Rainfall Estimation From Ground Radar and TRMM Precipitation Radar Using Hybrid Deep Neural Networks.

Remote sensing of precipitation is critical for regional, continental, and global water and climate research. This study develops a deep learning mechanism to link between point‐wise rain gauge measurements, ground‐based, and spaceborne radar reflectivity observations. Two neural network models are designed to construct a hybrid rainfall system, where the ground radar is used to bridge the scale gaps between rain gauge and satellite. The first model is trained for ground radar using rain gauge data as target labels, whereas the second model is for spaceborne Tropical Rainfall Measuring Mission (TRMM) Precipitation Radar (PR) using ground radar estimates as training labels. Data from 1 year of observations in Florida during 2009 are utilized to illustrate the application of this hybrid rainfall system. Validation using independent data in 2009, as well as 2‐year comparison against the standard PR products, demonstrates the promising performance and generality of this innovative rainfall algorithm. Plain Language Summary: The Tropical Rainfall Measuring Mission (TRMM) Precipitation Radar (PR) was the first spaceborne active sensor for observing precipitation over the tropics and subtropics. During its 17 years (1997–2014) in orbit and beyond, PR has been an important tool to characterize tropical precipitation microphysics and quantify rainfall rate over the globe. Ground validation is a critical component in the development of TRMM products. However, the ground‐based sensors have different characteristics from PR in terms of resolution, viewing angle, and uncertainties in the sensing environments, which are not taken into account in the operational parametric rainfall relations applied to PR measurements. This study develops a nonparametric machine learning technique for PR rainfall estimation. In the regions where substantial gauge and ground radar data are available, this approach can produce better rainfall estimates compared to the standard PR algorithm. In areas such as ocean and remote regions where no gauge or radar available, the proposed rainfall algorithm is easy to implement, and it can still produce reasonable estimates. With more and more gauges and radars being deployed and many of them become operational, this algorithm can be trained at different locations represented by different atmosphere properties to further improve the performance and generality. Key Points: Conventional parametric relationships between radar reflectivity Z and rain rate R are not sufficient to capture precipitation variabilitiesA hybrid deep neural network system is designed for improved space radar rainfall estimation [ABSTRACT FROM AUTHOR]