Analyzing the agreement of maps through spatial aggregations
尺度变换的正确率分析

Spatial aggregation of raster data based on majority and random rule were used in this study. To access the agreement of aggregation/scaling-up effects on landscape patterns, a classified TM imagery (8 cover types) covering 1.37 million ha with 30m resolution was aggregated incrementally from 30m to 990m. For proportions of most common cover types in majority rule-based aggregation increased slowly, whereas proportions of less common cover types decreased rapidly with increasing resolutions. For random rule-based aggregation, proportion of each cover type remained constant value. Kappa index for no ability, for location, for quantity and standard Kappa index decreased with increasing scales in majority and random rule-based aggregations. For Majority rule-based aggregation, Kappa index for quantity decreased gradually, but for random rule-based aggregation, it maintained 100%. Agreements of maps obtained from majority rule-based aggregation are higher than those from random rule-based aggregation. Agreements due to quantity increased in majority rule-based aggregation, but it maintained a fixed value 9.64% in random rule-based aggregation with increasing resolutions. Agreement due to chance maintained 12.50% in all aggregations. Agreement due to location obviously decreased, whereas error due to location substantially increased. There were no apparent changes in agreement and error due to location at stratum and substratum levels in all aggregations. To the contrary, agreement and error due to location at grid cell levels substantially increased in all aggregations. Agreement and error due to location at grid cell levels determined the agreement and error due to location, furthermore, agreement and error due to location determined the agreement and error of the whole map. If standard kappa was higher than 70% was considered satisfactory, the critical value in spatial scale was 210m for majority rule-based aggregation, and it was 150m for random rule-based aggregation. If a higher critical value was needed in study, the extent or classification system should be altered according to objective of study.