Abstract. The northwest European shelf (NWS) seas are environmentally and economically important, and an understanding of how their climate may change helps with their management. However, as the NWS seas are poorly represented in global climate models, a common approach is to dynamically downscale with an appropriate shelf sea model. We develop a set of physical marine climate projections for the NWS. We dynamically downscale 12 members of the HadGEM3-GC3.05 perturbed parameter ensemble (approximately 70 km horizontal resolution over Europe), developed for UKCP18, using the shelf sea model NEMO CO9 (7 km horizontal resolution). These are run under the RCP8.5 high-greenhouse-gas-emission scenario as continuous simulations over the period 1990–2098. We evaluate the simulations against observations in terms of tides, sea surface temperature (SST), surface and near-bed temperature and salinity, and sea surface height. These simulations represent the state of the art for NWS marine projections. We project an SST rise of 3.11 °C (± 2σ = 0.98 °C) and a sea surface salinity (SSS) freshening of −1.01 (± 2σ = 0.93; on the (unitless) practical salinity scale) for 2079–2098 relative to 2000–2019, averaged over the NWS (approximately bounded by the 200 m isobar and excluding the Norwegian Trench, the Skagerrak and Kattegat), a substantial seasonal stratification increase (23 d over the NWS seas), and a general weakening of the NWS residual circulation. While the patterns of NWS changes are similar to our previous projections, there is a greater warming and freshening that could reflect the change from the A1B emissions scenario to the RCP8.5 concentrations pathway or the higher climate sensitivity exhibited by HadGEM3-GC3.05. Off the shelf, south of Iceland, there is limited warming, consistent with a reduction in the Atlantic Meridional Overturning Circulation and associated northward heat transport. These projections have been publicly released, along with a consistent 200-year present-day control simulation, to provide an evidence base for climate change assessments and to facilitate climate impact studies. For example, we illustrate how the two products can be used to estimate climate trends, unforced variability and the time of emergence (ToE) of the climate signals. We calculate the average NWS SST ToE to be 2034 (with an 8-year range) and 2046 (with a 33-year range) for SSS. We also discuss how these projections can be used to describe NWS conditions under 2 and 4 °C global mean warming (compared with 1850–1900), as a policy-relevant exemplar use case.