论文标题
用机器学习得出的高分辨率欧洲每日土壤水分（2003-2020）
High-resolution European daily soil moisture derived with machine learning (2003-2020)
论文作者
论文摘要
机器学习（ML）已成为近年来生成大规模陆地数据的新工具。 ML可以学习输入和目标之间的关系，例如气象变量和原位土壤水分，然后在跨时空估算土壤水分，独立于先前的基于物理学的知识。在这里，我们使用长期的短期记忆进行了训练的耐用测量值，开发了欧洲（SOMO.ML-EU）的每日高分辨率（0.1°）土壤水分数据集。最终的数据集涵盖了三个垂直层和2003 - 2020年期。与以前的空间分辨率较低（0.25°）的版本相比，它在时间变化方面与独立的原位数据显示了更紧密的一致性，这表明当与高分辨率气象数据共同处理时，现场观察值的实用性增强了。与其他网格数据集进行区域比较还证明了SOMO.ML-eu描述土壤水分的变异性（包括干旱条件）的能力。结果，我们的新数据集将使需要高分辨率观察的土壤水分（例如水文和农业分析）受益。 somo.ml-eu可在figshare上找到。
Machine learning (ML) has emerged as a novel tool for generating large-scale land surface data in recent years. ML can learn the relationship between input and target, e.g. meteorological variables and in-situ soil moisture, and then estimate soil moisture across space and time, independently of prior physics-based knowledge. Here we develop a high-resolution (0.1°) daily soil moisture dataset in Europe (SoMo.ml-EU) using Long Short-Term Memory trained with in-situ measurements. The resulting dataset covers three vertical layers and the period 2003-2020. Compared to its previous version with a lower spatial resolution (0.25°), it shows a closer agreement with independent in-situ data in terms of temporal variation, demonstrating the enhanced usefulness of in-situ observations when processed jointly with high-resolution meteorological data. Regional comparison with other gridded datasets also demonstrates the ability of SoMo.ml-EU in describing the variability of soil moisture, including drought conditions. As a result, our new dataset will benefit regional studies requiring high-resolution observation-based soil moisture, such as hydrological and agricultural analyses. The SoMo.ml-EU is available at figshare.
