首页|Timeliness in forest change monitoring: A new assessment framework demonstrated using Sentinel-1 and a continuous change detection algorithm
Timeliness in forest change monitoring: A new assessment framework demonstrated using Sentinel-1 and a continuous change detection algorithm
扫码查看
点击上方二维码区域，可以放大扫码查看
原文链接
NSTL
Elsevier
The development of near-real time forest monitoring systems, which are used to create alerts for events such as logging or fire, often involves tradeoffs between accuracy and timeliness. In the context of forest monitoring, timeliness is measured by the lag between when a change occurs in the forest and the creation of an alert about the event. Conventional accuracy assessments quantify both false negative and false positive errors in change maps, but do not specify those rates as a function of lag. Recent near real-time (NRT) accuracy assessment methods summarize the relationship between lag and correctly identified change events, but do not integrate consideration of changes mapped where they have not occurred. Here, we propose an assessment framework that we call "Sigmoid Accuracy", which characterizes forest change detection accuracy as it relates to timeliness. Using a change monitoring algorithm that applies the Continuous Change Detection and Classification algorithm (CCDC) to Sentinel-1 radar data in Madagascar, we demonstrate how the assessment framework can be used to calculate a lag-dependent F1-Score to create a sigmoidal curve describing system performance as it relates to the time since a change event. We introduce two new performance metrics that define key moments along this curve: "Initial Delay", or the minimum time required to create an alert as observed in reference data, and the "Level Off Point", or the lag at which accuracy stabilizes. Our framework accommodates varying assessment designs, as demonstrated using pixel-, block-, and event-based agreement for the response design. These metrics provide a holistic way to evaluate and compare the usefulness of forest monitoring systems in real-world applications.
NSTL
Elsevier
