How Well Does NASA GEOS Model Perform in Simulating Dust Deposition into the Tropical Atlantic Ocean?

Massive dust emitted from North Africa can transport long distances across the tropical Atlantic Ocean, reaching the Americas. Dust deposition along the transit adds microorganisms and essential nutrients to marine ecosystem, which has important implications for biogeochemical cycle and climate. However, assessing the dust-ecosystemclimate interactions has been hindered in part by the paucity of dust deposition measurements and large uncertainties associated with oversimplified representations of dust processes in current models. We have recently produced a unique dataset of seasonal dust deposition flux and dust loss frequency into the tropical Atlantic Ocean at a nominal resolution of 200 km x 500 km by using the decade-long (2007-2016) record of aerosol three-dimensional distribution from four satellite sensors, namely CALIOP, MODIS, MISR, and IASI. On the basis of the ten-year average, the yearly dust deposition into the tropical Atlantic Ocean is estimated at 98-153 Tg. The dust deposition shows large spatial and temporal (on seasonal and interannual scale) variability. The satellite observations also yield an estimate of annual mean dust loss frequency of 0.052 ~ 0.078 d-1, a useful diagnostic that makes it possible to disentangle the dust transport and removal processes from the dust emissions when identifying the major factors contributing to the uncertainties and biases in the model simulated dust deposition. In this study, we use the dataset along with in situ and remote sensing observations to assess how well NASA GEOS model performs in simulating trans-Atlantic dust transport and deposition. We found that the GEOS modeling of dust deposition falls within the range of satellite-based estimates. However, this reasonable agreement in dust deposition is a compensation of the model's underestimate of dust emissions and overestimate of dust removal efficiency. Further, the overestimate of dust removal efficiency results largely from the model's overestimate of rainfall rate. Our results provide insights into the model's deficiencies at process level, which could better guide model improvements.