Essay --

Review of â€Å"Prediction Models for Annual Hurricane Counts† ELserner, J. (2006). Prediction Models for Annual US Hurricane Counts. American Meteorological Society, 2935-3951. HURRICANES This paper provides a Bayesian approach towards developing a prediction model for the occurrence of coastal hurricane activity based on historic hurricane data from 1851 to 2004 from US National Oceanic and Atmospheric Administration. A hurricane is defined as a tropical cyclone with maximum sustained (1min) 10-m winds of 65kt (33 m s-1) or greater. [1]A Hurricane landfall occurs when a storm passes over land after originating in water. A hurricane can make more than one landfall. A landfall may occur even when the exact centre of low pressure remains offshore(eye) as the eyewall of the hurricane extends a radial distance of 50km. The literature review in the paper suggests a significant effect of El Nino Southern Oscillations (ENSO) on the frequency of hurricanes forming over topics and a less significant effect over sub tropics. The North Atlantic Oscillation (NAO) also plays an important role in altering hurricane activity (Elsner 2003; Elsner et al. 2001; Jagger et al. 2001; Mur nane et al 2000) has been stated. The hurricane observations considered in the model fulfills the following criteria 1 The storm hits the US continent atleast once at hurricane intensity. 2 The storm is recorded in the US continent only except Hawaii, Puerto Rico, Virgin Islands The discrepancy associated with the available data of hurricanes is about the certainty of the records for before 1899 ie the hurricane record from 1851-1898 are less certain than records available after 1899. The challenge here is to achieve such a model that gives accurate predictions even if t... ...June. Therefore the partial season count excludes hurricanes of May (1 occurred) and June (19 occurred) from the total of 274 hurricanes from 1851 to 2004. A total of 20% data is eliminated from 274 hurricanes. MODEL FOR ANNUAL HURRICANE COUNT POISSON REGRESSION MODEL h≈ Poisson (lamdai ) lamdai =exp(ÃŽ ²o+ X`i ÃŽ ²) Ln(lamdai)= ÃŽ ²o+ X`i ÃŽ ² ÃŽ ²o and ÃŽ ² define a specific model and are calculated on Bayesian approach. The model assumes the parameters (intercept and coefficient) to have a distribution and that inference is made by computing the posterior probability density of the parameter conditioned on the observed data. The Bayesian approach combines Prior belief [ f(ÃŽ ²) ] and most frequent likelihood to give the posterior Density: f(ÃŽ ²|h) proportional f(h/ ÃŽ ²).f(ÃŽ ²) The posterior density talks about the belief of parameter values after considering the observed counts.