Mangroves in tropical and subtropical coasts are subject to episodic disturbances, notably from severe storms, leading to potential widespread vegetation mortality. The ability of vegetation to recover varies, and with disturbances becoming more frequent and severe, it is vital to track and project vegetation responses to support management and policy decisions. Prior studies have largely focused on binary mangrove mapping (i.e., presence or absence), while tracking conditions and condition change have not received sufficient attention. In this paper, we demonstrate a method based on dense time series Landsat images for continuous monitoring of mangrove conditions, where we track three kinds of post-disturbance mangrove conditions, including disturbed (disturbed, with rebound to the previous state within one growing season), recovering (undergoing natural recovery in longer than one growing season), and declining (showing long-term decline after disturbance). The method starts with disturbance detection using the DEtection and Characterization Of the tiDal wEtland change (DECODE) algorithm, an existing dense time series model designed to detect disturbances in tidal wetlands with adaptation to tidal fluctuations. This algorithm is well suited for the detection of tidal wetland disturbances but does not provide satisfactory post-disturbance monitoring results, due to the substantial variability in post-disturbance Landsat observations. To better monitor post-disturbance conditions, a new time series fitting approach, DECODER (DECODE and Recovery), is proposed for the recovery stage. Additionally, for temporal segments divided by disturbance events, we built a random forest classifier with temporal-spectral variables derived from the time series model to characterize mangrove conditions. Employing this approach in Florida's mangroves, we generated condition maps, such as dieback and recovery, with an overall accuracy of approximately 97.96 ± 0.86- [95 % confidence intervals]. Comparing post-hurricane conditions in Florida revealed that the increased frequency and severity of disturbances are challenging mangrove resilience, potentially diminishing their ability to recover and sustain ecosystem functions.

中文翻译：

---

使用密集 Landsat 时间序列追踪红树林状况变化


热带和亚热带海岸的红树林受到偶发性干扰，特别是来自强风暴的干扰，可能导致大面积植被死亡。植被的恢复能力各不相同，随着干扰变得越来越频繁和严重，追踪和预测植被响应以支持管理和政策决策至关重要。以前的研究主要集中在二元红树林制图（即存在或不存在）上，而跟踪条件和条件变化没有得到足够的重视。在本文中，我们展示了一种基于密集时间序列 Landsat 图像的红树林状况连续监测方法，其中我们跟踪了三种干扰后红树林状况，包括干扰（干扰，在一个生长季节内反弹到以前的状态）、恢复（在超过一个生长季节内经历自然恢复）和下降（在干扰后表现出长期下降）。该方法首先使用 tiDal wEtland 变化的检测和表征 （DECODE） 算法进行干扰检测，这是一种现有的密集时间序列模型，旨在检测潮汐湿地中的干扰并适应潮汐波动。该算法非常适合于检测潮汐湿地干扰，但由于干扰后 Landsat 观测值存在很大差异，因此无法提供令人满意的干扰后监测结果。为了更好地监测后干扰条件，为恢复阶段提出了一种新的时间序列拟合方法，即 DECODER （DECODE and Recovery）。 此外，对于按干扰事件划分的时间段，我们构建了一个随机森林分类器，其中包含从时间序列模型得出的时间光谱变量来表征红树林条件。在佛罗里达州的红树林中采用这种方法，我们生成了条件图，例如枯萎和恢复，总体精度约为 97.96 ± 0.86- [95% 置信区间]。比较佛罗里达州飓风过后的情况表明，干扰频率和严重程度的增加正在挑战红树林的复原力，可能会削弱它们恢复和维持生态系统功能的能力。