Adapt Precipitation Super-Resolution and Data Fusion Deep Learning Techniques for Operational Flood Forecasting
Researchers at UAH's Information Technology and Systems Center (ITSC) have developed a Cloud-based Analytic Framework (CAPRi) for conducting super-resolution of precipitation datasets from the Global Precipitation Measurement mission Validation Network (GPM VN) by using deep learning models. These models, consisting of Convolutional Neural Networks (CNNs), are used to enhance the resolution of GPM Dual-frequently Precipitation Radar (DPR) data for improved identification of convective scale precipitation features and rain-rate estimates, particularly outside the coverage of ground-based weather radars. UAH/ITSC will support NOAA's Cooperative Institute for Research to Operations in Hydrology (CIROH) by expanding its super-resolution precipitation research and by developing new data fusion methods for merging other precipitation datasets that consist of varying spatial and temporal resolutions. Enhanced spatial resolution rain rate products merged with other precipitation datasets will be used as supplemental inputs to NOAA’s National Water Model (NWM) for improved operational hydrologic forecasting. Positive impacts are expected to be the greatest in areas where ground-based radar precipitation data are lacking.
The goals of this project are to improve operational flood forecasting by (1) using deep learning for improving the resolution of space-based satellite precipitation observations in areas where ground-radar data is lacking; (2) using the super-resolution results for estimating rain rates and precipitation features; and (3) leveraging data fusion techniques for merging meteorological information from different sources with different spatial and temporal scales. The success of this subproject will advance the state of knowledge in deep learning for super-resolution by: (1) using non-image data, (2) accounting for large differences in scale both temporal and spatial, (3) developing novel techniques for computing metrics of success, and (4) generating a more complete picture of hydrologic processes. Improved rain rate products merged with other precipitation datasets would provide supplemental inputs to NOAA’s National Water Model for improved hydrologic forecasting; especially in areas where ground radar coverage is absent.
