Three-dimensional effects in satellite-equivalent observations over the Tropics at sub-mm wavelengths

By means of the Atmospheric Radiative Transfer Simulator (ARTS), a sensitivity study has been conducted to quantify the errors induced by neglecting three-dimensional (3D) atmospheres at sub-millimeter wavelengths. This work provides an overview of the 3D effects (beam-filling and horizontal photon transport effect) that will potentially affect the cloud ice retrievals of the upcoming Ice Cloud Imager (ICI) satellite mission that will be operated by the European Organization of Meteorological Satellites (EUMETSAT). The 3D synthetic scenes have been generated from two-dimensional (2D) CloudSat observations over Tropical latitudes using a stochastic approach. In order to conduct a comprehensive statistical analysis and quantify the magnitude of both effects, we randomly sampled CloudSat scenes for July 2015. Forward simulations have been conducted at frequencies 186.3 GHz and 668 GHz, resembling the ICI channels. In addition to the mean footprint size of ICI (15 km), a footprint of 6 km is employed as a basis for comparison. Three radiative transfer calculations have been conducted along the line of sight of the sensor: (a) a full 3D simulation (3D mode), (b) an Independent Beam approximation (IBA mode) resembling the independent column approximation, and (c) a one-dimensional (1D) simulation following the plane-parallel assumption (1D mode). The comparison between the 3D and IBA modes reveals no considerable discrepancies, with IBA simulations introducing mostly random errors and a slight overestimation (below 0.33K) of the simulated brightness temperatures that becomes evident at 668 GHz and the smaller footprint size (6 km). However, performing 1D simulations divulges a significant beam-filling effect that increases with frequency and footprint size. In brief, the errors induced by neglecting domain heterogeneities yield a median error of about 1.9 K with an interquartile range of [0.44, 6.68] K, but a considerable scatter, i.e., a root mean square error of about 15 K. Only for a footprint size of 6 km and a frequency of 186.3GHz, a rather good agreement is found between the two calculation modes, whereby the smallest discrepancies are found.