Revealing topoclimatic heterogeneity using meteorological station data

Climate is a crucial driver of the distributions and activity of multiple biotic and abiotic processes, and thus high-quality and high–resolution climate data are often prerequisite in various environmental research. However, contemporary gridded climate products suffer critical problems mainly related to sub-optimal pixel size and lack of local topography-driven temperature heterogeneity. Here, by integrating meteorological station data, high-quality terrain information and multivariate modelling, we aim to explicitly demonstrate this deficiency. Monthly average temperatures (1981–2010) from Finland, Sweden and Norway were modelled using generalized additive modelling under (1) a conventional (i.e. considering geographical location, elevation and water cover) and (2) a topoclimatic framework (i.e. also accounting for solar radiation and cold-air pooling). The performance of the topoclimatic model was significantly higher than the conventional approach for most months, with bootstrapped mean R2 for the topoclimatic model varying from 0.88 (January) to 0.95 (October). The estimated effect of solar radiation was evident during summer, while cold air pooling was identified to improve local temperature estimates in winter. The topoclimatic modelling exposed a substantial temperature heterogeneity within coarser landscape units (>5 °C/1 km−2 in summer) thus unveiling a wide range of potential microclimatic conditions neglected by the conventional approach. Moreover, the topoclimatic model predictions revealed a pronounced asymmetry in average temperature conditions, causing isotherms during summer to differ several hundreds of metres in altitude between the equator and pole facing slopes. In contrast, cold-air pooling in sheltered landscapes lowered the winter temperatures ca. 1.1 °C/100 m towards the local minimum altitude. Noteworthy, the analysis implies that conventional models produce biassed predictions of long-term average temperature conditions, with errors likely to be high at sites associated with complex topography.