Browsing School of Natural Resources and Agricultural Sciences by Subject "Remote sensing"
Now showing items 1-1 of 1
Assessment of LiDAR and spectral techniques for high-resolution mapping of permafrost on the Yukon-Kuskokwim Delta, AlaskaThe Yukon-Kuskokwim Delta (YKD) is one of the largest and most ecologically productive coastal wetland regions in the pan-Arctic. Formed by the Yukon and Kuskokwim Rivers flowing into the Bering Sea, nearly 130,000 square kilometers of delta support 23,000 Alaskan Natives living subsistence lifestyles. Permafrost on the outer delta commonly occurs on the abandoned floodplain deposits. Ground ice in the soil raises surface elevations on the order of 1-2 meters, creating plateaus on the landscape. Better drainage on the plateaus supports distinct Sphagnum-rich vegetation, which in turn protects the permafrost from rising air temperatures with low thermal conductivity during the summer. This ecosystem-protected permafrost is thus vulnerable to disturbances from rising air temperatures, vegetation mortality, and inland storm surges, which have been known to flood up to 37 km inland. This thesis assesses several novel techniques to map permafrost distribution at high-resolution on the YKD. Accurate baseline maps of permafrost extent are critical for a variety of applications, including long-term monitoring. As air and ground temperatures rise across the Arctic, monitoring landscape change is important for understanding permafrost degradation processes (e.g. thermokarst) and greenhouse gas dynamics from the local to global scales. This thesis separately explored the value of Light Detection And Ranging (LiDAR) and spectral datasets as tools to map permafrost at a high spatial resolution. Furthermore, this thesis sought to automate these processes, with the vision of high-resolution mapping over large spatial extents. Fieldwork was conducted in July 2016 to both parameterize and then validate the mapping efforts. The LiDAR mapping extent assessed a 135 km² area (~15% permafrost cover), and the spectral mapping extent assessed an 8 km² area (~20% permafrost cover). For the LiDAR dataset, the use of a simple elevation threshold informed by field ground truth values provided a permafrost map with 94.9% accuracy. This simple approach was possible because of the extremely flat terrain. For the spectral datasets, an ad-hoc masking technique was developed using a combination of texture analysis, principal component analysis, and morphological filtering. Two contrasting workflows were evaluated with fully-automated and semi-automated methods with mixed results. The highest mapping accuracy was 89.4% and the lowest was 79.1%, though the error of omission in mapping the permafrost remained high (7.02 - 59.7%) for most analyses. The spectral mapping algorithms did not replicate well across different high-resolution images, raising questions about the viability of using spectral methods alone to track thermokarst and landscape change over time. However, incorporating the spectral methods explored in this analysis with other datasets (e.g. LiDAR) has the potential to increase mapping accuracies. Both the methods and the results of this thesis enhance permafrost mapping efforts on the YKD, and provide a good first step to monitoring landscape change in the region.