New algorithm-backed tool offers accurate tracking for deforestation crisis

New tool uses algorithms for precise deforestation detection

Scientists use the National Land Cover Database to comprehensively view landscape changes. It transforms satellite images (Landsat) into detailed maps of different features. 

From 2001 to 2016, almost half of the land cover change in the contiguous United States (CONUS)— the 48 adjoining states located within North America, excluding Alaska and Hawaii— occurred in forests, as revealed by the database.

Jin emphasized that to ensure the quality of the database's products, accurately detecting the location and timing of forest disturbance is critical. To do this, Jin and her team combined 2-date and time-series change detection methods, which improved mapping efficiency, flexibility, and accuracy.

To explain simply, 2-date algorithms are more flexible and use richer spectral information to detect forest disturbances accurately by analyzing changes in image bands, indices, classifications, and combinations. 

However, they only work for one time period. They may need extra information to differentiate forest changes from other land cover changes.

Time-series algorithms consider spectral and long-term temporal information, simultaneously providing changes for multiple dates. However, adding a new date requires processing the entire algorithm, which can be challenging for continuous monitoring and may introduce inconsistencies.

Previous studies suggested ensemble approaches, like stacking, to enhance forest change mapping accuracy by combining different methods. Although effective, stacking requires significant computational resources and reference data for training.

In this latest study, the researcher's combined approach created the NLCD 1986-2019 forest disturbance product. It displays the most recent forest disturbance date within two to three-year intervals between 1986 and 2019.

"The TSUN index detects multi-date forest land cover changes and was shown to be easily extended to a new date even when new images were processed in a different way than previous date images," Jin stated.

The research team intends to enhance the tool by increasing the frequency of time measurements and generating an annual forest disturbance product covering the period from 1986 to the present. 

"Our ultimate goal is to automatically produce forest disturbance maps with high accuracy with the capability of continually monitoring forest disturbance, hopefully in real-time," Jin concluded.

The complete study was published in the Journal of Remote Sensing and can be found here. 

Study abstract:

The National Land Cover Database (NLCD) 2016 products show that, between 2001 and 2016, nearly half of the land cover change in the conterminous United States (CONUS) involved forested areas. To ensure the quality of NLCD land cover and land cover change products, it is important to accurately detect the location and time of forest disturbance. We designed a comprehensive strategy to integrate a continuous time series forest change detection method and a discrete 2-date forest change detection method to produce the NLCD 1986–2019 forest disturbance product, which shows the most recent forest disturbance date between the years 1986 and 2019 for every 2- to 3-year interval. This method, the Time-Series method Using Normalized Spectral Distance (NSD) index (TSUN), uses NSD to detect multi-date forest land cover changes and was shown to be easily extended to a new date even when new images were processed in a different way than previous date images. The discrete 2-date method uses the Multi-Index Integrated Change Analysis (MIICA) method to detect changes between 2-date images. A method based on confidence and object grouping was designed to combine the multiple MIICA outputs to improve change detection accuracy. Finally, an aggregation scheme was implemented to combine the TSUN output, the integrated MIICA results, and ancillary data to produce the NLCD 2019 forest disturbance 1986–2019 product. The initial accuracy assessments from 1,600 samples over 4 Landsat path/rows show that the producer's and user's accuracies of the 2001–2019 forest disturbance map are 76% and 74%, respectively.