Thursday, 1 February 2024: 9:30 AM
345/346 (The Baltimore Convention Center)
Quantitative Precipitation Estimation (QPE) is one of the most important applications of weather radar observations. However, accurate precipitation estimation is a challenging task due to the wide space time variability of precipitation. The radar-based rainfall is typically verified with in-situ instrumentation such as rain gauges. In this paper, we present a machine learning based QPE product using radar data and rain gauge data collected in United Arab Emirates (UAE) region. We used 6 C-band radar data in UAE area that operated plan position indicator (PPI) scan every 6 to 10 minutes. Also, we used rain gauge data recorded every 15 minutes which was collected from 79 rain gauges in the domain. To accomplish machine learning based QPE using radar data and rain gauge data, we implemented a 2 stage deep neural network. For the first stage, we trained deep neural network for rain/no-rain classification using vertical profile of radar moments and rain gauge data. The deep neural network outputs the classification of the rain/no-rain class. Then using the rain/no-rain class and vertical profile of radar moments as an input, we trained a new deep neural network for rainfall estimation. The performance of machine learning based rainfall estimation was compared to rainfall measurement from ground and the performance metrics are provided.

