COMPARISON OF HIMAWARI-8 SATELLITE RADIANCE DATA ASSIMILATION METHODS FOR RAINFALL PREDICTION IN EAST KALIMANTAN (CASE STUDY OF VERY HEAVY RAINFALL 2 – 4 JUNE 2019)

• Main Author: Abshor Mukhsinin, Huda
• Format: Theses
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/84286
• Institution: Institut Teknologi Bandung

Accurate weather prediction, particularly for very heavy rainfall events, is crucial for mitigating hydrometeorological disasters. However, the prediction of very heavy rainfall events remains challenging, as is the case in East Kalimantan, a region in the Indonesian Maritime Continent (IMC) with unique and complex weather system characteristics. Researchers have explored the use of numerical weather prediction models, ranging from regional to convective scale models (Convection-Permitting Models), but there is still a bias caused by initial conditions that do not accurately represent the real-time weather conditions. To address this, data assimilation is conducted to improve the initial conditions and generate more accurate simulations. The Himawari-8 satellite, equipped with the Advanced Himawari Imager (AHI) sensor, has the advantage of producing radiance data that capture atmospheric phenomena with high spatial and temporal resolution. Previous studies have shown that assimilating AHI radians data can enhance the accuracy of initial conditions, particularly for convection initiation, such as in the intensification of tropical cyclones and the formation of isolated convective cloud cells. However, research on AHI radiance data assimilation for regions with complex convection-dominated systems in the IMC is still limited and not thoroughly explored. In addition, data assimilation techniques significantly enhance the precision of initial conditions. While the Three-Dimensional Variational (3DVAR) method is favored for its computational efficiency, its reliance on homogeneous and isotropic background error covariance limits its ability to factor in actual flow-dependent. Conversely, the Ensemble Kalman Filter (EnKF) method, although accounting for flow-dependent, is hampered by a small ensemble size, potentially leading to spurious correlations. In response to this challenge, hybrid ensemble-variational (EnVar) approaches such as 3D-Ensemble-Variational (3DenVar) combine the strengths of both techniques, effectively overcoming the limitations of ensemble and static covariance. Consequently, this study seeks to evaluate and compare the impact of different assimilation methods on the enhancement of model prediction skills in the case of very heavy rainfall in East Kalimantan at a convective scale. iv Four schemes were designed on the Weather Research and Forecasting (WRF) model at 3 km resolution (Convection-permitting) comparing a scheme without assimilation (NODA) and three schemes with different assimilation methods. The three schemes are 3DVAR (model with 3DVAR assimilation method), HYBRID (model with 3DEnVar assimilation method), and DUALRES (model with 3DEnVardual resolution assimilation method). Assimilation was performed every 1 hour in the first 7 hours of model initiation time for a case study of very heavy rainfall in East Kalimantan on June 2 - 4, 2019. The analysis was conducted by comparing the prediction results on convection evolution, moisture flux enhancement, upper air profile improvement, and rainfall prediction skill in the metrics of probability of prediction (POD), threat score (TS), fraction skill score (FSS), and relative operating characteristic (ROC) curve along with area under the ROC curve (AUC). The results showed that the assimilation of AHI radians data on convectionpermitting models with hybrid 3DEnVar and 3DVAR techniques significantly improved the prediction skill of very heavy rain in East Kalimantan. The HYBRID and DUALRES schemes show a more significant improvement compared to the model without assimilation (NODA) and the 3DVAR scheme. The analysis results show that the assimilation of AHI data affects prognostic variables such as perturbation potential temperature, perturbation moist potential temperature, and water vapor mixing ratio significantly, resulting in larger and more sensitive changes in the initial conditions. DUALRES scheme excels in reducing biases in various atmospheric variables and improving evaluation metrics such as POD, TS, FSS, and AUC on the ROC curve, showing excellent rainfall event prediction capabilities. Overall, the assimilation of AHI radiation data with the 3DEnVar method improves the prediction sensitivity to complex atmospheric dynamics, produces more accurate rainfall distribution and intensity, and improves the diurnal pattern of rainfall in East Kalimantan