Monitoring land-cover and land-use dynamics in Fanjingshan National Nature Reserve. (October 2019)

• Record Type: Journal Article
• Authors: Tsai, Yu Hsin; Stow, Douglas; An, Li; Chen, Hsiang Ling; Lewison, Rebecca; Shi, Lei
• Is Part Of: Applied geography. Volume 111(2019)
• Publication Date: 2019-10
• Subjects: Land-cover and land-use change -- Payment for ecosystem services -- Protected area -- Landsat -- China -- Google earth engine; Geography -- Periodicals; Human geography -- Periodicals; Human ecology -- Periodicals; 910
• DOI: 10.1016/j.apgeog.2019.102077
• Languages: English
• ISSNs: 0143-6228

Abstract

Abstract: Fanjingshan National Nature Reserve (FNNR) in China is a biodiversity hotspot that is part of a larger, multi-use landscape where tourism, farming, grazing, and other land uses occur. Payment for ecosystem services (PES) programs that encourage afforestation on farmlands may be important drivers of land-cover and land-use change in the region that surrounds FNNR. Our objective is to monitor and examine vegetation and land-use changes, including PES-related afforestation, between 1989 and 2017. We utilize several image processing techniques, such as illumination normalization approaches to suppress terrain effects, and multi-seasonal image compositing to minimize persistent cloud cover. Ancillary data were also incorporated to generate reliable vegetation and land-use change information. A random forest machine learning image classification routine is implemented through the cloud-based Google Earth Engine platform and refined using optimal classifier parameter tuning. Land-use transitions are identified and mapped with the implementation of stable training sites, discrete image classification, and logical land-use transition rules. Accuracy assessment results indicate our change detection workflow provides a reliable methodology to remotely monitor long-term forest cover and land-use changes in this mountainous, forested, and cloud prevalent region. We quantify the area of new built development and afforestation land and found that most of the land transitions took place in reserve buffer and its adjacent environs. For example, less than 2 km 2 of new built was identified within the reserve boundary compared to 25 km 2 for the entire study area between 1995 and 2016. We also shed light on the strengths and weaknesses of using Google Earth Engine for land-cover and land-use change studies. This efficient and open-access technique is important not only for assessing environmental changes and PES efficacy, but also for evaluating other conservation policies elsewhere. Highlights: Remote monitoring in topographically complex and cloud prevalent area. Efficient cloud-based machine learning processing workflow. Multi-temporal composites to minimize persistent cloud cover effects. Improved accuracy with seasonal image composites, shade, and illumination normalization. Logical land-use transition rules for reliable land-use change monitoring.