An assessment of global forest cover maps using regional higher-resolution reference data sets

Six coarse-resolution global land cover maps are assessed for forest mapping using higher-resolution (i.e. 30-m, "Landsat" resolution) regional land cover products as reference. All products are aggregated spatially and thematically to derive fractional forest cover maps at 5-km spatial resolution. Forest-cover accuracy varies by product and continent. Compared with reference, GLC2000, GLCC and GlobCover consistently overestimate forest cover while MODIS VCF consistently underestimates in North America, South America and Africa. MODIS LC and UMD LC underestimate forest cover in North America and Africa but overestimate in South America. The products\´ root mean square errors (RMSE) against reference vary from 21.73% to 27.71% for North America, from 23.23% to 32.77% for South America, and from 20.10% to 29.76% for Africa.