Mapping tropical forests and deciduous rubber plantations in Hainan Island, China by integrating PALSAR 25-m and multi-temporal Landsat images

Abstract

Updated and accurate maps of tropical forests and industrial plantations, like rubber plantations, are essential for understanding carbon cycle and optimal forest management practices, but existing optical-imagery-based efforts are greatly limited by frequent cloud cover. Here we explored the potential utility of integrating 25-m cloud-free Phased Array type L-band Synthetic Aperture Radar (PALSAR) mosaic product and multi-temporal Landsat images to map forests and rubber plantations in Hainan Island, China. Based on structure information detected by PALSAR and yearly maximum Normalized Difference Vegetation Index (NDVI), we first identified and mapped forests with a producer accuracy (PA) of 96% and user accuracy (UA) of 98%. The resultant forest map showed reasonable spatial and areal agreements with the optical-based forest maps of Fine Resolution Observation and Monitoring Global Land Clover (FROM-GLC) and GlobalLand30. We then extracted rubber plantations from the forest map according to their deciduous features (using minimum Land Surface Water Index, LSWI) and rapid changes in canopies during Rubber Defoliation and Foliation (RDF) period (using standard deviation of LSWI) and dense canopy in growing season (using maximum NDVI). The rubber plantation map yielded a high accuracy when validated by ground truth-based data (PA/UA>86%) and evaluated with three farm-scale rubber plantation maps (PA/UA>88%). It is estimated that in 2010, Hainan Island had 2.11×106ha of forest and 5.15×105ha of rubber plantations. This study has demonstrated the potential of integrating 25-m PALSAR-based structure information, and Landsat-based spectral and phenology information for mapping tropical forests and rubber plantations.