Remote sensing of aerosols over North American land surfaces from POLDER and MODIS measurements

Abstract

We examine two different aerosol remote sensing approaches based on polarized ADEOS-1 POLDER measurements and on multispectral EOS Terra MODIS dark target retrieval. Satellite-derived aerosol optical depth (AOD) is assessed from comparisons with AERONET/AEROCAN ground-based sunphotometer AOD measurements over North America. The results show that the POLDER polarization retrieval method for AOD at 865 nm yields large scatter with a root mean square error (RMSE) of 0.11 for single pixels, and RMSE=0.092 for 3 � 3 windows and a systematic mean overestimation of +0.036 and +0.031 for single pixels and 3 � 3 windows, respectively. The correlation statistics can be significantly improved by employing the Size Weighted Aerosol Index (SWAI) defined as the product of the AOD times the Angstrom exponent, showing a correlation coefficient (R) of 0.68 with RMSE=0.072 for single pixels and R ¼ 0:75 with RMSE=0.065 for the 3 � 3 windows. We have checked the reliability of the POLDER inversion scheme by comparing the aerosol polarized radiance (i.e., corrected for ground and molecular contributions) with AOD sunphotometer measurements at 670 nm for five different sites (R ¼ 0:36). The MODIS capability for monitoring seasonal and long-term aerosol dynamics is assessed by analyzing the 8-day MODIS AOD product at 550 nm (Level-3 product). The retrieved AOD accuracy is variable depending on sites and observed AOD dynamic range; for the overall database (N ¼ 1200 points), R ¼ 0:37; RMSE=0.17 and Bias=+0.088. We illustrate how satellite-derived AOD images can be used to monitor spatial and temporal aerosol dynamics at the regional scale (anthropogenic pollution events over the North-Eastern American coast) and at the continental scale (dust storm events and mean seasonal background atmospheric aerosol loading). It appears that aerosol mapping from satellite images is still difficult over North America, particularly at high latitudes where AOD variations are generally relatively small. r 2004 Elsevier Ltd. All rights reserved.