<![CDATA[This study aimed to develop a temporal prediction model for atmospheric electric field variations using Long Short-Term Memory (LSTM) based on Electric Field Mill (EFM) data. Atmospheric electric field variation is an important indicator of charge accumulation in clouds and can support early warning systems for lightning potential. This study employed a descriptive quantitative approach using time-series data acquired from an ESP32-S3-based EFM system. The collected data included time, ADC value, raw voltage, filtered voltage, electric field value in kV/m, atmospheric status, and lightning potential. Before model training, the data were processed through low-pass filtering, Min-Max normalization, and sliding-window sequence formation with a window size of 60 observations. A Bidirectional LSTM architecture was applied for regression and classification tasks. The regression model produced an MSE of 0.558, RMSE of 0.747, and MAE of 0.580. Meanwhile, the classification model achieved an accuracy of 83.1% for atmospheric status classification and 99.20% for weather condition classification. These findings indicate that the proposed LSTM model can effectively learn temporal patterns in atmospheric electric field data and has potential to support the development of an early lightning warning system.]]>
Copyrights © 2026
