, 2007). In most climate change studies, GCMs have been used to project future climatic variables. However, due to a limitation of GCMs to incorporate local topography (spatial and temporal scales), the direct use of their outputs in impact studies on the local scale of e.g. hydrological catchments is

limited. To bridge the gaps between the climate model and local scales, downscaling is commonly used in practice. Dynamic downscaling and statistical downscaling are the most commonly used methods (Bergstrom, 2001, Fowler et al., 2007, Pinto et al., Ibrutinib 2010, Schoof et al., 2009 and Wilby et al., 1999). Dynamic downscaling by Regional Climate Models (RCMs) ensures consistency between climatological variables, however they are computationally expensive. Statistical downscaling models, on the other

hand, are based on statistical relationships and hence require less computational time. Extensive research has been carried out with both approaches (e.g., Chen et al., 2012, Maraun et al., 2010, Teutschbein et al., 2011 and Willems and Vrac, 2011). Besides the scale issue, there is often a clear bias in the statistics of variables produced by GCMs such as rainfall and temperature (Kay et al., 2006 and Kotlarski, 2005). Therefore hydrologically important variables need to be adjusted to obtain realistic time series for use in local impact studies (Graham et Dinaciclib al., 2007). A conventional way to adjust future time series is referred to as bias correction (Lenderink, 2007) where correction factors are derived by comparing the GCM output with observed weather variables in the reference period, and then applied to GCM output for future climate. While bias-correction generally reproduces the variability described by different climatic conditions simulated by GCM projections, one disadvantage is the assumption of stationarity, i.e. that the correction Phospholipase D1 factors do not change with time. As indicated by Rana et al. (2012), the rainfall intensity and frequency

for Mumbai is related to certain global climate indices such as the Indian Ocean Dipole, the El Niño-Southern Oscillation and the East Atlantic Pattern. These established connections between local rainfall and large-scale climate features suggest the possibility to statistically downscale GCM data directly to the local scale. The objective of this paper is to apply a statistical approach termed Distribution-based Scaling (DBS) technique, which has been tested and applied to RCM data, to scale GCM data. This includes the application of the DBS model to GCM projections for the area, an analysis of the scaling methodology and its applicability to GCM data, and finally assessment of the future impacts on the city of Mumbai due to climate change as projected by nine different GCM projections. The study is carried out for the city of Mumbai, (18°58′30″ N, 72°49′33″ E; formerly Bombay) the capital of Maharashtra state, located in the south-western part of India.