IMPROVING NUMERICAL WEATHER PREDICTION OF VERY HEAVY RAINFALL EVENTS AND THEIR ASSOCIATED CLOUD SYSTEMS OVER JAKARTA AREA USING C-BAND RADAR DATA ASSIMILATION

• Main Author: GUSTARI (NIM: 32409010); Tim Pembimbing : Prof. Safwan Hadi, Ph.D; Dr. Tri Wahyu Hadi, M.Sc; , INDRA
• Format: Dissertations
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/18704
• Institution: Institut Teknologi Bandung

Very heavy rainfall can be considered as strong weather disturbance and its occurrences often trigger severe floodings in the Greater Jakarta (Jabodetabek) area. Thus, accurate prediction is crucial to mitigate the negative impact of the meteorological events. While high resolution Numerical Weather Prediction (NWP) has become world standard method for high-impact and severe weather <br /> <br /> <br /> forecasting at regional and local scale, its application in the tropical region does not immediately improve forecast skill as expected. Results of several hindcast <br /> <br /> <br /> experiments using NWP in previous studies show significant discrepancies, especially in terms of phase, between predicted and observed cloud convection and rainfall events. Such discrepancies had likely occurred due to several determinants: (i)unsuitability of convective parameterizations, (ii)inappropriate initialization scheme, and (iii)insufficiency in data assimilation. <br /> <br /> <br /> The main objective of this research is to obtain a prototype of more accurate prediction system for forecasting very heavy rainfall events over Jakarta <br /> <br /> <br /> area with hourly time resolution. Therefore, accurate prediction of the phase of cloud convection and rainfall events is crucial. In this context, the overall improvement of the prediction was the main target of C-band Doppler radar data assimilation combined with more suitable convective parameterizations and initialization scheme. <br /> <br /> <br /> In order to understand the dynamical aspects of the heavy rainfall events, analysis of C-band radar data has also been performed. Cloud system from two cases of very heavy rainfall recorded on January 18, 2010 and February 13, 2010 have been investigated. Radar images of the two cases revealed that the former case is characterized by west-to-east (zonal) cloud movement with a parallel <br /> <br /> <br /> stratiform type, while the second case showed a north-to-south (meridional) cloud movement with a leading stratiform type. These results indicate that heavy rainfall events are associated with the development of mesoscale cloud system with rapidly changing formation, instead of large and sustained rainbands. <br /> <br /> <br /> A prototype of high resolution NWP system has been developed using Weather Research and Forecasting (WRF) model for downscaling of Global Forecast System (GFS) output. The model has been setup for fine horizontal <br /> <br /> <br /> resolution of 3 x 3 km over Jabodetabek area with three-level nesting of model domains (two outer domains used 9 km x 9 km and 27 km x 27 km grid spacing, respectively). Hindcast experiments for the aforementioned two cases were first conducted to obtain more detailed model setup. From a set of 30 experiments, it is found that the combination of Kain Fritsch convective parameterization and Lin microphysics scheme produced the most consistent results. Results of downscaling experiments (without data assimilation) using the selected model setup show that the high resolution NWP tends to produce false alarm of heavy rainfall events. This indicates that the phase error in numerical prediction was not remedied by better selection of convective parameterization. An attempt to improve the prediction system was then made by incorporating C-band Doppler radar data assimilation using 3DVar technique. <br /> <br /> <br /> In the implementation of radar data assimilation, the impact of cold and warm start initialization scheme was also investigated. It is found that warm start produced higher skill scores than cold start scheme. The total impacts of selected convective parametrization, warm start initialization, and radar data assimilation, are then evaluated using hindcast experiments involving the aforementioned two cases of heavy rainfall event. Results of the experiments are verified against station rainfall data. It is found that rainfall forecast with data assimilation showed improvement over those without data assimilation. This can particularly be seen in the use of radial velocity data for concentrated rain with strong zonal winds in the north coast on January 18, 2010. However, the positive impact of Doppler radial velocity assimilation using 3DVar techniques lasts only three to six hours after the initial time of prediction indicating that wind fields had rapidly changed along <br /> <br /> <br /> with the evolution of the cloud system. In order to improve the effectiveness of 3DVar data assimilation technique, a Rapid Update Cycle (RUC) procedure must be used in the prediction system. The application of RUC technique produced rainfall forecast with reduced phase discrepancies from observed data.