<![CDATA[At the BMKG (Indonesia's Meteorology, Climatology and Geophysics Agency) station in Tarakan City, manual recording and analysis of radiosonde observations pose challenges to efficiency and accuracy. The workflow includes automatic processing and classification of data. We designed a tool called the Radiosonde Data Recap and Analysis System (RASON) using the Random Forest model. This tool was designed in Python using PyQt5 (Python’s GUI framework), enabling data import, processing, and visualization. For the Random Forest classifier, the historical radiosonde datasets (844 soundings after cleaning) were used to train and validate between 2022 and 2023, and the independent dataset (733 soundings) from 2024 was used to evaluate out-of-sample performance. While it took more than 10 minutes to manually add the point data, its trained sample took up 20 points in 8 seconds. On several significant atmospheric stability metrics, including K Index (KI), Lifted Index (LI), Showalter Index (SI), Total Totals (TT), and Convective Available Potential Energy (CAPE), the Random Forest classification model presents an almost ideal classification score. Evaluation verified all performance metrics on each index as given; for instance, Precision, Recall, and F1-Score of 1.0000 were attained in each metric KI_Class and TT_Class, LI_Class and CAPE_Class achieved better than 0.99, SI_Class had an F1-Score of 0.9267 due to low precision compared to other indices. Each of the five indexes was achieved via overall classification (AllData_Class) using majority voting, producing Precision=0.9030, Recall=0.9877, and F1-Score=0.9369. An average score of 90 was obtained on the System Usability Scale (SUS), indicating the highest level of user satisfaction and usability. This is how it enables the processing of increasing amounts of radiosonde data to enhance decision-making for effective weather forecasting at BMKG Tarakan City.]]>
Copyrights © 2026
