EPIC OCRA/ROCINN: Retrieval of EPIC/DSCOVR cloud macrophysical parameters from measurements in the UV and oxygen A- and B- bands

Abstract:

In this work, we show the application of the OCRA (Optical Cloud Recognition Algorithm) and ROCINN (Retrieval Of Cloud Information using Neural Networks) algorithms to the sensor EPIC/DSCOVR to estimate three macrophysical parameters of liquid-water clouds: radiometric cloud fraction, cloud optical thickness and cloud-top height.

The estimation of the radiometric cloud fraction is performed by means of OCRA, which determines the cloud coverage as the weighted distance of the UVN pixel measurements to those of the expected clear-sky scenarios at every Earth location and time. The estimation of the cloud optical thickness and cloud-top height is done in the oxygen A- and B- bands by means of ROCINN, in which the radiative transfer model simulations are replaced by two artificial neural networks trained for the clear-sky and cloudy-sky scenarios. The training data is generated using EPIC-specific geometrical constraints on the model inputs in order to focus the training of the neural networks only on feasible model configurations.

The impact of the OCRA threshold value in the selection of the optimal OCRA parameters is discussed. From time evolution analysis it is found that the OCRA scaling factors present a yearly oscillating behaviour, while the OCRA offsets can be assumed constant. The regularisation requirements from the non-linear optimisation problem solved by ROCINN are also discussed. The EPIC OCRA/ROCINN daily cloud products are compared with the MODIS daily cloud products from Terra and Aqua, as well as the operational OCRA/ROCINN cloud products retrieved from TROPOMI/S5P (TROPOspheric Monitoring Instrument/Sentinel-5 Precursor).