Spatio-Temporal PM2.5 Forecasting in Thailand Using Encoder-Decoder Networks

PM2.5 is a type of particulate matter that contributes to air pollution in Thailand on yearly cycle. Exposure can cause acute health problems, including respiratory and cardiovascular diseases, as well an increased risk premature death. In this paper, we present spatio-temporal model based deep learning approach for concentration prediction via image-like at country-wide level. Our model: SimVP-CFLL-ML video model, called "SimVP". To enhance its performance when attempting predict high concentration, SimVP includes two major improvements i.e. cross-feature layer (CFLL) using 1x1 convolution learn feature correlation masking (ML) calculate loss specific locations. The experiment conducted data collected from the control department (PCD) sensor all (SFA). Results show our outperforms baselines. model’s F1 perforance 3.51% better than best baseline classifying class.