Synergy of satellite and ground based observations in estimation of particulate matter in eastern China

Estimating particulate matter (PM) from space is not straightforward and is mainly achieved using the aerosol optical depth (ADD) retrieved from satellite sensors. However. AOD is a columnar measure, whereas PM is a ground observation. Linking AOD and PM remains a challenge for air pollution monitoring. In this study, a back-propagation artificial neural network (BP ANN) algorithm trained with Bayesian regularization that benefited from the synergy of satellite- and ground-based observations was developed to estimate PM in eastern China. Correlations between observed and estimated PM (denoted by R) during the period 2007-2008 over seven individual sites were investigated comprehensively in terms of site scale, seasonal scale, particle size, and spatio-temporal scale. With respect to site differences, the Nanning site had the best results with 80.3% of cases having a moderate or strong correlation value. Lushan and Zhengzhou followed with results of 75% and 73.8%, respectively. Furthermore, R exhibited a significant seasonal variation characterized by a maximum (80.2%) during the autumn period, whereas no obvious differences in R for various spatial scales (spatial averaging schemes of MODIS AOD) were observed. Likewise, the ratio value for daily averaging (64.7%) was found to be better than those for the two hourly temporal averaging schemes (i.e., 61.1% for HA1 and 58.3% for HA2). In addition, PM, estimated from the ANN algorithm developed in this study had slightly higher R values than did PM10 and PM2.5. The planetary boundary layer (PBL) effect on PM estimation was decreasing R with increasing height of the PBL, which is consistent with previous studies. Comparisons of observed versus estimated PM10 mass time series implied that the ANN algorithm basically reproduced the observed PM concentration. However. PM mass at certain sites may be underestimated under the condition of high observed PM concentrations. (C) 2012 Elsevier B.V. All rights reserved.