Introduction

It is well-known that Madden-Julian Oscillation (MJO) is the most prominent mode of intraseasonal variability over the tropics with its largest amplitudes over the Indo-Pacific sector1,2,3. The east-west propagation of MJO in boreal winter and its impact over southeast Asia2, its interactions with higher-frequency variability such as the cold surges3, as well as the seasonal cycle, have been documented well4. For some parts of southeast Asia, boreal summer precipitation is dominated by the southwest monsoon3,4 and there are significant contributions from intraseasonal modes in boreal summer months (see Fig. 5 of Xavier et al.)4. Despite its importance, there are fewer studies5 in the literature on the role of northward/northeastward propagating boreal summer intraseasonal oscillations (BSISO6) and its impact on precipitation extremes over southeast Asia, compared to that of the eastward propagating MJO in the boreal winter season2,3,7.

Figure 1a shows the difference in the daily variance of the daily mean precipitation between boreal summer (May–September, MJJAS hereafter) and boreal winter (November–March, NDJFM). This shows a clear shift in the centres of active precipitation variability between seasons with changes in the seasonal monsoon circulations. In the boreal winter season, most of the precipitation variability occurs towards the eastern edges of the land masses (e.g., Peninsular Malaysia, Vietnam, Philippines) along with the seasonal changes in the intertropical convergence zones (ITCZ) over the equatorial regions and southern hemisphere. In boreal summer, the variances are of higher amplitude compared to that in boreal winter and dominate over the northern parts of the domain and along the western coasts (e.g., Myanmar, Gulf of Thailand, west of the Philippines and the northern South China Sea (SCS)). This also highlights that most of the precipitation variability is dependent on the direction and strength of the monsoon flow and its interactions with the land regions in both monsoon seasons.

Fig. 1: Spatial features of seasonal variance and temporal characteristics of BSISO.
figure 1

a Difference between boreal summer (MJJAS) and winter (NDJFM) variances of daily total precipitation (mm2 day−2). Positive (negative) values indicate higher variability in boreal summer (winter). Thirty days lead-lag composites of unfiltered precipitation (mm day−1) (b) and 20–90 days bandpass filtered precipitation as shaded and 850 hPa zonal winds as contours averaged between 100°–120° E longitudes (c) with reference to the peak dates in Phase 4 based on the BSISO index.

Full size image

Compared to the MJO, the intraseasonal variability over southeast Asia has not been studied extensively for its regional impacts. Even though the MJO Real-time Multivariate Monitoring (RMM) indices are produced for all seasons based on Empirical Orthogonal Functions (EOFs) of meridionally averaged fields of Outgoing Longwave Radiation (OLR) and 850 and 200 hPa zonal winds1, their value in determining and characterising northward or northeastward propagating boreal summer modes is somewhat limited6. The intraseasonal variability in boreal summer is largely dominated by the northward propagating BSISO over the Indo-Pacific sector8,9. Therefore, the MJO RMM indices based on zonal average winds and OLR to describe northward propagating BSISO modes may yield suppressed or misrepresented modes of variability. For real-time monitoring of the BSISO6 (L2013 hereafter) presented two sets of indices of BSISO based on the first 4 principal components derived from multivariate EOFs (MV-EOF) of OLR and 850–hPa zonal wind anomalies over an extended region 10° S–40° N, 40°–160° E, for May–October season. BSISO 1 and 2 pair represent the low-frequency northward/northeastward propagating modes with periods 30-90 days while the higher modes (BSISO 3 and 4 pair) characterise the westward propagating quasi-biweekly mode of the boreal summer6,10 that is more dominant during the pre-monsoon and onset phases. There are several other indices of summer intraseasonal oscillations in use8 for example, over the Indian sector, a regionally focused index of the Monsoon Intraseasonal Oscillations (MISO) is introduced and in use for real-time monitoring and forecasting11.

The BSISO modes6 are calculated over a large domain over the Indo-Pacific sector. These indices are used to describe the impact of BSISO over Indonesia5. They show that the increase in the probability of extreme summer precipitation is associated with enhanced large-scale moisture flux convergence and upward moisture transport induced by the active phases of BSISO5. This study uses large-scale BSISO 1 and 2 indices6 with periods between 30–60 days to study the regional impacts on extreme precipitation and its characteristics. Details of the data and detection methods of precipitation features are given in the section ‘Methods’. The characteristics of the northward propagation are described in the section ‘Characteristics of intraseasonal variability’ and their impact on precipitation extremes is discussed in the section ‘Impact on precipitation extremes’. Spatial features of precipitation extremes and their impact due to BSISO phases are given in ‘Precipitation properties’.

Results

Characteristics of intraseasonal variability

Figure 1b shows the lead-lag composites of unfiltered precipitation centred around the peak index days in phase 4, which corresponds to a peak in convective activity in the SCS (Fig. 2) averaged between 100°–120° E. This figure demonstrates the coherent northward propagation of precipitation and the slow evolution between a suppressed and active phase in daily precipitation fields without filtering. To add clarity to this, similar composites for 20–90 days bandpass filtered precipitation and 850 hPa winds are shown in Fig. 1c. The convective propagation initiates around 5° S and moves north at roughly 1° per day. It is clear that the phase of 850 hPa wind anomalies are at quadrature with the precipitation anomalies, indicating that the wind convergence and precipitation are strongly linked as noted in many studies over the Indian region9.

Fig. 2: Characteristics BSISO precipitation and 850 hPa wind anomalies.
figure 2

Composite anomalies of precipitation (mm day−1) and 850 hPa wind (m s−1) for the 4 BSISO phases as shown in (a–d) and the phases are labelled.

Full size image

The spatial details of precipitation in the different BSISO phases in the MJJAS season are shown in Fig. 2. The precipitation composites are similar to L2013, but instead of 8 phases of BSISO 1 and 2, only 4 phases are used here and with the high-resolution GPM-IMERG data, the features of regional convective activity are much more detailed. Here, the unfiltered precipitation for each phase is composited and the MJJAS climatological seasonal mean is removed to generate the anomalies. Phase 1 has an active convective phase over most of the northern parts of the domain, which clearly shows the north-south dipole in the sign of precipitation anomalies. The suppressed convection in Phase 3 over the northern parts of the domain is associated with anomalous anticyclonic circulation, which tends to oppose the prevailing mean southwesterly monsoon flow and is referred to as the monsoon break. In contrast, the anomalous circulation in Phases 1 and 4 tends to be cyclonic over most parts of the region, enhances the mean monsoon flow, and promotes the development of convection. It may be noted that these composites include Tropical Cyclone (TC) days and have contributions from the TCs as they are known to be modulated by the BSISO12. The northward/northeastward movement of convection is clearly shown in this figure. The most convectively active phase for Peninsular Malaysia, Sumatra, Java, and Borneo is Phase 3. The convective signals are most active south of 5° N, and the largest amplitudes are seen over the coastal and ocean grid points. Phases 4 and 1 move the convective regions further north, with the largest activity over SCS, the Gulf of Thailand, Vietnam, and the Philippines. There are also significant convective anomalies over the northern Bay of Bengal and the Myanmar coast in Phases 1 and 4.

Impact on precipitation extremes

Precipitation extremes in southeast Asia are known to be impacted by the MJO phases in boreal winter on regional and local scales2. Multi-week predictability of the MJO13,14 introduces a certain level of predictability on the probability of the occurrence of extremes in the region2. BSISO with similar large spatial scales and slow evolution suggests that they have similar predictability skills as that of the MJO15,16. This section examines how the phases of BSISO impact regional precipitation extremes. Figure 3 shows the 99th percentile values based on MJJAS daily precipitation data. The activity centres are similar to that of Fig. 1a with values above 40 mm day−1 over the Myanmar coast and the Gulf of Thailand. There are also high precipitation intensities in the coastal regions of the Northern Philippines as well as most oceanic regions in the north where Tropical cyclones are likely to be contributing to the extremes. The extreme precipitation values over Java and the southern parts of the domain are much weaker than those in the other regions in the MJJAS season.

Fig. 3: Climatological 99th percentile extreme values of precipitation (mm day−1) in the MJJAS season.
figure 3

The boxes shown are the regions used to construct the box plots in Fig. 5.

Full size image

The role of BSISO in modulating these precipitation extremes is shown in Fig. 4 using two measures as shown below:

$${P}_{intensity}^{phase}=100* \left({r}_{99}^{phase}-{r}_{99}^{clim}\right)/{r}_{99}^{clim}$$
(1)

where \({r}_{99}^{phase}\) is the 99th percentile value for each phase and \({r}_{99}^{clim}\) is the climatological 99th percentile value (Fig. 3). Figure 4a–d shows the percentage change in intensity of the 99th percentile value only for those days categorised as one of the 4 phases compared to the climatological value at each grid point as shown in Fig. 3. Patterns of change in the 99th percentile (Fig. 4a–d) values during the 4 BSISO phases are remarkably similar to the composite precipitation anomalies (Fig. 2) suggesting that there is a likely change in the probability distribution of precipitation during the different phases at different locations in the region. The patterns of change follow the phase propagation with about 50–60% increase in the intensity of precipitation extremes over Sumatra, Borneo, Peninsular Malaysia and Java Sea in Phase 3. A similar percentage reduction in extreme precipitation values is observed in the same regions in Phase 1. Over the oceanic regions in the northern parts, this increase/decrease is much larger (of the order of 70–90%).

Fig. 4: Impact of BSISO phases on the intensity and frequency of occurrance of extreme precipitation.
figure 4

Percentage change in the 99th percentile precipitation value (a–d; Eq. (1)) in the 4 BSISO phases in comparison with the climatological 99th percentile values shown in Fig. 3. Panels e–h shows the percentage change in the probability of occurrence of 99th percentile precipitation value in the 4 BSISO phases in comparison with the probability of occurrence of climatological 99th percentile values (\(P({r}^{clim}\ge {r}_{99}^{clim})=0.01\)). This quantity is shown in Eq. (2).

Full size image

The second measure on the percentage change in the probability of occurrence of precipitation exceeding the climatological 99th extreme precipitation value (Fig. 3) in each phase (\(P({r}^{phase}\ge {r}_{99}^{clim})\)), is written as:

$${p}_{occurr}^{phase}=100* \frac{\left[P({r}^{phase}\ge {r}_{99}^{clim})-P({r}^{clim}\ge {r}_{99}^{clim})\right]}{P({r}^{clim}\ge {r}_{99}^{clim})}$$
(2)

where \(P({r}^{clim}\ge {r}_{99}^{clim})=0.01\) by definition of the 99th percentile values. This measure is shown in Fig. 4e–h which shares broadly the same patterns as the changes in the 99th percentile value (Fig. 4a–d) with the same regions showing 50–60 increases in intensity, as well as a similar increase in the likelihood of occurrence of these extreme precipitation events. For example, in the Indochina region, these combined measures suggest that there is a 40–50% increase in the intensity of 99th percentile precipitation and a 60–80% increase in the probability of them occurring in Phase 4 of the BSISO.

The greatest increase and reduction in the values of extreme precipitation are observed in the ocean and coastal regions during the BSISO phases (Fig. 4). To better quantify the impact of BSISO phases over the major land masses in the region, box plots of precipitation (displaying 5, 25, 50, 75 and the 95th percentile values) over 5 regions marked in Fig. 3 are shown in Fig. 5 for the climatological precipitation as well as the 4 BSISO phases. Note that the percentiles displayed here are only for the land grid points in the boxes shown in Fig. 4. There are clear regional differences in the impact of BSISO in different phases for different domains. For Peninsular Malaysia and Borneo, the largest increase in mean and extreme precipitation is in Phase 3 and the largest suppression is in Phase 1. Similar but opposite impacts in these phases are observed for Indochina and the Philippines. Over Java, the mean precipitation in the MJJAS season is much smaller, and therefore the changes due to BSISO phases appear to be relatively weaker, but show a similar relative impact as over Peninsular Malaysia with an increase in extreme precipitation in Phase 3. How the precipitation extremes get modified under the large-scale dynamical background set by the BSISO phases requires knowledge of how the precipitation properties such as the precipitating area, intensity, etc. change in the different phases. The following section examines how the characteristics of precipitation change with the BSISO phases in detail.

Fig. 5: Box plots of precipitation (displaying 5, 25, 50, 75 and the 95th percentile values) over the land regions over 5 regions marked in Fig. 3 for the climatological values (black) and 4 BSISO phases.
figure 5

Note the non-linear scale on the y-axis.

Full size image

Precipitation properties

To characterise precipitation properties, an object-based feature detection method has been developed. Contiguous precipitation areas above the threshold values 5, 10, 15 and 20 mm day−1 are detected. Then a masking method for each of these thresholds was used to determine the precipitation statistics inside the contiguous area that covers the masked area exceeding the threshold value. Each of the contiguous precipitating areas is considered an object with its attributes such as date and time information, maximum, minimum, mean and standard deviation of precipitation values contained in the object polygon, their area, centroid, and the coordinates of the object area. Out of the 4.5 million total precipitation objects detected for the MJJAS seasons during the 2000–2020 period, each BSISO phase (with amplitude ≥1.0) contains approximately 18–20% of the objects.

Figure 6 shows the relationship between the object area (in units of the number of 0.1° grid boxes) and its mean (Fig. 6a) and the maximum precipitation values(Fig. 6b) inside the precipitation objects. There are multiple regimes of precipitation objects in Fig. 6a showing the most active regime which covers a few grid points and often with high precipitation values. These may be categorised as localised high-intensity precipitation events. There is also a regime that has mean precipitation between 10–20 mm day−1 that covers a range of spatial scales up to 10,000 grid points (~108 km2). There is also a low-probability regime in the spatial scales between 1000–20,000 grid points and with mean precipitation values up to 80 mm day−1. A similar relationship between the area of the object and the maximum precipitation values (Fig. 6b) shows the tail of the distribution on large scales more clearly. It shows the extreme precipitation values embedded inside these large-scale objects of the order of 5000–20,000 grid points, which is often associated with low-pressure systems and tropical storms. The domain used for object detection (Fig. 3) has ~192,000 grid points, and this figure shows that there is an apparent size limit of ~20,000 grid points for precipitation objects, suggesting that the largest systems in this domain may be tropical storms.

Fig. 6: Probability distribution of precipitation features.
figure 6

The joint probability distributions between the area (in 0.1-degree grid units in black and approximate area in km2 in red marked as x-axis labels) and mean (a) and area and maximum (b) precipitation (mm day−1) within detected precipitation objects are shown. The object area histograms are shown in the top panels of both (a)and (b), while the mean and maximum precipitation histograms are on the right in both (a) and (b), respectively.

Full size image

The way these precipitation regimes are impacted by four BSISO phases is examined in Fig. 7, which shows the difference in the joint distribution of area and mean precipitation for each phase from the climatological distribution (Fig. 6a). This figure only considers objects with centroids to the north of 5° N so that the opposing BSISO characteristics to the north and south of 5° N (Fig. 2) do not nullify the impact of its phases on the precipitation features in this figure. Phase 1 shows a clear increase in the probability of the regime with all spatial scales with mean precipitation up to 20 mm day−1 but the enhancement is limited to 10 mm day−1 at smaller scales (defined here as object area ≤100 grid points) and increases to scales up to 20 mm day−1 as the spatial scales become larger. This is associated with a comparable reduction in the probability of high-intensity smaller-scale (defined as object area ≤100 grid points) events in Phase 1. Phase 3 shows a similar but opposite behaviour with an enhancement of the probability of small-scale high-intensity objects and a reduction in the larger-scale medium-intensity objects. Phases 2 and 4 being the transition phases, also show the transition of the regimes between Phases 1 and 3, but are not described in detail here.

Fig. 7: Joint probability distribution between the area (in units of 0.1-degree grid) and mean precipitation value (mm day−1) inside all detected precipitation objects.
figure 7

The difference in the distribution for 4 BSISO phases from the climatological distributions (Fig. 6a) is shown in (a–d) and the phases are labelled on each panel.

Full size image

To show the link between extreme precipitation and its features, the locations and areas of only those precipitation objects that have a maximum value ≥40 mm day−1 are plotted in Fig. 8 for the 4 BSISO phases. The 40 mm day−1 threshold has been chosen which equates to some of the tropical cyclone-related 99th extreme value in the domain (Fig. 3). This threshold gives ~9700–13000 objects in each phase (Fig. 8). The small and large objects (defined as above, small = area ≤ 100, large = area > 100 grid points) are shown in different colours here to highlight their density and areas of occurrence. Generally, large precipitation objects are more likely to be observed over ocean points irrespective of the phases and there is an accumulation of small-scale convection over land and coastal regions. The impact of BSISO phases is to modify these patterns.

Fig. 8: Spatial distribution of small and large precipitation objects.
figure 8

Locations (centroids) and equivalent sizes of precipitation objects exceeding a max value of 40 mm day−1 for Phases 1–4 (a–d). Sizes of circles are proportional to the equivalent area of the objects. Blue circles indicate objects with an area less than 100 grid points and open orange circles indicate objects larger than 100 grid points. The radius of circles is proportional to the area each object covers. All blue circles are plotted with a constant radius to highlight the smaller objects. Panels e, f show the percentage distribution of small (defined as ≤100 grids) and larger (>100) convective systems over the land, ocean and coastal region during each of the 4 phases. Please note that these are for all objects irrespective of their precipitation intensities. The definition of land, coastal and ocean points are as: Land = proportion of water surface < 50%, Ocean = proportion of water surface = 100%, Coastal = proportion of water surface between 50 and 100%. With this definition, the domain contains 23.01% land, 68.66% ocean and 8.32% coastal points.

Full size image

As mentioned above (Fig. 7) there is a significant increase in the number of large precipitation objects in Phase 1 compared to Phase 3 over the northeast part of the domain. These are associated with an enhanced likelihood of tropical storms in the region during Phase 1 compared to Phase 3 as shown in Fig. 9. These are mainly associated with large objects in storm track areas around the Philippines, Vietnam, and other countries in the north. The fraction of large precipitation objects that are related to tropical storms is calculated by assuming that all large objects whose centroids fall within a 2.5° × 2.5° latitude-longitude region around a storm track are related to the storm. This fraction is computed as the ratio between large objects along the storm tracks with the total number of large objects in the whole season and each BSISO phase (Table 1). It may be noted that the total number consists of all large objects in the whole domain while the storms are mainly located in the northern parts of the domain. While considering only objects to the north of 5° N, these numbers are considerably higher (Table 1).

Fig. 9: Changes in the tracks of tropical storms.
figure 9

Tracks of tropical storms from IBTrACS29 for which any of the storm days occurs in a Phase 1 and b Phase 3. The colours indicate different storm categories: DS—Disturbance, TS—Tropical, ET—Extratropical, SS—Subtropical, NR—Not reported, MX—Mixture (contradicting nature reports from different agencies).

Full size image
Table 1 Percentage of large objects related to storm tracks to the total number of large objects in the whole season and for each BSISO phase
Full size table

Even in the presence of large-scale regimes, there are a reasonable number of small objects in the domain, particularly over the east coast of peninsular Malaysia, east of Sumatra, and in the storm track regions to the east of the Philippines. In contrast, Phase 3 has a substantial reduction in storm numbers (Fig. 9b), and therefore a proportional reduction in the number of large precipitation objects (Fig. 8c). The largest difference in the number of storm tracks is observed in the northern parts of the South China Sea between the Philippines and Vietnam (Fig. 9). On the other hand, the number of small objects shows a substantial increase in Phase 3 compared to Phase 1. This is reflected in the total number of features that show a 34% increase in the number of objects in Phase 3 (13161 objects compared to 9772 in Phase 1), a significant fraction of which is in the small-scale regime.

It may be noted that the small objects are rather randomly distributed in Phase 1 but in Phase 3, the majority of these small-scale objects are located in the land and coastal regions. Convection is more active in the southwestern part of the domain, including the central Indian Ocean to the eastern Indian Ocean6 in Phase 3 that promotes more large extreme precipitation objects (Fig. 8c). Phase 3 is the suppressed convective phase for the northern half of the domain (Fig. 2), the absence of any significant large-scale forcing or the related strong vertical wind shear can lead to the atmosphere becoming locally unstable due to a combination of an abundance of moisture, enhanced solar heating, and local variations in surface conditions. This instability can lead to the development of isolated convection which may lead to the conditions leading up to the subsequent phase of the BSISO. It may also be noted that a significantly large number of small objects are present in the coastal regions between Peninsular Malaysia and Sumatra in all phases. This is amplified in Phases 2 and 3 compared to other phases, which may be due to the increased presence of Sumatra squall lines17 in these phases.

To further quantify the impact of BSISO phases on the ocean, land, and coastal convection, a land/sea static mask relevant to IMERG precipitation 0.1 × 0.1 degrees V2 (GPM IMERG Land Sea Mask) at GES DISC18 is used to classify each grid point into the land, coastal or ocean grid points based on the following conditions19: Land—proportion of water surface < 50%; Ocean—proportion of water surface = 100%; Coastal—proportion of water surface 50 ≥ p < 100%. With this definition, the domain used in this study (90°–145° E, 10°S–25°N) contains 23.01% land, 68.66% ocean and 8.32% coastal points.

The percentage distribution of small and large precipitation objects (definitions as above) over the land, ocean and coastal regions during each of the 4 phases is shown in Fig. 8e, f. It is based on the location of the centroid of these objects to be in the ocean/land/coastal points based on the land-sea mask conditions described above. Please note that only objects with a maximum precipitation value greater than 40 mm day−1 as plotted in Fig. 8a–d are considered in this computation so that only the features related to severe precipitation extreme (99th percentile) are discussed. Consistent with the predominance of ocean grid points in the domain, ocean-based precipitation objects account for approximately 50–60% of all extreme precipitation events, while land-based objects contribute around 10–20% and coastal objects approximately 5–15%.

Figure 8e, f shows the percentage of the land/ocean/coastal convection for small and large extreme precipitation objects. For all the small objects (Fig. 8e) there is a consistent increase in the fraction of ocean/land and coastal convection from Phase 1 to Phase 3. This increase is 3% over the ocean, ~4% over land and about 2% over coastal points. This is also associated with about 8% reduction in large precipitation objects over the ocean (Fig. 8f), primarily over the storm tracks in the northern parts of the domain) with a 1–4% increase in the number of land and coastal precipitation objects from Phase 1 to 3. This shows the clear modulation of BSISO phases on the scales, intensities, and probabilities of occurrence of precipitation objects whether they are located over land, ocean, or coastal points. The notable increase in small coastal and land precipitation objects may be attributed to an enhancement of the diurnal cycle of convection ahead or during the convective phases of BSISO, as shown in the case of MJO20.

To establish the connection between the characteristics of the precipitation objects and extreme precipitation over the land regions for the BSISO phases (Fig. 5) more clearly, an example is shown in Fig. 10 showing the locations, size, and intensity of the precipitation objects for Phases 1 and 3 over Peninsular Malaysia, Borneo, Sumatra and Java (Fig. 10a, b) and the Philippines (Fig. 10c, d). Only the precipitation objects that have the maximum precipitation value greater than 40 mm day−1 as in the previous figures (Fig. 8) are shown. There are only a few of these extreme precipitation objects over land and coastal areas in Phase 1 (Fig. 10a) most of which are small localised objects clustered in the eastern regions of Kelentan and Trenganu, eastern parts of Sumatra and northern regions of Borneo. As noted in Fig. 5 both Peninsular Malaysia and Borneo have suppressed mean and extreme precipitation in this phase. Compared to other land regions in Fig. 10a, Borneo precipitation is contributed by larger (~1000–5000 grid points) objects with more intense precipitation values (Fig. 10a). This picture changes dramatically in Phase 3 which has more than double the number of extreme precipitation objects compared to phase 1 and contains more small objects over the land and coastal regions (Fig. 10b). Over the northern parts of Borneo, most of the extreme precipitation is contributed by more frequent, larger, and more intense precipitation objects in Phase 3. (Fig. 10b). Over the Philippines, on the contrary, Phase 1 has many more large-scale, more intense and more frequent extreme precipitation objects often associated with the tropical storms (Fig. 9a). In Phase 3 (Fig. 10b), the prevalence of large-scale oceanic objects diminishes significantly, giving way to intense precipitation events that predominantly affect land and coastal areas at both small and large scales.

Fig. 10: Location and intensity of extreme precipitation objects.
figure 10

Figure shows the location and intensity of precipitation objects which contain precipitation value greater than 40 mm day−1 for Peninsular Malaysia, Sumatra, Java and Borneo in Phase 1 (a) and Phase 3 (b). Similar plots for the Philippines are shown in (c, d). The size of the circles indicates the equivalent area of the precipitation object and the colour shows the maximum precipitation value in the object. Please note that the sizes of circles here are scaled down to 10% of their actual size.

Full size image

Discussion

Despite its importance in contributing to regional monsoon precipitation, the BSISO has not been studied extensively for its characteristic northward propagation and its impact on precipitation extremes over southeast Asia. This study addresses the role of BSISO over southeast Asia in modulating the extreme precipitation and associated precipitation features. The regional propagation features of the 4 BSISO phases6 show coherent northward/northeastward propagation of precipitation and a gradual transition between suppressed and active phases and a strong linkage between wind convergence and precipitation, consistent with previous studies of BSISO over other tropical regions.

The northward propagation of large-scale convection is categorised well using the 4 BSISO phases. The mean precipitation over Peninsular Malaysia, Borneo, Sumatra, and Java is suppressed in Phases 1 and 4 with a north-south dipole in precipitation anomalies where they are enhanced over Indochina and the Philippines. The opposite is true in the case of Phase 2/3. In addition to the changes in the mean precipitation values, there is a clear and discernible shift in the precipitation probability distributions in different BSISO phases. There are substantial increases in the frequency of occurrence of extreme precipitation as well as their intensity during the convective phases of the BSISO during its northeastward passage over the southeast Asia region. Changes in the 99th percentile values suggest that large-scale flow fields and moisture associated with the BSISO may be driving these changes, as in the case of MJO modifying regional extremes2.

To investigate how the precipitation characteristics change with BSISO phases, an object-based method to detect and characterise precipitation features has been presented. All precipitation features exceeding thresholds of 5, 10, 15 and 20 mm day−1 are detected and statistics and the features of individual precipitation objects are computed. This method gives a fresh look into the precipitation regimes that dominate the BSISO phases and hence contribute to the changes in precipitation distribution. Most of the precipitation in Phase 1 (active convection over SCS, northern Philippines and Indochina) is attributed to an increased large-scale convective regime associated with the increased frequency of tropical storms that are modulated by the BSISO phases. This is also linked to a reduction of localised small-scale high-intensity precipitation events. In contrast, Phase 3 (active convection over Peninsular Malaysia, Borneo, and Sumatra) shows an increase in smaller and medium-sized precipitation areas that produce high-intensity precipitation. A clear way of classifying the precipitation regimes therefore gives a detailed insight into how the large-scale envelope of the BSISO modifies the regional precipitation extremes through the precipitation properties.

The impact of BSISO phases on ocean, land, and coastal convection is investigated through the classification of precipitation objects based on their location on land, ocean, or coastal points. The results show that precipitation objects over the ocean dominate extreme precipitation (value ≥ 40 mm day−1), contributing 50–60% compared to 10–20% over land and 5–15% over coastal regions. Statistics for each BSISO phase reveal a consistent increase in the small extreme precipitation objects from Phase 1 to Phase 3 and a small increase in large precipitation over land and coastal regions. This is offset by a significant reduction in large ocean precipitation objects primarily in storm tracks. This modulation suggests that BSISO phases influence precipitation scales, intensities, and probabilities over different surface types, highlighting a potential link between BSISO phases and the enhancement of diurnal convection, similar to that of the MJO.

Further research will explore the predictability of BSISO phases. A similar approach as devised by L20136 can be employed to project the forecasted OLR values onto the EOFs 1 and 2 after removing the climatological and seasonal signals. Since their impact on regional extremes is unambiguous, the predictability of BSISO beyond the weather forecasting time scales will benefit the probabilistic forecasting of regional extremes in precipitation. Detailed analysis of the role of BSISO on regional processes—for example, their impact on tropical cyclone characteristics, and other mesoscale convective features such as the squall lines and rapidly develo** cumulus areas21)—is of importance and will be addressed in future studies. Potential modulation of these systems by the BSISO would suggest improved predictability of smaller-scale systems in boreal summer.

Methods

BSISO indices

As mentioned above, the MJO RMMs1 are not designed to meaningfully characterise the northward movement of convection as the EOFs are based on OLR and 850 and 200 hPa zonal winds zonally averaged between 15° S-15° N. The BSISO indices defined by L20136 are a pair of multivariate empirical orthogonal functions (MV-EOF) modes based on daily mean outgoing longwave radiation (OLR) and 850-hPa zonal wind (U850) anomalies over the ASM region (10° S-40° N, 40°-160° E). The BSISO 1 and 2, represent the northward and northeastward propagating ISO over the Asian summer monsoon region (ASM) during the boreal summer season (May to October) with quasi-oscillating periods of 30–60 days with similar periods as the eastward propagating MJO1,22. In this study, higher-frequency (BSISO 3 & 4) modes with eastward propagation with periods between 10–20 days are not considered. The BSISO 1 and 2-time series are used to define 8 phases of propagation1. However, to characterise the transition of phases over a relatively smaller Southeast Asia domain with reasonable overlaps when considering 8 phases, they are merged into 4 phases (Phases 1 & 8 as Phase 1, 2 & 3 as Phase2, 4 & 5 as Phase 3 and 6 & 7 as Phase 4).

Precipitation feature detection

To determine how large-scale modes of BSISO impact precipitation characteristics, a detection method is used (https://github.com/princekx/Features_monitor). It uses a threshold-based detection of contiguous grid points that exceed precipitation values greater than 5, 10, 15 and 20 mm day−1. Objects are detected using a simple grid masking for each threshold, and identified objects are extracted using the Python Scikit image library23. Each feature identified in the domain is marked as a unique object and labelled. Unlike existing cloud tracking algorithms24,25, the current method tries to identify all grid size precipitation features including single grid point features and their statistics, as they are in the original data, and no smoothing or segmentation methods are invoked. Tracking of objects in the time dimension is also not applied.

Each precipitation object contains attributes such as the mean, standard deviation, maximum and minimum precipitation within the object area; their geometric coordinates, area, perimeter, orientation, eccentricity, centroid, effective radius, etc. Then a Python Pandas data frame26 is constructed using these features of the object, which can then be used for statistical computations of the features of the object. The object areas are output as the number of grid points the object covers and no conversion to physical length units is performed.

Datasets

Daily total precipitation from NASA global precipitation measurement (GPM) integrated multi-satellite retrievals for GPM (IMERG)27 at 0.1° × 0.1° spatial resolution has been used for describing the features and statistics of the extremes. 850 hPa winds at 0.25° × 0.25° are obtained from the ERA5 reanalysis data28. The feature detection in GPM IMERG is done at 0.1° × 0.1° resolution and at 5 mm day−1 threshold, it generates 4.5 million precipitation objects in the MJJAS season for years 2000–2020 for the domain 10° S–25° N, 90°–145° E.

Land/Sea static mask relevant to IMERG precipitation 0.1 × 0.1 degrees V2 (GPM IMERG Land Sea Mask) at GES DISC18 to classify each grid point into the land, coastal or ocean grid points.

Tropical storm data for the West Pacific sector from The International Best Track Archive for Climate Stewardship (IBTrACS)29 have been used to demonstrate the impact of different BSISO phases on storm tracks and the number of storms that impact southeast Asia.