Introduction

Vegetation fires in tropical regions have multiple causes and consequences. Fires are widely used in land management and often, however started, develop into uncontrollably raging wildfires. Apart from local impacts on ecosystems and people, smoke from fires can have adverse impacts on human health at the regional scale during haze episodes (e.g. Chokkalingam et al. 2006; Dennis et al. 2005; Qadri 2001). Fires also constitute a significant contribution to greenhouse gas emissions. Carbon emissions from biomass burning in tropical regions have been estimated to equal up to 40% of global emissions from fossil fuels (Cochrane 2003).

Most of the main fire regions in the tropics have clear seasonal climates with long and distinct dry seasons (e.g. Northern Australia, Southern Africa and Sahel region). Insular Southeast Asia, on the other hand, has a relatively humid climate all year round. Undisturbed humid tropical forests are considered to be very resistant to fire, experiencing rare fires only during extraordinary dry periods (Uhl et al. 1988). Charcoal evidence suggests that fires have taken place in insular Southeast Asian forests during periodic droughts for thousands of years and references to infrequent vegetation fires can be found in historical documents since the late 19th century (Goldammer 2006). However, the use of fire in land clearing has escalated dramatically since the 1980s and lead to annual biomass burning with occasional severe fire episodes (Goldammer 2006, Silvius and Diemont 2007).

Continuing degradation and conversion of natural ecosystems in insular Southeast Asia (Fuller et al. 2004; Kauppi et al. 2006; Langner et al. 2007) increases fire occurrence mainly for four reasons. Fire is extensively used (1) in agricultural field preparation by small-holder farmers and (2) in land clearance by plantation companies (Bowen et al. 2001; Chokkalingam et al. 2006; Ketterings et al.1999; Simorangkir 2006). Both of these types of fires also frequently result in wildfires during dry spells. (3) Degraded and logged over forests are more susceptible to fire than natural forests due to drier microclimate and increased amount of dry material (Cochrane 2001; Siegert et al. 2001). (4) Drained peatlands become extremely susceptible to fire during drier periods (Rieley and Page 2005).

The environmental effects of fires in insular Southeast Asia are exacerbated by the considerable numbers of fires taking place in peatlands. Fires in peatland areas not only disturb unique ecosystems and impede ecological functions of peatlands (Rieley and Page 2005) but also cause haze and globally significant carbon emissions (e.g. Heil et al. 2006; Page et al. 2002). Carbon emissions from peatland fires are aggravated by drainage typically connected to peatland development. Drainage leads to exceptionally low water table levels during dry periods and enables combustion of the top layer of peat soil. The most alarming example of unsuccessful peatland development in Southeast Asia is undoubtedly the former Mega Rice Project in Central Kalimantan. It aimed in creating a large rice cultivation area but instead turned over 1 Mha of land into drained unmanaged peatlands experiencing yearly fire activity (Rieley and Page 2005).

Fire regime is a term used in vegetation fire science to categorize fire history in a landscape based on characteristics such as the spatial and temporal patterns of fire occurrence or intensity of the fires. The highly anthropogenic nature of biomass burning in insular Southeast Asia results in a range of fire regimes with greatly varying characteristics. Three major types of fire regimes have been identified (Bowen et al. 2001; Miettinen and Liew 2009): (1) small-holder burning, (2) large-scale land clearance and (3) wildfires (Fig. 1).

Fig. 1
figure 1

Examples of three major fire regimes in insular Southeast Asia: a small-holder burning in West Kalimantan, b large-scale land clearance in Riau and c wildfires in Central Kalimantan (Miettinen and Liew 2009). Burnt areas are visible in dark color and detected active fires marked as yellow dots. Satellite data SPOT 4 HRVIR and SPOT 5 HRG images (RGB:432; SPOT image © 2006 CNES)

Full size image

Both types 1 and 2 involve intentional use of fire, mainly for land clearance and soil preparation for agricultural purposes. However, whereas the first type (small-holder burning) is characterized by numerous individual fires resulting in a mosaic of small burnt patches rarely larger than 0.25 km2, the latter (large-scale land clearance) manifests itself by large continuous burnt areas in sizes of up to tens of square kilometers (Bowen et al. 2001). These fires are all purposely started by either small-holder farmers or plantation companies and burning is controlled, at least as long as the fire stays within the intended area. In satellite-based active fire (i.e. hotspot) detection, small-holder burning typically results in sporadically distributed single hotspots, whereas large-scale land clearance creates large clusters of hotspots (Fig. 1).

Fires burning out of control are considered wildfires (type 3). They may have started either accidentally or intentionally (arson), or escaped from intentional land clearance fires into areas that were originally not intended to be burnt. These fires usually result in irregularly shaped burnt areas. In active fire detection, wildfires create clusters of hotspots in varying shapes and sizes (Fig. 1).

The highly anthropogenic origin of vegetation fires in insular Southeast Asia combined with the natural characteristics of the region forms a basis for a complicated relationship between fire regimes and physical constraints like climate, land cover type and occurrence of peat soil. Numerous authors have investigated the variation of fire regimes and the underlying causes of fires from social and political perspective, emphasizing the influence of land management policies on fire activity in this region (e.g. Bowen et al. 2001; Chokkalingam et al. 2006; Dennis et al. 2005; Ketterings et al. 1999; Murdiyarso and Adiningsih 2006; Stolle et al. 2003). The effects of physical constraints on fire regimes have been studied less. However, it has been shown that interannual variations in regional climate affect the general severity of yearly fire activity considerably and that fire intensity varies between land cover types (Field et al. 2009; Langner and Siegert 2009; van der Werf et al. 2008). Furthermore, we believe that due to the varying fire vulnerability of natural, degraded and managed ecosystems on peat and mineral soils, land cover and occurrence of peat soil may greatly influence the distribution of fire regimes in this region.

The objective of this study is to analyze the influence of peatland and land cover on the occurrence and distribution of fire regimes in insular Southeast Asia. We analyze active fire data from 2008 (wet La Niña year) and 2009 (moderately dry El Niño year) with information on the extent of peat soil and land cover types to deepen understanding on the spatial and temporal patterns of vegetation fires in this region.

Materials and methods

Study area and climate

The study area includes Peninsular Malaysia and the islands of Sumatra, Borneo and Java (Fig. 2). This area is also known as the Sundaland. The tiny nation of Brunei in the northern coast of Borneo island is excluded from the analysis due to lack of data on the extent of peatlands. Also, the city state of Singapore was not included in the study area. Thereby, the study area is divided by two countries, Malaysia and Indonesia. It is home to more than 200 million people and has experienced rapid development during the past two decades. It must be noted, however, that more than half of the population lives in the island of Java.

Fig. 2
figure 2

Study area. Peatlands marked in dark gray. Note also the locations of two major cities (Pekanbaru and Palangkaraya) near extensive peatland areas. Precipitation statistics for these cities are given in Fig. 3

Full size image

Insular Southeast Asia has a relatively stable humid tropical climate. In general, the weather is hot and humid all year round, with a drier period of 23 months that, in the majority of the study area, occurs between June and October. The worst burning season closely follows the drier period. However, even during this period, rainfall exceeds evaporation in many parts of the region in normal years. Most of the more extreme dry seasons in the study area can be attributed to the El Niño-Southern Oscillation (ENSO; Corlett 2009).

ENSO is a periodic change in the atmospheric conditions and ocean temperatures in the tropical Pacific Ocean. Typically, El Niño (high ocean surface temperature in the Eastern Pacific) creates drier conditions in insular Southeast Asia, whereas La Niña (low ocean surface temperature in the Eastern Pacific) induces more rain in the study area. Of our two study years, 2008 was characterized by La Niña conditions and 2009 by El Niño conditions (National Weather Service 2010). Precipitation distributions for two towns situated in the most fire affected parts of the study area are presented in Fig. 3. It can be seen that, especially in Palangkaraya (Central Kalimantan province), El Niño conditions created a clear dry season between June and September. The consequences of El Niño are not as clear in Pekanbaru (Riau province) in 2009. Nevertheless, for three consecutive months from May to July, the monthly rainfall hardly exceeded 100 mm. This enabled high levels of fire activity.

Fig. 3
figure 3

Precipitation statistics for Pekanbaru and Palangkaraya in 2008 (black columns) and 2009 (gray columns). In Pekanbaru, the rainfall in 2008 and 2009 is rather similar. Note, however, the three consecutive months with low rainfall from May to July in 2009. In the precipitation statistics for Palangkaraya, a dry season between June and September 2009 is clearly visible

Full size image

The study area was originally covered by mostly tropical evergreen dipterocarp forests and had the highest biodiversity of all tropical regions of the world (Whitmore 1984). Other forest types include peatswamp forests in peatlands, heath forests in poor soils and various mountain forest types. Most areas of natural vegetation have now been either converted to agroecosystems or degraded; deforestation and land cover change continues making way for large-scale plantations and small-holder agriculture (Fuller et al. 2004; Kauppi et al. 2006; Langner et al. 2007).

Perhaps the most unique feature of the region is the broad extent of tropical peat soil. Around 40% of global tropical peatlands (15 Mha) can be found in the study area (Rieley and Page 2005; Selvaradjou et al. 2005; Wahyunto et al. 2003, 2004). Peatlands can be found mainly in the Northern and Southern coastal regions of Borneo and along the Eastern coast of Sumatra (Fig. 2). Peatlands cover around 15, 10 and 7 in Sumatra, Borneo and Peninsular Malaysia, respectively. Peatswamp forests have been shown to have several ecological functions in the region, and they contain unique flora and fauna (Rieley and Page 2005).

Active fire data

Active fire detection (i.e. hotspot detection) by satellite sensors is based on the detection of the thermal infrared (TIR) radiation emitted by fires. Hotspot detection is typically performed on a daily basis, and it can be considered as the most suitable and effective way to determine the seasonality, timing and interannual variation of fires at large scales (Eva and Lambin 1998). Moderate resolution imaging spectroradiometer (MODIS) active fire data were used in this study. The data were produced by the MODIS Rapid Response System (Davies et al. 2009) using the contextual fire detection algorithm, which utilizes 1-km resolution MODIS thermal bands (Giglio et al. 2003). Since the MODIS sensor is carried by two satellites (Terra and Aqua), it passes over each area four times a day (equatorial day passes around 10.30 for Terra and 13.30 for Aqua and night passes around 22.30 and 1.30, respectively). All fire detections from both Terra and Aqua satellites for the years 2008 and 2009 were used in this study.

Maps of the extent of peatlands

For Sumatra and Kalimantan, two recently published atlases by Wetlands International (Wahyunto et al. 2003, 2004) provided up-to-date information on the extent and locations of peatland areas at the scale of 1:700 000. For Malaysia, such recently published maps could not be found, and maps provided by the European Digital Archive of Soil Maps (Selvaradjou et al. 2005) were considered the best available source of information for delineation of peatlands. For Malaysian Peninsula, we used the 1970 1:800 000 Generalized Soil Map of Peninsular Malaysia, published by the Director-General of Agriculture, Peninsular Malaysia. For Sarawak, the 1968 1:500 000 Soil Map of Sarawak by the Sarawak Land and Survey Department was chosen. Finally, for Sabah, we used the 1974 1:250 000 Soils of Sabah-map by the British Government’s Overseas Development administration. Note that the island of Java is not covered with peatland maps since Java does not have any major peatland areas.

Land cover maps

Two different land cover maps were used in the study. A 500-m spatial resolution regional land cover map (Miettinen et al. 2008) produced with satellite data acquired during the first half of 2007 was used for mineral soil areas. Altogether, 250 MODIS Surface Reflectance Product images, Shuttle Radar Topography Mission (SRTM) 90 m version 3 digital elevation information (Jarvis et al. 2006) and several peatland distribution maps were used to produce this map. The classification scheme of 12 classes reflected the special characteristics of land cover in insular Southeast Asia (Table 1).

Table 1 Short description of land cover classes in the regional (Miettinen et al. 2008) and peatland (Miettinen and Liew 2010) land cover maps used in this study
Full size table

For areas on peat soil, a land cover map from a recent study on the peatlands in Sundaland was used (Miettinen and Liew 2010). This land cover map covered all the areas considered as peatland in the maps listed in the previous section. The peatland land cover map was produced using more than 120 high resolution (10–20 m) Satellite Pour l’Observation de la Terre (SPOT) 2, 4 and 5 satellite images acquired 2006–2008. The classification was based on visual image interpretation and resulted in a classification scheme of 10 classes specifically designed for Southeast Asian peatlands (Tables 1 and 2). The map was output in 100 m resolution.

Table 2 Land cover distribution in the peatlands of Sumatra, Borneo and Peninsular Malaysia (Miettinen and Liew 2010)
Full size table

Analysis

In addition to basic hotspot distribution and density statistics within the study area between land cover types on both mineral and peat soils, a value indicating the level of clustering of fire detections was calculated. The value was called hotspot proximity value, and it referred to the average number of active fire detections within a 5-km radius from any given hotspot. All hotpots detected within a calendar year (1st Jan–31st Dec) were taken into consideration when calculating the hotspot proximity value.

Results

General comparison of fire seasons 2008 and 2009

The total number of fire detections in the wet La Niña year (2008) was around 23 000, while in the drier El Niño year (2009), fire count surpassed 68,000 (Table 3). Sumatra and Borneo islands clearly dominated fire activity in both years. In 2009, these two islands together contained 95% of all active fire detections. The increase in fire numbers from 2008 to 2009 largely took place in the island of Borneo, which had nearly six times more fires in 2009 than in 2008. On Java, there was no difference in the level of fire activity between the two study years (Table 3; Fig. 4). This is likely due to the dominance of yearly small-holder land preparation burning: 78 and 76% (in 2008 and 2009, respectively) of fires in Java were detected in small-holder agriculture dominated mosaic and open land cover classes.

Table 3 Number of active fires detected, fire density and proximity values for 2008 and 2009
Full size table
Fig. 4
figure 4

Examples of fire distribution in the study region. Peatlands marked in gray color, hotspots indicated as small stars. The presence of fire clusters in Sumatran peatlands both in the La Niña year of 2008 and El Niño year of 2009 indicates deliberate large-scale land clearance burning. Note the absence of fires in the peatlands of Borneo in 2008, but abundance of wildfires (see justification in the text) in 2009. Small-holder dominated burning in Java shows very little variation between 2 years

Full size image

The increase in number of fires in Borneo from 2008 to 2009 was mainly due to a drastic rise in peatland fire activity. Fire density in the peatlands of Borneo was nearly ten times higher in 2009 than in 2008 (Table 3). In both years, peatland areas had the highest density and proximity values. This means that hotspots were not only concentrated in peatlands, but they were detected in groups indicating larger and longer lasting fires. In mineral soils, fire detections were typically more evenly spread out signifying small and short-lived fires (Fig. 4). This idea was further supported by the fact that in both peatlands and mineral soils, the proximity values roughly doubled from 2008 to 2009, although the rise in fire density was significantly higher in the peatland areas (Table 3). If hotspots are randomly spread out in mineral soils, increase in density automatically increases the proximity value. In peatlands, however, introduction of more fire clusters (and thereby more hotspots) raises the density values considerably but does not affect the proximity values in the same magnitude if the new clusters have similar characteristics as the earlier fire clusters.

Regardless of the great difference in fire numbers in 2008 and 2009, temporal progression of fire activity followed the same trend (Fig. 5). In both years, there was a peak in fire activity early in the year between 3°N and 1.5°S latitudes. This peak was largely due to fires in those areas of Sumatra that are covered from the Northeast monsoon by Malaysian Peninsula. The main fire season started to gather momentum in May and progressed from north to south peaking in August for areas north of 1.5°S latitude and in September for areas south of 1.5°S latitude (Fig. 5). This pattern can be seen in both 2008 and 2009, although the level of fire activity was several times higher in the El Niño year of 2009. The southward shift of the fire season generally trails the southward movement of the intertropical convergence zone over the study area.

Fig. 5
figure 5

Temporal distribution on fires in 2008 and 2009 in the study area. Numbers of active fire detections in four zones from north to south are presented (zone division: 7.5°N-3°N–1.5°S-6°S-10°S). Note: the peak in zone 1.5°S–6°S in September 2009 reaches over 18,000 hotspots

Full size image

Fire distribution on mineral soils

The more expansive mineral soil areas experienced the majority of fire activity in the study area (72 and 61% in 2008 and 2009: Table 3). However, fire densities and proximity values were generally lower on mineral soils than on peat soils (Table 3). This is caused by large surface area and high percentage of small-holder fire activity. Low intensity fire activity spread out widely over mineral soils can be witnessed in Fig. 4. Around three quarters of all fires in mineral soils (Table 4) were detected in fully or partially managed land cover types (plantations/secondary growth, mosaic landscape and lowland open areas). This highlights the importance of fire as a land preparation tool in this region.

Table 4 Active fire characteristics by land cover type in mineral soils
Full size table

It is important to notice, however, that 18 and 22% (in 2008 and 2009, respectively) of fires in mineral soils were detected in lowland evergreen forests (Table 4). These ecosystems are generally thought to be very resistant against fires and experience fires rarely in natural conditions. Occurrence of fires with high enough intensity to be detected by satellites in lowland evergreen forests is likely to lead to changes in land cover, regardless if they are wildfires or deliberate land clearance fires. Of all the 8,953 fire detections in lowland forests in 2009, 5,467 (61%) were detected in Borneo and 3,311 (37%) in Sumatra. Further investigation into the issue revealed that the proximity values for lowland forests in Borneo and Sumatra (49 and 45, respectively) were among the highest in mineral soils suggesting stronger clustering (and thereby larger fires) than in other land cover types in mineral soils. This can be taken as a sign of potential land conversion in the remaining lowland forests.

A few unusual observations in Table 4 deserve explanation. The small number of fire detections in peatswamp forests on mineral soils was caused by border issues between the two datasets used in land cover analysis and should be ignored. A detailed inspection revealed the high proximity value in upper montane forests in 2009 to be caused by false detections triggered by hot volcano craters in Java. And finally, it may be worth clarifying that the fires detected in water (Table 4) took place in seasonally wet areas (e.g. flood zones of rivers) that dry up during drier months and come highly susceptible to fire due to dry senescent organic material.

Fire distribution in peatlands

In general, peatland areas had significantly higher fire densities than mineral soils all over the study area in both 2008 and 2009. In 2009, around 40% of all fires were detected in peatlands that covered 10% of the study area (Table 3). It is important to notice, however, that there was large variation in fire distribution within peatland areas between the two study years. Sumatran peatlands had considerably more fires and higher proximity values than peatland areas in Borneo in the wet La Niña year of 2008. However, in 2009, the situation changed entirely, and the number and clustering of hotspots was higher in the peatlands of Borneo. Peninsular Malaysia not only had significantly less fires in peatland areas than Sumatra and Borneo, but the proximity values indicated smaller individual fire events.

The difference in fire characteristics between peatlands of Sumatra and Borneo can be explained by land cover structure. In the La Niña year of 2008, the highest proximity values were found in peatswamp forests, cleared areas and industrial plantations (Table 5). Large-scale fire activity in these land cover types suggests large-scale land clearance, typically to establish/renew industrial plantations. In the same year, fire activity was generally very low in Borneo where peatlands are dominated by degraded unmanaged ecosystems prone to wildfires (Table 2). This indicates lack of wildfires in the La Niña year of 2008, but continuing deliberate land clearance burning, predominantly in the peatlands of Sumatra (Fig. 4).

Table 5 Active fire characteristics by land cover type in the peatlands of Sumatra and Borneo
Full size table

In the El Niño year of 2009, on the other hand, fire counts and proximity values were generally highest in the fern/shrublands, secondary growth and degraded peatswamp forests of Borneo (Table 5). Note that nearly 40 times more fires were detected in the fern/shrublands of Borneo in 2009 than in 2008. This indicates high levels of wildfire activity in 2009. Similar trends in fire distribution between land cover types could be seen also in Sumatra, but in considerably smaller magnitude (Table 5). This can be explained by the fact that peatlands in Sumatra are more intensively managed and wildfires are not as widespread during dry conditions. Note also that fire counts for industrial plantations varied very little between the La Niña and El Niño years in both Sumatra and Borneo. These findings (1) highlight the strong annual variation of wildfire activity in unmanaged degraded peatland areas depending on yearly weather patterns and (2) reveal the underlying reason for the great difference in fire regimes of large-scale biomass burning in the peatlands of Sumatra and Borneo (Fig. 4).

Discussion and conclusion

The results of this study have underlined the influence of peat soil and land cover distribution on fire activity in insular Southeast Asia. Fire activity was not only concentrated on peatlands, but fires in peatlands were shown to be larger than fires in mineral soils. In addition, the results revealed highly different fire regimes in the peatlands of the two biggest islands in insular Southeast Asia, Sumatra and Borneo. These differences were attributed to more intensive land management in the peatlands of Sumatra. Together, the abovementioned findings enabled us to characterize the major trends of fire regime distribution in this region.

The three major fire regimes considered in this study (small-holder burning, large-scale land clearance and wildfires; Fig. 1) manifested themselves very differently depending on the general weather conditions of the year. In a wet year, the dominant large-scale fire regime was deliberate land clearance burning, and it took place especially in the peatlands of Sumatra. This result is supported by earlier findings on the use of fire in this region (Bowen et al. 2001; Chokkalingam et al. 2006; Dennis et al. 2005). Peatlands are considered to be particularly affected since they are currently the main target of large-scale land conversion activities in insular Southeast Asia and fire is still widely used as a land preparation method in peatland areas (Simorangkir 2006). This type of large-scale land clearance burning takes place annually in the extent that weather conditions permit. It currently occurs particularly in Sumatra due to more intensive peatland management (Miettinen and Liew 2010).

Large-scale wildfires, on the other hand, were practically absent in the La Niña year of 2008 due to ample rainfall. However, when El Niño conditions arose in 2009, especially the degraded peatlands of Borneo experienced widespread wildfire activity. The results of this study, thereby, confirmed earlier findings on the high fire vulnerability of degraded ecosystems in dry years (Cochrane 2001; Rieley and Page 2005; Siegert et al. 2001). This has a fundamental effect on fire activity in Borneo, where peatland areas are nowadays dominated by unmanaged and degraded landscape (Miettinen and Liew 2010). The impact of wildfires is smaller in Sumatra due to more intensive land management. Regardless of the fact that large-scale wildfires do not take place annually, they form a key fire regime in insular Southeast Asia and gather high amount of media attention especially during periodically occurring haze episodes.

Note that both of the large-scale fire regimes (land clearance and wildfires) currently take place especially in peatland areas. Combustion of carbon-rich peat soil during peatland fires worsens the global effects of insular Southeast Asian biomass burning considerably. It has been estimated that over 60% of the carbon released into the atmosphere during a severe fire episode in Indonesia in 1997–1998 came from burning peat soil (Heil et al. 2006). Due to the immense amount of carbon stored in peat soil in this region (Jaenicke et al 2008; Wahyunto et al. 2003, 2004), large-scale fire activity in insular Southeast Asian peatlands has a great potential to affect global climate change (Ballhorn et al. 2009; Page et al. 2002).

In addition to the abovementioned large-scale fire regimes, small-holder burning took place throughout the study area. It appeared to be the dominant fire regime in the island of Java, primarily because it is a densely populated island covered with small-holder dominated mosaic landscape. The results also suggested that small-holder burning would be the most stable of the fire regimes in insular Southeast Asia and would show less annual variation than other major fire regimes. However, this could not be fully confirmed since small-holder burning often intermingles with large-scale land clearance and wildfires discussed above. Therefore, further studies area needed to confirm this issue. In any case, although small-holder burning contributes significantly to the number of detected active fires in this region, it is likely to cause smaller effects than the other two fire regimes discussed above due to typically small burnt areas and small amount of burnt biomass.

This regional level study was based on the most recent land cover and peatland information currently available for the study area and a set of fire data for 2 years with highly different climate conditions. Further research on longer time period is needed to confirm the findings of this study. In addition, local-level investigation into the effects of land cover and peatland distribution on fire characteristics could deepen understanding of the regional level trends noticed in this study. It also has to be remembered that due to limitations imposed by the data used in this study, we were not able to include e.g. fire severity and fire effects into our analysis. Future studies should investigate methods to monitor the severity and effects of fires for improved estimation of regional to global consequences of fires.

Although this study has concentrated on the physical constraints of fire activity, we should never forget the importance of land management policies in the context of insular Southeast Asian vegetation fires. Numerous authors have emphasized land management and land use policies as the most important underlying cause of vegetation fires in insular Southeast Asia (e.g. Chokkalingam et al. 2006; Dennis et al. 2005; Ketterings et al. 1999; Murdiyarso and Adiningsih 2006; Stolle et al. 2003). Land management policies are to large extent behind the current land cover structure, which was shown to strongly affect fire regimes in this study. Furthermore, policies on the use of fire may affect fire occurrence even within similar land use. Langner and Siegert (2009) revealed variation in fire activity within Borneo between different countries. Similarly, low fire activity in Peninsular Malaysia noticed in this study may be partly due to fire policies.

It was shown in this study that fire regime characteristics are strongly connected to the occurrence of peat soil, land cover and weather patterns in this humid tropical region where fires in natural conditions are rare. This leads to high variability of fire activity both annually and over longer time range and greatly complicates interpretation of active fire data. The same land cover and soil type may show very different fire characteristics in wet and dry years. On the other hand, fire characteristics in any given area may vary significantly over time (regardless of the yearly weather conditions) depending on the current land management practices (Dennis et al. 2005). Therefore, information on land cover and occurrence of peatland is crucial for correct interpretation of regional active fire datasets in insular Southeast Asia. Without this information, the nature of the fires cannot be determined and reliable estimates of the effects of the fires cannot be given.

Overall the results of this study confirmed continuing use of fire as a tool in large-scale land conversion and management in insular Southeast Asia and persisting fire vulnerability of degraded areas. The current land cover distribution together with land management practices makes the region vulnerable to a catastrophic fire episode when exceptionally strong El Niño conditions arise. A region-wide fire episode under these circumstances would not only cause unprecedented damage to local ecosystems, societies and human health in insular Southeast Asia, but it would also cause globally significant carbon emissions into the atmosphere.