Cropping system features: Cropping systems in Bangladesh are very complex, highly intensive and diverse, and are continuously evolving and changing (Timsina and Connor, 2001). There are three main cropping seasons: (i) aman or kharif or monsoon (also called kharif-2) from June/July to September/ October, (ii) rabi or winter from October/November to February/ March, and (iii) spring or pre-kharif or pre-monsoon (also called kharif-1) from March/April to June/July. In kharif-2, rice (called transplanted aman or T. aman) is the predominantly grown crop (N90% of area), mostly under rainfed conditions. During the dry rabi season a wide range of crops, including rice (called boro), wheat, maize, pulses (chickpea, lentil and field peas), potatoes and oilseeds (e.g., mustard) are grown. In kharif-1, short-duration cultivars of maize, pulses (mungbean, cowpea) and rice (called aus) are grown. Boro and rabi (winter) maize are either fully or partially irrigated, while aman and aus rice and kharif-1 (spring) maize are predominantly rainfed, with some crops applied with partial irrigation. Wheat is also predominantly grown with full irrigation (~80%), with the remaining 20% under either partial irrigation or strictly rainfed. Thus, rice rice (R-R), rice-wheat (R-W), and rice-maize (R-M) are the dominant systems, which often include an additional crop such as aus rice or maize in kharif-1 mostly grown under partial irrigation or rainfed conditions. While R-R is the common rotation in tropical and sub-tropical areas with warm climates in Bangladesh and entire South Asia, R-W and R-M rotations are practiced in the subtropical areas with mild winters (Timsina and Connor, 2001; Timsina et al., 2010, 2011). Calculation of yield potential and yield gaps: Cereal supply and demand calculations Domestic supply of each of the three cereals for the base year 2010 was calculated as mean farmers' yield times the harvested area per crop for 2010 based on data reported by Bangladesh Bureau of Statistics - BBS (BBS, 2013) and FAO (FAOSTAT, 2015). Total maize consumption for Bangladesh in 2010, as estimated by the IMPACT model (Robinson et al., 2015), was very low in comparison to FAO food balance data, and predicted to increase little towards 2050. Since IMPACT maize consumption was estimated based on data collected prior to 2000 (Islam, 2003), it is likely that these estimates have not accounted for the rapid development of the poultry and fish production in Bangladesh during the last decade, resulting in increased demand for maize (Miah et al., 2013). Hence, we decided not to use the IMPACT data for maize but to use the FAO food balance. There was good agreement between IMPACT and FAO food balance data on rice and wheat consumption. Current total cereal demand (expressed in kg milled rice) was calculated by multiplying the 2010 UN population data (UN, 2015) with total demand per capita per cereal (inclusive that for animal and fish feeds) in 2010, derived from the FAO food balance for maize (FAOSTAT, 2015) and from IMPACT for wheat and rice (Robinson et al., 2015). To convert maize and wheat to milled rice equivalents, we used the ratios between the caloric contents of maize and wheat versus that of rice, i.e., 3600 kcal kg−1 milled rice grains, 2730 kcal kg−1 wheat grain and 3180 kcal kg−1 maize grain (FAOSTAT, 2015). Finally, SSR was calculated by dividing the current cereal production by the current cereal food and feed demand, assuming a waste of 15% based on the FAO food balance (FAOSTAT, 2015). Average Ya of rice, maize and wheat for 2010 in Bangladesh were calculated based on Ya data reported over the 2008–2012 time period (BBS, 2013). Yp (for all irrigated crops) and Yw (for rained rice) were simulated using ORYZA (v3) for rice (Bouman et al., 2001), Hybrid Maize for maize (Yang et al., 2004), and WOFOST for wheat (Wolf et al., 2011). These models were calibrated and validated for Bangladesh based on primary and secondary data for crop phenology, and soil and yield data recorded from well-managed experiments conducted in major agriculture districts (Comilla, Dinajpur, Gazipur, Rajshahi, Rangpur) during 2010 to 2014 (Hossain and Silva, 2013; Islam, 2016). Yw was not simulated for wheat and maize in this study because b20% of the total area cultivated with these two crops is rainfed. Instead, we estimated Yw of maize and wheat based on the existing field and modeling data from the literature (Ali et al., 2008; Carberry et al., 2011; Timsina and Humphreys, 2006). Cereal supply and demand calculations: Domestic supply of each of the three cereals for the base year 2010 was calculated as mean farmers' yield times the harvested area per crop for 2010 based on data reported by Bangladesh Bureau of Statistics - BBS (BBS, 2013) and FAO (FAOSTAT, 2015). Total maize consumption for Bangladesh in 2010, as estimated by the IMPACT model (Robinson et al., 2015), was very low in comparison to FAO food balance data, and predicted to increase little towards 2050. Since IMPACT maize consumption was estimated based on data collected prior to 2000 (Islam, 2003), it is likely that these estimates have not accounted for the rapid development of the poultry and fish production in Bangladesh during the last decade, resulting in increased demand for maize (Miah et al., 2013). Hence, we decided not to use the IMPACT data for maize but to use the FAO food balance. There was good agreement between IMPACT and FAO food balance data on rice and wheat consumption. Current total cereal demand (expressed in kg milled rice) was calculated by multiplying the 2010 UN population data (UN, 2015) with total demand per capita per cereal (inclusive that for animal and fish feeds) in 2010, derived from the FAO food balance for maize (FAOSTAT, 2015) and from IMPACT for wheat and rice (Robinson et al., 2015). To convert maize and wheat to milled rice equivalents, we used the ratios between the caloric contents of maize and wheat versus that of rice, i.e., 3600 kcal kg−1 milled rice grains, 2730 kcal kg−1 wheat grain and 3180 kcal kg−1 maize grain (FAOSTAT, 2015). Finally, SSR was calculated by dividing the current cereal production by the current cereal food and feed demand, assuming a waste of 15% based on the FAO food balance (FAOSTAT, 2015).