This project attempts to develop an adaptive and non-parametric Modified Mercalli Intensity (MMI) forecast model for the direct prediction of spatial distribution of MMI corresponding to an earthquake scenario for use in regional risk/damage assessment. The development of the model is based on recent advances in Adaptive Pattern Recognition (APR). The APR predictors of the MMI forecast model are to be constructed using neural networks implementation through supervised learning utilizing historical earthquake and regional geological data as training sets. The proposed model represents an improvement and simplification of the risk/damage assessment methodology in current use. By combining two steps, the attenuation of peak ground acceleration (PGA) and the PGA-MMI conversion, into one and using an optimal nonfunctional and incremental model, the accuracy of risk/damage analysis results are improved through the reduction of uncertainties, e.g., the PGA-MMI conversion, and the inclusion of additional site-specific independent variables affecting MMI levels. The model will be applied for regional risk assessment for the major metropolitan regions in California, namely the San Francisco Bay and the Greater Los Angeles metropolitan areas. A generalized computer code will also be developed and made available for application to these and possibly other seismic locations in the world at international levels.
