The impact of El Niño southern oscillation and Indian Ocean Dipole on the burned area in Indonesia

Abstract

Land and forest fire have been identified as one of the main problems contributing to forest biodiversity and Global Warming and well known as the phenomenon affected by El Niño Southern Oscillation (ENSO) and Indian Ocean Dipole (IOD). The total burned area becomes higher when either El Niño or positive IOD occur. This research aims to analyze and quantify the direct correlation of the Niño 3.4 and difference between west and east pole of IOD sea surface temperature anomaly (SSTA) to the burned area in Indonesia and the impact of ENSO and IOD of each category on the burned area. The correlation between spatial location with Niño 3.4 and difference IOD SST's will be analyzed using a heterogeneous correlation map. Meanwhile, the quantitative impact will be calculated based on the singular value decomposition analysis result to each year categories. The most significant impact of El Niño has occurred on Merauke following Kalimantan shows the strongest correlation between burned area and Niño 3.4 SST. However, the significant increase of burned area only occurred during very strong El Niño. Both areas can have double amount of burned area during peak fire in very strong El Niño. Moderate El Niño have the most diverse impact with the stronger one occurs on Kalimantan and Merauke. Weak El Niño can have a significant impact if occurred simultaneously with positive IOD. Even more, it can surpass the effect of a single Moderate El Niño. Meanwhile, the strongest IOD impact happened in the southern part of Sumatra.

Key points

• 1. Correlation map analysis between ENSO and IOD’s sea surface temperature with burned area.

• 2. Behaviour analysis of burned area in Indonesia in response to ENSO and IOD.

• 3. Identify and quantify the impact of ENSO-IOD’s teleconnection in Indonesia.

Introduction

In recent years, prolonged droughts and intensive use of tropical forests have increased land and forest fire events (FAO 2007). Land and Forest fires have been identified as one of the main environmental issues that have a long-term impact on biodiversity and global climate (Ganjam and Sudhakar 2015). In Indonesia, land and forest fires occur annually on Central Sumatra, South Sumatra, South Kalimantan, and Merauke (Nurdiati et al. 2019). The number of peatland areas in those regions produces high carbon emissions during the dry season (Osaki et al. 2016; Septiawan et al. 2019). The biggest forest fire event in Indonesia for the last 25 years happened in 1997. Most of the burned area in 1997 came from widespread peatland fires causing unusually high carbon emissions in that year (Page et al. 2002). In 1997, 9,755,000 hectares (ha) burned in Indonesia (Luca Tacconi 2003).

1997 El Niño year fire event causing dry peat smoulders for long periods and burns down to the water table. When this happens, the tree roots are exposed, and the peat and forest vegetation become unstable. Therefore, it generates peat subsidence, a massive tree falls, and the consequent loss of large forest areas, leading to more problems for flora and fauna (Harrison et al. 2009). Although the event happened in Indonesia, haze from the fires can extend to Malaysia, Singapore, and Thailand, shrouding them in smoke and affecting economic activities for millions of more people, resulting in billions of more dollars of financial losses (Harrison et al. 2009). Even more, the carbon emission from Indonesia's peatland wildfires makes a massive contribution to the increase of global emission (Hooijer et al. 2006). The second biggest one happened in 2015 with 4,604,569 ha were burned during the fire season (overall accuracy 84%) (Lohberger et al. 2017). Both of those two events happened during strong El Niño in each particular year. Moreover, El Niño events explain most of the year-on-year variation in fire (Edwards et al. 2020).

ENSO is a global phenomenon that describes the recurring pattern of climate variability in the eastern Pacific Ocean, characterized by SSTA and anomaly sea level change (Neelin et al. 1998; Hu et al. 2017). The warming phase of sea surface temperature (SST) describes the El Niño event, while the cooling phase describes the La Niña event. During El Niño events, SST in the Eastern Equatorial Pacific become hotter than normal conditions. Meanwhile, during La Niña conditions, SST in the East Equatorial Pacific are lower than normal conditions. The occurrence of El Niño causes the rainy season to come later and prolongs the dry season duration in Indonesia. Normally dry season in Indonesia occurred around may until October, while the rainy season occurred from November until April. Otherwise, La Niña impact on Indonesia causes the rainy season to come earlier and a shorter dry season. In the last 25 years, two strong El Niño happened in 1997 and 2015. Even though 2015 El Niño have highest Niño 3.4 index in those periods (L’Heureux 2016), 2015 El Niño event was not the most extensive fire event in the total burned area or fire emissions (Burton et al. 2020). A few moderate El Niño happened in 2002, 2006, and 2009 (Burton et al. 2020).

In the past few years, much research shows strong El Niño in 1997 and 2015 on Indonesia’s land and forest fire (Edwards et al. 2020; Tan et al. 2020; Yulianti and Hayasaka 2013). Many people in Indonesia, Southeast Asia, and worldwide expressed deep concern at the extent and severity of fires in Sumatra and Kalimantan forested areas in the late 1997 dry season. It is generally believed that the health of over 20 million people was adversely affected in Southeast Asia in late 1997 because causes were not well analyzed first (Byron and Shepherd 1998). Field et al. (2016) state that longer-term records of airport visibility in Sumatra and Kalimantan show that 2015 ranked after 1997 and alongside 1991 and 1994 as among the worst episodes on paper.

Another global phenomenon that affects the 2015 drought season in Indonesia is Indian Ocean Dipole (IOD) (Avia and Sofiati 2018). The IOD phenomenon has two phases: the dry phase called positive IOD and the wet phase known as negative IOD. A positive IOD is indicated by the SST of the western Indian Ocean (Africa) being warmer than the eastern Indian Ocean (Indonesia and its surroundings). In contrast, the opposite condition indicates the negative IOD (Li et al. 2002; McPhaden and Nagura 2014). In the transitional period of the dry season, coincided with the lasting period of the positive IOD phase, there will be increasingly severe drought and widespread Indonesia (Avia and Sofiati 2018). When positive IOD and El Niño occurred concurrently, Indonesian rainfalls decreased significantly than in the single event of positive IOD or El Niño (Nur’utami and Hidayat 2016). In general, positive (negative) phases of the IOD result in rainfall extremes becoming drier (wetter) than the normal phase. However, this generalization impacts ENSO and IOD for all Indonesian regions (Kurniadi et al. 2021). In addition, the relations between precipitation and land and forest fire were not uniform in Indonesia (Fanin and Werf 2016). Therefore, the impact of ENSO and IOD are different in each area of Indonesia, which frequently suffers from land and forest fire.

Other than global climate phenomena, land and forest fires in Indonesia are also affected by human behaviour. Examples of the land and forest fire caused by human behaviour are land clearing, shifting cultivation (agricultural waste), transmigration, reforestation, oil palm/rubber and other big-scale plantations (Luca Tacconi 2003). However, Kemen et al. (2019) showed that the cause of the fires across Indonesia from 2001 to 2016 has no apparent pattern. The leading causes of land and forest fire are vary depending on the location, such as oil palm plantation on Kalimantan and logging roads in Papua. Meanwhile, Sumatra has three leading causes: oil palm plantations, grassland/shrublands, and small-scale agriculture (Kemen et al. 2019). Furthermore, variation of leading causes is also found on an annual scale, such as oil plantation in 2013, small agriculture and plantation in 2015, grassland/shrubland in 2014 and 2016. Even though land and forest fire happened annually, the scale of the event is much smaller than during ENSO years (Luca Tacconi 2003). Therefore, this research only focused on causes of land and forest fire related to the climate.

Nurdiati et al. (2021a, b) examine the spatial and temporal analysis of the El Niño effect on forest and land fires on Kalimantan and Sumatra using the hotspot as a fire indicator. There is a direct spatial and temporal correlation between hotspot-ENSO SST and hotspot-IOD SST (Dafri et al. 2021). However, the result of those research cannot be used to directly describe the impact of ENSO and IOD on the burned area in Indonesia. Proportion of burned area recorded as hotspot is less than 0.39 (Tansey et al. 2008). Even though there are some improvements in the recent year, correlation between hotspot and burned area is different on each area caused by different characteristics of the region (Tansey et al. 2008; Vetrita and Cochrane 2020; Ardiansyah et al. 2017). Meanwhile, analyzing burned areas in Indonesia and assessing corresponding impacts, e.g., CO₂ emissions, can be potentially used for climate/economic damage assessments (Krasovskiy et al. 2018). Therefore, this research analyses the interrelationship patterns of burned area and precipitation in Indonesia with the influence of ENSO and IOD. The first objective of this research is to analyze the correlation between global phenomena (ENSO and IOD) to the burned area in Indonesia for each observed grid. After that, a detailed analysis was provided by looking at the different joint patterns for each possible ENSO and IOD category for each region with an annual forest fire. Lastly, this research indicated ENSO-IOD teleconnection impact on the land and forest fire in Indonesia. This research can explain how ENSO and IOD affect forest fires in Indonesia related to the burned area and monthly average precipitation clearly to contribute positively to forest fire modelling in the future.

Data and methodology

Data

This research’s land and forest fire indicator is monthly burned area data from 1997 to 2016 produced by Global Fire Emission Database (GFED), and MODIS burned area data from 2017 to 2020. The spatial information for the burned area data is 95° E–141° E and 6° N–11° S. The burned area variable in GFED4.1s data has a grid resolution of 0.25° × 0.25° for each data element, while in MODIS data has a grid resolution of 0.25° × 0.25° and 0.1° × 0.1°. Therefore, the GFED4.1s data is rescale using information of MODIS data from 0.25° × 0.25° to 0.1° × 0.1°. The value of those data represents the proportion of observed grid that burned. Therefore, the value will be converted into km² units (1 grid = 111,321 km²) multiplied by 100 to convert the value into ha. The data described in Giglio et al. (2013) and van der Werf et al. (2017), as well as can be accessed and downloaded for free from the Global Fire Data website.

Meanwhile, the precipitation variable uses a monthly precipitation average from ECMWF Reanalysis 5th Generation (ERA5) data. The spatial and temporal information for the data is same as burned area data. For the ENSO and IOD variables, this research uses SSTA in ERA5 data from 1991 to 2020 (Hersbach et al. 2019). The spatial information for both phenomena corresponds to the Niño3.4 index and DMI index. Nino 3.4 index is an anomaly of 190° E–240° E, 5° N–5° S SST (Trenberth 1997). The anomaly is based on the average of last 30 years data (1991–2020). Meanwhile, DMI index is anomalous SST gradient between west IOD (50° E–70° E, 10° N–10° S) and east IOD (90° E–110° E, 0° N–10° S) (Saji et al. 1997). Because of different size of spatial information, the east IOD SST is interpolated so that it has same spatial dimension with the west one. Both of data are monthly data with 0.25° × 0.25° spatial resolution, produced by the European Centre for Medium-Range Weather Forecast (ECMWF) and can be accessed for free (Hersbach et al. 2019).

Singular value decomposition

Singular value decomposition (SVD) is matrix decomposition method that factorize matrix $A$ in the form of:

$$A = U\Sigma V^{T} ,\;{\text{with}}\;\upsigma _{1} \ge\upsigma _{2} \ge \cdots\upsigma _{r} \ge 0.$$

(1)

$\Sigma$ represent the singular value matrix of $A$ with $\upsigma _{i}$ represent the singular value of ith mode. $U$ and $V$ represent left and right singular vector of $A$ respectively. Therefore, the equation can be written (Nicholson 2001):

$$A_{n \times p} = U_{n \times n} \left[ {\begin{array}{*{20}c}\upsigma & 0 \\ 0 &\upsigma \\ \end{array} } \right]V_{p \times p}^{T} .$$

(2)

The singular value decomposition in this research uses the cross-correlation concept to analyze the joint pattern between burned area and precipitation or SSTA. The idea is described by Navarra and Simoncini (2010). Suppose there are two matrices, namely matrix $P$ (burned area) and matrix $S$ (precipitation/SSTA), that have the same time span. In that case, a Cross-Correlation matrix is formed following the equation $C = PS^{T}$. The decomposition is applied to the matrix $C$. Therefore, $P$ and $S$ can have different spatial dimensions (number of rows). As long as the temporal dimension (number of columns) is the same, the decomposition process can be done. The result from SVD analysis of Cross-Correlation matrix will be called joint pattern to describe pattern of P that related specifically with pattern of S. Since it calculate related pattern of two variables, dominant pattern of each variable when calculated individually is not always become dominant joint pattern.

Empirical orthogonal analysis

The Empirical Orthogonal Function (EOF) analysis aims to transform the original correlated $p$ variables into $k$ orthogonal (uncorrelated) components. Let $X$ be a matrix of size $n \times p$ containing a data set with n number of variables and p time. SVD of $X$ with $rank(X) = r$ is the factorization as in Eq. (1) so that we get:

$$X = \mathop \sum \limits_{i = 1}^{r} u_{i} \sigma_{i} v_{i}^{T} .$$

(3)

The $V$ matrix of Eq. (1) is the EOF matrix or vector coefficient and $U\Sigma$ is the EOF score matrix or principal component. Principal component scores can be written:

$$z_{i} = u_{i}\upsigma _{i} ,i = 1,2, \ldots ,r.$$

(4)

The variance of ith principal components $i = 1,2, \ldots ,r$ is

$$\mu_{i} = {{\sigma_{i}^{2} } \mathord{\left/ {\vphantom {{\sigma_{i}^{2} } {\sum\limits_{i = 1}^{r} {\sigma_{i}^{2} } }}} \right. \kern-\nulldelimiterspace} {\sum\limits_{i = 1}^{r} {\sigma_{i}^{2} } }},$$

(5)

where i = 1, 2, …, r is the singular value of the matrix X. In practice, the k mode of the 1st EOF or the first principal component describes the largest proportion of variance. The 2nd EOF mode has the second-largest variance and so on. The kth EOF mode has the kth maximum variance independent of the previous EOF mode (Hannachi 2004).

Heterogeneous correlation maps

The spatial analysis used the Heterogeneous Correlation Map (HCM) to describe the correlation of hotspots in Indonesia with variances values of climate indicators. HCM hotspots in Indonesia have been obtained from the Pearson correlation between expansion coefficient of climate indicators variances with burned area matrix $r(V^{T} S,P)$ (Björnsson and Venegas 1997). Ranges of Pearson correlation value can be seen in Table 1.

Table 1 Ranges of Pearson correlation value

From the scratch three-dimensional matrix data (longitude × latitude × time), we rearrange that matrix to two-dimensional matrix data (space × time). Therefore, suppose that P and S are the two-dimensional matrices representing the burned area and SSTA with a matrix size of $n \times p$ and $m \times p$, respectively. By decomposing the cross-correlation matrix (PS^T) using SVD (Eq. 1), the U, $\Sigma$, and V matrices are constructed. The first EOF mode of HCM is the Pearson correlation between the expansion coefficient (the product of the first column of V notated by $v_{1}$ and the S matrix) and the burned area matrix P, i.e., $r(v_{1}^{T} S,P)$. The expansion coefficient is a time vector with a size of $1 \times p$. The HCM is obtained by calculating the Pearson correlation of the time vector $v_{1}^{T} S$ with every space dimension of matrix P so that the first EOF mode of HCM has a size of $n \times 1$. Remember that n is a spatial dimension consisting of longitude and latitude dimensions. Therefore, we can visualize the correlation map for each grid location in Indonesia by transforming the n-vectors to a longitude-latitude matrix. In this study, the value of m and p are 1160 and 240, respectively. Meanwhile, the value of m is 8241 and 6537 for the SSTA of ENSO and IOD, respectively. The result of HCM analysis will be called correlation pattern since it analysed the correlation of related pattern between two variables.

Euclidian distance and dynamic time wrapping

Euclidian distance and Dynamic Time Wrapping (DTW) are a well-known method that can compare two signals by finding the distance between the two signals (Bayarjargal et al. 2006; Lhermitte et al. 2011). Euclidian distance between $X = (x_{1} \ge x_{2} \ge \cdots \ge x_{n} )$ and $Y = (y_{1} \ge y_{2} \ge \cdots \ge y_{n} )$ can be calculated following the equation below:

$$D_{euclid} = \sqrt {\sum\limits_{i = 1}^{n} {(x_{i} - y_{i} )^{2} } } .$$

(6)

While the DTW distance of $X = (x_{1} \ge x_{2} \ge \cdots \ge x_{n} )$ and $Y = (y_{1} \ge y_{2} \ge \cdots \ge y_{n} )$ is defined as:

$$D_{tw} (X,Y) = d(m,n),$$

(7)

$$d(i,j) = \left\| {x_{i} - y_{i} } \right\| + \min \left\{ {d(i,j - 1),d(i - 1,j),d(i - 1,j - 1)} \right\},$$

(8)

$$d(0,0) = 0;\;d(i,0) = \infty ;\;d(0,j) = \infty ;\;i = 1,2, \ldots ,m;\;j = 1,2, \ldots ,n.$$

In this research, the Euclidian distance was used to get the different characteristics among each category in the same month. Meanwhile, the DTW distance is used to nullify the shift between the characteristic of the signal by comparing the value in the same index and comparing the value with the value of the previous and next index to find the optimum distance (Muller 2007).

Results and discussion

In this section, we present the analysis that is divided into two groups. The first one analyses correlation map between burned area and SSTA (Niño 3.4 and IOD). The second one explores spatial and temporal character of the burned area and precipitation for each year group divided based on the overall ENSO and IOD strength. As already mentioned before, ENSO and IOD’s impact cannot be generalized for all regions in Indonesia. The second analysis will be carried out for each region with recurrent land and forest fire events.

Analysis of correlation map between burned area and SSTA (ENSO and IOD)

SSTA of ENSO and IOD are variables that have a dominant joint pattern with a hotspot in Indonesia (Dafri et al. 2021; Nurdiati et al. 2021b). In this research, SSTA of ENSO and IOD is analyzed with burned area data from 1997 to 2016 to get correlation map that indicates the relation of ENSO and IOD with burned area in Indonesia respectively. The first analysis uses ENSO’s SSTA, and the second one uses IOD’s SSTA.

The correlation map analysis between burned area and ENSO’s SSTA result in one dominant pattern with 98.78% value of variance explains (Fig. 1). This pattern represents 98.78% of events from 1997 to 2020 in the correlation of burned area and ENSO. The pattern shows that the largest burned area happened in 1997; more than 15 million ha burned in a month (Fig. 1a). The second one occurred in 2015, with a maximum value of more than 11 million ha burned in a month. However, each grid location’s spatial correlations with ENSO’s SSTA were less than 0.5 (sufficient correlation). The higher correlation occurred from the centre to the eastern part of Indonesia (Fig. 1b). This result coincident with previous research which shows cold SSTA from centre to the eastern part of Indonesian sea during El Niño (Hendon 2003) resulting less precipitation (Nur’utami and Hidayat 2016), more burned area on the surrounding region (Pan et al. 2018).

Fig. 1

The correlation map between burned area and ENSO: a the HCM and burned area principal temporal pattern b violin plot of correlation coefficient on Riau (blue box), South Sumatra (red box), Kalimantan (black box), and Merauke (green box). Red dots and stars represent the average value and outlier of correlation distribution in each area

Maximum correlation between the temporal pattern of burned area and ENSO is 0.381 (sufficient correlation, Table 1). Those values represent the correlation between monthly total burned area in Indonesia and monthly Niño 3.4 index. However, the temporal pattern only captures the fire event in the middle to the end of the year. These periods represent the dry season that occurred caused by Australian Monsoon wind. Although the peak of El Niño occurs in winter, the burned areas do not appear in winter because Indonesia has already experienced the rainy season.

There are two land and forest fire events in Indonesia which happened in the beginning and the middle of the years (Septiawan et al. 2019). The second dry season which occurred in the middle of the year becomes longer when El Niño develops. Those different periods make land and forest fire patterns at the beginning of the year missing from the correlation map. However, these patterns can be identified in the violin plot of Riau by the negative value of correlation. These negative values represent the occurrence of burned area in Indonesia when ENSO is not yet developed.

The correlation map analysis between burned area and IOD also resulted in one dominant pattern with 96.15% value of variance explained. Similar to the previous analysis, the joint pattern only captured burned area in the middle to the end of the year. This period coincides with the impact of IOD on Indonesia’s monsoonal season, so the temporal plot also showed only from June to November for each year. The maximum spatial correlation between burned area and IOD is 0.464 (sufficient correlation), which occurred on South Sumatra. Due to the spatial location of east IOD’s SSTA (90° E–110° E, 0° N–10° S) which is cooler than normal during positive IOD is in the middle of Sumatra and Kalimantan which resulting less rainfall on the surrounding region. Therefore, the spatial correlation of surrounding regions higher compared to the ENSO. Those results can be seen in the violin plot of South Sumatra and Kalimantan, which shows a higher average (indicated by the red dot) and maximum value than the ENSO one. Otherwise, Merauke’s violin plot show the opposite behaviour with a lower average and minimum value than ENSO.

The violin plot of Riau (Fig. 2) shows higher maximum value of correlation than the ENSO one (Fig. 1), but the average is lower. Moreover, the minimum value belongs to negative weak correlation which indicate different land and forest fire event happened oppositely with IOD. As mentioned before, Riau is a region with two forest fire events in a year, which happened in the beginning (February until March) and middle (June until September) of the year. Considering the impact of positive IOD’s is start from June (Yulihastin et al. 2009), we can conclude that events that have weak positive correlation occurred in the middle to end of the year. The weak negative correlation represents land and forest fire events that happened at the beginning of the year. Due to the high dominance of events in the middle of the year, the first EOF mode cannot capture the beginning of the year events.

Fig. 2

The correlation map between burned area and DMI: a the HCM and burned area principal temporal pattern b violin plot of correlation coefficient on Riau (blue box), South Sumatra (red box), Kalimantan (black box), and Merauke (green box). Red dots and stars represent the average value and outlier of correlation distribution in each area

The low value of spatial and temporal correlation means that Indonesia’s land and forest fire influenced by many factors. Two of them are El Niño and positive IOD that affected the duration of Indonesian dry season in the particular year. However, the temporal pattern in Figs. 1 and 2 looks similar to each other. The first reason for this similarity is the very high dominance pattern in the burned area data. The second reason for this similarity is the interconnection (relationship) between ENSO and IOD. In 2011, Cai et al. stated that the Niño3.4 index was moderately correlated with the DMI index in June–August (JJA). Even more, the correlation becomes stronger in September–November (SON). The maximum positive correlation between ENSO and IOD occurred when the IOD led ENSO by 2 months, while the maximum negative correlation occurs when IOD is leading ENSO by 16 months (Stuecker et al. 2017).

Classification analysis of joint pattern between burned area and precipitation

This section analyses the joint pattern between burned area and precipitation by grouping the year 1997–2016 to each possible category that builds from ENSO and IOD categories (Behera et al. 2013; Anteneh et al. 2019). The year groups that obtained from 1997 until 2016 data are:

• 1. Normal year: 1998, 1999, 2001, 2003, 2010, 2012, 2013

• 2. Weak El Niño and normal IOD (WEN-NI): 2004, 2014, and 2018

• 3. Weak El Niño and positive IOD (WEN-PI): 2006 and 2019

• 4. Moderate El Niño and normal IOD (MEN-NI): 2002 and 2009

• 5. Very strong El Niño category and positive IOD (SEN-PI): 1997 and 2015

• 6. La Niña and normal IOD categories for 2000, 2005, 2007, 2008, 2011, 2017, and 2020

• 7. La Niña and negative IOD categories in 2016

The impact of La Niña in the 6th group increases the monthly precipitation in Indonesia and make the dry season shorter. Therefore, the correlation between burned area and precipitation decreases, making the analysis less relevant. Even more, the 7th group with the contribution of negative IOD makes the precipitation on Indonesia even higher and makes the correlation even more irrelevant. For that reason, the analysis will be carried out in groups 1–5. The previous research states that the recurrent land and forest fire occurred mainly on Riau, South Sumatra, Kalimantan, and Merauke. Therefore, the analysis will be carried out only in those specific regions following the average order correlation in ENSO HCM from the highest to lowest one.

Merauke

Land and forest fires on Merauke during strong El Niño in 2015 significantly impacted Timika. Regional Disaster Relief Agency (BPBD) of Merauke claimed 93 hotspots were detected from burning forests and land activities (Oja et al. 2019). Lohberger et al. (2017) show that more than 800 thousand ha burned, with around 500 thousand ha occurred on peatland. Those data contributed to the total burned area in Indonesia by approximately 20–25% during 2015’s fire event. HCM analysis in Sect. 3.1 also gives sufficient correlation between burned area and ENSO/IOD, indicating that Merauke is a region prone to land and forest fire events.

The temporal pattern between burned area and precipitation on Merauke shown by Fig. 3. Each line in burned area plot (Fig. 3a) represents an average of total burned area in the composite year respective to each category, while in precipitation plot (Fig. 3b) represents the average monthly precipitation of each analysed grid in the composite year. Figure 3 shows that forest fire events on Merauke commonly happened from August until November, except for WEN-PI and SEN-PI categories, which can end in December. The peak of the fire event occurred either in September or October, but mainly in September. Meanwhile, the monthly average precipitation decreases to less than 5 mm/day starting from June until October. Therefore, the peak fire event happened due to an accumulated dry condition from June to September. The high proportion of burned area that occurred in peatland and the geographic location of Merauke make the amount of burned area remain high in October. Therefore, forest fires on Merauke relatively have longer duration than other regions (Figs. 5, 7, and 10).

Fig. 3

Temporal of joint pattern for each category on Merauke. Burned area graph (a) represents the average total burned area of the composite year. In contrast, the precipitation graph (b) represents the average monthly average precipitation from each grid of the composite year for each category

The impact of IOD on Merauke is not visible in the burned area temporal pattern (Fig. 3). The WEN-PI burned area graph is lower than the WEN-NI graphs. In precipitation, positive IOD decreases the precipitation value until reaching around 1 mm/day in October, the end of the dry season. In other months, the impact is unclear when comparing WEN-NI and WEN-PI graphs. Considering the result of the HCM analysis in Sect. 3.1, which shows a weak to sufficient correlation on Merauke, we can state that positive IOD affects the joint pattern of burned area and precipitation of Merauke only in the end period of the dry season. When positive IOD occurs, the condition is relatively similar with other categories during the start of dry season (June–August), but it causes high burned area and low precipitation during the end of dry season (October–December).

By looking at burned area graph in Fig. 3, we can see that burned area on Merauke is affected by the strength of El Niño. On average, around 35 thousand ha was burned during peak fire events in the Normal year. The peak event of weak El Niño can result in around 0.13 million ha burned. The number of burned areas during moderate El Niño is also 0.11 million ha, and the strong El Niño will cause more than 0.28 million ha burned only during the peak event (Fig. 3). This result shows the sensitivity of land and forest fire on Merauke to the strength of El Niño, especially strong El Niño. Weak and moderate El Niño can cause two to three times area burnt compared to a normal year. Even more, a strong El Niño can lead to five times amount of burned area than normal year. Meanwhile, El Niño impact on Merauke’s precipitation has prolonged the duration of the dry season not only by holding the arrival of the upcoming rainy season but also making the dry season come earlier from May or June. This effect resulting in 2 months longer dry season compared to normal conditions. Therefore, the accumulated dry condition from early dry season results in a much more burned area than the normal condition.

Figure 4 shows the spatial of the first joint pattern between monthly average burned area (ba) and monthly average precipitation (prec) on Mearuke. Other than the normal category, the value subtracts with the normal category’s value to represent the difference between each category with the normal category. The difference in the spatial pattern of precipitation is inconsistent in WEN-NI, WEN-PI, and MEN-NI. However, this result coincides with the result in Fig. 3, which shows a relatively similar number of burned areas during those three categories. Even though the precipitation graph of MEN-NI, WEN-NI, and WEN-PI in Fig. 3 is lower than normal condition, the difference of average precipitation is only in the range of 0–1.5 mm/day (Fig. 4) per grid data. Even more, there are some increases of average precipitation during MEN-NI on same area. Therefore, we can consider that WEN-NI, WEN-PI, and MEN- have a similar impact on the burned area and precipitation of Merauke. Different from SEN-PI where the difference can be seen clearly in both Figs. 3 and 4.

Fig. 4

Spatial of joint Pattern from each category on Merauke. In a normal graph, the colour represents an average of total burned area for each grid in the composite year only during fire season (July–December). Other categories represent the difference between each category with value in a normal year

Kalimantan

Kalimantan, also known as Borneo, is the region with the most significant contribution to the total burned area in Indonesia for the last 25 years. In 2015 Kalimantan contributed 45–50% of the total burned area in Indonesia, with one-third of burned area on Kalimantan occurred on peat (Lohberger et al. 2017). Figure 5 shows that the number of burned areas from the peak fire event that commonly happened in September increases following the strength of El Niño and positive IOD. During a normal year, around 0.15 million ha was burned on average during the peak fire event. Following the strength of El Niño, the number of burned areas increases to be around 0.3 million ha during WEN-NI and around 0.75 million ha during MEN-NI. When positive IOD occurred during weak El Niño, around 0.7 million ha was burned during peak fire. However, the high number of burned areas in WEN-PI lasts until October, resulting in a longer duration of fire event than MEN-NI. Furthermore, SEN-PI doubled the impact of WEN-NI and WEN-PI, resulting in more than 1.5 million ha burned on average during the peak fire event and lasted until November.

Fig. 5

Temporal of joint pattern for each category on Kalimantan. Burned area graph (a) represents the average total burned area of the composite year. In contrast, the precipitation graph (b) represents the average monthly average precipitation from each grid of the composite year for each category

The impact of ENSO and IOD also appeared in the precipitation variable (Fig. 5b). During peak dry season (July–September), the monthly average of precipitation is less than 3 mm/day for all categories, except the normal year. Furthermore, the precipitation is below the peak of dry season in normal years from July until September, representing much more severe than normal year. El Niño and positive IOD prolonged the duration of the dry season by holding the arrival of the next rainy season. Besides the normal year, the large increases in precipitation represent the rainy season from October to November, which is 1 month later than the normal year. Figure 6 shows the spatial of the first joint pattern between burned area and precipitation for each category on Kalimantan. The normal category has the least value of variance explains caused of the less fire that happened in the category, so that correlation between the burned area and precipitation is reduce. Figure 6 (normal) shows that during a normal year, up to 700 ha was burned on one grid data during one fire season in the region in a range of 114°–116° E and 2°–4° S where average precipitation value during the dry season is around 4–7 mm/day. The number of burned areas in those regions is increased in WEN-NI and MEN-NI following the decreased precipitation. The decreased precipitation is less than 2 mm/day for each grid, the impact can increase burned area by around 2500 ha in WEN-PI and up to 5000 thousand ha per grid during SEN-PI.

Fig. 6

Spatial of joint Pattern from each category on Kalimantan. In a normal graph, the colour represents an average of total burned area for each grid in the composite year only during fire season (July–November). Other categories represent the difference between each category with value in a normal year

WEN-PI and SEN-PI is causing more decreased precipitation in the region in a range of 110°–114° E and 1.7°–4° S, causing more burned area in this region (Fig. 6). This result is related to the SST that contributes to positive IOD located on the west of Kalimantan. Therefore, the west part of South Kalimantan has less precipitation during positive IOD compared to normal IOD. From this result, we can conclude that the western burned area region on Kalimantan (110°–114° E, 1.7°–4° S) is more influenced by positive IOD, while the eastern one (114°–116° E, 2°–4° S) is affected by both El Niño and positive IOD. Not only increase the burned area of the western part, but WEN-PI also increases the burned area on the eastern part to the same level as the MEN-NI category. The impact is even more in SEN-PI with more decrease’s precipitation.

South Sumatra

For the last 25 years, South Sumatra has been commonly known as a region that suffers from land and forest fire for almost every year. Centre of the event is usually in Palembang. The high amount of peatland makes the fires difficult to extinguish, especially when long dry season occurred in 1997 and 2015 during the strong El Niño (Kirana et al. 2016). The joint pattern analysis between burned area and precipitation shown in Figs. 7 and 8. Figure 7 shows that the fire event on South Sumatra commonly happened from July to November, which is the entire monsoonal dry season on South Sumatra. Figure 7 shows that the peak of the fire event always happened in September for each category. The minimum precipitation value is less than 5 mm/day for each category. On average, the amount of burned area during one fire season is spread from 150 to 750 thousand ha. However, if we separate WEN-PI and SEN-PI from the categories, the amount of burned area is in the range of 100–200 thousand ha for each category. Meanwhile, the peak fire event in WEN-PI burnt more than 350 thousand ha, and the peak fire in SEN-PI has more than 700 thousand ha burned area during one fire season on average.

Fig. 7

Temporal of joint pattern for each category on South Sumatra. Burned area graph (a) represents the average total burned area of the composite year. In contrast, the precipitation graph (b) represents the average monthly average precipitation from each grid of composite year for each category

Fig. 8

Spatial of joint Pattern from each category on South Sumatra. In a normal graph, the colour represents an average of total burned area for each grid in the composite year only during fire season (July–November). Other categories represent the difference between each category with value in a normal year

From the burned area graph, we can see that El Niño and IOD that occurred coincidentally impacted the number of the burned area shown by the WEN-PI and SEN-PI graphs. Even weak El Niño can have double impact than that of moderate El Niño, if accompanied by positive IOD. The impact of El Niño and positive IOD is reinforced by joint spatial pattern for each category, which shows how the burned area spread over the region. Figure 8 shows that the most widespread increases in burned area and decreased precipitation occurs in WEN-PI and SEN-PI. This widespread will be causing more burned area in total, as shown by Fig. 7. WEN-NI and MEN-NI graph confirm the previous result that the impact of El Niño from weak to moderate is quite comparable if positive IOD did not occur (Fig. 7). The impact of El Niño is significantly noticeable when strong El Niño occurred. This statement is reinforced with analysis in Sect. 3.1, which shows that IOD strongly correlates with the burned area over South Sumatra while ENSO only has sufficient correlation. However, the impact of WEN-NI and MEN-NI in the average precipitation variable is less with only − 0.2 to 0.5 mm/day difference compared to normal year (Fig. 8). This result is the impact of both phenomena is prolonged the duration of South Sumatra’s dry season. At the same time, the precipitation in the peak condition remains similar value, as shown in Fig. 7.

From the normal years temporal pattern in Fig. 7b (the right panel), the dry season on South Sumatra happened when the average monthly precipitation value under 6 mm/day which is during June to September. When either El Niño or IOD occurs, the precipitation value on June is around 5 mm/day (beginning of dry season) and decreases in to around 2–3 mm/day during July to October which is significantly drier than normal year. Even more, the precipitation value become around 1 mm/day when both phenomena occurred simultaneously (Fig. 7, WEN-PI and SEN-PI). In addition, the dry season duration in WEN-PI and SEN-PI are 1 month longer than other categories with less than 6 mm/day precipitation until November, leading to much more burned areas during the dry season.

Riau

Riau is one of the regions in Indonesia that has a low rainy season. Therefore, Riau has two rainy seasons and two dry seasons in 1 year. The consequence of having two pairs of seasons is that the season’s duration is shorter than regions that only have one rainy season and dry season in 1 year. The duration of each season is around three months, including the transition period of the season. This short duration of the dry season makes the number of burned areas on Riau much less than other regions with annual land and forest fire. However, the two dry seasons on Riau also make two land and forest fire patterns occurred on Riau. Septiawan et al. (2019) describe patterns at the beginning of the year from January until March and in the middle of the year from June until August. Therefore, the singular value decomposition and empirical orthogonal analysis will be done separately to get a better result. The first one is done from January to April and May to December for the second one. In order to make an easier comparison between both dry seasons, the temporal plot for both analyses is joined. Meanwhile, the spatial pattern is shown separately. The joint pattern analysis of Riau is shown in Figs. 9 and 10.

Fig. 9

Temporal of joint pattern for each category on Riau with (a) represent burned area and (b) represent precipitation. The January–April graph is obtained from the 1st dry season analysis, while the May until December graph is obtained from the 2nd dry season, which is done separately

Fig. 10

Spatial joint Pattern on Riau (1st ba, 1st prec, 2nd ba, 2nd prec). In a normal graph, the colour represents an average of total burned area for each grid in the composite year only during fire season (January–March for the 1st dry season, May–November for the 2nd dry season). Other categories represent the difference between each category with value in a normal year

Figure 9 shows the combined joint pattern from both forest fire seasons on Riau. The worst condition for both dry seasons (February for the first dry season and June/July for the second dry season) is comparable with less than 4 mm/day on average for each grid. However, the range value for the second dry season is tighter (3.5–5.5 mm/day) than the first one (2–5.5 mm/day). The average monthly precipitation on Riau is 6.29 mm/day. Using 6.29 mm/day as an upper bound of dry season, duration of the second dry season is longer than the first one. The second dry season has 3–4 months of less than 6.29 mm/day precipitation, while the first dry season only has 1–2 months. Therefore, the fire season of the second one is also longer (Fig. 9a). This feature is not an effect of the separated analysis. The result is nearly identical when the singular value decomposition is applied to the whole data. Duration of the second fire season is even longer when strong Niño and positive IOD occurred, resulting in up to 6 months of high burned area. The impact of ENSO and IOD to burned area are significant in SEN-PI category, while the other categories have a comparable impact. However, all the categories have a similar impact on the precipitation of the second dry season. Meanwhile, the impact in the first dry season is inconsistent.

Figure 9 shows that the 1st fire season on Riau was not affected by El Niño’s strength or IOD’s strength. A normal year and MEN-NI categories show that almost 40 thousand ha were burned during peak fire events on average. The WEN-PI category has more than 20,000 ha burned during peak fire events, while SEN-PI has less than 20 thousand ha burned. This result corresponded with the dry season period before the El Niño and IOD effect occurred. The highest burned area occurred in WEN-PI category, which is influence by the fire’s event in 2014 caused by the cold phenomenon and the Intertropical Convergence Zone (ITCZ) contraction to the south and Madden Julian Oscillation (McBride et al. 2015). Those two events are resulting in more than 50 thousand on average during WEN-NI in February and March. The second fire season on Riau occurred from May until November. Figure 9 (left) shows that the peak of the events is different for each category from June to September. However, the burned area graph is likely affected by the El Niño and positive IOD. On average, the peak event from the WEN-NI is around 50,000 ha, 65,000 ha for WEN-PI, 55,000 ha for MEN-NI, and more than 80,000 ha for SEN-PI. Meanwhile, the peak fire event of the normal year has around 50 thousand ha burned in June. Previous research states that in June 2013, the total precipitation was far below normal (of 25 years average), 145.06 mm. The precipitation deficit starts from April, allowing the accumulation of dry-spell (number of days without drain) and causing many hotspots that lead to the burned area in June (Kusumaningtyas and Aldrian 2016). If we exclude the 2013 event from the normal category, the impact of ENSO and IOD are more consistent following the strength of ENSO and IOD.

Figure 10 shows the spatial of the first joint pattern between monthly average burned area (ba) and precipitation (prec) on Riau during the first and second dry season. The odd row represents burned area variable, and the even row represents the precipitation variable. The spatial pattern of the first dry season (1st and 2nd row) shows no consistent increases value of burned area or decreased value of precipitation following the increase of El Niño and IOD strength. Both temporal and spatial analysis of the first dry season explains the negative correlation in Figs. 1 and 2 on both spatial and violin plots of Riau. This result confirms the previous result that the first dry season on Riau was not affected by the strength of El Niño and IOD. The higher number of the burned area occurred anomalously and more correlated with the dry season condition during the respective years.

Figure 10 (3rd row and 4th row) shows that both burned area and precipitation in the second dry season affected by El Niño and IOD. It can be seen that the burned area increases and spread wider following the strength of El Niño and IOD, especially when both phenomena occur (SEN-PI and WEN-PI). Even though the difference is not as clear as burned area variable, the precipitation plot also shows wider area of decrease precipitation in the stronger El Niño and positive IOD categories. During WEN-NI and MEN-NI, some areas even have less burned area and higher precipitation than normal year. This result shows that the impact of El Niño is weaker than positive IOD on Riau, which can be seen by comparing the WEN-PI with MEN-NI spatial pattern. The WEN-PI gave a wider area with increases burned area and decreases precipitation compared to MEN-NI. This result is related to the location of Riau, which is closer to a location that contributed to IOD than El Niño.

Identifying the impact of teleconnection between ENSO and IOD

The previous section analysis shows that the peak from most of Indonesia’s land and forest fire events happened in September, except events that occurred on Riau that did not affect ENSO and IOD. For the other three regions, both land and forest fire and the dry season ended in November. ENSO and the IOD are positively correlated (Cai et al. 2011). Therefore, identifying the independent impact of ENSO and IOD that occurred in the same dry season is difficult. The correlation between ENSO and IOD is called ENSO-IOD teleconnection. In the past century, the ENSO-IOD teleconnection has been found to get stronger (Yuan et al. 2018). This section identifies the impact of ENSO-IOD teleconnection on the land and forest fire pattern in Indonesia.

The analysis will be provided using Dynamics Time Wrapping (DTW) and Euclidian distance methods. Both methods calculate the distance between signals in a normal year with a signal in each category to compare the whole signal easily among each category. The bigger value represents the bigger difference between both signals. The analysis is carried out to the mean and maximum signals to impact the average and worst conditions. The mean signal contains the average value of all analysed grid for each month, while the maximum signal contains the maximum value of analysed grid. Strong El Niño and positive IOD that occurred simultaneously had the strongest impact both on the peak and whole land and forest fire signal in all calculations. Therefore, the comparison will be focused on the WEN-NI, WEN-PI, and MEN-NI categories.

The yellow row in Table 2 is where El Niño and positive IOD occurred in the same year. Table 2 shows that, in general, the impact of El Niño and positive IOD that occurred simultaneously is bigger for all regions even compared with moderate El Niño. From the previous section analysis, both positive IOD and El Niño affect the dry season duration. Therefore, the dry season potentially becomes even longer when both phenomena happened compared to a single phenomenon. In general, the WEN-PI has higher impact than MEN-NI on both methods except on Merauke. However, the value of WEN-NI, WEN-PI, and MEN-NI of Merauke is not much difference among each other represent similar impact to the burned area. This result reinforces previous statement that IOD has less impact on Merauke which have overlapping temporal graph of burned area and precipitation during WEN-NI, WEN-PI, and MEN-NI (Fig. 3).

Table 2 DTW and Euclidian distance results for each category

The least difference also occurred on Riau reinforced the result of previous segment. Even so, the impact of WEN-PI is consistently higher than MEN-NI on Riau. The big difference is occurred on Kalimantan and South Sumatra. Even though Kalimantan has higher peak fire event during MEN-NI than WEN-PI (Fig. 5), MEN-NI impact through the whole signal is less than WEN-PI. This result related with longer fire season of Kalimantan during WEN-PI which is ended in December (Fig. 5). In line with previous result, South Sumatra is consistently affected by the occurrence of El Niño and positive IOD. Compared to normal year, WEN-PI have significantly higher value than WEN-NI or even MEN-NI. The occurrence of positive IOD during weak El Nino can lead to much severe dry season and fire event on South Sumatra.

This result indicated that ENSO-IOD teleconnection has a strong impact on Indonesia's land and forest fire signal. The effect is not always resulting in the biggest peak fire event, but a bigger fire event in the whole dry season caused by the longer duration. This result is supported with previous research about combined roles of the El Niño type and the IOD phase in modulating fire activities in Indonesia, especially on South Sumatra and Kalimantan (Pan et al. 2018). They found that carbon emissions are higher in 2006 (WEN-PI) than in 2009 (MEN-NI) in both southern Sumatra (83.8 versus 29.0 g·C·m⁻²·month⁻¹) and southern Kalimantan (104.8 versus 63.4 g·C·m⁻²·month⁻¹), even though 2006 is a weaker El Niño year by 0.17 °C in terms of Niño 3.4 index. Positive IOD during weak El Niño year could further contribute to the drier conditions and thus more intensive fire activities, which are usually unexpected from a weak El Niño year indicated by the relatively low Niño 3.4 index. (Pan et al. 2018).

This research is limited with the availability of burned area data in Indonesia which is fundamental to the climate analysis. Most of the classification only contains 2 years event in the last 24 years. Thus, it cannot be guaranteed that future event will have same pattern and characteristic explained in this research due to the lack of sample. However, the result is still can be used to give rough picture of dry season and land and forest fire condition during specific classification of ENSO and IOD based on last 24 years data. Considering future rainfall conditions in Indonesia will become drier than the historical condition during El-Nino and positive IOD (Qalbi et al. 2017) which can lead to more severe land and forest fire, various research needs to be done to help anticipate the event even if there are fundamental limitation.

Conclusion

ENSO and IOD are phenomena that positively correlate with land and forest fire and precipitation in Indonesia. In terms of burned areas, the strongest impact of El Niño is occurred on Merauke following by Kalimantan. The impact is diminished when affected in the western region, with the least impact appeared in land and forest fire on Riau (Fig. 1). Meanwhile, the strongest impact of IOD occurs in land and forest fire on South Sumatra and Kalimantan (Fig. 2). However, the severity of the peak event only increased dramatically when strong El Niño and positive IOD occurred simultaneously. Weak to moderate El Niño impact is depended on the characteristic of each region. The difference between impact between weak and moderate El Niño is insignificant on South Sumatra and Merauke, while it is significant on Kalimantan. Both El Niño and positive IOD phenomena make the duration of the dry season even longer than usual, resulting in more total burned areas (Figs. 5, 7, 9). Because of the same period of occurrence, the impact of weak El Niño combined with positive IOD could be higher than singular moderate El Niño for all regions. This result indicated the important effect of teleconnection between ENSO and IOD on Indonesia's precipitation and burned area. The high percentage of peatland during the fire’s event makes the impact of the long dry season caused by weak El Niño and positive can cause more problem than shorter but more severe dry season caused by moderate El Niño.