Temporal Prediction Model for CO and CO2 Pollutants using Long Short Term Memory

Forest and land fires in Indonesia cause smoke haze pollution in the Southeast Asian region. This situation has resulted in the air quality around the forest and peatland fires location being tripled from the hazardous to health to exceed 1000 in the Pollutant Standard Index (PSI). This study aims to create a predictive model of CO and CO2 concentration data from forest and land fires using the Long Short Term Memory method. The data used in this study are the HYSPLIT simulation results with data input are Sumatra’s hotspot data in 2019, and meteorological data in 2019. This research was carried out through several stages, namely data collection, selection of hotspots, dispersion simulations, merging data, data processing, data sharing, creating prediction models for CO and CO2 pollutant concentrations, and evaluation of the model with RMSE and R. This study produces a predictive model of CO pollutants that had the best R and RMSE values of 0.156977610 and 2.882038×10⁻⁸ respectively. Simultaneously, the CO2 pollutant prediction model has the best R and RMSE values of 0.6164270 and 1.815969×10⁻⁶, respectively.