3D radiative heating of tropical upper tropospheric cloud systems derived from synergistic A-Train observations and machine learning

• Published in: Atmospheric chemistry and physics 2021-01, Vol.21 (2), p.1015-1034
• Main Authors: Stubenrauch, Claudia J, Caria, Giacomo, Protopapadaki, Sofia E, Hemmer, Friederike
• Format: Article
• Language: English
• Subjects: Analysis; Anvils; Artificial neural networks; Atmospheric Infrared Sounder; Atmospheric models; Cirrus clouds; Cloud systems; Clouds; Convective clouds; Correlation coefficient; Correlation coefficients; Data; Dynamic meteorology; Emissivity; Heating; Heating rate; High temperature effects; Learning algorithms; Lidar; Lidar measurements; Lower atmosphere; Machine learning; Mesoscale convective systems; Meteorological satellites; Middle troposphere; Neural networks; Observational learning; Optical radar; Physics; Radar; Radar measurement; Radiative heating; Railway tracks; Regions; Regression analysis; Regression models; Remote sensing; Sciences of the Universe; Short wave radiation; Sky; Surface properties; Surface temperature; Tropical climate; Tropical environments; Tropopause; Troposphere; Tropospheric clouds
• ISSN: 1680-7324; 1680-7316; 1680-7324
• DOI: 10.5194/acp-21-1015-2021

Upper tropospheric (UT) cloud systems constructed from Atmospheric Infrared Sounder (AIRS) cloud data provide a horizontal emissivity structure, allowing the convective core to be linked to anvil properties. By using machine learning techniques, we composed a horizontally complete picture of the radiative heating rates deduced from CALIPSO lidar and CloudSat radar measurements, which are only available along narrow nadir tracks. To train the artificial neural networks, we combined the simultaneous AIRS, CALIPSO and CloudSat data with ERA-Interim meteorological reanalysis data in the tropics over a period of 4 years. The resulting non-linear regression models estimate the radiative heating rates as a function of about 40 cloud, atmospheric and surface properties, with a column-integrated mean absolute error (MAE) of 0.8 K d−1 (0.5 K d−1) for cloudy scenes and 0.4 K d−1 (0.3 K d−1) for clear sky in the longwave (shortwave) spectral domain. Developing separate models for (i) high opaque clouds, (ii) cirrus, (iii) mid- and low-level clouds and (iv) clear sky, independently over ocean and over land, leads to a small improvement, when considering the profiles. These models were applied to the whole AIRS cloud dataset, combined with ERA-Interim, to build 3D radiative heating rate fields. Over the deep tropics, UT clouds have a net radiative heating effect of about 0.3 K d−1 throughout the troposphere from 250 hPa downward. This radiative heating enhances the column-integrated latent heating by about 22±3 %. While in warmer regions the net radiative heating profile is nearly completely driven by deep convective cloud systems, it is also influenced by low-level clouds in the cooler regions. The heating rates of the convective systems in both regions also differ: in the warm regions the net radiative heating by the thicker cirrus anvils is vertically more extended, and their surrounding thin cirrus heat the entire troposphere by about 0.5 K d−1. The 15-year time series reveal a slight increase of the vertical heating in the upper and middle troposphere by convective systems with tropical surface temperature warming, which can be linked to deeper systems. In addition, the layer near the tropopause is slightly more heated by increased thin cirrus during periods of surface warming. While the relative coverage of convective systems is relatively stable with surface warming, their depth increases, measured by a decrease of their near-top temperature of -3.4±0.2 K K−1. Fi