A Global Merged Diurnal Ice/Snow Cloud Product from Spaceborne Passive Microwave Observations and Its Applications to Model Evaluation

Ice cloud and floating snow play critical roles in Earth’s energy budget and hydrological cycle. Their diurnal variation is tightly coupled with convection development life cycle, hence it also greatly impacts the diurnal cycle of surface precipitation and top of the atmosphere radiation. Due to the high degree of freedom of ice crystal microphysical properties, remote sensing of ice/snow cloud is challenging for passive spaceborne sensors.

In this work, we present a global diurnal ice/snow cloud product by merging three spaceborne passive microwave sensor observations together (GPM-GMI, NPP-ATMS, and MT-SAPHIR). This dataset includes ice water path (cloud ice + falling snow), cloud top height (CTH) and cloud bottom height (CBH) at pixel level between 2015 – 2016, and monthly gridded values at 2deg X 2deg X 2 hours grid scale. The convolutional neural network (CNN) approach is adopted for the algorithm development by learning from collocated CloudSat observations, and the Monte Carlo dropout method is used for uncertainty estimation. A customized loss-function is developed to retrieve cloud mask and mass together.

We evaluated the retrieval at collocated pixels as well as against other independent field campaign and ground-based measurements. Diurnal and semi-diurnal distributions of the IWP will be presented. We will also demonstrate how we use this product to evaluate model performance on capturing the general distribution and diurnal variation of the frozen hydrometeors in the atmosphere.