Land transitions from multivariate remotely sensed time series: Using seasonal trends to characterize and categorize land cover changes in Alaska over the 2001-2009 period

The Earth is undergoing climate, environmental and land cover changes at an increasing pace. Monitoring land cover change is crucial as land transitions affects the Earth system's functioning by disrupting biogeochemical cycles such as water and carbon and influencing surface energy exchanges. Within this context, documenting changes in Alaska is paramount because the region has a large global reservoir of carbon stored in its soils and its standing biomass. Under warming scenarios of climate change, Alaska is poised to undergo changes that will turn the region from a carbon sink to a carbon source, an event potentially affecting the entire Earth's system.;Remote Sensing provides remarkable datasets to study and track land transitions over large areas such as Alaska. In particular, the literature suggests that, using land surface temperature (LST), normalized difference vegetation index (NDVI) and albedo (ALB) variables that are closely related to biophysical and surface processes, provide valuable information to detect, characterize and categorize land transitions. However, there are many methodological challenges in using such data time series. First, noise affects the signal making it difficult to extract useful temporal patterns to study land transitions. Second, time series are typically short posing problems when applying existing methods from classical statistics. Third, mapping requires grouping areas into land cover change categories which is an arduous task that necessitates combining temporal information from several time series into meaningful classes while simultaneously dealing with noise and large spatial variability.;This dissertation addresses these methodological challenges by conducting three independent pieces of research organized around three core chapters. All three chapters use the Seasonal Trend Analysis (STA) method to extract temporal information in the form of nine seasonal trends from the NDVI, LST and ALB time series derived from the Moderate Resolution imaging Spectroradiometer (MODIS) over the 2001-2009 time period. The first chapter focuses on the process of detection of changes and introduces the use of segmentation in conjunction with STA to deal with noise and the high spatial variability. Segments are groups of spatially contiguous pixels undergoing similar trends and constitute more natural and meaningful units of analysis than pixels. Using such segments, it was found that most of Alaska (40.71%) is undergoing significant land transitions. The second chapter focuses on the process of characterization of change related to fire disturbances by using Principal Component Analysis (PCA) to organize seasonal trends extracted from STA. Results indicate that land transitions related to fire disturbance may be characterized as changes in four seasonal trends: decrease in average NDVI (NDVI_A0), increases in Albedo (average, ALB_A0 and annual variability ALB_A1) and increase in annual surface variability (LST_A1). The third chapter presents a method to group areas of changes into meaningful large clusters by using the ChainCluster algorithm with segments. Results indicate that clusters are related to the WWF 2001 ecoregions and broadly agree with environmental changes recorded in the literature.;Thus, the dissertation offers three contributions to the field of Land Change Science and Arctic System Science by: 1) documenting the types and extent of land transitions in Alaska, a key region in global environmental change; 2) contributing to the advancement of methods in large area mapping and land cover monitoring; 3) documenting biophysical changes associated to land transitions such as fire using three remotely sensed biophysical variables.