MAGARA: A Multi-Angle Geostationary Aerosol Retrieval Algorithm

Abstract. For over 40 years, the Geostationary Operational Environmental Satellite (GOES) system has provided frequent snapshots of the Western Hemisphere, with its data used for a variety of tasks ranging from weather forecasting to wildfire detection. Located on the GOES-16, GOES-17, and GOES-18 platforms, the Advanced Baseline Imager (ABI) is the first GOES-series imager that meets the precision requirements (e.g., ≥ 10 bits per datum) for high-quality, aerosol-related research. Here, we present a pixel-level (up to 1 km) Multi-Angle Geostationary Aerosol Retrieval Algorithm (MAGARA) that leverages the ABI instruments on the GOES-16 and GOES-17 platforms, as well as the differences in autocorrelation time-scales between surface reflectance, aerosol type, and aerosol loading. MAGARA retrieves pixel-level aerosol loading and fine-mode fraction at up to the cadence of the measurements (10 minutes), fine-and-coarse mode aerosol particle properties at a daily cadence, and surface properties under a framework that combines the unique information content in multi-angle radiances (e.g., sensitivity to aerosol type from multiple scattering-angle observations) with the robust surface characterization inherent to temporally tiled algorithms such as the MAIAC method.

We present three case studies, tiling radiances for several days over the Desert Southwest (2 cases) and the Pacific Northwest (1 case). We observed/retrieved smoke from the following major fires: the Camp Fire (November 5^th–12^th, 2018), the Williams Flats Fire (July 29^th–August 8^th, 2019), and the Kincade Fire (October 23^rd–November 1^st, 2019). Because GOES-17 was not making observations during the Camp Fire, we present this as a unique case demonstrating the efficacy of the multi-angle algorithm using only a single ABI sensor. We compare MAGARA retrievals of fine-mode (FM) AOD, coarse-mode (CM) AOD, and single-scattering albedo (SSA) with coincident AErosol RObotic NETwork (AERONET) spectral deconvolution algorithm (SDA) and inversion retrievals for the same period. We also compare MAGARA results against bias-corrected NOAA GOES-16 and GOES-17 retrieved 550 nm AOD.

For the 8,443 coincidences of MAGARA and the NOAA bias-corrected product with AERONET, MAGARA (NOAA bias-corrected product) 550 nm AOD error statistics are as follows: median-absolute error (MAE) = 0.016 (0.021), root-mean-squared error (RMSE) = 0.040 (0.049), and linear correlation coefficient (r) = 0.785 (0.666). At pixel-level resolution, the disparity between MAGARA and the NOAA bias-corrected product increases substantially, with MAGARA suffering less degradation in the results, likely due to lower pixel-to-pixel noise.

We report the following over-land MAGARA 500 nm fine-mode fraction error statistics for the 384 MAGARA/AERONET coincidences with MAGARA 500 nm AOD > 0.3: MAE = 0.031, RMSE = 0.100, and r = 0.902. Combined with the presented figures of daily averaged retrieved aerosol particle properties, this suggests that MAGARA has good sensitivity to fine-mode fraction over land, especially for smoky regions.

We also compare retrievals of MAGARA spectral single-scattering albedo with AERONET. Results suggest that a 1-parameter bias correction can substantially reduce MAGARA errors at high AOD. For the MAGARA retrieved spectral AOD > 0.5 (n = 116), this bias correction reduces MAE by 65 % (0.028 → 0.010), RMSE by 50 % (0.030 → 0.015), and improves correlation by 0.03 (0.84 → 0.87).

MAGARA performs best in regions where surface reflectance varies over long-time scales with minimal clouds. This represents a large portion of the western half of the US, much of North-Central Africa and the Middle East, some of Central Asia, and much of Australia. For these regions, aerosol type and aerosol loading on time scales as short as 10 minutes could allow for novel research into aerosol-cloud interactions, improvements to air-quality modeling and forecasting, and tighter constraints on direct aerosol radiative forcing.