Generating Synthetic Values on Automatic Weather Station Dataset using GAIN

The Automatic Weather Station (AWS) is an automated instrument designed to gather meteorological parameters within a given environment. The dataset obtained from AWS often contains missing values due to various factors such as sensor malfunction or transmission issues. Nonetheless, the data produced by AWS plays a vital role in weather analysis and prediction. In this study, the authors employ the Generative Adversarial Imputation Network algorithm to generate synthetic values for the missing data in the AWS dataset. The dataset is divided into four schemes, each exhibiting different distributions of missing values at 5, 25, 50, and 80 gaps, respectively. The synthetic values generated by the algorithm are subsequently compared to the ground truth values using the Root Mean Square Error (RMSE) and Mean Absolute Error (MAE) metrics. The findings demonstrate that for the air pressure parameter, both the 5 gaps and 25 gaps models exhibit RMSE and MAE values below 2 hPa. Similarly, the models for the air temperature parameter demonstrate RMSE and MAE values below 2°C. As for the relative humidity parameter, the models achieve RMSE and MAE values below 15%. Therefore, both the 5 gaps and 25 gaps models satisfy the error threshold established by the Badan Meteorologi Klimatologi, dan Geofisika for AWS data quality control.