SENSITTVITYAND RELIABILITY ANALYSIS OF ESTIMATION PROCEDURES OF UNOBSERVED TRIGGER FACTOR FOR LANDSLIDE HAZARD MAPPING

摘要

For better supporting landslide hazard zonation, this paper presents a sensitivity and reliability analysis of estimation procedures of unobserved trigger factor (i.e., latent variable) on landslide occurrences based on Structural Equation Modeling (SEM) approach. Through a measurement equation in the path diagram defined between the trigger factor and causal factors, the scores of unobserved trigger factor are estimated, that are delineated on a Trigger Factor Influence (TFI) map. As the sensitivity and reliability analysis of estimation procedures, let us consider the following models: i)Model A: using Maximum Likelihood estimation (ML) and Generalized Least Squares (GLS) to estimate the parameters in the path diagram and the latent valuable scores of trigger factor, respectively; and ii)Model B: using ML and GLS to the same as above. The experimental results are summarized as follows: i)Based on the statistical fit measures (i.e., GFI, AGFI, and RMSEA), both Models can be accepted and; ii)The success rates of Model A with respect to training data are higher than that of Model B. Those results corroborate that the higher the trigger factor value of Model A, the more hazardous the subarea affected by landslide occurrences. As a final outcome, the differences of trigger factor values between Models A and B are delineated on the difference map (DIF map). Based on the above quantitative evaluation results, the pixels on the DIF map that indicate the difference values of "Model A > Model B" and "Model A < Model B" can be interpreted as "reliable- and unreliable-side assessment" subareas, respectively. Such approach on sensitivity and reliability analysis is effective in estimating unobserved trigger factor for landslide hazard mapping.