Satellite remote sensing enhances model predictions by providing insights into terrestrial and hydrological processes. While data assimilation techniques have proven promising, there is a lack of standardized and effective approaches for integrating multiple observations simultaneously. This study presents a novel assimilation framework, the multi-observation local ensemble-Kalman-filter (MoLEnKF), designed to effectively integrate multiple variables, even at scales different than the model. Evaluation of MoLEnKF in the Amazon River basin includes assimilation experiments with remote sensing data only, including water surface elevation (WSE), terrestrial water storage (TWS), flood extent (FE), and soil moisture (SM). MoLEnKF demonstrates improvements in a scenario where regions lack in-situ hydroclimatic records and when assuming uncertainties of large-scale hydrologic-hydrodynamic models. Assimilating WSE outperforms daily discharge and water-level estimations, achieving 38% and 36% error reduction, respectively. However, the monthly evapotranspiration estimate achieves the greatest error reduction by assimilating SM with 11%. MoLEnKF always remains in second position in a ranking of error and uncertainty reduction, providing an intermediate condition, being able to holistically outperform univariate experiments. MoLEnKF also outperform state-of-the-art models in many cases. This study suggests potential improvements, urging exploration of correlations between assimilated variables and adaptive localization methods based on seasonality. The flexibility and the elegant way of expressing the LEnKF equations by MoLEnKF facilitates their application with different types of variables, compatible with large-scale hydrologic-hydrodynamic models and missions such as SWOT. Its robustness ensures easy replicability worldwide, facilitating hydrological reanalysis and improved forecasting, establishing MoLEnKF as a valuable tool for the scientific community in hydrological research.

中文翻译：

---

用于大规模水文水动力预测的多卫星数据同化：亚马逊盆地的概念验证


卫星遥感通过提供对陆地和水文过程的洞察来增强模型预测。虽然数据同化技术已被证明很有前景，但缺乏同时整合多个观测结果的标准化和有效方法。这项研究提出了一种新颖的同化框架，即多观测局部集成卡尔曼滤波器（MoLEnKF），旨在有效地整合多个变量，即使在与模型不同的尺度上也是如此。亚马逊河流域MoLEnKF的评估仅使用遥感数据进行同化实验，包括水面高程（WSE）、陆地蓄水量（TWS）、洪水范围（FE）和土壤湿度（SM）。 MoLEnKF 展示了在地区缺乏原位水文气候记录以及假设大规模水文-水动力模型不确定性的情况下的改进。同化 WSE 的效果优于每日流量和水位估算，分别减少了 38% 和 36% 的误差。然而，通过同化 SM，月蒸散量估算实现了最大误差减少 11%。 MoLEnKF 在误差和不确定性降低排名中始终保持第二位，提供了一个中间条件，能够整体优于单变量实验。 MoLEnKF 在许多情况下也优于最先进的模型。这项研究提出了潜在的改进，敦促探索同化变量和基于季节性的自适应定位方法之间的相关性。 MoLEnKF 表达 LEnKF 方程的灵活性和优雅方式促进了它们在不同类型变量中的应用，与大规模水文-水动力模型和 SWOT 等任务兼容。 其稳健性确保在全球范围内轻松复制，促进水文再分析和改进预报，使 MoLEnKF 成为科学界水文研究的宝贵工具。