Combining neural networks and geostatistics for landslide hazard assessment of San Salvador metropolitan area, El Salvador

Abstract

This contribution describes the creation of a landslide hazard assessment model for San Salvador, a department in El Salvador. The analysis started with an aerial photointerpretation from Ministry of Environment and Natural Resources of El Salvador (MARN Spanish acronym), where 4792 landslides were identified and georeferenced along with 7 conditioning factors including: geomorphology, geology, rainfall intensity, peak ground acceleration, slope angle, distance to road, and distance to geological fault. Artificial Neural Networks (ANN) were utilized to assess the susceptibility to landslides, achieving results where more than 80% of landslide were properly classified using in-sample and out of sample criteria. Logistic regression was used as base of comparison. Logistic regression obtained a lower performance. To complete the analysis we have performed interpolation of the points using the kriging method from geostatistical approach. Finally, the results show that is possible to derive a landslide hazard map, making use of a combination of ANNs and geostatistical techniques, thus the present study can help landslide mitigation in El Salvador.

Prediction Landslide Location Area Using ANN (Case study in Bangli Regency, Bali Indonesia)

“…In equation ( 3), i p is the output of a neural network with * k input characteristics and * j neurons. This research i p was representative of the event of landslide probability (Rios et al, 2016).…”

“…The MLP consists of three parts: the input layers as neurons representing the value of data; the hidden layer, which demonstrates the network training process; and finally, the output layer. In previous study cases, the MLP was employed to predict the landslide areas (Mohamed et al, 2015;Rios et al, 2016).…”