Prediction of 24-hour-average PM2.5 concentrations using a hidden Markov model with different emission distributions in Northern California

► A hidden Markov model with different non-Gaussian distributions is developed to match data characteristics. ► The method is applied to the prediction of PM2.5 exceedance days in Concord, CA and Sacramento, CA. ► Results show that the HMM can predict most exceedances correctly and reduce false alarms significantly.