• Aldrian, E., and R. D. Susanto, 2003: Identification of the three dominant rainfall regions within Indonesia and their relation to sea surface temperature. Int. J. Climatol., 23, 14351452, doi:10.1002/joc.950.

    • Search Google Scholar
    • Export Citation
  • Alexandersson, H., 1986: A homogeneity test applied to precipitation data. J. Climatol., 6, 661675, doi:10.1002/joc.3370060607.

  • Boyard-Micheau, J., P. Camberlin, N. Philippon, and V. Moron, 2013: Regional scale rainy season onset detection: A new approach based on multivariate analysis. J. Climate, 26, 89168928, doi:10.1175/JCLI-D-12-00730.1.

    • Search Google Scholar
    • Export Citation
  • Buishand, T. A., 1982: Some methods for testing the homogeneity of rainfall records. J. Hydrol., 58, 1127, doi:10.1016/0022-1694(82)90066-X.

    • Search Google Scholar
    • Export Citation
  • Camberlin, P., and M. Diop, 2003: Application of daily rainfall principle component analysis to the assessment of the rainy season characteristics in Senegal. Climate Res., 23, 159169, doi:10.3354/cr023159.

    • Search Google Scholar
    • Export Citation
  • Choi, G., and Coauthors, 2009: Changes in means and extreme events of temperature and precipitation in the Asia-Pacific Network region, 1955–2007. Int. J. Climatol., 29, 19061925, doi:10.1002/joc.1979.

    • Search Google Scholar
    • Export Citation
  • ECA&D Project Team, 2012: European Climate Assessment & Dataset Algorithm Theoretical Basis Document (ATBD), version 10.5. ECA&D, De Bilt, Netherlands, 45 pp.

  • Hamada, J.-J., M. D. Yamanaka, J. Matsumoto, S. Fukao, P. A. Winarso, and T. Sribimawati, 2002: Spatial and temporal variations of the rainy season over Indonesia and their link to ENSO. J. Meteor. Soc. Japan, 80, 285310, doi:10.2151/jmsj.80.285.

    • Search Google Scholar
    • Export Citation
  • Hamed, K. H., and A. R. Rao, 1998: A modified Mann–Kendall trend test for autocorrelated data. J. Hydrol., 204, 182196, doi:10.1016/S0022-1694(97)00125-X.

    • Search Google Scholar
    • Export Citation
  • Hartmann, D., and Coauthors, 2013: Observations: Atmosphere and surface. Climate Change 2013: The Physical Science Basis, T. Stocker et al., Eds., Cambridge University Press, 159–255.

  • Haylock, M. R., and J. McBride, 2001: Spatial coherence and predictability of Indonesia wet season rainfall. J. Climate, 14, 38823887, doi:10.1175/1520-0442(2001)014<3882:SCAPOI>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Held, I. M., and B. J. Soden, 2006: Robust responses of the hydrological cycle to global warming. J. Climate, 19, 56865699, doi:10.1175/JCLI3990.1.

    • Search Google Scholar
    • Export Citation
  • Klein Tank, A. M. G., and Coauthors, 2002: Daily dataset of 20th-century surface air temperature and precipitation series for the European Climate Assessment. Int. J. Climatol., 22, 14411453, doi:10.1002/joc.773.

    • Search Google Scholar
    • Export Citation
  • Klein Tank, A. M. G., F. Zwiers, and X. Zhang, 2009: Guidelines on analysis of extremes in a changing climate in support of informed decisions for adaptation. Climate Data and Monitoring WCDMP-No. 72, WMO/TD-1500, 52 pp. [Available online at http://www.wmo.int/pages/prog/wcp/wcdmp/wcdmp_series/documents/WCDMP_72_TD_1500_en_1.pdf.]

  • Liebmann, B., S. J. Camargo, A. Seth, J. A. Marengo, L. M. V. Carvalho, D. Allured, C. S. Fu, and R. Vera, 2007: Onset and end of the rainy season in South America in observations and the ECHAM 4.5 atmospheric general circulation model. J. Climate, 20, 20372050, doi:10.1175/JCLI4122.1.

    • Search Google Scholar
    • Export Citation
  • Lintner, B. R., M. Biasutti, N. S. Diffenbaugh, J.-E. Lee, M. J. Niznik, and K. L. Findell, 2012: Amplification of wet and dry month occurrence over tropical land regions in response to global warming. J. Geophys. Res., 117, D11106, doi:10.1029/2012JD017499.

    • Search Google Scholar
    • Export Citation
  • Marteau, R., V. Moron, and N. Philippon, 2009: Spatial coherence of monsoon onset over western and central Sahel (1950–2000). J. Climate, 22, 13131324, doi:10.1175/2008JCLI2383.1.

    • Search Google Scholar
    • Export Citation
  • Moron, V., A. W. Robertson, and R. Boer, 2009: Spatial coherence and seasonal predictability of monsoon onset over Indonesia. J. Climate, 22, 840850, doi:10.1175/2008JCLI2435.1.

    • Search Google Scholar
    • Export Citation
  • Naylor, R. L., D. S. Battisti, D. J. Vimont, W. P. Falcon, and M. B. Burke, 2007: Assessing risks of climate variability and climate change for Indonesian rice agriculture. Proc. Natl. Acad. Sci. USA, 104, 77527757, doi:10.1073/pnas.0701825104.

    • Search Google Scholar
    • Export Citation
  • Nguyen-Le, D., J. Matsumoto, and T. Ngo-Duc, 2015: Onset of the rainy season in the eastern Indochina Peninsula. J. Climate, 28, 56455666, doi:10.1175/JCLI-D-14-00373.1.

    • Search Google Scholar
    • Export Citation
  • Nieuwolt, S., 1989: Estimating the agricultural risks of tropical rainfall. Agric. For. Meteor., 45, 251263, doi:10.1016/0168-1923(89)90047-6.

    • Search Google Scholar
    • Export Citation
  • Peng, S., and Coauthors, 2004: Rice yield decline with higher night temperature from global warming. Proc. Natl. Acad. Sci. USA, 101, 99719975, doi:10.1073/pnas.0403720101.

    • Search Google Scholar
    • Export Citation
  • Pettitt, A. N., 1979: A non-parametric approach to the change-point problem. Appl. Stat., 28, 126135, doi:10.2307/2346729.

  • Polade, S., D. Pierce, D. Cayan, A. Gershunov, and M. Dettinger, 2014: The key role of dry days in changing regional climate and precipitation regimes. Sci. Rep., 4, 4364, doi:10.1038/srep04364.

    • Search Google Scholar
    • Export Citation
  • Purseglove, J. W., Ed., 1988: Tropical Crops—Monocotyledons. Longman Scientific and Technical, 607 pp.

  • Schmidt, F. H., and J. Van der Vecht, 1952: East Monsoon fluctuations in Java and Madura during the period 1880-1940. Kementerian Perhubungan Verhandelingen No. 43, Djawatan Meteorologi dan Geofisik, Djakarta, Indonesia, 45 pp.

  • Seager, R., N. Naik, and L. Vogel, 2012: Does global warming cause intensified interannual hydroclimatic variability? J. Climate, 25, 33553372, doi:10.1175/JCLI-D-11-00363.1.

    • Search Google Scholar
    • Export Citation
  • Sen, P. K., 1968: Estimates of the regression coefficient based on Kendall’s tau. J. Amer. Stat. Assoc., 39, 13791389, doi:10.2307/2285891.

    • Search Google Scholar
    • Export Citation
  • Siswanto, S., R. Jilderda, G. Van der Schrier, G. J. Van Oldenborgh, and B. Van den Hurk, 2016: Trend of temperature, extreme precipitation, and diurnal rainfall changes in the urbanized Jakarta city during the past 130 years. Int. J. Climatol., doi:10.1002/joc.4548, in press.

    • Search Google Scholar
    • Export Citation
  • Smith, I. N., L. Wilson, and R. Suppiah, 2008: Characteristics of the Northern Australian rainy season. J. Climate, 21, 42984311, doi:10.1175/2008JCLI2109.1.

    • Search Google Scholar
    • Export Citation
  • Suppiah, R., 1993: ENSO phenomenon and 30-50 day variability in the Australian summer monsoon rainfall. Int. J. Climatol., 13, 837851, doi:10.1002/joc.3370130803.

    • Search Google Scholar
    • Export Citation
  • Van den Besselaar, E. J. M., and Coauthors, 2015: International Climate Assessment & Dataset (ICA&D): Climate services across borders. Bull. Amer. Meteor. Soc., 96, 1621, doi:10.1175/BAMS-D-13-00249.1.

    • Search Google Scholar
    • Export Citation
  • Vecchi, G. A., and B. J. Soden, 2007: Global warming and the weakening of the tropical circulation. J. Climate, 20, 43164340, doi:10.1175/JCLI4258.1.

    • Search Google Scholar
    • Export Citation
  • Von Neumann, J., 1941: Distribution of the ratio of the mean square successive difference to the variance. Ann. Math. Stat., 12, 367395, doi:10.1214/aoms/1177731677.

    • Search Google Scholar
    • Export Citation
  • Von Storch, H., and F. W. Zwiers, 1999: Statistical Analysis in Climate Research. Cambridge University Press, 484 pp.

  • Wheeler, T., and J. Von Braun, 2013: Climate change impacts on global food security. Science, 341, 508513, doi:10.1126/science.1239402.

    • Search Google Scholar
    • Export Citation
  • Wijngaard, J. B., A. M. G. Klein Tank, and G. P. Können, 2003: Homogeneity of 20th century European daily temperature and precipitation series. Int. J. Climatol., 23, 679692, doi:10.1002/joc.906.

    • Search Google Scholar
    • Export Citation
  • Wilks, D. S., 1995: Statistical Methods in the Atmospheric Sciences: An Introduction. Academic Press, 467 pp.

  • Williamson, F., R. Allan, A. D. Switzer, J. C. L. Chan, R. J. Wasson, R. D’Arrigo, and R. Gartner, 2015: New directions in hydro-climatic histories: Observational data recovery, proxy records and the Atmospheric Circulation Reconstructions over the Earth (ACRE) initiative in Southeast Asia. Geosci. Lett., 2, doi:10.1186/s40562-015-0018-z.

    • Search Google Scholar
    • Export Citation
  • Zhang, H., A. Moise, P. Liang, and L. Hanson, 2013: The response of summer monsoon onset/retreat in Sumatra-Java and tropical Australia region to global warming in CMPI3 models. Climate Dyn., 40, 377399, doi:10.1007/s00382-012-1389-x.

    • Search Google Scholar
    • Export Citation
  • View in gallery

    Maps with (left) station locations for daily minimum temperature and (right) daily sums of precipitation. Figure valid for blended data. The green dots represent stations for which the actual daily data are available online; daily data from the red dots are not accessible. Climate indices data, aggregated over time, are accessible for all stations.

  • View in gallery

    Number of stations with daily precipitation (rr), daily maximum (tx), average (tg), and minimum (tn) temperature against time. The lines for the daily minimum and maximum temperatures coincide. Figure valid for blended data.

  • View in gallery

    The seasonal cycle of the median (red) and 75th percentile (green) of daily precipitation based on the 1971–2000 climatology. The vertical bars show the climatological mean yearday of the onset of the wet season, following the definitions of Liebmann et al. (2007, L), Smith et al. (2008, S), or the BMKG definition (B). Stations shown are (a) Dagupan; (b) Phu Lien; (c) Blang Bintang–Banda Aceh; (d) Jakarta; (e) the National Weather Service Office (WSO) in PohnPei, Micronesia; and (f) Darwin International Airport, Northwest Territory, Australia. For Banda Aceh, the climatological period 1981–2010 is used because there is too much missing data prior to 1980 for this station.

  • View in gallery

    Onset and retreat dates of the wet season for Jakarta, using the (a) Liebmann et al. (2007), (b) BMKG, and (c) Smith et al. (2008) definitions. The bottoms and tops of the gray bars denote the onset and retreat of the wet season, respectively (in yearday). The length of the gray bar is therefore a measure of the length of the wet season. The red lines show low-pass-filtered onset and retreat dates using the Lowess smoother function with parameters f = 1/5 and iter = 3.

  • View in gallery

    Correlations between the yearday of the onset the wet season between the (a) Liebmann et al. (2007) and BMKG, (b) Liebmann et al. (2007) and Smith et al. (2008), and (c) BMKG and Smith et al. (2008) definitions.

  • View in gallery

    Climatological mean in onset dates of the wet season for Southeast Asia in yearday using data from the period 1981–2010. Onset dates defined by the (a) Liebmann et al. (2007), (b) BMKG, and (c) Smith et al. (2008) definitions.

  • View in gallery

    Standard deviation of the onset date of the wet season over Southeast Asia, calculated over 1981–2010. Onset dates defined by the (a) Liebmann et al. (2007), (b) BMKG, and (c) Smith et al. (2008) definitions.

  • View in gallery

    Composite of the onset of the wet season for (left) El Niño and (right) La Niña conditions, as a deviation from the 1981–2010 climatology. The Liebmann et al. (2007) index is used in this plot.

  • View in gallery

    Trends in the onset date (days decade−1) of the wet season over Southeast Asia, calculated over 1971–2012. Gray circles denote stations with nonsignificant trends (at the 75% level). Onset dates defined by the (a) Liebmann et al. (2007), (b) BMKG, and (c) Smith et al. (2008) definitions.

  • View in gallery

    Trends in (left) sums of precipitation (mm decade−1) and (right) the number of wet days (precipitation ≥1 mm), calculated for the wet season (DJF) over the 1971–2012 period. Colored dots are significant at the 75% level. Note the change in scale in the trends of precipitation sums between the wet and the dry seasons.

  • View in gallery

    Probability plots for monthly precipitation sums over the periods 1971–90 and 1991–2010 for the (left) DJF and (right) AMJ seasons. The blue lines give the fitted gamma distribution based on the months with nonzero precipitation. The plots relate to (a),(b) west Java and Banten, (c),(d) central Java and Yogyakarta, and (e),(f) east Java and Bali.

  • View in gallery

    Median values for the period (in days) between the onset of the wet season and a potential false start Values are calculated over the 1981–2010 period. Right panel for Java only.

  • View in gallery

    Trends in daily minimum temperature (°C decade−1), calculated over the 1971–2012 period: (left) DJF and (right) JJA. Colored dots are significant at the 95% level.

  • View in gallery

    Probability plots for daily minimum temperatures over the periods 1971–90 and 1991–2010 for the (left) DJF and (right) JJA seasons. The plots relate to (a),(b) Cilacap, (c),(d) Tacloban, and (e),(f) Suphan Buri.

  • View in gallery

    The fraction of stations vs the interstation distance in the blending.

  • View in gallery

    The fraction of stations vs the interstation distance in the blending.

  • View in gallery

    Maximum correlations between the first multicombination PC and the onset date based on a combination of N, P, and C for each station.

  • View in gallery

    Maps of (top) N, (middle) P, and (bottom) C for which the correlation with the first multicombination PC is highest.

  • View in gallery

    Median values for the period (in days) between the first day of a 5-day period receiving 40 mm or more and the first day of a 5-day period receiving 40 mm or more and having no dry spell of 7 days or longer in the next 30 days. Values are calculated over the 1981–2010 period. Right panel for Java only.

All Time Past Year Past 30 Days
Abstract Views 0 0 0
Full Text Views 302 272 10
PDF Downloads 246 222 5

Observed Trends and Variability in Climate Indices Relevant for Crop Yields in Southeast Asia

View More View Less
  • 1 Badan Meteorologi, Klimatologi dan Geofisika, Jakarta Pusat, Indonesia
  • | 2 Royal Netherlands Meteorological Institute (KNMI), De Bilt, Netherlands
  • | 3 Badan Meteorologi, Klimatologi dan Geofisika, Jakarta Pusat, Indonesia
© Get Permissions
Full access

Abstract

Climate indices are analyzed using a newly developed dataset with station-based daily data for Southeast Asia. With rice the staple food of the diet in the region, the indices used are aimed at agriculture, specifically rice production, and include the onset of the wet season and the nighttime temperature. Three indices are used to estimate the onset of the wet season. Despite a quantitative lack of similarity between these indices (although they are strongly correlated), the progression of the wet season over the area matches existing descriptions. Trends in the onset date of the wet season calculated over 1971–2012 are only statistically significant for a few stations; there are no signs that a wide spread delay as anticipated by future climate scenarios is already taking place. A positive trend in the nighttime temperature over the region is observed with values up to 0.7°C decade−1. For a selection of stations the change in distribution of nighttime temperatures is analyzed when comparing the 1971–90 period with the 1991–2010 period. They show a shift of the median to higher temperatures, and the decline in the number of relatively cool nights is stronger than the increase in the number of relatively warm nights.

Corresponding author address: G. van der Schrier, Royal Netherlands Meteorological Institute (KNMI), P.O. Box 201, 3730 AE De Bilt, Netherlands. E-mail: schrier@knmi.nl

Abstract

Climate indices are analyzed using a newly developed dataset with station-based daily data for Southeast Asia. With rice the staple food of the diet in the region, the indices used are aimed at agriculture, specifically rice production, and include the onset of the wet season and the nighttime temperature. Three indices are used to estimate the onset of the wet season. Despite a quantitative lack of similarity between these indices (although they are strongly correlated), the progression of the wet season over the area matches existing descriptions. Trends in the onset date of the wet season calculated over 1971–2012 are only statistically significant for a few stations; there are no signs that a wide spread delay as anticipated by future climate scenarios is already taking place. A positive trend in the nighttime temperature over the region is observed with values up to 0.7°C decade−1. For a selection of stations the change in distribution of nighttime temperatures is analyzed when comparing the 1971–90 period with the 1991–2010 period. They show a shift of the median to higher temperatures, and the decline in the number of relatively cool nights is stronger than the increase in the number of relatively warm nights.

Corresponding author address: G. van der Schrier, Royal Netherlands Meteorological Institute (KNMI), P.O. Box 201, 3730 AE De Bilt, Netherlands. E-mail: schrier@knmi.nl

1. Introduction

An increase in extreme climatic events such as the intensity and number of heavy rainfall days and prolonged periods of hot days has been linked to global warming (Hartmann et al. 2013). These changes in climate affect human lives and have a large impact on society as a whole. This is reflected in the slowing down of the long-term reduction in the prevalence of undernutrition, which is partly due to extreme climatic events (Wheeler and Von Braun 2013). This is particularly relevant for Southeast Asia because agriculture and food security are vulnerable sectors already, given that the Global Hunger Index 2012 (http://www.ifpri.org/publication/2012-global-hunger-index) attaches the level “serious” to the largest part of Southeast Asia, with only Malaysia and Thailand having reached the “moderate” level.

Tropical warming will increase the water vapor content in the atmosphere and this is anticipated to cause an increase in precipitation, with wet areas getting wetter and dry areas getting drier (the “rich get richer”) as a result of a weakening of the hydrological cycle (Held and Soden 2006; Vecchi and Soden 2007). In addition to changes in the mean, the variability of precipitation will increase under global warming (Seager et al. 2012), especially for Southeast Asia, meaning that occasional dry spells may occur even more frequently (Polade et al. 2014; Lintner et al. 2012). These changes in the hydrological cycle affect the monsoon rains, which are vitally important for agriculture in the region. The onset of the wet season is anticipated to be delayed and the duration shortened by approximately 10% in climate scenario simulations for the area between the Sumatra–Java archipelago and Northern Australia (Zhang et al. 2013). This is in line with an earlier study by Naylor et al. (2007), who concluded that rice production is at risk in the two most dominant rice-producing areas in Indonesia by an increased probability of a 30-day delay in the onset of the wet season as a result of changes in the mean climate, which they anticipate happening by 2050.

One aspect of societal vulnerability to precipitation variability relates to delays in the onset of the wet season, which causes the rice crop to be planted later in the season, thus extending the period before the main rice harvest. Furthermore, delayed planting of the main wet season crop may not be compensated by increased planting later in the crop year, giving a reduced rice crop yield (Naylor et al. 2007). The relation between a delayed onset of the wet season and a reduced rice crop is quantified for some Southeast Asian countries in Table 1, which shows a negative correlation between paddy rice yield from the country and the onset date for a selected station in that country. The selection of stations is made on the basis of completeness and quality of the record and if the station is in a rice-producing area. The correlations indicate a relation (although rather weak) between the two.

Table 1.

Correlations between the national accumulated paddy rice yield and the onset date of the wet season for a selected station in three countries in Southeast Asia. The selected stations are in rice-producing areas. The significance level of the correlation is given parentheses and the rice production figures have been detrended prior to correlation (using a second-degree polynomial for Indonesia and a linear function for Vietnam and Thailand). The onset date is calculated using the Liebmann index (introduced in section 3). Paddy rice yield data sourced from the Food and Agriculture Organization of the United Nations (http://faostat.fao.org/).

Table 1.

The aim of this study is to establish a climatology of the onset of the wet season across Southeast Asia and to quantify possible trends in this onset. There are many different definitions of the onset of the wet season. In this study we use three indices to capture some of the variety associated with the choice of index. The use of climate indices contrasts with the approach followed recently by Nguyen-Le et al. (2015), who determine onset dates of the wet season using an EOF analysis.

Some attention will be given to false starts of the wet season and we will quantify the climatology of such false starts and look into possible trends in this risk. False starts of the wet season are a common and major problem for farmers, who need additional resources for new seedlings when the first have withered away. Analyzing false starts is motivated by the possibility that a coincidental combination of the delayed onset of the wet season, as seen in future climate scenario simulations (Zhang et al. 2013; Naylor et al. 2007), with more extreme precipitation events, as are expected to occur under climate change, may confuse farmers into thinking that the wet season has an early start.

Using data from a field experiment in the Philippines, Peng et al. (2004) reported for the period 1979–2003 a rice yield decline by 10% for each 1°C increase in minimum (~nighttime) temperature during the growing season, whereas the effects of maximum temperatures on crop yield were insignificant. The high nighttime temperature results in a high respiration rate, reducing net dry-weight gains (Purseglove 1988). This links decreasing rice yields with increased nighttime temperature and a further aim of this study is to quantify trends in nighttime temperature over Southeast Asia.

This study will focus on an analysis of climate indices that describe a particular aspect of climate—both changes in the mean and the extremes. The choice of the analyzed climate indices relates to their relevance for agriculture and, in particular, the rice crop yield. Relating changes in the actual rice crop yield to climate variability is beyond the scope of this study.

The paper is organized as follows. Section 2 introduces the dataset used, section 3 introduces the climate indices studied, and section 4 gives the results. The study is summarized and concluded in section 5.

2. Data and method

a. Southeast Asian climate assessment and dataset

Data and analyses for this study are based on the Southeast Asian Climate Assessment and Dataset (SACA&D, http://sacad.database.bmkg.go.id). SACA&D is a climate-monitoring tool and data portal specifically aimed at Southeast Asia, combing the Association of Southeast Asian Nations (ASEAN) countries with the WMO’s region V, and aims to put particular emphasis on changes in extremes. The data in SACA&D are brought together through cooperation of the national meteorological services of the region. SACA&D provides uniform analyzing methodologies to daily observational series from as many stations from the Southeast Asian region as possible. At the time of writing (January 2016), SACA&D receives data from 23 participants in 15 countries and the SACA dataset contains in total 5926 series of observations. These series are observed at 4066 meteorological stations. The most dense coverage is provided for precipitation and maximum and minimum temperature (Fig. 1), but other elements like relative humidity, sunshine duration, and wind are included in the dataset as well.

Fig. 1.
Fig. 1.

Maps with (left) station locations for daily minimum temperature and (right) daily sums of precipitation. Figure valid for blended data. The green dots represent stations for which the actual daily data are available online; daily data from the red dots are not accessible. Climate indices data, aggregated over time, are accessible for all stations.

Citation: Journal of Climate 29, 7; 10.1175/JCLI-D-14-00574.1

The infrastructure of SACA&D is based on its European counterpart, the European Climate Assessment and Dataset [ECA&D; information online at http://www.ecad.eu; Klein Tank et al. (2002)], which has formed the backbone of the climate data node in the Regional Climate Centre (RCC) for WMO region VI (Europe and the Middle East) since 2010. Both SACA&D and ECA&D are members of the International Climate Assessment and Dataset (ICA&D), which is part of a pilot project of WMO’s Global Framework for Climate Services (GFCS) (Van den Besselaar et al. 2015).

Within SACA&D two types of daily data exist. One is “blended” data, and the other is “nonblended” data. The data collected from the participating countries may contain gaps or may contain separate time series from discontinued and relocated stations. In the blending, gaps in a daily series are filled with observations from nearby stations, and irregular time series from a discontinued station and its replacement are rejoined, provided that they are within 25-km distance and that height differences are less than 50 m. Only the blended series are further analyzed in SACA&D and are used for the calculation of the climate indices and trends.

The motivation for the blending step is to provide records that are as long and as complete as possible. The requirement that a rather short distance should exist between stations when blending is applied, which means that the climatology of the donating station is likely to be similar to that of the receiving station, except perhaps in very complex terrain. Another argument is that if the donating station has a very different climatology than the receiving station, and the amount of data donated to the receiving station is large enough to significantly change the character of the receiving station, then the homogeneity test (see section 2b) would detect this mismatch between stations and flag the series. Flagged series are not used in any trend analysis (avoiding the possibility of mistaking a trend due to a mismatch of stations for climate change).

Some statistics on the percentage of rain gauges that have blended data and the distribution of the interstation distances between the receiving and donating stations is given in appendix A.

Figure 2 shows the number of stations per year for precipitation and temperature having blended data. This figure shows rapid growth in the number of stations from the early 1950s onward, to a maximum of 381 stations for temperature and a maximum of 3835 for precipitation in 1990 (status January 2016). The sharp decline observed in the most recent years is due to the fact that many series do not extend up to a very recent date. Note that the number of stations for the first half of the twentieth century is very low. Finally, Fig. 2 shows that the growth of the number of stations is not monotonic, especially from the 1970s onward. This is either an indication that stations are discontinued during that period and other stations have started to produce data, or that only parts of series are digitally available (or shared with SACA&D). Both result in a situation with relatively many stations producing rather short time series.

Fig. 2.
Fig. 2.

Number of stations with daily precipitation (rr), daily maximum (tx), average (tg), and minimum (tn) temperature against time. The lines for the daily minimum and maximum temperatures coincide. Figure valid for blended data.

Citation: Journal of Climate 29, 7; 10.1175/JCLI-D-14-00574.1

b. Data quality and homogeneity tests

The data in all series are quality checked using a set of rules documented in the Algorithm Theoretical Basis Document (ECA&D Project Team 2012). These tests include sanity tests, outlier tests, and consistency tests. Data are flagged as “useful” (i.e., data that have passed the test), “suspect” (i.e., data that failed to pass the test) or “missing.” Only the data with a useful quality flag are used in further analysis. No corrections or adjustments are made to the data.

Another test applied to the data is the homogeneity test. Long climatological series often contain variations or discontinuities due to nonclimatic factors, like site relocations, or changes in observational procedures or in the surroundings of the station. High quality metadata are generally lacking when trying to identify these changes. As these types of inhomogeneities can distort or hide the climatic signal, statistical techniques for homogeneity testing are applied. Within SACA&D, the approach of Wijngaard et al. (2003) is used to classify the homogeneity of the temperature and precipitation series. The procedure applies four tests: the standard normal homogeneity test (Alexandersson 1986), the Buishand range test (Buishand 1982), the Pettitt test (Pettitt 1979), and the Von Neumann ratio test (Von Neumann 1941). The temperature series are tested using the annual mean of the diurnal temperature range (maximum temperature–minimum temperature) and the annual mean of the absolute day-to-day differences of the diurnal temperature range. Precipitation series are tested using the annual wet day count (using a threshold of 1 mm). The use of derived annual variables avoids autocorrelation problems with testing daily series.

If the null hypothesis of no break in the series is rejected by none or one test, the series is classified as useful; if two tests reject the null hypothesis, the series is classified as doubtful; and if three or four tests reject the null hypothesis, the label suspect is given to the series. SACA&D does no homogeneity corrections; it only flags series. Trends shown via SACA&D and in this study are from series whose homogeneity has been classified as useful and doubtful(and which have sufficient data over the period when the trend is calculated).

The percentage of temperature and precipitation series flagged as suspect is 10.8% for precipitation series and 23.9% for temperature series. However, most series are too incomplete to reliably assess homogeneity. This is the case for nearly 73% of the precipitation series and 64% for the temperature series. This highlights the need for data rescue efforts in this region as much of the historic data are still only available in hard copy format, even from as recent as one or two decades ago (Williamson et al. 2015).

3. Quantifying the wet season onset

a. Definition of onset

There are scores of different definitions of the onset (and end) of the wet season. Here, we will use three indices to provide some measure of the variations associated with the choice of index. Popular are definitions relating the onset date to the first day of a wet spell with a given amount of accumulated precipitation, without being followed by a dry spell in the subsequent weeks. The thresholds used are empirically derived (Moron et al. 2009; Marteau et al. 2009). A recent study (Boyard-Micheau et al. 2013) generalizes such approaches to be applicable to diverse climatological regions by parameterizing the thresholds empirically to make them uniquely suitable to the local climate. An application of this index on the current dataset is briefly described in appendix B.

An example of this type of onset definition is used operationally at the Indonesian National Meteorological Service [Badan Meteorologi, Klimatologi dan Geofisika (BMKG)]. It defines the onset of the wet season as the first day, after 1 September, of three consecutive 10-day periods with cumulative precipitation ≥ 50 mm in each of the 10-day periods. The end of the wet season is defined in a similar way, but now in terms of three consecutive 10-day periods with accumulated precipitation < 50 mm. The data availability requirement is that in the 10-day periods no data are missing.

The second index lacks such empirical thresholds. It is based on work by Liebmann et al. (2007) and calculates
e1
where is the daily precipitation sum on day n and is an average daily precipitation sum; a simple average of the total amount of rainfall during a year divided by the number of days in a year. The beginning of the wet season is defined as the absolute minimum of A, indicating that the daily precipitation total from that date onward is larger than the average daily precipitation. Similarly, the end of the wet season is defined as the absolute maximum of A. In contrast to Liebmann et al. (2007), the present study uses the average daily precipitation sum as defined from the first day of the climatologically driest month over a 12-month period. The approach followed by Liebmann et al. (2007) is to use an average daily precipitation calculated over a climatological mean period. The 12-month period from the first day of the climatologically driest month is also the period used in summation (1). Calculating using data for the running year ensures that onset and end dates are found, also in excessively wet or dry years, which may not be the case when climatological values of are used. The drawback is that this makes the index a diagnostic (rather than a prognostic) metric. This index is calculated when 350 days of the year or more are nonmissing.

This approach has many similarities with the work of Camberlin and Diop (2003), who calculate a principal component analysis for daily rainfall over Senegal and use cumulative values for the principal component (PC1) time series (which also shows a distinct minimum and maximum) to determine the onset and retreat of the wet season.

A similar approach has been adopted by Smith et al. (2008) for Northern Australia, who simply relate the onset date to the date where 15% of the accumulated precipitation of the running year is reached, starting from the first day of the climatologically driest month. The end date is related to the 85% threshold. This index is calculated when 350 days of the year or more are nonmissing.

The use of data from the running year as reference, as done in this study for the index defined by Smith et al. (2008) or in the calculation of the daily averaged rainfall in the definition of Liebmann et al. (2007), will give different results compared to the use of climatological values. In particular, the interannual variability associated with the onset date will be reduced when using data from the running year, since each year is scaled to the interannual variability of the rainfall amount of the running year.

The three indices used are referred to as the BMKG index, the Liebmann index, and the Smith index for the sake of simplicity.

b. Comparison of onset definitions

Figure 3 shows the median and 75th percentile of daily precipitation sums for six stations in the Southeast Asian region, using data from the 1971–2000 period. The plots have been made by gathering data centered on a 5-day window. In these plots, mean onset dates for the three indices used are plotted, averaged over the same period.

Fig. 3.
Fig. 3.

The seasonal cycle of the median (red) and 75th percentile (green) of daily precipitation based on the 1971–2000 climatology. The vertical bars show the climatological mean yearday of the onset of the wet season, following the definitions of Liebmann et al. (2007, L), Smith et al. (2008, S), or the BMKG definition (B). Stations shown are (a) Dagupan; (b) Phu Lien; (c) Blang Bintang–Banda Aceh; (d) Jakarta; (e) the National Weather Service Office (WSO) in PohnPei, Micronesia; and (f) Darwin International Airport, Northwest Territory, Australia. For Banda Aceh, the climatological period 1981–2010 is used because there is too much missing data prior to 1980 for this station.

Citation: Journal of Climate 29, 7; 10.1175/JCLI-D-14-00574.1

The stations Phu Lien, Vietnam, and Blang Bintang–Banda Aceh, Indonesia (Figs. 3b,c), show a weak wet season in the first half of the year and a stronger one in the second half, with drier periods in between. All three indices relate to the start of the stronger wet season; a situation where an index sometimes relates to the weaker wet season is not observed. However, considerable differences in the means of these indices are present.

Based on the examples in Fig. 3, no index shows a systematic bias in relation to the others. The estimates of the onset date cluster for Darwin, Australia (Fig. 3f), and Dagupan, Philippines (Fig. 3a), but there are large differences between the indices for Phonpei, Micronesia (Fig. 3e), and Phu Lien (Fig. 3b) of approximately 50 days and to a lesser extent for Jakarta, Indonesia (Fig. 3d) (approximately 30 days).

For Jakarta, located on Java, which is the main rice production area in Indonesia, the BMKG and Smith indices gives values that are similar to those given by Naylor et al. (2007, their Fig. 1), who used an agronomical-based index and related the onset to a threshold of 200-mm accumulated precipitation, starting from 1 August. Using the same agronomical arguments, Schmidt and Van der Vecht (1952) place this threshold at 350 mm and argue that when this amount has fallen, the soil is generally sufficiently moistened to allow the farmers to prepare the seed beds for the wet season rice crop. This higher threshold shifts the onset date some 3 weeks later. The rich variety in onset definitions is understandable when even validation data for calculated onset dates can vary by such margins.

Figure 4 shows the onset and end dates for each year at Jakarta. Although the three indices are on average broadly similar, the interannual variability is very different, with the variability in the BMKG index the strongest and that in the Smith index the weakest. This can be understood by considering the onset dates using the rainfall of an ordinary year, , and the onset dates of another year, which has the same daily rainfall but with additional artificial noise, . The noise is randomly distributed and, averaged over the year, zero. When accumulating the rainfall using the two records, the dates during which 15% of the total annual accumulated rainfall is reached will not be far apart, since the accumulated noise, , will be much smaller than the accumulated rainfall, . However, the onset date for the Liebmann index relates to a point in time when the accumulated difference between actual daily and annually averaged precipitation has a minimum, meaning that the actual daily and annually averaged precipitation are very similar. In this situation, the accumulated noise will have a relatively large impact and will strongly affect the timing of the onset date. A similar argument applies to the BMKG index for which 10-day sums are used to determine the onset. These short-term periods are relatively sensitive to small variations in daily precipitation.

Fig. 4.
Fig. 4.

Onset and retreat dates of the wet season for Jakarta, using the (a) Liebmann et al. (2007), (b) BMKG, and (c) Smith et al. (2008) definitions. The bottoms and tops of the gray bars denote the onset and retreat of the wet season, respectively (in yearday). The length of the gray bar is therefore a measure of the length of the wet season. The red lines show low-pass-filtered onset and retreat dates using the Lowess smoother function with parameters f = 1/5 and iter = 3.

Citation: Journal of Climate 29, 7; 10.1175/JCLI-D-14-00574.1

Figure 5 gives maps of the correlations between the year-to-year changes in the three indices for the onset of the wet season. Correlations are calculated over the 1981–2010 period. The strongest relation is generally found between the Liebmann index and the Smith index. A distribution of the correlation values is distinctly bimodal, with the maximum numbers of stations correlating at 0.82 and 0.91. The distribution of correlations between the Liebmann index and the BMKG index is unimodal with a peak at 0.81, which is also the case for the distribution of correlations between the BMKG index and the Smith index at 0.80. Correlations between the BMKG-index and the other (nonparametric) indices are generally lower at the southern boundary of the domain, over Northern Australia.

Fig. 5.
Fig. 5.

Correlations between the yearday of the onset the wet season between the (a) Liebmann et al. (2007) and BMKG, (b) Liebmann et al. (2007) and Smith et al. (2008), and (c) BMKG and Smith et al. (2008) definitions.

Citation: Journal of Climate 29, 7; 10.1175/JCLI-D-14-00574.1

4. Results

a. Onset of the wet season

A description of the temporal development of the monsoon over the region has been given in earlier publications (Hamada et al. 2002; Aldrian and Susanto 2003; Moron et al. 2009). For the current dataset, Fig. 6 gives a map of the climatological yearday (based on the 1981–2010 period) of the onset of the wet season at each station as calculated by the three indices. Figure 6 shows that locally large differences exist between the estimates of the indices. It also shows that while the BMKG index appears noisier, there is less spatial homogeneity in the onset date than in the Liebmann index or the Smith index (although over central Sumatra and the Malay Peninsula these latter two also seem noisy). Interesting is the pronounced delay in the onset dates along the eastern seaboard of the Philippines and parts of the eastern seaboard of Vietnam compared to stations more inland, which relates to having different seasonal cycles between coastal and more inland stations in these areas and that rainfall for the coastal stations is mostly associated with trade winds. This delay is stronger in the Liebmann index than in the Smith or BMKG indices.

Fig. 6.
Fig. 6.

Climatological mean in onset dates of the wet season for Southeast Asia in yearday using data from the period 1981–2010. Onset dates defined by the (a) Liebmann et al. (2007), (b) BMKG, and (c) Smith et al. (2008) definitions.

Citation: Journal of Climate 29, 7; 10.1175/JCLI-D-14-00574.1

The onset dates show a progression from Vietnam and Thailand in the northwest toward Northern Australia in the south. For Indonesia, the progression of the onset of the rainy season matches earlier descriptions (Aldrian and Susanto 2003; Moron et al. 2009) with the onset the earliest in north Sumatra, progressing southward following the migration of the ITCZ.

Figure 6b shows a distinct lack of stations in Northern Australia and to a lesser extent in Thailand for the BMKG index. This relates to the use of fixed thresholds in the definition used by BMKG, which are adequate for the Indonesian situation but apparently fall short for other regions in Southeast Asia.

Figure 7 shows the standard deviation of the onset date of the wet season over Southeast Asia calculated over the 1981–2010 period. The vast differences between the standard deviation of the indices can be anticipated (section 4b). Even the most conservative index in this respect, the Smith index, shows areas where the standard deviation in the onset date is between 20 and 30 days. The Liebmann and Smith indices both show an increase in standard deviation for the Australian stations near 20°S. This must be related to the absence of a well-defined peak in the rainfall for that region (Smith et al. 2008). The Liebmann index shows the highest variability in the onset dates between 10°S and 10°N.

Fig. 7.
Fig. 7.

Standard deviation of the onset date of the wet season over Southeast Asia, calculated over 1981–2010. Onset dates defined by the (a) Liebmann et al. (2007), (b) BMKG, and (c) Smith et al. (2008) definitions.

Citation: Journal of Climate 29, 7; 10.1175/JCLI-D-14-00574.1

Much of the variability in the onset date will relate to the El Niño–Southern Oscillation (ENSO) phenomenon (Haylock and McBride 2001; Naylor et al. 2007; Moron et al. 2009). Figure 8 shows a composite of the deviation of the onset dates with respect to the 1981–2010 climatology for El Niño and La Niña years, using the Liebmann index focused on Java because of the high station density on that island. Table 2 shows the El Niño and La Niña years used in this composite, based on (updated) data from Suppiah (1993). The classification is made on the basis of the Southern Oscillation index and eastern Equatorial Pacific sea surface temperatures.

Fig. 8.
Fig. 8.

Composite of the onset of the wet season for (left) El Niño and (right) La Niña conditions, as a deviation from the 1981–2010 climatology. The Liebmann et al. (2007) index is used in this plot.

Citation: Journal of Climate 29, 7; 10.1175/JCLI-D-14-00574.1

Table 2.

El Niño and La Niña years during the period 1971–2010 used in the composite analysis of Fig. 8, based on (updated) data from Suppiah (1993).

Table 2.

Figure 8 shows a general delay in onset dates in El Niño years, with delays for some stations of up to 40 days. Interesting is that the onset date in this figure is not uniformly affected by El Niño conditions, with a large spread in the El Niño–related delay and many stations showing a modest delay of approximately 5–10 days. For the La Niña years the reverse is observed with a general advancement of the onset day. A similar picture is seen for the complete region.

b. Trends in the onset date of the wet season and other precipitation characteristics

Trends in the onset date of the wet season over the 1971–2012 period are shown in Fig. 9, where trends that have passed the 75% confidence level are shown in color. Trends have been calculated using the nonparametric Sen’s slope or median trend calculation (Sen 1968). The estimate of the trend is the median of the set of slopes , joining pairs of , where , of the time series where and denote time and value, respectively. This method is more resistant to outliers than the traditional least squares–based approach. Significance of the trend, accounting for the serial correlation, is estimated following Hamed and Rao (1998).

Fig. 9.
Fig. 9.

Trends in the onset date (days decade−1) of the wet season over Southeast Asia, calculated over 1971–2012. Gray circles denote stations with nonsignificant trends (at the 75% level). Onset dates defined by the (a) Liebmann et al. (2007), (b) BMKG, and (c) Smith et al. (2008) definitions.

Citation: Journal of Climate 29, 7; 10.1175/JCLI-D-14-00574.1

Stations with trends that fail to meet the significance level are shown in gray. Only a few stations show trends that are only weakly significant, and two out of three indices show, only over east Java, a concerted delay in the onset of the wet season of around 10 days decade−1.

The number of stations with statistically significant trends using the median trend calculations and the significance testing described above is lower than what would result from a trend analysis using ordinary least squares and a t test in which autocorrelation is taken into account, following Von Storch and Zwiers (1999). We relate this difference to the strong interannual variations in the onset dates. Both methods, however, show no wide-spread delay in onset date of the wet season but the least squares approach indicates that the delay in the onset date for east Java is shared by all three indices and for a few more stations (at the 95% significance level).

The hydroclimate over Java changes in more respects than only the delay in the onset of the wet season. Trends have been calculated over the 1971–2012 period for the total amount of rain received in the wet season [December–February (DJF)]. Figure 10a shows these trends, where the colored stations show results that are significant at the 75% level. This figure shows that only a small number of the stations reveal trends that can be separated from the internal variability. For the wet season, a common trend seems absent, with some stations showing an increase and others a decrease in total precipitation. For the dry season, no stations are found with a trend at the 75% significance level. Trends in the number of wet days, where a wet day is defined as having ≥1-mm precipitation (Fig. 10b), show a general positive pattern for the wet season. Again, the dry season shows a decrease in the number of wet days.

Fig. 10.
Fig. 10.

Trends in (left) sums of precipitation (mm decade−1) and (right) the number of wet days (precipitation ≥1 mm), calculated for the wet season (DJF) over the 1971–2012 period. Colored dots are significant at the 75% level. Note the change in scale in the trends of precipitation sums between the wet and the dry seasons.

Citation: Journal of Climate 29, 7; 10.1175/JCLI-D-14-00574.1

Earlier, Naylor et al. (2007) observed in climate scenario simulations a shift in the probability distribution of precipitation toward wetter conditions around the year 2050 when aggregating rainfall over the provinces of Java. In Fig. 11 their analysis is repeated using data from the 1971–90 and 1991–2010 periods in order to establish if such an anticipated shift in the probability distribution is already observed. Results for the wet season are shown, as well as or the April–June season when dry season planting typically occurs (Naylor et al. 2007). For each area, monthly precipitation totals for the relevant months within the analyzed periods are gathered for all stations that comply with two restrictions. One is that they should have data for 75% of the days within the period 1971–2010 and they should also be flagged as useful or doubtful regarding their homogeneity in the 1951–2011 period. The monthly precipitation totals are then aggregated over all stations within each province and a simple probability distribution is plotted. The bin size used is 1 cm.

Fig. 11.
Fig. 11.

Probability plots for monthly precipitation sums over the periods 1971–90 and 1991–2010 for the (left) DJF and (right) AMJ seasons. The blue lines give the fitted gamma distribution based on the months with nonzero precipitation. The plots relate to (a),(b) west Java and Banten, (c),(d) central Java and Yogyakarta, and (e),(f) east Java and Bali.

Citation: Journal of Climate 29, 7; 10.1175/JCLI-D-14-00574.1

For west and central Java, no clear shift in the probability distribution is observed. A gamma distribution (Wilks 1995) is fitted using all months with nonzero precipitation and this distribution is plotted along with the observed data. Although the plots are rather noisy, the distributions for the two periods are very similar, which is also reflected in the similarity of the fitted gamma distributions. This confirms the results of Fig. 10, where only a few stations show significant trends in precipitation totals.

For east Java, a shift toward drier conditions seems plausible where the median of the distribution has shifted to the drier end of the spectrum and the distribution has become less positively skewed. This relates to the observed reduction in large monthly rainfall totals in the more recent period, which is strongest in the wet season. In the April–June period an increase in the number of months with no precipitation and a decrease in the number of months with high amounts of rain are observed for the recent period. However, the amount of data available for the east Java/Bali area is low, with only nine stations that comply with the restrictions, casting doubt on the robustness of this result.

c. False starts of the wet season

The onset of the wet season is often difficult to establish in the field because of “false rain”; an isolated rainfall event preceding the expected onset date but followed by a dry spell. A dry period between the first day of a wet spell and the onset date may be hazardous to the farmer if the dry spell lasts too long and if no means of irrigation are accessible. Following a suggestion from Moron et al. (2009), a false start of the wet season is identified in this study by the first day of a 5-day sequence with accumulated precipitation of at least 40 mm preceding the onset date. The length of the consecutive dry days spell between the false start and the onset of the wet season is now analyzed. Here, a dry day is one with ≤1-mm accumulated precipitation. This threshold relates to the internationally agreed upon distinction between a wet and a dry day (Klein Tank et al. 2009). In this approach we take the calculated onset of the wet season, using the Liebmann index, to be a reference and we look for a wet spell preceding this date. The argument for using the Liebmann index is that this index is nonparametric and may be more applicable to diverse climatological regions than an agronomical index using fixed thresholds.

Note that in tropical areas, the use of a higher threshold than 1 mm in distinguishing between a dry and wet day may be more realistic. Values like 2.5–5 mm might be more suitable as that amount is easily evaporated and not stored in the soil. Using higher threshold values in this analysis would increase the probability of reporting a false start.

Using this definition, Fig. 12 gives the median values of the length of the consecutive dry days between the false start and the wet season onset calculated over the 1981–2010 period. The index used for the onset of the wet season is the Liebmann index. Figure 12 shows that for many stations in Thailand, Vietnam, and the Philippines the median value of the length of the dry spell is relatively low, although this value reaches a fortnight for some stations. In southern Sumatra, Java, Papua New Guinea, and Northern Australia, the median value of many stations increases to values of up to 30 days. Figure 12 indicates that the area in Southeast Asia just south of the equator down to Northern Australia has a higher risk of long periods of consecutive dry days between a false start and the actual onset of the wet season than do other areas in this region.

Fig. 12.
Fig. 12.

Median values for the period (in days) between the onset of the wet season and a potential false start Values are calculated over the 1981–2010 period. Right panel for Java only.

Citation: Journal of Climate 29, 7; 10.1175/JCLI-D-14-00574.1

The pattern in the length of the consecutive dry days between the false start and the onset of the wet season over Java (Fig. 12b) seems to reflect the topography, with generally small values north of the mountain range and generally larger values south of it.

While dry spells shorter than approximately 7 days can be considered small enough to be harmless (Nieuwolt 1989), a false start to the rainy season preceding the expected date by a very large number of days is not likely to fool the experienced farmer. Nevertheless, Fig. 12 confirms that the dry spell between a false start of the rainy season and the observed onset date is approximately of the same magnitude (perhaps somewhat smaller) than the observed standard deviation in the onset dates (Fig. 7). This emphasizes the need for high quality seasonal forecasts for guidance with agricultural planning.

No coherent signal in statistically significant trends in the false start has been detected (not shown).

A problem with the approach used to quantify false starts is that it mixes two different definitions of onset: one based on 5-day precipitation sums and one on an analysis of the accumulated anomalous daily precipitation (the Liebmann index). Perhaps a more consistent approach would be to define a false start of the wet season by calculating the period between the first day of a 5-day period receiving at least 40 mm of rain, and the first day of a 5-day sequence with at least this amount of rain and followed by a period of 30 days without a dry spell of 7 days or longer. The 7-day period is motivated by the work of Nieuwolt (1989), who argues that a dry spell in tropical areas lasting longer than a week would already have serious consequences for crops.

An analysis of the climatology of false starts using this alternative definition is presented in appendix C. The maps for the median value of the false start for the analyzed stations are noisier compared to the maps that are shown later in this section. We relate this to the “binary” character of the alternative false start definition, where fixed thresholds are used in determining if a false start is observed or not. The Liebmann index, because of its integrative character, is more robust to small variations in precipitation as discussed in section 3b.

A third possible way to define false starts of the wet season is proposed by Camberlin and Diop (2003), who analyzed false starts using an index similar to the Liebmann index, but defining the false start as a local minimum in the cumulative anomalous precipitation curve prior to the absolute minimum in this curve.

d. Changes in nighttime temperatures

Next to changes in the hydroclimate, changes in the nighttime temperatures are also relevant for food security in Southeast Asia (Peng et al. 2004). This section is focused on that aspect.

In tropical regions, annual mean maximum and minimum temperatures have increased since the mid-1950s (Choi et al. 2009). Figure 13 shows trend maps for DJF- and JJA-averaged minimum temperatures, calculated over the period 1971–2012. It shows that many stations see a statistically significant increase in the minimum temperature for both the DJF and JJA seasons, indicating that the nights are warming. This is confirmed when trends are calculated for the extremes in climate, for instance the temperature of the warmest nights (TNx) and the number of warm nights (TN90p) (not shown). The converse is true for the number of cold nights (TN10p), which decreases over the region, and the lowest minimum temperature (TNn), which increases across the region (not shown). Northern Australia in JJA is the exception; for some stations a decrease in minimum temperatures, and its extremes, is observed.

Fig. 13.
Fig. 13.

Trends in daily minimum temperature (°C decade−1), calculated over the 1971–2012 period: (left) DJF and (right) JJA. Colored dots are significant at the 95% level.

Citation: Journal of Climate 29, 7; 10.1175/JCLI-D-14-00574.1

For three of the stations shown on the map, the possibility exists that the trends in nighttime temperatures are affected by the growth of the urban area in the vicinity of the meteorological stations. These stations are Lang/Hanoi (ID = 305) and Vinh (ID = 307) in Vietnam and Tacloban (ID = 521) in the Philippines. The other stations on the map in Fig. 13 are located either in the countryside, at airstrips, or near small settlements or small townships. A Google Maps satellite view of these stations can be obtained via the SACA&D website (http://sacad.database.bmkg.go.id/utils/stationdetail.php?stationid=N, with N the station id).

Note that there is a difference in trends between the number of warm nights, based on the daily minimum temperature, and the number of warm daytimes, which is based on the daily maximum temperature. For many stations, the change in the increase of the number of warm nights is stronger than in the number of warm daytimes. For some stations, this increase can be partly related to the rapid expansion of urban area around the stations, like in Jakarta (Siswanto et al. 2016). Note that Jakarta is not included in Fig. 13; the homogeneity of this station is classified as suspect and therefore it is left out of the trend analysis.

Figure 14 shows probability plots for three stations in the region for which the homogeneity of the temperature data is flagged as useful: Cilacap, Indonesia; Tacloban; and Suphan Buri, Thailand. These stations show a strong increase in their mean values of the daily minimum temperature when comparing data from 1971–90 with the 1991–2010 period. Figure 14 shows that the increase in the mean relates to a shift in the probability to higher values (e.g., Cilacap in the DJF season), but a change toward a more skewed probability distribution is observed as well (e.g., Cilacap in JJA).

Fig. 14.
Fig. 14.

Probability plots for daily minimum temperatures over the periods 1971–90 and 1991–2010 for the (left) DJF and (right) JJA seasons. The plots relate to (a),(b) Cilacap, (c),(d) Tacloban, and (e),(f) Suphan Buri.

Citation: Journal of Climate 29, 7; 10.1175/JCLI-D-14-00574.1

5. Summary and conclusions

Using a newly developed dataset with station-based daily data for the Southeast Asian region, climatology and trends in climatic indices are analyzed that bear some relevance to food security. With rice the staple food of the diet in the region, these indices are specifically targeted at this food source. The amount of precipitation and variations in the length (and start) of the wet season strongly influence agricultural production (Naylor et al. 2007) and three indices for the onset of the wet season are compared.

These three indices are that in operational use at the Indonesian Agency for Meteorology, Climatology and Geophysics (BMKG), the index proposed by Liebmann et al. (2007), and the index used earlier in Northern Australia by Smith et al. (2008). The climatologies of the three indices are not very similar, with differences that reach up to 50 days. While the temporal correlation between the indices is generally very high, the quantitative estimates of the year-to-year onset of the wet season vary between the indices, with the Smith et al. (2008) index having the lowest interannual variability of the three.

Trends in the onset date of the wet season calculated over 1971–2012 are only statistically significant at the 75% level for a few stations. Despite signaling delays in the onset of the rainy season for future climate scenarios (Zhang et al. 2013), no wide-spread or coherent evidence is found that this has already taken place. A possible exception may be Java, where stations in west Java show a delay of approximately 10 days decade−1 in the onset date in two of the three indices. Trends are only significant at the 75% level and are not wide spread.

For Java, only a few stations show trends in the total amount of precipitation or in the number of wet days, which can be separated from the internal variability. For the wet season (DJF), a common trend in the amount of precipitation seems absent but a concerted trend toward more rainy days seems present.

A common problem for farmers in Southeast Asia involves false starts of the rainy season, wherein an isolated rainfall event precedes the expected onset date but is followed by a dry spell. Using a definition suggested by Moron et al. (2009), the climatology of the length of the dry spell over the 1981–2010 period is given. It is observed that the median length of this dry spell is, for many areas, similar or smaller than the standard deviation on the onset date of the wet season. This illustrates the dilemma for the farmer who has no access to reliable forecasts of the onset of the wet season in judging when to plant seedlings and the relevance of this issue to agriculture. The area in Southeast Asia just south of the equator up to Northern Australia has a higher risk of long periods of consecutive dry days between a false start and the actual onset of the wet season than do other areas in this region.

No coherent signal is observed in trends of the length of the dry spells associated with the false starts of the wet season.

High nighttime temperature has an adverse effect on rice crop yields (Peng et al. 2004). However, some areas, for example the north of Lao People's Democratic Republic, may benefit from higher winter temperatures since seedling nursery in dry season rice in can fail because of low temperatures. A positive trend in the average daily minimum temperature, representative of the nighttime temperature, over the region is observed with values of up to 0.8°C decade−1. For a selection of stations the change in the distribution of nighttime temperatures is analyzed when comparing the 1961–90 period with the 1991–2010 period. All stations show a shift of the median to higher values and some stations also show a distribution that is more negatively skewed in the most recent period than in the earlier period. This relates to a decline in the number of relatively cool nights, which is stronger than the increase in the number of relatively warm nights.

In conclusion, the analysis of intraseasonal and interannual variations in rainfall and changes in the distribution of nighttime temperatures makes clear that more detailed and reliable insights into the effects of climate variability and climate change can only be obtained when a sufficient number of observations are available. Further efforts to make daily data from rain gauges and weather or climatological stations available for research are necessary, as is the inclusion of newly digitized data, in order to provide a historical framework. This requires a concerted effort by the countries in the region.

Acknowledgments

We thank all the data contributors to SACA&D. Geert Jan van Oldenborgh and Cees Stigter are thanked for stimulating discussions. Brant Liebmann, Joseph Boyard-Micheau, and two anonymous reviewers are thanked for their constructive remarks, especially the reviewer directing our attention to alternative definitions of the false start in the wet season. Digitization and construction of SACA&D are partly funded by the Digitisasi Data Historis (DiDaH) project, a joint project between the national meteorological services of Indonesia (BMKG) and the Netherlands (KNMI). The research leading to these results has received funding from the Didah project and the European Union, Seventh Framework Programme (FP7/2007-2013 and SPA.2013.1.1-02), under Grant Agreements 242093 (EURO4M) and 607193 (UERRA). GvdS acknowledges the support of the Royal Netherlands Embassy in Jakarta, Indonesia, through a Joint Cooperation Programme between Dutch and Indonesian research institutes. Data used in this study can be accessed online (http://sacad.database.bmkg.go.id).

APPENDIX A

Details on the Blending of Station Records

Blending is a step in the processing of the daily station data where data from nearby stations are used to fill gaps or extent series in order to have series that are as continuous and as long as possible. The requirements for blending are that the receiving and the donating stations are not more than 25 km apart and have no more than a 50-m-elevation difference.

Figure A1 shows the distribution of the percentage of the precipitation time series with data filled in from the blending procedure against the number of stations. This figure shows that only 14.4% of the precipitation stations have not 1 day of blended data. The distribution also shows a maximum near 64%. The observation that many records have more than half their data filled in from nearby stations points to the fact that many records in SACA&D are actually fairly short. These short records are blended with each other, which is only possible in areas with a sufficiently high station density, like Java and Sumatra. Obviously, in this case the blended series cannot be seen as independent from each other since they share a large fraction of their data.

Fig. A1.
Fig. A1.

The fraction of stations vs the interstation distance in the blending.

Citation: Journal of Climate 29, 7; 10.1175/JCLI-D-14-00574.1

Figure A2 shows the distribution of the interstation distance for all cases when blending is applied. The distribution is fairly flat, with a median value just below 10 km.

Fig. A2.
Fig. A2.

The fraction of stations vs the interstation distance in the blending.

Citation: Journal of Climate 29, 7; 10.1175/JCLI-D-14-00574.1

APPENDIX B

Results Derived from the Boyard–Micheau Approach

A novel way to calculate onset (and cessation) of the rainy season is provided by Boyard-Micheau et al. (2013). They employ an agronomical definition where the onset is determined as the first day of N consecutive days receiving at least P millimeters without a dry spell lasting n days and receiving less than p millimeters in the following C days.

This method is applied to the station set used in this study, provided the station has sufficient data, and results are given for reference. Similar to Boyard-Micheau et al. (2013), the values of n and p are fixed at 10 days and 5 mm, respectively. Following Boyard-Micheau et al. (2013), the onset day is calculated for experiments where N (2, 3, 4, 5), where P (10, 15, 20, 25, 30, 40, 50) and C (20, 30). The combination of N, P, and C yields 56 experiments. The onset day is calculated for stations that have sufficient data in the 40-yr 1961–2000 period. Finally, results are presented for the long rains (where applicable) and what Boyard-Micheau et al. (2013) call the multicombination PC, which is simply an average of all 56 experiments.

There are 1672 stations with sufficient data for this analysis in the domain. The undefined onset dates were replaced by an average of the nonmissing onset dates computed over the 40 yr for the combination of N, P, and C at the station. This approach of replacing a missing onset date is an alternative to the possible approaches given by Boyard-Micheau et al. (2013).

The first PC explains 9.91% of the total variance over the domain, which is considerably less than the percentages obtained by Boyard-Micheau et al. (2013) over the eastern Africa area.

It is interesting to see which local-scale thresholds maximize the correlation with the regional-scale onset. The first multicombination PC is correlated, for each station, with all 56 combinations of N, P, and C. The combination with the highest correlation (shown in Fig. B1) is used for visualization in the maps for the “optimal” choice for N, P, and C in Figs. B2a–c.

Fig. B1.
Fig. B1.

Maximum correlations between the first multicombination PC and the onset date based on a combination of N, P, and C for each station.

Citation: Journal of Climate 29, 7; 10.1175/JCLI-D-14-00574.1

Fig. B2.
Fig. B2.

Maps of (top) N, (middle) P, and (bottom) C for which the correlation with the first multicombination PC is highest.

Citation: Journal of Climate 29, 7; 10.1175/JCLI-D-14-00574.1

Figure B1 shows stations with relatively high correlations in northeastern Australia, Farther west and north, the correlations decrease, with the lowest values over Java and Sumatra.

Similar to Boyard-Micheau et al. (2013), considerable variations for N, P, and C are seen in small areas, and [again similar to the findings of Boyard-Micheau et al. (2013)] the largest interannual variation is determined by P. This indicates that that the amount of precipitation received by a station dominates the onset date.

APPENDIX C

An Alternative Definition for the False Start of the Wet Season

Figure C1 gives the median values of the period between the first day of a 5-day period with accumulated rainfall of ≥40 mm and the first day of a 5-day period with accumulated rainfall of ≥40 mm and followed by a 30-day period without a dry spell lasting 7 days or longer. Data are calculated over the 1981–2010 period, requiring at least 20 yr with sufficient data to calculate a value for the false start. Figure C1 shows a rather noisy pattern. For many stations in Southeast Asia the median value of the period between the false start and the actual onset of the wet season reaches 10–15 days. The median value increases to values of up to 30 days for some stations.

Fig. C1.
Fig. C1.

Median values for the period (in days) between the first day of a 5-day period receiving 40 mm or more and the first day of a 5-day period receiving 40 mm or more and having no dry spell of 7 days or longer in the next 30 days. Values are calculated over the 1981–2010 period. Right panel for Java only.

Citation: Journal of Climate 29, 7; 10.1175/JCLI-D-14-00574.1

The pattern in the length of the consecutive dry days between the false start and the onset of the wet season over Java (Fig. 12b) shows large variability, but shows coherent areas with generally higher or lower values. Averaged over Java, the length of the period between the false start and the actual onset of the wet season is between 9 and 10 days.

REFERENCES

  • Aldrian, E., and R. D. Susanto, 2003: Identification of the three dominant rainfall regions within Indonesia and their relation to sea surface temperature. Int. J. Climatol., 23, 14351452, doi:10.1002/joc.950.

    • Search Google Scholar
    • Export Citation
  • Alexandersson, H., 1986: A homogeneity test applied to precipitation data. J. Climatol., 6, 661675, doi:10.1002/joc.3370060607.

  • Boyard-Micheau, J., P. Camberlin, N. Philippon, and V. Moron, 2013: Regional scale rainy season onset detection: A new approach based on multivariate analysis. J. Climate, 26, 89168928, doi:10.1175/JCLI-D-12-00730.1.

    • Search Google Scholar
    • Export Citation
  • Buishand, T. A., 1982: Some methods for testing the homogeneity of rainfall records. J. Hydrol., 58, 1127, doi:10.1016/0022-1694(82)90066-X.

    • Search Google Scholar
    • Export Citation
  • Camberlin, P., and M. Diop, 2003: Application of daily rainfall principle component analysis to the assessment of the rainy season characteristics in Senegal. Climate Res., 23, 159169, doi:10.3354/cr023159.

    • Search Google Scholar
    • Export Citation
  • Choi, G., and Coauthors, 2009: Changes in means and extreme events of temperature and precipitation in the Asia-Pacific Network region, 1955–2007. Int. J. Climatol., 29, 19061925, doi:10.1002/joc.1979.

    • Search Google Scholar
    • Export Citation
  • ECA&D Project Team, 2012: European Climate Assessment & Dataset Algorithm Theoretical Basis Document (ATBD), version 10.5. ECA&D, De Bilt, Netherlands, 45 pp.

  • Hamada, J.-J., M. D. Yamanaka, J. Matsumoto, S. Fukao, P. A. Winarso, and T. Sribimawati, 2002: Spatial and temporal variations of the rainy season over Indonesia and their link to ENSO. J. Meteor. Soc. Japan, 80, 285310, doi:10.2151/jmsj.80.285.

    • Search Google Scholar
    • Export Citation
  • Hamed, K. H., and A. R. Rao, 1998: A modified Mann–Kendall trend test for autocorrelated data. J. Hydrol., 204, 182196, doi:10.1016/S0022-1694(97)00125-X.

    • Search Google Scholar
    • Export Citation
  • Hartmann, D., and Coauthors, 2013: Observations: Atmosphere and surface. Climate Change 2013: The Physical Science Basis, T. Stocker et al., Eds., Cambridge University Press, 159–255.

  • Haylock, M. R., and J. McBride, 2001: Spatial coherence and predictability of Indonesia wet season rainfall. J. Climate, 14, 38823887, doi:10.1175/1520-0442(2001)014<3882:SCAPOI>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Held, I. M., and B. J. Soden, 2006: Robust responses of the hydrological cycle to global warming. J. Climate, 19, 56865699, doi:10.1175/JCLI3990.1.

    • Search Google Scholar
    • Export Citation
  • Klein Tank, A. M. G., and Coauthors, 2002: Daily dataset of 20th-century surface air temperature and precipitation series for the European Climate Assessment. Int. J. Climatol., 22, 14411453, doi:10.1002/joc.773.

    • Search Google Scholar
    • Export Citation
  • Klein Tank, A. M. G., F. Zwiers, and X. Zhang, 2009: Guidelines on analysis of extremes in a changing climate in support of informed decisions for adaptation. Climate Data and Monitoring WCDMP-No. 72, WMO/TD-1500, 52 pp. [Available online at http://www.wmo.int/pages/prog/wcp/wcdmp/wcdmp_series/documents/WCDMP_72_TD_1500_en_1.pdf.]

  • Liebmann, B., S. J. Camargo, A. Seth, J. A. Marengo, L. M. V. Carvalho, D. Allured, C. S. Fu, and R. Vera, 2007: Onset and end of the rainy season in South America in observations and the ECHAM 4.5 atmospheric general circulation model. J. Climate, 20, 20372050, doi:10.1175/JCLI4122.1.

    • Search Google Scholar
    • Export Citation
  • Lintner, B. R., M. Biasutti, N. S. Diffenbaugh, J.-E. Lee, M. J. Niznik, and K. L. Findell, 2012: Amplification of wet and dry month occurrence over tropical land regions in response to global warming. J. Geophys. Res., 117, D11106, doi:10.1029/2012JD017499.

    • Search Google Scholar
    • Export Citation
  • Marteau, R., V. Moron, and N. Philippon, 2009: Spatial coherence of monsoon onset over western and central Sahel (1950–2000). J. Climate, 22, 13131324, doi:10.1175/2008JCLI2383.1.

    • Search Google Scholar
    • Export Citation
  • Moron, V., A. W. Robertson, and R. Boer, 2009: Spatial coherence and seasonal predictability of monsoon onset over Indonesia. J. Climate, 22, 840850, doi:10.1175/2008JCLI2435.1.

    • Search Google Scholar
    • Export Citation
  • Naylor, R. L., D. S. Battisti, D. J. Vimont, W. P. Falcon, and M. B. Burke, 2007: Assessing risks of climate variability and climate change for Indonesian rice agriculture. Proc. Natl. Acad. Sci. USA, 104, 77527757, doi:10.1073/pnas.0701825104.

    • Search Google Scholar
    • Export Citation
  • Nguyen-Le, D., J. Matsumoto, and T. Ngo-Duc, 2015: Onset of the rainy season in the eastern Indochina Peninsula. J. Climate, 28, 56455666, doi:10.1175/JCLI-D-14-00373.1.

    • Search Google Scholar
    • Export Citation
  • Nieuwolt, S., 1989: Estimating the agricultural risks of tropical rainfall. Agric. For. Meteor., 45, 251263, doi:10.1016/0168-1923(89)90047-6.

    • Search Google Scholar
    • Export Citation
  • Peng, S., and Coauthors, 2004: Rice yield decline with higher night temperature from global warming. Proc. Natl. Acad. Sci. USA, 101, 99719975, doi:10.1073/pnas.0403720101.

    • Search Google Scholar
    • Export Citation
  • Pettitt, A. N., 1979: A non-parametric approach to the change-point problem. Appl. Stat., 28, 126135, doi:10.2307/2346729.

  • Polade, S., D. Pierce, D. Cayan, A. Gershunov, and M. Dettinger, 2014: The key role of dry days in changing regional climate and precipitation regimes. Sci. Rep., 4, 4364, doi:10.1038/srep04364.

    • Search Google Scholar
    • Export Citation
  • Purseglove, J. W., Ed., 1988: Tropical Crops—Monocotyledons. Longman Scientific and Technical, 607 pp.

  • Schmidt, F. H., and J. Van der Vecht, 1952: East Monsoon fluctuations in Java and Madura during the period 1880-1940. Kementerian Perhubungan Verhandelingen No. 43, Djawatan Meteorologi dan Geofisik, Djakarta, Indonesia, 45 pp.

  • Seager, R., N. Naik, and L. Vogel, 2012: Does global warming cause intensified interannual hydroclimatic variability? J. Climate, 25, 33553372, doi:10.1175/JCLI-D-11-00363.1.

    • Search Google Scholar
    • Export Citation
  • Sen, P. K., 1968: Estimates of the regression coefficient based on Kendall’s tau. J. Amer. Stat. Assoc., 39, 13791389, doi:10.2307/2285891.

    • Search Google Scholar
    • Export Citation
  • Siswanto, S., R. Jilderda, G. Van der Schrier, G. J. Van Oldenborgh, and B. Van den Hurk, 2016: Trend of temperature, extreme precipitation, and diurnal rainfall changes in the urbanized Jakarta city during the past 130 years. Int. J. Climatol., doi:10.1002/joc.4548, in press.

    • Search Google Scholar
    • Export Citation
  • Smith, I. N., L. Wilson, and R. Suppiah, 2008: Characteristics of the Northern Australian rainy season. J. Climate, 21, 42984311, doi:10.1175/2008JCLI2109.1.

    • Search Google Scholar
    • Export Citation
  • Suppiah, R., 1993: ENSO phenomenon and 30-50 day variability in the Australian summer monsoon rainfall. Int. J. Climatol., 13, 837851, doi:10.1002/joc.3370130803.

    • Search Google Scholar
    • Export Citation
  • Van den Besselaar, E. J. M., and Coauthors, 2015: International Climate Assessment & Dataset (ICA&D): Climate services across borders. Bull. Amer. Meteor. Soc., 96, 1621, doi:10.1175/BAMS-D-13-00249.1.

    • Search Google Scholar
    • Export Citation
  • Vecchi, G. A., and B. J. Soden, 2007: Global warming and the weakening of the tropical circulation. J. Climate, 20, 43164340, doi:10.1175/JCLI4258.1.

    • Search Google Scholar
    • Export Citation
  • Von Neumann, J., 1941: Distribution of the ratio of the mean square successive difference to the variance. Ann. Math. Stat., 12, 367395, doi:10.1214/aoms/1177731677.

    • Search Google Scholar
    • Export Citation
  • Von Storch, H., and F. W. Zwiers, 1999: Statistical Analysis in Climate Research. Cambridge University Press, 484 pp.

  • Wheeler, T., and J. Von Braun, 2013: Climate change impacts on global food security. Science, 341, 508513, doi:10.1126/science.1239402.

    • Search Google Scholar
    • Export Citation
  • Wijngaard, J. B., A. M. G. Klein Tank, and G. P. Können, 2003: Homogeneity of 20th century European daily temperature and precipitation series. Int. J. Climatol., 23, 679692, doi:10.1002/joc.906.

    • Search Google Scholar
    • Export Citation
  • Wilks, D. S., 1995: Statistical Methods in the Atmospheric Sciences: An Introduction. Academic Press, 467 pp.

  • Williamson, F., R. Allan, A. D. Switzer, J. C. L. Chan, R. J. Wasson, R. D’Arrigo, and R. Gartner, 2015: New directions in hydro-climatic histories: Observational data recovery, proxy records and the Atmospheric Circulation Reconstructions over the Earth (ACRE) initiative in Southeast Asia. Geosci. Lett., 2, doi:10.1186/s40562-015-0018-z.

    • Search Google Scholar
    • Export Citation
  • Zhang, H., A. Moise, P. Liang, and L. Hanson, 2013: The response of summer monsoon onset/retreat in Sumatra-Java and tropical Australia region to global warming in CMPI3 models. Climate Dyn., 40, 377399, doi:10.1007/s00382-012-1389-x.

    • Search Google Scholar
    • Export Citation
Save