Assessment of Ensemble Rainfall Nowcasting: A Case Study of the U.S. Southeast Coast during Hurricane Ian

Several counties were affected by severe flooding caused along the U.S. Southeast coastline by Hurricane Ian, in 2022. An accurate forecast of hourly and sub-hourly precipitation is essential for emergency operations and the planning of response actions during such natural disasters.

Sub-hourly forecast of rainfall is critical, especially in urban areas where the hydrologic response is rapid. This is particularly critical when extreme events are compounded with ocean surges during extreme events like Hurricane Ian. In this study, the ensemble-based Lagrangian INtegra-Difference equation model with Autoregression (LINDA) model was applied to predict sub-hourly rainfall based on the Multi Radar/Multi Sensors (MRMS) data. The purpose of this study is to assess the effectiveness of the nowcasting model to predict sub-hourly rainfall using a retrospective validation of Hurricane Ian.

First, the Hybrid Single-Particle Lagrangian Integrated Trajectory (HYSPLIT) model was used to backtrack the rainfall, and determine the appropriate start location and time to initialize the nowcasting which is done retrospectively. Based on the ensemble spread and skill metrics, the efficiency of nowcasting techniques was assessed. Probability of Detection (POD), False Alarm Ratio (FAR), and Critical Success Index (CSI) were used to evaluate predictions of the spatial range of hurricane rainfall against MRMS data. The accuracy of the predicted precipitation depth was evaluated using three different metrics: Root-Mean-Square-Error (RMSE), Correlation Coefficient (CC), and Mean Average Error (MAE). A good agreement was found between the rainfall nowcast and the observed rainfall data for the first two forecasting steps (10 and 20 minutes) with CSI values exceeding 0.8 for the ensemble spread and 0.65 for the ensemble skill. For the following time steps, satisfactory prediction performance was obtained with CSI values ranging between 0.70 and 0.81 for the ensemble spread, and between 0.28 and 0.59 for the ensemble skill. There is, however, a tendency to underestimate the spatial extent of hurricane rainfall, as well as high rainfall rates, which could be attributed to rapid changes in rainfall intensity and wind speed during hurricanes.