Blind-testing experiments for interpreting spatial-prediction patterns of landslide hazard
Free (open access)
Volume 6 (2016), Issue 2
193 - 208
ANDREA G. FABBRI & CHANG-JO CHUNG
In this contribution, we analyse three separate databases in case study areas each suggestive of particular strategies to better portray their predictive power. A database in north-eastern Spain is used to separate sub-areas, with hopefully more compatible geomorphologic settings. Another database in central Portugal offers the opportunity of representing the uncertainty of predicted hazard class membership via iterative cross-validation with systematically partitioned landslide occurrences. A third database in central Slovenia is used to interpret the predictive qualities of two dynamic types of landslides: one that is relatively well predicting and the other one poorly predicting. The diversity of the experiments and their results, point at strategies of blind-testing, still unexploited in spatial prediction modelling, that are not necessarily limited to the landslide hazard domain.
blind testing, cross-validation, empirical likelihood ratios, landslide hazard, prediction patterns, prediction-rate curves, target pattern