• Beck, J., F. Bouttier, L. Wiegand, C. Gebhardt, C. Eagle, and N. Roberts, 2016: Development and verification of two convection-allowing multi-model ensembles over western Europe. Quart. J. Roy. Meteor. Soc., 142, 28082826, https://doi.org/10.1002/qj.2870.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Bednarczyk, C. N., and B. C. Ancell, 2015: Ensemble sensitivity analysis applied to a southern plains convective event. Mon. Wea. Rev., 143, 230249, https://doi.org/10.1175/MWR-D-13-00321.1.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Birch, C. E., J. H. Marsham, D. J. Parker, and C. M. Taylor, 2014a: The scale dependence and structure of convergence fields preceding the initiation of deep convection. Geophys. Res. Lett., 41, 47694776, https://doi.org/10.1002/2014GL060493.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Birch, C. E., D. J. Parker, J. H. Marsham, D. Copsey, and L. Garcia-Carreras, 2014b: A seamless assessment of the role of convection in the water cycle of the West African monsoon. J. Geophys. Res. Atmos., 119, 28902912, https://doi.org/10.1002/2013JD020887.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Birch, C. E., M. J. Roberts, L. Garcia-Carreras, D. Ackerley, M. J. Reeder, A. P. Lock, and R. Schiemann, 2015: Sea-breeze dynamics and convection initiation: The influence of convective parameterization in weather and climate model biases. J. Climate, 28, 80938108, https://doi.org/10.1175/JCLI-D-14-00850.1.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Birch, C. E., S. Webster, S. C. Peatman, D. J. Parker, A. J. Matthews, Y. Li, and M. E. E. Hassim, 2016: Scale interactions between the MJO and the western Maritime Continent. J. Climate, 29, 24712492, https://doi.org/10.1175/JCLI-D-15-0557.1.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Bousquet, O., D. Barbary, S. Bielli, S. Kebir, L. Raynaud, S. Malardel, and G. Faure, 2020: An evaluation of tropical cyclone forecast in the southwest Indian Ocean basin with AROME-Indian Ocean convection-permitting numerical weather predicting system. Atmos. Sci. Lett., 21, e950, https://doi.org/10.1002/asl.950.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Bowler, N., A. Arribas, K. Mylne, K. Robertson, and S. Beare, 2008: The MOGREPS short-range ensemble prediction system. Quart. J. Roy. Meteor. Soc., 134, 703722, https://doi.org/10.1002/qj.234.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Bush, M., and Coauthors, 2020: The first Met Office Unified model–JULES regional atmosphere and land configuration, RAL1. Geosci. Model Dev., 13, 19992029, https://doi.org/10.5194/gmd-13-1999-2020.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Cafaro, C., T. H. A. Frame, J. Methven, N. Roberts, and J. Bröcker, 2019: The added value of convection-permitting ensemble forecasts of sea breeze compared to a Bayesian forecast driven by the global ensemble. Quart. J. Roy. Meteor. Soc., 145, 17801798, https://doi.org/10.1002/qj.3531.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Cafaro, C., and Coauthors, 2021: Do convection-permitting ensembles lead to more skillful short-range probabilistic rainfall forecasts over tropical East Africa? Wea. Forecasting, 36, 697716, https://doi.org/10.1175/WAF-D-20-0172.1.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Chang, C. P., P. A. Harr, and H. J. Chen, 2005: Synoptic disturbances over the equatorial South China Sea and western Maritime Continent during boreal winter. Mon. Wea. Rev., 133, 489503, https://doi.org/10.1175/MWR-2868.1.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Clark, P., N. Roberts, H. Lean, S. P. Ballard, and C. Charlton-Perez, 2016: Convection-permitting models: A step-change in rainfall forecasting. Meteor. Appl., 23, 165181, https://doi.org/10.1002/met.1538.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Dey, S. R. A., G. Leoncini, N. M. Roberts, R. S. Plant, and S. Migliorini, 2014: A spatial view of ensemble spread in convection permitting ensembles. Mon. Wea. Rev., 142, 40914107, https://doi.org/10.1175/MWR-D-14-00172.1.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Dipankar, A., and Coauthors, 2020: SINGV: A convective-scale weather forecast model for Singapore. Quart. J. Roy Meteor. Soc., 146, 41314146, https://doi.org/10.1002/qj.3895.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Ferrett, S., G. Yang, S. J. Woolnough, J. Methven, K. Hodges, and C. E. Holloway, 2020: Linking extreme precipitation in Southeast Asia to equatorial waves. Quart. J. Roy. Meteor. Soc., 146, 665684, https://doi.org/10.1002/qj.3699.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Gebhardt, C., S. E. Theis, M. Paulat, and Z. Ben Bouallègue, 2011: Uncertainties in COSMO-DE precipitation forecasts introduced by model perturbations and variation of lateral boundaries. Atmos. Res., 100, 168177, https://doi.org/10.1016/j.atmosres.2010.12.008.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Golding, B. W., and Coauthors, 2014: Forecasting capabilities for the London 2012 Olympics. Bull. Amer. Meteor. Soc., 95, 883896, https://doi.org/10.1175/BAMS-D-13-00102.1.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Hagelin, S., J. Son, R. Swinbank, A. McCabe, N. Roberts, and W. Tennant, 2017: The Met Office convective-scale ensemble, MOGREPS-UK. Quart. J. Roy. Meteor. Soc., 143, 28462861, https://doi.org/10.1002/qj.3135.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Hamada, J.-I., S. Mori, H. Kubota, M. D. Yamanaka, U. Haryoko, S. Lestari, R. Sulistyowati, and F. Syamsudin, 2012: Interannual rainfall variability over northwestern Jawa and its relation to the Indian Ocean dipole and El Niño–Southern Oscillation events. SOLA, 8, 6972, https://doi.org/10.2151/sola.2012-018.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Hanley, K. E., D. J. Kirshbaum, N. M. Roberts, and G. Leoncini, 2013: Sensitivities of a squall line over central Europe in a convective-scale ensemble. Mon. Wea. Rev., 141, 112133, https://doi.org/10.1175/MWR-D-12-00013.1.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Heng, B. C. P., and Coauthors, 2020: SINGV-DA: A data assimilation system for convective-scale numerical weather prediction over Singapore. Quart. J. Roy. Meteor. Soc., 146, 19231938, https://doi.org/10.1002/qj.3774.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Hohenegger, C., and C. Schar, 2007: Atmospheric predictability at synoptic versus cloud-resolving scales. Bull. Amer. Meteor. Soc., 88, 17831794, https://doi.org/10.1175/BAMS-88-11-1783.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Holloway, C. E., S. J. Woolnough, and G. M. S. Lister, 2013: The effects of explicit versus parameterized convection on the MJO in a large-domain high-resolution tropical case study. Part I: Characterization of large-scale organization and propagation. J. Atmos. Sci., 70, 13421369, https://doi.org/10.1175/JAS-D-12-0227.1.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Huffman, G. J., and Coauthors, 2019: NASA Global Precipitation Measurement (GPM) Integrated Multi-satellitE Retrievals for GPM (IMERG). Algorithm Theoretical Basis Doc., version 06, 38 pp., https://gpm.nasa.gov/sites/default/files/document_files/IMERG_ATBD_V06.pdf.

  • Johnson, S. J., and Coauthors, 2016: The resolution sensitivity of the South Asian monsoon and Indo-Pacific in a global 0.35° AGCM. Climate Dyn., 46, 807831, https://doi.org/10.1007/s00382-015-2614-1.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Kain, J. S., and Coauthors, 2008: Some practical considerations regarding horizontal resolution in the first generation of operational convection-allowing NWP. Wea. Forecasting, 23, 931952, https://doi.org/10.1175/WAF2007106.1.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Kendon, E. J., N. M. Roberts, C. A. Senior, and M. J. Roberts, 2012: Realism of rainfall in a very high-resolution regional climate model. J. Climate, 25, 57915806, https://doi.org/10.1175/JCLI-D-11-00562.1.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Khan, S., and V. Maggioni, 2019: Assessment of level-3 gridded Global Precipitation Mission (GPM) products over oceans. Remote Sens., 11, 255, https://doi.org/10.3390/rs11030255.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Lestari, S., A. King, C. Vincent, D. Karoly, and A. Protat, 2019: Seasonal dependence of rainfall extremes in and around Jakarta, Indonesia. Wea. Climate Extremes, 24, 100202, https://doi.org/10.1016/j.wace.2019.100202.

    • Search Google Scholar
    • Export Citation
  • Lim, S. Y., C. Marzin, P. Xavier, C. P. Chang, B. Timbal, S. Yee Lim, C. Marzin, and P. Xavier, 2017: Impacts of boreal winter monsoon cold surges and the interaction with MJO on Southeast Asia rainfall. J. Climate, 30, 42674281, https://doi.org/10.1175/JCLI-D-16-0546.1.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Lin, J. L., and Coauthors, 2006: Tropical intraseasonal variability in 14 IPCC AR4 climate models. Part I: Convective signals. J. Climate, 19, 26652690, https://doi.org/10.1175/JCLI3735.1.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Loken, E. D., A. J. Clark, M. Xue, and F. Kong, 2017: Comparison of next-day probabilistic severe weather forecasts from coarse- and fine-resolution CAMs and a convection-allowing ensemble. Wea. Forecasting, 32, 14031421, https://doi.org/10.1175/WAF-D-16-0200.1.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Love, B. S., A. J. Matthews, and G. M. S. Lister, 2011: The diurnal cycle of precipitation over the Maritime Continent in a high-resolution atmospheric model. Quart. J. Roy. Meteor. Soc., 137, 934947, https://doi.org/10.1002/qj.809.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Matsumoto, J., and Coauthors, 2017: An overview of the Asian Monsoon Years 2007–2012 (AMY) and multi-scale interactions in the extreme rainfall events over the Indonesian Maritime Continent. The Global Monsoon System: Research and Forecast, C.-P. Chang et al., Eds., World Scientific Series on Asia-Pacific Weather and Climate, Vol. 9, World Scientific, 365–385.

    • Crossref
    • Export Citation
  • Met Office, 2018: Iris: A Python library for analysing and visualising meteorological and oceanographic data sets, v.2.0. Accessed 5 October 2018, https://scitools.org.uk/.

  • Mittermaier, M., N. Roberts, and S. A. Thompson, 2013: A long-term assessment of precipitation forecast skill using the fractions skill score. Meteor. Appl., 20, 176186, https://doi.org/10.1002/met.296.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Miura, H., M. Satoh, T. Nasuno, A. T. Noda, and K. Oouchi, 2007: A Madden-Julian oscillation event realistically simulated by a global cloud-resolving model. Science, 318, 17631765, https://doi.org/10.1126/science.1148443.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Mohd Nor, M. F. F., C. E. Holloway, and P. M. Inness, 2020: The role of local orography on the development of a severe rainfall event over western peninsular Malaysia: A case study. Mon. Wea. Rev., 148, 21912209, https://doi.org/10.1175/MWR-D-18-0413.1.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Mori, S., H. Jun-Ichi, M. D. Yamanaka, N. Okamoto, F. Murata, N. Sakurai, and H. Hashiguchi, 2004: Diurnal land–sea rainfall peak migration over Sumatera Island, Indonesian Maritime Continent, observed by TRMM satellite and intensive rawinsonde soundings. Mon. Wea. Rev., 132, 20212039, https://doi.org/10.1175/1520-0493(2004)132<2021:DLRPMO>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Neale, R., and J. Slingo, 2003: The Maritime Continent and its role in the global climate: A GCM study. Mon. Wea. Rev., 16, 834848, https://doi.org/10.1175/1520-0442(2003)016<0834:TMCAIR>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Pearson, K. J., R. J. Hogan, R. P. Allan, G. M. S. Lister, and C. E. Holloway, 2010: Evaluation of the model representation of the evolution of convective systems using satellite observations of outgoing longwave radiation. J. Geophys. Res., 115, D20206, https://doi.org/10.1029/2010JD014265.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Pearson, K. J., G. M. S. Lister, C. E. Birch, R. P. Allan, R. J. Hogan, and S. J. Woolnough, 2014: Modelling the diurnal cycle of tropical convection across the ‘grey zone.’ Quart. J. Roy. Meteor. Soc., 140, 491499, https://doi.org/10.1002/qj.2145.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Peatman, S. C., A. J. Matthews, and D. P. Stevens, 2014: Propagation of the Madden–Julian oscillation through the Maritime Continent and scale interaction with the diurnal cycle of precipitation. Quart. J. Roy. Meteor. Soc., 140, 814825, https://doi.org/10.1002/qj.2161.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Peatman, S. C., J. Methven, and S. J. Woolnough, 2015: Propagation of the Madden–Julian oscillation and scale interaction with the diurnal cycle in a high-resolution GCM. Climate Dyn., 45, 29012918, https://doi.org/10.1007/s00382-015-2513-5.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Peatman, S. C., J. Methven, and S. J. Woolnough, 2018: Isolating the effects of moisture entrainment on convectively coupled equatorial waves in an aquaplanet GCM. J. Atmos. Sci., 75, 31393157, https://doi.org/10.1175/JAS-D-18-0098.1.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Porson, A. N., S. Hagelin, D. F. A. Boyd, N. M. Roberts, R. North, S. Webster, and J. C. Lo, 2019: Extreme rainfall sensitivity in convective-scale ensemble modelling over Singapore. Quart. J. Roy. Meteor. Soc., 145, 30043022, https://doi.org/10.1002/qj.3601.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Porson, A. N., and Coauthors, 2020: Recent upgrades to the Met Office convective-scale ensemble: An hourly time-lagged 5-day ensemble. Quart. J. Roy. Meteor. Soc., 146, 32453265, https://doi.org/10.1002/qj.3844.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Qian, J. H., 2008: Why precipitation is mostly concentrated over islands in the Maritime Continent. J. Atmos. Sci., 65, 14281441, https://doi.org/10.1175/2007JAS2422.1.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Raynaud, L., and F. Bouttier, 2017: The impact of horizontal resolution and ensemble size for convective-scale probabilistic forecasts. Quart. J. Roy. Meteor. Soc., 143, 30373047, https://doi.org/10.1002/qj.3159.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Rezacova, D., P. Zacharov, and Z. Sokol, 2009: Uncertainty in the area-related QPF for heavy convective precipitation. Atmos. Res., 93, 238246, https://doi.org/10.1016/j.atmosres.2008.12.005.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Roberts, B., I. L. Jirak, A. J. Clark, S. J. Weiss, and J. S. Kain, 2019: Postprocessing and visualization techniques for convection-allowing ensembles. Bull. Amer. Meteor. Soc., 100, 12451258, https://doi.org/10.1175/BAMS-D-18-0041.1.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Roberts, N. M., and H. W. Lean, 2008: Scale-selective verification of rainfall accumulations from high-resolution forecasts of convective events. Mon. Wea. Rev., 136, 7897, https://doi.org/10.1175/2007MWR2123.1.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Sato, T., H. Miura, M. Satoh, Y. N. Takayabu, and Y. Wang, 2009: Diurnal cycle of precipitation in the tropics simulated in a global cloud-resolving model. J. Climate, 22, 48094826, https://doi.org/10.1175/2009JCLI2890.1.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Schwartz, C. S., and R. A. Sobash, 2019: Revisiting sensitivity to horizontal grid spacing in convection-allowing models over the central and eastern United States. Mon. Wea. Rev., 147, 44114435, https://doi.org/10.1175/MWR-D-19-0115.1.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Schwartz, C. S., and Coauthors, 2009: Next-day convection-allowing WRF model guidance: A second look at 2-km versus 4-km grid spacing. Mon. Wea. Rev., 137, 33513372, https://doi.org/10.1175/2009MWR2924.1.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Schwartz, C. S., G. S. Romine, K. R. Smith, and M. L. Weisman, 2014: Characterizing and optimizing precipitation forecasts from a convection-permitting ensemble initialized by a mesoscale ensemble Kalman filter. Wea. Forecasting, 29, 12951318, https://doi.org/10.1175/WAF-D-13-00145.1.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Schwartz, C. S., G. S. Romine, R. A. Sobash, K. R. Fossell, and M. L. Weisman, 2015: NCAR’s experimental real-time convection-allowing ensemble prediction system. Wea. Forecasting, 30, 16451654, https://doi.org/10.1175/WAF-D-15-0103.1.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Schwartz, C. S., G. S. Romine, K. R. Fossell, A. Sobash, and M. L. Weisman, 2017: Toward 1-km ensemble forecasts over large domains. Mon. Wea. Rev., 145, 29432969, https://doi.org/10.1175/MWR-D-16-0410.1.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Stensrud, D. J., J. Bao, and T. T. Warner, 2000: Using initial condition and model physics perturbations in short-range ensemble simulations of mesoscale convective systems. Mon. Wea. Rev., 128, 20772107, https://doi.org/10.1175/1520-0493(2000)128<2077:UICAMP>2.0.CO;2.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Sun, X., and Coauthors, 2020: A subjective and objective evaluation of model forecasts of Sumatra squall events. Wea. Forecasting, 35, 489506, https://doi.org/10.1175/WAF-D-19-0187.1.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Sunilkumar, K., A. Yatagai, and M. Masuda, 2019: Preliminary evaluation of GPM-IMERG rainfall estimates over three distinct climate zones with APHRODITE. Earth Space Sci., 6, 1321–1335, https://doi.org/10.1029/2018EA000503.

    • Crossref
    • Export Citation
  • Supari, F., E. Salimun, E. Aldrian, A. Sopaheluwakan, and L. Juneng, 2018: ENSO modulation of seasonal rainfall and extremes in Indonesia. Climate Dyn., 51, 25592580, https://doi.org/10.1007/s00382-017-4028-8.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Takahashi, H. G., and T. Yasunari, 2008: Decreasing trend in rainfall over Indochina during the late summer monsoon: Impact of tropical cyclones. J. Meteor. Soc. Japan, 86, 429438, https://doi.org/10.2151/jmsj.86.429.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Tan, J., G. J. Huffman, D. T. Bolvin, and E. J. Nelkin, 2019: Diurnal cycle of IMERG V06 precipitation. Geophys. Res. Lett., 46, 13 58413 592, https://doi.org/10.1029/2019GL085395.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Tan, M. L., and Z. Duan, 2017: Assessment of GPM and TRMM precipitation products over Singapore. Remote Sens., 9, 720, https://doi.org/10.3390/rs9070720.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Tan, M. L., and H. Santo, 2018: Comparison of GPM IMERG, TMPA 3B42 and PERSIANN-CDR satellite precipitation products over Malaysia. Atmos. Res., 202, 6376, https://doi.org/10.1016/j.atmosres.2017.11.006.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Villafuerte, M. Q., and J. Matsumoto, 2015: Significant influences of global mean temperature and ENSO on extreme rainfall in Southeast Asia. J. Climate, 28, 19051919, https://doi.org/10.1175/JCLI-D-14-00531.1.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Vincent, C. L., and T. P. Lane, 2018: Mesoscale variation in diabatic heating around Sumatra, and its modulation with the Madden–Julian oscillation. Mon. Wea. Rev., 146, 25992614, https://doi.org/10.1175/MWR-D-17-0392.1.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Vogel, P., P. Knippertz, A. H. Fink, A. Schlueter, and T. Gneiting, 2018: Skill of global raw and postprocessed ensemble predictions of rainfall over northern tropical Africa. Wea. Forecasting, 33, 369388, https://doi.org/10.1175/WAF-D-17-0127.1.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Walters, D., and Coauthors, 2017: The Met Office Unified Model Global Atmosphere 6.0/6.1 and JULES Global Land 6.0/6.1 configurations. Geosci. Model Dev., 10, 14871520, https://doi.org/10.5194/gmd-10-1487-2017.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Wang, J., J. Chen, J. Du, Y. Zhang, Y. Xia, and G. Deng, 2018: Sensitivity of ensemble forecast verification to model bias. Mon. Wea. Rev., 146, 781796, https://doi.org/10.1175/MWR-D-17-0223.1.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Weller, E., K. Shelton, M. Reeder, and C. Jakob, 2017: Precipitation associated with convergence lines. J. Climate, 30, 31693183, https://doi.org/10.1175/JCLI-D-16-0535.1.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Wilson, D. R., A. C. Bushell, A. M. Kerr-Munslow, J. D. Price, and C. J. Morcrette, 2008: PC2: A prognostic cloud fraction and condensation scheme. I: Scheme description. Quart. J. Roy. Meteor. Soc., 134, 20932107, https://doi.org/10.1002/qj.333.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Woodhams, B. J., C. E. Birch, J. H. Marsham, C. L. Bain, N. M. Roberts, and D. F. A. Boyd, 2018: What is the added value of a convection-permitting model for forecasting extreme rainfall over tropical East Africa? Mon. Wea. Rev., 146, 27572780, https://doi.org/10.1175/MWR-D-17-0396.1.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Wu, P., M. Hara, J. Hamada, M. D. Yamanaka, and F. Kimura, 2009: Why a large amount of rain falls over the sea in the vicinity of western Sumatra Island during nighttime. J. Appl. Meteor. Climatol., 48, 13451361, https://doi.org/10.1175/2009JAMC2052.1.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Wu, P., A. A. Arbain, S. Mori, J. Hamada, M. Hattori, F. Syamsudin, and M. D. Yamanaka, 2013: The effects of an active phase of the Madden–Julian oscillation on the extreme precipitation event over western Java island in January 2013. SOLA, 9, 7983, https://doi.org/10.2151/sola.2013-018.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Xavier, P., R. Rahmat, W. K. Cheong, and E. Wallace, 2014: Influence of Madden–Julian oscillation on Southeast Asia rainfall extremes: Observations and predictability. Geophys. Res. Lett., 41, 44064412, https://doi.org/10.1002/2014GL060241.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Yamanaka, M. D., 2016: Physical climatology of Indonesian Maritime Continent: An outline to comprehend observational studies. Atmos. Res., 178–179, 231259, https://doi.org/10.1016/j.atmosres.2016.03.017.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Yang, G.-Y., and J. Slingo, 2001: The diurnal cycle in the tropics. Mon. Wea. Rev., 129, 784801, https://doi.org/10.1175/1520-0493(2001)129<0784:TDCITT>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • View in gallery
    Fig. 1.

    Malaysia (black boxes), Indonesia (blue boxes), and the Philippines (red boxes) forecast domains for 2.2-km forecasts (dashed lines) and 4.5 km-forecasts (solid lines). The full domain shows the 8.8-km domain. For analysis, the 8.8-km forecasts are subset to the relevant smaller regional domains.

  • View in gallery
    Fig. 2.

    GPM-IMERG (dashed lines) and 4.5-km forecast (solid lines) 95th percentile for (a) 24-hourly accumulations, (b) 3-hourly accumulations, and (c) mean of 24-hourly precipitation accumulations as a function of forecast lead time. All values are converted to rain rate in mm h−1. Values shown here are the thresholds used for FSS calculation (see section 3). For forecasts the percentile are calculated over all ensemble members, all forecasts, and all grid points in the domain.

  • View in gallery
    Fig. 3.

    GPM-IMERG and 4.5-km forecast 95th percentile precipitation for (left) Malaysia, (center) Indonesia, and (right) the Philippines. The 95th percentile is calculated for 24-hourly accumulations and is displayed as mm h−1. Forecast precipitation is shown for lead day 1 (hours 0–24), lead day 3 (hours 48–72), and lead day 5 (hours 96–120).

  • View in gallery
    Fig. 4.

    The eFSS of 24-hourly accumulated precipitation exceeding 95th percentile aggregated over all 4.5-km forecasts in October 2018–March 2019 as function of spatial scale (x axis) and lead time (y axis) for the (a) Malaysia model domain, (b) Indonesia model domain, and (c) Philippines model domain. The red line shows the eFSS = 0.5 “skillful” contour.

  • View in gallery
    Fig. 5.

    GPM-IMERG and 4.5-km forecast 95th percentile precipitation for (left) Malaysia, (center) Indonesia, and (right) Philippines. The 95th percentile is calculated for 3-hourly accumulations at 1200–1500 UTC (local evening) and is displayed as mm h−1. Forecast precipitation is shown for lead day 1 (hours 12–15), lead day 3 (hours 60–63), and lead day 5 (hours 108–111).

  • View in gallery
    Fig. 6.

    The eFSS of 3-hourly accumulated precipitation exceeding 95th percentile aggregated over all 4.5-km forecasts in October 2018–March 2019 as a function of spatial scale (x axis) and lead time (y axis) for the (a) Malaysia model domain, (b) Indonesia model domain, and (c) Philippines model domain. The red line shows the eFSS = 0.5 “skillful” contour for 3-hourly forecasts. The blue line shows the eFSS = 0.5 contour for 3-hourly “shifted forecasts” where the forecast time stamp has been shifted a day ahead of observations. The green line shows the FSS = 0.5 contour for a persistence forecast such that lead time corresponds to the lag of the persistence forecast (see methods).

  • View in gallery
    Fig. 7.

    Localized FSS (LFSS) for the 72-km neighborhood-scale 3-hourly accumulated precipitation exceeding 95th percentile aggregated over all 4.5 km forecasts in October 2018–March 2019 for the (left) Malaysia model domain, (center) the Indonesia model domain, and (right) Philippines model domain. LFSS is calculated for lead hours 12–15, 60–63, and 108–111. The red line shows the LFSS = 0.5 contour.

  • View in gallery
    Fig. 8.

    The eFSS calculated using land and ocean points for the 72-km neighborhood-scale 3-hourly accumulated precipitation exceeding 95th percentile aggregated over all 4.5-km forecasts in October 2018–March 2019 for (a) the Malaysia model domain, (b) the Indonesia model domain, and (c) Philippines model domain. Solid lines show eFSS over land and dashed lines show eFSS over ocean.

  • View in gallery
    Fig. 9.

    dFSS of 3-hourly accumulated precipitation exceeding 95th percentile aggregated over all 4.5-km forecasts in October 2018–March 2019 as function of spatial scale (x axis) and lead time (y axis) for the (a) Malaysia domain (note that local time is UTC + 8), (c) Indonesia domain, and (e) Philippines domain. The red line shows the dFSS = 0.5 contour for 3-hourly forecasts. (b),(d),(f) The difference between eFSS and dFSS for the same forecasts as in (a), (c), and (e), such that a negative value indicates the ensemble is underspread and a positive value indicates the ensemble is overspread.

  • View in gallery
    Fig. 10.

    (a)–(c) FSS and the difference with (d)–(f) dFSS at a spatial scale of 144 km of 3-hourly accumulated precipitation exceeding 95th percentile aggregated over all forecasts in October 2018–March 2019 as a function of lead time (x axis) for the (a),(d) Malaysia model domain, (b),(e) Indonesia model domain, and (c),(f) Philippines model domain. The color of the lines indicates the type of forecast; 2.2-km CP (black), 4.5-km CP (blue), 8.8-km CP (red), and 8.8-km GA (green). The horizontal dashed line in (a)–(c) indicates a skillful forecast (eFSS > 0.5). In (d)–(f), negative values indicated an underspread ensemble, positive values indicate an overspread ensemble. Analysis in (d)–(f) is performed over a subregion of the stated 2.2-km region (see figure discussion).

All Time Past Year Past 30 Days
Abstract Views 448 152 0
Full Text Views 297 190 37
PDF Downloads 395 252 40

Evaluating Convection-Permitting Ensemble Forecasts of Precipitation over Southeast Asia

Samantha FerrettaDepartment of Meteorology, University of Reading, Reading, United Kingdom

Search for other papers by Samantha Ferrett in
Current site
Google Scholar
PubMed
Close
,
Thomas H. A. FrameaDepartment of Meteorology, University of Reading, Reading, United Kingdom

Search for other papers by Thomas H. A. Frame in
Current site
Google Scholar
PubMed
Close
,
John MethvenaDepartment of Meteorology, University of Reading, Reading, United Kingdom

Search for other papers by John Methven in
Current site
Google Scholar
PubMed
Close
,
Christopher E. HollowayaDepartment of Meteorology, University of Reading, Reading, United Kingdom

Search for other papers by Christopher E. Holloway in
Current site
Google Scholar
PubMed
Close
,
Stuart WebsterbMet Office, Exeter, United Kingdom

Search for other papers by Stuart Webster in
Current site
Google Scholar
PubMed
Close
,
Thorwald H. M. SteinaDepartment of Meteorology, University of Reading, Reading, United Kingdom

Search for other papers by Thorwald H. M. Stein in
Current site
Google Scholar
PubMed
Close
, and
Carlo CafaroaDepartment of Meteorology, University of Reading, Reading, United Kingdom

Search for other papers by Carlo Cafaro in
Current site
Google Scholar
PubMed
Close
Full access

Abstract

Forecasting rainfall in the tropics is a major challenge for numerical weather prediction. Convection-permitting (CP) models are intended to enable forecasts of high-impact weather events. Development and operation of these models in the tropics has only just been realized. This study describes and evaluates a suite of recently developed Met Office Unified Model CP ensemble forecasts over three domains in Southeast Asia, covering Malaysia, Indonesia, and the Philippines. The fractions skill score is used to assess the spatial scale dependence of skill in forecasts of precipitation during October 2018–March 2019. CP forecasts are skillful for 3-h precipitation accumulations at spatial scales greater than 200 km in all domains during the first day of forecasts. Skill decreases with lead time but varies depending on time of day over Malaysia and Indonesia, due to the importance of the diurnal cycle in driving rainfall in those regions. Skill is largest during daytime when precipitation is over land and is constrained by orography. Comparison of CP ensembles using 2.2-, 4.5-, and 8.8-km grid spacing and an 8.8-km ensemble with parameterized convection reveals that varying resolution has much less effect on ensemble skill and spread than the representation of convection. The parameterized ensemble is less skillful than CP ensembles over Malaysia and Indonesia and more skillful over the Philippines; however, the parameterized ensemble has large drops in skill and spread related to deficiencies in its diurnal cycle representation. All ensembles are underspread indicating that future model development should focus on this issue.

© 2021 American Meteorological Society. For information regarding reuse of this content and general copyright information, consult the AMS Copyright Policy (www.ametsoc.org/PUBSReuseLicenses).

Corresponding author: Samantha Ferrett, s.j.ferrett@reading.ac.uk

Abstract

Forecasting rainfall in the tropics is a major challenge for numerical weather prediction. Convection-permitting (CP) models are intended to enable forecasts of high-impact weather events. Development and operation of these models in the tropics has only just been realized. This study describes and evaluates a suite of recently developed Met Office Unified Model CP ensemble forecasts over three domains in Southeast Asia, covering Malaysia, Indonesia, and the Philippines. The fractions skill score is used to assess the spatial scale dependence of skill in forecasts of precipitation during October 2018–March 2019. CP forecasts are skillful for 3-h precipitation accumulations at spatial scales greater than 200 km in all domains during the first day of forecasts. Skill decreases with lead time but varies depending on time of day over Malaysia and Indonesia, due to the importance of the diurnal cycle in driving rainfall in those regions. Skill is largest during daytime when precipitation is over land and is constrained by orography. Comparison of CP ensembles using 2.2-, 4.5-, and 8.8-km grid spacing and an 8.8-km ensemble with parameterized convection reveals that varying resolution has much less effect on ensemble skill and spread than the representation of convection. The parameterized ensemble is less skillful than CP ensembles over Malaysia and Indonesia and more skillful over the Philippines; however, the parameterized ensemble has large drops in skill and spread related to deficiencies in its diurnal cycle representation. All ensembles are underspread indicating that future model development should focus on this issue.

© 2021 American Meteorological Society. For information regarding reuse of this content and general copyright information, consult the AMS Copyright Policy (www.ametsoc.org/PUBSReuseLicenses).

Corresponding author: Samantha Ferrett, s.j.ferrett@reading.ac.uk

1. Introduction

The nations of Southeast Asia are susceptible to devastating impacts of heavy rainfall such as flooding and landslides. Deep convection is central to extreme rainfall intensity in the region (e.g., Matsumoto et al. 2017) and it also plays an active part in the dynamics of the larger-scale atmospheric phenomena that dominate in the region. There are many contributing factors to the occurrence of convective rainfall events on a range of spatial and temporal scales from the longest to the shortest such as El Niño–Southern Oscillation (Hamada et al. 2012; Villafuerte and Matsumoto 2015; Supari et al. 2018), the Madden–Julian oscillation (MJO; Wu et al. 2013; Peatman et al. 2014; Xavier et al. 2014; Birch et al. 2016; Vincent and Lane 2018; Lestari et al. 2019), cold surges (Chang et al. 2005; Lim et al. 2017), equatorial waves (Ferrett et al. 2020), and tropical cyclones (Takahashi and Yasunari 2008).

Rainfall variability in many regions of Southeast Asia has a strong diurnal cycle, a result of local land–sea-breeze circulations, surface heating during the day and the delayed response of deep convection (e.g., Mori et al. 2004; Yamanaka 2016). Coarse-grid models (including all current operational global models) rely on convection parameterizations and are known to struggle to capture accurately tropical rainfall features (e.g., Neale and Slingo 2003; Johnson et al. 2016), such as the diurnal cycle (Yang and Slingo 2001; Love et al. 2011), the propagation of the MJO and equatorial waves (Lin et al. 2006; Holloway et al. 2013; Peatman et al. 2015, 2018), and other aspects of convection (Qian 2008; Pearson et al. 2010, 2014). Consequently, there has been considerable effort to improve the representation of these processes in models, namely by increasing model resolution so that deep convection can be explicitly simulated by the dynamical core of the model reducing the need for convection parameterizations. In a numerical weather prediction context, there is a trade-off between resolution, model domain size and ensemble size. Resolving deep convection requires a model capable of representing nonhydrostatic, compressible dynamics and a horizontal (and vertical) grid spacing that is much less than the depth of deep convective updrafts (10–20 km). If achieving this resolution requires the domain to be too small, the behavior of systems developing in the domain will be almost completely specified by the lateral boundary conditions imposed by the parent, lower-resolution model, in which the model is nested and there is no benefit in the high-resolution prediction. If the domain is large enough to allow the interior solution of the high-resolution model to deviate from that of its parent, then a compromise must be made on resolution and ensemble size in order to obtain an ensemble forecast in near–real time.

The current state-of-the-art for operational numerical weather prediction is that deep convective motions are only partially resolved; such models have horizontal grid spacing on the order of 1–10 km and are described as “convection permitting” (CP) and the ensembles are small (~10 members). Often shallow moist convection is not resolved and may be parameterized as part of the convective regime of the boundary layer scheme, although the approach taken to turbulence and boundary layer parameterization also depends on model resolution. While CP forecasts are more computationally expensive, they are better able to represent processes that drive convection (Clark et al. 2016). Studies have shown significant improvement in the initiation of convection (Mittermaier et al. 2013; Birch et al. 2014a,b, 2015; Woodhams et al. 2018), the diurnal cycle of convection (Sato et al. 2009; Love et al. 2011; Birch et al. 2015), and large-scale modes that drive convection (Miura et al. 2007; Holloway et al. 2013) in CP models compared to those with parameterized convection.

Forecasting at convective scales is inherently uncertain, even at short lead times (Hohenegger and Schar 2007). This uncertainty can be associated with many things, such as model physics, initial conditions, or boundary conditions. Therefore, ensembles of forecasts are used to account for uncertainty. These convective-scale ensembles have been developed in regions worldwide (e.g., Gebhardt et al. 2011; Golding et al. 2014; Schwartz et al. 2015; Hagelin et al. 2017; Roberts et al. 2019) with obvious improvements in forecast accuracy compared to a single forecast (Hagelin et al. 2017). While there have been many studies of the benefits of using ensemble forecasts to predict the risk of high impact weather in the extratropics (Hanley et al. 2013; Bednarczyk and Ancell 2015) and CP ensembles have been shown to add value to forecasts of mesoscale phenomena such as sea breezes (Cafaro et al. 2019), there are fewer studies examining how CP ensembles may benefit forecasts of extreme rainfall in the tropics more widely and Southeast Asia specifically. A few recent studies have focused on CP ensemble forecasts over Singapore and the surrounding region (Porson et al. 2019; Sun et al. 2020). These studies focused on comparing the effect of global models in which CP ensembles are nested on forecast skill and spread (Porson et al. 2019) and comparing objective and subjective evaluation methods of forecasts of squall lines (Sun et al. 2020). Both studies evaluate against observations in relatively small regions around Singapore (approximately 400 km × 400 km).

In this study CP ensemble forecasts are one-way nested in limited area domains within the operational Met Office Global and Regional Ensemble Prediction System (MOGREPS) using three horizontal resolutions. Three domains within the Southeast Asia region are examined, including peninsular Malaysia, Java, and the Philippines. The aim of this study is to quantify the usefulness of CP ensemble forecasts of precipitation in this region, the scale dependence of forecast skill as well as the role of the diurnal cycle in forecast skill.

Descriptions of the CP ensembles and other datasets used for the analysis are provided in section 2. The methods for evaluation of forecasts using observations are outlined in section 3. Section 3 also provides details of the construction of a variation of the ensemble forecast, and a persistence forecast, that are used to assess the role of the diurnal cycle in forecast skill. Section 4 provides the results of the study, detailing the spatial-scale dependence in the skill of the forecasts, the role of the diurnal cycle and the spread of the ensembles in relation to mean forecast error as a function of lead time. Results are summarized in section 5.

2. Data

a. Ensemble forecasts

The CP ensemble forecasts consist of 18 ensemble members and were created by nesting limited area simulations using the Met Office Unified Model (MetUM) within the 18-member operational global ensemble of MOGREPS (MOGREPS-G). The CP ensembles were initialized twice daily (0000 and 1200 UTC) over a period of six months spanning October 2018–March 2019, producing hourly forecast output in three domains corresponding to Malaysia, Indonesia, and the Philippines (Fig. 1) out to 120 h, except for 2.2-km forecasts, which are run for 60 h. Only the 0000 UTC forecasts are shown here; however, analysis has also been carried out for the 1200 UTC forecasts with similar results after the initial model spinup period, albeit displaced by 12 h.

Fig. 1.
Fig. 1.

Malaysia (black boxes), Indonesia (blue boxes), and the Philippines (red boxes) forecast domains for 2.2-km forecasts (dashed lines) and 4.5 km-forecasts (solid lines). The full domain shows the 8.8-km domain. For analysis, the 8.8-km forecasts are subset to the relevant smaller regional domains.

Citation: Weather and Forecasting 36, 4; 10.1175/WAF-D-20-0216.1

The forecasts use horizontal grid spacing of 2.2, 4.5, and 8.8 km and are nested in MOGREPS-G that has a horizontal grid spacing of 20 km at the equator and 70 vertical levels. The global ensemble initial conditions are derived from the ensemble transform Kalman filter (ETKF) method and a stochastic parameterization scheme is also used in the global ensemble (Bowler et al. 2008). Each member of the limited area ensembles is obtained by one-way nesting from a MOGREPS-G forecast (dynamical downscaling). No additional stochastic perturbation scheme is used within the CP forecasts. The MetUM dynamical core solves a nonhydrostatic, deep atmosphere equation set using a semi-implicit, semi-Lagrangian time-stepping method and has 80 vertical levels. The first Regional Atmosphere and Land (RAL1) science configuration of the dynamics and physics schemes of the atmosphere and land used for the CP simulations in tropical regions (RAL1-T) is documented in Bush et al. (2020). RAL1-T is the tropical subversion of RAL1. The tropical version is required since the midlatitude version (RAL1-M) has relatively weak turbulent mixing and stochastic perturbations, which causes convection to initiate too early and convective cells to be small in the tropics. To account for this, RAL1-T uses the prognostic cloud prognostic condensate (PC2) cloud scheme (Wilson et al. 2008), has an interactive boundary layer free-atmosphere mixing length and has no stochastic boundary layer perturbations.

An 8.8-km ensemble with parameterized convection is also included and uses the configuration of the operational global atmosphere, version 6 (GA6), documented in Walters et al. (2017). This ensemble is referred to throughout as the GA ensemble. Before analysis is carried out all forecasts are regridded to a common 9-km grid using an area-weighted conservative regridding scheme included in the Python library Iris developed by the Met Office (2018).

b. GPM-IMERG

To verify forecasts, rainfall is taken from the Integrated Multisatellite Retrievals for GPM (GPM-IMERG; Huffman et al. 2019). The product used is Level 3 half-hourly Final Run Precipitation at a resolution of 0.1° and combines precipitation estimates from GPM constellation satellites (see https://gpm.nasa.gov/missions/GPM/constellation) and Global Precipitation Climatology Centre (GPCC) precipitation rain gauges. Precipitation estimates from passive microwave radiometers are combined with estimates from infrared data from geostationary weather satellites. Analyses of monthly GPCC gauge accumulations are then used to reduce biases in the multisatellite monthly averages where available. Results using this dataset are referred to in this paper as “GPM” for simplicity.

Before any analysis takes place, the GPM precipitation field is also converted to an average hourly rain rate and is interpolated from a 0.1° grid to a common 9-km grid. Note that this is a slightly higher resolution than the native grid but the area-weighted conservative interpolation scheme maps between staggered grids with similar spacings without affecting integrals over larger scales. It should also be noted that generally both heavy rainfall and rainfall over ocean tend to be underestimated by GPM (Tan and Duan 2017; Kahn and Maggioni 2019; Sunilkumar et al. 2019; Tan and Santo 2018). Nonetheless, studies suggest that in the Philippines, unless examining very heavy rainfall (99th percentile), GPM captures rainfall relatively well (Sunilkumar et al. 2019). The Singapore diurnal cycle of rainfall is well represented in GPM (Tan et al. 2019) and Tan and Santo (2018) also concluded that GPM was a reliable precipitation source for a flooding event in Malaysia during 2014–15. A recent study finds that GPM precipitation is similar to local precipitation for percentiles between the 85th and 95th and concludes that IMERG can be used for forecast evaluation of precipitation up to the 95th percentile (N. De Silva et al. 2021, personal communication).

3. Methods

a. FSS

The fractions skill score (FSS; Roberts and Lean 2008) is a metric that compares two gridded fields and measures the degree of correspondence as a function of spatial scale. To calculate the FSS the regridded (see section 2) forecast and observation fields are converted into binary fields (1 or 0) based on values in each grid cell being above or below a threshold. In this study a threshold of the 95th percentile of rainfall is used, calculated using all grid cells (including zero values) over the regridded domains and the six months spanned by the forecasts. The threshold is calculated separately using GPM-IMERG and the ensemble forecast data for the GPM and forecast binary fields, respectively. The threshold varies depending on time of day in the case of GPM and depending on lead time in the case of forecasts. Percentiles are used for the threshold choice in calculations (rather than a fixed rain-rate threshold) to counteract the influence of intensity bias that effects the frequency of “event” occurrence.

A neighborhood length N (number of grid cells) is defined and is used to convert regularly gridded fields into fractions based on how many grid cells within neighborhoods of size N × N have cell values exceeding the threshold (see Fig. 2 in Roberts and Lean 2008). The FSS is given as
FSS(N)=1MSE(N)MSE(N)ref,
where
MSE(N)=1NxNyi=1Nxj=1Ny[O(N)i,jM(N)i,j]2,and
MSE(N)ref=1NxNy[i=1Nxj=1NyO(N)i,j2+i=1Nxj=1NyM(N)i,j2],
such that O(N) and M(N) are the fields of fractions for neighborhood length N for observations and models, respectively; Nx and Ny are the number of neighborhoods in the full domain along the longitude and latitude axes, respectively. MSE stands for mean square error. MSE(N) is the MSE between the observed and modeled fractions, and MSE(N)ref is the maximum possible error from the two fields.

This calculation can be computed for varying N to obtain FSS as a function of neighborhood size. The aim is to allow for the fact that the location of features on the smallest scales in CP models, associated with convective updrafts, is highly unpredictable, while the probability that a certain fraction of neighborhood area is occupied by precipitation cells (reaching an intensity threshold) may be predicted skillfully. Furthermore, in the absence of data assimilation for high-resolution observations (such as radar) there is no information in the initial conditions on the smaller scales. The FSS value must lie between 1 and 0 with an FSS of 1 indicating a perfect forecast. FSS is expected to increase with neighborhood size. A threshold of FSS can be used to determine the minimum spatial scale at which the forecast is considered skillful. This is taken to be 0.5 if observed rainfall frequency is sufficiently low (see Roberts and Lean 2008).

Using this calculation as a basis, this study examines the ensemble-aggregated FSS (eFSS), the dispersion FSS (dFSS), and the localized FSS (LFSS). Because this study deals with an ensemble, it is required that FSS for all ensemble members is summarized, so to do this the average of MSE(N) and MSE(N)ref is taken over all ensemble members and all forecasts in the six-month period prior to calculation of the FSS using Eq. (1), resulting in the eFSS (Dey et al. 2014).

To obtain a measure of ensemble spread in relation to skill, the dFSS is also calculated. The method for this is very similar to that of the eFSS except rather than compare ensemble members to observations they are compared to the control member of the ensemble (Rezacova et al. 2009; Dey et al. 2014). This provides a metric that demonstrates how similar precipitation patterns are between members of the ensemble. It is possible that model bias can influence metrics of spread (e.g., Wang et al. 2018). However, this FSS-based metric uses a percentile threshold, therefore accounting for precipitation intensity biases as part of the calculation.

Finally, a localized FSS (Woodhams et al. 2018) is used to determine regions in the domain that have particularly high or low values of skill relative to the full domain. For this metric, the MSE(N) and MSE(N)ref are calculated for each neighborhood (using the same domainwide threshold precipitation rate, calculated from a percentile of the data from all forecasts at all grid points, as described above) but are not averaged over the domain prior to FSS calculation.

b. Persistence forecast and shifted forecast

Forecast skill should be benchmarked against some more simple climatological or persistence-based reference forecast, since a useful forecast has to provide greater skill than these basic methods. The FSS is calculated for three variations of forecast over October 2018–March 2019. The three variations are the following:

  • The standard CP ensemble forecasts.

  • A “shifted ensemble forecast” where forecast time stamps (for all lead times) are shifted by adding one day such that the forecast is verified against observations that occur a day later than the actual forecast verification times.

  • A persistence forecast based on forward extrapolation from observations.

The shifted forecast tests how much potential predictability comes from the similarity of the observed diurnal cycle from one day to the next (e.g., the pattern of precipitation over mountains and coasts). If the standard ensemble were no more skillful than the shifted forecast, it would imply that the model gives little useful information beyond its representation of the diurnal cycle in that flow regime and season.

The persistence forecast uses no model data but instead uses hourly GPM precipitation from the day prior to forecast initialization and replicates this for every 24-h period of the forecast. If the observed precipitation at each location were dominated by its diurnal cycle, the persistence forecast would be almost perfect, while the shifted forecast would have the same skill as the standard ensemble and comparison of both these forecasts with observations would reflect model bias. However, longer-time-scale variability reduces the skill in the persistence forecast.

4. Results

a. Rainfall climatology and 4.5-km forecast bias

GPM 95th percentile rainfall amounts in the Malaysia and Indonesia domains have a pronounced diurnal variation (dashed lines in Fig. 2b). Each domain has 2 peaks in rainfall: one around 0800 UTC and one around 2000 UTC. Local time in Malaysia and the Philippines is UTC + 8 h and Indonesia local time is UTC + 7 h so these times correspond to local early evening when rainfall is mostly over land and local early morning when rainfall is mostly over ocean (Mori et al. 2004). The morning peak is larger in the Malaysia domain than in the Indonesia domain. This is because rainfall off the northwest coast of Sumatra is relatively strong during early morning. The Philippines also has a variation throughout the day at similar times, but it is not as pronounced as that in the other two domains (Fig. 2b). Only the 95th percentile is shown here, but the diurnal variation is consistent across other intensities, such as average rainfall (not shown).

Fig. 2.
Fig. 2.

GPM-IMERG (dashed lines) and 4.5-km forecast (solid lines) 95th percentile for (a) 24-hourly accumulations, (b) 3-hourly accumulations, and (c) mean of 24-hourly precipitation accumulations as a function of forecast lead time. All values are converted to rain rate in mm h−1. Values shown here are the thresholds used for FSS calculation (see section 3). For forecasts the percentile are calculated over all ensemble members, all forecasts, and all grid points in the domain.

Citation: Weather and Forecasting 36, 4; 10.1175/WAF-D-20-0216.1

Percentiles are also calculated for forecasts. Note here that there is no averaging involved in this calculation; the percentiles are calculated over all ensemble members, all available forecasts, and all grid points in the domain. For 3-hourly accumulations, the forecasts tend to underestimate the 95th percentile rainfall in comparison to GPM (Fig. 2b; Table 1). A similar result was found by Woodhams et al. (2018) examining the 95th percentile rainfall in East Africa forecasts. In contrast, the 95th percentile and mean of 24-hourly accumulations (Figs. 2a,c) are overestimated, as is the 3-hourly 99th percentile (Table 1). Woodhams et al. (2018) find that in Africa, at higher percentiles, such as 99%, the CP model has higher values than observed. This is consistent with other studies that find that CP models in the extratropics overestimate rainfall amounts and persistence at very high intensities, but underestimate rainfall at lower intensities (e.g., Kendon et al. 2012). However, bias varies with lead time and CP models tend to have large amounts of rainfall after spinup. Note also that GPM is known to underestimate heavy rainfall events (Tan and Duan 2017; Sunilkumar et al. 2019) that will impact the extent to which the model shows “bias” in comparison, though N. De Silva et al. (2021, personal communication) find that 95th percentile rainfall over the Maritime Continent in GPM is suitable for forecast verification.

Table 1.

The average over all lead times of 95th and 99th percentiles (pc) for 3-hourly accumulations of GPM and 2.2-km CP, 4.5-km CP, 8.8-km CP, and 8.8-km GA forecasts for all three regions (using 2.2-km domains in Fig. 1). All values are converted to rain rate in mm h−1.

Table 1.

Forecast precipitation rapidly increases during the initial hours of the forecast as the model spins up, then declines with lead time (Fig. 2b). Variations with the diurnal cycle are captured in the forecasts and peaks tend to occur at the correct times of day; however, the difference between the morning and evening peaks in rainfall is larger than in GPM for the Malaysia and Indonesia domains. This suggests that rainfall over the ocean is not as well captured in the model, or is overestimated in GPM, or a combination of both. Since both heavy rainfall and rainfall over ocean tend to be underestimated by GPM (Tan and Duan 2017; Khan and Maggioni 2019; Sunilkumar et al. 2019) this suggests the bias largely lies with the ensemble.

The 95th percentile of precipitation in Malaysia forecasts also shows a relatively large decrease following the first day of forecasts (Fig. 2). Such abrupt declines between day 1 and day 3 are less evident in the Indonesia and Philippines domains (Fig. 2). This visible drop in forecast precipitation for Malaysia may be a result of the spatial distribution of precipitation in the Malaysia domain between ocean and land. Indeed, when performing the percentile analysis for only land points and only ocean points this decline is visible for the ocean rainfall but less so for the land rainfall (not shown). This may be a result of differing drivers of rainfall in the regions. In the Malaysia domain there is strong rainfall off the east coast of Sumatra, mainly driven by convergence because of Sumatran orography (Wu et al. 2009). Rainfall in the Indonesia domain is also largely driven by convergence lines between landmasses, though Wu et al. (2009) suggest that differences in orography of nearby landmasses may explain differences in ocean rainfall across Southeast Asia, such as between the east coast of Sumatra and the west coast of Borneo. Therefore, the representation of orography may be a large factor in diurnal cycle rainfall biases.

Other aspects of the ensemble may also contribute to biases, such as initial conditions, or the dynamical configuration and parameterizations of the parent ensemble. The drop in rainfall is also found by Dipankar et al. (2020) in similar CP forecasts covering Singapore and surrounding regions, such as peninsular Malaysia and Sumatra (SINGV). Dipankar et al. (2020) find that rainfall in a version of the model driven by the global Unified Model (UM) declines over ocean following 24 h. This was not the case in a version of the model driven by the ECMWF operational deterministic forecast. It is suggested that this is primarily a result of the lateral boundary conditions and a dry bias in global UM forecasts. Further analysis would be required to determine the precise causes of bias here, and why they differ between regions. Examining the spatial pattern of extreme precipitation in subsequent figures will highlight the differences between the domains further. At longer lead times, rainfall in the Malaysia and Indonesia domains continues to decrease, whereas the Philippines rainfall remains well captured at all lead times.

In GPM, the highest values of 95th percentile rainfall are over ocean, particularly over the west coast of Sumatra (Figs. 3a,b). On the first day of forecasts the largest precipitation amounts occur over mountainous regions of Sumatra and Java (Figs. 3d,e), in contrast to GPM. There is some heavier rainfall off the west coast of Sumatra where the observed rainfall peaks, but they are smaller than the land rainfall amounts. The 95th percentile of rainfall in the Philippines domain is well replicated with comparable rainfall amounts to GPM in the south of the Philippines and the northeast region of Borneo (Fig. 3f). Rainfall over mountains is heavier in the forecasts than observed, explaining the discrepancies in the diurnal variations of rainfall (Figs. 2a,b).

Fig. 3.
Fig. 3.

GPM-IMERG and 4.5-km forecast 95th percentile precipitation for (left) Malaysia, (center) Indonesia, and (right) the Philippines. The 95th percentile is calculated for 24-hourly accumulations and is displayed as mm h−1. Forecast precipitation is shown for lead day 1 (hours 0–24), lead day 3 (hours 48–72), and lead day 5 (hours 96–120).

Citation: Weather and Forecasting 36, 4; 10.1175/WAF-D-20-0216.1

As lead time increases the 95th percentile of rainfall off the west coast of Sumatra decreases (Fig. 3g). Rainfall over the ocean here decreases at a faster rate than that over land between lead day 1 and lead day 3, explaining the large decrease in the rainfall shown in Fig. 2b after lead day 1. The rainfall in the ocean to the north of Java and in the Strait of Malacca does not show this decrease and even increases slightly. As mentioned previously it seems likely there is a meteorological phenomenon in this area that is more accurately captured than equivalent phenomena occurring west of Sumatra and results in this differing behavior. One possibility is convergence lines in the Strait of Malacca resulting from land–sea breezes (Weller et al. 2017; Mohd Nor et al. 2020) that are relatively well represented and persist in the forecast. At longer lead times rainfall over mountainous regions continues to decrease (Figs. 3j,k). Rainfall amounts over the Philippines show less change as lead time increases (Fig. 3l), as is also indicated in Fig. 2.

b. Skill of 4.5-km ensemble forecasts of daily precipitation accumulations

The spatial scales over which forecasts of 24-hourly accumulated rainfall can be considered skillful is assessed using the eFSS to determine the smallest scale for useful forecast information from each of the ensembles. Forecast skill of 24-hourly precipitation exceeds the skillful threshold of 0.5 (red line in Fig. 4) at spatial scales greater than around 150 km on the first day of forecasts for all three domains. Skill in the Philippines domain is slightly higher and so forecasts are considered skillful at spatial scales greater than around 50 km. Skill decreases as lead time increases such that by the end of the 5-day forecasts only spatial scales exceeding 350 km are considered skillful. It should be noted that here the ensemble-average FSS is examined for comparison with ensemble spread, not a metric based on the probabilistic output of the FSS. The skill of the probabilistic ensemble output is slightly higher (not shown) and so the probabilistic forecasts will be able to be displayed usefully at slightly smaller scales than these (approximately 50–100 km smaller depending on lead time and region).

Fig. 4.
Fig. 4.

The eFSS of 24-hourly accumulated precipitation exceeding 95th percentile aggregated over all 4.5-km forecasts in October 2018–March 2019 as function of spatial scale (x axis) and lead time (y axis) for the (a) Malaysia model domain, (b) Indonesia model domain, and (c) Philippines model domain. The red line shows the eFSS = 0.5 “skillful” contour.

Citation: Weather and Forecasting 36, 4; 10.1175/WAF-D-20-0216.1

While this is useful information for forecasting daily rainfall totals, variations in rainfall with the diurnal cycle and between land and sea mean that forecast skill is likely to vary with both location and time of day, particularly in Malaysia and Indonesia. To further understand contributors to forecast skill it is important to also examine diurnal variations in forecast skill, as well as spatial variations in skill.

c. The role of the diurnal cycle in 4.5-km forecast skill

As mentioned previously, rainfall has a strong diurnal cycle during the day and can be defined by two peaks: one during local early morning when rainfall is mostly over the ocean (not shown), and one during local early evening when rainfall is over land (Fig. 5). In GPM during early evening the highest values of rainfall are over Sumatra and Borneo (Figs. 5a,b). The rainfall over Sumatra tends to be located around orographic features: mountains run down the west side of Sumatra. There are also large amounts of rainfall just off the west coast of Sumatra. There are smaller amounts of rainfall over peninsular Malaysia, Java, and the southern Philippines (Figs. 5a–c).

Fig. 5.
Fig. 5.

GPM-IMERG and 4.5-km forecast 95th percentile precipitation for (left) Malaysia, (center) Indonesia, and (right) Philippines. The 95th percentile is calculated for 3-hourly accumulations at 1200–1500 UTC (local evening) and is displayed as mm h−1. Forecast precipitation is shown for lead day 1 (hours 12–15), lead day 3 (hours 60–63), and lead day 5 (hours 108–111).

Citation: Weather and Forecasting 36, 4; 10.1175/WAF-D-20-0216.1

For forecasts, 12 h after forecast initialization the spatial pattern of 95th percentile precipitation is similar to that in GPM but with varying amounts. Precipitation in Sumatra is most strong in the northwest (Fig. 5d), unlike GPM where larger amounts of precipitation run the full length of the island. Forecasts do not capture the precipitation off the west coast of Sumatra (Fig. 5d), though they still have some precipitation over the ocean, similar to GPM. Three days into the forecast precipitation over the ocean is decreased (Fig. 5g) but the precipitation over land is still relatively strong. Examining early morning in the forecasts also shows that precipitation over the ocean around Sumatra decreases following the first day forecast (not shown), explaining the large drop in Malaysia domain precipitation shown in Fig. 2a. By the final day of the forecast precipitation amounts have slightly decreased but the spatial pattern remains the same (Figs. 5j–l)

To examine how forecast skill depends on the diurnal cycle in the region, the skill of the three ensembles is examined. The shading in Fig. 6 demonstrates eFSS for the standard ensemble forecast, but the eFSS = 0.5 “skillful threshold” contours are shown for all three forecast variations described in section 3. Forecasts with skill exceeding 0.5, regions to the right of the colored lines in Fig. 6, are considered skillful for the given spatial scale and lead time. Forecasts with lines further to the left in the figure are therefore more skillful at smaller spatial scales.

Fig. 6.
Fig. 6.

The eFSS of 3-hourly accumulated precipitation exceeding 95th percentile aggregated over all 4.5-km forecasts in October 2018–March 2019 as a function of spatial scale (x axis) and lead time (y axis) for the (a) Malaysia model domain, (b) Indonesia model domain, and (c) Philippines model domain. The red line shows the eFSS = 0.5 “skillful” contour for 3-hourly forecasts. The blue line shows the eFSS = 0.5 contour for 3-hourly “shifted forecasts” where the forecast time stamp has been shifted a day ahead of observations. The green line shows the FSS = 0.5 contour for a persistence forecast such that lead time corresponds to the lag of the persistence forecast (see methods).

Citation: Weather and Forecasting 36, 4; 10.1175/WAF-D-20-0216.1

Forecast skill of 3-hourly precipitation is strongly tied to the diurnal cycle for the Malaysia forecasts (Fig. 6a) and the Indonesia forecasts (Fig. 6b). The Philippines forecast skill shows some link to the diurnal cycle, but this is less pronounced than that of the other two domains (Fig. 6c). Skill tends to be largest in the daytime when precipitation is over land and smallest at night when precipitation is offshore. It is likely that this is due to precipitation that is constrained by topography, and therefore more predictable, during the day. For the first day forecasts are considered skillful at spatial scales greater than approximately 200 km (Fig. 6). After the first day forecast skill begins to decrease as lead time increases. On day 5 of the forecasts there is skill on spatial scales greater than around 400 km. These scales are comparable to analysis by Dey et al. (2014) that showed skill on scales greater than 200 km for forecasts of extreme rainfall in the United Kingdom using a similar nested 2.2-km ensemble. Woodhams et al. (2018) also find that a deterministic CP model forecast of extreme rainfall over East Africa is skillful at spatial scales around 275–300 km.

The reliance of skill on time of day leads to a question: how much of forecast skill is driven simply by diurnal variations? The ensemble forecast is compared with the other two forecast variations described in section 3b in order to gain an idea of the role of persisting weather in the forecast skill, and how much value is added by the dynamical ensemble forecast. It is found that the standard ensemble forecast is more skillful than the persistence forecast for all three domains, indicating that the ensemble forecast contains more information about the weather occurring in the future than can be inferred from local knowledge of the diurnal cycle observed on the day before making the forecast.

The “shifted forecast” (see section 3b) tests how much predictability comes from the characteristics of the diurnal cycle in the current flow regime and season. The standard forecast is substantially more skillful than the shifted forecast in all three domains (Fig. 6). This implies that substantial skill in the precipitation forecast is associated with phenomena with multiday time scales, or that the forecast is skillfully predicting day-to-day variation in the characteristics of the diurnal cycle that are conditional on the large-scale environment. It is expected that as lead time increases the standard ensemble forecast skill must tend toward the skill of the shifted forecast as the model forecast increasingly becomes no better than a forecast of a different day.

Interestingly, for Malaysia (Fig. 6a) and Indonesia (Fig. 6b) the shifted forecast is more skillful than the persistence forecast at all lead times beyond the initial 3 h. During the first 24 h, the difference is small. Moving into day 2, the persistence forecast becomes substantially worse than the shifted forecast, implying that the precipitation field has multiday variability that is captured by the model. This is particularly true for the Philippines (Fig. 6c), which has a large drop in the skill of the persistence forecast following day 1 relative to the model. In conclusion, even at 5 days lead time, the ensemble forecast is still more skillful than a forecast based on persistence, as well as the shifted forecasts, that mainly capture the diurnal cycle.

Skill is increased when precipitation is located over land, perhaps because it is more constrained by orography. To examine this further, an LFSS (see section 3a) can be calculated (Fig. 7). It should be noted that while the red line in Fig. 7 indicates a threshold of 0.5 in keeping with Fig. 6, this should not be considered a threshold of “skillfulness” in this context. Rather, the LFSS is a tool to understand the distribution of skill across the full domain, in relation to the eFSS.

Fig. 7.
Fig. 7.

Localized FSS (LFSS) for the 72-km neighborhood-scale 3-hourly accumulated precipitation exceeding 95th percentile aggregated over all 4.5 km forecasts in October 2018–March 2019 for the (left) Malaysia model domain, (center) the Indonesia model domain, and (right) Philippines model domain. LFSS is calculated for lead hours 12–15, 60–63, and 108–111. The red line shows the LFSS = 0.5 contour.

Citation: Weather and Forecasting 36, 4; 10.1175/WAF-D-20-0216.1

During early evening higher skill tends to be located over land where the forecast has most precipitation, such as in the northwest region of Sumatra (Fig. 7a) and over Java (Fig. 7b). Comparing the spatial patterns of skill to the 95th percentile of precipitation (Fig. 3) shows that the spatial patterns are similar in these regions. There is much less skill over ocean for the Malaysia and Indonesia domains (Figs. 8a,b). Peaks in skill over ocean also tend to be lower than those over land (Fig. 8). This supports the idea that most skill comes from precipitation that is spatially constrained, and results in the diurnal variation of skill as precipitation moves from ocean to land.

Fig. 8.
Fig. 8.

The eFSS calculated using land and ocean points for the 72-km neighborhood-scale 3-hourly accumulated precipitation exceeding 95th percentile aggregated over all 4.5-km forecasts in October 2018–March 2019 for (a) the Malaysia model domain, (b) the Indonesia model domain, and (c) Philippines model domain. Solid lines show eFSS over land and dashed lines show eFSS over ocean.

Citation: Weather and Forecasting 36, 4; 10.1175/WAF-D-20-0216.1

There are also regions of high skill that have lower 95th percentile rainfall in both GPM and forecasts, such as the north of the Philippines (Fig. 7c). This skill is likely tied to synoptic-scale variability, that is, tropical cyclones. This region of skill is also relatively independent of the diurnal cycle and is still present at other times of day (not shown), supporting this hypothesis.

As lead time increases, skill decreases rapidly but is still highest in mountainous regions and in the north of the Philippines (Figs. 7d–i). Localized skill analysis suggests that the diurnal cycle is more important for skill in Malaysia and Indonesia since rainfall is more spatially constrained by orographic features when over land. For the Philippines, skill is more likely to come from synoptic-scale systems that are not as dependent on time of day.

d. Ensemble spread–skill relationship for 4.5-km forecasts

The dFSS (Dey et al. 2014) is used to assess spatial differences between ensemble members, or “ensemble spread,” in relation to forecast skill (see section 3a). If the ensemble is “well calibrated” then the dFSS and eFSS should be the same. If the dFSS is smaller than the eFSS, then the ensemble is overspread (underconfident) and if the dFSS is larger than the eFSS, then the ensemble is underspread (overconfident). For Malaysia forecasts the dFSS is larger than eFSS (Figs. 9a,b) at all lead times and spatial scales, indicating that ensemble members are too similar to one another compared to the difference between the forecasts and observations. This is also true for Indonesia and the Philippines (Figs. 9c–f). Ensembles are underspread in relation to the skill, and displacement errors in forecast rainfall features are typically larger than the differences between ensemble members. This is consistent with previous studies that find MetUM convective-scale ensembles to be underspread (Porson et al. 2020; Cafaro et al. 2021), and being underspread is a common error for CP ensembles generally (Schwartz et al. 2014; Beck et al. 2016; Raynaud and Bouttier 2017). Problems with spread can be linked to model errors (e.g., Stensrud et al. 2000) and initial conditions and lateral boundary conditions from the parent ensemble. However, finding the underlying cause in this case would require further study.

Fig. 9.
Fig. 9.

dFSS of 3-hourly accumulated precipitation exceeding 95th percentile aggregated over all 4.5-km forecasts in October 2018–March 2019 as function of spatial scale (x axis) and lead time (y axis) for the (a) Malaysia domain (note that local time is UTC + 8), (c) Indonesia domain, and (e) Philippines domain. The red line shows the dFSS = 0.5 contour for 3-hourly forecasts. (b),(d),(f) The difference between eFSS and dFSS for the same forecasts as in (a), (c), and (e), such that a negative value indicates the ensemble is underspread and a positive value indicates the ensemble is overspread.

Citation: Weather and Forecasting 36, 4; 10.1175/WAF-D-20-0216.1

The spread also varies with the diurnal cycle, consistent with the variation in ensemble skill. Ensemble spread is smallest (larger dFSS) during local evening, with dFSS peaking around 2000–2300 local time (lead 12–15), particularly at longer lead times (Figs. 9a,c,e). Consistent with the skill results discussed previously, this is when precipitation is typically over land and is more spatially constrained, and therefore more similar between ensemble members. When precipitation is over ocean there is slightly more ensemble spread, indicated by reduced dFSS.

e. The role of resolution and convection parameterization in forecast skill and spread

The 4.5-km ensembles examined in the earlier sections are part of a larger set of nested ensembles that include ensembles with a 2.2- and 8.8-km horizontal grid spacing, but the same model levels (see section 2a), and provide an opportunity to examine the role of resolution in forecast skill. Analysis of an 8.8-km ensemble with parameterized convection (GA ensemble) is also included to examine how much skill is gained from partially resolving convection with the dynamical core. Resolution tends to play a fairly minor role in forecast skill for all three regions (Fig. 10). This is also the case when using a higher rainfall percentile threshold, such as the 99th percentile (not shown). The 2.2-, 4.5-, and 8.8-km CP ensembles all have similar variation in skill as a function of lead time and in this measure the 2.2- and 4.5-km forecasts are barely distinguishable. This was found to be true evaluating across the range of neighborhood scales from grid scale to 288 km (results are only shown for 144 km, the smallest scale for which the eFSS exceeds 0.5 for all domains in the first day). During early evening, when skill in the 4.5 km ensemble (blue line in Fig. 10) is highest, the 8.8-km CP ensemble (red line) has less skill in the Malaysia domain than the 2.2- and 4.5-km ensembles (Fig. 10a), but in general the differences are small. A possible reason for this is that resolution may be less important at longer lead times where large-scale conditions dominate, and so resolution variations have little effect on skill.

Fig. 10.
Fig. 10.

(a)–(c) FSS and the difference with (d)–(f) dFSS at a spatial scale of 144 km of 3-hourly accumulated precipitation exceeding 95th percentile aggregated over all forecasts in October 2018–March 2019 as a function of lead time (x axis) for the (a),(d) Malaysia model domain, (b),(e) Indonesia model domain, and (c),(f) Philippines model domain. The color of the lines indicates the type of forecast; 2.2-km CP (black), 4.5-km CP (blue), 8.8-km CP (red), and 8.8-km GA (green). The horizontal dashed line in (a)–(c) indicates a skillful forecast (eFSS > 0.5). In (d)–(f), negative values indicated an underspread ensemble, positive values indicate an overspread ensemble. Analysis in (d)–(f) is performed over a subregion of the stated 2.2-km region (see figure discussion).

Citation: Weather and Forecasting 36, 4; 10.1175/WAF-D-20-0216.1

There is a larger difference in the skill between the CP ensembles and the GA ensemble (green line in Fig. 10), though this is region dependent. For the Malaysia domain the GA ensemble is less skillful across all lead times, except for the first few hours of forecasts, during the spinup period (Fig. 10a). Parameterized convection has less effect on skill in the Indonesia domain, though skill peaks earlier in the day compared to the CP ensemble skill (Fig. 10b). A note to make here is that the diurnal cycle of precipitation in Malaysia and Indonesia is not as well captured by the GA ensemble (not shown); the observed peaks in rainfall tend to occur a few hours earlier in the GA ensemble, around 0600 UTC (early afternoon local time) and 1800 UTC (early morning local time). The afternoon skill peak therefore coincides with a time when rainfall is increasing in observations and decreasing in the GA ensemble. This peak in skill in the Indonesia domain likely reflects a time when rainfall patterns match better than other times of day, but not for the correct reasons; rainfall in the GA ensemble does not move over ocean during night in the same way as in observations and the CP ensembles (not shown). Since rainfall in the Malaysia and Indonesia domains is largely driven by diurnal variations it is expected that the skill of the GA ensemble is less than the CP ensembles that capture diurnal variations more accurately.

While there is little difference between the skill in the CP ensembles it is important to note that rainfall is reduced in lower-resolution 8.8-km CP ensembles (Table 1). Peaks also tend to occur slightly later in the day in these ensembles, particularly for the Malaysia domain, with the early morning peak hardly being captured at all (not shown). This is indicative of larger errors in the diurnal cycle of precipitation in the lower-resolution ensembles, despite little difference being shown in the FSS measure of spatial skill. It is therefore important not to consider all CP ensembles to have equal value for operational forecasting despite similar FSS values.

In the Philippines domain there is almost no difference in skill between the CP ensembles at all resolutions (Fig. 10c). However, following the first 12 h of the forecast the GA ensemble has slightly more skill than the CP ensembles, in contrast to the other two regions. A possible reason for this is that precipitation patterns associated with certain, likely larger-scale, phenomena are represented more accurately in the GA ensemble. The CP forecasts spin up at initialization, and this may explain the higher skill of the GA model since less spinup may occur from the initial conditions. This is a surprising result since previous studies find that drivers of Philippines rainfall, such as tropical cyclones, can be represented more accurately by CP models (Bousquet et al. 2020). A full analysis of the differences in the representation drivers of rainfall between CP and GA ensembles is outside of the scope of this study, but this result certainly highlights an area of future research.

Resolution also has a small impact on ensemble spread (Figs. 9d–f). All three CP ensembles are underspread to a similar degree at the 144-km neighborhood scale. Note here that ensemble spread is shown over a subregion with boundaries 2° away from the 2.2-km boundaries to avoid a larger influence of lateral boundary conditions on the 2.2-km ensemble than the larger domain ensembles. When performing analysis over the full 2.2-km ensemble domain the spread of the 2.2-km ensemble is smaller than that of the 4.5- and 8.8-km CP ensembles (not shown) as a result of the role of lateral boundary conditions on ensemble spread. Ensemble spread in the GA ensemble is relatively similar to the CP ensembles at most times of day but reduces during early afternoon (local time; approximately 0400 UTC) when observed rainfall is moving from over ocean to over land (Figs. 10d,e). This suggests that the ensemble spread in the spatial patterns of rainfall is much less than the CP ensemble as a result of the parameterizations used. In the Philippines, where rainfall is not as strongly driven by the diurnal cycle, ensemble spread is similar in the GA ensemble and the CP ensembles, particularly during the first 2–3 days of the forecast.

5. Summary and conclusions

Convection-permitting (CP) ensemble forecasts in the tropics have, until now, been relatively uncommon. Recently, the Met Office has developed such systems over Singapore (Porson et al. 2019) and East Africa (Cafaro et al. 2021). Model development has now extended to further domains in Southeast Asia, covering Malaysia, Indonesia, and the Philippines. In this study the skill of newly developed CP ensembles over these regions has been examined using the fractions skill score (FSS), with a particular focus on the role of the diurnal cycle and quantifying the spread of the ensemble in relation to skill.

The skill of forecasts of precipitation in the Malaysia and Indonesia domains are strongly linked to the diurnal cycle, peaking during local early evening when rainfall is over land and dropping in local early morning when rainfall is over ocean (Fig. 6). The diurnal variation suggests that orographic and coastal features play a role in skill, such that precipitation over mountainous regions is spatially constrained and therefore better captured by the model. Land–sea contrast is also important to the characteristics of the rainfall. Examination of maps of the localized fractions skill score (LFSS) confirms this, as the largest skill tends to be over mountainous regions and skill is much lower over the ocean (Figs. 7 and 8). The diurnal cycle is less strong in the Philippines domain, suggesting that other modes of variability drive precipitation there. Skill decreases as lead time increases, and forecasts are skillful at spatial scales greater than 400 km for all lead times and all domains, providing a suggested scale for future forecast display.

The large role played by the diurnal cycle motivates the question as to how much of the skill is potentially predictable simply by representing well the diurnal cycle in the region. Vogel et al. (2018) found that forecasts of precipitation in West Africa that were based on climatological precipitation outperformed operational ensemble forecasts that used parameterized convection schemes. Here, comparisons are made to persistence forecasts to examine how the CP ensembles perform relative to a forecast from recently observed rainfall. In this case all three ensembles have much higher skill than a persistence forecast, suggesting that there is significant added value in the CP ensemble compared to a forecast based on observed rainfall from the previous day. Skill for a “shifted forecast,” in which the forecast date stamps are shifted one day later and therefore verified against observations from one day after the true verification date, is also presented to determine how much of the model skill is related to simulation of the diurnal cycle. The skill of the shifted forecast is larger than the persistence forecast, but less than the standard ensemble forecast, for lead times longer than one day. This suggests that increased CP ensemble forecast skill is associated with phenomena characterized by multiday time scales that are captured by the model, highlighting the usefulness of the CP ensemble in this region.

Variation of horizontal grid spacing, across 2.2, 4.5, and 8.8 km, plays a fairly small role in ensemble skill assessed using the fractions skill score over the full range of neighborhood scales. These results agree with previous studies that also find 1–2-km resolution forecasts rainfall in the United States to perform as well as 4-km forecasts (Kain et al. 2008; Schwartz et al. 2009; Loken et al. 2017). However, other studies found that increased resolution resulted in higher forecast skill for U.S. rainfall (Schwartz et al. 2017; Schwartz and Sobash 2019). Schwartz and Sobash (2019) suggest that this could be because of higher quality initial conditions than in previous studies. Given the computational cost of higher-resolution forecasts this is a result to be considered in future forecast model development focusing on this region.

The 8.8-km GA ensemble using a parameterized convection scheme has less skill compared to the CP ensembles, but only in the regions where rainfall patterns are strongly driven by the diurnal cycle, that is, Malaysia and Indonesia. In the Philippines region, CP ensembles have slightly lower skill than the GA ensemble suggesting that CP forecasts add limited value when forecasting spatial patterns of rainfall in the Philippines region. This is likely due to the higher-latitude location of the Philippines and hence stronger synoptic-scale forcing of rainfall and weaker dependence on local thermal contrasts in forcing convective systems. Also, there are no additional high-resolution observations assimilated into the CP forecasting system and the initial conditions for each forecast member are obtained simply by interpolating the parent global member onto a finer grid. The only benefit that could come from the higher resolution is an improvement in the representation of the lower boundary and weather system dynamics. The higher-resolution forecasts need to spin up and this may explain the apparently higher skill of the 8.8-km model with parameterized convection, rather than explicit. The model dynamics with parameterization is more similar to the global model parent and therefore less spinup may occur from the initial conditions. An important note and caveat to make here is that while resolution plays a small role in skill measured using the FSS there are many other reasons why a higher-resolution model is preferable. Other aspects of rainfall, aside from spatial pattern, such as intensity and timing, are of importance and should be considered when choosing models for an operational forecast.

The spread of the ensemble is examined using the dispersion fractions skill score (dFSS), which compares ensemble member pairs to obtain a measure of the spatial differences within the ensemble. Spread also varies with the diurnal cycle such that there is less ensemble spread when precipitation is over land, again supporting the idea that convection is more constrained by orography at that time (Fig. 9). The ensemble is underdispersive. Spread is 59%, 61%, and 33% less than ensemble mean forecast error on average over 4.5-km forecasts for Malaysia, Indonesia, and the Philippines domains, respectively, confirming that these ensembles suffer from a persistent drawback of CP ensembles. While using percentiles for the threshold choice in calculations (rather than a fixed amount threshold) somewhat counteracts the influence of rainfall intensity bias, there are further possible causes of reduced ensemble spread, such as the influence of initial and lateral boundary conditions and model errors. Studies have found various ways to improve ensemble spread, such as combining ensembles (Beck et al. 2016) and time-lagging ensemble member forecasts (Porson et al. 2020). Porson et al. (2019) assessed the difference in skill and spread in an ensemble covering Singapore nested in two different global ensembles. They find that spread is sensitive to initial conditions at the beginning of the forecast and lateral boundary conditions toward the end of the forecast. While beyond the scope of the present study, investigating why the ensembles analyzed here are underspread would be an interesting topic of future research. Assimilation of high-resolution data to generate a higher-resolution regional analysis would be expected to improve initial conditions for the CP ensembles, introducing spread associated with observed mesoscale features and acting to reduce artificial forecast spinup effects that may reduce skill. However, assimilation in the convection-dominated regime in the tropics is challenging, due to convective instability and the high variance on small spatial and temporal scales, and prototype systems assimilating satellite radiance data are currently under development in the Southeast Asia region (Heng et al. 2020).

Despite the lack of sufficient spread in the CP ensembles shown here, the forecasts have significant skill beyond that shown by a persistence forecast based on observed climatology. Aside from improvements to ensemble spread, future work with these forecasts should focus on sources of skill in the three different regions, including the role of large-scale conditions in forecast skill. Understanding drivers of forecast skill will lead to improved forecasts of high-impact weather in the region and should therefore be a high priority in future.

Acknowledgments

This work and SF, THAF, CEH, JM, and THMS were supported by the Weather and Climate Science for Services Partnership (WCSSP) Southeast Asia as part of the Newton Fund. We thank three anonymous reviewers for insightful comments and suggestions on this work. The GPM-IMERG data were provided by the NASA Goddard Space Flight Center’s Precipitation Measurement Missions Science Team and Precipitation Processing System, which develops and computes GPM-IMERG as a contribution to GPM, and archived at the NASA GES DISC.

REFERENCES

  • Beck, J., F. Bouttier, L. Wiegand, C. Gebhardt, C. Eagle, and N. Roberts, 2016: Development and verification of two convection-allowing multi-model ensembles over western Europe. Quart. J. Roy. Meteor. Soc., 142, 28082826, https://doi.org/10.1002/qj.2870.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Bednarczyk, C. N., and B. C. Ancell, 2015: Ensemble sensitivity analysis applied to a southern plains convective event. Mon. Wea. Rev., 143, 230249, https://doi.org/10.1175/MWR-D-13-00321.1.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Birch, C. E., J. H. Marsham, D. J. Parker, and C. M. Taylor, 2014a: The scale dependence and structure of convergence fields preceding the initiation of deep convection. Geophys. Res. Lett., 41, 47694776, https://doi.org/10.1002/2014GL060493.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Birch, C. E., D. J. Parker, J. H. Marsham, D. Copsey, and L. Garcia-Carreras, 2014b: A seamless assessment of the role of convection in the water cycle of the West African monsoon. J. Geophys. Res. Atmos., 119, 28902912, https://doi.org/10.1002/2013JD020887.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Birch, C. E., M. J. Roberts, L. Garcia-Carreras, D. Ackerley, M. J. Reeder, A. P. Lock, and R. Schiemann, 2015: Sea-breeze dynamics and convection initiation: The influence of convective parameterization in weather and climate model biases. J. Climate, 28, 80938108, https://doi.org/10.1175/JCLI-D-14-00850.1.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Birch, C. E., S. Webster, S. C. Peatman, D. J. Parker, A. J. Matthews, Y. Li, and M. E. E. Hassim, 2016: Scale interactions between the MJO and the western Maritime Continent. J. Climate, 29, 24712492, https://doi.org/10.1175/JCLI-D-15-0557.1.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Bousquet, O., D. Barbary, S. Bielli, S. Kebir, L. Raynaud, S. Malardel, and G. Faure, 2020: An evaluation of tropical cyclone forecast in the southwest Indian Ocean basin with AROME-Indian Ocean convection-permitting numerical weather predicting system. Atmos. Sci. Lett., 21, e950, https://doi.org/10.1002/asl.950.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Bowler, N., A. Arribas, K. Mylne, K. Robertson, and S. Beare, 2008: The MOGREPS short-range ensemble prediction system. Quart. J. Roy. Meteor. Soc., 134, 703722, https://doi.org/10.1002/qj.234.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Bush, M., and Coauthors, 2020: The first Met Office Unified model–JULES regional atmosphere and land configuration, RAL1. Geosci. Model Dev., 13, 19992029, https://doi.org/10.5194/gmd-13-1999-2020.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Cafaro, C., T. H. A. Frame, J. Methven, N. Roberts, and J. Bröcker, 2019: The added value of convection-permitting ensemble forecasts of sea breeze compared to a Bayesian forecast driven by the global ensemble. Quart. J. Roy. Meteor. Soc., 145, 17801798, https://doi.org/10.1002/qj.3531.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Cafaro, C., and Coauthors, 2021: Do convection-permitting ensembles lead to more skillful short-range probabilistic rainfall forecasts over tropical East Africa? Wea. Forecasting, 36, 697716, https://doi.org/10.1175/WAF-D-20-0172.1.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Chang, C. P., P. A. Harr, and H. J. Chen, 2005: Synoptic disturbances over the equatorial South China Sea and western Maritime Continent during boreal winter. Mon. Wea. Rev., 133, 489503, https://doi.org/10.1175/MWR-2868.1.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Clark, P., N. Roberts, H. Lean, S. P. Ballard, and C. Charlton-Perez, 2016: Convection-permitting models: A step-change in rainfall forecasting. Meteor. Appl., 23, 165181, https://doi.org/10.1002/met.1538.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Dey, S. R. A., G. Leoncini, N. M. Roberts, R. S. Plant, and S. Migliorini, 2014: A spatial view of ensemble spread in convection permitting ensembles. Mon. Wea. Rev., 142, 40914107, https://doi.org/10.1175/MWR-D-14-00172.1.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Dipankar, A., and Coauthors, 2020: SINGV: A convective-scale weather forecast model for Singapore. Quart. J. Roy Meteor. Soc., 146, 41314146, https://doi.org/10.1002/qj.3895.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Ferrett, S., G. Yang, S. J. Woolnough, J. Methven, K. Hodges, and C. E. Holloway, 2020: Linking extreme precipitation in Southeast Asia to equatorial waves. Quart. J. Roy. Meteor. Soc., 146, 665684, https://doi.org/10.1002/qj.3699.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Gebhardt, C., S. E. Theis, M. Paulat, and Z. Ben Bouallègue, 2011: Uncertainties in COSMO-DE precipitation forecasts introduced by model perturbations and variation of lateral boundaries. Atmos. Res., 100, 168177, https://doi.org/10.1016/j.atmosres.2010.12.008.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Golding, B. W., and Coauthors, 2014: Forecasting capabilities for the London 2012 Olympics. Bull. Amer. Meteor. Soc., 95, 883896, https://doi.org/10.1175/BAMS-D-13-00102.1.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Hagelin, S., J. Son, R. Swinbank, A. McCabe, N. Roberts, and W. Tennant, 2017: The Met Office convective-scale ensemble, MOGREPS-UK. Quart. J. Roy. Meteor. Soc., 143, 28462861, https://doi.org/10.1002/qj.3135.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Hamada, J.-I., S. Mori, H. Kubota, M. D. Yamanaka, U. Haryoko, S. Lestari, R. Sulistyowati, and F. Syamsudin, 2012: Interannual rainfall variability over northwestern Jawa and its relation to the Indian Ocean dipole and El Niño–Southern Oscillation events. SOLA, 8, 6972, https://doi.org/10.2151/sola.2012-018.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Hanley, K. E., D. J. Kirshbaum, N. M. Roberts, and G. Leoncini, 2013: Sensitivities of a squall line over central Europe in a convective-scale ensemble. Mon. Wea. Rev., 141, 112133, https://doi.org/10.1175/MWR-D-12-00013.1.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Heng, B. C. P., and Coauthors, 2020: SINGV-DA: A data assimilation system for convective-scale numerical weather prediction over Singapore. Quart. J. Roy. Meteor. Soc., 146, 19231938, https://doi.org/10.1002/qj.3774.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Hohenegger, C., and C. Schar, 2007: Atmospheric predictability at synoptic versus cloud-resolving scales. Bull. Amer. Meteor. Soc., 88, 17831794, https://doi.org/10.1175/BAMS-88-11-1783.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Holloway, C. E., S. J. Woolnough, and G. M. S. Lister, 2013: The effects of explicit versus parameterized convection on the MJO in a large-domain high-resolution tropical case study. Part I: Characterization of large-scale organization and propagation. J. Atmos. Sci., 70, 13421369, https://doi.org/10.1175/JAS-D-12-0227.1.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Huffman, G. J., and Coauthors, 2019: NASA Global Precipitation Measurement (GPM) Integrated Multi-satellitE Retrievals for GPM (IMERG). Algorithm Theoretical Basis Doc., version 06, 38 pp., https://gpm.nasa.gov/sites/default/files/document_files/IMERG_ATBD_V06.pdf.

  • Johnson, S. J., and Coauthors, 2016: The resolution sensitivity of the South Asian monsoon and Indo-Pacific in a global 0.35° AGCM. Climate Dyn., 46, 807831, https://doi.org/10.1007/s00382-015-2614-1.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Kain, J. S., and Coauthors, 2008: Some practical considerations regarding horizontal resolution in the first generation of operational convection-allowing NWP. Wea. Forecasting, 23, 931952, https://doi.org/10.1175/WAF2007106.1.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Kendon, E. J., N. M. Roberts, C. A. Senior, and M. J. Roberts, 2012: Realism of rainfall in a very high-resolution regional climate model. J. Climate, 25, 57915806, https://doi.org/10.1175/JCLI-D-11-00562.1.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Khan, S., and V. Maggioni, 2019: Assessment of level-3 gridded Global Precipitation Mission (GPM) products over oceans. Remote Sens., 11, 255, https://doi.org/10.3390/rs11030255.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Lestari, S., A. King, C. Vincent, D. Karoly, and A. Protat, 2019: Seasonal dependence of rainfall extremes in and around Jakarta, Indonesia. Wea. Climate Extremes, 24, 100202, https://doi.org/10.1016/j.wace.2019.100202.

    • Search Google Scholar
    • Export Citation
  • Lim, S. Y., C. Marzin, P. Xavier, C. P. Chang, B. Timbal, S. Yee Lim, C. Marzin, and P. Xavier, 2017: Impacts of boreal winter monsoon cold surges and the interaction with MJO on Southeast Asia rainfall. J. Climate, 30, 42674281, https://doi.org/10.1175/JCLI-D-16-0546.1.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Lin, J. L., and Coauthors, 2006: Tropical intraseasonal variability in 14 IPCC AR4 climate models. Part I: Convective signals. J. Climate, 19, 26652690, https://doi.org/10.1175/JCLI3735.1.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Loken, E. D., A. J. Clark, M. Xue, and F. Kong, 2017: Comparison of next-day probabilistic severe weather forecasts from coarse- and fine-resolution CAMs and a convection-allowing ensemble. Wea. Forecasting, 32, 14031421, https://doi.org/10.1175/WAF-D-16-0200.1.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Love, B. S., A. J. Matthews, and G. M. S. Lister, 2011: The diurnal cycle of precipitation over the Maritime Continent in a high-resolution atmospheric model. Quart. J. Roy. Meteor. Soc., 137, 934947, https://doi.org/10.1002/qj.809.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Matsumoto, J., and Coauthors, 2017: An overview of the Asian Monsoon Years 2007–2012 (AMY) and multi-scale interactions in the extreme rainfall events over the Indonesian Maritime Continent. The Global Monsoon System: Research and Forecast, C.-P. Chang et al., Eds., World Scientific Series on Asia-Pacific Weather and Climate, Vol. 9, World Scientific, 365–385.

    • Crossref
    • Export Citation
  • Met Office, 2018: Iris: A Python library for analysing and visualising meteorological and oceanographic data sets, v.2.0. Accessed 5 October 2018, https://scitools.org.uk/.

  • Mittermaier, M., N. Roberts, and S. A. Thompson, 2013: A long-term assessment of precipitation forecast skill using the fractions skill score. Meteor. Appl., 20, 176186, https://doi.org/10.1002/met.296.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Miura, H., M. Satoh, T. Nasuno, A. T. Noda, and K. Oouchi, 2007: A Madden-Julian oscillation event realistically simulated by a global cloud-resolving model. Science, 318, 17631765, https://doi.org/10.1126/science.1148443.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Mohd Nor, M. F. F., C. E. Holloway, and P. M. Inness, 2020: The role of local orography on the development of a severe rainfall event over western peninsular Malaysia: A case study. Mon. Wea. Rev., 148, 21912209, https://doi.org/10.1175/MWR-D-18-0413.1.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Mori, S., H. Jun-Ichi, M. D. Yamanaka, N. Okamoto, F. Murata, N. Sakurai, and H. Hashiguchi, 2004: Diurnal land–sea rainfall peak migration over Sumatera Island, Indonesian Maritime Continent, observed by TRMM satellite and intensive rawinsonde soundings. Mon. Wea. Rev., 132, 20212039, https://doi.org/10.1175/1520-0493(2004)132<2021:DLRPMO>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Neale, R., and J. Slingo, 2003: The Maritime Continent and its role in the global climate: A GCM study. Mon. Wea. Rev., 16, 834848, https://doi.org/10.1175/1520-0442(2003)016<0834:TMCAIR>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Pearson, K. J., R. J. Hogan, R. P. Allan, G. M. S. Lister, and C. E. Holloway, 2010: Evaluation of the model representation of the evolution of convective systems using satellite observations of outgoing longwave radiation. J. Geophys. Res., 115, D20206, https://doi.org/10.1029/2010JD014265.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Pearson, K. J., G. M. S. Lister, C. E. Birch, R. P. Allan, R. J. Hogan, and S. J. Woolnough, 2014: Modelling the diurnal cycle of tropical convection across the ‘grey zone.’ Quart. J. Roy. Meteor. Soc., 140, 491499, https://doi.org/10.1002/qj.2145.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Peatman, S. C., A. J. Matthews, and D. P. Stevens, 2014: Propagation of the Madden–Julian oscillation through the Maritime Continent and scale interaction with the diurnal cycle of precipitation. Quart. J. Roy. Meteor. Soc., 140, 814825, https://doi.org/10.1002/qj.2161.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Peatman, S. C., J. Methven, and S. J. Woolnough, 2015: Propagation of the Madden–Julian oscillation and scale interaction with the diurnal cycle in a high-resolution GCM. Climate Dyn., 45, 29012918, https://doi.org/10.1007/s00382-015-2513-5.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Peatman, S. C., J. Methven, and S. J. Woolnough, 2018: Isolating the effects of moisture entrainment on convectively coupled equatorial waves in an aquaplanet GCM. J. Atmos. Sci., 75, 31393157, https://doi.org/10.1175/JAS-D-18-0098.1.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Porson, A. N., S. Hagelin, D. F. A. Boyd, N. M. Roberts, R. North, S. Webster, and J. C. Lo, 2019: Extreme rainfall sensitivity in convective-scale ensemble modelling over Singapore. Quart. J. Roy. Meteor. Soc., 145, 30043022, https://doi.org/10.1002/qj.3601.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Porson, A. N., and Coauthors, 2020: Recent upgrades to the Met Office convective-scale ensemble: An hourly time-lagged 5-day ensemble. Quart. J. Roy. Meteor. Soc., 146, 32453265, https://doi.org/10.1002/qj.3844.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Qian, J. H., 2008: Why precipitation is mostly concentrated over islands in the Maritime Continent. J. Atmos. Sci., 65, 14281441, https://doi.org/10.1175/2007JAS2422.1.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Raynaud, L., and F. Bouttier, 2017: The impact of horizontal resolution and ensemble size for convective-scale probabilistic forecasts. Quart. J. Roy. Meteor. Soc., 143, 30373047, https://doi.org/10.1002/qj.3159.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Rezacova, D., P. Zacharov, and Z. Sokol, 2009: Uncertainty in the area-related QPF for heavy convective precipitation. Atmos. Res., 93, 238246, https://doi.org/10.1016/j.atmosres.2008.12.005.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Roberts, B., I. L. Jirak, A. J. Clark, S. J. Weiss, and J. S. Kain, 2019: Postprocessing and visualization techniques for convection-allowing ensembles. Bull. Amer. Meteor. Soc., 100, 12451258, https://doi.org/10.1175/BAMS-D-18-0041.1.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Roberts, N. M., and H. W. Lean, 2008: Scale-selective verification of rainfall accumulations from high-resolution forecasts of convective events. Mon. Wea. Rev., 136, 7897, https://doi.org/10.1175/2007MWR2123.1.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Sato, T., H. Miura, M. Satoh, Y. N. Takayabu, and Y. Wang, 2009: Diurnal cycle of precipitation in the tropics simulated in a global cloud-resolving model. J. Climate, 22, 48094826, https://doi.org/10.1175/2009JCLI2890.1.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Schwartz, C. S., and R. A. Sobash, 2019: Revisiting sensitivity to horizontal grid spacing in convection-allowing models over the central and eastern United States. Mon. Wea. Rev., 147, 44114435, https://doi.org/10.1175/MWR-D-19-0115.1.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Schwartz, C. S., and Coauthors, 2009: Next-day convection-allowing WRF model guidance: A second look at 2-km versus 4-km grid spacing. Mon. Wea. Rev., 137, 33513372, https://doi.org/10.1175/2009MWR2924.1.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Schwartz, C. S., G. S. Romine, K. R. Smith, and M. L. Weisman, 2014: Characterizing and optimizing precipitation forecasts from a convection-permitting ensemble initialized by a mesoscale ensemble Kalman filter. Wea. Forecasting, 29, 12951318, https://doi.org/10.1175/WAF-D-13-00145.1.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Schwartz, C. S., G. S. Romine, R. A. Sobash, K. R. Fossell, and M. L. Weisman, 2015: NCAR’s experimental real-time convection-allowing ensemble prediction system. Wea. Forecasting, 30, 16451654, https://doi.org/10.1175/WAF-D-15-0103.1.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Schwartz, C. S., G. S. Romine, K. R. Fossell, A. Sobash, and M. L. Weisman, 2017: Toward 1-km ensemble forecasts over large domains. Mon. Wea. Rev., 145, 29432969, https://doi.org/10.1175/MWR-D-16-0410.1.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Stensrud, D. J., J. Bao, and T. T. Warner, 2000: Using initial condition and model physics perturbations in short-range ensemble simulations of mesoscale convective systems. Mon. Wea. Rev., 128, 20772107, https://doi.org/10.1175/1520-0493(2000)128<2077:UICAMP>2.0.CO;2.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Sun, X., and Coauthors, 2020: A subjective and objective evaluation of model forecasts of Sumatra squall events. Wea. Forecasting, 35, 489506, https://doi.org/10.1175/WAF-D-19-0187.1.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Sunilkumar, K., A. Yatagai, and M. Masuda, 2019: Preliminary evaluation of GPM-IMERG rainfall estimates over three distinct climate zones with APHRODITE. Earth Space Sci., 6, 1321–1335, https://doi.org/10.1029/2018EA000503.

    • Crossref
    • Export Citation
  • Supari, F., E. Salimun, E. Aldrian, A. Sopaheluwakan, and L. Juneng, 2018: ENSO modulation of seasonal rainfall and extremes in Indonesia. Climate Dyn., 51, 25592580, https://doi.org/10.1007/s00382-017-4028-8.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Takahashi, H. G., and T. Yasunari, 2008: Decreasing trend in rainfall over Indochina during the late summer monsoon: Impact of tropical cyclones. J. Meteor. Soc. Japan, 86, 429438, https://doi.org/10.2151/jmsj.86.429.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Tan, J., G. J. Huffman, D. T. Bolvin, and E. J. Nelkin, 2019: Diurnal cycle of IMERG V06 precipitation. Geophys. Res. Lett., 46, 13 58413 592, https://doi.org/10.1029/2019GL085395.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Tan, M. L., and Z. Duan, 2017: Assessment of GPM and TRMM precipitation products over Singapore. Remote Sens., 9, 720, https://doi.org/10.3390/rs9070720.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Tan, M. L., and H. Santo, 2018: Comparison of GPM IMERG, TMPA 3B42 and PERSIANN-CDR satellite precipitation products over Malaysia. Atmos. Res., 202, 6376, https://doi.org/10.1016/j.atmosres.2017.11.006.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Villafuerte, M. Q., and J. Matsumoto, 2015: Significant influences of global mean temperature and ENSO on extreme rainfall in Southeast Asia. J. Climate, 28, 19051919, https://doi.org/10.1175/JCLI-D-14-00531.1.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Vincent, C. L., and T. P. Lane, 2018: Mesoscale variation in diabatic heating around Sumatra, and its modulation with the Madden–Julian oscillation. Mon. Wea. Rev., 146, 25992614, https://doi.org/10.1175/MWR-D-17-0392.1.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Vogel, P., P. Knippertz, A. H. Fink, A. Schlueter, and T. Gneiting, 2018: Skill of global raw and postprocessed ensemble predictions of rainfall over northern tropical Africa. Wea. Forecasting, 33, 369388, https://doi.org/10.1175/WAF-D-17-0127.1.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Walters, D., and Coauthors, 2017: The Met Office Unified Model Global Atmosphere 6.0/6.1 and JULES Global Land 6.0/6.1 configurations. Geosci. Model Dev., 10, 14871520, https://doi.org/10.5194/gmd-10-1487-2017.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Wang, J., J. Chen, J. Du, Y. Zhang, Y. Xia, and G. Deng, 2018: Sensitivity of ensemble forecast verification to model bias. Mon. Wea. Rev., 146, 781796, https://doi.org/10.1175/MWR-D-17-0223.1.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Weller, E., K. Shelton, M. Reeder, and C. Jakob, 2017: Precipitation associated with convergence lines. J. Climate, 30, 31693183, https://doi.org/10.1175/JCLI-D-16-0535.1.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Wilson, D. R., A. C. Bushell, A. M. Kerr-Munslow, J. D. Price, and C. J. Morcrette, 2008: PC2: A prognostic cloud fraction and condensation scheme. I: Scheme description. Quart. J. Roy. Meteor. Soc., 134, 20932107, https://doi.org/10.1002/qj.333.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Woodhams, B. J., C. E. Birch, J. H. Marsham, C. L. Bain, N. M. Roberts, and D. F. A. Boyd, 2018: What is the added value of a convection-permitting model for forecasting extreme rainfall over tropical East Africa? Mon. Wea. Rev., 146, 27572780, https://doi.org/10.1175/MWR-D-17-0396.1.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Wu, P., M. Hara, J. Hamada, M. D. Yamanaka, and F. Kimura, 2009: Why a large amount of rain falls over the sea in the vicinity of western Sumatra Island during nighttime. J. Appl. Meteor. Climatol., 48, 13451361, https://doi.org/10.1175/2009JAMC2052.1.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Wu, P., A. A. Arbain, S. Mori, J. Hamada, M. Hattori, F. Syamsudin, and M. D. Yamanaka, 2013: The effects of an active phase of the Madden–Julian oscillation on the extreme precipitation event over western Java island in January 2013. SOLA, 9, 7983, https://doi.org/10.2151/sola.2013-018.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Xavier, P., R. Rahmat, W. K. Cheong, and E. Wallace, 2014: Influence of Madden–Julian oscillation on Southeast Asia rainfall extremes: Observations and predictability. Geophys. Res. Lett., 41, 44064412, https://doi.org/10.1002/2014GL060241.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Yamanaka, M. D., 2016: Physical climatology of Indonesian Maritime Continent: An outline to comprehend observational studies. Atmos. Res., 178–179, 231259, https://doi.org/10.1016/j.atmosres.2016.03.017.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Yang, G.-Y., and J. Slingo, 2001: The diurnal cycle in the tropics. Mon. Wea. Rev., 129, 784801, https://doi.org/10.1175/1520-0493(2001)129<0784:TDCITT>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
Save