| Summary: | The ground thermal regime of Arctic and Subarctic regions is impacted by climate and local scale factors. Snow cover has been linked to differences in ground temperatures over short distances (< 5 m) because of its thermal buffering properties which slows energy exchanges between the ground and atmosphere. Accurate derivation and characterization of the effects of snow cover on the ground thermal regime is essential for predicting the future impacts of climate change in northern Canada. This thesis presents a machine learning-based method for estimating snow cover from local ground surface temperature (GST) and air temperature measurements and was tested using modelled and in situ data. Results were compared against two other commonly used snow prediction methods, which select thresholds of either 1) the standard deviation of GST or 2) the difference between the standard deviations of air and surface temperatures. The machine learning method showed better performance for the modelled data and comparable performance with the in situ data compared to the other techniques. Variations in snow and ground temperatures were further analyzed with extensive field investigations at two field sites in coastal Labrador which includes a permafrost probability analysis. Results showed that mean annual ground surface temperatures (MAGST) was significantly correlated (p < 0.05) at our southern site with microclimate indices (freezing n-factor [r=-0.70], surface offset [r=0.99], nival offset [r=0.71]) while our northern site had significant correlations with both in situ ecosystem focused indices (snow depth [r=0.78], snow water equivalent [r=0.77]) and microclimate indices (freezing n-factor [r=-0.93], surface offset [r=0.99], nival offset [0.93]). Permafrost probability results showed a 10% likelihood across all 35 logger locations with all probable permafrost locations within our northern site at tundra and wetland ecotypes of low snow accumulation. Overall, this research will improve our ability to model snow-ground ...
|
|---|