A comparison of forest cover maps in Mainland Southeast Asia from multiple sources: PALSAR, MERIS, MODIS and FRA

Abstract The uncertainty in tracking tropical forest extent and changes substantially affects our assessment of the consequences of forest change on the global carbon cycle, biodiversity and ecosystem services. Recently cloud-free imagery useful for tropical forest mapping from the Phased Array Type L-band Synthetic Aperture Radar (PALSAR) onboard the Advanced Land Observing Satellite (ALOS) has become available. We used PALSAR 50-m orthorectified mosaic imagery in 2009 and a decision tree method to conduct land cover classification and generate a 2009 forest map, which was evaluated using 2106 field photos from the Global Geo-referenced Field Photo Library ( http://www.eomf.ou.edu/photos ). The resulting land cover classification had a high overall accuracy of 93.3% and a Kappa Coefficient of 0.9. The PALSAR-based forest map was then compared with three existing forest cover products at three scales (regional, national, and continental): the Food and Agriculture Organization of the United Nations (FAO) Forest Resources Assessments (FRA) 2010, Global Land Cover Map with MERIS (GlobCover) 2009, and the MODIS Terra + Aqua Land Cover Type product (MCD12Q1) 2009. The intercomparison results show that these four forest datasets differ. The PALSAR-based forest area estimate is within the range (6.1–9.0 × 10 5  km 2 ) of the other three products and closest to the FAO FRA 2010 estimate. The spatial disagreements of the PALSAR-based forest, MCD12Q1 forest and GlobCover forest are evident; however, the PALSAR-based forest map provides more details (50-m spatial resolution) and high accuracy (the Producer's and the User's Accuracies were 88% and 95%, respectively) and PALSAR can be used to evaluate MCD12Q1 2009 and GlobCover 2009 forest maps. Given the higher spatial resolution, PALSAR-based forest products could further improve the modeling accuracy of carbon cycle in tropical forests.