Development of urban surface models for improved aerosol retrieval
Min Oo*
a,b
, Matthias Jerg
a
, Ana J. Picon
a,b
, Eduardo Hernandez
a,b
, Barry Gross
a
, Fred Moshary
a
, Samir Ahmed
a
a
City College of City University of New York, EE Department, 140
th
Street & Convent Ave, New York, NY 10031
b
Graduate Center of City University of New York, 365 Fifth Avenue, New York, NY 10016
ABSTRACT
A combination of CIMEL radiometer and MODIS measurements are used to correct surface albedo models. In
particular, we show through an analysis of hyperspectral high resolution Hyperion data that the correlation coefficient
assumption underestimates ground albedo resulting in an overestimate of the VIS optical depth and operational collect 5
surface model shows an incorrect trend between the MVI index and the surface correlations. Preliminary radiative
transfer calculations based on the same model show that this mechanism can help explain the observed overestimation
and the corrected models have been implemented for NYC and Mexico City with significantly improved AOD.
Key words: urban ground albedo, aerosol retrieval over land, satellite remote sensing
1. INTRODUCTION
It is well known that accurate global characterization of Aerosol Optical Depth (AOD) is essential in quantifying energy
balance for climate change studies [1]. On the other hand, regional and local characterization of aerosols at Earth’s
surface is essential for monitoring surface air quality [2-5] and effects on human health [6]. While satellite retrieval
products are used for both purposes [2, 6], assumptions which are used to derive non-biased global aerosol may induce
errors in specific regions [7], including highly urbanized megacities. As a result, using biased AOD measurements
directly or through assimilation into Air Quality Forecasts can lead to further errors in PM2.5 monitoring and
forecasting.
As urbanization continues to grow, retrieval of aerosols within these megacities becomes more important.
However, retrieval of AOD by satellite remote sensing measurements over land is complicated by the fact that the Top
of Atmosphere (TOA) reflectance is a combination of the desired atmospheric path reflectance as well as the ground
reflectance. To avoid this problem, AOD retrieval with the Moderate Resolution Imaging Spectroradiometer (MODIS)
instrument as described in the MODIS Collection 4 (C004) algorithm documentation [7] was focused on using “dark”
pixels such as dense dark vegetation in an image as a way of isolating the aerosol contribution [8-9]. In fact, the surface
reflectance ratios that connect the visible (VIS) ground albedo to the shortwave infrared (SWIR) albedo were set to
values that were suitable mainly for vegetation. These conditions are not realistically met over large urban areas such as
the New York City megalopolis as well as other surfaces. In addition, the requirement of dark pixels is a prime reason
for the low resolution AOD retrieval (at 10 km) since bright pixels must be removed prior to processing. Therefore,
loosening the restriction can ultimately improve the spatial resolution of the AOD product which is especially relevant
for urban observations.
Several improvements were pursued in trying to improve on the Collection 4 aerosol retrieval over land
algorithm in Collection 5 (i.e. C005) [10-12] such as the inclusion of Polarization into the LUT (look up table)
generation, [13]. However, special interest was paid to developing surface models that could be connected to land
surface types. This effort was based on global comparisons of AERONET (AErosol RObotic NETwork) sky radiometer
data and MODIS TOA reflectances. In performing these comparisons, it was found helpful to obtain global aerosol
models based on cluster analysis techniques. Once these models were obtained and their climatology examined, selected
aerosol models (for the fine mode aerosols) were chosen which were best linked to specific regions. From this analysis,
an empirical relationship was constructed to describe the surface VIS-SWIR ratios which were shown to be functions of
surface type (as defined by a Modified Vegetation Index (MVI) which is not sensitive to atmospheric uncertainty) and
to a lesser extent on scattering angle. However, we find that significant errors occur when this globally trained model is
applied to regional urban areas and regional refinements must be implemented. This is best illustrated in figure 3 where
MODIS AOD retrieval is compared to collocated AERONET AOD measurements. We note in particular a very strong
over-bias in the AOD. On the other hand, MISR inter-comparisons as illustrated in figure 1 show a very different
Atmospheric and Environmental Remote Sensing Data Processing and Utilization V: Readiness for GEOSS III,
edited by Mitchell D. Goldberg, Hal J. Bloom, Proc. of SPIE Vol. 7456, 74560B · © 2009 SPIE
CCC code: 0277-786X/09/$18 · doi: 10.1117/12.826294
Proc. of SPIE Vol. 7456 74560B-1