Agriculture in Bangladesh Agriculture is a pillar of Bangladesh’s economy, using more than70% of land area (FAOSTAT, 2009) and accounting for nearly 20% of gross domestic product and 65% of the labor force, employed primarily on small-holder farms (Yu et al., 2010). Rice production occurs on more than 80% of agricultural lands and is grown in three growing seasons that span the entire year and are synonymous with their rice crop: aman, boro, and aus (BBS, 2005, 2008). Aman grows during the monsoon season when rainfall is plentiful. Boro is grown during the dry season, after floods have receded, and is restricted to irrigated areas. Aus production takes advantage of rainfall during the spring transition to ward the monsoon that enables a short growing season, although the sufficiency of rains varies from year to year. Rice cultivation makes up nearly 95% of cereal production in Bangladesh, with wheat a prominent dry-season crop grown with irrigation (Bangladesh Bureau of Statistics; BBS, 2008). Rice is threatened by floods during the pre-monsoon (aus) and monsoon (aman) seasons and by heatwaves and water scarcity during the dry monsoon (boro) season. Boro is currently the most productive season, followed by aman, with aus considerably lower due mostly to smaller planted area (BBS, 2008). Historical cereal production has been increasing for decades, rising from about 10 million metric tons (MT) in the early1970s to almost 30 MT in 2001 (USDA, 2008). Annual variations and long-term trends in agricultural production come from many factors, including climate events (particularly floods), resource limitations, socio political events, and the implementation of modern agricultural practices, so isolating the sensitivity of historical agricultural production to climate factors is challenging. 3. Biophysical approach and model framework 3.1. Models and methods Biophysical process models in this study occupy a range of scales, with crop models focusing on representative farms throughout Bangladesh, river floods simulated over the entire GBM Basins, and coastal inundation determined along the Bangladesh coastline. Differences in the computational resources available for each model limited the ability to set up identical ensembles of climate drivers for each set of model simulations. However, the full range of crop model experiments provided a framework for integrating the highest possible number of river flood and sea level rise simulations and their impact on agricultural lands. This modeling framework was described in Yu et al. (2010), and is therefore only summarized below and in the Supplementary Online Material Appendices. Climate scenarios are generated according to local historic conditions and simulated climate changes from global climate model (GCM) output contributed to the World Climate Research Programme’s (WCRP’s) Coupled Model Inter comparison Project phase 3 (CMIP3) multi-model dataset (Meehl et al., 2007) at the Program for Climate Model Diagnosis and Inter comparison (PCMDI;http://www-pcmdi.llnl.gov). These climate simulations were analyzed by the studies reviewed in the Inter governmental Panel on Climate Change (IPCC) Fourth Assessment Report (AR4).Climate change scenarios are generated by comparing a given GCM’s 2040–2069 period (referred to as the ‘‘2050s’’) with control simulations of that GCM over a 1970–1999 baseline, and then imposing these changes on historical observations (daily rainfall and maximum and minimum temperatures from BMD observations; solar radiation data and short gaps were filled using the Weather-man Utility; Hoogenboometal., 2003).This delta method (described more completely in Wilby et al., 2004, and Yu et al., 2010) offsets many of the common GCM biases but assumes no change in high-frequency variability or the frequency of rain events. Both a relatively high (A2) and low (B1) future emissions pathway were analyzed and compared to the baseline period, with crop models utilizing representative CO2 concentrations of 556 ppm, 498 ppm, and 345 ppm, respectively, set according to projected or observed concentrations in the central year of 30-year the period being studied (SRES, 2000). Following Yu et al. (2010), an ensemble of climate scenarios were created for each sub-region from 16 GCMs and 2 emissions scenarios, capturing a consistent temperature rise and wide uncertainty among projected precipitation changes.The MIKE BASIN hydrologic model was employed over theentire GBM Basins that drain through Bangladesh to simulate river floods for each baseline and future year. Flood protection infrastructure in Bangladesh was also taken into account by a Bangladeshi flood model (Nishat and Rahman, 2009; Hopson and Webster, 2009). Coastal inundation in the 2050s was simulated bythe UK Department for Environment, Food, and Rural Affairs incollaboration with IWM and CEGIS (DEFRA, 2007) using the MIKE21 Two-Dimensional Estuary Model and mean sea levels of 27 cm (A2) and 8 cm (B1) above baseline. More details about the flood models are provided in Appendix II of the Supplementary Online Material. Process-based crop model simulations were run with the Crop Environment RE source Synthesis (CERES) rice and wheat models from the Decision Support System for Agrotechnology Transfer (DSSAT v4.5.0.030; Hoogenboom et al., 2003; Jones et al., 2003). CERES models require detailed soil profiles, farm-level management practices, genetic coefficients describing the cultivar, and daily meteorological conditions (rainfall, solar radiation, and maximum and minimum temperature). CERES models have been applied extensively in Bangladesh, although less frequently for climate change applications (e.g., Karim et al., 1996; Karim et al., 1994, 1998; Hussain, 2006; Basak et al., 2009). These models include the beneficial effects of enhanced carbon dioxide concentrations on plant growth and are sensitive to the phenological stage of plant development when particular stresses occur, but the exact relationships remain uncertain. For this study the CERES model was calibrated for Bangladeshi sub-regions using soil and cultivar information provided by S.G.Hussain (2008), additional soil profiles drawn from Brammer (1996), and farm-level management according to practices recommended by the Bangladesh Rice Research Institute (BRRI, 2007). Crop model simulations for this study utilize modern cultivars and relatively high fertilizer application rates and assume pest-, salinity-, and disease-free conditions. This management configuration follows the recommendations of the Bangladesh Rice Research Institute (BRRI, 2007; but is a more intensive management than is typical for many farms in the region. Mahmood et al. (2003)noted the large gap between actual and potential yields in Bangladesh, with actual yield averages approximately 1/6th of the potential yields produced under unstressed, high-input conditions protected from floods. Projections of future yields provide an indication of the magnitude and direction of climate impacts which will occur on top of the anticipated reduction in this yield gap through economic development.