Satellite data for upscalling urban air pollution in Malaysia

Abstract

Air pollution has been recognised as one of risk factors that exert adverse effect on human health, climate change and environmental deterioration. Since particulate matter less than 10μm (PM10) is used as an important air quality indicator in Malaysia, it is crucial to map the PM10 spatial distributions especially over Malaysian cities which encounter higher pollutant concentration due to trans-boundary and local sources. PM10 has been widely estimated using satellite data but the coarse resolution of 10km tends to average out the spatial variation especially in cities. Therefore, in this study we tested the Aerosol Optical Depth (AOD550) product from Moderate Resolution Imaging Spectrometer (MODIS) sensor at 3km to estimate PM10 concentration over Peninsular Malaysia. The performance of AOD550 at 3km product was validated with AOD retrieved from AERONET stations and an accuracy of R2 = 0.58 and RMSE = 0.13 was obtained. PM10 was estimated over the cities for the period 2007-2011 using MODIS AOD550 and meteorological variables (surface temperature, relative humidity, atmospheric stability, wind speed, wind direction). In consideration of their complicated relationship and non-linear mechanism that may exist between the variables artificial neural network (ANN) technique was utilized to develop an empirical model to estimate PM10 concentrations. Result shows that the empirical model developed using ANN was moderately robust with R2 of 0.41 and RMSE = 12.99 μgm-3. However, PM10 estimated with the model over the cities was validated using an independent set of field data produced an acceptable accuracy with R2 = 0.39 and RMSE = 10.95 μgm-3. The inclusion of meteorological parameters improved the prediction and the result obtained allow us to map pollution levels in Malaysia.