Forests create unique microclimates that have the potential to serve as microrefugia for species under climate change. Yet, our understanding of the heterogenous thermal patterns at the forest floor of complex landscapes (e.g., in mountain forests) remains incomplete. We here used Light Detection and Ranging (LiDAR) for predicting summer temperature offsets in a mountain forest landscape in the European Alps. We calibrated models on a network of 150 microclimate loggers that were combined with data from 15 meteorological stations to estimate the maximum, mean, and minimum temperature offsets, using LiDAR-derived metrics of forest structure and topography as predictors. Models predicted summer temperature offsets with an R²/RMSE of 0.50/3.15 °C for maximum temperature, 0.51/0.41 °C for mean temperature and 0.55/0.57 °C for minimum temperature. Forest canopy openness and elevation were most important for predicting temperature offsets. The mean offset ranged from – 1.9 °C to 2.7 °C (mean of - 0.3 °C), but both minimum and maximum offsets varied considerably, with some forests even having warmer maximum and colder minimum temperatures than open areas. This was particularly prominent in forests of the subalpine zone, which are characterized by open canopies and a considerable presence of coniferous shrubs. In contrast, submontane forests with largely closed canopies had mostly colder maximum and warmer minimum temperatures within forests compared to open areas. Analysing the development of temperature offsets with time since disturbance, we found that recently disturbed forests had higher maximum temperatures compared to open areas, but they recovered to closed forest conditions within two decades. We conclude that mountain forests exhibit complex microclimate patterns that vary strongly with forest type and canopy openness. We further highlight that disturbances are an important driver of spatiotemporal dynamics in forest microclimate. Finally, temperature offset maps such as the ones generated here have strong potential to improve the robustness of species distribution models and to assess climate risks for biodiversity.