Significant Precipitation Anomalies over Indonesia in the Aftermath of iod Events

Indian Ocean Dipole (IOD) is one of the prominent climate phenomena affecting Indonesia. The well-documented impacts of IOD are dry and wet anomalies during positive and negative IOD events, respectively, which are clearly detected during the development and mature phases of IOD in the boreal summer to autumn (June-November). IOD impacts on other seasons were questionable because of unclear associated signals. Here, we find significant precipitation anomalies in the aftermath of IOD events, which reach a peak in the following spring (March-May). The delayed pattern is distinctive with the typical impact of IOD. During positive IOD, the quick recovery and warming of IOD-induced sea surface temperature (SST) in the eastern Indian Ocean due to increasing shortwave radiation is suggested to play a role. When the influence of El Niño-Southern Oscillation (ENSO) is linearly removed, the aftermath impacts of IOD in the western regions appear to be suppressed because of weaker positive (SST) anomalies. While the eastern regions show anomalous increase on precipitation due to anomalous increase of convergence. This finding reveals the complex impacts of IOD and ENSO and may benefit to developing a better seasonal forecast over Indonesia.


Introduction
Precipitation in Indonesia is modulated by many global factors, such as monsoon and other climate variabilities (Aldrian & Susanto, 2003).One of the climate variabilities that affects Indonesia's precipitation is Indian Ocean Dipole (IOD).IOD is an atmosphere-ocean phenomenon characterized by strong sea surface temperature (SST) dipole on Tropical Indian Ocean accompanied by changes in the equatorial wind direction (Saji, et al., 1999;Webster et al., 1999).IOD events can be identified by the change in Dipole Mode Index (DMI), calculated from the difference of SST anomalies on western (50°E-70°E, 10°S-10°N) and south-eastern (90°-110°E, 10°S-EQ) Tropical Indian Ocean (Saji, et al., 1999).IOD has two phases, the positive and negative phases.On positive phase, eastern Tropical Indian Ocean has lower SST anomaly than the western Tropical Indian Ocean.Easterly wind anomaly appear on Tropical Indian Ocean, and eastern Africa has increased convective activities, while western Indonesia has reduced convection (Webster et al., 1999).On the negative phase, Eastern SST anomaly 1245 (2023) 012032 IOP Publishing doi:10.1088/1755-1315/1245/1/012032 2 is higher, and westerly wind anomaly appears with lower magnitude than positive phase's easterly wind anomaly (Vinayachandran et al., 2002).
Previous studies have been conducted on the effects of IOD on June-November to the precipitation over Indonesian region during its phases.On positive IOD, precipitation on Indonesia is lower, while it increases on negative IOD (Ashok et al., 2003;Saji et al., 1999;Vinayachandran et al., 2002).Regional view of the IOD effects show bigger impacts over the central and southern part of Sumatra, Java Island, and smaller region of influence on Borneo, Sulawesi, and Papua.When the effect of El Niño-Southern Oscillation (ENSO) is removed, affected regions reduced to only Sumatra, Java, and Borneo (Kurniadi et al., 2021).However, on other seasons the impact of IOD is still unclear.Fig. 1 shows increased precipitation after positive IOD events.
This study investigates the pattern of precipitation on Indonesian region related to IOD events, focused on other seasons after the events have occurred (December to March).We employed total correlation and partial correlation method to describe IOD and precipitation relationships using Dipole Mode Index and other variables explained on Section 2. In Section 3, the results presented and explained and then summarized in section 4.

Data
In this study, we analyzed precipitation, SST, velocity potential on 850mb, and outgoing longwave radiation (OLR) data.DMI and Nino 3.4 indices were used to determine IOD and ENSO years.We obtained SST data from ECMWF Reanalysis v5 (ERA5) Reanalysis dataset, precipitation data from Global Precipitation Climate Project (GPCP) monthly dataset, velocity potential data from NCEP/NCAR Reanalysis Monthly Means dataset, and OLR data from NOAA Interpolated Outgoing Longwave Radiation dataset.The time period for the DMI and Nino 3.4 ranges from 1979 to 2021 and for other variables we use data from 1979-2020.
The intensity of IOD is measured from SST anomaly differences between the Western Tropical Indian Ocean (50°E-70°E and 10°S-10°N) and South-Eastern Tropical Indian Ocean (90°E-110°E and 10°S-0°N).These values are used as indices for IOD (DMI) (Saji, et al., 1999).We could then identify IOD years and its phases using these indices.DMI is accessible from NOAA ESRL Physical Sciences Laboratory website (https://psl.noaa.gov/gcos_wgsp/Timeseries/DMI/).Nino 3.4 indices are calculated from the mean of SST on Nino 3.4 region (5°N-5°S, 150°W-90°W).Five-month running mean calculated from these indices can be used to identify ENSO years and phases (Trenberth, 1997).This study uses Nino 3.4  The GPCP Monthly Dataset provides precipitation data merged from satellite and in-situ observation with the spatial resolution of 2.5°.Dataset is accessible from the website https://psl.noaa.gov/data/gridded/data.gpcp.html.
The NCEP/NCAR Reanalysis 1 project provides atmospheric data on a spatial resolution of 2.5°.This study used this dataset to obtain monthly velocity potential data on 850 mb pressure height.Dataset is accessible from the website https://psl.noaa.gov/data/gridded/data.ncep.reanalysis.html.
OLR data were obtained from NOAA Interpolated Outgoing Longwave Radiation dataset and has the spatial resolution of 2.5° and monthly temporal resolution.OLR values in this dataset came from satellite observations with gaps filled by spatial and temporal interpolation.Data is accessible from the website https://psl.noaa.gov/data/gridded/data.olrcdr.interp.html.

Method
At first, we calculated climate variable anomalies constructed from removing its monthly climatological mean and then smoothed with three-month running average to remove intraseasonal signals from these values.We then quantified the relations between IOD and precipitation anomaly by calculating their correlation coefficients.Since precipitation anomaly is influenced by IOD and ENSO phases, partial correlation can be applied to remove the influence of ENSO.Partial correlation shows the relations between two variables (in this case, IOD and precipitation) after the relation of other variables (ENSO) is removed, or supposed to be constant (Cramer, 1946).For this study, the equation for partial correlation is given by the formula where  , .
is the partial correlation of precipitation anomaly (PR) and DMI after removing Nino 3.4 influence,  , is the correlation between PR and DMI,  , is the correlation between DMI and Nino 3.4, and  , is the correlation between PR and Nino 3.4.Correlation between two variables are calculated using Pearson's correlation coefficient formula: where  is the Pearson correlation coefficient.We also applied Student's t test to determine area with over 95% confidence level.The test statistic for  :  , .= 0 : where  is the number of conditioning variables (Nino 3.4).Since in this study we need to determine the relationships of IOD to precipitation anomalies not only when dipole mode events occured, but also after the events have transpired, we applied both total and partial correlation of DMI to precipitation anomalies and other climate variables on the periods of September-November (lag +0 month) and two seasons after dipole mode events, December-February (lag +3 month) and March-May (lag +6).

Results
Partial correlation coefficients of IOD (DMI) with variables precipitation, SST, OLR and velocity potential anomalies were calculated using ENSO (Nino 3.4) as controlled variable to see relationship of IOD after ENSO influence is removed.The results were also compared with total (bivariate) correlation coefficients between DMI and other variables.Using correlation method, September-November averaged values of DMI and Nino 3.4 were used to test their relationship with September to November (lag +0 months), December to following year's February (lag +3 months), and following year's March to May (lag +6 months) averaged values of other mentioned variables.

Precipitation
Total correlation of DMI and PR (Fig. 3) shows negative correlations within the majority of the Indonesian region on September-November (lag +0 month) indicating decreased precipitation anomalies on positive IOD events, while precipitation anomalies will be increased on negative IOD events.This shows similar results with previous study (Kurniadi et al., 2021).Partial correlation shows smaller regions of influence and magnitudes.On lag +3 months (from December to February of the following year), partial correlation in the region also shows smaller region influenced by IOD compared to total correlation.The change is quite large in comparison, region with correlation magnitude ≥ 0.2 reduced by more than half of region shown from total correlation.There are small regions with positive correlations (Sumatra and Papua) that indicates positive precipitation anomalies occured after positive IOD events and negative precipitation anomalies occured after negative IOD events on these regions.
On lag +6 months (following year's March to May), total correlation shows majority of Indonesian regions are positively correlated with DMI, especially around the islands of Java and Sumatra, indicating increased precipitation anomalies occured after positive IODs, and reduced precipitation anomalies after negative IODs.Partial correlation shows smaller correlation magnitudes around Java and Sumatra, while eastern Indonesian region has stronger relationship in terms of correlation magnitude compared to the results shown by total correlation.Java and Sumatra region has the magnitude around 0.2-0.4 while eastern Indonesian region has magnitudes above 0.4.Fig. 3 Spatial distribution of time-lagged total correlation coefficients (a-c) and partial correlation coefficients (d-f) of DMI and PR.From left column to right, average of September-November DMI values are correlated with average precipitation anomalies on September-November (a,d), December-February (b,e), and March-May (c,f).Dotted lines show area with 95% confidence level.

Sea Surface Temperature
Fig. 4 Shows the total and partial correlation of DMI and SST anomalies.On lag +0 month, total correlation shows most of the sea region has negative correlation with magnitude ≥ 0.6, which indicates the heating of sea surface around indonesia is highly influenced by the IOD.Partial correlation shows the region affected by IOD is reduced in size, and the magnitude of the correlation coefficients are reduced.
On lag +3 and +6 months of total and partial correlation shows reduced area of influence with positive correlation coefficients after the effect of ENSO is removed.This indicates stronger heatings occured after IOD events and the heatings are influenced both by IOD and ENSO events.Stronger heatings affects evaporations and cloud formations on these affected regions.Fig. 4 Same as in Fig. 3 but with SST anomalies.

Outgoing Longwave Radiation
Total correlation of OLR (Fig. 5) on lag +0 months shows positive correlation coefficients between DMI and OLR anomalies on Indonesian region.On positive IOD events, the formations of clouds are reduced while on negative IOD events cloud formations are increased.Partial correlation result has smaller magnitudes of coefficient compared to total correlations, but still yields positive signs around the region.
On lag +3 months partial correlation result yields smaller regions of influence compared to total correlation.This means there is significant influence of ENSO on the OLR anomalies.From December to February after Positive IOD events more clouds appear on the region, smaller in magnitude than during the IOD events which will reduce the heating of the surface.The opposite pattern will occur after negative IODs, with smaller magnitude compared to during negative IOD events.
On lag +6 months total correlation yields large region with positive correlation around Java Island, and after ENSO effect is removed, the affected region around Java Island is reduced.Previous test with SST shows that the area around Java Island has lower SST heating after the effect of ENSO is removed which explains smaller magnitude on the region.Reduced heating will lead to smaller evaporation, and resulted in smaller precipitation anomaly around March-May.However, the pattern in the eastern region of Indonesia is different.After the effect of ENSO is removed, the region has negative correlation coefficients compared to total correlation result.On lag +3 months partial correlation resulted in increased region with positive correlation coefficients on Sumatra, Java and part of Borneo.And on lag +6 months,there are regions with positive correlation coefficients on eastern Indonesia.This means after positive IOD events convection around these areas are increased, causing more clouds to appear around the eastern region.Fig. 6 Same as in Fig. 3 but for velocity potential anomalies.

Summary
The result of our analysis shows that after positive IOD events, there are significant precipitation anomaly on Indonesian region.After the effect of ENSO is removed, the magnitude and the affected region of the anomaly around Java Island is reduced, while around eastern Indonesia precipitation anomaly increased in both magnitude and affected region.
Analysis of variables SST, OLR, and velocity potential shows the two regions mentioned above has different mechanisms that affects their precipitation anomalies.Around Java Island, after the events of IOD changes on the SST heatings occured.After the effect of ENSO is removed, the magnitude is reduced and the decreases the rate of the cloud formation.Eastern Indonesian region shows different mechanism, after the effect of ENSO is removed, there are increased convective activities which resulted in more coulds to form above the region.

Fig. 2
shows the 3-monthly time series of DMI and Nino 3.4 index over 43 years.Both the DMI and Nino 3.4 tends to fluctuate at timescales larger than 1 year period, indicating interannual phenomena.

Fig. 2
Fig. 2 Time series of seasonal mean (3 monthly) of (a) DMI and (b) Nino 3.4 indices during the period of 1979-2021.Dotted lines are the threshold for identifying years of each respective climate variabilities based on the definition of Verdon and Franks (2005) for IOD events (a) and the definition of Trenberth (1997) for ENSO events (b).

7 3. 5 .
Velocity Potential Fig.6shows total and partial correlation of DMI and velocity potential anomaly.During positive (negative) IOD events, Indonesian region has decreased (increased) convection.Partial correlation shows smaller magnitude which indicates the decrease (increase) of convection activities are also influenced by ENSO.