To provide estimates of fish production potential under climate change, we used a combination of atmospheric, hydrological, ocean circulation, and ocean biogeochemical models, driving changes in ocean productivity and fisheries potential. These models produce yearly time-steps and spatially resolved results. The Bay of Bengal (BoB) physico-biogeochemical model simulates the cycling of the main nutrients through the benthic and planktonic pelagic ecosystems. Outputs from this model drive two fisheries models: a size-spectrum model to provide time-series of total marine fish production by size, and a species-based model to compute total marine production of the main species by size. The effects of human activities and fisheries management policies were explored through fishing mortality scenarios. Details of each model are provided below. Climate and hydrological models Previous climate modeling studies in this region have tended to focus on the wider Indian subcontinent rather than Bangladesh.Most studies project a generic increase in atmospheric temperature, annual rainfall, and heavy precipitation events. Climate data for this study were taken from the UK Met Office regional climate model (RCM) HadRM3P, which is dynamically downscaled from the global circulation model HadCM3 (Caesaret al., 2015). The green-house gas emission scenario used here is the Special Report on Emissions Scenarios (SRES) A1B (IPCC, 2007), a medium-high emissions scenario developed for the Intergovernmental Panel on climate Change (IPCC) 3rd and 4th Assessment Reports, which still underpins research into climate impacts. To capture some of the model uncertainties, we considered three different climate pro-jections from a 17-member ensemble of HadCM3 runs (Caesaret al., 2015). The three climate model runs selected correspond to a range of possible future outcomes for the BoB, between the stand-ard (Q0), drier and warmer (Q16), and intermediate rainfall and temperature (Q8) projections (Caesaret al., 2015). Q0, Q8, andQ16 have a successively higher sensitivity to greenhouse gas forcing because of the different parameter values used in the general circulation model for these ensemble members. As delta regions are particularly sensitive to precipitation and river run-off, outputs from an Integrated Catchment Model (INCA, Whitehead et al., 2015a,b)were used to determine run-off and associated nutrient loadings from the delta rivers into the BoB for each projection. The model simulates factors controlling flow and water quality dynamics in both land and stream components of river catchments. The INCA model application took account of both climatic scenarios (Q0, Q8, Q16) and patterns of upstream water use according to three socio-economic scenarios (Less Sustainable, LS, Business as Usual, BaU, More Sustainable, MS; Whitehead et al., 2015a,b). We used the results of the Q0-BAU, Q8-LS, and Q16-MS INCA simulation runs to capture the variation of the simulated river flows and nutrient loads. The rivers in the Ganges–Brahmaputra–Meghna delta region (Figure2) account for 40% of flow into the model domain. For all other rivers, for which INCA data were not available, data were extracted from global databases (Global NEWS,http://marine.rutgers.edu/globalnews/datasets.htmandDai and Trenberth Global River Flow and Continental Discharge Ocean dynamics and biogeochemistry A regional POLCOMS-ERSEM coupled model (Holt and James,2001; Blackford et al. 2004; Holtet al., 2009) was used to project both the physical state of the ocean (temperature, salinity, currents, light level) and the biogeochemistry and lower trophic levels of the marine foodweb in the BoB. The model simulates four phytoplankton functional types, three zooplankton functional types and bacteria, as well as three size classes of particulate organic matter and dissolved and semi-labile organic matter. Four nutrients(C, P, N, and Si) are explicitly tracked within the model. The model domain covers the coastal area of the whole BoB (778–1048W, 1.38S–238N), and its width is from the coast to 200 km beyond the edge of the continental shelf (Figure2). The model uses a rectangular grid with a horizontal resolution of 0.18and 42 vertical levels distributed according to bottom depth. At the atmospheric boundary, it was forced using 3-hourly and daily outputs from the Had RM3 Pregional climate model described above, and physical conditions at the open ocean boundary were set using monthly outputs from theHadCM3 GCM.Nutrient values at the ocean boundary were fixed to values from the World Ocean Atlas (Garcia et al., 2010); since future projections are not available for these variables, the values were kept constant during the run. The boundary is advective, so although nutrient values are kept constant, nutrient losses and gains at the boundaries are allowed. Keeping nutrient levels at the boundary fixed could have some effect on primary production, but since production here is very low (Martin and Shaji,2015), the effect is likely to be small compared with changes in more productive zones nearer the coast.For each climate dataset, the model was run continuously for 1971–2099. Model outputs including temperature, salinity, current speeds, primary production, dissolved oxygen, pH, and plankton biomass were recorded at daily intervals and used to run the fish production models.The BoB POLCOMS-EREM model was validated by comparing model outputs with in situ measurements of temperature and salinity and with satellite values of surface chlorophyll. Temperature and salinity data were taken from the World Ocean Database (Boyer et al., 2013), using all available data for the model domain for the period 1993 to 2009. Monthly aggregated satellite chlorophyll data for 1997–2009 were taken from the database of the Ocean Colour Climate Change Initiative (Hollmann et al., 2013;http://www.esa-oceancolour-cci.org). For validation purposes, the model was run with forcing from reanalysis data (ERA interim, Dee et al.,2011, and GLORYS, Ferry et al., 2012).