This project's intellectual merits stem from integrating meteorological and geographical research methods to investigate two fundamental questions about tropical cyclone (TC) structure and precipitation production: 1) how skillful are satellite and modeling datasets in representing precipitation structures of TCs compared to ground-based observations, and 2) what are the dominant large-scale patterns of moisture that surround Atlantic hurricanes.  

To explore question one, we conducted an observational analysis of reflectivity from space-based radar (the Global Precipitation Measurement Mission's Dual-frequency Precipitation Radar (DPR)) and ground-based radars in the Weather Surveillance Radar 1988 Doppler (WSR-88D) network. For 24 instances when scans from both viewing angles were available for TCs over the United States, we calculated differences in matched points from individual ground radars and mosaics that incorporated all ground-based radars that detected the storm, and calculated differences after placing values onto a common grid. We also grouped similar reflectivity values into regions and performed object-based analyses. We found that the differences between DPR and ground radars varied across reflectivity values, from being closer-matched (within 1-2 dBZ) when both detected lower reflectivity (20-30 dBZ) to DPR being higher when both detected higher reflectivity, especially in the core regions of the storm. The mosaicking strategy that produced the smallest differences was to retain the maximum value when multiple WSR-88D observations existed, but values can be too high when more than 4 minutes elapses between scans of neighboring radars due to fast-moving and rapidly changing convective regions of hurricane eyewalls. We also compared observed (from WSR-88D observations that are rain gauge-corrected [Stage IV]) and modeled rain rates from two high-resolution hurricane forecasting models [Hurricane Analysis and Forecast System (HAFS) model and the basin-scale Hurricane Weather Research and Forecasting (HWRF-B) model] using 2020 North Atlantic landfalling TCs. This model verification used object-based methods, which quantify and compare the shape and size of the forecast and observed precipitation. The spatial biases suggested that the model forecasts were too intense even though there was a negative intensity bias for both models, indicating there may be an inconsistency between the precipitation configuration and the maximum sustained winds in the model forecasts. Additionally, the HAFS model struggled with forecasting stratiform precipitation during the first six hours after initialization, a spin-up issue that will likely be addressed with a more advanced data assimilation scheme.  

For question two, we used two approaches to determine the dominant large-scale moisture patterns values of moisture surrounding Atlantic hurricanes and relate these to environmental conditions, storm intensity, and rainfall production. A machine-learning methodology combined with a cluster analysis of 4652 snapshots of total precipitable water revealed four moisture patterns: 1) symmetrically distributed high moisture  accompanied by high rain rates and strong winds, 2) asymmetrically pattern of high moisture to the southeast and low moisture to the northwest sides, 3) moderate levels of moisture in a symmetrical pattern accompanied by low rain rates in a compact shape, and 4) low moisture with an asymmetrical pattern, which was also accompanied by asymmetrically-distributed rain rates, strong vertical wind shear, lower sea surface temperatures, and a lower maximum intensity. The two high moisture patterns were most often present when hurricanes moved over land, while hurricanes in the low moisture pattern usually complete an extratropical transition over the ocean. When applying an Empirical Orthogonal Function analysis that extracts similar patterns but requires independent observations, and omitting cases that interact with land, experience strong vertical wind shear, and/or are from adjacent time steps, we found that the first two dominant patterns explain more than half the variance in moisture. Projecting the original data onto these groups reveals cases that vary in the same direction (positive) or opposite (negative), resulting in four subgroups. Three of these four subgroups exhibited strong similarity to the patterns found via the machine-learning analysis, suggesting that our results are robust.  

In terms of broader impacts, five graduate students (4 from underrepresented groups) and one REU undergraduate student received funding to contribute research to the project. One manuscript has been published, one is under review, six are in preparation for submission, and five of these eight manuscripts have graduate students as lead authors. Team members have presented results at invited seminars, classroom guest lectures, regional and national Meteorology, Geography, and Geosciences conferences, and at a meeting of researchers affiliated with the National Oceanic and Atmospheric Administration. Thirteen of the presentations were delivered as posters or talks by students. Undergraduate and graduate students enrolled in research courses at each of the three universities exchanged knowledge and benefitted from the expertise of the other PIs during a semester-long collaborative teaching effort. The increased understanding of how moisture affects TC rainband development and rainfall distributions will aid forecasters as they interpret satellite products and model output.

Last Modified: 11/04/2024
Modified by: Corene J Matyas