16. JP 01 Real Time Alignment and Distribution of Weather Radar Data with Rain Gauge Data for Deep Learn


Quantitative Precipitation Estimation (QPE) based on weather radar observations plays a significant role in the understanding of weather events, especially in real-time, where fast evolving phenomena like convective storm cells can be dangerous. We wish to demonstrate QPE using deep learning as an alternative approach to empirical relationship equations between rainfall rate and reflectivity which were developed in the past. QPE using radar reflectivity is one of the possible applications of deep learning in the weather radar field. Preprocessing this data and saving it in real-time on cloud would let the users skip the time-consuming preprocessing step and assist them to directly get to the deep learning phase. To train and test deep learning models with radar data, we must align rain gauge data in space and time. This data preprocessing requires time and resource consuming processes that involve downloading, extracting, gridding, aligning the radar data with respect to every gauge in the region. If this preprocessed dataset was readily available in real-time, deep learning can be easily performed on it by anyone without going through the heavy computations required in the process. EarthCube’s CHORDS tool is a real-time data service that can be used to store preprocessed data on cloud so that it can be accessed whenever and wherever required. In this work, we demonstrate the steps involved in preprocessing such as accessing WSR-88D radar and NASA-TRMM rain gauge data, Cartesian gridding of radar data, aligning the radar data with gauge data in real-time. This aligned data is stored on cloud using CHORDS, so that it can be readily available to users who wish to use it for deep learning. The notebook will also demonstrate the procedure for storing and retrieving the dataset from CHORDS server and an example of the deep learning process on the downloaded dataset.