We evaluate trends in terrestrial water storage over 1950–2100 in CMIP6 climate models against a new global reanalysis from assimilating GRACE and GRACE-FO satellite observations into a hydrological model. To account for different timescales in our analysis, we select regions in which the influence of interannual variability is relatively small and observed trends are assumed to be representative of the development over longer periods. Our results reveal distinct biases in drying and wetting trends in CMIP6 models for several world regions. Specifically, we see high model consensus for drying in the Amazon, which disagrees with the observed wetting. Other regions show a high consensus of models and observations suggesting qualitatively correctly simulated trends, e.g., for the Mediterranean and parts of Central Africa. A high model agreement might therefore falsely indicate a robust trend in water storage if it is not assessed in light of the observed developments. This underlines the potential use of maintaining an adequate observational capacity of water storage for climate change assessments.

我们在 CMIP6 气候模型中评估了 1950-2100 年陆地储水的趋势，并根据将 GRACE 和 GRACE-FO 卫星观测同化到水文模型中的新全球再分析。为了在我们的分析中考虑不同的时间尺度，我们选择了年际变化影响相对较小的地区，并且假设观察到的趋势代表了较长时间的发展。我们的结果揭示了世界多个地区的 CMIP6 模型中干燥和润湿趋势的明显偏差。具体来说，我们看到亚马逊地区对干燥的模型共识很高，这与观察到的润湿不一致。其他地区显示出高度一致的模型和观测结果，表明定性地正确模拟了趋势，例如，地中海和中非部分地区。因此，如果不根据观察到的发展进行评估，高模型一致性可能会错误地表明蓄水的强劲趋势。这强调了保持足够的蓄水观测能力用于气候变化评估的潜在用途。