Satellite‐derived aerosol optical depth over dark water from MISR and MODIS: Comparisons with AERONET and implications for climatological studies

Abstract

[1] Although the current Multiangle Imaging Spectroradiometer (MISR) and Moderate Resolution Imaging Spectroradiometer (MODIS) satellite passive remote sensing midvisible aerosol optical thickness (AOT) products are accurate overall to about 0.05 or 20%, they differ systematically on a global, monthly average basis, by about 0.03 to 0.05. Some key climate change and other applications require accuracies of 0.03 or better. The instruments are sufficiently stable and well characterized, and have adequate signal-tonoise, to realize such precision. However, assumptions made in the current standard aerosol retrieval algorithms produce AOT biases that must be addressed first. We identify the causes of AOT discrepancies over dark water under typical, relatively low AOT conditions and quantify their magnitudes on the basis of detailed analysis. Examples were selected to highlight key issues for which there are coincident MISR, MODIS, and Aerosol Robotic Network (AERONET) observations. Instrument calibration and sampling differences, assumptions made in the MISR and MODIS standard algorithms about ocean surface boundary conditions, missing particle property or mixture options, and the way reflectances used in the retrievals are selected each contribute significantly to the observed differences under some circumstances. Cloud screening is also identified as a factor, though not fully examined here, as are the relatively rare high-AOT cases over ocean. Specific algorithm upgrades and further studies indicated by these findings are discussed, along with recommendations for effectively using the currently available products for regional and global applications.

Response to “Toward unified satellite climatology of aerosol properties. 3. MODIS versus MISR versus AERONET”

“…Aside from global retrieval coverage differences [e.g., 12], MISR and MODIS do not observe, or perform aerosol retrievals on, top-of-atmosphere radiances from exactly the same locations within their 17.6 km and 10 km retrieval regions, respectively. Where variability occurs on spatial scales of ~10 km or less, this leads to sampling differences even for "collocated" observations [11]. MISR and MODIS data are often complementary, so differences do not always represent algorithm issues; sometimes they offer a deeper understanding of realistic natural scenes.…”

“…Explanations for the most prominent differences between the MISR Standard aerosol products and validation data are given in several of the papers cited above, making use of other studies where individual effects for dust [13], spherical absorbing and non-absorbing particles [7], thin cirrus [17], and algorithmic issues [11] are explored for representative cases with the MISR Research aerosol retrieval algorithm. Detailed analysis of MODIS issues is presented in [2,4,14,20,21,[24][25][26][27].…”

“…This metric is especially sensitive to the largest outliers, which can be caused by retrieval error or by sampling differences, as mentioned above. In addition, although the sampling differences tend to produce outliers of both signs, AOD retrieval errors are often systematic rather than random [10,11], and most are localized geographically within regions where certain algorithmic assumptions are violated [12, Figure 6; 29, Section 3.4]. As a statistical tool, the RSTD is most appropriate for assessing random error about an expected value; interpretation of RSTD values under other circumstances is not always straightforward.…”

“…MISR and MODIS data are often complementary, so differences do not always represent algorithm issues; sometimes they offer a deeper understanding of realistic natural scenes. For example, case studies that take detailed account of instrument sampling have shown where MISR and MODIS-retrieved AOD differences are due to aerosol spatial variability [9,11].…”

“…The MISR and MODIS instrument teams, as well as many other research groups, have performed such evaluations using a variety of methods, ranging from theoretical sensitivity studies to comparisons with other satellite and suborbital data sets [e.g., 2,[6][7][8][9][10][11][12][13][14][15][16][17][18][19][20][21][22][23]. Over the past ten years, the exchange of data, information, and ideas represented by this body of work has yielded better characterization of existing products, as well as algorithm upgrades.…”