Combined IASI-NG and MWS Observations for the Retrieval of Cloud Liquid and Ice Water Path: A Deep Learning Artificial Intelligence Approach

Abstract

A neural network (NN) approach is proposed to combine future infrared (IASI-NG) and microwave (MWS) observations to retrieve cloud liquid and ice water path. The methodology is applied to simulated IASI-NG and MWS observations in the period January–October 2019. IASI-NG and MWS observations are simulated globally at synoptic hours (00:00, 06:00, 12:00, 18:00 UTC) and on a regular spatial grid (0.125° × 0.125°) from ECMWF 5-generation reanalysis (ERA5). The state-of-the-art σ-IASI and RTTOV radiative transfer codes are used to simulate IASI-NG and MWS observations, respectively, from the earth's state vector given by ERA5. A principal component analysis of the simulated IASI-NG observations is performed. Accordingly, a NN is developed to retrieve cloud liquid and ice water path from a combination of 24 MWS channels and 30 IASI-NG PCs. Validation indicates that this combination results in liquid and ice water path retrievals with overall accuracy of 1.85 10−2 kg/m2 and 1.18 10−2 kg/m2, respectively, and 0.97 correlation with respect to reference values. The root-mean-square error (RMSE) for CLWP results in about 30% of the mean value (5.91 10−2 kg/m2) and 22% of the variability (1-sigma). Similarly, the RMSE for CIWP results in about 41% of the mean value (2.91 10−2 kg/m2) and 22% of the variability. Two more NN are developed, retrieving cloud liquid and ice water path from microwave observations only (24 MWS channels) and infrared observations only (30 IASI-NG PCs), demonstrating quantitatively the advantage of using the combination of infrared and microwave observations with respect to either one alone.

Keywords

• Artificial neural networks
• Fourier transform infrared
• low earth orbit satellites
• microwave radiometry