Event

The UWI Department of Physics hosts the MPhil Assessment Seminar of Ms. Nkese Mc Shine, on Thursday 4^th November, 2010, at 2:45 pm, at the Faculty of Science and Agriculture, Room 413. Ms. Mc Shine will present on the topic Caribbean Precipitation: an Alternative Prediction Approach, under supervisor Dr. Ricardo Clarke. 

Abstract

Research work on the Caribbean climate is performed on a large scale as a result of limited knowledge and access to relevant data sets. Many oceanic and atmospheric variables influence the precipitation in the Caribbean as a result of the geographic location. Large scale analyses have been used in most of the linear precipitation prediction models for the Caribbean. These models however, either under predict or over predict for certain seasons. At present, there are proposed prediction models for both the dry and rainy seasons. The two statistical models for the dry season are: Canonical Correlation Analysis (CCA) and Jamaican rainfall index. The CCA model is based on the El Niño Southern Oscillation (ENSO) signal. For the rainy season there are separate models for the early wet season (May, June and July) and late wet season (August, September and October) using two predictands. These are the Caribbean Rainfall Variability Index (CPINDX) and the Detrended Caribbean Precipitation Index (DCPINDX) for both seasons. The hit rates for the rainy season models are between 70-75%. An Artificial Neural Network (ANN) is a non-linear prediction model used to analyze large scale data with numerous parameters. A hybrid approach is produced when the deterministic and stochastic models are combined. Therefore, an artificial neural network would be trained to predict Caribbean precipitation. For the same data input, this non linear model is expected to predict more accurately than standard models. With this new approach, a better precipitation prediction can be used in the prevention of the effects of drought and flooding caused by intense variations in the Caribbean precipitation cycles.