Detecting and modelling dynamic landuse change using multitemporal and multi-sensor imagery

It is now common to use data from two or more sensors for land cover change detection. Since the spatial and spectral resolutions of different sensors vary significantly, the ability to discriminate the land cover also varies greatly. In this paper the applications of landuse change detection including area statistics, temporal trajectories and spatial pattern are discussed. The area statistics show the general landuse change pattern, but with quite significant uncertainty. The results of this study show that if the area of detected landuse change accounts for less than 5% of the total area, the uncertainty of change detection can be very significant. Temporal trajectory analysis was also conducted with the particular focus on the analysis of unchanged and "stable" change trajectories, because they generally show the trend of landuse change that is irreversible. Unstable change trajectories, on the other hand, show relatively less significance since they largely contain reversible temporary changes (e.g. seasonal cropping and bare ground) and classification errors. The study results show overall accuracy of 85-90% with Kappa coefficients of 0.66-0.78 in classification and change detection. On spatial patterns, the landuse pattern metrics demonstrate a reasonable result, but most other patch metrics do not show recognisable patterns.