INTERNATIONAL JOURNAL OF CLIMATOLOGY Int. J. Climatol. (2013) Published online in Wiley Online Library (wileyonlinelibrary.com) DOI: 10.1002/joc.3770 Impact of climate change on rainfall in Northwestern Bangladesh using multi-GCM ensembles Dipak Kumar, D. S. Arya, * A. R. Murumkar and M. M. Rahman Department of Hydrology, Indian Institute of Technology, Roorkee, India ABSTRACT: Teesta River basin, located in the northwest of Bangladesh, is more vulnerable to floods if compared to other parts of the country. In this context, daily rainfall data of ten raingauge stations located in the catchment of the Jamuneswari River, part of the Teesta River basin, were analysed to study the impact of climate change on rainfall. Length of wet and dry series and mean monthly rainfalls along with their variances were used for validating Long Ashton Research Station Weather Generator (LARS-WG). The analysis was carried out for A1B, A2 and B1 emission scenarios using 15 Global Climate Models GCMs simulations for the periods 2011 – 2030 centred at 2020, 2046 – 2065 centred at 2055 and 2080 – 2099 centred at 2090. The analysis of the data shows that the uncertainty in the prediction increases with the timescale. It was also found that the variability in the predictions is smaller in annual values followed by seasonal. Ensemble of seasonal analysis shows that most of the GCM are in agreement for changes in monsoon season. The LARS-WG has reasonable skill to downscale the point rainfall data and the results obtained so are useful to analyse the impact of climate change on the hydrology of the basin. KEY WORDS rainfall; GCM; statistical downscaling; ensemble; LARS-WG; climate change Received 15 October 2012; Revised 16 May 2013; Accepted 19 May 2013 1. Introduction Atmosphere–ocean coupled Global Climate Models (GCMs) simulate the present and future climate of Earth under different climate change scenarios (SRES, 2000). The computational grid of GCMs is very coarse thus they are unable to skillfully model the sub-grid scale climate features like topography or clouds (Wilby et al ., 2002). Hence, there is a need for downscaling from coarse resolution of a GCM to a very fine resolution or even at the station scale. The available downscaling methodologies are broadly grouped into statistical and dynamical categories. Among the statistical downscaling methods, the use of stochastic weather generators is very popular. They are computationally less demanding, simple to apply and provide station scale information (Coulibaly et al ., 2005; Kilsby et al ., 2007). The weather generators are basically statistical models which are used to generate a long synthetic time series, fill in missing data and produce different realizations of the same data (Wilby, 1999). They employ random number generators and use the observed time series of a station as input. Stochastic weather simulation is not new and has a history starting from 1950s, as reported by Racsko et al . (1991). Among some researchers who contributed to its evolution are Bruhn et al . (1980), Nicks and Harp (1980), Richardson (1981), Richardson and Wright (1984) and * Correspondence to: D. S. Arya, Department of Hydrology, Indian Institute of Technology, Roorkee, India. E-mail: dsarya@gmail.com Schoof et al . (2005). Wilby (1999) has presented a com- prehensive review of its theory and evolution over time. Weather generators have been employed to obtain long time series of hydro-meteorological variables which are then used by crop growth model to forecast agricultural production (Riha et al ., 1996; Hartkamp et al ., 2003) and assessment of risk associated with climate variabil- ity (Semenov, 2006; Bannayan and Hoogenboom, 2008). When the climate scientists started looking for low cost, computationally less expensive and less demanding and quick methods for impact assessment, the weather gen- erator emerged as the most viable solution (e.g. Wilks, 1992; Wilks and Wilby, 1999). Long Ashton Research Station Weather Generator (LARS-WG) is a stochastic weather generator specially designed for climate change impact studies (Semenov and Barrow, 1997). It has been tested for diverse climates and found better than many others (Semenov et al ., 1998). A recent study by Semenov (2009) tested the LARS-WG for rainfall mod- elling at different sites across the world and has shown its ability to model rainfall extremes with reasonable skill. Over the past 15 years, ensemble forecasting became established in Numerical Weather Prediction Centers as a response to the limitations imposed by the inherent uncertainties in the prediction process. The ultimate goal of ensemble forecasting is to predict quantitatively the probability of the state of the atmosphere in future. This article also presents a multi-GCM and multi-scenarios (A1B, A2 and B1) ensemble forecast of rainfall during  2013 Royal Meteorological Society