Abstract Precipitation is a fundamental forcing variable in land surface modeling, controlling several hydrological and biogeochemical processes (e.g., runoff, carbon cycling, evaporation, transpiration, groundwater recharge, and soil moisture). However, precipitation estimates from rain gauges, ground-based radars, satellite sensors, and numerical models are affected by significant uncertainties, which can be amplified when exposed to highly non-linear land model physics. This work tests the hypothesis that precipitation data from different sources can be optimally merged to minimize the hydrologic response error in surface soil moisture simulations and maximize their correlation with ground observations (multi-objective optimization problem). This hypothesis is tested by merging three precipitation products (one satellite product, a ground-based dataset, and model-base estimates) that force a land surface model trained to minimize soil moisture anomalies. A Monte Carlo-based algorithm is developed to generate weights to linearly combine these precipitation datasets. Optimal combinations of weights are identified by minimizing the errors and maximizing the correlation between the model simulated soil moisture and the satellite-based SMOS soil moisture product. The proposed methodology has been tested over Oklahoma where high-quality, high-resolution (independent) ground-based soil moisture observations are available for validation purposes. Results show that there exist optimal combinations of these precipitation datasets that provide smaller errors and larger correlation coefficients between modeled soil moisture estimates and ground-based data with respect to forcing the land surface model with single precipitation datasets. Specifically, combining three precipitation products from different sources provides the largest correlation coefficient and the lowest root mean square error at several locations across Oklahoma.

中文翻译：

---

一种基于蒙特卡罗的多目标优化方法，用于合并不同的降水估计以进行地表建模

摘要 降水是地表建模中的一个基本强迫变量，控制着几个水文和生物地球化学过程（例如，径流、碳循环、蒸发、蒸腾、地下水补给和土壤水分）。然而，来自雨量计、地基雷达、卫星传感器和数值模型的降水估计受到重大不确定性的影响，当暴露于高度非线性的陆地模型物理时，这些不确定性可能会被放大。这项工作检验了以下假设：可以最佳合并来自不同来源的降水数据，以最大限度地减少表层土壤水分模拟中的水文响应误差，并最大限度地提高它们与地面观测的相关性（多目标优化问题）。该假设通过合并三个降水产品（一个卫星产品，一个基于地面的数据集和基于模型的估计），强制训练地表模型以最大限度地减少土壤水分异常。开发了一种基于蒙特卡罗的算法来生成权重以线性组合这些降水数据集。通过最小化误差和最大化模型模拟土壤水分与基于卫星的 SMOS 土壤水分产品之间的相关性来确定权重的最佳组合。所提议的方法已经在俄克拉荷马州进行了测试，在俄克拉荷马州，高质量、高分辨率（独立）的地面土壤湿度观测可用于验证目的。结果表明，这些降水数据集存在最佳组合，在模拟土壤水分估计值和地基数据之间提供较小的误差和较大的相关系数，相对于强制使用单个降水数据集的地表模型。具体而言，将来自不同来源的三种降水产品组合在一起，可在俄克拉荷马州的多个地点提供最大的相关系数和最低的均方根误差。