| This research focused on the development of methods that draw on the spatial autocorrelation, spectral content, hierarchical relationships and hypertemporal content inherent in remotely sensed datasets to model the dynamics of post-disturbance forest regeneration. Spectral and spatial components of Landsat 7 Enhanced Thematic Mapper Plus (ETM+) data were used to discriminate forest stand classes. The application of the spatial component of the data improved the classification by 36%, compared to a classification based on only the spectral content of the imagery. Therefore, the spatial component of the data allowed for the best discrimination of forest successional classes. The use of hyperspectral data allowed for mapping of seven gradients of seedling regeneration density; here too, the spatial content of the hyperspectral data were responsible for a 10% improvement in classification accuracies. Harmonic analysis of hypertemporal remotely sensed data provided a replicable method of quantifying and monitoring the temporal dynamics of forested landscapes, including post-fire regenerating forests and forest impacted by anthropogenic disturbances such as harvesting. The method showed that naturally regenerating forests are more temporally diverse than harvested and replanted forests, this is evident in the interannual and seasonal temporal trends. The hierarchical object-oriented image analysis outperformed the per-pixel classification of Landsat imagery by over 10%, however, the method was not as effective as the application of hyperspectral data, that showed improvement in accuracies of approximately 8%. Statistical analysis of the indirect optical method and the allometric method of obtaining leaf area index (LAI) estimates of coniferous temperate forest, demonstrated that the allometric approach produced superior fitting models. The LAI geostatistically-based cokriged models were superior compared to aspatial methods. An aspatial LAI model for post-fire regenerating forest class provided an R2 of 0.83, thus, about 17% of the variability of the predicted variables was still unexplained by the model. The geostatistically-based cokriged model showed a lower percentage of the predicted variable unexplained by the models at less than 2%, for the post-fire regenerating forest. Finally, geovisualizations were used to disseminate the research results of the spatiotemporal forest landscape change to decision makers and the public. |
