Soil moisture data is essential for estimating groundwater storage anomalies (GWSA) from the Gravity Recovery and Climate Experiment (GRACE) data, but the general lack of direct in-situ root-zone soil moisture observations has typically resulted in a reliance on modelled soil moisture estimates instead. These model-simulated soil moisture profiles – upper (0 to 10 cm), lower (10 to 100 cm), and deep layers (100 to 200 cm), are characterized by large uncertainties due to the simplification and parameterization of soil moisture processes in hydrological models. It is thus crucial to account for these uncertainties and understand how they affect the estimation of groundwater storage changes based on GRACE data. In this study, we evaluated the contributions and impacts of different soil moisture profiles on GRACE-derived groundwater storage (between 2002 and 2016) modelling uncertainties over the Murray Darling Basin (MDB) using statistical and machine learning regression. We observed that the lower layer exhibited the strongest correlation with base GWSA, particularly during 2006 to 2009 (r = 0.99, RMSE = 7.50 mm). Bootstrap analysis indicated that the lower layer consistently had the largest absolute coefficient weights, signifying its predominant influence on GWSA predictions. The deep layer contributed the least during 2010 to 2013, while the upper layer was highly dynamic and introduced a 26.8 % more uncertainty rating when compared to the lower layer. Regression analysis showed the lower layer maintained the smallest confidence interval widths, confirming its reliability. The Monte Carlo resampling corroborated these findings, with the lower layer maintaining the most consistent relationship with base GWSA across all periods. The lower layer’s steadier state and lower susceptibility to surface disturbances provided more accurate predictions than other layers. This study advances the modelling of groundwater storage from space by improving our understanding of the uncertainties introduced by the different soil moisture layers. It will be helpful for better and accurate freshwater reporting and management.

中文翻译：

---

在 GRACE 地下水建模中揭示土壤水分不确定性


土壤水分数据对于从重力恢复和气候实验 （GRACE） 数据中估计地下水储存异常 （GWSA） 至关重要，但普遍缺乏直接的原位根区土壤水分观测通常会导致依赖建模的土壤水分估计。这些模型模拟的土壤水分剖面——上层（0 至 10 厘米）、下层（10 至 100 厘米）和深层（100 至 200 厘米）由于水文模型中土壤水分过程的简化和参数化，具有较大的不确定性。因此，考虑这些不确定性并了解它们如何影响基于 GRACE 数据对地下水储量变化的估计至关重要。在这项研究中，我们使用统计和机器学习回归评估了不同土壤水分剖面对 GRACE 衍生的地下水储存（2002 年至 2016 年）的贡献和影响，对墨累达令盆地 （MDB） 的不确定性进行建模。我们观察到，下层与基础 GWSA 的相关性最强，尤其是在 2006 年至 2009 年期间 （r = 0.99，RMSE = 7.50 mm）。Bootstrap 分析表明，下层始终具有最大的绝对系数权重，表明其对 GWSA 预测的主要影响。在 2010 年至 2013 年期间，深层的贡献最小，而上层是高度动态的，与下层相比，引入的不确定性等级高出 26.8%。回归分析显示，下层保持了最小的置信区间宽度，证实了其可靠性。蒙特卡洛重采样证实了这些发现，下层在所有时期与基 GWSA 保持最一致的关系。 与其他层相比，下层的稳定状态和较低的对地表干扰的敏感性提供了更准确的预测。这项研究通过提高我们对不同土壤水分层引入的不确定性的理解，推进了从太空对地下水储存进行建模。这将有助于更好、更准确的淡水报告和管理。