Ecological Modelling 266 (2013) 103–117 Contents lists available at ScienceDirect Ecological Modelling jo ur nal ho me page: www.elsevier.com/locate/ecolmodel Performance of tree phenology models along a bioclimatic gradient in Sweden Cecilia Olsson a,∗ , Kjell Bolmgren b , Johan Lindström c , Anna Maria Jönsson a a Department of Physical Geography and Ecosystem Science, Lund University, SE-223 62 Lund, Sweden b Swedish National Phenology Network, Swedish University of Agricultural Sciences, SE-360 30 Lammhult, Sweden c Centre for Mathematical Sciences, Division of Mathematical Statistics, Lund University, SE-221 00 Lund, Sweden a r t i c l e i n f o Article history: Received 18 January 2013 Received in revised form 20 June 2013 Accepted 22 June 2013 Available online 3 August 2013 Keywords: Tree phenology Budburst model Temperature sums Forcing units a b s t r a c t Tree phenology has been recognized as an important indicator of climate change, and a wide range of budburst models have been developed. The models differ in temperature sensitivity, and the choice of model can therefore influence the result of climate impact assessments. In this study we compared the ability of 15 models to simulate budburst of the main forest tree species in Sweden. Records on the timing of budburst, available for 1873–1918 and 1966–2011, were used for model evaluation. The predefined models, having different chilling, competence and forcing modules, represented different hypothesis on temperature impact on tree phenology. We extracted the model-specific forcing units accumulated by the observed day of budburst, and tested for covariation with bio-climatic gradients. For all tree species, most models indicated a negative relation between forcing requirement and latitude, which may indicate provenance specific adaptations. The thermal continentality index, which in Sweden is highly correlated with latitude, did provide some additional explanation for the period of 1873–1918 but not for the period of 1966–2011. For most model- and tree species combinations, temperature anomalies explain a significant part of the variability in forcing units accumulated at day of budburst. This indicates that the budburst models were not able to fully track the response to inter-annual variations in temperature conditions, probably due to difficulties in capturing species and provenance specific chilling requirement, day length response and impact of spring backlashes. © 2013 Elsevier B.V. All rights reserved. 1. Introduction Plant phenology is influenced by temperature, and has been rec- ognized as an important indicator of climate change (Menzel et al., 2006). Europe and North America have been experiencing a length- ening of the growing season since 1950s, due to earlier springs and later autumns (Menzel and Fabian, 1999; Ahas et al., 2002; Menzel and Sparks, 2006; Menzel et al., 2008). The trend is more pronounced at high latitudes than further south (White et al., 1999). The temperature response is however species-specific, and tree species growing in the same region may respond differently. The growing season may for instance become longer for Fagus sylvatica but shorter for Quercus robur in a warmer climate, due to differences in leaf senescence (Kramer, 1995). Many boreal and temperate tree species require a chilling period with cold temperatures to break winter rest, before warm forcing temperatures can trigger bud- burst (Cannell and Smith, 1986). The chilling requirement is an ∗ Corresponding author at: Sölvegatan 12, 223 62 Lund, Sweden. Tel.: +46 46 222 48 90. E-mail address: Cecilia.Olsson@nateko.lu.se (C. Olsson). adaptation to prevent early onset of growth, minimizing the risk of frost damage due to spring backlashes (Cannell and Smith, 1983). The length of the growing season influences the biochemical cycles of nitrogen, carbon and water (Ibá ˜ nez et al., 2010), thereby affect- ing the productivity of terrestrial ecosystems (Rotzer et al., 2004; Noormets et al., 2009). The performance of ecosystem models simulating plant–atmosphere interaction and ecosystem productivity are enhanced by accurate predictions of phenological events (White et al., 1999; Leinonen and Kramer, 2002; Rotzer et al., 2004; Kucharik et al., 2006; Jeong et al., 2012; Migliavacca et al., 2012; Richardson et al., 2012). Tree phenology is also important when modeling species distribution, as it can affect survival and repro- ductive success (Chuine and Beaubien, 2001). A perfect model should have high accuracy, high generality and high reality; qual- ities that are difficult to combine (Levins, 1968). A wide range of different phenology models have been presented including empir- ical models, intermediate empirical models and process-based models (Vegis, 1964; Cannell and Smith, 1983; Kobayashi and Fuchigami, 1983; Hänninen, 1990; Hunter and Lechowicz, 1992; Chuine, 2000; Schaber and Badeck, 2003; Caffarra et al., 2011). The model structure is of main importance for its temperature 0304-3800/$ – see front matter © 2013 Elsevier B.V. All rights reserved. http://dx.doi.org/10.1016/j.ecolmodel.2013.06.026