A practical sampling method for assessing accuracy of detected land cover/land use change: Theoretical analysis and simulation experiments

Accuracy assessment plays a crucial role in the implementation of change detection, which is commonly used to track land surface changes and ecosystem dynamics. There are currently two major indicators for accuracy assessment of change detection: the binary change accuracy (ca) and the overall transition accuracy (ta). The overall transition accuracy has been recommended over change accuracy, because the binary change accuracy does not consider the accuracy of the types of changes of the underlying land cover classes. However, the application of overall transition accuracy has been limited by the challenge of collecting enough representative samples with a practical sampling strategy to meet the users' requirement of precision. This study provides an iterative sampling framework to ensure that the precision of the estimated overall transition accuracy meets the users' predefined requirement. We use a set of simulated change maps to comprehensively examine the effectiveness and robustness of the proposed sampling strategy. The simulation-based results demonstrate that the proposed framework can achieve satisfactory performance for transition accuracy assessment and it is robust against different properties of classification results and target landscapes, including the degree of fragmentation, proportions of land cover types, and temporal correlation of the classification error between individual dates. The effectiveness, robustness and practicality of the proposed sampling strategy will enable producers and users of land cover/land use change maps to get reliable and meaningful accuracy assessment for further applications.