Experimental tropical cyclone forecasts using a variable-resolution global model

• Published in: Monthly Weather Review
• Other Authors: Zarzycki, Colin (author), Jablonowski, Christiane (author)
• Format: Article in Journal/Newspaper
• Language: English
• Published: American Meteorological Society 2015
• Subjects: Pacific; North Atlantic
• Online Access: http://nldr.library.ucar.edu/repository/collections/OSGC-000-000-022-180; https://doi.org/10.1175/MWR-D-15-0159.1

Tropical cyclone (TC) forecasts at Δx~14-km horizontal resolution (0.125°) are completed using variable-resolution (V-R) grids within the Community Atmosphere Model (CAM). Forecasts are integrated twice daily from 1 August to 31 October for both 2012 and 2013, with a high-resolution nest centered over the North Atlantic and eastern Pacific Ocean basins. Using the CAM version 5 (CAM5) physical parameterization package, regional refinement is shown to significantly increase TC track forecast skill relative to unrefined grids (Δχ~55 km, 0.5°). For typical TC forecast integration periods (approximately 1 week), V-R forecasts are able to nearly identically reproduce the flow field of a globally uniform high-resolution forecast. Simulated intensity is generally too strong for forecasts beyond 72 h. This intensity bias is robust regardless of whether the forecast is forced with observed or climatological sea surface temperatures and is not significantly mitigated in a suite of sensitivity simulations aimed at investigating the impact of model time step and CAM’s deep convection parameterization. Replacing components of the default physics with Cloud Layers Unified by Binormals (CLUBB) produces a statistically significant improvement in forecast intensity at longer lead times, although significant structural differences in forecasted TCs exist. CAM forecasts the recurvature of Hurricane Sandy into the northeastern United States 60 h earlier than the Global Forecast System (GFS) model using identical initial conditions, demonstrating the sensitivity of TC forecasts to model configuration. Computational costs associated with V-R simulations are dramatically decreased relative to globally uniform high-resolution simulations, demonstrating that variable-resolution techniques are a promising tool for future numerical weather prediction applications.