Objectives <p>Winter mortality of honey bee colonies remains high and reduces early-spring pollination. Fondant feeding is a common emergency intervention, yet quantitative descriptions of colony and microclimate responses—especially homogeneous time-series—are scarce. Milder winters and unstable springs further elevate starvation risk and complicate overwintering. We present a homogeneous time-series dataset that captures the thermal response of overwintering <i>Apis mellifera</i> colonies to fondant feeding under these conditions. The dataset fills this gap and supports reproducible protocols for winter and early-spring fondant feeding.</p> Data description <p>The dataset covers Ukraine (mild winter, unstable spring; one colony) and Canada (record-warm winter; three colonies; four feeding events). Data were collected non-invasively from identical Internet of Things (IoT) hives equipped with internal/external temperature (°C) and relative humidity (%) sensors, a load-cell scale, and a solar-powered acquisition and transmission unit. Measurements were recorded at fixed intervals to ensure homogeneous time series. The repository provides raw and cleaned data, a variable dictionary, cleaning/imputation rules, an event log, and a validation script, enabling quantitative assessment for forecasting colony development and for designing scheduled fondant feeding to help reduce winter losses and prepare colonies for early-spring pollination.</p>