Exploring the impact of model and data uncertainties in the detection and
attribution of upper-ocean warming


Large-scale increases in upper ocean heat content (OHC) are evident in the observational record of the past fifty years. Several studies have made use of well-established detection and attribution (D&A) methods to demonstrate that the observed changes are consistent with model-based estimates of the OHC response to increasing concentrations of greenhouse gases, and inconsistent with model estimates of natural variability. The recent identification of systematic XBT biases in observational OHC records has led to new estimates of global scale OHC variability and trends. This provides motivation for a re-examination of previous D&A findings. In the present study, we use these newer estimates of OHC change to further examine the causes of ocean warming. We perform a comprehensive assessment of the sensitivity of ocean heat content D&A results to: 1) measurement biases in observations; 2) incomplete, time- varying spatial coverage of observational data; 3) methods used to remove residual drift from model simulations; and 4) the inclusion or neglect of volcanic forcings in model simulations. Previous D&A studies involving OHC changes have relied on simulations from one or two individual models. Our D&A analysis is conducted in a multi-model framework, in which results from over a dozen coupled models are used to estimate the response to external forcing and the noise of natural internal variability. We find that signal-to-noise ratios show some sensitivity to the model and data uncertainties considered here. However, our results suggest that the positive identification of a human-caused warming signal in observed OHC changes is generally robust to these uncertainties. This is consistent with findings from previous "single model" D&A studies based on earlier versions of observational OHC datasets (which did not include XBT bias corrections).