bul 2004 
TOWARDS AN UNDERSTANDING OF UNCERTAINTY 
IN GREENHOUSE FOREST ASSESSMENTS 
S. D. Jones™ *, G. Richards ®, K. Lowell ¢, P. Woodgate 9, L. Buxton ® 
* School of Mathematical and Geospatial Sciences, RMIT University, 
GPO Box 2476V, Melbourne 3001, VIC Australia simon.jones@rmit.edu.au 
? National Carbon Accounting System, Australian Greenhouse Office, 
Canberra, A.C.T. Australia gary.richards@greenhouse.gov.au 
“ Centre de recherche en géomatique, Université Laval, 
Pav. Casault Ste-Foy, Québec G1K 7P4 Canada Kim.Lowell@scg.ulaval.ca 
3 Cooperative Research Centre for Spatial Information, 
University of Melbourne, VIC 3010 Australia peterww@unimelb.edu.au 
Commission TS WG VIU/6 Monitoring and Modelling Global Change 
KEY WORDS: Remote Sensing, Vegetation, Land Cover, Analysis, Monitoring, Global-Environmental-Databases 
ABSTRACT: 
Australia's National Carbon Accounting System provides information on land-based sources and sinks of greenhouse gases to fulfil 
international reporting obligations under the Kyoto Protocol, as well as providing annual estimates to Australia's National 
Greenhouse Gas Inventory. Manifold is an understanding of change in forest area: afforestation, reforestation and deforestation 
events. Using a thirty-year archive of Landsat imagery (1972-2002), a set of 12 continent-wide land cover maps, and associated 
change layers for the 11 intervals was created. A continuous probability network was then used to estimate the probability of a pixel 
belonging to Forest or Non-Forest classes for each of these 12 dates. These Forest/Non-forest classifications, from successive dates, 
were then compared on a pixel-by-pixel basis to identify areas of No Change (Forest), No Change (Non-forest), Deforestation, and 
Regrowth. To gain an understanding of the uncertainty in these change maps, and so that improvements could be made in the 
mapping technique, a fuzzy evaluation methodology was developed and implemented. A network of ~300 aerial photographs was 
co-registered to the database and more than ~12,000 points were compared using photo interpretation to validate the matching pixels 
on each respective change map. The classes used for the photographic interpretation were Definitely Forest, Probably Forest, 
Unsure, Probably Non-forest, and Definitely Non-forest. Australia-wide the error rates were very low. The ‘definite’ errors for 
forest were ~2% and ‘definite’ errors for non-forest ~4%. Hotspots of uncertainty in forest change errors did emerge however in 
some forested areas (up to 5.7%). To improve the temporal classification process, a performance analysis was undertaken that cross- 
referenced reported change in forest area with reported errors in classification. This process will be repeated with each continent- 
wide land cover map update to provide progressive improvement in the change maps. 
1. INTRODUCTION 
Fundamental to accounting for carbon change in land use is an 
understanding of the change in land cover. The impact of an 
event associated with land cover change may continue over 
many years and vary with time since the event took place. It is, 
therefore, necessary to monitor change in land cover over 
extended periods of time. To be considered in the NCAS 
accounting framework, the land cover change must also be 
shown to be directly human-induced, that is a deliberate, not an 
indirect natural event (Furby, 2002). 
1.1 Background 
Article 4.1(1) of the United Nations Framework Convention on 
Climate Change (UNFCCC) commits Australia to produce an 
annual inventory of national greenhouse gas emissions 
according to the Revised 1996 Intergovernmental Panel on 
Climate Change (IPCC) Guidelines for National Greenhouse 
Gas Inventories’. The inventory reports human-induced 
greenhouse gas emissions, by sources and removals by sinks, 
not controlled by the Montreal Protocol, in six sectors: energy, 
industrial processes, solvent and other produot use, agriculture, 1.2 Continental Database 
land use change and forestry. Reducing the levels of 
  
uncertainty previously associated with estimates of land use 
change emissions is essential as they are a significant 
component of Australia’s greenhouse gas emissions profile. In 
1998, to achieve the above, the Australian Government 
commenced the development of its National Carbon 
Accounting System (NCAS). 
  
* Corresponding author. 
Landsat data were used to create a continent-wide (Australia 
including Tasmania) database of Forest / Non-Forest Landcover 
for 12 time periods, spanning 1972 to 2002 (1972, 1978, 1980, 
1985, 1988, 1989, 1991, 1992, 1995, 1998, 2000 and 2002). 
Forests are defined as having a minimum of 20% tree crown 
cover and a minimum height of 2 metres at maturity. A 
United Nations Framework Convention on Climate Change, Article 4, para. 3. 
747