MISR Research Aerosol Algorithm: refinements for dark water retrievals

Abstract

Abstract. We explore systematically the cumulative effect of many assumptions made in the MISR Research Aerosol retrieval algorithm, with the aim of quantifying the main sources of bias and uncertainty over ocean, and correcting them to the extent possible. 1132 coincident, surface-based sun photometer spectral aerosol optical depth (AOD) measurements are used for validation. Based on comparisons between these data and our baseline case (similar to the MISR Standard algorithm, but without the "modified linear mixing" approximation), for mid-visible AOD < 0.10, a high bias of 0.024 (0.032 for blue) is reduced by about half in the blue and green bands when (1) ocean surface under-light is included and the whitecap reflectance is increased for the red band, (2) physically based adjustments in particle microphysical properties and mixtures are made, (3) an adaptive pixel selection method is used, (4) spectral uncertainty is estimated from vicarious calibration, and (5) minor radiometric calibration changes are made for the red and NIR wavelengths. Applying (6) more stringent cloud screening (setting the maximum non-clear fraction to 0.50) brings all median spectral biases below 0.01. Large surface-modeling uncertainties preclude the use of both the blue and green MISR bands for over-ocean aerosol retrievals, even at mid-visible AOD higher than 0.2. When all adjustments are included (except more stringent cloud screening) and a modified acceptance criterion is used, the RMSE decreases for all wavelengths by 10–26% for the Research Algorithm itself, and 12–35% compared to the Version 22 MISR Standard Algorithm (SA). At mid-visible wavelengths, 86% of AOD data falls within 0.05 or 20% of validation values; 61% of blue band AOD data, and > 68% of green, red, and NIR values fall within 0.03 or 10%. For Ångström exponent (ANG): 68% of 1117 validation cases for AOD > 0.01 fall within 0.275 of the sun photometer values, compared to 49% for the SA. ANG RMSE decreases by 16% compared to the SA, and the median absolute error drops by 36%.