Monitoring of wetland inundation dynamics in the Delmarva Peninsula using Landsat time-series imagery from 1985 to 2011
Monitoring of wetland inundation dynamics in the Delmarva Peninsula using Landsat time-series imagery from 1985 to 2011
复制标题
DOI：
10.1016/j.rse.2016.12.001
复制
发表时间：
2017-03
影响因子：
13.5
通讯作者：
Huiran Jin;Chengquan Huang;M. Lang;I. Yeo;S. Stehman
中科院分区：
文献类型：
--
作者：
Huiran Jin;Chengquan Huang;M. Lang;I. Yeo;S. Stehman
湿地提供重要的生态系统服务，其提供在很大程度上受洪水和土壤饱和度波动的控制。洪水是高度动态的，随着时间的推移会因多种驱动因素（包括降水和蒸散）而发生很大变化。这项研究的重点是开发一个实用有效的框架，利用机载激光雷达强度数据对湿地淹没动态进行区域性长期监测（Lang等人，2013）和Landsat时间序列图像。从1985年到2011年，在美国东海岸的整个Delmarva半岛，每年生成一次亚像素水分数（SAR）地图，显示每个30米像素内的地表水百分比。利用历史上的高分辨率航空摄影对这些地图进行了全面的准确性评估，以确定参考条件。评估结果表明，开放水域样本的均方根误差（RMSE）估计值为7.78%（地图上该区域的平均均方根误差约为40%）。此外，一个单独的准确性评估，针对湿地的洪水（即存在或不存在水）产生了93%的整体准确性。由此得出的精确度表明，大地卫星数据可以进行校准，以准确地提取区域尺度的长期水信息。洪水淹没的特点进行了研究，相对于不同的湿地类型定义的水分状况和优势植被类型，以及不同的物理驱动程序。结果表明，潮汐湿地通常表现出更强烈的淹没比非潮汐湿地，和更高程度的淹没与紧急湿地相比，以木本植被为主的湿地地区。变化驱动因素分析显示，潮汐对沿海淹没产生了统计学显著影响，r2值为32-36%和p < 0.01，而内陆湿地地区的淹没变化部分由降水驱动，r2值为25-34%和p < 0.08。由于最新的陆地卫星图像档案在全球范围内都可以获得，而且激光雷达数据越来越便宜，因此可以很容易地实施所开发的框架，以生成地球仪许多区域的连续洪水记录，以协助正在进行的和未来的湿地水文学和湿地管理研究。
Wetlands provide important ecosystem services, the provision of which is largely controlled by fluctuations in inundation and soil saturation. Inundation is highly dynamic and can vary substantially through time in response to multiple drivers, including precipitation and evapotranspiration. This research focused on developing a practical and effective framework for regional, long-term monitoring of wetland inundation dynamics using airborne LiDAR intensity data (Lang et al., 2013) and Landsat time-series imagery. Subpixel water fraction (SWF) maps indicating the percent of surface water within each 30-m pixel were generated on an annual basis over the entire Delmarva Peninsula on the East Coast of the United States from 1985 to 2011. Comprehensive accuracy assessments of the SWF maps were conducted using historical high-resolution aerial photography to determine the reference condition. The assessment resulted in an estimated root mean square error (RMSE) of 7.78% for the sample of open water areas (mean SWF was ~ 40% for this region of the map). Moreover, a separate accuracy assessment targeting inundation in wetlands (i.e. presence or absence of water) yielded an overall accuracy of 93%. Accuracies derived indicated that Landsat data can be calibrated to accurately extract long-term water information at the regional scale. Characteristics of inundation were examined with respect to different wetland types defined by water regime and dominant vegetation types, as well as different physical drivers. Results showed that tidal wetlands typically exhibited more intensive inundation than nontidal wetlands, and a higher degree of inundation was associated with emergent wetlands compared to wetland areas dominated by woody vegetation. Analysis of change drivers revealed that tide exerted a statistically significant influence on coastal inundation withr2values of 32–36% andp< 0.01, whereas inundation changes in inland wetland areas were in part driven by precipitation withr2values of 25–34% andp< 0.08. Because an up-to-date archive of Landsat imagery is globally available and LiDAR data are becoming increasingly more affordable, the developed framework can be easily implemented to generate a continuous inundation record in many regions of the globe to assist in ongoing and future studies focused on wetland hydrology and wetland management.