Predicting Hurricane-induced Intense Rainfall Trajectories using GNSS

Abstract

The tropospheric products of Global Navigation Satellite Systems (GNSS) can be used to determine the density and distribution of water vapor in the atmosphere, and therefore has been used to monitor precipitation events. This study analyzes GNSS-based precipitable water vapor (PWV) measurements calculated from the NOAA Continuously Operating Reference Stations (CORS) GNSS observations, to 1) track spatiotemporal variability of PWV, 2) identify the abnormal fluctuations in PWV level before the arrival of hurricane at a ground station, and 3) predict the path and relative intensity of hurricane-induced rainfalls. Firstly, we examined PWV perturbations with the local atmospheric elements including temperature and pressure, and relative humidity and revealed the relationship between the atmospheric parameters and the formation process of a severe precipitation. The CORSs are then classified into a training set and a test dataset. Numerically analyzed meteorological constituents for the training dataset were used to derive a PWV prediction model by applying a multivariate regression approach. The PWV prediction model quantifies the relationship among a spatiotemporal hurricane intensification, the PWV rate of change, and meteorological variables. To avoid the correlation effect between these variables, a principal component regression (PCR) was applied. From the PWV prediction model, the PWV at each test station was predicted with 12 and 24 hours of time scale. The PCR is then applied to test the dataset, and the model’s residuals, which are the discrepancy between the model and measurements, are calculated to verify the model. The residual of the predicted model is a key factor to determine the trajectory of hurricane-induced rainfall and its intensity. By analyzing the distribution pattern of the predicted PWV residuals, their magnitude, and the observed PWV at the test site, the probable locations of intense rainfalls due to the storm front passage can be identified. For a robust analysis considering the uncertainty from the measurement noise and other error sources in the GNSS-derived PWV, we defined a grid in the test site that allows evaluating multiple stations’ PWV prediction/measurements. The grid size was determined with the consideration of the test site and the geometric distribution of available CORSs, which their GNSS observations were used as the data feeds into the prediction model. Because various hurricanes have their own spatial and temporal characteristics, the approach is assessed for two different hurricanes occurred in the same location and showed different types of rainfall events that are Hurricane Mathew in 2016, and Hurricane Irma in 2017. Because both hurricanes landed in Florida and proceeded to Georgia, and South Carolina, the model’s performance can be evaluated under similar geographic and climate characteristics of the study area. The results were validated by the radar reflectivity map and reported NHC hurricane landfall centers. The results showed that for both hurricanes, the highly probable locations of heavy precipitation by the grid-based prediction coincide to the grids with the minimum residuals of the prediction model. In addition, the negative correlation between the residuals of PWV measurements with the prediction model and the magnitude of precipitation was revealed. The magnitude of the predicted model residuals was used for hurricane tracking and applied to evaluation of the storm relative intensity. The study showed that predicted locations (grids) were contained at maximums of less than 25% and 32% of total residuals in the area for 12h and 24h prediction time lags, respectively. This study demonstrates the effectiveness of the statistical model for forecasting the intense precipitation path at least several hours before the arrival of a storm that can be applied to a hazard early warning system.