Neural network based automated interpretation algorithm for combined geophysical soundings in coastal zones

The fresh water availability in coastal aquifers is an important problem faced by a major part of world's population dwelling there. For in situ and dynamic characterization of seawater encroachment into coastal aquifers, electrical geophysical methods are better suited. However, the resolving power of such data in distinguishing saline sands from moist clays in the subsurface is very poor. To meet this aspect and also the problem of analyzing voluminous data sets, we propose a feed forward back-propagation neural network (BPNN) based approach for the analysis of combined vertical electrical and induced polarization soundings. Our method is tested on synthetic data computed from available geo-electric sections and prevailing subsurface geological information of coastal aquifers of East Coast of India. The synthetic data are comprised of 18 combined soundings spread over five profiles. 15 out of 18 are used for training the BPNN, while 3 are used for testing. The trained BPNN (one node each in each of the input and output layers and 18 hidden nodes) showed 84.85% accuracy in testing phase for distinguishing clays from saline sands. Our method is also tested on real data concerning a shaly aquifer in Bahia, Brazil yielding an overall accuracy of 84.9%, comparable to that of synthetic case; thereby validating our approach.