Today, two types of ENSO can be distinguished: (1) the canonical or conventional El Niño, which is characterized by SST anomalies located in the eastern Pacific near the South American coast (Rasmusson and Carpenter, 1982), and (2) the central Pacific El Niño or El Niño Modoki (Larkin and Harrison, 2005; Ashok et al., 2007; Kug et al., 2009; Ashok and Yamagata, 2009; Gushchina and Dewitte, 2012). In 2003, a new definition of the conventional El Niño was accepted by the National Oceanic and Atmospheric Administration (NOAA) of the USA, in referring to the warming of the Pacific region between 5∘ N–5∘ S and 170–120∘ W. According to Ashok et al. (2007) the central Pacific El Niño, or El Niño Modoki, – i.e., unusually high SST – occurs roughly in the region between 160∘ E–140∘ W and 10∘ N–10∘ S.

As criteria to assess the intensity of ENSO events, a wide range of indexes based on different combinations of sea level pressure and SST data in various areas of the Pacific is used. For diagnostics of the central Pacific El Niño, the SST anomalies (in ∘C) in the Nino4 region (5∘ N–5∘ S and 160∘ E–150∘ W) are commonly used (Fig. 1). The monthly SST anomalies (in ∘C) in the Nino3.4 region (5∘ N–5∘ S and 170–120∘ W) are used to diagnose both types of El Niño phenomenon: canonical and central Pacific (Download Climate Timeseries, 2013).

The tropical rainforest selected for the study is situated near the village of Bariri in the southern part of the Lore Lindu National Park of Central Sulawesi in Indonesia (1∘39.47′ S and 120∘10.409′ E or UTM 51S 185482 M east and 9816523 M north) (Fig. 1). The site is located on a large plateau of several kilometers in size at about 1430 m above sea level surrounded by mountain chains rising above the plane by another 300 to 400 m. Within 500 m around the tower the elevation varies between 1390 and 1430 m. Wind field measurement with a sonic anemometer indicate a slope of around 2–3∘, which is similar to many FLUXNET sites. About 1000 m to the east of the experimental site, the forest is replaced by a meadow; in all other directions it extends for several kilometers (Ibrom et al., 2007).

According to the Köppen climate classification the study area relates to tropical rainforest climate (Af) (Chen and Chen, 2013). Weather conditions in the region are mainly influenced by the ITCZ. During the wet season (typically, from November to April), the area is influenced by very moist northeast monsoons coming from the Pacific. Maximum precipitation during the observation period from January 2004 to July 2008 was observed in April – with 258.0±148.0 mm month-1. The drier season usually lasts from May to October. The precipitation minimum was observed in September with 195.0±48.0 mm month-1. The September–October period was also characterized by maximal incoming solar radiation, up to 650±47.0 MJ m-2 month-1, mainly because of a significant decrease in convective clouds, due to the reversing of an oceanic northeast monsoon to a southeast monsoon blowing from the Australian continent. The mean annual precipitation amount exceeded 2000 mm. The mean monthly air temperature varies between 19.4 and 19.7 ∘C. The mean annual air temperature was 19.5 ∘C (Falk et al., 2005; Ibrom et al., 2007).

The vegetation at the experimental site is very diverse and representative of the mountainous rainforest communities of Central Sulawesi. There are about 88 different tree species per hectare. Among the dominant species are Castanopsis accuminatissima BL. (29 %), Canarium vulgare Leenh. (18 %) and Ficus spec. (9.5 %). The density of trees, with diameter at breast height larger than 0.1 m, is 550 trees per ha. In addition, there is more than times that number of smaller trees per hectare with a stem diameter smaller than 0.1 m. The total basal area of trees reached 53 m2 per ha. Leaf area index (LAI) is about 7.2 m2 m-2. LAI was estimated using an indirect hemispherical photography approach with a correction for leaf clumping effects. The height of the trees, with diameters at breast height larger than 0.1 m, varies between the lowest at 12 m and the highest at 36 m. The mean tree height is 21 m (Ibrom et al., 2007).

CO2 and H2O fluxes were measured from 2004 to 2008 within the framework of the STORMA project (Stability of Rainforest Margins in Indonesia, SFB 552), supported by the German Research Foundation (DFG). Eddy covariance equipment for flux measurements was installed on a meteorological tower of 70 m height at 48 m, i.e., ca. 12 m higher than the maximal tree height. The measuring system consists of a three-dimensional sonic anemometer (USA-1, Metek, Germany) and an open-path CO2 and H2O infrared gas analyzer (IRGA, LI-7500, LI-COR, USA) (Falk et al., 2005; Ibrom et al., 2007; Panferov et al., 2009). The open-path IRGA was chosen due to its smaller power requirements compared to closed-path sensors. The sensor was calibrated with calibration gases two times per year and showed no considerable sensitivity drift within 1 year of operation. Turbulence data were sampled at 10 Hz and stored as raw data on an industrial mini PC (Kontron, Germany). All instruments were powered by batteries, which were charged by solar panels, mounted on the tower. The system is entirely self-sustaining and has been proven to run unattended over a period of several months. Post-field data processing on eddy covariance flux estimates was carried out strictly according to the established recommendations for data analysis (Aubinet et al., 2012). In addition to the procedures described in Falk et al. (2005) and Ibrom et al. (2007), we corrected the flux data for CO2 or H2O density fluctuations due to heat conduction from the open-path sensor (Burba et al., 2008; Järvi et al., 2009) using the suggested method as described in Reverter et al. (2011).

The system operated ca. 70 % of the time. About 30 % of the measured flux data were negatively affected by rain and other unfavorable conditions and removed. From nighttime ecosystem respiration data, a friction velocity (u∗) threshold value of 0.25 m s-1 was estimated (Aubinet et al., 2000), i.e., at u∗ values above this threshold the measured nighttime flux became independent of u∗. Nighttime flux values that were measured at u∗ < 0.25 m s-1 were removed, which left 15 % of the measured nighttime flux data in the data set. In order to fill the gaps in the measured net ecosystem exchange (NEE) and evapotranspiration, net radiation and sensible and latent heat flux records as well as to quantify GPP, RE and forest canopy transpiration, the process-based Mixfor-SVAT model (Olchev et al., 2002, 2008) was used.

Mixfor-SVAT is a one-dimensional model of the energy, H2O and CO2 exchange between vertically structured mono- or multi-specific forest stands and the atmosphere. The main model advantage is its ability both to describe seasonal and daily patterns of CO2 and H2O fluxes at individual tree and entire ecosystem levels and to estimate the contributions of soil, different forest layers, and various tree species to the total ecosystem fluxes taking into account individual structure, biophysical properties and responses of plant species to changes in environmental conditions. The model also allows us to take into account the non-steady-state water transport in the trees, rainfall interception, dew generation, turbulence and convection flows within the canopy and plant canopy energy storage. As model input the measured meteorological variables (air temperature, water vapor pressure, wind speed, precipitation, CO2 concentration, global solar radiation) are used. The model was tested with long-term meteorological and flux data from different experimental sites including the investigated forest under well-developed turbulent conditions and showed a good agreement over a broad spectrum of weather and soil moisture conditions (Olchev et al., 2002, 2008; Falk et al., 2005; Falge et al., 2005). Using the model is superior to common statistical gap-filling approaches because these depend on calibration under all relevant weather conditions, including those that were systematically excluded when the open-path sensor did not work, e.g., under rain. For this reason one might argue that statistical gap filling is biased by calibration during dry weather conditions. The process-based model is, however, able to take these weather situations into account because it is based on general physical principles. As was shown in previous studies, the model is able to predict both CO2 and water fluxes under various weather and soil moisture conditions at sites where closed-path sensors were used (Oltchev et al., 1996; Falge et al., 2005).

Air temperature, relative humidity and horizontal wind speed were measured at four levels above and at two levels inside the forest canopy using ventilated and sheltered thermo-hygrometers and cup anemometers (Friedrichs Co., Germany) installed on the tower. Short- and long-wave radiation components were measured below and above the canopy with CM6B and CG1 sensors (Kipp & Zonen, The Netherlands). Rainfall intensity was measured on top of the tower with a tipping bucket in a Hellman-type rain gauge. To fill the gaps in measuring records the meteorological data from an automatic meteorological station, situated about 900 m away from the tower outside the forest on a nearby meadow, were used. For the analysis, the monthly mean air temperature values and monthly sums of precipitation and solar energy were calculated.

To estimate the possible impact of ENSO events on CO2 and H2O fluxes in the tropical rainforest at Bariri the temporal variability in monthly NEE, GPP, RE and ET in periods with different ENSO intensity was analyzed. To quantify the ENSO impacts on meteorological parameters and fluxes and to distinguish them from effects caused by the seasonal migration of the ITCZ, the intra-annual patterns of CO2 and H2O fluxes as well as meteorological conditions during the measuring period were also evaluated.

In the first step, to assess the possible impact of ENSO events on meteorological parameters (global solar radiation (G), precipitation amount (P), air temperature (T) and CO2 and H2O fluxes), the correlation between the absolute values of monthly G, P, T, NEE, GPP, RE, ET and monthly SST anomalies in the Nino4 and Nino3.4 regions (Nino4 and Nino3.4 indexes) were analyzed.

In the second step, we analyzed the correlation between the deviations of monthly meteorological parameter and flux values from their monthly averages over the entire measuring period and the Nino4 and Nino3.4 indexes. The deviation in the case of GPP (ΔGPP) was estimated as ΔGPPMonth,Year=GPPMonth,Year-1N∑Year=20042008GPPMonth,Year, where GPPMonth,Year is total monthly GPP for a particular month (January to December) and corresponding year (2004 to 2008); 1N∑Year=20042008GPPMonth,Year is monthly GPP for this particular month averaged for the entire measuring period (2004 to 2008); N is number of years. Positive values in ΔGPP, ΔRE, and ΔNEE indicate GPP, RE higher and NEE (carbon uptake) lower than average.

The typical timescale of the full ENSO cycle is estimated to be about 48–52 months (Setoh et al., 1999), whereas the timescale of the main meteorological parameters (global solar radiation (G), precipitation amount (P), air temperature (T)) is characterized by much higher month-to-month variability even after annual trend filtering. In order to filter the high-frequency oscillation in the time series of atmospheric characteristics and monthly NEE, GPP, RE, and ET anomalies, the simple centered moving average smoothing procedure was applied. The moving averages (MAs) of variables were calculated over 7 months (centered value ±3 months).

Statistical analysis included both simple correlation and cross-correlation analysis (Chatfield, 2004). Cross-correlation analysis was used to take into account the possible forward and backward time shifts in maximal anomalies of meteorological parameters and CO2 and H2O fluxes in respect to time of the ENSO culmination. To describe the relationships between atmospheric fluxes and meteorological parameters, the monthly non-smoothed values were used.

Eddy covariance flux measurements in tropical mountainous conditions are challenging. Our tower and eddy covariance system was designed to minimize power consumption by using an open-path sensor, which had the consequence that rainy conditions systematically caused gaps in the flux data. To minimize a potential bias on the flux sums, we used a process-based forest model that is not biased by a lack of data in wet conditions in the same way as the statistical gap-filling algorithms often used are (Reichstein et al., 2005, see also Sect. 2.3). The weather in the tropics typically has a relatively high percentage of calm nights. The selected forest is located on a plateau in a mountainous region, and this increases the risk of CO2-rich air draining downhill in calm nights. We investigated this effect very carefully and found that the CO2 fluxes showed a very clear u∗ threshold above which the nighttime CO2 emission rates did not depend on u∗ anymore. Using only data from nights with sufficient turbulence (u∗ >u∗ threshold value), we minimized advection and drainage affecting the NEE estimates. Also, here we benefitted from the use of the process-based model for gap filling. We then analyzed the statistical relationships between our gap-filled monthly fluxes with climate anomaly indices and corroborated these analyses also with midday NEE data only. As time data are independent from nighttime data, we made sure that our analysis was not affected by nighttime flux loss. The correlations with midday data and ENSO indices were very similar to those with daily mean NEE data. This demonstrated the robustness of our analysis.

In addition we compared the model-predicted mean annual soil respiration rate with soil CO2 efflux data that were measured in the study region with soil chambers (van Straaten et al., 2011). The Mixfor-SVAT model estimated an average annual soil respiration rate of 1110±30 g C m-2 yr-1 for the investigated site. This value was very close to the measured average soil CO2 efflux of the Central Sulawesi region of 1170 g C m-2 yr-1, which shows realistic behavior of the model.

The relatively high annual NEE sums need further investigation. After applying all corrections including the correction for open-path sensor heating and after gap filling, we found an average annual uptake of 782±24 g C m-2 yr-1 (standard deviation between 5 different years). This value is higher than the range found in lowland rain forests, i.e., ranging from, e.g., 75 to 538 g C m-2 yr-1 (Luyssaert et al., 2007). The clarification of this very interesting phenomenon, maybe relating to the site history and regrowth after selected use of large individual trees by the local population, does not, however, lie within the scope of this article.

The main components of carbon and water balances in the tropical rainforest showed a high correlation between Nino4 and Nino3.4 SST anomalies and ΔGPPMA and ΔETMA values over the entire measuring period. The smoothing procedure allowed us to remove the high-frequency month-to-month oscillations in the time series of atmospheric characteristics. These are caused by local and regional circulation processes that are not directly connected with ENSO activity and thus disturb the analysis. The relationships between ΔGPPMA, ΔETMA and Nino4 and Nino3.4 indexes are governed via the dependency of the incoming solar radiation on ENSO development – surface water warming in Nino3.4 and 4 regions generally results in a decrease in cloudiness above the study region and thus in an increase in incoming solar radiation. The high correlation of monthly GPP and ET rates with incoming and absorbed solar radiation at this site is well described (e.g., Ibrom et al., 2008). The effects of monthly air temperature and precipitation changes on ΔGPP and ΔET variability are, on the contrary, relatively poor. ΔTMA, ΔPMA and ENSO intensity are not very much related.

The cross-correlation analysis (Fig. 5) shows that the ΔGPPMA and ΔGMA have a small 2–3 month backward shift relative to the course of Nino4 SST, i.e., the maxima in GPPMA occur earlier than ENSO culmination in the central Pacific (Nino4 SST anomaly). The maximal values of ΔEMA occurred simultaneously with El Niño and La Niña culminations. Such an effect of El Niño episodes on G can be explained, as mentioned above, by a decrease in the cloud cover in the region of Indonesia, due to the El Niño-associated shift in the Walker circulation cell and the corresponding zone of deep convection from the maritime continent of Indonesia toward the dateline, following SST anomaly displacement. El Niño usually begins in April, and toward August–September the ascending branch of the Walker cell leaves Indonesia and migrates eastward to the Pacific. Therefore, 3–4 months before the El Niño culmination in December–January, a decrease in cloud amount is observed over Indonesia. The weakening of El Niño, in turn, leads to a backward, westward shift in the intensive convection zone. It can result in increasing precipitation amounts in the region during the second half of the wet period after passing the maximal El Niño activity, in the gradual increase of the cloudiness, and in a decrease in incoming solar radiation. The opposite effect takes place during the La Niña with similar phase shift: simultaneously, with the spreading of a negative SST anomaly over the Pacific, the increasing of deep convection over Indonesia occurs, which results in an increase in cloudiness and precipitation, being more pronounced as it falls into the dry period of the year. The lower panels of Fig. 4 indicate, however, that the decrease in radiation due to an increase in cloudiness does not depend linearly on La Niña intensity, reaching a saturation state at approximately -20 to -30 MJ m-2 month-1.

A relatively poor correlation between ΔTMA patterns and ENSO activity and an insignificant influence of ΔT on ΔGPP and ΔET can mainly be explained by the small intra-annual amplitude of the air temperature in the study area not exceeding 1.0 ∘C as well as by the low dependence of the air temperature on incoming solar radiation. The mean monthly temperatures ranged between 19.5 and 20.5 ∘C in the intra-annual development. Maximal air temperatures did not exceed 28.5 ∘C, even on sunny days. Such optimal thermal conditions with high precipitation amounts provide sufficient soil moistening and relatively comfortable conditions for tree growth during the whole year. As was already mentioned, even during the El Niño culmination in 2005–2006 the ET / PET did not decrease below 0.74, (ET / PET)MA>0.81, and the relative soil water content of the upper 30 cm horizon was always higher than 80 %.

The analysis of absolute and relative changes in GPP and ET during the periods of maximal El Niño and La Niña activities showed that GPP during the El Niño culminations of 2005 and 2007 increased by about 20 g C m-2 month-1 (6–7 %). ΔGPPMA was about 9 g C m-2 month-1 (2–3 %), ΔET – about 40 mm month-1 (about 30 %) and ΔETMA – about 10 mm month-1 (6–7 %). Thus, the maximal ΔGPP was 2 times lower than the mean annual amplitude of GPP (Fig. 2). The maximal ΔET was equal to the annual amplitude of ET (Fig. 2). During the La Niña culmination of 2008 the maximal relative changes in GPP were higher than the relative changes observed during El Niño events: ΔGPP was about -22 g C m-2 month-1 (8 %), ΔGPPMA – about -12 g C m-2 month-1 (4 %). The maximal decrease in ΔET in the period was relatively small: ΔET decreased by about -12 mm month-1 (10 %) and ΔETMA decreased by about -5 mm month-1 (4 %). ΔET was about 3 times lower than the mean annual amplitude of ET. Interestingly the radiation-dependent GPP (as represented by the smoothed 7-month mean) does not demonstrate any prolonged constant period during La Niña phases though the radiation does. During the first cold event the GPP reduction is not as strong as during the second one, although the G reductions are nearly of the same strength. It could be assumed that in the first case the effect of radiation decrease on GPP was compensated by other factors, like a slight increase in the air temperature.

Additionally, we investigated the influence of other climatic anomalies in the region on CO2 and H2O fluxes in the tropical rainforest, such as the Madden–Julian oscillation (MJO) and the Indian Ocean Dipole (IOD). The MJO is characterized by an eastward propagation of large regions of enhanced and suppressed deep convection from the Indian Ocean toward the central Pacific (Zhang, 2005). Each MJO cycle lasts approximately 30–60 days and includes wetter (positive) and drier (negative) phases. The outgoing long-wave radiation (OLR) measured at the top of the atmosphere is commonly used as an estimation of deep convection intensity in the tropics. It was recently shown that 6–12 months prior to the onset of an El Niño episode, a drastic intensification of the MJO occurs in the western Pacific (Zhang and Gottschalck, 2002; Lau, 2005; Hendon et al., 2007; Gushchina and Dewitte, 2011). Furthermore, MJO behavior varies significantly during the ENSO cycle: it is significantly decreased during the maxima in conventional El Niño episodes, while it is still active during the peak phase of central Pacific events. MJO rarely occurs during La Niña episodes (Gushchina and Dewitte, 2012). As MJO is strongly responsible for intra-seasonal variation of precipitation in the study region, the occurrence of MJO events was compared to the significant anomalies of the ET / P ratio and of key meteorological variables. No evidence of MJO influence is observed: the positive and negative anomalies of the ET / P ratio are associated with positive, negative and zero anomalies of OLR, filtered in the MJO interval. Also, no significant relation emerged from the correlation analysis.

Correlations between the MJO index (Wheeler and Kiladis, 1999; Gushchina and Dewitte, 2011) and the deviations of key meteorological parameters from monthly averages during the study period were very low: r2=0.03 for T, r2=0.03 for P and r2=0.01 for G (p>0.05, in all cases).

The IOD is characterized by changes in the SST in the western Indian Ocean, resulting in intensive rainfall in the western part of Indonesia during the positive phase and a corresponding precipitation reduction during the negative phase (Saji et al., 1999). To find a possible influence of IOD events on temporal variability in meteorological parameters and CO2 and H2O fluxes, the monthly mean IOD index (Dipole Mode Index, DMI) was used. Results showed that with respect to the western part of Indonesia situated close to Indian Ocean, the IOD phenomenon has no significant impact on meteorological conditions and fluxes in the area of Central Sulawesi.

Our case study showed a high sensitivity of the main components of CO2 and H2O fluxes in the investigated mountainous tropical rainforest in Bariri to El Niño and La Niña phenomena as well as a low sensitivity to IOD and MJO events. The time lag between the respective indices and their effect on the fluxes at our site indicates that the timing and the extent of the effects are site specific. The fluxes respond to the local weather and only indirectly to the large-scale weather anomalies, i.e., in the same way that the local weather is affected by the large-scale weather phenomena. The observed phenomena are thus not representative of all mountainous forest sites in the tropics. The conclusion is that large-scale weather anomalies do have systematic effects on local fluxes, but the timing and the extent are likely to differ across different regions.

Even though remote-sensing analyses have shown that the site is representative of the region (Ibrom et al., 2007; Propastin et al., 2012), the response to ENSO might differ in the region due to differences in altitude and land use (Erasmi et al., 2009). In general, anthropogenic deforestation has removed most parts of lowland forests so that the remaining forest cover consists mostly of mountainous forests. At the moment, there are no other FLUXNET sites situated in the equatorial mountainous rainforests of southeast Asia with which we could directly compare our findings and investigate whether a similar response to ENSO can be observed. Most of the existing FLUXNET sites (AsiaFlux) are not comparable with the investigated site as they are situated in subequatorial and tropical climate zones. These are characterized by a higher seasonality of air temperature and precipitation compared to our equatorial site. Thus, our site provides a unique opportunity to investigative the response of an equatorial mountainous rainforest to ENSO in the western Pacific region.