Characteristics of seasonal precipitation isotope variability in Indonesia

The few previous studies of precipitation isotopes (δ18O and δD) in Indonesia, based on low spatial resolution observation datasets, have found several types of patterns in their seasonal variabilities. This study conducted high spatial resolution rainfall sampling and investigated the temporal characteristics of precipitation isotope in Indonesia. Rainfall samples were collected weekly from 33 stations in Indonesia. Cluster analysis showed that Indonesia could be divided into four types based on the seasonal variability of the precipitation of δ18O. The majority of stations showed seasonal patterns in the variability of δ18O, characterized by high values in the dry season (July–October) as type 1. Type 2 also showed one peak of high δ18O but in the longer period (June– November) was similar to type 1 stations. A region of Northwest Indonesia, comprising North and Central Sumatra and western Borneo, was identified as type 3, having two peaks of high δ18O values in January–February and May–August. Another pattern of variability was the anti-monsoonal type, indicated by low δ18O in May–July found in east part of Indonesia. Asia-Australia monsoon regime was the main factor that controls seasonal δ18O variability. This research showed that stable isotope in precipitation could correspond to precipitation climatology in Indonesia.


INTRODUCTION
Because of its geographic location, Indonesia is recognized as part of the "Maritime Continent" (Ramage, 1968), which refers to a region in the tropics with many islands and peninsulas surrounded by numerous seas.The Indonesian part of this region is sometimes called the Indonesian Maritime Continent (IMC).IMC is located in very warm sea water that produces significant convective activities, Correspondence to: Halda A. Belgaman, Kumamoto University, 2-39-1 Kurokami, Chuo-ku, Kumamoto 860-8555, Japan.E-mail: halda.aditya@bppt.go.id playing the role as a heat source for Earth's atmospheric circulation (Yamanaka, 2016).The complex arrangement and interactions of land and sea produce significant seasonal variations of precipitation.The wet season (December-February) in Indonesia is related to the Asian monsoon, which brings water vapor across the region from the Northern Hemisphere.The dry season (June-October) is linked to the Australian monsoon, which brings drier air from the Southern Hemisphere.In addition to the effects of the Asian-Australian monsoon, precipitation variability in Indonesia is also affected by the location of the Intertropical Convergence Zone (ITCZ).Furthermore, as well as seasonal variabilities, precipitation in Indonesia also shows several other patterns of variation on different timescales such as intra-annual and inter-annual variability due to regional and local disturbances.Factors described above (Asia-Australia Monsoon and ITCZ) along with Madden-Jullian Oscillation (MJO) and El-Nino Southern Oscillation at intra-annual and inter-annual time-scales respectively have been proved as a factor that determines the precipitation climatology in Indonesia (e.g.Hamada et al., 2002;Hidayat and Kizu, 2010).Climate classification study based on long-term seasonal precipitation variability in Indonesia had been done by several researchers (e.g.Hamada et al., 2002;Aldrian and Susanto, 2003).Aldrian and Susanto (2003) divided Indonesia into three climatological regions namely annual, semi-annual and local patterns.The annual pattern comprised central and south Indonesia; identified with one peak and one trough of precipitation amount because of strong influences of two monsoons, namely the wet Asian monsoon from November to March and the dry Australian monsoon from May to September.The semi-annual pattern located in the northwest of Indonesia has two peaks of precipitation amount in October-November and in March to May (MAM), those two peaks are associated with the movement of the intertropical convergence zone (ITCZ).The last pattern called local pattern covering Maluku area, has one peak of precipitation amount in June-July and one trough in November-February.Meanwhile, Hamada et al. (2002) divided Indo-nesia into five types of precipitation variability.In this study, we want to differentiate Indonesia climatology type based on stable isotope in precipitation variability.
Stable isotope in precipitation has been widely used in climate science as proxy data for the temperature and amount of precipitation to reproduce the past climate.Various meteorological and climatological factors can be combined to produce unique variability in precipitation isotope value.Isotopic concentrations are expressed as the difference between the measured ratio of the sample and reference over the measured ratio of the reference designated using the delta unit (δ) notation.Ratios of isotopic composition in water are reported in comparison with the Vienna Standard Mean Ocean Water reference standard.In this procedure, an isotopic ratio is reported as the relative deviation on the standard value, given by the following equation: Where R denotes the ratio of the heavy to light isotope.Applied to the pair of ratios HD 16 O/H 2

16
O and H 2 18 O/H 2 16 O, δD and δ 18 O were used, and the units of the ratios are permil or parts of thousand (‰).Dansgaard (1964) reviewed and identified several "effects" on precipitation isotope variability due to several meteorological and geological parameters (i.e.temperature, latitude, altitude, and continental effects).In particular, in the tropical region in which Indonesia is located, an "amount effect" was identified (e.g.Rozanski et al., 1993;Araguás-Araguás et al., 1998).The amount effect refers to the negative correlation between the amount of precipitation and the values of the stable isotopes.Studies on the amount effect in the IMC have shown that it happens because of a fractionation process that occurs across large regions and over extended periods of time (e.g.Cobb et al., 2007;Risi et al., 2008;Kurita et al., 2009Kurita et al., , 2011)).The isotopic signal in precipitation is influenced by a combination of factors, such as precipitation amount, temperature, vapor source, and atmospheric circulation (Vuille et al., 2003).
Studies on the stable isotope of precipitation using observational data from the IMC have been reported previously (e.g.Ichiyanagi et al., 2005;Kurita et al., 2009;Fudeyasu et al., 2011).Seasonal and annual variabilities of δ 18 O in precipitation around Indonesia have been reported by Suwarman et al. (2013) based on data from six observation stations located within the IMC, by Moerman et al. (2013) using data from one observation station in North Borneo Island, Malaysia, and by Permana et al. (2016) based on data from six stations in Papua Island, Indonesia.Considering the vast scale of Indonesia, it was recognized that a greater number of observation points were needed for complete understanding of the spatial distribution of the seasonal variability of δ 18 O across the region and the factors that controls the variability.Therefore, this study was conducted to investigate the seasonal variability of δ 18 O in precipitation in Indonesia using sampling points with a high spatial distribution.

Precipitation sampling and isotopic content measurement
From the end of 2010 until the end of 2012, precipitation samples were collected weekly from 33 meteorological and climatological observation stations in Indonesia belonging to the Indonesian Agency for Meteorology, Climatology, and Geophysics (BMKG).The precipitation samples were collected manually using buckets and stored in 6 ml glass vials with screw caps.Table SI lists name, abbreviations and location of the stable isotope precipitation sampling stations.
δ 18 O and δD in precipitation were measured at the Hydrology Laboratory of Kumamoto University using a cavity ring-down spectrometer (Picarro L2120-i water isotope analyzer).To calibrate the isotopic composition of the rainfall samples, three water standards (Aqua Standard ® DOW-SLW2-ICE2) were analyzed at the beginning and end of each analysis.An internal water standard was also analyzed at the start of each analysis to monitor instrument drift.The standard error of long term reproducibility of the machine against the water standard values are ±0.08 ‰ and ±0.22 ‰ (1σ) for δ 18 O and δ 2 H respectively.
Overall, 2251 bottled samples were measured.Quality control was conducted using d-excess values to determine isotopic enrichment by evaporation during the period of storage between collection and measurement.Ultimately, 2056 samples passed the quality control threshold.

Reanalysis data
An analysis of the vertically integrated moisture transport during the observation period (2010-2012) was performed to emphasize the monsoonal influence on seasonal δ 18 O variability over Indonesia.Moisture flux was obtained by vertically integrating the National Centers for Environmental Prediction (NCEP) reanalysis data of specific humidity and wind fields.The NCEP reanalysis data (Kalnay et al., 1996) were provided by NOAA/OAR/ESRL PSD of Boulder, Colorado, USA (http://www.esrl.noaa.gov/psd/).

Cluster analysis
To analyze the seasonal variability of δ 18 O in precipitation, we calculated the monthly anomaly of δ 18 O for all stations.The anomaly value at each station was derived from the δ 18 O monthly average subtracted by the annual average.The monthly δ 18 O anomaly was then put through the cluster analysis (CA) procedure to determine the groups/clusters.
CA is one of the statistical technique that has been used to make a spatial grouping of observation stations in the field of climatology.CA combines individuals or objects into clusters such that objects within the same cluster are more similar to one another than they are to objects in other clusters.The hierarchical method is the most commonly implemented CA procedures, involving a series of clustering decisions that combine observations into a hierarchy or a tree-like structure.
Ward's method (Ward, 1963), which is one of the most used hierarchical procedures in CA technique in climate research (Kalkstein et al., 1987), calculates the distance as

PRECIPITATION ISOTOPE VARIABILITY IN INDONESIA
the distance of all clusters to the grand average of the sample (sum of the square).At each step, the two clusters that are combined are those that reduce the increase in the total sum of the squares across all variables in all clusters.We used the statistical software OriginPro to conduct the cluster analysis with Ward's method.
The results and progress during CA are conventionally illustrated using dendrograms.A dendrogram is a tree diagram that represents the sequence and the distance at which the observations are clustered.Groups can be selected from the clusters of the dendrogram.The selection of the final cluster solution requires substantial judgment on the part of the researcher, and thus it is often considered a procedure that is too subjective (Hair et al., 2009).One method used to help determine the final cluster solution is to examine the changes in the homogeneity measure to identify significant increases related to the merging of different clusters.
The characteristics of seasonal δ 18 O and precipitation amount variability for each cluster are shown in Figure 3. Cluster 1 is identified by one peak of high δ 18 O in July-October and low δ 18 O in November-June (Figure 3a).The peak terms refer to the δ 18 O value or precipitation amount that was observed above their annual average.Low amounts of precipitation in cluster 1 are observed in July-September, and peak precipitation can be seen in December-February.Cluster 2 is similar to cluster 1 but distinguished by a longer peak of high δ 18 O from June-October and low δ 18 O in December-May (Figure 3b).The peak amount of precipitation in cluster 2 occurs in November-May, and the lowest precipitation can be seen during June-October.Cluster 3 is identified by two peaks of high δ 18 O in January-February, and May-July; and two peaks of low δ 18 O in March-April, and August-December (Figure 3c).The peak amount of precipitation can be observed in two seasons: January-April, and August-December.Even though cluster 4 does not show clear seasonal variability in the quantity of precipitation, the seasonal variability of δ 18 O does show a clear pattern, i.e. low δ 18 O in May-July and high δ 18 O in November-February (Figure 3d).
The peaks (δ 18 O) in clusters 1 and 2 during June-November is related to the Australian monsoon that brings dry air across Indonesia from Northern Australia; these particular months are recognized as the dry season in Indonesia as shown in moisture flux analysis (Figures 4c and d) and 850 hPa relative humidity map (Figures S1c and S1d).In contrast, the smaller δ 18 O value during December-May is related to the Asian monsoon.This brings moist air (Figures S1a and S2b) across Indonesia from the South China Sea and western Indian Ocean (Figures 4a and b), resulting in enhanced rainfall in the Indonesia region.The wet season in clusters 1 and 2 leads to lighter (smaller) δ 18 O, whereas the dry season causes heavier (higher) δ 18 O.
The two peaks of precipitation amount in cluster 3 are associated with southward and northward movements of the ITCZ (Aldrian and Susanto, 2003), which control the amount of rainfall in this region.As for the unclear precipitation trend in cluster 4, Hamada et al. (2002) claimed that there are no specific wet and dry seasons at Biak, Papua (Northwest of Jayapura) but that rainfall is abundant throughout the year, which was confirmed by the findings of this study.In contrast, precipitation isotope variability showed an antimonsoonal pattern, consistent with the monsoonal regime observed by Chang et al. (2005), in this area in June-August.Moreover, seasonal precipitation isotope variability in Peleliu (Northeast of Papua) also showed a similar pattern (Belgaman et al., 2016).
Figure 5 shows scatter plots of δ 18 O anomaly and precipitation anomaly for each cluster to describe the amount effect.The amount effect was explored in two timescales; first in all season or annual amount effect which is the correlation in a whole year, and secondly with correlation for a seasonal (DJF, MAM, JJA, and SON) timescale.The amount effect can be seen clearly for all seasons in clusters 1 (r = -0.94,p < 0.0001) and 2 (r = -0.90,p < 0.0001) in Figures 5a and b, with slopes of -9.06 ‰/mm and -7.06 ‰/mm, respectively.In cluster 3 (Figure 5c), even though all seasonal data show a negative correlation between δ 18 O and precipitation amount (slope: -3.37 ‰/mm), it's not statistically significant (r = -0.43,p = 0.16) and there is positive correlation during September-November (r = 0.99, p < 0.05).In cluster 4 (Figure 5d) an all year amount effect was not observed and not statistically significant (r = 0.08, p = 0.79).A seasonal negative correlation between precipitation and δ 18 O is observed but it's also not statistically significant during December-February (r = -0.41,p = 0.67), March-May (r = -0.48,p = 0.73), and September-November (r = -0.43,p = 0.72).In June-August negative correlation is not observed (r = 0.83, p = 0.38).Based on these characteristics, the seasonal variability of δ 18 O in Indonesia is considered closely related to the amount effect in type/cluster 1 and 2 stations.Amount effect also influences cluster 3 stations, although it is not statistically significant, and no amount effect is observed in seasonal δ 18 O variability in cluster 4 station in this study.The amount effect observed in this study is found based on a monthly averaged dataset.Previous studies have also shown that the amount effect around Indonesia can be detected over larger regions and longer time scales.The amount effect would be observed if the isotopic content was correlated with regional precipitation amount rather than local (station based) precipitation amount.The negative relationship between isotopic composition and precipitation over the Indonesian region would be larger if they were correlated with a monthly time scale rather than a daily or weekly time-scale (e.g.Risi et al., 2008;Kurita et al., 2009Kurita et al., , 2011;;Moerman et al., 2013;Permana et al., 2016).
As stated in the introduction section, δ 18 O variability can be controlled by one or a combination of meteorological factors such as precipitation amount, temperature, and humidity at the vapor source region.Clark and Fritz (1997) stated that due to low seasonal variability of temperature at tropical marine stations, the seasonal variation of δ 18 O at these stations correlates poorly with temperature, owing to the strong seasonality of monsoon precipitation.Even though we show that precipitation amount was controlling the δ 18 O variability in some parts of Indonesia, it was not the only factor.Suwarman et al. (2013) explain in detail the mechanism of vapor origin region affecting the seasonal δ 18 O variability throughout Indonesia.Another finding from Belgaman et al. (2016) shows that vapor originated from the Indian Ocean greatly affects δ 18 O variability at North Sumatera region during MJO events.
In this research, we showed that even though observation period was relatively short (± 3 years), precipitation isotope value (δ 18 O) could describe precipitation climatology condition in Indonesia made by Aldrian and Susanto (2003) (Figure S2).In general, this finding complemented that of Suwarman et al. (2013) who performed a classification of δ 18 O seasonal variability across Indonesia based on only six stations.

CONCLUSIONS
The CA method was used to distinguish the spatial grouping of precipitation isotopic content (δ 18 O) seasonal variability.In Indonesia, four clusters/groups of δ 18 O variability were identified.Clusters 1 and 2 had similar seasonal patterns with the highest/heavier δ 18 O in the dry season or during July to October for cluster 1 and during June to November in cluster 2. The lowest/lighter δ 18 O was in the wet season (November-May for cluster 1 and December-May for cluster 2).Cluster 3 had two peaks of δ 18 O value in January-February and May-August.Cluster 4 had an antimonsoonal pattern with the lightest δ 18 O in May-July.
Spatially, the distribution of four clusters/regions based on δ 18 O variability shows a similar majority with spatial distribution of classification based on precipitation amount by Aldrian and Susanto (2003).Clusters 1 and 2 correspond to region A (annual type precipitation), and cluster 3 located in Northwest Indonesia corresponds to region B.
This research found that the amount effect related to Asia-Australia monsoon regime was the main controller of seasonal variability of δ 18 O over clusters 1 and 2 stations.Meanwhile, precipitation amounts in cluster 3 and cluster 4 are not the factors that control the isotopic content over that region at this study.
A recent study (Lekshmy et al., 2014) has shown that the depletion of δ 18 O in monsoon rain is not related directly to the amount of rainfall but to large-scale organized convection.Further investigation is needed to explain the seasonal spatial variability of δ 18 O in Indonesia and to examine the correlation between δ 18 O spatial variability and water vapor transport over longer observational periods.

Figure 1 .
Figure 1.Dendrogram for the 30 Indonesian Agency for Meteorology, Climatology and Geophysics (BMKG) stable isotope observation stations (vertical axis shows Euclidean distance)

Figure 3 .
Figure 3. Composite monthly average of precipitation and δ 18 O from observation stations: (a)-(d) clusters 1-4, respectively.Horizontal dashed lines indicate annual average value for δ 18 O (blue) and precipitation amount (red)