ENSO and monsoon variability during the past 1.5 kyr as reflected in sediments from Lake Kalimpaa, Central Sulawesi (Indonesia)

The climate of Sulawesi is driven by the monsoon system as well as the El Niño-Southern Oscillation (ENSO). Until now, mechanisms and long-term variations of these complex interacting climate processes have been poorly understood. This paper uses a sediment record from Lake Kalimpaa to investigate long-term rainfall trends of the past ~1500 years. Granulometric and geochemical data provide indications for an increasingly wetter climate (higher rainfall intensities and/or mean rainfall) on centennial to millennial time scales from approximately ad 560 to the 20th century. Highest rainfall intensities probably occurred at the end of the ‘Little Ice Age’ (LIA). The trend towards wetter conditions during this time could also be detected in other palaeoclimatic studies from the region. A plausible explanation for these observations is the southward displacement of the Intertropical Convergence Zone (ITCZ) associated with changes in monsoon dynamics. However, comparison of the results with other proxy and model data indicates that the long-term rainfall variability in Central Sulawesi is also determined by variations in ENSO. During the 20th century, the climate signal in the Kalimpaa record is superimposed by human impact. Moreover, the data suggest that two periods of disturbance occurred within the lake catchment between about ad 1090–1190 and ad 1450–1620. Comparison with fire frequency derived from macro-charcoal analysis indicates that these events were caused by forest fires which likely took place during periods of drought. Broadly simultaneous drought periods have been detected in records from East Java suggesting a regional drought occurrence affecting at least East Java and Sulawesi.


Introduction
Sulawesi belongs to the Greater Sunda Islands located in the Indo-Pacific warm pool (IPWP), which is the largest storage of warm surface ocean water globally (Gagan et al., 2004;Oppo et al., 2009). Consequently, this area is a substantial source of latent heat and has a great importance for understanding modern climate dynamics (Hope, 2001).
Several rainfall regimes occur in Indonesia in response to, for example, variations in the monsoon system and the El Niño-Southern Oscillation (ENSO; Aldrian and Susanto, 2003). The seasonal rainfall in eastern Indonesia is mainly driven by the intensity of the Australian-Indonesian summer monsoon (Griffiths et al., 2010), whereas the ENSO cycle (Diaz and Kiladis, 1992) is responsible for the complex interannual climatic variability in this region. While ENSO cold events (La Niña) enhance the rainfall in Sulawesi, ENSO warm events (El Niño) result in dry periods, which occasionally have led to extreme droughts in the past (D'Arrigo et al., 2006;Keil et al., 2008;Salafsky, 1994). These drought events facilitated the occurrence of forest fires in tropical rainforests, as recorded in Borneo and Sulawesi during the strong El Niño of 1997/1998 (Rowell and Moore, 2000;Sastry, 2002;Siegert and Hoffmann, 2000).
The annual influence of the monsoon is dominant in most parts of Sulawesi, where the wet season coincides with the Australian-Indonesian summer monsoon from October to March and the dry season with the Asian summer monsoon from April to September (Aldrian and Susanto, 2003;Gunawan, 2006;Figure 1b-c). An 'anti-monsoonal' rainfall pattern has been observed in the northeast of Sulawesi. This area representing only a vast minority of the island is characterized by inverse occurrences of wet seasons from April to August and dry seasons from September to March as well as a more pronounced impact of ENSO (Aldrian and Susanto, 2003;Aldrian et al., 2004;Gunawan, 2006). The mechanism underlying this pattern is the strong influence of the ocean and its sea surface temperatures (SST) in this region. Due to the Indonesian throughflow (ITF), marine currents from a Pacific warm pool northeast of New Guinea can enter the Indian Ocean ENSO and monsoon variability during the past 1.5 kyr as reflected in sediments from Lake Kalimpaa, Central Sulawesi (Indonesia) (Godfrey, 1996;Morey et al., 1999). These currents not only mainly pass the Makassar Strait but also influence the SST in the Molucca Sea northeast of Sulawesi (Figure 1a). During the boreal winter, the ITF brings cooler water from the warm pool to the Molucca Sea inhibiting the formation of a convective zone in the northeastern part of Sulawesi (wet season in the monsoonal regions of Indonesia). In contrast, during the boreal summer, warmer water supplied from the warm pool promotes the formation of a convective zone in northeastern Sulawesi (dry season in most parts of Sulawesi; Aldrian and Susanto, 2003;Godfrey, 1996;Gordon et al., 1999;Gunawan, 2006;Morey et al., 1999). According to Gunawan (2006), Central Sulawesi is characterized by a mixture of both monsoonal and 'anti-monsoonal' types of rainfall and by the strong influence of ENSO.
Lore Lindu National Park, rainfall is mainly generated orographically (Gunawan, 2006). During boreal winters, the Australian-Indonesian summer monsoon brings moist air masses to Central Sulawesi reaching the island from the northwest. In the mountains around Palu Valley, the ascending air leads to orographic rainfall with a strong correlation between altitude and precipitation totals. In contrast, dry air masses reach Sulawesi during austral winters, when precipitation is lower and solely generated orographically (Gunawan, 2006).
Thus far, neither the mechanisms nor the long-term variations of these complex interacting climate processes have been completely understood. Palaeoenvironmental data are needed to achieve more detailed knowledge of palaeoclimatic changes in Sulawesi; there are only a few palaeoclimatic studies from this region covering the past millennia. Marine records are available from Makassar Strait (Newton et al., 2006;Oppo et al., 2009;Tierney et al., 2010;Figure 1a) and Kau Bay (Halmahera; Langton et al., 2008); lacustrine sediments from East Java (Lake Lamongan: Crausbay et al., 2006;Lake Lading: Konecky et al., 2013;Lake Logung: Rodysill et al., 2012) provide terrigenous archives with a suitable temporal resolution. Palynological evidence is available from two sites in the Besoa Valley, Central Sulawesi, and the pollen data of one of these exhibit a cooling trend during the 'Little Ice Age' (LIA; Kirleis et al., 2011). However, palaeoclimatic studies from Sulawesi are rare and cover longer time frames with lower temporal resolutions (Lake Tondano: Dam et al., 2001; Wanda site near Lake Matano: Hope, 2001;Figure 1a). Therefore, it is necessary to carry out further investigations on palaeoenvironmental archives from Sulawesi to enhance the spatial and temporal resolution of proxy data.
This study aims to contribute to a better understanding of palaeoclimatic changes in Sulawesi during the past 1500 years. The main objective is to investigate various sedimentological characteristics of a core from Lake Kalimpaa (Central Sulawesi; Figure 1a) in order to provide indications for long-term rainfall trends, palaeoenvironmental changes and fire history in the lake catchment.

Site description
Lake Kalimpaa (1°19′34.8″S, 120°18′31.9″E; Figure 1a), sometimes also referred to as Danau Tambing, is located at 1660 m a.s.l. in the Lore Lindu National Park in Central Sulawesi, Indonesia. It occupies an area of 6.5 ha and has a maximum water depth of 6.6 m. The lake is surrounded by small reed belts and a swamp area to the north ; it has an inflow from the northeast and an outflow to the southwest. Lake Kalimpaa is located in a mountain pass at the end of a side valley of the Palu Valley. Surrounding mountaintops in the northeast and southwest exceed altitudes of 2000 m a.s.l. The lake's catchment area of approximately 1.8 km 2 extends mainly to the north. The geological setting is characterized by the Pliocene Kambuno Granite (Leemhuis, 2005;Priadi et al., 1994;Simandjuntak et al., 1991) and its weathering products. Regional vegetation is characterized by montane rainforest which in the Lore Lindu National Park is presently characterized by a 90% intact canopy cover (Kirleis et al., 2011) and is dominated by Fagaceae, mostly represented by the two genera Lithocarpus and Castanopsis (Culmsee et al., 2010).
There is no representative meteorological station in the vicinity of Lake Kalimpaa itself, but there are records for the mountains of western Central Sulawesi, and these can be used as representative of the catchment. Because of the close proximity to the equator, temperatures are almost constant throughout the year. Daytime temperatures at higher altitudes range from 16°C to 22°C; annual precipitation is around 2000-3000 mm (Weber, 2006).

Materials and methods
A 211-cm-long composite sediment record consisting of three overlapping sections (KAL 1-1, KAL 1-2, KAL 1-3) was recovered from Lake Kalimpaa using a Livingstone piston corer (Livingstone, 1955). Palaeomagnetic analyses showed that only KAL 1-1 and KAL 1-2 (153 cm composite length) were suited for palaeoenvironmental reconstruction since no reliable chronology could be established for the oldest part of the core . Each section was split, photographed and described lithologically. The composite profile was compiled based on macroscopic marker layers. The chronology was adapted from Haberzettl et al. (2013;Figure 2a). Several sets of bulk samples which had been sent for radiocarbon dating revealed inconsistencies, that is, ages were not in stratigraphic order. Since hard-water effects can be excluded for Lake Kalimpaa, another kind of The age-depth model for the Lake Kalimpaa record based on calibrated radiocarbon ages which are presented as medians referring to the 2σ ranges . The grey data points are regarded as outliers and have been excluded from the final agedepth model . (b) Comparison of declination and inclination of the Lake Kalimpaa record  with the CALS3k.4 model output for the location of Lake Kalimpaa (Korte and Constable, 2011). reservoir effect is assumed. Probably, the dated material of the bulk samples comprised both autochthonous and allochthonous organic matter. Therefore, only the youngest ages were used for the age-depth model that is based on linear interpolation Figure 2a). Despite these dating uncertainties, the created age-depth model seems to be a first-order approximation which is corroborated by magnetostratigraphy. The comparison of palaeosecular variation data of the Kalimpaa record with the output of the CALS3k.4 model (Korte and Constable, 2011) shows many isochronic similarities for inclination as well as for declination ( Figure 2b). Therefore, the age-depth model can be used as basis for further multi-proxy approaches on the Lake Kalimpaa record. A marked increase in sedimentation rate is obvious for the most recent sediments. While the mean sedimentation rate between 151 and 44 cm is about 0.8 mm/yr, for the youngest sediments between 44 cm and the top of the core, the age-depth model provides a sedimentation rate of ~9.2 mm/yr on average.
After palaeomagnetic measurements, u-channels of KAL 1-1 and KAL 1-2 were subsampled for subsequent geochemical and granulometric analyses. Grain size distribution was measured at an interval of 1 cm (N = 153) with a Laser Diffraction Particle Size Analyser (Beckman Coulter LS 13320) utilizing the Aqueous Liquid Module and 10 s ultrasonic for dispersion. The Fraunhofer diffraction theory was used for optical modelling of light scattering (De Boer et al., 1987). Organic matter and carbonates were removed with H 2 O 2 (30%) and HCl (10%), respectively, and sodium pyrophosphate solution (Na 4 P 2 O 7 ·10 H 2 O; 0.1 M) was used as a dispersion medium. The measurements were carried out in several runs until a reproducible signal was obtained. The first reproducible run was taken for further statistical treatment. The grain size fractions were calculated according to Ad-hoc AG Boden (2005). The skewness (ϕ-scale) was determined by the logarithmic method of moments by means of a modified version of Gradistat 4.2 (Blott and Pye, 2001).
With the use of syringes, sediment samples were obtained at ~4 cm intervals (N = 41) from the core and freeze-dried. Aliquots were ground and analysed for total organic carbon (TOC) and total nitrogen (TN) contents with a CNS elemental analyser (Euro EA 3000). TOC was determined after the destruction of carbonates with 2 M H 3 PO 4 . Subsequently, the molar TOC/TN ratio was calculated.
Additionally, mineralogical investigations on the pestled samples (N = 10) were carried out using an x-ray diffractometer (D8-Discover, Bruker AXS) equipped with a CuKα x-ray tube and a gas proportional counter (HI-STAR area detector, Bruker AXS). The evaluation of the data was performed by means of the software Match! 2.0.9 and MacDiff 4.2.6.
Macro-charcoal particles (>150 µm) were counted for samples evenly spaced at 1 cm intervals along the upper part of the sediment core (first 145 cm). The samples (1 cm 3 each) were prepared following the method of Stevenson and Haberle (2005), which is a modification of a method developed by Rhodes (1998). Weak hydrogen peroxide (6% H 2 O 2 ) was used to partially digest and bleach organic material in the sediment samples when counted under a binocular dissecting microscope. The sample preparation procedure aims to ensure that little particle fragmentation occurs during preparation. Results are expressed as number of charred particles per cubic centimetre.
Charcoal raw data were interpolated to constant 5 years, corresponding approximately to the median temporal resolution.
Interpolated charcoal concentrations (number of particles per cubic centimetre) were multiplied by estimated sedimentation rate (cm 2 /yr) to obtain the charcoal accumulation rate (CHAR, particles/cm 2 /5 yr) of each sample. Low-frequency variations in a charcoal record (C background ) represent changes in charcoal production, sedimentation, mixing and sampling. C background was estimated with a locally weighted regression using a 100-year window in order to maximize the signal-to-noise index and the goodness-of-fit between the empirical and the modelled C noise distributions (Higuera et al., 2009). C background was subtracted to obtain a residual series, C peak . It is assumed that C peak is composed of two subpopulations (Higuera et al., 2008(Higuera et al., , 2009): C noise , representing variability in sediment mixing, sampling and analytical and naturally occurring noise, and C fire , representing charcoal input from local fires. A Gaussian mixture model was used for each sample to identify the C noise distribution. The 99th percentiles of the C noise distributions were considered as thresholds separating samples into 'fire' and 'non-fire' events. The peaks which passed the threshold criterion were subjected to a 'Poisson minimum-count' screening in order to eliminate the peaks that result from statistically insignificant variations in charcoal counts. Peak fire episodes refer to one or more fires occurring within the time span of the charcoal peak. Past fire regime characteristics were inferred based on the temporal pattern of identified charcoal peaks via calculation of fire frequencies smoothed with a 200year window. All statistical treatments were done using the program CharAnalysis (Higuera et al., 2009).

Palaeorainfall proxies
Grain size can be used as a proxy for variations in transport energy or lake levels and, hence, climate variability (Conroy et al., 2008). In detail, high rainfall intensities and/or amounts possibly result in enhanced erosion in the catchment as well as an increased transport capacity and competence of the tributary which might lead to the deposition of coarser clastic material in the lake (Håkanson and Jansson, 1983). According to Nichols (2009), grain size can yield information on the flow velocity and hence the runoff during the time of sediment deposition.
In the Lake Kalimpaa catchment, sediment erosion and transport are likely driven by runoff. Channel erosion and denudation of soil material are probably the major causes for the transport of sediments, although dense vegetation cover accompanied by high interception protects the soil surface.
It is assumed that the deposition of coarse grain sizes reflects periods or events characterized by high runoff and, hence, higher rainfall intensities. In contrast, the deposition of fine grain sizes is linked to periods of lower average rainfall and, thus, lower transport energy of the inflow. In such a small system, probably lower runoff facilitates the deposition of finer particles due to the absence of turbulences in the lake water which are caused by the inflow during periods of greater rainfall and runoff.
The skewness (Sk) of grain size distributions of deposited sediments is a result of the composition of the source material as well as the energy level of the transport process. If transport processes exhibit a high energy level, the deposited sediments become coarser, and their grain size distributions are more positively skewed (on ϕ-scale). If the energy level is low, the deposited sediments become finer and more negatively skewed (McLaren and Bowles, 1985).
It is supposed that the main source of Al, Ca, K, Mg and Ti are the minerals of the Kambuno Granite and its weathering products. K is mainly associated with K-feldspars, biotite and muscovite as well as, to a lower extent, with illite. Mg is related to biotite and clay minerals, such as montmorillonite; Ca occurs, for instance, in plagioclase and illite. Ti is chemically immobile and occurs mainly in heavy accessory minerals like rutile and ilmenite, which are extremely resistant against weathering (Goldich, 1938;Li et al., 2003;Shotyk et al., 2001). Therefore, Ti was used in many other studies on lake sediments as an indicator for the input of clastic, terrigenous material (e.g. Haberzettl et al., 2005;Kasper et al., 2012;Whitlock et al., 2007) -a process that is often driven by precipitation and runoff in the lake catchment.
During the chemical weathering of feldspars and micas to clay minerals, K, Ca and Mg get dissolved. In contrast, Al, which is part of feldspars and micas as well as clay minerals, is nearly insoluble and less mobile than the alkali and alkaline earth elements (Middelburg et al., 1988;Nesbitt et al., 1980). Kaolinite and gibbsite, which are common weathering products of granites under tropical conditions (West and Dumbleton, 1970), contain Al but no K, Ca and Mg. Therefore, Al is used as denominator in various element/Al ratios in the following to compensate variable depletion effects which are primarily caused by concentrations of the redox sensitive element Fe.
In many other studies, element/Al ratios like K/Al and Ti/Al are related to weathering intensities in the source area, input pathways or the strength of transport processes (Boyle, 1983;Lückge et al., 2001;Müller et al., 2001;Zabel et al., 2001). Engstrom and Wright (1984) and Mackereth (1966) found that alkali and alkaline earth elements (e.g. Ca, K and Mg) accumulate in lake sediments during periods of intense erosion, when mineral matter is transported into the lake. In contrast, low values of alkali and alkaline earth elements in lake sediments occur when erosion is low, and leaching of the catchment soils is dominant. Granite is usually deeply weathered in tropical climates, and thus, unweathered minerals like feldspars are more prominent in areas with higher slope angles (Ruxton, 1959). Therefore, the source area of feldspars and micas is possibly the steeper slopes around the lake, and their transport into the sediment occurs via terrigenous runoff during periods of high erosion. Accordingly, K/Al, Ca/Al and Mg/ Al reflect the variability of the proportion of chemically less weathered feldspars and micas to clay minerals, especially kaolinite. The Ti/Al ratio mainly reflects changes in grain size (Boyle, 1983;Zabel et al., 2001). Ti is associated with coarser material, while Al rather represents the fine-grained fraction, so that the Ti/ Al ratio is linked to the strength of fluvial transport which reflects hydrological variability. Therefore, Ti/Al, K/Al, Mg/Al and Ca/Al ratios may be used as proxies for palaeorainfall similar to other studies on tropical lake sediments (Felton et al., 2007;Warrier and Shankar, 2009). High ratios are interpreted as periods and/or events of high erosion and, accordingly, high rainfall intensities in the lake catchment.

Proxies for catchment disturbances and changes in redox conditions
Fire frequency data derived from macro-charcoal analysis are used to reconstruct the fire history around Lake Kalimpaa. Such forest fires may cause substantial disturbances within the lake catchment regarding changes in vegetation and sedimentological processes. Enhanced erosion as well as an increased supply of organic matter entering the lake can be the consequences of these disturbances.
Fe, Mn, P, S, TOC and TN are typically part of a common reaction and transport cycle in sediments (Van Cappellen and Wang, 1996). The significantly higher presence of allogenic Fe and Mn is characteristic of tropical lakes (Crowe et al., 2008). In soil samples from the upper horizons of nine sites in the catchment area of Lake Kalimpaa, average contents of ~2.5% of Fe and ~340 ppm of Mn were measured (Markussen, 2000). As it will be demonstrated later, Fe and Mn concentrations are significantly higher in certain sections of the sediment core than those in the topsoil. This cannot be explained solely by an increased input rather by the accumulation of these elements due to another process(es), such as redox reactions in the sediment and/or the water column. Since this is accompanied by relatively high TOC/ TN ratios, which can indicate a shift to a more terrestrial origin of the organic matter (Haberzettl et al., 2008;Mayr et al., 2005;Meyers, 1994), it is hypothesized that sharp increases in TOC, TN, S, Fe and Mn concentrations result from the supply of organic material from the catchment. This promotes reducing conditions in the sediment due to microbial decomposition. In contrast, low molar TOC/TN ratios likely arise from a comparatively low input of terrestrial organic matter or a higher occurrence of phytoplankton, which is typically characterized by TOC/TN ratios ranging from 4 to 10 (Kasper et al., 2013;Meyers, 1994).
The microbiological decomposition of organic material in lake water and sediment is an oxygen-consuming process that produces anoxic environments (Davison, 1993). In lakes, Fe and Mn occur in various oxidation states (Fe II and Fe III ; Mn II , Mn III and Mn IV ) depending on the given redox conditions. Under oxidizing conditions, both elements exhibit low solubility, where Mn is soluble at higher redox potentials than Fe (Sigg and Stumm, 1996). Fe that enters lakes via rivers occurs mainly in particulate form (>99%; Salomons and Forstner, 1984) as ferric (oxyhydr) oxide or bound in the lattice of micas and clay minerals. Mn is also predominantly supplied by solids, such as MnO 2 , and to lower amounts as dissolved Mn 2+ (Davison, 1993;Engstrom and Wright, 1984). There seems to be a succession of redox processes in the sediment record from Lake Kalimpaa, during which particulate Fe(III)-(oxyhydr)oxides are dissolved to Fe 2+ which precipitates as amorphous Fe(II)-sulphide or, for instance, mackinawite, or it becomes mobile and reprecipitates as Fe(III)-(oxyhydr)oxide at the oxic/anoxic boundary. The identification of Fe(II)-sulphides is inferred from the black sediment colour and its association with high Fe and S bulk concentrations (Emerson, 1976;Engstrom and Wright, 1984). The affinity of P to be adsorbed on the surfaces of Fe-(oxyhydr)oxides is known from the literature (e.g. Buffle et al., 1989;Lijklema, 1980). According to Lopez et al. (2006), high Fe and P values are indicative for the precipitation and accumulation of authigenic Fe. Moreover, P is also associated with organic matter.

Core lithology and mineralogy
The core sections KAL 1-1 and 1-2 consist mainly of finely laminated silts with a few homogeneous sections and distinct sand layers. Based on obvious changes in sediment structure, colour and macroscopic grain size, six lithological units can be distinguished (Figure 3). Unit I (153-128 cm; ~ad 560-1090) exhibits finely laminated blackish grey and light grey silts. Unit II (128-118 cm; ~ad 1090-1190) is characterized by homogeneous black sediments composed of fine silts to fine sandy coarse silts. Finely laminated, light greyish to grey layers consisting of medium to fine sandy coarse silts occur in unit III (118-93 cm; ~ad 1190-1450; Figure 3). Mica particles, probably muscovite, are conspicuous between ~110 and 40 cm (~ad 1275-1965). Partially laminated, dark brown to black fine sandy coarse silts occur in unit IV (93-77 cm; ~ad 1450-1620). Unit V (77-48 cm; ~ad 1620-1910) exhibits a greyish colour and contains partially laminated, fine sandy coarse silts with an intercalated sand layer. The uppermost unit VI (48-0 cm; ~ad 1910-2006; Figure 3) consists of brownish medium to fine sandy silts. It is assumed that human impact in the catchment area (e.g. by road construction, camp site) superimposes any effects of climatically induced changes during the 20th century . Nevertheless, it cannot be ignored that climate could have also caused the marked increase in sedimentation rate for the most recent sediments. Since the distinction between climatic and human-driven impacts seems not possible with certainty, no palaeoclimatic interpretation has been done for unit VI in the following.
X-ray diffractograms of all measured samples show comparable spectra. The most common sediment minerals are quartz, alkali feldspars, plagioclase, muscovite, biotite (chemically unweathered minerals from the Kambuno Granite), kaolinite and, to a lesser extent, illite (weathering products of the Kambuno Granite). In some samples, goethite and gibbsite were detected, which are typical for weathered granites in tropical regions as well as clay minerals (West and Dumbleton, 1970). Furthermore, x-ray diffraction spectra indicate that carbonates are of minor importance in this record which was confirmed by multiple negative tests with hydrochloric acid.

Grain size analysis
The core sections KAL 1-1 and 1-2 are dominated by medium to coarse silt and accessorily contain some fine sand layers. With a mean of ~78.3% in all samples, silt is the most common fraction followed by sand (~13.5%) and clay (~8.2%). A correlation matrix of the calculated grain size fractions (Table 1) exhibits positive correlations between clay (Cl), fine silt (FSi) and medium silt (MSi). Coarse silt (CSi) and fine sand (FSa) are also positively correlated but negatively correlated to the aforementioned. Moreover, there is a positive correlation between medium sand (MSa) and coarse sand (CSa) which are only sporadically present in the record (Table 1). On this basis, the grain size fractions can be grouped into classes, namely, one of Cl + FSi + MSi and one of CSi + FSa. They show an opposing trend (Figure 3c) but depend on each other and, thus, probably represent the same palaeoenvironmental signal. In contrast, the third class consisting of MSa + CSa is interpreted as an additional signal.
Units I and II (153-118 cm; ~ad 560-1190) are characterized by low but slightly increasing CSi + FSa and Sk values as well as high but slightly decreasing Cl + FSi + MSi values (Figure 3c

Macro-charcoal and fire frequency
A total of 11 fire episodes occurred locally during the past 1300 years, and 7 additional fire episodes failed to pass the screen

Palaeorainfall in Central Sulawesi
Based on variations in the established palaeorainfall proxies (K/ Al, Ca/Al, Mg/Al, Ti/Al, CSi + FSa, Sk), the Lake Kalimpaa  record likely reveals a centennial to millennial-scaled trend towards wetter conditions characterized by higher rainfall intensities and possibly higher mean rainfall from ~ad 560 to the 20th century. The period from ~ad 560-1090 was characterized by drier conditions, while an increasingly wetter climate can be inferred from ~ad 1090 to 1910 (Figure 4). A long-term trend towards wetter conditions is also observed in sediments from two other lakes in East Java (Konecky et al., 2013;Rodysill et al., 2012; Figure 6e). Rodysill et al. (2012) explain the trend as a consequence of migration of the Intertropical Convergence Zone (ITCZ) with increased precipitation during its southward displacement. In comparison, Konecky et al. (2013) conclude that the migration of the ITCZ influences the climate variability on multidecadal to centennial time scales, while it is the strengthening of the Walker circulation and its associated changes in ENSO variability that produces the increasingly wetter climate during the last millennium. Yan et al. (2011) arrive at similar conclusions, namely, precipitation changes in response to the combined influence of the migration of the ITCZ and the position and strength of the Pacific Walker circulation in the western tropical Pacific.
The palaeorainfall proxies obtained from the Lake Kalimpaa record show long-term similarities with ENSO variability from the eastern Pacific region as well as the South American continent (Figure 6a-c). An ENSO record derived from the sediments of Laguna Pallcacocha, southern Ecuador, reveals that the number of El Niño events per 100 years decreased since ~ad 1150 until today (Moy et al., 2002) consistent with increasing terrigenous input to Lake Kalimpaa (Figure 6a-c). Another lacustrine archive from Galápagos exhibits a decreased ENSO frequency between ad 1300 and 1850 (Conroy et al., 2008). From Kau Bay (Halmahera; Figure 1a), a region where rainfall is strongly dependent on ENSO, it is known that the El Niño activity decreased steadily from ~ad 1300 (Langton et al., 2008). A long-term trend starting from ~ad 560 onwards is visible in both the established palaeorainfall proxies for the Kalimpaa record and the output of the coupled ocean atmosphere model from Clement et al. (2000; Figure 6a and b) which reflects the number of El Niño events occurring over a 500-year interval. These similarities of the ENSO and CSi + FSa data (Figure 6a-c) indicate that palaeorainfall in Central Sulawesi is related to the variability of ENSO on centennial to millennial time scale and, thus, changes in the Pacific Walker circulation.
Two studies carried out at Makassar Strait (Newton et al., 2006;Oppo et al., 2009; Figures 1 and 5d) revealed that the SST in this region was ~0.5-1.5°C lower during the LIA (~ad 1550-1850) compared with modern SST and those during the 'Medieval Warm Period' (MWP; ~ad 900-1300; Oppo et al., 2009). The authors infer that a cooling of the North Pacific surface water, which is transported by ocean currents through the Makassar Strait, is responsible for lower SST during this period. These cooler SST phases are associated with the southward displacement of the ITCZ during the LIA, which led to wetter conditions at Makassar Strait (Newton et al., 2006), similar to the region around Lake Kalimpaa in Central Sulawesi (Figure 6a-d). Sachs et al. (2009) note that the ITCZ reached its southernmost position between ad 1400 and 1850 with the result that drier conditions occurred north of the equator and wetter conditions in the southern tropics (Newton et al., 2006). Tierney et al. (2010), who examined marine sediments off the coast of Southwest Sulawesi (Figure 1a), inferred Figure 6. Comparison of (a) coarse silt and fine sand data (CSi + FSa) representing terrigenous input to Lake Kalimpaa with other studies; (b) modelled number of El Niño events using input data from the eastern Pacific region. Shown is the number of events in 500-year overlapping windows (overlapping every 10 years; digitized from Clement et al., 2000); (c 1 ) red colour intensity of sediments from Laguna Pallcacocha, Ecuador, representing the layers deposited during El Niño events (grey) and (c 2 ) the resulting modelled number of El Niño events per 100 years (black; Moy et al., 2002); (d) reconstructed sea surface temperatures of Makassar Strait derived from Mg/Ca ratios (Newton et al., 2006); and (e) leaf wax δD from Lake Lading (East Java) as palaeorainfall proxy (Konecky et al., 2013). The periods of the 'Little Ice Age' (LIA; ad 1550 to 1850) and the 'Medieval Warm Period' (MWP; ad 900-1300) according to Oppo et al. (2009) are shown on the right. that the hydrological variability and the monsoon strength in the IPWP are dependent on migrations of the ITCZ. These authors suggest teleconnections between ENSO and the monsoon with a weak Indian monsoon and a more El Niño-like mean state during the MWP and a strong Indian monsoon and a more La Niña-like mean state during the LIA (Tierney et al., 2010). It seems therefore that the southward displacement of the ITCZ may have caused higher rainfall intensities and possibly higher mean rainfall around Lake Kalimpaa during the LIA.
Both the monsoon and ENSO climate systems are interacting through teleconnections (Ju and Slingo, 1995;Soman and Slingo, 1997;Torrence and Webster, 1999). Variations in the Pacific Walker circulation and/or SST anomalies during El Niño events could influence the global scale divergence, which can result in a shift of the ITCZ and, thus, changes of the monsoon dynamics (Ju and Slingo, 1995;Soman and Slingo, 1997;Torrence and Webster, 1999). The position of the ITCZ as well as the mean state of ENSO/Pacific Walker circulation, therefore, may be considered interactive and co-responsible for the long-term rainfall variability in Central Sulawesi on centennial to millennial time scale. The interpretation of the Kalimpaa record suggests a general weakening of the Pacific Walker circulation and a more El Niño-like mean state from ~ad 560 to 1090. In contrast, the strengthened Pacific Walker circulation and the southward displacement of the ITCZ probably are the causes for the wetter climate in Central Sulawesi during the LIA, which is consistent with the interpretation of Yan et al. (2011).

Palaeoenvironmental disturbance events in the catchment of Lake Kalimpaa
In the Lake Kalimpaa record, peaks in Fe, Mn, P, S, TOC and TN occur almost simultaneously with increased fire frequency in unit II (128-118 cm; ~ad 1090-1190) and unit IV (93-76 cm; ~ad 1450-1620; Figure 5). The most prominent peaks of these elements coincide in unit II with macro-charcoal and MSa + CSa ( Figure 5). Increased fire frequency data indicate periods during which forest fires occurred more frequently in the drainage basin of Lake Kalimpaa. Hence, it is assumed that these disturbance events are a potential cause for the increased supply of organic material into the lake, which in turn may lead to enhanced TOC, TN and TOC/TN values. A similar pattern showing increased TOC and charcoal values was found in the sediments from Lago dell'Accesa (Tuscany, Italy; Vannière et al., 2008). As a result of microbial decomposition of the organic matter, anoxic conditions prevailed at Lake Kalimpaa and, thus, enhanced the formation of black sediment layers which contain high Fe, Mn, P and S amounts.
As noted, the deposition of MSa + CSa seems to reflect a different process compared with the other grain size fractions (Table  1). Due to the simultaneous occurrence of the most prominent peaks in MSa + CSa and macro-charcoal ( Figure 5), the deposition of these coarse materials is interpreted as an input signal after catchment disturbance (Cerdà and Lasanta, 2005) which is likely associated with a forest fire around ~ad 1090 ( Figure 5). Thus, the fire likely facilitated the erosion and deposition of MSa + CSa.
Following the first period of inferred increasing fire frequency (~ad 1090-1210), changes in the vegetational composition occurred within the catchment. These alterations are reflected within the units II and III (123-95 cm; ~ad 1140-1430) and were characterized by an expansion of Weinmannia which probably acts as secondary forest species in the Kalimpaa drainage basin (Biagioni et al., unpublished data). In contrast, Weinmannia pollen concentrations remained low after the second period of disturbance (~ad 1460-1620). However, pollen data of fast-growing pioneer species like Macaranga and Mallotus increased during this period indicating disturbance in the forest (Biagioni et al., unpublished data). Weinmannia increased again in unit VI (47-35 cm; ~ad 1930-1975), when the vegetation cover in the catchment area was likely disturbed due to the road construction ( Figure 5).
Considering the macro-charcoal and sedimentological findings from KAL 1-1 and 1-2, at least two disturbance events occurred in the catchment area that are reflected in units II andIV (~ad 1090-1190;~ad 1450-1620. The changes in fire frequencies between ~ad 1090-1190 and ~ad 1450-1620 have affected the geochemical composition of the lacustrine sediments. The alterations within the catchment during the 20th century are possibly caused by human impact .
From archaeological and palynological investigations in Central Sulawesi, it has been concluded that anthropogenic impact in the Besoa Valley, ~25 km south of Lake Kalimpaa, started ~2000 years ago (Kirleis et al., 2011(Kirleis et al., , 2012 when the montane rainforest was replaced by grassland. However, considerable human modifications of the landscape in the catchment of Lake Kalimpaa are not assumed before the 20th century. Natural fires occurring during drought periods seem to be the most likely triggers of the assumed disturbances in the Kalimpaa drainage basin. Probably, drought stress accompanied by increased plant mortality (McDowell et al., 2008;Zach et al., 2010) fostered the forest fires as it is known from eastern Borneo during the strong El Niño event from 1997/1998 (Siegert and Hoffmann, 2000;Van Nieuwstadt and Sheil, 2005), when fires also occurred on Sulawesi (Rowell and Moore, 2000;Sastry, 2002). Droughts in Indonesia result from the failure of the monsoon, which often coincides with ENSO warm events (D'Arrigo et al., 2006). According to Quinn et al. (1978), over 90% of droughts in Indonesia during the period from ad 1861 to 1976 are associated with a warm phase of ENSO.
The ~800 year-record obtained from Lake Lamongan (Crausbay et al., 2006; Figure 1a) reveals two periods of multidecadal drought from ~ad 1275-1325 and ~ad 1450-1650 (Figure 7), probably as a result of ENSO variations. More recent investigations on the timing of droughts in East Java have been carried out by Rodysill et al. (2013) who use U-series dating and suggest that the onset of the latter drought at Lake Lamongan was more than 300 years later around ~ad 1790. In sediments from Lake Logung (also East Java; Figure 1a) spanning the past ~1400 years, the long-term trend towards wetter conditions was superimposed by four decadal to centennial-scale droughts between ~ad 930-1130930- , ~ad 1460930- -1640930- , ~ad 1790930- -1860930- and ~ad 1985930- -2008930- (Rodysill et al., 2012 Figure 7). The authors discuss these drought occurrences in relation to both migration of the ITCZ and variability in ENSO. Two of the four droughts (~ad 1460-1640 and ~ad 1790-1860) took place when the ITCZ was displaced to the south, a period that is actually characterized by a wetter mean climate on centennial to millennial time scale. These droughts hence represent unusual events on an interannual to multidecadal time scale.
Comparisons of the data from Lake Kalimpaa with the drought occurrences observed by Crausbay et al. (2006) and Rodysill et al. (2012Rodysill et al. ( , 2013 show that disturbance events at Lake Kalimpaa partly coincide with drought periods in East Java. The older two of four drought periods observed in the Lake Logung record match temporally with disturbance events at Lake Kalimpaa (Figure 7;Rodysill et al., 2012). Therefore, it is suggested that they are regional in spatial extent since age differences are within the range of dating uncertainties. The second drought period (~ad 1450-1620) at Lake Kalimpaa coincides well with findings from Lake Logung (~ad 1460-1640) and the radiocarbon dated drought at Lake Lamongan (~ad 1450-1650; Crausbay et al., 2006) but not with the same drought period applying U-series dating  Figure 7).
These similarities indicate the occurrence of two drought periods that found their expression at least in Central Sulawesi and East Java during this time. Thus, it seems likely that the long-term trend towards higher rainfall intensities and possibly higher mean rainfall was superimposed by individual, interannual to multidecadal-scaled drought periods which were likely associated with intense ENSO warm events. The drought indicated from ~ad 1450 to 1620 occurred during a period when the ITCZ was displaced to the south (Sachs et al., 2009), and wetter conditions were prevailing at Lake Kalimpaa on the centennial to millennial time scale. This may indicate that the movement of the ITCZ is not the main trigger for interannual to multidecadal drought occurrences in Central Sulawesi rather the variability of the Pacific Walker circulation and, hence, ENSO.

Conclusion
The Lake Kalimpaa record is one of only a few terrestrial archives from Sulawesi providing information on palaeoenvironmental and palaeorainfall changes as well as fire history throughout the past ~1500 years. The two main conclusions of this study are the following: (a) in Central Sulawesi, a long-term trend towards wetter conditions (higher rainfall intensities and/or mean rainfall) probably occurred on the centennial to millennial time scale starting from ~ad 560 to the 20th century with highest rainfall intensities during the LIA and (b) two disturbance events (~ad 1090-1190 and ~ad 1450-1620) caused by forest fires occurred in the catchment area of Lake Kalimpaa. A comparison with other records exhibits that the long-term trend towards wetter conditions is associated with migrations of the ITCZ and the millennial-scale variability of ENSO/Pacific Walker circulation. The disturbance events around Lake Kalimpaa are probably related to regional droughts affecting at least East Java and Sulawesi. The occurrence of droughts in Indonesia, which can be accompanied by forest fires, is mostly caused by the failure of the monsoon during ENSO warm events (El Niño years). While a major human impact in the Kalimpaa catchment cannot be excluded completely before the 20th century, this seems rather unlikely considering the regional correlation of drought periods at these times. Figure 7. Comparison of the Fe data from Lake Kalimpaa indicative for disturbance events with total inorganic carbon (TIC) data representing periods of drought at Lake Logung, East Java (Rodysill et al., 2012). The black bars on the right represent periods of drought (based on radiocarbon dating) observed from Lake Lamongan, also East Java (Crausbay et al., 2006). The grey bar on the right represents the time period of the more recent drought at Lake Lamongan on the basis of U-series dating . The greyish bars (representing the core units II and IV of the Kalimpaa record which were deposited when disturbance events occurred in the catchment) are linked to periods of drought obtained from the Lake Logung record. the modified version of the Gradistat 4.2 software. Moreover, we would like to acknowledge Tina Trautmann for supporting the XRD measurements. Michael E. Meadows (University of Cape Town) is thanked for helpful comments on the manuscript as well as for improving the English language. Finally, we would like to thank the two reviewers for their suggestions which helped to improve this contribution distinctively.

Funding
The coring campaign was carried out within the ELUC (Environmental and land-use change in Sulawesi, Indonesia) subproject, which was part of the Collaborative Research Centre SFB 552 'Stability of Rainforest Margins in Indonesia' (STORMA) project and was funded by the DFG.