Sub-daily soil moisture estimate using dynamic Bayesian model averaging
Sub-daily soil moisture estimate using dynamic Bayesian model averaging
复制标题
使用动态贝叶斯模型平均估算次日土壤湿度
DOI：
10.1016/j.jhydrol.2020.125445
复制
发表时间：
2020-11
影响因子：
6.4
通讯作者：
Ruochen Sun
中科院分区：
文献类型：
--
作者：
Yong Chen;Huiling Yuan;Yize Yang;Ruochen Sun
从卫星产品和次日时间尺度的模型模拟中准确估计土壤湿度(SM)仍然是一个挑战。本研究提出了一个用于合并次日模型产品的通用动态贝叶斯模型平均(BMA)框架。与传统的BMA方法相比，本研究为BMA成员引入了自适应权值(随时间动态变化)。在前人评价工作的基础上，从8个模型产品中筛选出一个模型产品子集作为BMA成员。对2017年江淮流域表层SM(0-10 cm)模式产品进行了亚日(6-h)时间尺度的动态BMA试验。将结果与自动SM观测(ASMOS)进行了比较，ASMOS具有前所未有的高空间和时间分辨率(104km2像素内多达7个台站；每小时)。由于天气模式和模型的性能随着时间的推移而变化，因此确定最佳训练周期对于获得适应快速天气状况变化的BMA权重至关重要。然后对训练长度(天数)的敏感性进行了检验，证明了BMA训练周期使用的最佳数据长度约为80天。利用确定性和概率性检验方法，结合ASMOS、8个全球模式产品和中国气象局的区域陆地资料同化系统(CLDAS)产品，对动态BMA估计的SM进行了综合评估。为了更好地比较不同产品的概率分布，提出了用累积分布函数(CDF)一致性直方图和更客观的度量一致性偏差(CD)来诊断两个SM CDF(如BMA估计和观测CDF)的一致性。在确定性(Kling-Gupta有效性、相关性、系统偏差和偏差调整均方根误差)和概率验证方法(CD、QQ图和可靠性)方面，动态BMA估计SM的表现优于任何BMA成员，甚至优于基准产品CLDAS。研究表明，动态BMA框架为SM模型产品的融合提供了一种新的解决方案。合并后的SM和BMA组合概率分布可以进一步用于干旱监测和预测。
Accurate estimation of soil moisture (SM) from satellite products and model simulations at sub-daily timescale remains a challenge. This study proposes a general dynamic Bayesian model averaging (BMA) framework for merging sub-daily model products. Compared to the traditional BMA method, this study introduces adaptive weights (dynamically variant with time) for BMA members. Based on the previous evaluation work, a subset of model products is selected from eight model products as BMA members. The dynamic BMA experiment is performed for the surface SM (0–10 cm) model products at sub-daily (6-h) timescale in 2017 over the Yangtze-Huaihe river basin. The results are compared with the automatic SM observations (ASMOs) with unprecedented high spatial and temporal resolution (up to 7 stations within a 104km2pixel; hourly). Because weather pattern and model performance change over time, the determination of an optimal training period is critical to obtain adaptive BMA weights for rapid weather regime changes. The sensitivity of training length (days) is then examined, and the optimum data length used in the BMA training period proves to be about 80 days. With deterministic and probabilistic verification metrics, the dynamic BMA estimated SM is comprehensively evaluated against the ASMOs, eight global model products, and the CMA’s (China Meteorological Administration) regional Land Data Assimilation System (CLDAS) product. To better compare the probability distribution of different products, the cumulative distribution function (CDF) consistency histogram and a more objective metric consistency deviation (CD) are proposed to diagnose the consistency of two SM CDFs (e.g., the BMA estimated and the observed CDF). In terms of both the deterministic (the Kling-Gupta efficiency, correlation, system bias, and bias adjusted root-mean square error) and probabilistic verification methods (CD, QQ-plots, and reliability), the dynamic BMA estimated SM outperforms any BMA members and even the benchmark product CLDAS. This study demonstrates that the dynamic BMA framework provides a new solution for merging SM model products. The merged SM and the BMA combined probability distribution can be further used for drought monitoring and prediction.
登录
查看更多内容
影响因子：
4.7
作者：
James M. Leach;P. Coulibaly
通讯作者：
James M. Leach;P. Coulibaly
影响因子：
3.8
作者：
Shahrbanou Madadgar;H. Moradkhani
通讯作者：
Shahrbanou Madadgar;H. Moradkhani
DOI：
10.1016/j.jag.2015.06.009
发表时间：
2016-06
期刊：
Int. J. Appl. Earth Obs. Geoinformation
影响因子：
--
作者：
Abdolaziz Rahmani;S. Golian;L. Brocca
通讯作者：
Abdolaziz Rahmani;S. Golian;L. Brocca
影响因子：
3.8
作者：
S. Shukla;A. Steinemann;D. Lettenmaier
通讯作者：
S. Shukla;A. Steinemann;D. Lettenmaier
影响因子：
8
作者：
C. Mass;David Ovens;Kenneth J. Westrick;B. Colle
通讯作者：
C. Mass;David Ovens;Kenneth J. Westrick;B. Colle