Impacts of the GLM measurements aboard the GOES-R series satellites on severe convective precipitation detection and retrievals

Compared to the legacy GOES Imager, the Advanced Baseline Imager (ABI) aboard the GOES-R series provides much higher resolution measurements in both spatial and temporal dimensions, which benefits many quantitative applications such as precipitation detection and estimation. However, it is still a challenge to accurately quantify the cloud water content and the near-surface precipitation intensity using parametric relations based on the Infrared (IR) radiation and converted cloud-top brightness temperature. Comparison between ground-based rainfall observations and the operational GOES-16 (formerly known as GOES-R) rainfall products indicates that the uncertainty in GOES-16 rain retrievals is significant, and the uncertainty varies at different geophysical locations. This research develops a deep convolutional neural network-based ensemble machine learning framework to enhance precipitation detection and retrievals through incorporating the Geostationary Lightning Mapper (GLM) data, with an emphasis on severe convective precipitation events. The saliency maps are used to interpretate the underlying model physics and quantify the impacts of input features, i.e., cloud-top brightness temperature at multiple ABI channels and GLM flash rate, on the precipitation detection and estimation performance. Extensive experiments are performed to understand the flash characteristics and impacts on precipitation retrievals by investigating the saliency maps on the activation of a specific neuron and different groups of neurons. The results demonstrate that this explainable machine learning framework can predict precipitation location, intensity, and amount very well, and the inclusion of GLM data can not only better predict the location of sever convective storms, but also improve the accuracy of estimated precipitation intensity.