Novel change detection methods for multi-date digital imagery applied to south florida vegetation

Remote sensing using multidate imagery allows for change detection and the analysis of important landscape processes over time. Multidate image analysis has been used to map, measure, monitor and model important changes related to topics including deforestation, loss of wetlands, drought and flooding, and urban change. Existing change detection methods have proven themselves valuable, but are limited in terms of the patterns they can detect, the need for analyst intervention, and ease of interpretation. As the volume of remotely sensed data increases and the price of data and computing facilities decreases, new techniques are needed for the rapid automated or semi-automated identification of change patterns. This research presents a number of novel methods for analyzing and visualizing change in remotely sensed data sets. One approach includes the application of parametric measures (standard deviation, range, slope) to a time series. A second approach involves the visualization of data transformed into the temporal shape domain. The third approach involves the classification of temporal patterns by neural networks. The novel techniques were proven using synthetic data, then applied to anniversary AVHRR NDVI composite images of South Florida from 1989 through 1993. For the Florida data, the results from the novel methods were compared with the results of standard methods including an unsupervised classification, principal components analysis, and write function memory insertion. A comparison of results indicates that the novel methods do uncover information that is different from, but consistent with, the standard methods. The novel methods are able to detect specific change patterns that the standard methods cannot. The novel methods are easier to interpret than the standard methods, and can contribute to the interpretation of the standard methods.