A Deep Learning Approach to Radar‐Based QPE.

In this study, we propose a volume‐to‐point framework for quantitative precipitation estimation (QPE) based on the Quantitative Precipitation Estimation and Segregation Using Multiple Sensor (QPESUMS) Mosaic Radar data set. With a data volume consisting of the time series of gridded radar reflectivities over the Taiwan area, we used machine learning algorithms to establish a statistical model for QPE in weather stations. The model extracts spatial and temporal features from the input data volume and then associates these features with the location‐specific precipitations. In contrast to QPE methods based on the Z–R relation, we leverage the machine learning algorithms to automatically detect the evolution and movement of weather systems and associate these patterns to a location with specific topographic attributes. Specifically, we evaluated this framework with the hourly precipitation data of 45 weather stations in Taipei during 2013–2016. In comparison to the operational QPE scheme used by the Central Weather Bureau, the volume‐to‐point framework performed comparably well in general cases and excelled in detecting heavy‐rainfall events. By using the current results as the reference benchmark, the proposed method can integrate the heterogeneous data sources and potentially improve the forecast in extreme precipitation scenarios. Plain Language Summary: Quantitative precipitation estimation (QPE) is a method of approximating the amount of rain that has fallen at a location or across a region. In most use cases, weather service providers use radar signals to estimate the amount of precipitation through the formula describing the relationship between radar reflectivity and the size of raindrop particles, Z–R relation. The state‐of‐the‐art QPE methods with adjusted Z–R relation are robust and accurate in general. Yet, they consider only the radar signal at a given location, which represented a point‐to‐point framework. In this study, we proposed a volume‐to‐point alternative that estimates the amount of rain by considering the signals in a broader spatial region and a longer time‐span. By using a deep neural network to process the large data volume, we demonstrated that the proposed method performed comparably well in general cases and excelled in detecting heavy‐rainfall events. This method could improve advanced warning for flash flooding and make water resource management more effective. In ongoing research, we seek to extend our approach to integrate the heterogeneous data sources and to be applied to precipitation forecasting. Key Points: We proposed a novel deep learning approach for estimating precipitation from a large data volume and demonstrated it with QPE from aggregated radar dataIn comparison to the operational QPE scheme, the proposed framework performed comparably well in general cases and excelled in detecting heavy‐rainfall eventsThe proposed framework can be further extended to integrate heterogeneous data sources and potentially improve the QPF in extreme precipitation scenarios [ABSTRACT FROM AUTHOR]