Numerical weather prediction (NWP) capabilities in the Maritime Continent are not as developed as in the midlatitudes. Countries in the region do not develop their own modelling systems; rather they adapt models primarily developed for the midlatitudes. Due to the complexity of processes involved in the region, this adaptation is non‐trivial. In this article the developments made by the Meteorological Service Singapore (MSS) and the United Kingdom Met Office (UKMO) to implement a convective‐scale NWP system for short‐range weather prediction for Singapore and the surrounding regions are presented. In particular, this article describes the changes to the initial model configuration, which was based on the UKMO's convective‐scale NWP system (the UKV), to produce operational forecasts over this region. Results presented here demonstrate the benefit of convection‐permitting simulations over convection parametrized simulations and show that the model performance is greatly affected by the choice of driving model, the cloud scheme, and the turbulence scheme.

Extreme rainfall is one of the primary meteorological hazards in Singapore, as well as elsewhere in the deep tropics, and it can lead to significant local flooding. Since 2013, the Meteorological Service Singapore (MSS) and the United Kingdom Met Office (UKMO) have been collaborating to develop a convective-scale Numerical Weather Prediction (NWP) system, called SINGV. Its primary aim is to provide improved weather forecasts for Singapore and the surrounding region, with a focus on improved short-range prediction of localized heavy rainfall. This paper provides an overview of the SINGV development, the latest NWP capabilities at MSS and some key results of evaluation. The paper describes science advances relevant to the development of any km-scale NWP suitable for the deep tropics and provides some insights into the impact of local data assimilation and utility of ensemble predictions.

To improve the wind speed forecasts at turbine locations and at hub height, this study develops the WRFDA system to assimilate the wind speed observations measured on the nacelle of turbines (hereafter referred as turbine wind speed observations) with both 3DVAR and 4DVAR algorithms. Results exhibit that the developed data assimilation (DA) system helps in greatly improving the analysis and the forecast of wind turbine speed. Among three experiments with no cycling DA, with 2-h cycling DA, and with 4-h cycling DA, the last experiment generates the best analysis, improving the averaged forecasts (from T + 9 to T + 24) of wind speed over all wind farms by 32.5% in the bias and 6.3% in the RMSE. After processing the turbine wind speed observations into superobs, even bigger improvements are revealed when validating against either the original turbine wind speed observations or the superobs. Taken the results validated against the superobs as an example, the bias and RMSE of the forecasts (from T + 9 to T + 24) averaged over all wind farms are reduced by 38.8% and 12.0%, respectively. Compared to the best-performed 3DVAR experiment (4-h cycling and superobs), the experiment following the same DA strategy but using 4DVAR algorithm exhibits further improvements, especially for the averaged bias in the forecasts of all wind farms, and the changing amount in the forecasts of the enhanced wind farms. Compared to the control experiment, the 4DVAR experiment reduces the bias and RMSE in the forecasts (from T + 9 to T + 24) by 54.6% (0.66 m s−1) and 12.7% (0.34 m s−1).

Sumatra squalls are important rain-bearing weather systems that affect Singapore and southern Peninsular Malaysia. The performance of forecasts for 63 past squall events is evaluated using a subjective evaluation by forecasters and an objective evaluation based on the fractions skill score (FSS). The purpose of this study is to investigate whether an objective procedure can reproduce the main results of the subjective evaluation. A convection permitting version of the Met Office (UKMO) Unified Model (UM), configured for a limited domain in the southern region of the South China Sea, is used with two driving global deterministic models: the UM and the European Centre for Medium-Range Weather Forecasts (ECMWF) model. Subjective and objective evaluation scoring methods for the two limited-area forecasts of the UM are compared, and it is shown that the objective procedure can reasonably emulate the scores produced by the forecasters in the context of parameters that are of direct relevance to the forecast process. This indicates that automated objective verification methods may be a reasonable alternative to resource intensive subjective evaluations for some cases. The robustness of the objective results is investigated using 7 months of data, and issues of statistical significance are considered.