Abstract
There is worldwide interest in monitoring changes in forest area for carbon budgeting and assessing biodiversity changes. For cost-effective monitoring over large areas, automated mapping is necessary. However, the accuracy of automated mapping is rarely high enough to permit accurate estimation of changes in forest area when there is small change. In this article, we show how accurate area statistics of small change can be achieved by combining automated change mapping with stratified random sampling. The stratified random sampling can be made efficient by exhaustive sampling of a stratum with contiguous areas of apparent change greater than or equal to some lower limit (5 ha in the case presented here) and splitting the forest-to-forest stratum into large and small spectral change. Efficiency of the method is high when the exhaustively sampled stratum contains a high proportion of all real change. A regional example from New Zealand of mapping change in forest area is given. Acceptable accuracy (afforestation is estimated to ±7% of reported afforestation) from acceptable effort (10,702 sample points) is obtained, representing an efficiency gain of 78% (in terms of sample size) over simple random sampling. The method can be applied to national estimation of forest change to the same accuracy with an efficiency gain of approximately 70%. It provides a more rapid alternative to multipurpose forest inventories, which depend on intensive and expensive field data, and has significant implications for carbon accounting at regional and national scales, where rapid production of usable estimates are required for carbon exchange markets.