Estimation of Cloud Condensation Nuclei (CCN) from SPEXone on PACE using a deep neural network retrieval algorithm

Proper proxies for CCN are vital to provide accurate constraints for Aerosol-Cloud Interactions (ACI) in climate models. An effective proxy for CCN is the column number of aerosol particles that surpasses a predetermined threshold radius (Nccn). This CCN proxy has been estimated from PARASOL using level 2 aerosol microphysical and/or optical property retrievals. With the launch of SPEXone on Plankton, Aerosol, Cloud, ocean Ecosystem (PACE) satellite, further improvements on the Nccn retrievals are expected. For example, retrieved refractive index can be used to estimate the volume fraction of aerosol-water, which can help deduce the dry aerosol size distibution and subsequently dry CCN. Further, the retrieved Aerosol-Layer Height (ALH) can be used to estimate the boundary layer (BL) contribution of Nccn (Nccn (BL)) which is better suited for quantifying ACI as it is more related to CCN at cloud base than the total column.

The estimation of Nccn from physics based MAP algorithms can be challenging given its dependance on multiple retrieved aerosol parameters. We have implemented a deep neural network (NN) algorithm as an extension for the Remote sensing of Trace gas and Aerosol Products (RemoTAP)-NN algorithm to directly retrieve dry Nccn and Nccn (BL) from SPEXone measurements. The algorithm is trained on synthetic SPEXone measurements based on 3 aerosol modes which are fine mode, insoluble coarse/dust mode and soluble coarse mode. It has been validated using synthetic SPEXone measurements, simulated based on the 7 mode aerosol model from the ECHAM-HAM global aerosol-climate model. The performance of the NN algorithm was compared with RemoTAP classical algorithm.

The NN algorithm retrieved dry Nccn has a relative RMSE of 0.197 over the ocean and 0.301 over the land whereas dry Nccn estimated by RemoTAP level-2 retrievals for the same synthetic measurements has a relative RMSE of 0.382 over ocean and 0.559 over land. Nccn (BL) retrieved from the NN algorithm has a relative RMSE of 0.349 and 0.825 over the ocean and the land respectivey. The relative RMSE of Nccn (BL) derived from the RemoTAP classical algorithm is 1.039 and 1.233 over the ocean and land respectively. Our study demonstrates that the NN algorithm can accurately retrieve Nccn, outperforming the capabilities in classical algorithms.