STATISTICAL MODELING OF ATLANTIC HURRICANE ACTIVITY USING ATMOSPHERIC REANALYSES AND IPCC SIMULATIONS AND PROJECTIONS
STATISTICAL MODELING OF ATLANTIC HURRICANE ACTIVITY USING ATMOSPHERIC REANALYSES AND IPCC SIMULATIONS AND PROJECTIONS
Files
Publication or External Link
Date
2013
Authors
Kim, Kye-Hwan
Advisor
Nigam, Sumant
Citation
DRUM DOI
Abstract
Atlantic hurricane activity has increased in recent decades leading to extensive investigation of its links with rising greenhouse gases (GHG). The International Panel on Climate Change (IPCC) is investigating how global-regional climate will change in the 21st century (21C) in response to GHG emissions but the deployed climate model grids are too coarse to resolve hurricanes. Projections of hurricane activity must thus be obtained, indirectly, from regional downscaling of climate using statistical and dynamical approaches; the former is adopted here.
Hurricane counts are reconstructed in 20C and projected in 21C from statistical modeling with Atlantic Sea-Surface Temperature (SST), static-stability, and zonal-wind shear as predictors. Optimal predictor definitions, including geographical domains, are identified from observational analysis (1958-2005). The viability of the statistical approach is demonstrated from the successful reconstruction of hurricane counts in both training and independent periods, with reconstruction-observation correlations (0.72-0.86) higher than Kim and Webster's (2010; the best statistical model to date).
Statistical models for counts were also developed for each of the five analyzed IPCC- the-fifth-Assessment (AR5) models, based on their 1958-2005 simulations. The long-term trend in counts was modeled using multivariate linear regression with predictors from the ensemble-mean simulation. On the other hand, predictors from several ensemble members were used via the best subset regression technique when modeling the interannual-to-decadal count variability. Focusing on the observationally rich 1958-2005 period provided critical evaluation and a basis for selection of a credible model-subset. The 21C projections of hurricane counts by this model subset will be deemed more trustworthy than the average of all AR5-based models.
Modeling of Atlantic hurricane activity with IPCC-AR5 predictors shows a stronger count-trend in the 21C, principally, from increasing SSTs; AR5-models disagree on the trend in zonal-wind shear which is a less influential predictor in the AR5-based models compared to observations.
Decadal predictions of hurricane activity for the independent but observations-available 2006-2010 period show the promise of the best subset regression models, especially with predictors from the IPCC-Decadal (2005 ocean-initialized) experiments. According to this model, the 2013 hurricane season will be slightly more active (+1 count) but not as much as NOAA's forecast (+3 counts).