Quantifying the influence of the Madden–Julian oscillation on rainfall extremes during the northeast monsoon season of the Philippines

Abstract

This study investigates the impact of the Madden–Julian Oscillation (MJO) on the extreme rainfall events across 11 eastern coastal stations during the northeast monsoon season (November to March) over the Philippines from 1979 to 2019. The contribution of synoptic systems to these extreme rainfall events such as tropical cyclones (TCs), low-pressure systems (LPS), cold surges (CS), and other disturbances as they coincide with a strong and active MJO were quantified. The results show that the probability of extreme rainfall occurrence increases first to as much as 20% in the southernmost stations in Phase 4. Then, it increases to more than 60% in central-eastern stations in Phase 6. The extreme rainfall events were then classified into: MJO-only, TC-MJO, TC-nonMJO, LPS-MJO, LPS-nonMJO, CS-MJO, CS-nonMJO, and others. The percentage contribution of MJO only, TC-MJO, LPS-MJO, and CS-MJO to the total extreme rainfall events ranges from 9% to 16%, 0% to 3%, 2% to 4%, 1% to 9%, respectively. The relationship between MJO and flooding events in the Philippines was also examined. About 28 flood events or 266 flooding days were identified during the analysis period, wherein 50% of these events coincidentally occurred during strong and active phases of MJO.

INTRODUCTION

The Madden–Julian Oscillation (MJO; Madden & Julian, 1971) is a well-known intraseasonal oscillation (ISO) mode with a periodicity within 30–90-day time scales that can modulate local to regional climate features, including extreme rainfall events (Hsu et al., 2017; Pullen et al., 2015; Schreck III, 2021; Wei et al., 2020; Xavier et al., 2014). Although the influence of MJO may be present in all seasons, this ISO mode was found to be weaker during boreal summer and stronger during boreal winter season (e.g., Kikuchi, 2021).

Previous studies that examined the influence of MJO on extreme rainfall events found higher frequency of such events during active phases of the MJO (Da Silva & Matthews, 2021; Jones et al., 2004; Liang et al., 2022; Xavier et al., 2014). Jones et al. (2004) found that the global frequency of extreme rainfall events is 40% higher during active phases of MJO. The convective anomalies associated with the active MJO may alter the background conditions along its path and vicinity, favoring the occurrence of extreme rainfall events. Although the MJO's eastward propagation is mainly confined along the equatorial region, its impact is also felt in the extra-tropical regions through emanation of Rossby waves or interaction with the prevailing background circulation (Lim et al., 2017; Zhou et al., 2020).

Extreme rainfall events, which are associated with devastating floods, are one of the hazards that affect the Philippines. Various factors that contribute to these extreme rainfall events have been explored in recent studies and include, but not limited to, tropical cyclones (TCs), cold surges (CS), shearlines, MJO, BSISO, and non-TC vortices (Abdillah et al., 2021; Bagtasa, 2017; Cayanan et al., 2011; Chen et al., 2015a, 2015b; Cruz & Narisma, 2016; Olaguera, Matsumoto, et al., 2021; Olaguera, Caballar, et al.,  2021; Olaguera, Cruz, et al.,  2022; Olaguera, Manalo, et al.,  2022; Olaguera, et al., 2023; Olaguera & Matsumoto, 2020; Pullen et al., 2015; Racoma et al., 2022). Aside from the copious amounts of rainfall that each of the above factors may bring, these factors may also interact with the background circulation or with one another, which makes the extreme rainfall events in the Philippines more complex and severe (Olaguera, Cruz, et al., 2022).

Studies that examined the impact of MJO on the climate of the Philippines remain limited and mostly focusing on its impact on TC activity, interaction with a CS event, and diurnal cycle of rainfall (e.g., Bagtasa, 2020; Natoli & Maloney, 2019; Pullen et al., 2015), although the Philippines is usually included in studies that examined the impact of MJO over the broad Southeast Asian region (e.g., Lim et al., 2017; Xavier et al., 2014). Still, the impact of MJO on extreme rainfall events during the boreal winter (henceforth, referred to as northeast monsoon) season of the Philippines is not clear. In addition, how much of these extreme rainfall events are attributable to MJO alone or its interaction with synoptic disturbances needs to be quantified. Therefore, the objective of the present study is to build on these previous studies to quantify the influence of MJO on extreme rainfall events over the Philippines. The rest of the paper is organized as follows. Section 2 describes the different data sets and methodologies used in the study. The results and discussions are presented in Section 3. Summary and concluding remarks are provided in Section 4.

DATA AND METHODOLOGY Data

The following data sets from 1979 to 2019 were used in this study:

1. Daily rainfall records provided by the Philippine Atmospheric, Geophysical and Astronomical Services Administration (PAGASA) for the 11 eastern coast stations (Alabat, Casiguran, Catarman, Daet, Legazpi City, Tayabas, Guiuan, Surigao City, Hinatuan City, Infanta, Virac), shown in Figure 1. These stations have less than 20% missing data during the analysis period and were specifically chosen because they belong to the same climate type (i.e., Type 2; Kintanar, 1984) and where the impact of the northeast monsoon season is more pronounced;
2. Daily reanalysis data (JRA-55; Kobayashi et al., 2015) from the Japan Meteorological Agency (JMA) with 1.25° × 1.25° horizontal grid resolution;
3. Six hourly TC tracks from the Joint Typhoon Warning Center (JTWC);
4. Six hourly tracks of Low pressure systems (LPS) based on the JRA-55 reanalysis data set from Vishnu et al. (2020), which can be accessed at https://zenodo.org/record/3890646#.Y5naaHZBzrd. Vishnu et al. (2020) used an automated Lagrangian pointwise tracker called TempestExtremes (Ullrich & Zarzycki, 2017) to track these disturbances over the global tropics (35° S–35° N). The LPS data set that they provided are based on their optimized tracking algorithm using the 850 hPa stream function of the horizontal wind and 850 hPa relative humidity. Further details on the tracking algorithm are provided in Vishnu et al. (2020);
5. Historical flood events archived by the Dartmouth Flood Observatory (DFO), which can be accessed at http://floodobservatory.colorado.edu. This flood archive has data starting from 1985 to present;
6. Daily MJO index developed by Kikuchi (2020), which is available at https://iprc.soest.hawaii.edu/users/kazuyosh/Bimodal_ISO.html.

FIGURE 1. The topography of the Philippines (shades; m) and the location of the 11 PAGASA stations along the eastern coast. The topography data came from the National Aeronautics and Space Administration Shuttle Radar Topography Mission v3 (SRTMv3).

Methodology

We defined the northeast monsoon season as the months spanning from November of a particular year to March of the succeeding year (NDJFM), which covers the northeast monsoon season of the Philippines. For example, the northeast monsoon for the year 1980 is composed of November 1979 and January to March 1980. Easterly to northeasterly winds at low levels are dominant in this season; hence, higher rainfall amounts are found along the eastern coast of the country (Matsumoto et al., 2020). The analysis period spans from 1979 to 2019, which covers 40 seasons.

TC days were defined as the days when a TC is located within 1100 km from the PAGASA stations. This distance was statistically derived by previous studies such as Kubota and Wang (2009) and Bagtasa (2017) assuming that the TC rain rate depends only on the distance between the station and the TC center. Only the TCs with at least Tropical Storm (≥34 knots) categories were used to define TC days. Tropical Depressions (TDs) are included in the LPS days.

Previous studies that examined the impact of LPS on rainfall used a 500 km distance from the station location (e.g., Dang-Quang et al., 2016; Olaguera et al., 2023). To estimate the optimum distance by which an LPS can induce rainfall over the Philippines, we followed the same methodologies of Kubota and Wang (2009) and Bagtasa (2017) in determining the optimum distance for TC influence. First, we divided the daily rainfall into four quarters to match the 6-hourly LPS tracks. Then, the rainfall is binned into 100 km intervals (Kubota & Wang, 2009). Figure 2 shows the mean daily rainfall as a function of LPS distance and the stations. Similar to the relationship between TC distance and rainfall over the Philippines found by Kubota and Wang (2009), the rainfall rate decreases with LPS distance. When the distance between the stations and LPS center is beyond 1100 km, the rainfall rate becomes independent of LPS distance. Based on this result, LPS days were defined as the days when a LPS is located within 1100 km from the PAGASA stations.

FIGURE 2. Mean LPS-induced rainfall (mm day−1) across the 11 PAGASA stations vs the radius from the LPS center.

Composites of rainfall were also computed for the eight phases of MJO. Only the days during the strong phases (i.e., amplitude >1) were used in the composites. The rest of the days when the MJO is weak were considered as non-MJO days. The extreme rainfall days were defined as the days when their rainfall amounts exceeded the 95th percentile value of all rainy days during NDJFM.

The changes in the frequency of extremes were calculated as follows:[Image Omitted. See PDF]where the ΔPMJO is the percentage change in the cumulative probability of rainfall (X) exceeding the 95th percentile threshold (Xc) due to MJO. PMJO is X exceeding the 95th percentile threshold for a given MJO phase, while the Pnon-MJO is for the non-MJO days. A bootstrapping procedure was used to assess the significance of the percentage change in the frequency of occurrence of extremes (Chen & Zhai, 2017; Efron & Tibshirani, 1994; Matsueda & Takaya, 2015). The probability of occurrence of extreme rainfall days were calculated in each NDJFM season for both the MJO and non-MJO days, producing 40 pairs. These 40 pairs were randomly resampled, and the average probabilities were calculated for both the MJO and non-MJO days. This procedure was repeated 1000 times. The ΔPMJO was considered significant when the average probability of occurrence of extreme rainfall days during active MJO days was higher in 950 out of the 1000 samples. This significance test was applied for each phase of the MJO.

RESULTS AND DISCUSSION MJO impact on rainfall extremes

Power spectral analysis using fast Fourier Transform was applied on the rainfall time series of the 11 stations to determine whether the impact of MJO can be observed in these stations. The annual cycle and linear trends were removed first before applying a 10% tapering on the data. Unfortunately, this spectral analysis cannot be applied on time series with missing values; hence we removed the northeast monsoon seasons with at least 1-day missing data in generating the mean spectra in each station, as shown in Figure 3. The available years for each station are summarized in Table S1. The mean spectra show three dominant periods above the 95th percentile at MJO (30–60-days), quasi-biweekly (10–25-days), and synoptic (2–10-days) time scales. Significant peaks at the MJO time scale can be seen in Catarman (Figure 3c), Daet (Figure 3d), Tayabas (Figure 3f), Virac (Figure 3g), Legazpi City (Figure 3h), and Surigao City (Figure 3j) stations. No significant peaks at this time scale were found in the mean spectra of Alabat (Figure 3a), Casiguran (Figure 3b), Infanta (Figure 3e), Guiuan (Figure 3i), and Hinatuan City (Figure 3k) stations. We carefully checked the annual spectra of these stations and found that the MJO signal is actually present in some years (not shown) in these stations. This means that the mean spectra tend to hide the MJO signal in these stations. In the succeeding sections, we still used all these 11 stations in the analysis. Furthermore, the explained variance of MJO ranges from 9% to 12%, with the highest found in Surigao City station (12%; Figure 3j).

FIGURE 3. The mean power spectra (solid line) of NDJFM rainfall from 1979 to 2019 for the 11 PAGASA stations. The dashed line indicates the Markov red noise spectrum. The dotted and dot-dashed lines indicate the 5th and 95th percentile confidence levels, respectively, estimated using the lag −1 autocorrelation. Differences in the number of years used in generating the mean spectra are noted in the main text. The numbers inside the parentheses refer to the explained variance of the MJO mode (30–60-days) relative to the total seasonal variance expressed as percentage in each station.

To further elucidate how the MJO modulates the occurrence of extreme rainfall events during the northeast monsoon season of the Philippines, we examined the probability density functions (PDFs) of rainfall amounts in each station during the different phases of MJO (Figure 4). These PDFs are represented by box plots consisting of the 5th (minimum), 25th (lower quartile), 75th (upper quartile), and 95th (maximum) percentiles. The median and 75th percentile of rainfall amounts for the northern stations (i.e., Casiguran, Infanta, Tayabas, and Alabat) are skewed toward lower values across all phases, although the 95th percentiles are higher in some phases such as in Phase 5 in Infanta station and Phase 6 in Alabat station, relative to the non-MJO days. The PDFs have large interquartile ranges between Phases 4 and 6 in Infanta, Daet, Legazpi City, and Catarman stations. For the southern stations (i.e., Guiuan, Surigao City, Hinatuan City), the median, 75th, and 95th percentiles are skewed toward higher values between Phases 4 and 6, and toward lower values between Phases 1 and 3.

FIGURE 4. Probability distribution functions of average rainfall (mm day−1) in the 11 PAGASA stations for the eight MJO phases and non-MJO (nMJO) days.

The spatial distribution of the changes in the probability of extreme rainfall occurrence in the different phases of MJO is illustrated in Figure 5. The probability of extreme rainfall occurrence decreases by 20%–80% in most stations in Phases 1, 2, and 3, except in Virac station in Phase 1, wherein it increases by less than 20%. Surigao and Hinatuan City stations have the lowest percentage decrease (< 20%) compared to the other stations in Phase 3. In Phase 4, only Surigao and Hinatuan City stations show an increase in the probability of extreme rainfall occurrence up to 60%, while the rest of the stations show a decrease in the probability of extreme rainfall occurrence up to 60%. Between Phases 5 and 6, most of the central and southern stations show an increase in the probability of extreme rainfall occurrence, with Surigao City station having the largest change above 80% in Phase 5. From Phases 4 to 6, the MJO transits from the Maritime Continent to the western North Pacific (WNP), which could enhance the northeast monsoon and lead to more extreme rainfall events over the eastern coast of the Philippines (Bagtasa, 2020; Lim et al., 2017). Between Phases 7 and 8, a decrease in the probability of rainfall occurrence can be observed across all stations by 20%–60%.

FIGURE 5. Percentage changes in the probability of rainfall at the 95th percentile extremes for the eight phases of MJO. Upward (downward) triangles indicate increase (decrease). Shaded triangles indicate stations that exceed the 95% confidence level.

In general, the probability of extreme rainfall occurrence can be seen first in the southern stations (i.e., Surigao City and Hinatuan City stations) at Phase 4 before expanding northward to the central stations in Phases 5 and 6. Four stations (i.e., Daet, Alabat, Casiguran, and Infanta), particularly those located to the north, did not show increases in the probability of extreme rainfall occurrence. Interestingly, the magnitude of the probability of extreme rainfall occurrence has lower magnitude in the above mentioned four stations in Phases 4 to 6 compared to those in Phases 1 to 3, suggesting that some of the extreme rainfall events in these stations may still be modulated by MJO but, on average, are more dominated by other disturbances aside from MJO (Figure 4).

To further elucidate the cause of this spatial difference in the occurrence of extreme rainfall events between the north and southern stations, we obtained the composites of vertically integrated moisture flux convergence (VIMFC) and 850 hPa winds, as shown in Figure 6. The significance of the composites was obtained using t-test and by comparing the strong MJO days (amplitude >1) in each phase with weak MJO days (amplitude ≤1). During Phases 1 to 3, a large anticyclonic circulation is anchored over the eastern Philippines, which induces enhanced divergence over the Philippines. The presence of anomalous westerly to southwesterly winds in Phase 3 to the north of the Philippines indicates weakening of the northeasterly winds, which may consequently lead to less extreme rainfall events. There is also an anomalous northeasterly wind over the southern Philippines from Phase 2 to Phase 3, which may favor the occurrence of extreme rainfall events in this region. During Phases 4 to 6, enhanced VIMFC and northeasterly winds can be seen from the central to south-eastern Philippines, while suppressed VIMFC is apparent to the north. The region with enhanced VIMFC also appears to move from south to the central-eastern Philippines and in Phase 6, an anomalous cyclonic circulation appears over the central-eastern Philippines. Lim et al. (2017) noted that the MJO may provide additional moisture supply in the atmosphere over Southeast Asia in these phases, facilitating conditional instability, which favors the formation of synoptic disturbances, and increased rainfall. In the succeeding phases, the VIMFC weakens, which may lead to the decrease in extreme rainfall events over the eastern coast of the Philippines. In general, the suppressed VIMFC to the north in all phases of the MJO and enhanced VIMFC to the south during Phases 4 to 6 leads to the spatial difference in the probability of extreme rainfall occurrence between these two regions.

FIGURE 6. Composites of vertically integrated moisture flux convergence (VIMFC; shades; ×10−5 kg m−2 s−1) and 850 hPa winds (vectors; scale is 2 m s−1) for the extreme rainfall days in the eight phases of MJO (amplitude >1) relative to the weak MJO (amplitude ≤1). The numbers inside the parenthesis are the number of extreme rainfall days used for the composites in each phase. The + markings and bold vectors indicate statistical significance at the 95% confidence level by t-test.

Contribution of TC, LPS, CS, and other disturbances to extreme rainfall events

In this section, we identified the extreme rainfall events in each station and classified (in order) them into:

1. MJO-only (amplitude >1 and Phases 4–6);
2. TC-MJO (i.e., when a TC affects the Philippines during a strong [i.e., amplitude >1] and active MJO [i.e., between Phases 4 and 6]);
3. TC-nonMJO (due to TC only and without the influence of MJO);
4. LPS-MJO (when an LPS affects the Philippines during a strong and active MJO similar to TC-MJO);
5. LPS-nonMJO (due to LPS only and without the influence of MJO);
6. Cold-Surge-MJO (CS-MJO; when there is an enhanced northerly wind (i.e., days with averaged meridional winds exceeding one standard deviation of all NDJFM days from 1979 to 2019 along 15° N, between 122.5 and 130° E; Abdillah et al., 2021; Olaguera, Caballar, et al., 2021) and during strong and active MJO);
7. CS-nonMJO (due to cold surge only); and
8. Others (such as shearlines, which are formed at the trailing-end, along, or ahead of a cold front; Olaguera, Matsumoto, et al., 2021).

The extreme rainfall events were classified first as MJO-only, then the remaining days are classified into TC-MJO, TC-nonMJO, and so on. The percentage contributions of each classification were computed relative to the total number of extreme rainfall events in each station (Figure 7). Only the stations with an increase in the probability of extreme rainfall occurrence in Figure 4 are shown here (i.e., Alabat, Catarman, Virac, Legazpi City, Guiuan, Surigao City, and Hinatuan City stations).

FIGURE 7. Percentage contribution of TC-MJO, LPS-MJO, CS-MJO, MJO-only, TC-nonMJO, LPS-nonMJO, CS-nonMJO, and Others to the total number of extreme rainfall events in each station. The descriptions of these classifications are discussed in the main text. Only the stations with an increase in the probability of extreme rainfall occurrence in Figure 5 are shown here.

High percentage of extreme rainfall events are due to TC-nonMJO for Alabat (22%), Guiuan (12%), Legazpi City (15%), and Virac (26%) stations. Surigao and Hinatuan City stations both have 4% percentage contributions from TC-nonMJO. Such low percentage contributions from TCs alone over Mindanao Island were also noted by Cinco et al. (2016), Bagtasa (2017), and Olaguera et al. (2023). The contribution from TC-MJO ranges from 0% to 6%, with the largest in Virac station (6%). The contribution from LPS-nonMJO is also small and ranges from 4% to 13%. The contribution from the LPS-MJO is small and ranges from 2% to 4%, with the largest in Surigao City station. The contribution from CS-MJO ranges from 1% to 9%, with the largest contribution in the southernmost stations (i.e., Surigao and Hinatuan City stations). The contribution from CS-nonMJO ranges from 3% to 18%, also with high contributions in the southernmost stations. This result is consistent with Lim et al. (2017) who also found higher CS related extremes over this region during the boreal winter season. The contribution from MJO-only ranges from 9% to 16%, with the largest contribution found in Hinatuan City station (16%). As for the others, the contribution ranges from 38% to 49%, with the largest contribution found in Legazpi and Hinatuan City stations.

Relationship between MJO and flood events in the Philippines

In this section, we examined the flood events over the Philippines from the DFO archive during the analysis period and their relationship with MJO (Figure 8). There were 28 flood events or a total of 266 flooding days identified. About 7%, 10%, and 11% of the total flooding days are attributable to TC-MJO, LPS-MJO, and CS-MJO, respectively, while 9%, 6%, and 6% of the total flooding days are attributable to TC-nonMJO, LPS-nonMJO, and CS-nonMJO, respectively. The MJO alone contributes by about 22%, which is relatively larger compared to the contributions of TC-nonMJO, LPS-nonMJO, and CS-nonMJO. Still, the largest contribution to the total flooding days comes from other disturbances (30%). In general, about 50% of the flooding days are attributable to MJO or when it is coincident with synoptic disturbances.

FIGURE 8. Distribution of all flood events during NDJFM season recorded by the Dartmouth Flood Observatory from 1985 to 2019.

SUMMARY AND DISCUSSION

This study examined the relationship between the probability of extreme rainfall occurrence and MJO during the northeast monsoon season over the Philippines from 1979 to 2019. The results show that the probability of extreme rainfall occurrence increases first in the southern stations by no more than 20% in Phase 4 before expanding to the central-eastern stations between Phases 5 and 6. In these phases, the MJO is traversing the Maritime Continent and western North Pacific. When the convection associated with MJO is enhanced (suppressed) over the Maritime Continent and western North Pacific, an anomalous cyclonic (anticyclonic) circulation appears in the vicinity of the Philippines, which may be viewed as a Matsuno-Gill pattern response due to equatorial heating (Bagtasa, 2020). This anomalous cyclonic (anticyclonic) circulation enhances (suppresses) the moisture flux leading to higher (lower) rainfall amounts along the eastern coast of the Philippines.

The contribution of MJO-only, TC-MJO, LPS-MJO, CS-MJO, and other disturbances to the total extreme rainfall events ranges from 9% to 16%, 0% to 6%, 2% to 4%, 1% to 9%, 38% to 49%, respectively. It is apparent that the largest contribution comes from other disturbances. What systems are included in the “other disturbances”? To answer this question, we teased out further the extreme events caused by other disturbances and found that 656 days (16%) out of the 4011 extreme days due to other disturbances matched with the non-TC vortices in Olaguera et al. (2023). Hence, the category “other disturbances” used in this study also include disturbances that were not classified as LPS. Shearlines (e.g., Olaguera Matsumoto, et al., 2021), mesoscale convective systems (MCSs; Cheng et al., 2023; Feng et al., 2021; Lagare et al., 2023) are the other systems that can also induce heavy to extreme rainfall events over the Philippines, particularly in Mindanao Island during the northeast monsoon season. The relationship between MJO and flooding events in the Philippines was also quantified. In total, about 50% (133 flooding days) of the flooding events during the study period are attributable to MJO or when it is coincident with synoptic disturbances.

AUTHOR CONTRIBUTIONS

Lyndon Mark Payanay Olaguera: Conceptualization; data curation; formal analysis; investigation; methodology; validation; writing – original draft. John A. Manalo: Conceptualization; data curation; formal analysis; investigation; methodology; validation; visualization; writing – original draft. Alwin Bathan: Formal analysis; investigation; methodology; writing – original draft. Jun Matsumoto: Formal analysis; funding acquisition; investigation; methodology; supervision; writing – original draft.

ACKNOWLEDGEMENTS

Part of this study was supported by Grant-in-Aid for Scientific Research (No. 23H00030; PI Jun Matsumoto, 20H01386; PI Yoshiyuki Kakikawa of Kobe University), and 22H04938; PI Kei Yoshimura of the University of Tokyo. L.M.P. Olaguera was supported by the Regional Climate Systems of the Manila Observatory's project: High-definition Clean Energy, Climate, and Weather Forecasts for the Philippines.

CONFLICT OF INTEREST STATEMENT

The authors declare no conflicts of interest.

DATA AVAILABILITY STATEMENT

The data used in this study may be provided upon request. Please send an email to [email protected] for data requests.