Natural hazard analysis often makes use of alphanumeric data sets whose content is mainly expressed using natural language expressions that describe where and when events such as eruptions, avalanches, floods , etc. took place. However, a lot of expressions used in natural language are vague and imprecise and especially those which are used to locate places (around, near to, north of, etc.) and dates (between 1710 and 1711, at the beginning of the century, etc.). Some works propose approaches to define a spatial expansion representing the localization of any vague spatial information. Yet, these approaches do not allow to define any spatial or temporal localization that takes into account their physical environment: topographical or meteorological elements of the studied ground, nature and the characteristics of the observed phenomenon, etc. We propose a methodology that allows to improve the localization and the cartographic representation of imperfect spatial and temporal information. This methodology relies on a specific extension of Spatial ML and TimeML markup languages and an ontology for the integration, quantification and representation of imperfect spatio-temporal information expressed in natural language in order to locate natural hazard phenomena. In this paper, we present the elements of the STELLA (Spatial - TEmporal LocaLizAtion) methodology and framework .