Satellite-based vegetation optical depth as an indicator of drought-driven tree mortality
Satellite-based vegetation optical depth as an indicator of drought-driven tree mortality
复制标题
基于卫星的植被光学厚度作为干旱驱动的树木死亡率的指标
DOI：
10.1016/j.rse.2019.03.026
复制
发表时间：
2019-06
影响因子：
13.5
通讯作者：
K. Rao;W. Anderegg;A. Sala;J. Martínez‐Vilalta;A. Konings
中科院分区：
文献类型：
--
作者：
K. Rao;W. Anderegg;A. Sala;J. Martínez‐Vilalta;A. Konings
在气候变化下，干旱导致的树木死亡事件预计会增加。然而，植被对气候干旱胁迫响应的高度空间变异性，以及缺乏可在大尺度上监测的具有生理意义的胁迫变量，限制了对树木死亡的监测和建模。在这项研究中，我们验证了被动微波遥感通过植被光学厚度估计的相对含水率(RWC)可以作为树木死亡的经验指标的假设，该指标既可以综合植物干旱胁迫的变化，又可以在大范围内获得。这一假设在美国加利福尼亚州最近的一次严重干旱中得到了验证。相对含水率与死亡率的阈值关系比气候缺水--一种常用的死亡率指标--更强，尽管由于数据的空间分辨率较低(0.25°或约25 千米)，这两种关系都存在噪声。此外，加州北部和南部地区的RWC阈值比CWD阈值更一致。用描述地形、气候和植被特征的32个变量的随机森林回归(机器学习)预测森林死亡程度，即死亡面积分数，具有令人满意的精度-决定系数Rest2= 0.66，均方根误差 = 0.023。重要的是，RWC在估计死亡率方面的重要性是模型中任何其他变量的两倍以上，证实了它与死亡率的密切联系。此外，RWC显示出中等的帮助预测死亡率的能力，RWC的相对重要性比死亡率提前一年测量的类似于在死亡率年测量的其他相关解释变量。这项研究的结果提出了一种很有希望的新方法来估计与死亡风险有关的森林干旱胁迫。
Drought-induced tree mortality events are expected to increase in frequency under climate change. However, monitoring and modeling of tree mortality is limited by the high spatial variability in vegetation response to climatic drought stress and lack of physiologically meaningful stress variables that can be monitored at large scales. In this study, we test the hypothesis that relative water content (RWC) estimated by passive microwave remote sensing through vegetation optical depth can be used as an empirical indicator of tree mortality that both integrates variations in plant drought stress and is accessible across large areas. The hypothesis was tested in a recent severe drought in California, USA. The RWC showed a stronger threshold relationship with mortality than climatic water deficit (CWD) – a commonly used mortality indicator – although both relationships were noisy due to the coarse spatial resolution of the data (0.25° or approximately 25 km). In addition, the threshold for RWC was more uniform than that for CWD when compared between Northern and Southern regions of California. A random forests regression (machine learning) with 32 variables describing topography, climate, and vegetation characteristics predicted forest mortality extent i.e. fractional area of mortality (FAM) with satisfactory accuracy-coefficient of determinationRtest2= 0.66, root mean square error = 0.023. Importantly, RWC was more than twice as important as any other variable in the model in estimating mortality, confirming its strong link to mortality rates. Moreover, RWC showed a moderate ability to aid in forecasting mortality, with a relative importance of RWC measured one year in advance of mortality similar to that of other relevant explanatory variables measured in the mortality year. The results of this study present a promising new approach to estimate drought stress of forests linked to mortality risk.