Observations of Aerosols, Clouds, and Radiation over the Amazon Rainforest while constraining the Large-Scale Synoptic Variability Using Self-Organizing Maps

The inaccurate representation of aerosol, cloud, and radiation properties and their co-variability with synoptic conditions over the Amazon rainforest within Earth System models leads to deficiencies in global climate simulations. This is addressed using an unsupervised machine learning technique called the Self-Organizing Map (SOM). The SOM is used to classify each day from 2000 to 2019 into nodes that represent patterns within the synoptic conditions over the Amazon. The dominant synoptic modes are represented by two sets of SOM nodes which were found to predominantly comprise the thermodynamic regimes of the Amazon’s dry and wet season. Satellite retrievals of clouds, aerosols, and top-of-atmosphere (TOA) radiative fluxes for these SOM nodes are compared to tie seasonal thermodynamic and synoptic patterns to cloud, aerosol, and radiation properties observed over the Amazon.

The SOM was trained on the 20-year dataset of 850 mb geopotential height anomalies from the European Centre Medium-Range Weather Forecasts Reanalysis (ERA5). Daily satellite retrievals of cloud and aerosol properties were obtained from Moderate Resolution Imaging Spectroradiometer (MODIS; MOD08_D3) and the all-sky TOA fluxes were obtained from Clouds and the Earth’s Radiant Energy System (CERES; Syn1deg). Ground-based observations from the GoAmazon2014/15 field campaign are used to characterize the thermodynamic regimes over the Amazon.

SOM nodes 0-3 had positive height anomalies and predominantly comprised days from the dry season. In contrast, SOM nodes 12-15 had negative height anomalies and comprised days from the wet season. Lower cloud fraction (CF), cloud optical depth, and cloud tops were observed in the dry season nodes compared to the wet season nodes with regional differences in aerosol optical depth (AOD). Atmospheric water vapor increased with AOD until AOD reached 0.3 and CF remained nearly constant for AOD above 0.1. This suggests the 3-D cloud effect was minimal and this dataset can be used to examine aerosol-cloud interactions. The wet season nodes showed higher TOA observed shortwave and net flux and lower longwave flux compared to the dry season. Further analysis is ongoing to examine regional variability in aerosol, cloud, and radiation interactions while constraining the synoptic variability.