Abolafia-Rosenzweig R, Livneh B, Small E E, Kumar S V
Department of Civil, Environmental, and Architectural Engineering University of Colorado Boulder Boulder CO USA.
Cooperative Institute for Research in Environmental Science (CIRES) University of Colorado Boulder Boulder CO USA.
J Adv Model Earth Syst. 2019 Nov;11(11):3670-3690. doi: 10.1029/2019MS001797. Epub 2019 Nov 17.
Knowledge of irrigation is essential to support food security, manage depleting water resources, and comprehensively understand the global water and energy cycles. Despite the importance of understanding irrigation, little consistent information exists on the amount of water that is applied for irrigation. In this study, we develop and evaluate a new method to predict daily to seasonal irrigation magnitude using a particle batch smoother data assimilation approach, where land surface model soil moisture is applied in different configurations to understand how characteristics of remotely sensed soil moisture may impact the performance of the method. The study employs a suite of synthetic data assimilation experiments, allowing for systematic diagnosis of known error sources. Assimilation of daily synthetic soil moisture observations with zero noise produces irrigation estimates with a seasonal bias of 0.66% and a correlation of 0.95 relative to a known truth irrigation. When synthetic observations were subjected to an irregular overpass interval and random noise similar to the Soil Moisture Active Passive satellite (0.04 cm cm), irrigation estimates produced a median seasonal bias of <1% and a correlation of 0.69. When systematic biases commensurate with those between NLDAS-2 land surface models and Soil Moisture Active Passive are imposed, irrigation estimates show larger biases. In this application, the particle batch smoother outperformed the particle filter. The presented framework has the potential to provide new information into irrigation magnitude over spatially continuous domains, yet its broad applicability is contingent upon identifying new method(s) of determining irrigation schedule and correcting biases between observed and simulated soil moisture, as these errors markedly degraded performance.
灌溉知识对于保障粮食安全、管理日益枯竭的水资源以及全面理解全球水和能量循环至关重要。尽管了解灌溉很重要，但关于灌溉用水量的一致信息却很少。在本研究中，我们开发并评估了一种新方法，该方法使用粒子批量平滑器数据同化方法来预测每日至季节性的灌溉量，其中将陆面模型土壤湿度以不同配置应用，以了解遥感土壤湿度特征如何影响该方法的性能。该研究采用了一系列综合数据同化实验，以便对已知误差源进行系统诊断。同化零噪声的每日综合土壤湿度观测数据，得到的灌溉量估计值与已知真实灌溉量相比，季节性偏差为0.66%，相关性为0.95。当综合观测数据具有不规则的观测间隔和类似于土壤湿度主动被动卫星（Soil Moisture Active Passive）的随机噪声（0.04 cm/cm）时，灌溉量估计值的季节性偏差中位数<1%，相关性为0.69。当施加与NLDAS - 2陆面模型和土壤湿度主动被动卫星之间的偏差相当的系统偏差时，灌溉量估计值显示出更大的偏差。在这个应用中，粒子批量平滑器的性能优于粒子滤波器。所提出的框架有可能在空间连续区域提供关于灌溉量的新信息，但其广泛适用性取决于确定灌溉时间表的新方法以及校正观测和模拟土壤湿度之间的偏差，因为这些误差会显著降低性能。
