Optimal Threshold Value Determination for Land Change Detection

Recently data mining techniques have emerged as an important technique to detect land change by detecting the sudden change and/or gradual change in time series of vegetation index dataset. In this technique, the algorithms takes the vegetation index time series data set as input and provides a list of change scores as output and each change score corresponding to a particular location. If the change score of a location is greater than some threshold value, then that location is considered as change. In this paper, we proposed a two step process for threshold determination: first step determine the upper and lower boundary for threshold and second step find the optimal point between upper and lower boundary, for change detection algorithm. Further, by engaging this process, we determine the threshold value for both Recursive Merging Algorithm and Recursive Search Algorithm and presented a comparative study of these algorithms for detecting changes in time series data. These techniques are evaluated quantitatively using synthetic dataset, which is analogous to vegetation index time series data set. The quantitative evaluation of the algorithms shows that the Recursive Merging (RM) method performs reasonably well, but the Recursive Search Algorithm (RSA) significantly outperforms in the presence of cyclic data.