Flood risk under future climate in data sparse regions: Linking extreme value models and flood generating processes Yves Tramblay a,⇑ , Ernest Amoussou b,c , Wouter Dorigo d , Gil Mahé a a IRD – HydroSciences Montpellier, UMR5569, Université de Montpellier 2, Case courrier MSE, Place Eugène Bataillon, 34095 Montpellier cedex 5, France b Département de Géographie et Aménagement du Territoire de l’Université de Parakou, BP 123 Parakou, Benin c Laboratoire Pierre PAGNEY, Climat, Eau, Ecosystème et Développement (LACEEDE), Université d’Abomey-Calavi, 03 BP1122 Cotonou, Benin d Remote Sensing Research Group, Department of Geodesy and Geoinformation Vienna University of Technology, Vienna, Austria article info Article history: Received 22 May 2014 Received in revised form 16 July 2014 Accepted 26 July 2014 Available online 12 August 2014 This manuscript was handled by Konstantine P. Georgakakos, Editor-in-Chief, with the assistance of Ashish Sharma, Associate Editor Keywords: GEV Soil moisture Non-stationary RCM Flood risk summary For many areas in the world, there is a need for future projections of flood risk in order to improve the possible mitigation actions. However, such an exercise is often made difficult in data-sparse regions, where the limited access to hydrometric data does not allow calibrating hydrological models in a robust way under non-stationary conditions. In this study we present an approach to estimate possible changes in flood risks, which incorporates flood generating processes into statistical models for extreme values. This approach is illustrated for a West African catchment, the Mono River (12,900 km 2 ), with discharge, precipitation and temperature data available between 1988 and 2010 and where the dominant flood gen- erating process is soil saturation. A soil moisture accounting (SMA) model, calibrated against a merged surface soil moisture microwave satellite dataset, is used to estimate the annual maximum soil saturation level that is related to the location parameter of a generalized extreme value model of annual maximum discharge. With such a model, it is possible to estimate the changes in flood quantiles from the changes in the annual maximum soil saturation level. An ensemble of regional climate models from the ENSEM- BLES–AMMA project are then considered to estimate the potential future changes in soil saturation and subsequently the changes in flood risks for the period 2028–2050. A sensitivity analysis of the non-stationary flood quantiles has shown that with the projected changes on precipitation (2%) and temperature (+1.22°) under the scenario A1B, the projected flood quantiles would stay in the range of the observed variability during 1988–2010. The proposed approach, relying on low data requirements, could be useful to estimate the projected changes in flood risks for other data-sparse catchments where the dominant flood-generating process is soil saturation. Ó 2014 Elsevier B.V. All rights reserved. 1. Introduction The vulnerability to floods has increased in Africa during the recent decades (Douglas et al., 2008; Di Baldassarre et al., 2010). Therefore, future projections of flood risks using climate models are needed in order to improve the mitigation actions. Probably the most common method to evaluate the climate change impacts on hydrology is the top-down approach: the outputs of general cir- culation models (GCM) are downscaled to the catchment of inter- est and subsequently run into a hydrological model to evaluate the climate change impacts. It is well known that there are potentially great uncertainties at the different levels of this type of approach, namely in the downscaling methods (Fowler et al., 2007; Teng et al., 2012) and the validity of hydrological models under different climatic conditions (Wilby, 2005; Peel and Blöschl, 2011). Indeed, a great problem faced when conducting climate change impact stud- ies on flood risks in Africa is the lack of data available to calibrate hydrological models in a robust manner. The same statement applies to other data-sparse regions of the world. Physical-based models could be useful tools to estimate the climate change impacts on hydrological processes (Dankers and Feyen, 2009; Cornelissen et al., 2013) but they require a great amount of hydro- metric and physiographic characteristics data that is often not available in the target catchments. On the other side, conceptual hydrological models require less input data but several studies have warned about their use under non-stationary conditions (Vaze et al., 2010; Amoussou et al., 2014). Indeed the calibrated parameters of conceptual models may be dependent of climatic conditions (Merz et al., 2011). Beside top-down modeling chains and their potential limita- tions, a growing number of studies have considered bottom-up http://dx.doi.org/10.1016/j.jhydrol.2014.07.052 0022-1694/Ó 2014 Elsevier B.V. All rights reserved. ⇑ Corresponding author. E-mail address: ytramblay@gmail.com (Y. Tramblay). Journal of Hydrology 519 (2014) 549–558 Contents lists available at ScienceDirect Journal of Hydrology journal homepage: www.elsevier.com/locate/jhydrol