The goal of the project is to develop accurate and easy to use seismic hazard forecast models utilizing the neural network approach. The neural network is a very versatile and efficient approach. The main advantages of using the neural network approach in this project are that (1) it is not necessary to predecide about the mathematical form for the model in this approach, (2) it can include a large number of variables without unduly increasing the computational effort and (3) it can easily update and improve the model as more data becomes available. It is proposed to develop models to predict the spatial distribution of Modified Mercelli Intensity for the western, mid- western and eastern United States. Also the models to predict the ground motion parameters such as peak ground acceleration and response spectra will be developed for the western United States. The proposed study will be conducted over a period of two years. During the first year, data collection, processing, and model formulation will be undertaken and generic models for each of the three regions of the country will be formulated. The second year will be devoted to rigorous testing and fine-tuning of the models. Sensitivity studies will be performed to identify the important input variables.. Comparison with conventional seismic hazard/risk analysis models will be made to evaluate the performance of the proposed models. Finally, application computer softwares will be developed for public dissemination.
