Comparing Different Spectral Indices in Monitoring Forest Disturbances Using Landsat Time Series Imagery

Southern Appalachian Mountains has been recently threatened by large-scale disturbances, such as wildfire and insect pest, which can alter the forest ecosystem structures and functions. Hemlock Woolly Adelgid (Adelges tsugae Annand, HWA) is a non-native pest that causes widespread foliar damage and potentially irreversible tree mortality in eastern (Tsuga Canadensis L.) and Carolina (Tsuaga caroliniana Engelm.) hemlocks throughout eastern United States. To better understand the implications of hemlock declines in biogeochemical cycles in forest landscapes, it is important to map the hemlock morbidity distribution and its recovery patterns in space and time. Recently, several change detection techniques using all available satellite images have been developed for land use and land cover changes; however, there are few studies in application of forest disturbance monitoring covering from pre-disturbance to full recovery stages. In this study, we used all available Landsat satellite images to investigate and compare the performance of Tasseled Cap Transformation (TCT) based indices, NDVI, and Disturbance Index (DI) in capturing spectral-temporal trajectory of forest disturbance (fire and HWA infestation) and following regenerations process in southern Appalachians. For each Landsat pixel, the temporal trajectories of the metrics were fitted using a time series analyses separating inter-annual (low frequency) disturbance patterns and seasonal phenology (high frequency) signals. We estimated the temporal dynamic of disturbance and following recovery based on the residuals between observed values and predicted values of the model. In addition, we investigated the performance of all metrics in capturing the infestation intensity and further validated those with individual hemlock tree locations identified from high-resolution aerial photos. Investigating the performance of all metrics indicated that the overall performance of TC wetness (TCW), DI and NDVI were more accurate in detecting both disturbance timing and HWA infestation intensity. However, NDVI capability in detecting the infestation intensity was not consistently good as DI and TCW. Despite the overall accurate performance of TCW in characterizing disturbance properties, this index has some limitations in detecting forest regeneration pattern, and DI and NDVI were better predictors of recovery stage characteristics. Our results suggested that using frequent Landsat observations and fitting time series analysis can improve both the accuracy of forest disturbance and recovery characterizing.