The accelerated warming Arctic climate may alter the surface energy balance locally and regionally of which a changing land surface temperature (LST) is a key indicator. Modelling current and anticipated changes of the surface energy balance requires an understanding of the spatio-temporal interactions between LST and land cover. This paper investigated the accuracy of MODIS LST V5 1 km level 3 product and its spatio-temporal sensitivity to land cover properties in a Canadian High Arctic permafrost landscape. Land cover ranged from fully vegetated moss/segde grass tundra to sparsely vegetated bare soil and barren areas. Daily mean MODIS LST were compared to in-situ radiometer measurements over wet tundra for three summers and two winters in 2008, 2009, and 2010. MODIS LST showed an accuracy of 1.8°C and a RMSE of 3.8°C in the total observation period including both summer and winter. Agreement was lowest during summer 2009 and freeze-back periods which were associated with prevailing overcast conditions. A multi-year anomaly analysis revealed robust spatio-temporal patterns taking into account the found uncertainty and different atmospheric conditions. Summer periods with regional mean LST larger than 5°C showed highest spatial diversity with four distinct anomaly classes. Dry ridge areas heated up most whereas wetland areas and dry barren surfaces with high albedo were coolest. Mean inter-annual differences of LST anomalies for different land cover classes were less than 1°C. However, spatial pattern showed fewer positive anomalies in 2010 suggesting differences in surface moisture due to inter­annual differences in the amount of end-of-winter snow. Presented summer LST anomalies might serve as a baseline against which to evaluate past and future changes in land surface properties with regard to the surface energy balance. Sub-temporal heterogeneity due to snow or ice on/off as well as the effect of subpixel water bodies has to be taken into account. A multi-sensor approach combining thermal satellite measurements with high-resolution optical and radar imagery therefore promises to be an effective tool for a dynamic, process-based ecosystem monitoring scheme.