Can Coastal Upwelling Trigger a Climate Mode? A Study on Intraseasonal‐Scale Coastal Upwelling Off Java and the Indian Ocean Dipole

Coastal upwelling along the southern coast of Java brings cold and nutrient‐rich subsurface water to the surface. We explored whether the upwelling could trigger the onset of the Indian Ocean Dipole (IOD) by supplying cold water to the southeastern tropical Indian Ocean. We used satellite‐based daily chlorophyll‐a concentration (Chl‐a) data during 2003–2020 as a proxy of the coastal upwelling. We focused on first Chl‐a bloom that occurred in April‐June, the onset phase of the positive IOD (pIOD). We found that the timing and strength of the upwelling signals were significantly correlated with the subsequent IOD peaks. We diagnosed processes associated with the upwelling affecting sea surface temperature (SST) in the southeastern Indian Ocean. Results indicate that after the cold‐water upwelling south of Java, westward surface temperature advection plays a role in anomalously cooling the SST in the southeastern Indian Ocean and setting a favorable condition for the subsequent pIOD development.

Various factors can set up a favorable background condition for the onset of the pIOD, such as the El Niño-Southern Oscillation (ENSO) (e.g., Annamalai et al., 2003), biennial monsoon variability in the Indian Ocean (Crétat et al., 2018), and anomalous oceanic and atmospheric condition in the southern Indian Ocean (Fischer et al., 2005;Zhang et al., 2020). Based on these previous achievements, the present study focuses on the coastal upwelling along the southern coast of Java (e.g., Susanto et al., 2001;Wyrtki, 1962), the spatial scale of which is one to two orders smaller than these large-scale factors. Could the coastal upwelling actively contribute to the onset and development of pIOD events, not only as a dependent variation of the large-scale air-sea coupled interaction associated with the IOD? This hypothesis is based on the expansion of cold SSTA from off Sumatra and Java to the southeastern Indian Ocean from May to August observed by satellite and in situ moored buoy data (Horii et al., 2008(Horii et al., , 2009Vinayachandran et al., 2007) during the 2006 pIOD.
On the cold SSTA south of Java, Delman et al. (2016) pointed out that the cooling around May-July was a precursor of a pIOD which is driven by upwelling Kelvin waves. While several studies used numerical models and investigated the ocean temperature variations along the coasts of Sumatra and Java associated with IOD (Delman et al., 2018;Du et al., 2008;Halkides & Lee, 2009), no observational study has directly studied the impact of the seasonal timing of the upwelling onset on the IOD, and assessed the possible contribution of the coastal upwelling to SST variations over the eastern Indian Ocean. We explore the possibility that the anomalous cold water along the coasts leads to subsequent large-scale IOD onset and development. If this possibility holds, understanding the coastal upwelling in the southeastern Indian Ocean could contribute to a better understanding of the IOD and thus a better prediction.
To clarify the coastal upwelling variations around May-June, we need to resolve intraseasonal-scale variations. Past studies showed that there is a short-term occurrence of the upwelling due to anomalous southeasterly alongshore winds (Cao et al., 2019;Chen et al., 2015;Horii et al., 2016) during a certain phase of intraseasonal variation, referred to as Madden-Julian oscillation (MJO) (Madden & Julian, 1994) or intraseasonal oscillation (ISO) (Lawrence & Webster, 2002). However, compared to the open ocean (e.g., McPhaden et al., 2009), there have been fewer such hydrographic time series in the coastal area. Analyses by satellite-observed SST (e.g., Hendiarti et al., 2004) or daily sea level variations (Horii et al., 2016(Horii et al., , 2018 cannot necessarily discriminate the SST cooling due to coastal upwelling. Satellite-based Chl-a data could be a proxy for the coastal upwelling (Siswanto et al., 2020), but because of the large missing values, most previous studies used monthly averages (e.g., Iskandar et al., 2009;Susanto & Marra, 2005). Recently, Xu et al. (2021) used daily interpolated data and outlined the intraseasonal and interannual surface Chl-a variations in the southeastern Indian Ocean.
The present study makes daily time series for the coastal upwelling signals based on the satellite-based Chl-a data. We herein primarily report relationships on the coastal upwelling signal in the onset phase of the IOD and its subsequent evolution. We also diagnose the process of how and to what extent the coastal upwelling can contribute to the SSTA variation in the southeastern Indian Ocean.

Data Sets
We used daily merged surface Chl-a data provided by the Copernicus Marine Service (Garnesson et al., 2019). We also used monthly Chl-a data from the Sea-viewing Wide Field-of-view Sensor (SeaWiFS) for 1998-2007 to observe the climatology. The daily wind data was obtained from the Cross-Calibrated Multi-Platform (CCMP) satellite-based ocean surface wind data set (Mears et al., 2019). The SST data were provided by the National Oceanic and Atmospheric Administration (NOAA) daily optimum interpolation (OI) data set (Reynolds et al., 2007). Using the data set, we calculated Dipole Mode Index (DMI; Saji et al., 1999) as the difference in SSTA between western (50°−70°E, 10°S-10°N) and eastern regions (90°−110°E, 10°S-0°) (Figure 1). To diagnose the ocean surface heat advection, we used the Ocean Surface Current Analysis-Real Time (OSCAR) data (Bonjean & Lagerloef, 2002). To investigate surface heat flux variation, we used the TropFlux data set (Praveen Kumar et al., 2012). To estimate the thermocline variation south of Java, we used hourly sea level data from a tidal station (7.75°S, 109.02°E) along the coast ( Figure S1 in Supporting Information S1). Following the procedure of Horii et al. (2016), we prepared a daily sea level anomaly (SLA) time series with barometric corrections and tides removed.

A Proxy of the Coastal Upwelling
To make a proxy of the coastal upwelling south of Java, we used a Chl-a data set (OCEANCOLOUR_GLO_ CHL_L3_REP_OBSERVATIONS_009_085). The original grids are 1/24° × 1/24° grids (approximately 4.6 km). This product is based on a multiple ocean color sensors but is limited to daily observations, that is, a space-time interpolation is not applied. Due to the large number of missing values off Sumatra, we focused on the region south of Java, where the missing values were relatively small. We analyzed the period from 2003 to 2020 during which the moderate resolution imaging spectroradiometer (MODIS) is incorporated into the merged data.
As a proxy of the coastal upwelling, the average Chl-a data south of Java was computed as follows. (a) A spatial Gaussian filter with an e-folding scale of 1/4° was applied to the original Chl-a data set to produce data with 1/8° × 1/8° grids. Here, the data was calculated only if there were more than 50% data around each grid. (b) A temporal linear interpolation was applied for a missing period within 4 days. These procedures can be applicable here because we focus on time scales more than intraseasonal variability (>20 days) while ignoring ocean submesoscale disturbances. We confirmed that the results did not change even if we traded the order of the (a) and (b) above. (c) The interpolated Chl-a data are averaged over the coastal region ( Figure 1 and Figure S1 in Supporting Information S1), defined as inside the line between (8.0°S, 106°E) and (9.5°S, 114°E), which is approximately 100 km from the Java coast. The spatial scale is appropriate for observing the coastal upwelling signals (Horii et al., 2018). After (a) and (b), available data in the region during April-September in each year was 91.5% on average, and ranged from 75.6% (2010) to 99.7% (2019) ( Table S1 in Supporting Information S1).  Figure S1 in Supporting Information S1 for the map). The light blue and red shades indicate time series of 31-day running-mean Dipole Mode Index (DMI; right axis).
10.1029/2022GL098733 4 of 9 Finally, we estimated the error range using a statistical method (see Text S1 and Figure S2 in Supporting Information S1 for details). The error ranges are shown in Figure S3 in Supporting Information S1.

Comparison to SST and Sea Level Variations
The first peaks of Chl-a from April to August in each year were mostly accompanied by strengthening of upwelling-favorable alongshore winds, SST cooling, and lowering of SLA ( Figure 2 and Figure S2 in Supporting Information S1). Based on our previous study, the lowering SLA is consistent with the shoaling thermocline south of Java (Horii et al., 2018). For example, the Chl-a peaks (>2.0 mg/m 3 ) in mid-June in 2006 and 2007 were concurrent with SST cooling and SLA lowering ( Figure S3 in Supporting Information S1). Here, we defined the first significant Chl-a signal in each year based on the two standard deviations (STD) of Chl-a variability from April to June (0.45). Based on the first signals, we made a composite time series of temporal changes in SSTA and SLA ( Figure 2). The average peaks of the SST cooling and SLA lowering were approximately −3.2 (°C/month) and −39 (cm/month), respectively, and these signals are statistically significant at the 90% level. Significant lowering in SLA first appeared from day −16, and then SST cooling and Chl-a increase occur concurrently. Intraseasonal-scale variation (e.g., Cao et al., 2019) mainly explained these variations ( Figure  S4 in Supporting Information S1). We conclude that coastal upwelling signals south of Java can be observed from the Chl-a proxy during the onset phase of the IOD. We also obtained the same results using a higher resolution (1/20° × 1/20°) SST data set of Group for High-Resolution Sea Surface Temperature (GHRSST; https://podaac.jpl.nasa.gov) instead.

Coastal Upwelling Signals and Subsequent IOD Events
The daily Chl-a time series for the coastal region south of Java, the proxy of the coastal upwelling, shows weak variations from December to March and has seasonal peaks from August to October in most years ( Figure 1). The development and peak of Chl-a are concurrent with southeasterly monsoon winds over the southeastern Indian Ocean (Figure 2). The year-to-year variations show that the Chl-a signals are generally larger in pIOD years such as 2006 and 2019, while the signals are smaller in negative IOD (nIOD) years such as 2010 and 2016. In several pIOD years (Table S1 in Supporting Information S1), significant intraseasonal-scale Chl-a signals above 0.5 mg/m 3 appeared in April−June, earlier than the seasonal coastal upwelling period in July−October. This indicates an early supply of cold water due to a short-term upwelling from April to early June in these years.
We focused on the earlier coastal upwelling signals than the basin-scale SSTA developments in several pIOD years ( Figure 1) and investigated the relationship between the timing of the first Chl-a signal and the subsequent IOD condition in each year (Figure 3a). A daily time series enabled us to observe a significant correlation between the timing of the first upwelling signals and the subsequent IOD peaks: coastal upwelling signals before early June tended to be followed by pIOD events with DMI > 0.5. On the other hand, in neutral years and nIOD years, most of the coastal upwelling signals occurred after mid-June. Most of the first Chl-a blooms are significant, even when taking into account the error ranges ( Figure S3 in Supporting Information S1). Note that the result did not change when the threshold of 2STD was replaced from 3STD to 5STD, whereas this relationship did not hold for 1STD because a sporadic weak Chl-a signal was detected in January-March.  (Table S1 in Supporting Information S1). (b) As in (a), but for Chl-a south of Java. (c) As in (a), but for temporal changes in sea surface temperature anomalies (SSTA) (blue; °C/month) and sea level anomaly (SLA) (black; cm/month) south of Java. The SSTA was averaged in the same region as the Chl-a. Five-day running-mean filters were applied. The values with dots are statistically significant at the 90% level. The shadings represent the 25th-75th percentile ranges.
We also investigated the relationship between the Chl-a amplitude observed in April-June and the subsequent IOD condition (Figure 3b). The Chl-a averaged in April-June was significantly correlated with the subsequent DMI peaks. The average Chl-a signals greater than 0.4 mg/m 3 during April-June were followed by pIOD events. On the other hand, pIOD events did not develop when the average Chl-a was less than 0.35 mg/m 3 , except in 2012. Among 10 pIOD cases (blue), no significant relationship was obtained between the preceding Chl-a signal and the subsequent DMI peaks. The results did not change with the minor modification of the averaged period of Chl-a or for the case in which we used the average DMI around the peaks (August-October).
To determine whether the coastal upwelling signals depend on the large-scale ocean-atmosphere condition, we also checked the correlation between the average Chl-a (April-June) and the Nino-3.4 index or DMI in the same period. The correlation between Chl-a and Nino-3.4 was −0.30, whereas that with DMI was +0.44. This suggests that in the onset phase of the IOD (April-June), it is unlikely that the coastal upwelling signals occurred due to the large-scale change of the tropical Pacific and Indian Ocean associated with ENSO and IOD, respectively. The Chl-a signals in April-June were concurrent with intraseasonal-scale local SST and SLA variations ( Figure S3 in Supporting Information S1).

Anomalous Surface Temperature Advection Associated With the Coastal Upwelling
To investigate whether and to what extent the early occurrence of the coastal upwelling plays a role to anomalously cool the SST in the eastern Indian Ocean, we observe composite SSTA (Figure 4) based on the first Chl-a signal in each year (Table S1 and Figure S3 in Supporting Information S1). After the upwelling around April-June in pIOD years, significant cold SSTA develop in the southeastern Indian Ocean, including the coastal regions ( Figure 4a). The largest cooling is observed south of Java, reaching −1.5 °C in 1 month. The wider distribution of the cold SSTA than the spatial scale of the coastal upwelling is due to the surface heat flux variation. The period (day 0 to approximately +30) follows the easterly phase of the ISO in which the stronger southeasterly alongshore winds enhanced the latent surface heat cooling there (Cao et al., 2019;Horii et al., 2016). For the upwelling in other years, the development of the cold SSTA south of Java is similar to that in pIOD years, although the cooling is confined south of Java, whereas anomalous SST warming is also observed in the west of 100°E.
To diagnose the processes on the expanded cold SSTA observed in the onset phase of the pIOD, we investigate the contribution of anomalous cold-water advection. Anomalous advections are expected to be due to offshore-ward Ekman flow driven by alongshore southeasterly wind anomalies (Wirasatriya et al., 2020) and enhanced northwestward south Java currents (Quadfasel & Cresswell, 1992). Using observational data sets, we estimated the anomalous surface temperature advections. The horizontal mixed-layer temperature gradient and surface currents were estimated using the SST and OSCAR data set, respectively. See the Supporting Information (Text S2 in Supporting Information S1) for details. Although the estimated horizontal temperature advection contains errors of approximately 0.5°C/month in the southeastern Indian Ocean (Horii et al., 2009), it is useful to diagnose SST variations associated with IOD (Horii et al., 2013). We also diagnosed the contribution of surface heat flux variations, assuming a mixed-layer depth (MLD) of 30 ± 10 m (Keerthi et al., 2013).
After the first signal of coastal upwelling in pIOD years, significant cold temperature advection anomalies extend westward from the coastal region to the open ocean. The regional peak around 10°-7°S, 98°-103°E is consistent with that in the SSTA change. The anomalous cold advection is primarily explained by the nonlinear zonal advection (Text S2 and Figure S5 in Supporting Information S1). Anomalously enhanced westward surface currents bring cold temperature anomalies from east to west. The pair of climatological (anomalous) westward currents and anomalous (climatological) temperature gradients also has a secondary role in the cooling ( Figure S5 in Supporting Information S1). On the other hand, cold advection anomalies were observed only south of Java for other years and warm advection anomalies were prominent west of 107°E (Figure 4d).
We diagnose the anomalous contribution of the surface heat flux and horizontal temperature advection to SSTA changes before and after the coastal upwelling signals for the southeastern sector (100°−110°E, 10°S-5°S) (Figures 4a-4d) of the eastern pole of the IOD. The flux contribution was estimated by the net surface heat flux. On the composite of pIOD years, SSTA changes and advection anomalies have similar amplitudes and show significant cooling from day −1 to day +5 and from day −3 to day +12, respectively ( Figure 4e). The flux occasionally contributes to the cooling around day −15, but shows warming after day +18. The total flux does not contribute to cooling in the region. Neither surface heat flux nor advection explains the peak SSTA cooling around day 0. Ocean vertical processes that are not quantified here must play a role in the cooling. Nevertheless, the integrated contribution from the advection for 1 month after the first coastal upwelling signal (day 0) is −0.8°C and is equivalent to the observed SSTA change (−1.0°C). On the other hand, there are no significant SSTA changes in the composite time series for other years (Figure 4f), as the timing of SSTA cooling varies among the cases. From day −30 to day −2, anomalous warm advections from the west due to eastward surface current anomalies significantly contribute to the warming. Unlike pIOD years, anomalous heat flux mainly by the enhanced latent heat flux significantly contributes to the temperature change of −0.5 to −1.0°C with peaks at day 0 and day +15. Although these flux contributions were estimated assuming a MLD of 30 m, the conclusions are the same in case of the MLD of 20 m or 40 m.
These results indicate that intraseasonal-scale coastal upwelling around April to early June in pIOD years leads to SSTA cooling in the southeastern Indian Ocean on a temporal scale approximately 1 month thereafter. This anomalous cold advection prevails only in the pIOD years when an earlier occurrence of the anomalously cold coastal waters is effectively expanded to the open ocean in the form of nonlinear advection anomalies. Implications of the cooling for the IOD onset and development are further discussed in the next section.

Discussion and Conclusions
A daily proxy of the coastal upwelling south of Java showed some relationships between the early occurrence of the upwelling and the subsequent IOD evolution (Figure 3). Anomalous cold advections to the open ocean associated with the coastal upwelling in pIOD years were also demonstrated ( Figure 4). Further question should be how this signal could promote the development of the basin-scale IOD. An earlier supply of cold temperature anomalies into the eastern edge of the tropical Indian Ocean may have an implication for triggering ocean-atmosphere interaction. Since the climatological SST in the eastern equatorial Indian Ocean has its peak in April-May, the cold SSTA south of Java appearing in this period results in a large SST gradient with a spatial scale of approximately 1,000 to 2,000 km. This large SST gradient can provide a favorable condition for Bjerknes feedback, which is essential for the development of the IOD.
The SST cooling by the anomalous cold advections in the southeastern Indian Ocean is observed only in the pIOD years, whereas intraseasonal-scale coastal upwelling signals are present in most years (e.g., Cao et al., 2019). This indicates that the background conditions in the eastern Indian Ocean around April-May are essential for the onset and development of the IOD (e.g., Song et al., 2008). The present study suggests that it would be important to have intraseasonal-scale coastal upwelling under the condition of a shallow thermocline in the eastern Indian Ocean. This leads to subsequent anomalous coolings that expand over a wider area due to nonlinear cold advection anomalies. As a partial evidence of this speculation, significant interrelationship among intraseasonal-scale southeasterly wind anomalies and subsequent SLA and SST decreases during the first Chl-a bloom period was mainly observed in the pIOD years ( Figure S6 in Supporting Information S1). Note that in some years, large-scale dipole SSTA (pIOD) have already appeared before the first upwelling signals, such as in 2007, 2017, and 2018 ( Figure 1). In these cases, the basin-scale SSTA and atmosphere-ocean interaction started by surface heat flux (e.g., Wang et al., 2020), and the coastal upwelling plays a secondary role in further cooling the SSTA. Further studies including the use of numerical models are needed to better quantify the coastal upwelling and its potential impact on IOD onset.
Finally, we would like to note that the relationship between the alongshore southeasterly winds/coastal upwelling south of Java and the subsequent pIOD is similar to that between westerly wind bursts and subsequent El Niño in the tropical Pacific in terms of multi-scale interactions. Although these are opposite anomalous conditions, these are examples of how climate modes develop due to intraseasonal-scale forcing on favorable background fields. Thus, understanding the coastal upwelling signals will help predict pIOD events as well as WWB on El Niño events. We recognize that appropriate observation of the eastern boundary region in the tropical Indian Ocean will contribute to Indian Ocean climate study.