Cumulonimbus Prediction using Artificial Neural Network Back Propagation with Radiosonde Indeces

This research purpose an accurate quantitative forecasting of Cumulonimbus (Cb) Cloud development events. Neural network will use to develop a quantitative cumulonimbus events forecasting model that highly improve forecasting skill especially in Tropical Area. Cumulonimbus storm cells can produce torrential rain of a convective nature and flash flooding, as well as straight-line winds. Most storm cells die after about 20 minutes, when the precipitation causes more downdraft than updraft, causing the energy to dissipate. If there is enough solar energy in the atmosphere as in tropical are, the moisture from one storm cell can evaporate rapidly—resulting in a new cell forming just a few miles from the former one. This can cause thunderstorms to last for several hours. Cumulonimbus clouds can also bring dangerous storms which bring lightning, thunder, and torrential ice. The technique is Back propagation neural network (BPN), the advantages using this engine are more reasonably to estimated a large class of functions and efficient than numerical differentiation. The BPN inputs were not only raw sounding observation data also derived indices value, using Principle component analysis (PCA) as feature selection processing. PCA has been widely known as method to reduce the dimension of input to multivariate data by minimizing loss of information input. In this research PCA is used to reduce the dimension of input to the BPN and reconstruct the new input data. Clustering and initialization process centers on neural network done with the Self Organizing Map (SOM) technique and the determination of the weight of the hidden center during the learning process using the algorithm Recursive Orthogonal Least Square. The initial result show that the combination of dataset engine are workable, the proposed technique result in better accuracy of prediction and can implemented in operational used for early warning system to reduce destruction impact from weather hazard in tropical area