MACHINE LEARNING BASED CUMULONIMBUS CLOUD PREDICTION WITH RADIOSONDE DATA APPROACH AT EL TARI KUPANG INTERNATIONAL AIRPORT

• Main Author: Dwi Putra, Novianto
• Format: Final Project
• Language: Indonesia
• Subjects: Geologi, hidrologi & meteorologi
• Online Access: https://digilib.itb.ac.id/gdl/view/75622
• Institution: Institut Teknologi Bandung

Cumulonimbus cloud is one of the most dangerous convective clouds for aviation. Heavy rain, tornadoes and turbulence can occur due to the presence of Cumulonimbus clouds. Upper air conditions have an influential impact on cloud formation. Radiosonde observations can be used to predict the presence of Cb clouds in the short term. This study aims to predict the existence of Cumulonimbus (Cb) clouds using 5 radiosonde data indexes Showalter Index (SI), Lifted Index (LI), K Index, Total–Totals and CAPE based on a Machine Learning approach. In this study, indexes originating from upper air observations were processed by normalizing the data and then the radiosonde index–based prediction model data was trained using Machine Learning to predict the presence of Cumulonimbus clouds, after the prediction model was built, the output results were displayed in the dichotomy table and then assessed for accuracy, detection and resulting errors. Based on data processing, the prediction of the presence of Cb clouds using radiosonde indices is quite good when implemented on test data. Effect of Convective Available Potential Energy (CAPE), Machine Learning can predict the presence of Cb clouds up to 97%. When the model does not use CAPE, the model can predict only up to 95%. Training and testing schemes based on wet and dry months provide better Probability of Detection (POD) results of 98% in wet months with CAPE and 90% without CAPE. The addition of CAPE can also reduce the False Alarm Rate, the lowest error rate is 4% when adding CAPE and 10% when without CAPE. It can be concluded that predictions of Cumulonimbus clouds using radiosonde data based on Machine Learning are quite reliable in use.