Structural deficits in large ensembles limit detection and attribution of terrestrial water storage

Determining whether terrestrial water storage trends and extremes reflect external forcing or internal climate variability requires a credible counterfactual baseline: how much water storage could vary in a world shaped only by internal climate variability. We test whether two widely used single-model large-ensemble archives, CESM2-LENS2 and IPSL-CM6A-LR, provide that baseline by benchmarking their basin-scale variability against the 23-year GRACE and GRACE-FO terrestrial water storage record from 2002 to 2024 across 184 global river basins, drawing on 18,258 simulated model-years. Both ensembles frequently fail to reproduce the observed amplitude, dispersion, persistence, and severity of extremes. A four-metric benchmark classifies 78 to 82% of basins as incompatible with the model distributions. These results show that null hypotheses for internal climate variability should be tested against observations before being used in detection and attribution to avoid biased inference about emergence and external forcing. They also indicate that the main limitation is not ensemble size but model structure, implying that progress depends more on improving the representation of hydrologic storage, transport, and memory in Earth system models than on generating ever larger ensembles. Model ensembles often fail to capture basin-scale terrestrial water storage variability and extremes due to model structure, according to a study benchmarking CESM2-LENS2 and IPSL-CM6A-LR against GRACE and GRACE-FO records across 184 global river basins from 2002 to 2024.