A comparison of fine resolution census and image-based national forest inventories: A case study of Rwanda

Abstract

Comparisons of national forest inventories have been undertaken at coarse resolutions in Europe, but not at fine resolutions and not in Africa. This study compares the consecutive (1990, 2000, 2005, 2010), census-based Food and Agriculture Organisation Forest Resources Assessments (FRAs) of Rwanda with contemporary fine-resolution, image-based inventories. Natural forest and forest plantation were extracted from 1985 to 1989 topographic maps, Africover 2002 and Centre of Geographic Information Systems and Remote Sensing (CGIS) 2007. The natural forest area estimates over the years obtained by image-based inventories are fairly constant from the earliest inventory (topo-1988; 140 000 ha), through Africover (135 500 ha) to CGIS (125 300 ha). FRA 2000 (46 000 ha), 2005 (61 000 ha) and 2010 (62 000 ha) estimates are lower, whereas FRA 1990 (164 000 ha) reported the largest extent of natural forest. The extent of forest plantations is at its lowest level in Africover (68 600 ha). However, CGIS (114 000 ha) detected a larger area of forest plantation, more similar to the levels reported in the FRA 1990 (125 000 ha) and topo-1988 (125 200 ha). The three recent FRAs (2000: 261 000 ha; 2005: 419 000 ha; 2010: 373 000 ha) report larger plantation extents. For the image-based inventories, we could estimate the uncertainty in each step of the inventory, while for the census-based cases, the required transparency is lacking. Uncertainty may be caused by definition of forest, forest versus forestland, image classification errors, image resolution, forest fragmentation or time-lags between imagery and reporting. In our pairwise comparison of the census and contemporary image-based inventories, all uncertainties are added. The FRA 2000 (307 000 ha) reports a larger total forest area than Africover (204 100 ha). The uncertainties of Africover add up to maximally 64 100 ha, which may explain part of the difference. The uncertainties in the CGIS inventory add up to nearly 15 000 ha and cannot explain the difference between FRA 2005 (480 000 ha) and GCGIS 2007 (239 300 ha).We conclude that FRAs starting in 2000 over-report the forest plantation area and under-estimate the extent of natural forest. Finally, we suggest technical and organizational improvements for national forest inventories. © Institute of Chartered Foresters, 2013. All rights reserved.