Few coastal ecosystems remain untouched by direct human activities, and none are unimpacted by anthropogenic climate change. These drivers interact with and exacerbate each other in complex ways, yielding a mosaic of ecological consequences that range from adaptive responses, such as geographic range shifts and changes in phenology, to severe impacts, such as mass mortalities, ecological regime shifts and loss of biodiversity. Identifying the role of climate change in these phenomena requires corroborating evidence from multiple lines of evidence, including laboratory experiments, field observations, numerical models and palaeorecords. Yet few studies can confidently quantify the magnitude of the effect attributable solely to climate change, because climate change seldom acts alone in coastal ecosystems. Projections of future risk are further complicated by scenario uncertainty – that is, our lack of knowledge about the degree to which humanity will mitigate greenhouse-gas emissions, or will make changes to the other ways we impact coastal ecosystems. Irrespective, ocean warming would be impossible to reverse before the end of the century, and sea levels are likely to continue to rise for centuries and remain elevated for millennia. Therefore, future risks to coastal ecosystems from climate change are projected to mirror the impacts already observed, with severity escalating with cumulative emissions. Promising avenues for progress beyond such qualitative assessments include collaborative modelling initiatives, such as model intercomparison projects, and the use of a broader range of knowledge systems. But we can reduce risks to coastal ecosystems by rapidly reducing emissions of greenhouse gases, by restoring damaged habitats, by regulating non-climate stressors using climate-smart conservation actions, and by implementing inclusive coastal-zone management approaches, especially those involving nature-based solutions.

Twelve climate models and observations are used to attribute the global mean surface temperature (GMST) changes from 1900 to 2014 to external climate forcings. The external forcings are decomposed into the effects of the well-mixed greenhouse gas concentration variation, the effects of anthropogenic aerosol concentration changes, and the effects of natural forcings. First, a convolutional neural network (CNN) is trained to estimate the simulated historical GMST from single-forcing experiments using outputs from the multi-model ensemble. We then use this CNN to solve the attribution problem using an original variational inversion approach. The variational inversion is first validated using historical climate simulations as pseudo-observations. Then we perform an inversion from observations. This provides a distribution of the GMST resulting from the three forcings. For 2014, inversions estimate that the greenhouse gases changes are responsible for a GMST anomaly within [0.8 $ {}^{\circ } $ C,1.9 $ {}^{\circ } $ C], while anthropogenic aerosols and natural forcings anomalies are within [−0.7 $ {}^{\circ } $ C,−0.1 $ {}^{\circ } $ C] and [−0.1 $ {}^{\circ } $ C,0.3 $ {}^{\circ } $ C], respectively. The method designed here can be adapted and extended to attribute the changes of other variables or to focus on the regional scale.