Enhanced flight planning and calibration for UAV based thermal imaging: Implications for canopy temperature and transpiration analysis

Malkin Gerchow ^1* Kathrin Kühnhammer ^1,2 Alberto Iraheta ¹ John D. Marshall ³ Matthias Beyer ¹

• ¹ Technical University of Braunschweig, Braunschweig, Germany
• ² University of Freiburg, Freiburg, Baden-Wurttemberg, Germany
• ³ University of Gothenburg, Gothenburg, Västergötland, Sweden

Leaf and canopy temperature have long been recognized as important indicators of plant water status because leaves cool when water is transpired and warm up when leaf stomata close and transpiration is reduced. Unmanned aerial vehicles (UAVs) open up the possibility to capture high resolution thermal images of forest canopies at the leaf scale. However, a careful calibration procedure is required to convert the thermal images to absolute temperatures, in addition, at high spatial resolution, the complexity of forest canopies leads to challenges in stitching overlapping thermal images into an orthomosaic of the forest site. In this study, we present a novel flight planning approach in which the locations of ground temperature references are directly integrated in the flight plan. Six UAV flight campaigns were conducted over a tropical dry forest in Costa Rica. For each flight five different calibration methods were tested. The most accurate calibration was used to analyze the tree canopy temperature distributions of five tree species. From the distribution we correlated its mean, variance, 5th and 95th percentile against individual tree transpiration estimates derived from sapflow measurements. Our results show that the commonly applied calibration provided by the cameras manufacturer (factory calibration) and empirical line calibration were less accurate than the novel repeated empirical line calibration and the factory calibration including drift correction (MAE 3.5 °C vs. MAE 1.5 °C). We show that the orthomosaic is computable by directly estimating the thermal image orientation from the visible images during the structure from motion step. We found the 5th percentile of the canopy temperature distribution, corresponding to the shaded leaves within the canopy, to be a better predictor of tree transpiration than the mean canopy temperature (R² 0.85 vs. R² 0.60). Although these shaded leaves are not representative of the whole canopy, they may be the main transpiration site in the heat of the day. Spatially high-resolution, validated temperature data of forest canopies at the leaf scale have many applications for ecohydrological questions, e.g., the estimation of transpiration, for comparing plant traits and modeling of carbon and water fluxes by considering the entire canopy temperature distribution.