1 Introduction

Ecosystems are continuously changing in a way that is difficult to understand and to predict. Studies of climate changes (IPCC, 2007), land-use changes, biodiversity erosion (MEA, 2005), and energy management (Dalgaard et al., 2006) would strongly benefit from models aiming at better handling landscape dynamics observed on continents (and in oceans). By landscape dynamics, we imply all kinds of land-cover changes occurring at mesoscale (tens of hectares or km²) on Earth. By models, we imply numerical (algorithm-based) and analytical (equation-based) approaches able to handle landscape elements and their dynamics. More precisely, a landscape is usually referred to as an object assembling elements of various natures (fields, forests, buildings, hedges, roads…) in interaction, simultaneously changing through many spatial and temporal scales (Forman and Godron, 1986*; Turner and Gardner, 1991*).

Scientists model landscapes for at least two reasons: to better understand landscape dynamics themselves (hereafter called intrinsic needs), and to offer realistic frames to host other ecological, biological, sociological and/or physical processes (extrinsic needs) (Rounsevell et al., 2012). The role and status of models and modelling is itself an instantiation of a wider debate concerning representation and explanation (Clifford, 2007). For example, we may model urbanization and agriculture because they are two of the most important drivers of rapid changes in biodiversity worldwide (Benton et al., 2002; Cincotta et al., 2000). One of the advantages of studying landscapes as dynamic land covers is that models of this nature are not bound to some specific processes such as sectors of the economy or of biogeochemical behaviours. It helps to characterize the processes involved. Conversely, it is dangerous to infer landscape functioning on the sole basis of its observed structure (Schröder and Seppelt, 2006*; Shochat et al., 2006), as the pattern-process debate reminds it to us. This paper concerns intrinsic needs mainly, and reviews terrestrial landscape models (LM) from socio-ecological as well as methodological perspectives (Collins et al., 2011).

Comprehensive studies state that agricultural landscapes today cover approximately 39 – 50% of continental areas (and are increasing), that forested landscapes still cover 30% of them and are rapidly disappearing, that urban areas (3 – 5%) and arid or semi-arid (15 – 25%) areas are continuously growing, too (Paudel and Yuan, 2012*; Vitousek et al., 1997). Numerous LM reviews may be found in the literature (Baker, 1989; Berling-Wolff and Wu, 2004*; Scheller and Mladenoff, 2007*; Verburg et al., 2004*), but each of them is focusing on one of the four main landscape types mentioned above. Yet, these landscape types are in deep interaction on continents (Lambin, 1997*; Verburg and Veldkamp, 2004*). Furthermore, it is to be expected that models used for one landscape type may be useful (or at least may feed concepts) for any of the other types. Can we identify common processes behind such landscape similarities and interactions?

In this discussion paper, we address the question whether it is useful and feasible to build a comprehensive theory of landscape dynamics; and in case of a positive answer, which concepts are today relevant for this program. Hence, our objective is to show the probable unity behind the landscape diversity and to advocate the urgent need of a comprehensive theory to handle this unity. We aim at offering a critical state-of-the-art by the use of a double-entry analysis grid (matrix), focusing on the one hand on the four main terrestrial landscape types and, on the other hand, on the most relevant LM characteristics we identified. About the latter, we discuss in particular explicit or neutral models (Gardner et al., 1987*; With and King, 1997*), patchy or grid-based (Costanza and Voinov, 2004*; Kotliar and Wiens, 1990*), multi- or mono-scale models (Pascual and Guichard, 2005*; Thomas et al., 2008*) and landscapes with or without linear networks (Proulx et al., 2005*; Pumain, 2006*). We outline each case of this analysis grid by a list of associated processes and one or two examples described in more details, often taken from anthropogenic (i.e., man-made) landscapes. We then explore on the basis of this analytical grid which model characteristics would be useful for which landscape type. We did not intend to exhaustively compare papers dedicated to landscape modelling. We rather focused on theoretical and applied literature contributing to this first proposal of a unified landscape dynamics theory.

We finally discuss a set of proposals on the basis of the previous landscape types and model characteristics that are presumably linked, and discuss the need for a comprehensive theory to interpret landscape dynamics (Lambin, 1997*), as made by some authors for modelling scales and entities with LM (Agarwal et al., 2002*; Gaucherel and Houet, 2009; Haase and Schwarz, 2009*; Verburg et al., 2006). We more specifically insist on anthropogenic landscapes, for the main reason that these mosaics usually superimpose in a complex way human decisions concerning already present natural processes. Such a theory should be in the continuation of landscape ecology concepts (Forman and Godron, 1986*), but based on original mathematical formalisms providing new insights for the land change science (Turner et al., 2007). For example, it could be inspired from complexity approaches (Bolliger et al., 2005*; Crawford et al., 2005; Solé and Bascompte, 2006*). We discuss the implication of our suggestion for future research, which is needed to understand, predict and project the dynamics of these landscapes in relation to local and global environmental changes.