HYDROLOGICAL PROCESSES Hydrol. Process. 25, 3760–3773 (2011) Published online 16 April 2011 in Wiley Online Library (wileyonlinelibrary.com) DOI: 10.1002/hyp.8101 Pattern-process relationships in surface hydrology: hydrological connectivity expressed in landscape metrics Bartel H. J. Van Nieuwenhuyse, 1 * Michael Antoine, 2 Guido Wyseure 1 and Gerard Govers 1 1 Katholieke Universiteit Leuven, Department of Earth and Environmental Sciences, Celestijnenlaan 200E, B-3001 Leuven, Belgium 2 Universit´ e catholique de Louvain, Earth and Life Institute, Croix du Sud 2 bte 2, B-1348 Louvain-la-Neuve, Belgium Abstract: The degree of hydrological connectivity is mainly determined by the spatial organisation of heterogeneity. A meaningful and aggregate abstraction of spatial patterns is one of the promising means to gain fundamental insights into this complex interaction and can, moreover, be used as a tool to acquire a profound understanding of the major controls of catchment hydrology. In order to disclose such controls, pattern-process relationships and the explanatory power of landscape metrics were tested by simulating the runoff of differently patterned virtual basins, generated by neutral landscape models and fractal networks and solved by a surface hydrological model composed of kinematic wave routing and Green-Ampt infiltration. A total of 23 landscape metrics quantified the spatial patterns and were subsequently related to the functional connectivity, assessed as the proportion of internal runoff generation constituting the hydrological response at the outlet. Landscape metrics allowed the identification of dominant features of heterogeneity that explained the observed connectivity, and to disclose changes in control with class abundance. Therefore, landscape metrics are a useful tool for basin comparison and classification in terms of the dominant processes and the corresponding model structure requirements. Copyright  2011 John Wiley & Sons, Ltd. KEY WORDS hydrological connectivity; landscape metric; pattern-process relationship; neutral landscape model; virtual experiment Received 26 June 2010; Accepted 8 March 2011 INTRODUCTION “Hydrological connectivity describes the internal ‘link- ages’ (Warner, 2005) between runoff generation in upper parts of the catchment and the receiving water” (Croke et al., 2005) and depends on the degree to which the rain- fall duration and intensity allow the transfer of runoff over the hillslope and into channels, to then overcome transmission losses and finally generate some catchment runoff (Bracken and Croke, 2007). A major challenge accompanying hydrological connectivity, and the pro- cesses occurring at catchment scale in general, lies in dealing with the spatial heterogeneity of rainfall and land characteristics (Beven, 2002; Sivapalan, 2005; Hopp and McDonnell, 2009). McDonnell et al. (2007) and Siva- palan et al. (2003a), among others, argue that the popular way of characterising landscape heterogeneity in ever greater detail for highly calibrated distributed models, the so called reductionist approach, will not move the hydro- logical science beyond its current status quo. Although this approach is often successful in predicting outflows, reductionist models do not always do so for the right reason (Kirchner, 2006) and have thus failed to fulfil their most important promise: extrapolation beyond the *Correspondence to: Bartel H. J. Van Nieuwenhuyse, Katholieke Uni- versiteit Leuven, Department of Earth and Environmental Sciences, Celestijnenlaan 200E, B-3001 Leuven, Belgium. E-mail: Bartel.Vannieuwenhuyse@ees.kuleuven.be calibration range (Beven, 2001). Further progress will critically depend on building models that do better in representing landscape heterogeneity: we need to get the reasons right (supposing that the predictions will then eventually follow), instead of supplying yet more “grist for the calibration mill” (Sivapalan, 2005). Although small-scale processes are fairly well under- stood, upscaling to the hillslope or catchment scale is not straightforward, because processes emerge that cannot be handled with spatial averaging (Harman and Siva- palan, 2007) or re-parameterisation of conservative model structures (Beven, 2001). Depending on the stand one takes towards hydrological modelling (Pappenberger and Beven, 2005), one can adhere to the reductionist prin- ciple and persist in its physical validity even though this is untenable in the face of limited computational resources (Sivapalan, 2005) and a shortage of appro- priate, highly detailed field information (Summer, 1998; Wooldridge and Kalma, 2001). Alternatively, one can be more pragmatic and tune overparameterized models until they are behavioural (Beven, 1993), but parame- ter calibration suffers from equifinality (Beven, 2006) while model constants often do not remain constant under changing boundary conditions (Kirchner, 2006). Owing to these reasons, distributed modelling no longer promises to provide many new scientific insights con- cerning the functioning of the system that is studied. A possible solution might be the development of so-called Copyright  2011 John Wiley & Sons, Ltd.