Predicting local colonization and extinction dynamics from coarser-scale surveys Jennifer Moody-Weis, Janis Antonovics, Helen M. Alexander and Diana Pilson J. Moody-Weis (moody-weisj@william.jewell.edu) and H. M. Alexander, Dept of Ecology and Evolutionary Biology, Univ. of Kansas, 1200 Sunnyside Ave, Lawrence, KS 66045, USA. (Present address of J. M.-W.: 500 College Hill, Campus Box 1059, William Jewell College, Liberty, MO 64068, USA.)  J. Antonovics, Dept of Biology, Gilmer Hall, Univ. of Virginia, Charlottesville, VA 22905, USA.  D. Pilson, School of Biological Sciences, Univ. of Nebraska, Manter Hall, Lincoln, NE 68588-0118, USA. The demand for methods to translate information between spatial scales (i.e. size of observational units and the total area of study) has intensified given increased recognition that empirical data collection and practical applications occur at scales ranging from individual organisms to landscapes. For example, there has been considerable interest in ‘‘scaling- down’’ methods that have been successful at predicting fine-scale species’ distributions from coarse-scale distributional maps. Here, we describe the application of scaling-down methods to the estimation of colonization and extinction rates in metapopulations using long-term, large-scale data sets of two roadside plant species, Helianthus annuus and Silene latifolia. Fine-scale data collected from roadside populations were aggregated to generate data at several increasingly coarse scales. The relationships between occupancy, colonization, or extinction and the scale of measurement (scale- curves) were determined using the standard ‘‘fully-nested’’ method and the ‘‘stratified random sampling’’ method. Both methods were successful at predicting not only occupancy, but also the dynamic metapopulation processes of extinction and colonization (R 2 values, averaged across species and methods, were 88.5, 69.3, and 88.8%, respectively, for occupancy, extinction, and colonization). Scaling-down generated more accurate predictions in Helianthus (average R 2  88.4) compared to Silene (average R 2  63.4), and in both species, scaling-down generated more accurate predictions for occupancy and colonizations compared to extinctions. This is one of the first demonstrations that dynamic processes are scalable, and provides a useful methodology for dealing with the logistical challenges of collecting fine-scale data over large geographic areas when studying metapopulation processes or range limits. While some ecological phenomena are scale invariant (Steele and Forrester 2005), many are dependent upon the spatial scale (i.e. extent and resolution) of observations and analysis (Levin 1992, Englund et al. 2001). Scale dependence of ecological phenomena is problematic be- cause, out of necessity, ecological studies are generally conducted at single spatial scales, but predictions and management decisions may occur at other spatial scales (May 1989, Wiens 1989, Schulze 2000). Scaling can occur in two directions. Scaling-up methods use data collected at fine scales to infer coarse-scale patterns and processes, while scaling-down methods use data collected at coarse scales to infer patterns and processes at finer scales. Although scaling- up can be challenging (Schulze 2000, Englund and Hamba ¨ck 2004), such methods are commonly applied to a wide range of ecological questions (Rastetter et al. 1992, Inouye 2005, Urban 2005). For example, photosynthetic rates of individual leaves can be incorporated into models that predict the CO 2 uptake of an entire forest (Rastetter et al. 1992). Scaling-down had generally been thought impossible (Hartley et al. 2004). However, recent studies suggest that fine-scale species’ abundances can be predicted from coarse-scale occupancy data gathered from distribu- tional maps (Kunin 1998, He and Gaston 2000, Kunin et al. 2000, Kallimanis et al. 2002, Cousens et al. 2004, Halley et al. 2004, Hartley et al. 2004, Tosh et al. 2004), although some approaches remain controversial (He and Gaston 2007, Conlisk et al. 2007). Scaling-down methodologies could be particularly useful to conservation efforts. For example, in many cases fine-scale data on rare or invasive species’ abundance are needed, but such data are difficult to collect and knowledge of most species is limited to coarse- scale distributional maps (Kunin 1998, Tosh et al. 2004). The need for scaling-down methods is expected to increase as technological advances in GPS, GIS, and remote sensing increase the availability of large-scale ecological data, most often collected at coarser scales (Withers and Meentemeyer 1999). Most scaling-down methodologies predict species dis- tributions using the box-counting method, in which a species distribution is examined using different-sized quad- rants (‘‘boxes’’) and scale-area curves are generated through Ecography 31: 6172, 2008 doi: 10.1111/j.2008.0906-7590.05282.x # 2008 The Authors. Journal compilation # 2008 Ecography Subject Editor: Thorsten Wiegand. Accepted 15 January 2008 61