Results from an ice stress buoy deployed near the center of a multi‐year floe in the Viscount Melville Sound of the Canadian Arctic Archipelago between 10 October 2010 and 17 August 2011 are presented. The position record indicates the landlocked season was approximately 5 months, from 18 January to 22 June, when the sea ice was fast to Melville Island and Victoria Island. Thermal stresses (ranging from −84 to 66 kPa) dominate the internal stress record, with only a few dynamic stress events (∼50 kPa) recorded before the landlocked season. Intriguingly, the thermal stresses are isotropic before the landlocked ice onset and anisotropic during the landlocked season. Two possible causes to explain anisotropy in thermal stresses are considered: preferred c axis alignment of the ice crystal, and land confinement associated with the nearby coastline. The orientation of the principal stresses indicates that land confinement is responsible for the anisotropy. The stress record also clearly shows the presence of residual compressive stresses at the melt onset, suggesting a viscous creep relaxation time constant of several days. Finally, results show an interesting reversal in the sign of the correlation (from negative to positive) between surface air temperature and thermal stress after the onset of surface melt. We attribute this to the onset of water infiltration within sea ice after which colder night temperature leads to refreezing and compressive stresses. To the best of the authors' knowledge, this is the first time that anisotropic thermal stresses have been reported in sea ice.

We present a 1.5‐D thermal stress model that takes into account the effect of land confinement, which causes anisotropy in thermal stresses. To this end, we fix the total strain in the direction perpendicular to the coastline to its value at landlocked ice onset. This prevents thermal expansion in the direction perpendicular to the coastline and therefore induces larger thermal stresses in this direction. The simulated stresses best match the observations, when a Young's Modulus of 0.5 GPa and a relaxation time constant of 8 days are used. This simulation gives root‐mean‐square errors of 13.0 and 13.1 kPa (∼15%) in the major and minor principal stresses, respectively. The simulated anisotropic component of thermal stress also generally agrees with observations. The optimal Young's Modulus is in the low range of reported values in the literature, and the optimal relaxation time constant (8 days) is larger than the largest relaxation time constant reported in the literature (5 days). A series of experiments are done to examine the model sensitivity to vertical resolution, snow cover, and the parameterizations of Young's Modulus and viscous creep. Results show that a minimum of one and three layers in the snow and ice, respectively, is required to simulate the thermal stresses within 15% error of the value assessed with the higher‐resolution control simulation. This highlights the importance of resolving the internal snow and ice vertical temperature profile in order to properly model the thermal stresses of sea ice.