Mechanistically mapping near-surface temperature in the understory of temperate forests: A validation of the microclima R package against empirical observations

• Published in: Agricultural and forest meteorology 2024-03, Vol.346, p.109894, Article 109894
• Main Authors: Brusse, Théo, Lenoir, Jonathan, Boisset, Nicolas, Spicher, Fabien, Dubois, Frédéric, Caro, Gaël, Marrec, Ronan
• Format: Article
• Language: English
• Subjects: Biodiversity and Ecology; Climate change; Environmental Sciences; Forest microclimates; Global Changes; Mechanistic models; Temperature loggers; Vegetation cover
• ISSN: 0168-1923; 1873-2240
• DOI: 10.1016/j.agrformet.2024.109894

•Mechanistic microclima model properly predicts near-ground temperatures.•Strong correlation between predicted temperatures and in-situ logger data.•Vegetation height affects RMSE between measured and predicted temperatures.•Difference between predicted and measured temperatures increase in open ecosystems. Temperature conditions matter for ground-dwelling biodiversity. However, contrary to ambient-air temperatures as measured by weather stations, there is no global network available yet for measuring microclimate temperatures as perceived by organisms living near the ground. To predict microclimate temperatures near the ground, mechanistic models have been recently developed. Here, we aim at testing the ability of the microclima package in R to make mechanistic predictions of real temperature conditions near the ground. Focusing on a network of 45 temperature loggers measuring hourly air temperature near the ground (1-m height) inside and outside the forest of Compiègne, in northern France, we generated hourly maps of near-ground air temperature, as predicted by the microclima package, covering the exact same period: February 2018 to October 2019. Our results show a strong correlation between hourly temperatures as predicted by the model and hourly temperatures as measured by loggers (R² = 0.88). We also found that vegetation height and the Normalized Difference Vegetation Index (NDVI) influence the root mean square error (RMSE) as well as the slope coefficient between measured and predicted temperatures. For instance, increasing vegetation height reduces the RMSE and the slope coefficient between measured and predicted temperatures. Sensors placed in open habitats outside the forest or under low forest canopy height tended to measure higher temperatures than those predicted by the model. Because sensors placed outside forests are likely biased by overheating due to incoming solar radiation, the predictive accuracy of the microclima model cannot be quantified in a fair manner. Better and more in-situ data outside forests are needed. Alternatively, the microclima package could be tailored to mimic sensor overheating and better reflect the temperature as measured by sensors near the ground in open conditions—(e.g., 3D structure of the vegetation, sliding window approach).