Non-stationary analysis of dry spells in monsoon season of Senegal River Basin using data from Regional Climate Models (RCMs)

The Senegal River Basin, located in West Africa, has been affected by several droughts since the end of the 1960s. In its valley, which is densely populated and highly vulnerable to climate variability and water availability, agricultural activities provide the livelihood for thousands of people. Increasing the knowledge about plausible trends of drought events will allow to improve the adaptation and mitigation measures in order to build “adaptive capacity” to climate change in West Africa. An innovative methodology for the non-stationary analysis of droughts events, which allows the prediction of regional trends associated to several return periods, is presented. The analyses were based on Regional Climate Models (RCMs) provided by the European ENSEMBLES project for West Africa, together with observed data. A non-stationary behaviour of the annual series of maximum length of dry spells (AMDSL) in the monsoon season is reflected in temporal changes in mean and variance. The non-stationary nature of hydrometeorological series, due to climate change and anthropogenic activities, is the main criticism to traditional frequency analysis. Therefore, in this paper, the modelling tool GAMLSS (Generalized Additive Models for Location, Scale and Shape), is applied to develop regional probability density functions (pdfs) fitted to AMDSL series for the monsoon season in the Senegal River Basin. The skills of RCMs in the representation of maximum length of dry spells observed for the period 1970–1990, are evaluated considering observed data. Based on the results obtained, a first selection of the RCMs with which to apply GAMLSS to the AMDSL series identified, for the time period 1970–2050, is made. The results of GAMLSS analysis exhibit divergent trends, with different value ranges for parameters of probability distributions being detected. Therefore, in the second stage of the paper, regional pdfs are constructed using bootstrapping distributions based on probabilistic models. In general, an increase in the mean and variance statistics of AMDSL at regional level are predicted, thereby increasing the lengths of dry spells associated with a low probability of occurrence (related to high return period) in the monsoon season.