Bayesian modelling of water resource stress and drought monitoring in southeastern France
Résumé
Drought is a frequent phenomenon in the Mediterranean environment. Recent studies have shown a decrease in rainfall and an increase in temperature in the Mediterranean climate (de Marsily, 2010; Giorgi, 2006; Milano et al., 2013). These trends could worsen by the end of the 21st century (Arnell, 2003; Milano et al., 2013). They would lead to a decrease in water resources. In addition, anthropogenic pressure is increasing in each basin (increasing per capita water consumption). New operational and local means of monitoring and management must be engage.
Depending on the discipline, drought is defined in different ways. Classically (Mishra and Singh, 2010; Wilhite and Glantz, 1985) four main types (atmospheric, edaphic, hydrological and societal) are distinguished. Nevertheless, these different situations must always be defined in terms of social needs, in a given place and time (Charre, 1977; Dorize, 1990). It is necessary to make a clear distinction between what relates to the hydrosystem (aridity, etc.) and the sociosystem (water supply, agricultural water, etc.). Drought, will be defined in this work as a stress between natural supply and anthropic water needs, in a given place and time.
The quantification of water deficit by mass balances is not adapted to large geographical and time scales for managers of small rivers such as the Gardons in the Cévennes. In this work, we propose a new way of measure water stress. This method is based on probability distributions (PD) and on conditional probabilities. First, we need to determine a causal network. Then, From Bayes theorem, we propose a Bayesian network (see Pearl, 2008, for technical aspects) for drought monitoring.
Bayesian networks are not frequently used in the study of natural hazards (Kaikkonen et al., 2021) and even less so in the study of drought. This article proposes a first conceptual formalization of a Bayesian network allowing the monitoring, on a large geographical scale (small catchment area) and over short periods of time (week, month), of the water stress in a Mediterranean context.
The process of creating this model follows several steps (Pollino and Henderson, 2010): (1) determination of the network boundaries, elements and objectives; (2) choice of variables and formatting; (3) definition of the network structure (graph); (4) characterisation of the distributions associated with the variables, (5) the causality system and (6) validation of the network. Each of these steps will be explained.
The Gardon is a pilot basin and representative of water stress problems in a Mediterranean context. The data used for the Bayesian network is centralised in a GIS database called HydroSec.
Origine | Fichiers produits par l'(les) auteur(s) |
---|