Mapping microclimate temperatures in open ecosystems using UAVs: a comparison between thermal, correlative, and mechanistic approaches

Christophe Metsu, Sander Raets, Marijke Thoonen, Stef Haesen , Reinout Vandenabeele, Karlien Moeys, Sam Ottoy, Koenraad Van Meerbeek 

Open terrestrial ecosystems exhibit pronounced fine scale thermal heterogeneity, yet spatially continuous microclimate data at biologically relevant heights and scales remain scarce. Here, we evaluate three unoccupied aerial vehicle (UAV) informed approaches for mapping microclimate temperatures, including land surface temperature (LST) and near surface air temperature (T13cm), in an open heathland ecosystem. The approaches comprise a correlative gradient boosted model (GBM), the mechanistic microclimate model microclimf, and UAV based thermal remote sensing, with LST physically converted to near surface air temperature. The comparison was conducted across four summer UAV campaigns in 2024 under clear sky and overcast conditions. All spatial predictor variables, including topography, vegetation structure, land cover, and NDVI, were derived from UAV borne thermal, multispectral, and LiDAR sensors. The GBM showed the closest agreement with in situ TOMST TMS 4 logger measurements for near surface temperature (RMSE = 2.22 °C), followed by the UAV based thermal conversion (3.37 °C) and microclimf (4.43 °C). Differences among approaches were systematic: microclimf generally underestimated near surface temperatures, particularly beneath solitary trees, whereas the UAV based thermal approach tended to predict higher temperatures, reflecting sensitivity to vegetation parameterization, spatial alignment, and potential radiative warming of shielded loggers. UAV derived thermal observations captured more extreme LST values and sharper spatial contrasts than either correlative or mechanistic models, revealing fine scale thermal mosaics characteristic of these open ecosystems. Overall, our results demonstrate that UAV based thermal remote sensing, particularly when integrated with multispectral and LiDAR derived structural information and physically based temperature conversions, provides complementary value to established microclimate modelling approaches by resolving thermal extremes and spatial variability that are otherwise smoothed or overlooked.