2022 Volume 100 Issue 3 Pages 475-492
Sub-daily extreme precipitation in Jakarta exhibits trends that are related to local temperature, seasonal tropical monsoon circulations, and other environmental drivers. An analysis of 81 years of hourly rainfall between 1900 and 2010 shows a significant increase of about doubling the number of short-duration rainfall events in the wet season. In recent decades, rainfall is found to be higher in intensity and shorter in duration relative to preceding decades. These short-duration rain showers develop typically between afternoon and late night or during early morning hours. Changing short-duration rainfall characteristics throughout the last century are partly attributed to changes in the surface environment of urban Jakarta. A recent temperature increase and land surface drying in the city, in combination with a small increase in the atmospheric moisture content, promote intensified atmospheric convection. A combination of rain gauge data with upper-air observations collected during 2002–2016 reveals that surface warming in the urbanized city accompanied by enhanced availability of moisture results in an increase in convective available potential energy, which contributes to enhanced intense precipitation. Super Clausius–Clapeyron scaling (CC) of high-intensity rainfall is attributed to high near-surface temperature and atmospheric moisture content in the morning. This super-CC scaling is present in a relatively small range of surface temperature values. Results of this study are in agreement with earlier findings exploring the intensification of extreme morning precipitation and a temporal shift of the diurnal convective maximum from late afternoon to late night/early morning in response to local warming. For a delta city such as Jakarta with abundant convection and heavy precipitation, a well-maintained rainfall database is crucial to assist urban flood early warning.
In most climate regions of the planet, extreme precipitation is expected to increase non-uniformly with anthropogenic global warming (IPCC 2021). An increase in the atmospheric temperature by 1°C increases the water holding capacity of an air parcel by approximately 7 % (Trenberth 2011), according to the so-called Clausius–Clapeyron (CC) relationship, expressing the relation between the moisture-holding capacity of the atmosphere, temperature, and pressure (Lenderink and van Meijgaard 2010; O'Gorman 2015). Model projections show that global mean precipitation increases with temperature at a rate lower than the atmospheric moisture increase (Held and Soden 2006; IPCC 2013) and is estimated to be approximately 1–3 % per degree of global warming (Wentz et al. 2007; Lambert et al. 2008). Uncertainties of increases in frequency and intensity of extreme precipitation are considerable and regionally varying (Meehl et al. 2007; Jones et al. 2010).
Using hourly observations of temperature and precipitation, Lenderink and van Meijgaard (2008) showed that in the Netherlands relatively mild precipitation extremes (the 90th percentiles) increase with temperature at a rate close to CC (1CC, approximately ∼7 % °C−1). For stronger extremes (99th percentiles) a super-CC relation with temperature exists (almost 2CC, or approximately ∼14 % °C−1). Similar results were found for Belgium and Switzerland (Lenderink and van Meijgaard 2010). A comparison between the Netherlands and Hongkong showed that the scaling closely follows a 2CC for higher percentiles in both climate zones, but in Hongkong, this scaling is confined to dew point temperatures below 23°C (Lenderink et al. 2011). Similar observations were explored for other regions, e.g., in Germany (Haerter and Berg 2009; Berg et al. 2013), Australia (Jones et al. 2010), Japan (Utsumi et al. 2011), and America (Shaw et al. 2011). The intensity-dependent CC scaling is generally less explicit for daily precipitation (Lenderink and van Meijgaard 2008), and even negative scaling relationships between precipitation extremes and temperature were found at most stations over mainland China (Miao et al. 2016).
Although in the extra-tropics intensified precipitation extremes are observed in response to a temperature increase following thermodynamic concepts, the sensitivity of precipitation extremes to warming in the tropical regions remains uncertain. This is probably related to specific feedbacks associated with the strong convection (O'Gorman 2015), resulting from contributions of dynamical or microphysical processes (Muller and Takayabu 2020). The contribution of dynamical processes mainly consists of changes in atmospheric (vertical) motions. Additionally, feedback on microphysical processes leads to altered precipitation efficiency (Muller and Takayabu 2020). These feedbacks are suggested to be at least partly caused by responses to local warming (Lau et al. 2017; Lutsko and Cronin 2018; Muller and Takayabu 2020).
Over Indonesia, increasing trends of both frequency and intensity of extreme daily precipitation have been found in the last decades, together with spatially coherent increasing trends of various temperature indices (Supari et al. 2017). For Jakarta, a century-long record of high-quality data was used (Siswanto et al. 2016). Jakarta's temperature has increased at a rate exceeding the global mean temperature over the past century, in response to both global warming and rapid urbanization (Siswanto et al. 2016). The heaviest 1 % of all precipitation events also exhibits an increasing trend, particularly during the wet season. These events have become roughly 2.4 times more likely over the last 115 years (Siswanto et al. 2015). By contrast, the annual mean precipitation does not display a strong trend.
Torrential rain events of > 200 mm day−1 are rare in the region and are usually associated with major flood events. As an example of a corresponding extreme precipitation event, extensive flooding struck the city on the 2020 new year's eve after a record-breaking daily extreme precipitation of 377 mm recorded at Halim PK Airport. Similar rainfall events led to the February 2015 and January 2014 Jakarta flooding. Frequently the Indonesian Agency for Meteorology, Climatology and Geophysics (BMKG) is asked to provide guidance on changes in precipitation extremes, their relation to urban flooding events, the impact of climate change, and urbanization on these trends.
These extreme events are usually generated from rapidly developing large-scale convective systems induced by particular regional atmospheric circulation features. A strong background Asian monsoon system, orographic lifting, proximity of tropical cyclones (or tropical low-pressure system), equatorial atmospheric waves, or patterns with high regional sea surface temperatures may enhance the formation of extreme convection over Jakarta. These confounding factors are responsible for a strong seasonal to inter-annual variability of the occurrence frequency and intensity of Jakarta's extreme precipitation events. At shorter multi-day to weekly time scales, the Asian monsoon activity is governed by the emergence of cold surges (CS) or cross-equatorial northerly surges (CENS), and the presence of the Intertropical Convergence Zone (ITCZ), which may coincide with CS or CENS events. The Madden–Julian Oscillation (MJO) is the main source of variability at the sub-seasonal timescale, whereas on the inter-annual time-scale, the well-known El Niño–Southern Oscillation (ENSO) and Indian Ocean Dipole (IOD) are dominant (see, e.g., Chang et al. 2005; Hattori et al. 2011; Wu et al. 2015; Koseki et al. 2013; Siswanto et al. 2015; Mori et al. 2018; Lestari et al. 2019). Under favorable atmospheric conditions, convective clouds develop at high altitudes, often as a spin-off from a mesoscale convective system (Nuryanto et al. 2019). This occasionally generates large torrential downpours with repeated showers that can last up to a few hours.
Observation-based relationships between sub-daily precipitation and temperature may contain useful guidance on the sensitivity of tropical precipitation extremes to local warming. Sub-daily precipitation extremes show a stronger response to changes in local atmospheric temperature changes than daily-mean precipitation (Westra et al. 2014; Barbero et al. 2019). These temperature changes may arise from both urbanization and global warming, and thus, both trends can affect the local precipitation characteristics (McCarthy et al. 2010) in the tropical and energetic monsoon region of Jakarta.
This paper analyzes the precipitation–temperature scaling for the Jakarta area. We assess trends in sub-daily duration, intensity, frequency, and timing of rainfall and relate it to surface conditions (near-surface temperature, urbanization, atmospheric moisture, and convection characteristics) and large-scale climatic oscillations. We use more than 100 years of daily and more than 80 years of hourly observational records from Jakarta Observatory. Additionally, 15 years of upper-air sounding observations collected between 2002 and 2016 at Soekarno–Hatta International Airport (SHIA) are included in the analysis. This long observational record contributes to the statistical robustness of the assessment of the drivers of extreme precipitation and long-term trends therein.
The main dataset used in this study comprises hourly surface measurements at Batavia (Jakarta) Observatory between the period of 1900–1980 and 2000–2010. This study does not analyze the period 1981–1999 because of missing digitized data, although observations were continued for this period (e.g., daily/pentad rainfall amount is available). The dataset is built from old Jakarta Observatory yearbook archives stored at the Royal Netherlands Meteorological Institute (KNMI) and BMKG. Digitization activities occurred under the joint BMKG/KNMI project Digitisasi Data Historis1 and successfully rescued 130-year daily (Siswanto et al. 2016) and 100-year hourly precipitation and temperature time series between 1866 and 1980 (929182 hourly records in total with 10 % missing values; see Fig. S1 on Supplementary Material).
Hourly dew point temperature records are only available for the period 1866–1966 (806414 hourly records, < 30 % missing values including no data between 1967 and 1980). Missing values are mostly related to the years with no observational activity (15 years) due to political issues. Daily records of dew point temperature are available for the entire time period. A homogeneity check of the hourly dew point time series has been applied using the RHtestsV4 package2. No significant change points were found in the mean or variance of the yearly and monthly time series, besides the episodes with no-observation years. Detailed information regarding the data history, quality control, missing values, homogeneity procedures, and trends of temperature and precipitation datasets can be found in (Siswanto et al. 2016). The continuity check of gradual changes in temperature during the period 1991–1999 with neighboring station data has been verified as documented in their Supplementary Material. Readers are also referred to Brandsma and Jilderda (2011) and Können et al. (1998) for a historical overview of the observation site and its datasets.
For the period 2002–2016, atmospheric soundings are collected twice per day at 00 UTC (07 LT) and 12 UTC (19 LT) at the SHIA at Cengkareng (20 km distance from Jakarta). These were analyzed for convective characteristics before three hourly accumulated rainfall events recorded at the SHIA, Pondok Betung, Tanjung Priok, and Kemayoran between 2014 and 2016. Standard meteorological temperature and humidity observations taken at in situ observatories in and around the greater city area of Jakarta are collected for 1985–2010.
Temperature and humidity are aggregated for both urban and rural environments. Urban data represent the average of measurements taken at BMKG's meteorological stations within the Jakarta city, namely, Kemayoran/BMKG headquarter (6.15°S, 106.84°E, WMO ID: 96745), Tanjung Priok harbor (6.11°S, 106.88°E, WMO ID: 96741), and Halim PK Airport (6.27°S, 100.89°E, WMO ID: 96747, ICAO ID: WIHH) indicated as locations 1, 2, 3, on the map in Fig. 1. Rural data are represented by the average of measurements taken at BMKG's meteorological stations at Darmaga Bogor (6.50°S, 106.75°E, WMO ID: 96753) and Citeko (6.70°S, 106.85°E, WMO ID: 96751) indicated as locations 7 and 8 on the map in Fig. 1. Locations 5 and 6 were not included in the analysis because they were not clearly classified as urban or rural.
The Greater Jakarta map in the setting of the Java Island topographical map. The Greater Jakarta, or the Jakarta metropolitan area (in the inset), known locally as Jabodetabekpunjur (an acronym of Jakarta–Bogor–Depok–Tangerang–Bekasi–Puncak–Cianjur), is the most populous metropolitan area in Indonesia. The area is crossed by a network of 13 rivers, flowing from mountainous and hilly areas in the Bogor and Cianjur regions to Jakarta through the alluvial and coastal lowlands of North Jakarta. BMKG operates eight meteorological stations that are classified as urban (red dot) or rural (green dot) in this study.
Hourly precipitation data was mainly digitized from pluviograph paper strip charts of the 19th and 20th centuries, with a rainfall interval of 0.1 mm. In the analysis, we defined an hourly rainfall event with a threshold of 0.2 mm h−1 (Manton et al. 2001). Hourly rainfall under this threshold is used as a separator of different events. Hourly precipitation data are stratified into cumulative multi-hour amounts in different ways. For most analyses, data are classified into rainfall duration categories for which frequency and intensity distributions are calculated. A rainfall duration class is determined by the number of consecutive hours with precipitation intensity exceeding a given threshold. For the precipitation–temperature scaling, hourly precipitation data are binned into air and dew point temperature intervals. For each bin, the 90th, 99th, and 99.9th precipitation percentiles are computed to determine the hourly extreme precipitation.
Long-term trends of precipitation characteristics are assessed by comparing three 10-year climate periods, referring to the early 20th century (1901–1910), mid-century (1971–1980), and recent (2001–2010) episodes. We compare the seasonally mean diurnal cycles of rainfall and temperature for these climate periods. Long-term changes in contrasting climate characteristics between the city of Jakarta and surrounding non-urban areas are investigated using the geographic setting of rural station data depicted in Fig. 1. Spatially aggregated temperature time series are constructed after correcting for station altitude differences with respect to the altitude of Jakarta Observatory Station (8 m above sea level) using a standard lapse rate of −0.65°C 100 m−1. The hourly relative humidity is computed from observed hourly temperature and dew point temperature. The time-averaged relative humidity is calculated from hourly relative humidity data. All analyses are performed with R3.
First, we present the intensity–frequency–duration relation of Jakarta sub-daily rainfall events. Figure 2 shows cumulative rainfall (intensity) and the number of events (frequency) for different duration classes (hourly bins). The analysis is performed separately for all rainfall data and extreme rainfall events exceeding the 95th percentile. The 95th-percentile thresholds are calculated for every rainfall duration class. Results are shown for the wet (NDJFM) and dry (MJJAS) monsoon seasons in the three distinct climate episodes 1901–1910, 1971–1980, and 2001–2010.
Precipitation intensity–frequency–duration relation of (A) mean of all rainfall events and (B) extreme precipitation for (left) wet and (right) dry monsoon regimes for three different climate periods (colored lines). A rainfall event is defined as the occurrence of hourly rainfall ≥ 0.2 mm, and extreme rainfall is represented by the class of cumulative hourly rainfall exceeding the 95th percentile threshold for each duration class. Shading for the climate periods indicates the standard errors of the means per rainfall duration class. Vertical bars (labeled at the right axis) indicate the total number of wet events. Single events in a given rainfall duration class are not displayed.
For each duration class, the frequency of events shows a gradual increase over the three climate episodes (vertical bars in Fig. 2). This is most pronounced for the wet season and short-duration (1–3 h) events and is visible for both mean and extreme rainfall events. Cumulative rainfall with a duration between 2 h and 6 h has increased over time in the wet season, particularly for short-duration events. This signature is also visible for extreme rainfall, although with more noise. The number of short-duration events increases in both seasons, and these events tend to become more intense. In the dry season, no significant changes in the frequency are found.
To assess a temporal trend, we compute the annual average and the distribution of precipitation amount and frequency for each duration category. We distinguish short- (1–3 h), intermediate- (4–6 h), and long-duration (6–8 h) continuous rain. Short-duration rainfall generally comprises convective showers and is a dominant class in the tropics. Figure 3A indicates a significant positive change of frequency of short durations events during the wet season. An approximate doubling in the number of events is observed over the last century, which is statistically significant (p < 0.01), with a rapid increase of approximately 25 % over the period 2001–2010. A small frequency increase is also observed for the intermediate duration class, whereas longer duration events exhibit a downward frequency trend. For longer duration events the negative trend has relatively low confidence (p < 0.1) because of the small number of events in this category. In the dry season, no notable change in the frequency of each rainfall category is shown.
Annual frequency (A, left panels) and intensity (B, right panels) of short (1–3 h; top), intermediate (4–6 h; middle), and long rainfall events (> 6 h; bottom) for the wet season (NDJFM, blue lines/symbols) and dry season (MJJAS, red lines/symbols). The boxplot of the annual variability of monthly maximum rainfall represents the interquartile range (IQR), with the median shown as a horizontal line within the box. Vertical dashed lines indicate the extent of 1.5 times the IQR. Blue (red) thick lines represent low-pass-filtered annual NDJFM (MJJAS) series using the Loess smoother, with span parameter = 3/10 (compatible with 3 years moving average). Round dots represent individual outliers. Panel C shows the inter-annual variability of the Oceanic Niño Index (ONI) and Southern Oscillation Index (SOI) to indicate ENSO [El Niño (red) and La Niña (blue) events], and the Dipole Mode Index (DMI) to indicate IOD events.
For the intensity, Fig. 3B shows the annual distribution of monthly maximum hourly rainfall for each duration category. Short- and intermediate-duration categories show a significant intensity increase by 50 % (significant at p < 0.01) for both seasons, whereas no intensity trend is observed for the longer duration events. Contrary to the event frequency, the wet and dry seasons display a similar rainfall intensity distribution. Besides a systematic trend in short-duration rainfall, an apparent inter-annual to decadal variation is present in the frequency and intensity of Jakarta rainfall. This inter-annual to decadal variability is not clearly correlated to El Niño/La Niña or IOD events (Fig. 3C). This is in accordance with Lestari et al. (2019) who found only a weak relation between Jakarta's extreme precipitation and ENSO characteristics, particularly during the rainy season. Observed rainfall characteristics of the wider Maritime Continent do show a correlation with ENSO (e.g., Lestari et al. 2019; Supari et al. 2018) and the Pacific Decadal Oscillation (Yanto et al. 2016).
3.2 Diurnal timing of showersThe diurnal timing of extreme rain events is highly relevant for disaster preparedness and safety. Here, we present the change in the start time distribution of short-duration rainfall events over the diurnal cycle (Fig. 4). The start time of these short-duration rainfall events is sampled in 6-hourly diurnal time windows and displayed in 10-year aggregates. The morning time is defined as the starting time for the rainfall falling between 07 LT and 12 LT, afternoon rainfall events between 13 LT and 18 LT, night-time between 19 LT and 24 LT, and after midnight between 01 LT and 06 LT. Figure 4A shows that approximately 60 % of all rainfall events occur in the wet season (NDJFM), whereas approximately 20 % in the dry season (MJJAS). Most of these rainfall events occur in the afternoon (3 % and 1.5 % of all-time windows for the wet and dry seasons, respectively). The most visible change over the period from the early 20th century to the present is a steep increase in rainfall events occurring in the night-time and after midnight time windows during the wet season and a sharp decrease in afternoon rainfall events during the dry season. Night-time, after midnight, and morning rainfall events show a moderate increase over time.
(A) Fraction of short-duration rainfall events in diurnal time windows for each 10-year period from 1901 to 1980, and 2001 to 2010. Diurnal time windows are indicated by “morning” [rainfall starting hour between 07 and 12 local time (LT)], “afternoon” (13–18 LT), “night” (19–24 LT), and “after midnight” (01–06 LT), and are labeled separately for wet (NDJFM) and dry (MJJAS) seasons. (B) same as A but valid for short-duration rainfall in the category heavy shower (hourly rainfall > 10 mm). Note that there is a period gap between the period 1971–1980 and 2001–2010.
Concerning the timing of extreme short-duration rainfall events, we use a heavy shower threshold of > 10 mm h−1 (Kahlig 1993; also used by USGS4). Figure 4B shows that heavy shower events increase in each time window and in both seasons over the 1901–2010 time period, although with varying rates. Again, night-time heavy shower events exhibit a steeper increase for both seasons, and stronger increases are found for all heavy shower events in the wet season. These heavy shower events account for 6 % of all rainfall events, and many of them (∼50 %) occur during the wet season.
3.3 Changes in urban environment in relation to extreme precipitation a. Urban surface warming and drying as temperature risesRapid growth and urbanization have introduced increasing occasions of problematic flood conditions regularly experienced by communities. Urban expansion can also contribute to changing patterns of regional temperature and humidity, on top of the global warming trend (Hoegh-Guldberg et al. 2018). Figure 5A shows that annual mean relative humidity has decreased by approximately 6 % (p < 0.01) over the 20th century but with a considerable inter-annual and decadal variation. A significant increase in the surface temperature of approximately 1.6°C (p < 0.01) over the same period is observed (Fig. 5B). Dew point temperature remained virtually constant during most of the 20th century.
Time series of (A) annual average surface relative humidity, and (B) surface air temperature and dew point temperature collected at Jakarta Observatory from 1900 to 2010. Different sources of observations (see main text) are indicated with varying color shades. Gray lines show annual means calculated from daily means, black lines show low-pass-filtered annual data using the Loess smoother (compatible with 10-years moving average). Shaded colors indicate the standard errors of the annual means. (C) Annual average of relative humidity (blue lines) and temperatures (red lines) for urban (thick color) and rural (thin color) locations. Urban data represent the average of measurements taken at BMKG's meteorological stations 1, 2, and 3 in the Jakarta city area, whereas the rural data represent the average of locations 7 and 8 (see Fig. 1).
To assess the potential impact of urban development on regional meteorology, the difference in surface temperature and relative humidity trends between urban and rural areas is presented in Fig. 5C. Urban data represent the average of measurements taken at BMKG's meteorological stations within the Jakarta city (Kemayoran, Tanjung Priok, and Halim PK), whereas rural data are represented by the average of measurements taken at BMKG's meteorological stations at Darmaga and Citeko, Bogor (Fig. 1). Over the past 30 years, the decreasing trend in relative humidity in the rural area is similar to the trend observed in the city of Jakarta. Nevertheless, the steep temperature increase shown in the urban area is not present in the rural area where only a small non-significant positive trend is shown. This suggests an impact of the urban expansion of Jakarta on the regional warming trend.
Figure 6 displays the diurnal cycles of temperature, dew point temperature, and relative humidity for the urban Jakarta environment, contrasting the 1900–1980 climatology with the recent decade 2000–2010. As noted by (Siswanto et al. 2016), the air temperature increase is largest during the early morning and smallest at midday (Fig. 6A). Also, significant changes in morning dew point (Fig. 6B) and relative humidity (Fig. 6C) are shown. The diurnal cycle of relative humidity in the recent decade shows a different signature during night-time than in the climatological period, with a maximum in the early morning just before sunrise and low relative humidity during the afternoon and evening. The dew point temperature exhibits only a small variation throughout the day during the recent decade, and the early morning minimum is no longer present.
Mean diurnal cycle of (A) temperature, (B) dew point temperature, and (C) relative humidity for Jakarta Observatory Station for the 1900–1966 climatology and the recent decade 2000–2010. Corresponding shading and error bars indicate one standard deviation from the mean.
The combination of urban surface warming and drying in Jakarta city is consistent with earlier findings on the thermal environment in Jakarta by (Widyasamratri et al. 2014). Their results are based on seven temperature and humidity observation sites during a field campaign in the 2012 transition period between dry and the pre-wet monsoon season. Higher temperatures and lower humidity were observed around noon over densely urbanized areas than over the sparsely urbanized areas. A near-surface decrease in the humidity of the urban Jakarta atmosphere can be related to a reduction of evaporation over strongly urbanized surfaces. Additionally, the elevated inner-city temperature and reduced soil moisture availability contribute to lower urban relative humidity (e.g., Blake et al. 2011).
b. Scaling of extreme precipitation with local temperatureAtmospheric saturation vapor pressure increases with air temperature at 7 % °C−1 following the CC relationship. Increases in precipitation intensity following a rate exceeding CC are labeled as super-CC scaling (1CC and 2CC denote the scaling of 7 % and 14 % per degree warming, respectively). Following Lenderink and van Meijgaard (2010), Fig. 7 shows the results of a scaling analysis of extreme hourly precipitation as a function of temperature (T) and dew point temperature (Td). After sorting all hourly precipitation events in 1°C bin intervals of T and Td, the 90th, 99th, and 99.9th percentiles for each temperature bin are computed. To allow some time for convection to respond to environmental surface conditions, corresponding T and Td values are taken 4 h before the precipitation events.
Extreme precipitation as a function of temperature (T, left) and dew point temperature (Td, right) 4 h before the rainfall events at Jakarta Observatory for the 90th, 99th, and 99.9th percentiles (indicated by blue, green, and red, respectively) collected over the period 1900–2010. Shading represents 2 times the standard deviation derived from bootstrap resampling (n = 1000). Background dashed lines denote the 1CC and 2CC scaling relationships.
Figure 7A shows the results of this analysis for air temperature. It indicates a 2CC scaling of the 90th percentile over a temperature interval ranging between 22°C and 31°C. For the higher percentiles, this 2CC relation is found in a smaller temperature range (23–25°C), returning to a 1CC scaling at higher temperatures. The scaling relation between the dew point and the hourly rainfall extremes displays a 2CC relation until Td exceeds ∼25°C for all precipitation intensity categories. The dew point range in which 2CC applies is wider than the previous findings of Lenderink and van Meijgaard (2010), who detected 2CC scaling for a 12–22°C Td range for the Netherlands and Hongkong.
Observations at other locations, such as in South Korea (Park and Min 2017), the French Mediterranean region (Drobinski et al. 2018), and the Alps of south-eastern Austria (Schroeer and Kirchengast 2018), showed similar scaling behavior with varying (dew point) temperature ranges. A more recent study by Ali et al. (2021) shows that in a global dataset of over 7000 hourly precipitation gauges across six macro-regions, a consistently strong relationship between hourly precipitation extremes and dew point temperature is found at scaling rates from 1CC to 2CC at more than 60 % of the gauges, peaking in the tropics at a median rate of ∼1.5CC. 2CC scaling implies additional dynamical feedbacks on precipitation formation, such as large-scale dynamics, orography, and local features of atmospheric convection. Also, various physical mechanisms may be responsible for the limited extent to which 2CC applies.
Generally, the convection triggered in the (early) morning hours occurs under lower temperature conditions than afternoon convection, whereas moisture availability is sufficient. Intense convective precipitation must be fueled by available moisture, which may be sourced from advection and evaporation from surrounding land and ocean areas. During daytime under the dry surface and high-temperature conditions, the supply of moisture may be limited (Changnon 1981; Vogel et al. 2018).
3.4 Convective instability and extreme precipitationExtreme rainfall in Jakarta is governed by local thermodynamics, but the presence of super-CC scaling indicates that specific atmospheric convection and dynamical processes may intensify the precipitation formation process. We analyze the atmospheric convection and its possible dynamical interactions from 930 selected atmospheric sounding data from SHIA between 2002 and 2016. The soundings are selected for favorable conditions for triggering convection in a subsequent couple of hours. For both morning (00 UTC, 07 LT) and evening (12 UTC, 19 LT) soundings the convective available potential energy (CAPE) is calculated. Higher CAPE indicates an unstable planetary boundary layer and middle troposphere, which is favorable for convection by the rapid ascent of warm air parcels, particularly when abundant moisture in the lower troposphere is available (Seeley and Romps 2015).
The degree of convective instability and intensity of precipitation in this region is presumably dependent on the state of the Madden–Julian Oscillation MJO (Wu et al. 2015). Also, air mass advection by strong north-westerlies resulting from an emergence of northerly CS or CENS (Chang et al. 2005; Hattori et al. 2011; Yulihastin et al. 2019) during the peak of the Asian monsoon circulation (December to February) may play a role. However, CS frequency decreases when the MJO is present (Chang et al. 2005), and the MJO wind pattern tends to be opposite to the CS wind pattern.
In our analysis, we identify the role of MJO events on rainfall in Jakarta. For this, the Real-time multivariate MJO (RMM) index (Wheeler and Hendon 2004) is used, and MJO is indicated by RMM phase 3, 4, 5, or 6 over the Maritime Continent with an amplitude greater than 1 to ensure their significant appearance. We define MJO-triggered rainfall events as rainfall on a day where the rain is evenly distributed in all weather stations in Jakarta (Kemayoran, Halim PK, Tanjung Priok, Pondok Betung, or SHIA) with daily rainfall exceeding at least 5 mm day−1 or when daily precipitation exceeding 50 mm day−1 recorded in one of these stations.
Figure 8 shows the relationship between accumulated 3-hourly daytime and night-time rainfall and preceding surface temperature, dew point, and CAPE for days with and without a strong forcing from MJO. The results confirm that in general more rainfall and higher intensity occurs in conditions with MJO forcing (Fig. 8A) than without an MJO background (Fig. 8B), particularly for heavy and extreme rainfall categories (≥ 50 mm 3 h−1 and ≥ 100 mm 3 h−1). Nonetheless, its effect on sub-daily rainfall intensity is unclear (Fig. S2 on Supplementary Material). In the morning hours, CAPE is generally low. Rainfall intensity is distributed in narrow ranges of temperature and dew point. During MJO events, higher values of CAPE (1000–2500 J kg−1) are observed in the evening as temperature and dew point also show higher values, but rainfall intensity is not coherently associated with CAPE. More intense (≥ 50 mm 3 h−1 and ≥ 100 mm 3 h−1) rainfall is observed at a higher temperature, dew point, and CAPE, which preferably occurs at late afternoon or evening moments. Night-time rainfall is observed within a wider temperature range 23–30°C (dew point temperature range 22–26°C), whereas morning rainfall occurs commonly within a somewhat smaller temperature range 23–27°C (dew point temperature range 22–25°C). During MJO events the dependence of particularly night-time CAPE on local surface (dew point) temperature is stronger. A high CAPE (1000–2500 J kg−1) is commonly found when an MJO-dominated atmospheric background coincides with the monsoon systems in India. Strong pseudo-adiabatic buoyancy up to the upper troposphere is observed during an active monsoon period (Thomas et al. 2018).
CAPE (heatmap) and 3-hourly cumulative rainfall classes (circles) as a function of surface temperature (T) and dew point (Td), for daytime (upper panels) and night-time (bottom panels) for periods (A) with and (B) without large-scale dynamic MJO forcing. Daytime rainfall denotes 3-hourly cumulative rainfall in the 07–13 LT time window, associated with the morning sounding at 07 LT; night-time rainfall denotes 3-hourly cumulative in the 19–24 LT time window associated with the evening sounding at 19 LT. Cumulative rainfall is collected from 3-hourly rainfall records at SHIA, Pondok Betung, Tanjung Priok, and Kemayoran stations between 2002 and 2016. Only data with an intensity exceeding 10 mm (3 h)−1 time interval are classified and plotted.
In the subset of days without MJO forcing (Fig. 8B), intense rainfalls are generally less frequent than with MJO forcing. Most rainfall events occur in a somewhat more concentrated temperature range of 23–29°C. Night-time rainfall occurs at higher temperatures and dew points than during daytime. High CAPE values (1000–2000 J kg−1) occur within the 26–28°C temperature range in combination with high dew points. The response of precipitation to local surface temperature/dew point temperature is stronger under conditions without MJO forcing.
The dependence of extreme precipitation on local surface (dew point) temperature is governed by both thermodynamic and atmospheric dynamics aspects. Night-time rainfall events occur regardless of the MJO dynamic forcing. However, intense rainfall produced from greater convective instability is more frequently observed in a dynamic atmospheric background. Dong et al. (2019) pointed out that for Chinese stations in the south-eastern regions, a geographically coherent structure is found in the dependence of precipitation intensity to temperature-induced changes in available water vapor and atmospheric convection. From our observations over the Jakarta area, changes in convective instability related to surface warming do not show strong signals or markable changes in the likelihood for heavy or extreme precipitation.
Increasing short-duration rainfall extremes in a warmer climate have been widely addressed by recent research (e.g., Westra et al. 2014; Park and Min 2017; Bürger et al. 2019; Yu et al. 2020; Fowler et al. 2021). Our observations from historical data since 1900 show a significant increase in short-duration rainfall extremes in Jakarta (Figs. 2–4, 7), a representative for tropical humid-monsoon regions. Also, we observe an enhanced surface temperature trend in the urbanized area compared to surrounding rural stations. Annual average relative humidity monotonically decreases in both urban and rural areas (Fig. 5), whereas specific humidity and dew point increase slightly over time. We find that short-duration rainfall intensity and frequency increases particularly in the wet seasons. Jakarta is a delta city where water vapor is available abundantly from the surrounding tropical oceans. This leads to high values of the total water column and precipitation frequency.
For the Jakarta location, the increase in short-duration rainfall (extremes) during the wet season may be related to an enhanced land-ocean contrast in the response of moisture convergence [precipitation (P)–evaporation (E)] to global warming. Byrne and O'Gorman (2015) pointed out a larger change in P-E changes over the ocean than over land. Changes over the ocean broadly follow a thermodynamic scaling of the atmospheric moisture convergence: the so-called “wet-gets-wetter, dry-gets-drier” mechanism. This does not simply apply over the land area, where the P–E response is constrained by limited moisture availability. Changes in atmospheric circulation can overrule local precipitation statistics and modify horizontal gradients of temperature and fractional changes in relative humidity over land. During the wet season, a dominant north-westerly Asian monsoon transports abundant moisture evaporated from the sea toward the coastal city of Jakarta, which favors heavy precipitation.
Our analysis shows that short-duration rainfall events have a preference to occur in the night to morning hours, and its frequency increases significantly in more recent decades in Jakarta. Nevertheless, early morning rainfall is also observed to increase in the wet season. Excessive rainfall in the city center or near the coast that occurs in the early morning time (Yulihastin et al. 2019; Nuryanto et al. 2019) led to big flooding events in 2013, 2014, 2015 (Siswanto et al. 2015 2016, 2017), and latest in 2020. During 1992–1999, boundary layer radar data analyzed by (Renggono et al. 2001) showed a small shift from stratiform to deep convective precipitation during December to February between 03 LT and 06 LT at Serpong, 37 km southwest off Kemayoran Jakarta. In the Maritime Continent, the typical afternoon precipitation peaks over land are complemented with convective morning precipitation peaks, which are governed by the ocean (Mori et al. 2004; Qian 2008; Yamanaka et al. 2018). This is driven by sea-land breezes interaction with the prevailing background monsoon flows (Hadi et al. 2002; Araki et al. 2006).
The urbanization of the Greater Jakarta area has likely contributed to the increase in surface temperature (Fig. 5) and subsequently to an increase in CAPE. Modeling studies confirm high CAPE values in the presence of high near-surface temperatures (Rasmussen et al. 2020; Meredith et al. 2019). Nevertheless, additional (dynamical) processes other than CAPE affect convection and rainfall formation. A modeling study by Kanda et al. (2001) showed that urbanization affected the intensity and position of a local cumulus cloud formation in the Tokyo metropolitan area through enhancement of local upward motions and changes in the near-surface horizontal pressure gradient between urban and sub-urban areas. The urban heat and dry island may deform or change the behavior of sea breeze circulation (Martilli 2003; Tokairin et al. 2010).
Low-level moisture convergence and wind convergence anomalies tend to be larger over a drier (warmer) surface and can lead to precipitation anomalies, as demonstrated in a low-resolution general circulation model with an idealized continent situated on the equator (Cook 1994). At the latitude of the ITCZ, dry convection drives anomalous low-level convergence, which is larger over a drier (warmer) surface. In the case of Jakarta, similar warm surface conditions in the city may generate mesoscale convective systems during night-time (Mori et al. 2018), which leads to intense rainfall events.
The warmer city surface may also interact with converging local circulations (land–sea breeze), and further promote the repeated development of convective systems that generate rainfall in the early morning (Siswanto et al. 2016). A modeling study on impacts of land-use changes on local meteorological conditions in the Jakarta area by (Tokairin et al. 2010) revealed that in the present-day land-use scenario over Jakarta the sea breeze develops at an earlier moment in the diurnal cycle than for historic land use conditions, caused by heat advection from the new urban areas in South Jakarta into the old city (Central Jakarta). To investigate the complex combination of mechanisms leading to heavy precipitation in the Jakarta area, additional analyses are required, for instance, by regional modeling work where different urbanization scenarios are included to explore its impact on the development of extreme precipitation. This modeling set-up would require substantial tuning of the experimental design and computing resources and is out of scope for the present study.
Our analysis indicates that both local factors and the large-scale dynamics background contribute to the rainfall characteristics. Local extreme rainfall–temperature scaling is stronger when the large-scale atmospheric forcing is smaller. This is consistent with the finding of (Magan et al. 2020) for the rainfall–temperature sensitivity (scaling) in tropical Australia, who demonstrated this scaling to vary with weather types. Wet monsoon weather types result in 1CC scaling, whereas dry-weather types (with small remote advection) produce up to twice a 1CC scaling. The increase in convective precipitation extremes with surface warming is also governed by the stratification of the overlying atmosphere. Super-CC scaling has been found in model experiments where the surface warming was extended uniformly with height, which reduces the static stability of a moist atmosphere leading to faster updrafts (Attema et al. 2014). Exploring the mechanisms of climate change-induced convective activity in the midlatitudes, strong increases of CAPE in climate change model simulations were found to be associated with near-uniform vertical temperature changes (Byrne and O'Gorman 2015).
Observed increases of available moisture in Jakarta (particularly in the early morning), together with convergence processes of the local circulation or moisture advection from large-scale atmospheric circulation features, promote the repeated development of large convective systems that lead to intensification of morning precipitation and rainfall peak shifting from afternoon to night-time. Meredith et al. (2019) showed that future intensification of intense rainfall events has a strong diurnal signal, with the mid-morning period displaying the largest intensification. This leads to a shift of the diurnal convective precipitation maximum from late afternoon to the overnight/morning period under a strong climate change scenario.
Using a Jakarta data record with a length exceeding 100 years, characteristics of sub-daily precipitation extremes are analyzed in conjunction with temperature and relative humidity readings. A significant frequency increase of approximately 25 % in short-duration (1–3 h) rainfall events was observed in 2001–2010 compared with the start of the 20th century, and the cumulative rainfall amount related to these events has increased by 50 %. The most intense rainfall events are getting shorter in duration. Apart from a change in duration and intensity of events, a shift in the diurnal timing of the events from the afternoon to later in the night and into the early morning is observed.
Changes in the surface environment of urban Jakarta throughout the last century probably contributed to increases in CAPE. Surface warming in the urbanized city together with additional moisture availability resulting from large-scale atmospheric features in the lower atmosphere may result in warmer and more unstable air aloft, particularly in the mid-troposphere, implying a larger buoyancy into the mid-troposphere and larger and higher convective updrafts. A direct impact of urbanization on rainfall characteristics is not demonstrated by our observational analysis and could be explored in numerical simulation experiments.
For a subset of our data intense precipitation was found to increase with temperature at a rate approximating twice the CC (2CC) scaling, constrained to a limited temperature regime and high precipitation intensity thresholds. Increased atmospheric moisture in the early morning may promote 2CC temperature scaling triggering strong morning rainfall events, whereas 1CC temperature scaling is found for less extreme rainfall events. Increased extreme rainfall intensity during the night and morning time (when the air is still humid) is in agreement with the hypothetical precipitation response to considerably warmer and moister climate conditions than the present day. Given the location of Jakarta as a delta city, the abundant moisture evaporated from the sea is a continuous source of low-level moisture convergence that may develop into intense convection resulting in heavy precipitation. The study highlights the need for an appropriate and well-maintained sub-daily rainfall database, which gives the opportunity to accurately monitor the impact of a changing climate on floods in urban areas.
The supplement for this article consists of two figures, Figs. S1 and S2. Figure S1 (left panel) shows the missing value and presents hourly data for temperatures, dew point, and precipitation at Jakarta Observatory during 1866–1980 (right panel). Example of homogeneity check for the 1866–1966 dew point time series using RHVtest4 (see text for explanation). Figure S2 shows the relation between Jakarta rainfall and MJO phases. Step gray line chart indicates the frequency of the day when the rainfall is at least 5 mm day−1 is evenly distributed in all weather stations within the urban Jakarta area (Kemayoran, Halim PK, Tanjung Priok, Pondok Betung, or SHIA) or when daily precipitation exceeding 50 mm day−1 recorded in one of these stations. Only MJO events with an amplitude greater than 1 were selected. Boxplots indicate the sub-daily rainfall intensity that occurred in Jakarta corresponding to each MJO phase. The colored star at the end of the data series outside the boxplot for each sub-daily rainfall shows an outlier, which is the highest value of the entire data series. In terms of extreme events, this value can be identified as extreme rainfall with an extraordinary intensity far from its normal range.
This study was supported by the Joint Cooperation Program between KNMI (the Netherlands), BMKG (Indonesia), PusAir (Indonesia), and Deltares (the Netherlands). We would like to thank Mr. Agie Wandala (Center for Meteorology, BMKG) and the Meteorological Station of the SHIA for providing the upper-air data used in this study.
1 https://www.idare-portal.org/data/digitisasi-data-historisdidah
2 http://etccdi.pacificclimate.org/software.shtml
3 https://www.R-project.org
4 https://water.usgs.gov/edu/
