Sensitivity analysis of land cover change prediction model in the presence of aleatory and epistemic imperfection
Résumé
Making accurate decision about forthcoming Land Cover Change (LCC) are generally complex. Besides, input parameters for LCC prediction systems are varied and married by imperfection that have a significant influence on out results of these systems. This imperfection is divided into two classes: aleatory imperfection and epistemic imperfection. Studying the effect of these parameters on systems output can help improving decision. Sensitivity Analysis (SA) has an important role in the identification and reduction of the imperfection. In literature, Sobol indices, are most popular. However, they have computational cost and time demanding. Recently, the Derivative-based Global Sensitivity Measure (DGSM) appears to overcome this problem. In this paper, we present a SA approach to address both types of imperfections related to LCC prediction model taking into account correlation among parameters. Performances of the proposed approach are proved using several real-world data sets representing the Port region, Reunion Island. Experiments made demonstrate the effectiveness and the efficiency of the proposed approach.