Evaluating choices in multi-process landscape evolution models

Abstract

The interest in landscape evolution models (LEMs) that simulate multiple landscape processes is growing. However, modelling multiple processes constitutes a new starting point for which some aspects of the set up of LEMs must be re-evaluated. The objective of this paper is to demonstrate the practical significance of, and possibilities for such re-evaluation. We first discuss which simplifications must be made to set up LEMs. Then, simplifications of particular interest to the modelling of multiple processes are identified. Finally, case studies from New Zealand, Belgium and Croatia explore the performance of different model versions under several common choices for model and data simplifications. In these case studies we illustrate methods to make the choices, typically by i) comparing the results of different model versions, ii) assessing model validity or iii) indicating the sensitivity of models for different simplifications. The results indicate that LEM performance is strongly dependent on multi-process related choices, and that performance indicators can be used for ex-post testing of the influence of these choices. In particular, we demonstrate the significance of simplifications regarding the number of processes, presence of sinks and temporal resolution.

Introduction

Many processes result in landscape change. How these landscape processes act and interact is a topic of continuing interest for geomorphology. Landscape evolution models (LEMs) can play an important role in pursuing this interest, given their potential role in hypothesis generation about landscape evolution and interactions between processes in space and time (e.g. Coulthard, 2001). Both the development of LEMs themselves, as the interpretation of simulation results, may lead to improved understanding of landscape evolution.

Within current literature, no clear definition of LEMs is provided (Pazzaglia, 2003). Arguably, such a definition should focus on objectives, rather than methods. A recent review on models focussing on fluvial and hillslope processes is given by Tucker and Hancock (2010). In addition, Brasington and Richards (2007) suggest that LEMs characteristically focus on landscape dynamics at spatial extents of 10⁰–10² km² and temporal extents of 10¹–10³ years. This focus is an expression of both (diminishing) computational and (more permanent) data constraints as much as that of the objectives. Therefore, adaptation of these extents is warranted. Here, we choose to discuss LEMs that focus on landscape dynamics at spatial extents larger than 10⁰ km² and at temporal extents between 10¹ and 10⁴ years. Consequently, we operate at the lower end of geological timescales, where we safely may reduce the influence of climate, base level change, tectonics, and weathering.

LEM studies that model a single landscape process outnumber those that model multiple processes. Recent work has been done on for example water erosion and deposition (Takken et al., 1999, Schoorl and Veldkamp, 2001, Collins et al., 2004), tillage erosion (Heuvelink et al., 2006), landsliding (Claessens et al., 2005, Claessens et al., 2007), weathering (Heimsath et al., 1997, Minasny and McBratney, 2001), soil creep (Minasny and McBratney, 1999, Minasny and McBratney, 2001, Minasny and McBratney, 2006), fluvial action (Coulthard et al., 1998, Coulthard et al., 2000, Coulthard et al., 2002) and dune formation (Baas, 2007, Baas and Nield, 2007). To a large degree these studies aimed at development and validation of process descriptions. For other landscape processes, empirical identification of controls is ongoing and landscape-scale process descriptions are still premature. Examples are wind erosion (Okin et al., 2006), gelifluction (Harris et al., 2003), solifluction (Matsuoka et al., 2005) and frost weathering (Williams and Robinson, 2001). At the same time, the interest in studying multiple processes and their interactions in the landscape is growing, particularly in combination with the increased focus on reduced complexity modelling in fluvial geomorphology (Brasington and Richards, 2007), which stresses the simplification of process knowledge required for landscape-scale LEMs.

Simplification or reducing complexity is part of every modelling exercise, with single-process LEMs included. Recent LEM studies have used reduced complexity as a way to focus on multiple processes and their interactions. The combination of water and tillage erosion received early attention in agricultural landscapes in Belgium (Govers et al., 1996, Peeters et al., 2006) and later in Spain (Schoorl et al., 2004). Follain et al. (2006) simulated the combined effect of soil creep and water erosion in an agricultural landscape in France; Coulthard and Van de Wiel, 2006, Van de Wiel et al., 2007 combined several processes in the fluvial domain in Wales; and Temme and Veldkamp (2009) combined water erosion, biological and frost weathering, creep and solifluction in a study of a valley in South Africa. Coulthard and Baas (2008) combined aeolian dune activity and fluvial activity using an example from Mongolia.

Attention for the interactions between geomorphic landscape processes and land use and cover change processes is also increasing (Veldkamp et al., 2001). Claessens et al. (2009) modelled interactions and feedback mechanisms between landscape processes water erosion and deposition and tillage on the one hand and land use change on the other hand.

Frameworks for LEMs increasingly include multiple processes in a modular setup. Frameworks that combine processes include CHILD (Tucker et al., 2001), CAESAR (Coulthard et al., 1998), LAPSUS (Schoorl et al., 2000, Schoorl et al., 2002), SIBERIA (Willgoose et al., 1990), CASCADE (Braun and Sambridge, 1997) and WATEM (Van Oost et al., 2000).

The modelling of multiple processes in a dynamic landscape constitutes a new starting point for which common model and data simplifications have to be re-evaluated. The objective of this paper is to perform such re-evaluation. First, we discuss the simplifications that are necessary to set up case-specific optimal LEMs and identify simplifications of particular importance for multi-process LEMs. Then, we use case studies to illustrate how some of these simplifications affect multi-process LEM performance. Simplifications regarding the number of processes, the presence of sinks and temporal resolution are considered. The case studies are intended to help choose appropriate simplifications in other studies.

The focus of this paper is on the effects of model and data simplifications in multi-process model setup, not on the descriptions of individual processes or their validity. Therefore, the discussion of the process descriptions is kept brief and no topical conclusions are drawn in the presented case studies. Furthermore, the focus is on LEMs as defined previously, particularly those that use digital elevation models (DEMs) as their digital landscape, although the results may be applicable wider afield.