Four Landsat path/row footprints in purple, which were selected for accuracy assessment, are overlaid on the National Land Cover Database (NLCD) Forest Disturbance Date 1986-2019 science product.
Newswise — Every minute, the planet forfeits a segment of woodland comparable to a soccer pitch due to logging, fires, insect invasion, illness, wind, drought, and additional causes. In a freshly released analysis, scientists from the U.S. Geological Survey Earth Resources Observation and Science (EROS) Center unveiled an all-encompassing plan to identify the timing and locations of forest disruption on a vast level and offer a more profound comprehension of forest alteration.
The study was published on Feb. 28 in the Journal of Remote Sensing.
"Our approach yields enhanced precision in land cover mapping and revision," expressed Suming Jin, a physical scientist affiliated with the EROS Center.
In order to grasp the overall transformation of a dynamic terrain, researchers depend on the National Land Cover Database, which converts Earth-observation satellite (Landsat) imagery into detailed maps of distinct characteristics, pixel by pixel. From 2001 to 2016, the database revealed that approximately 50% of land cover alteration in the contiguous United States was attributed to changes in wooded regions.
"Ensuring the accuracy of National Land Cover Database's land cover and land cover change products requires precise identification of forest disturbance in terms of location and timing," emphasized Jin.
Jin and the research team devised a technique to identify forest disturbance on an annual basis. This method combines the advantages of a time-series algorithm and a 2-date detection method, resulting in enhanced efficiency, flexibility, and accuracy for operational mapping of vast regions. This innovative approach enables better forest management, policy-making, and various other applications to be carried out with greater effectiveness.
The extensive utilization of Landsat data in forest disturbance detection is due to its prolonged historical record, exceptional spatial and radiometric resolutions, open and freely accessible data policy, and its suitability for generating comprehensive mosaic images for various seasons, spanning continental or even global scales.
"We require algorithms that can generate consistent forest disturbance maps for extensive regions, aiding in the production of multi-epoch National Land Cover Database," Jin stated. "Additionally, these algorithms need to be scalable to enable us to track forest changes over extended periods of time."
The "2-date forest change detection" method is frequently employed, which entails comparing images captured on two different dates. On the other hand, the "time-series algorithm" leverages Landsat time series data to provide observations at a yearly or even monthly frequency.
Typically, 2-date forest change detection algorithms offer greater flexibility compared to time-series methods and make use of more comprehensive spectral information. The 2-date method excels at identifying changes between image bands, indices, classifications, and combinations, thereby enabling more precise detection of forest disturbances. However, it should be noted that the 2-date method only identifies changes within a single time period and often necessitates supplementary information or additional processing to differentiate forest changes from other types of land cover changes.
In contrast, time-series-based forest change detection algorithms leverage both spectral and long-term temporal information, allowing them to detect changes for multiple dates concurrently. However, these methods typically necessitate reprocessing every step of the time series algorithm when a new date is added, which can be burdensome for continuous monitoring updates and may introduce inconsistencies.
Prior research has suggested ensemble approaches to enhance the accuracy of forest change mapping, such as the technique of "stacking," which involves combining the outputs of various mapping methods. Stacking has demonstrated effectiveness in reducing both omission and commission errors. However, it should be noted that this method is computationally demanding and relies on reference data for training purposes.
Jin and the team's approach integrated the advantages of 2-date change detection methods and the continuous time-series change detection method, specifically the Time-Series method Using Normalized Spectral Distance (NSD) index (TSUN). This combination aimed to enhance the efficiency, flexibility, and accuracy of operational mapping for extensive regions. By implementing this hybrid technique, the researchers successfully generated the NLCD 1986-2019 forest disturbance product. This product provides information on the most recent forest disturbance date within two-to-three-year intervals between the years 1986 and 2019.
"The TSUN index excels at detecting land cover changes in forests across multiple dates, and it has been demonstrated to be easily adaptable to new dates, even when new images are processed differently from previous date images," Jin explained.
The research team plans to improve the tool by increasing the time frequency and produce an annual forest disturbance product from 1986 to present.
"Our ultimate objective is to achieve high-precision automatic generation of forest disturbance maps, while also possessing the capability to continually monitor forest disturbances, ideally in real-time," Jin expressed.
This research received support from the USGS-NASA Landsat Science Team Program, specifically for the project aimed at advancing the monitoring and characterization of land surface change in the contiguous United States towards near real-time capabilities.
RSS