prone probability classes (below) compared in both wet and dry years.    5). During wet years with more 180 precipitation and rainy days, the reserve area was able to limit fire. However, in dry 181 years, when the rainfall is low the presence of protected area borders as a mean of 182 land-use management was ineffective to prevent burning events [34], causing an 183 increase of fire occurrences within and surrounding the reserve. 184 Our findings confirm that the accumulated rainfall and the length of the dry season 185 influence the annual area burned in protected areas [35]. we provided here fine-scale  peat area to fire. Yet our study has not explored whether any seasonal changes of NDVI 202 influence the susceptibility to fire due to limited amount of clear satellite images (i.e less 203 than 10% cloud cover) obtained per-season. We concluded that due to the lack of 204 multitemporal NDVI series, we could not ascertain vegetation phenology. Hence, the 205 complex interactions between vegetation, land use and climate characteristics cannot be 206 explained [40].

207
In our study, the approach to use NDVI and EVI as vegetation indexes is limited by 208 the relationship between fire; these indexes cannot be understood as a cause and effect 209 relationship. We collected environmental variables (including NDVI and EVI) in the 210 same year of the fire without considering the temporal relationship of fire occurrence.

211
However, environmental variables would be better observed prior to a burning event to 212 aid in determining cause-effect relationships between vegetation indexes and fire 213 occurrences. Vegetation seasonal variability which indicates the shift from moist to dry 214 conditions can help to identify the primary bioclimatic drivers related to fuel dynamics 215 [40]. Implementing high-quality resolution of the multi-spatiotemporal image to depict 216 the role of vegetation biomass will provide a substantial contribution to understanding 217 the spatial pattern of fire [41]. simulate both top and below-ground fire spreading mechanisms, further exploration of 227 farmers decision-making process to burn land for agricultural conversion is required, as 228 fire is employed as a tool in agricultural-conflict scenarios pertaining to land 229 rights/ownership [42][43][44]. 230 We rarely found repeated fire in an area during the 15-year period and very few 231 recurrent fires in the same grid cell. This phenomenon implies that the peatlands 232 self-restored within the 15 year period and intermittent fire was avoided. It is commonly 233 believed that isochronal fire is a function of land conversion into agricultural uses [38]. 234 The absence of recurrent fires means that degradation of the PSWR only occurred 235 recently. Commonly fire use by locals only occurs when initially opening/preparing land 236 during the onset of the planting period [42][43][44]. When the initial fires occurred 237 anthropogenic activities were also present which effect flooding rates in the grid cells; 238 therefore cessation of land conversion prevented recurrent fires from being ignited. [45]. 239 Extension of observations, to exceed a plantation rotation may provide better 240 understanding of fire use for land conversion and resulting recurrent fire.

241
Protected area ineffectiveness also stems from the presence of canal networks, whilst 242 such networks provide access they also drain peatlands and increase the likelihood of it 243 burning [36][37][38]. This evidence shows the bottom-up mechanism. Canal construction as 244 a part of peatland conversion is commonly followed by the construction of drainages, 245 and serves as a proxy of human disturbance to predict fire in the PSWR and 246 surrounding landscape. The patterns of the fire illustrated the canal scars still play 247 essential roles on fire during dry years as waterways give access to fire, representing the 248 human-induced activities [46,47] including forest conversion into plantations [48]. 249 Furthermore, canals and roads as a proxy of human presence serve as valuable 250 predictors of fire occurrence in our study area, as well as elsewhere [2,49]. Our approach, 251 however requires additional land cover and land use analysis [50][51][52] to assess causal 252 factors of fire in the surrounding landscape of PSWR. Recent perspectives pertaining to 253 the role of hydrological drought on fire concurrences indicate additional studies on such 254 dynamics in disturbed and undisturbed areas may influence our findings. Additionally 255 research should seek to ascertain the motivations of anthropogenic land burning [53] to 256 aid in understanding humans role in dry year fire amplification [2,46,54].

257
Fire presence within the reserve during wet and mostly dry years indicate the 258 reserves biodiversity is under serious threat through habitat destruction and isolation 259 within their boundaries [49,55]. Responses of biodiversity, such as mammals, to and 260 following fire incidences provide insight of the reserves effectiveness in maintaining such 261 biodiversity, which was unfortunately beyond the scope of the present study. A 262 particular emphasis on the behavioral response of Sumatran elephants as a flagship 263 species in PSWR, using GPS collaring [56]and population projection studies [57] 264 would gain insight into the effect of fire on endangered species conservation.

265
Furthermore, population viability analysis for the elephant in the reserve and 266 surrounding area using a modeling approach [22,52,58] will provide guidance for 267 management alternatives, currently lacking such data.

268
This study provided insights on spatial and temporal fire occurrence predictors, 269 including human interference and climatic variations. Nevertheless, knowledge on the 270 detailed mechanism of fire occurrence and its causal factors remain incomplete.

271
Individual/Agent-based modeling is one promising approach to understanding the 272 complex phenomenon of wildfire, as well as predicting fire-prone areas in the future 273 [59][60][61][62][63]. Development of such models would be beneficial for applied science 274 development on fire ecology and management. Predictions from such dynamic models, 275 incorporating bottom-up and top-down processes would allow these complex processes 276 to be incorporated, providing dynamic information applicable to management fire 277 mitigation policy development. Also since vegetation and social conditions are highly 278 dynamic in the tropical peatlands [42], using such dynamics model with relatively short 279 temporal scale will enhance the prediction .  [20]. 283 However, our study shows that a protected area in a peatland ecosystem was not devoid 284 of dry season fires. We also noticed the surrounding landscape becomes critical for the 285 protected areas if adjacent areas could not hold fire occurrences. The surrounding area 286 is key in mitigating fire, particularly during dry years. In addition, raising awareness of 287 people surrounding the reserve to use fires wisely and practice sustainable peatland  The PSWR has crucial roles on biodiversity conservation in Indonesia as habitat for 293 the endangered species of Sumatran Elephant. This subspecies of elephant has a large 294 homerange [57] which consequently needs various habitat types and larger habitat size. 295 The landscape approach of considering composition and quality of land surrounding a 296 protected area has emerged recently in biodiversity conservation [22], and urgent 297 implementation has been recommended. Encouragement of privately protected areas 298 [64,65] and village-based protected areas such as a sacred forest [66,67] in the 299 surrounding area of the reserve may enhance the effectiveness of biodiversity 300 conservation and fire mitigation. In case of fire incidents within the PSWR, management 301 should ensure safe areas from fire as a wildlife refugia system are present [68,69].

302
Since the PSWR is also surrounded by canal systems and rivers for local 303 transportation access to the reserve is relatively easy. Suggestions or the proposal to 304 change the reserve into mixed-used protected areas to deal with pressure from 305 agricultural and timber extraction [70] should be carefully considered. Attention to 306 evaluate the effectiveness of the BKSDA Sumatera Selatan to manage the reserve 307 [71][72][73][74] with particular emphasis on fire management;should be considered; prior to 308 contemplating changing the status of protected areas into other protection types or 309 other land use types. Strengthening the ability of rangers to detect and handle fires 310 when they occur, will also enhance the effectiveness of protected areas' management.