ESRL/PSD Seminar Series

Abstract

Rainfall products derived based on satellite measurements have proven to be very useful for regional and/or global hydrologic modelling and climate studies. A number of precipitation products at multiple space and time scales have been developed using Infrared (IR) brightness temperature information measured by geostationary satellites and observations from passive microwave (PMW) sensors on board the low earth orbit satellites. Typical examples include the CMORPH products developed using the NOAA/CPC morphing technique, which produces global precipitation mapping by combining existing space-based observations and retrievals. However, the accuracy of satellite products is often limited due to the spatial and temporal sampling as well as the parametric retrieval algorithms, particularly for heavy rain in extreme events or in orographic rainfall conditions. On the other hand, ground-based radar is more mature for quantitative precipitation estimation, especially after the implementation of dual-polarization technique and further enhanced by urban scale dense radar networks. This study presents the high performance rainfall system developed for an urban weather radar network over the Dallas-Fort Worth (DFW) area. A machine learning mechanism is introduced to improve satellite-based rainfall retrievals by incorporating the ground radar observations from the DFW network. In particular, the PMW-based retrievals and IR data serve as input to the machine-learning model, whereas the high-quality ground radar rainfall products are used as target to train the model. The trained model is evaluated using operational CMORPH products and surface rainfall measurements from gauge networks during independent (testing) precipitation events.