Innovative pheno-network model in estimating crop phenological stages with satellite time series
Innovative pheno-network model in estimating crop phenological stages with satellite time series
复制标题
DOI：
10.1016/j.isprsjprs.2019.04.012
复制
发表时间：
2019-07
影响因子：
12.7
通讯作者：
C. Diao
中科院分区：
文献类型：
--
作者：
C. Diao
大规模远程监测作物物候发育对于安排农场管理活动和估计作物产量至关重要。跟踪作物物候进展对于了解农业对环境压力和气候变化的反应也至关重要。在过去的十年中，遥感图像的时间序列越来越多地用于监测农作物的季节性生长动态。各种基于曲线拟合的物候学方法已被开发出来，以估计关键的物候过渡日期。然而，这些物候学方法通常是参数化的，通过对作物物候过程进行数学假设，并且通常需要长达一年的卫星观测来进行参数训练。这些假设和限制使得这些方法不足以在重云污染地区或复杂的农业系统中进行物候监测。本研究的目的是使用基于复杂网络的物候模型（即，“pheno-network”）。创新的物候网络模型是非参数的，没有数学定义的物候假设，可以构建部分年份的遥感数据。基于网络理论的物候网络模型以光谱定义的节点和边来描述复杂的物候过程。它提供了一个创新的网络表示模型的光谱反射率的作物在整个生长季节的时间动态。以美国伊利诺伊州的玉米和大豆为例，设计了表型网络模型，估算了它们沿着叶片衰老轨迹的物候转换日期。结果表明，玉米的估计转换日期与其地面观测成熟期有很强的相关性。对于大豆而言，估计的转换日期与其落叶期密切相关。表型网络模型显示出在复杂的农业多样化和集约化系统中推进物候监测的显著潜力。
Large-scale remote monitoring of crop phenological development is vital for scheduling farm management activities and estimating crop yields. Tracking crop phenological progress is also crucial to understand agricultural responses to environmental stress and climate change. During the past decade, time series of remotely sensed imagery has been increasingly employed to monitor the seasonal growing dynamics of crops. A variety of curve-fitting based phenological methods have been developed to estimate critical phenological transition dates. However, those phenological methods are typically parametric by making mathematical assumptions of crop phenological processes and usually require year-long satellite observations for parameter training. The assumption and constraint make those methods inadequate for phenological monitoring in heavy cloud-contaminated regions or in complex agricultural systems. The objective of this study is to estimate crop phenological stages with satellite time series using a complex network-based phenological model (i.e., “pheno-network”). The innovative pheno-network model is non-parametric without mathematically defined phenological assumptions and can be constructed with partial-year remote sensing data. Rooted in network theory, the pheno-network model characterizes the complex phenological process with spectrally defined nodes and edges. It provides an innovative network representation to model the temporal dynamics of spectral reflectance of crops throughout the growing season. With corn and soybean in Illinois as a case study, the pheno-network model was devised to estimate their phenological transition dates along the leaf senescence trajectory from 2002 to 2017. Results indicated that the estimated transition dates of corn had strong correlation with its ground-observed mature stage. As for soybean, the estimated transition dates were closely associated with its dropping leaves stage. The pheno-network model shows marked potential to advance phenological monitoring in complex agricultural diversified and intensified systems.