The evolution and impacts of ‘complexity notions’ in landscape architecture

Abstract

Complexity notions, i.e. ideas or methods that incorporate concepts and rationales from complexity science as analogies or models, frequently appear in landscape architectural discourses. However, debates have arisen about the legitimacy and relevance of complexity notions in landscape architecture. Are complexity notions an ephemeral fashion or derived from the inherent needs of landscape architecture research and practice? What role do complexity notions play in the development of landscape architecture? To answer these questions, we conducted a three-phase review of the complexity notions in landscape architectural theories and practices since early 20th century. We concluded that complexity notions in landscape architecture are a long-standing and increasingly significant subject rather than a passing fad. Complexity notions serve as an exploratory system rather than tyrannical dogma. Addressing the increasing complexity of landscapes and inspired by up-to-date complexity theories, incorporating adaptive learning processes is becoming a new paradigm in landscape research and practice.

Keywords:

• Landscape architecture
• design theory
• landscape criticism
• landscape history
• landscape practice
• complexity science
• complex adaptive system
• resilience
• design process
• feedback learning

Introduction

In Human Ecology: Following Nature’s Lead, Frederick Steiner (2002) pointed out that ‘landscapes are dynamic entities defined by their interacting parts and their integrative whole’ (p. 86). This definition highlighted the ‘complexity’ of landscapes. In today’s science, complexity describes a property of a system with many interacting parts, of which the whole is greater than the sum of its parts (Chaffee & Mcneill, 2007). Over the past few decades, concepts like chaos, self-organisation, and complex adaptive systems known in the field of ‘complexity science’ or ‘complexity theory’ have frequently appeared in landscape architectural discourses. In this article, ‘complexity notions’ refers to the notions in landscape architectural theories or practices that incorporate concepts and rationales from complexity science as analogies or models. Complexity notions see landscapes as holistic systems that emerge from the interactions of numerous autonomous natural and cultural agents, and favour approaches that treat landscapes as nonlinear and unpredictable dynamic systems, as opposed to the mechanistic, reductionist, and determinist views and methodologies in classical science.

However, there have been debates about the legitimacy and relevance of complexity notions in landscape architecture. On the one hand, many academics and practitioners believe that complexity notions reflect or promote landscape architecture’s progress. For example, Koh (1982) articulated that the emergence of ecological design in landscape architecture signified a major paradigm shift from reductionistic to holistic and evolutionary worldviews and methods. Lystra (2014) noticed that cybernetics, the study of complex system control through feedback, has influenced landscape architecture since the 1960s to pay more attention to landscape changes and design processes. Today, ecological design has moved away from a mechanistic model and towards complex adaptive systems thinking (Lister, 2015; Liu & Zhang, 2018; Reed & Lister, 2014). The pursuit of ‘resilience’, ‘adaptation’, and ‘plasticity’ in today’s design and design processes demonstrates a conjunction of landscape architecture and complexity science (Mugerauer & Liao, 2012).

On the other hand, the use of complexity analogies in landscape architecture has also been criticised. Riley and Brown (1995) regarded chaos or complexity science analogies in landscape architectural discussions as ‘intellectual hipness’. Anne Whiston Spirn criticised the tendency of landscape architects to draw broadly from scientific disciplines as authorities, especially ‘without examining and reconciling the beliefs and traditions on which they are based’ (Spirn, 1997). Despite criticisms, complexity notions have appeared repeatedly in landscape discourses in the past two decades. Meanwhile, contemporary discourses have warned about the limitations of metaphorical understandings of complex systems in the design and planning realm (Lister, 2018), which may obscure important features of the scientific models of complex systems (Pickett, Cadenasso, & McGrath, 2013).

Given the above debates, this article aims to address two questions: (1) Are complexity notions an ephemeral fashion or derived from the inherent needs of landscape architecture research and practice? (2) What role do complexity notions play in the development of landscape architecture? By answering these questions, we argue that complexity notions have been a long-standing and increasingly significant subject in this field, while they need to be founded on a scientific understanding of complex systems, as well as understood in the context of landscape architecture itself, to address the challenges in contemporary landscape research and practice.

To answer these questions, we conducted a three-phase critical review of complexity notions in landscape architecture since the early 20th century. The materials reviewed are primarily the theoretical discussions and practices in landscape architecture’s own context that consciously incorporated complexity notions, with some scientific discourses related to complexity science briefly mentioned. Instead of being exhaustive, the material selection is intended to be illustrative of the evolution of complexity notions in landscape architecture. Although some amount of historical literature is involved, an emphasis is put on the phase from the 1990s to the present to provide implications for current and future research and practice. Although both physical/ecological and cultural/human aspects of the landscape are important, this article primarily focuses on the physical/ecological aspects.

The emergence of complexity notions (early 20th century)

The scientific understanding of complexity originated in the early 20th century when scientists noticed ‘spontaneous order’ or ‘emergence’ in biological or social evolution. According to Wheeler (1928, p. 14), ‘emergence’ is ‘a novelty of behaviour arising from the specific interaction or organisation of a number of elements…thereby constitute a whole, as distinguished from their mere sum’.

In the planning and design field, the earliest notion of complexity in its scientific sense can be found in the concepts of ‘Regional Planning’ developed and practised by Patrick Geddes and his followers in the early 20th century. Known as a biologist and town planner, Geddes was also the first European who claimed to be a ‘landscape architect’ (Turner, 2005, p. 378). In an article published in Nature in 1882, Geddes presented an explanation of evolution based on interspecies cooperation, which resembled what scientists call emergence or self-organisation today, a central rationale of complexity science (Batty & Marshall, 2017). Later, Geddes (1915, p. 198) applied this cooperative evolutionary perspective to the formation of regional landscapes by human–nature interaction, where ‘Place, work, and folk—environment, function, and organism—are thus no longer viewed apart, but as the elements of a single process—that of healthy life for the community and the individual’. Lewis Mumford, Geddes’s disciple, described this process using the term ‘emergent evolution’. He explicitly stated that the term ‘emergent’ was used in the definite sense that Wheeler used it (Mumford, 1989, p. 29).

In practice, Geddes (1915) believed that 19th-century science was ‘too static and analytic’ to be applied in the social and civic fields (pp. 398–399), for ‘town planning is not something which can be done from above, on general principles easily laid down’ (p. 205). Instead, he promoted a new science he called ‘Civics’ that embraced concrete survey and action—‘it is from facts surveyed and interpreted that we gain our general ideas of the direction of Evolution, and even see how to further this’ (Geddes, 1915, p. vi). Based on this viewpoint, Geddes advocated for exhibitions to help the public understand the various relationships in the landscape and transform it through local actions. This was best exemplified in his Edinburgh project, where surveys were conducted at the regional scale, while incremental improvements were initiated in many communities in its old town (Figure 1).

Figure 1. Patrick Geddes’s regional survey and incremental improvements in Edinburgh (Patrick Geddes Collections at the University of Edinburgh, Coll-1167).

Although regional-scale human–nature harmony had already been an important concern of earlier scholars and practitioners, they usually held a static and dualistic view and preferred master planning in a strongly top-down manner. For example, American environmentalist John Wesley Powell (1879), in the Report on the Lands of the Arid Region of the United States, emphasised conservation over agricultural development and suggested ‘the redemption of these lands will require extensive and comprehensive plans, for … individual farmers, being poor men, cannot undertake the task’ (p. viii). Frederick Law Olmsted saw natural landscapes as a remedy for the health and morality problems caused by city living. In his Park System planning for many American cities, it was a persistent desire of Olmsted’s to create ‘an opposite class of conditions [between parks and the city]’ and ‘an antithesis of objects of vision [of the parks] to those of the streets and houses’ (Olmsted & Hubbard, 1928, p. 249). In contrast, Regional Planning dissolved the spatial discrimination between human and nature and considered landscapes as evolutionary wholes underpinned by many inseparable ‘Place–Work–Folk’ units (Figure 2). Hence, planning is an extension of landscapes’ evolution.

Figure 2. Frederick L. Olmsted’s master plan of the Park System in Boston (top, retrieved from National Park Services Olmsted Archives, public domain) demonstrated a contrast between nature and city, while Patrick Geddes’ Valley Section (bottom, reproduced based on Hall, 2014, p. 157) showed the Place–Work–Folk units as a whole in a dynamic landscape.

In sum, Regional Planning embodied preliminary complexity notions. Although Geddes had a biology background, he did not simply apply evolutionary or emergence theories as an outside authority. Rather, the complexity notions in Regional Planning originated from the study of human activities, as well as the practice that dealt with urban expansion, in landscape regions. However, until the mid-20th century, such complexity notions were not well understood and accepted, while landscape practice was still dominated by reductionist, top-down master planning.

Complexity notions as a ‘belief system’ (mid-20th century to 1990s)

In the mid-20th century, a series of scientific breakthroughs (such as chaos theory, self-organisation, and fractal geometry) revealed the structure and dynamics of complex systems, resulting in the rise of complexity science. Among schools of thought and terminological preferences in complexity science, the theory of the complex adaptive system (CAS) developed by the Santa Fe Institute has been the most widely accepted (Figure 3). A CAS emerges from the self-organisation of a number of interacting agents and assumes network structures with nested hierarchies and unpredictable, nonlinear dynamics (Chaffee & Mcneill, 2007).

Figure 3. A complex adaptive system has a nested hierarchical network structure, which emerges from the bottom-up self-organisation of many interacting agents, endowed with indeterminate dynamics and adaptability. Diagram drawn by author.

Multi-scale connections in ecological planning

The emergence of complexity science led to paradigm shifts in many scientific disciplines, with ecology possibly having the most significant linkages to landscape architecture. Established in the early 20th century, the classical paradigm (known as the ‘equilibrium paradigm’) in ecology assumes that ecosystems are closed, self-regulating systems with predictable dynamics and a single equilibrium state. However, the new paradigm, or ‘non-equilibrium paradigm’, established in the 1970s considers ecological systems as open systems with probabilistic dynamics and multiple or no stable state(s) (Wu & Loucks, 1995).

Ian L. McHarg was the most noted pioneer who introduced ecology to landscape architecture. Due to the limitations of his time, McHarg’s theory based on ‘fitness’ assumed an equilibrium state, while the famous map-overlay method attempted to identify the fittest locations for different land uses in order to pursue and maintain the equilibrium (Herrington, 2010). Furthering McHarg’s ideas, Steiner, Young, and Zube (1988) pointed out that individual fitness must be viewed in the larger context of the fitness of the whole and that fitness should be dynamic and sustainable rather than static. Steiner also repelled the pursuit of standardisation and rationalism in ecological planning. In The Living Landscape, Steiner (2008) wrote that ‘[c]haos theory emphasises the complexity of our planet and its natural and social systems’, and therefore ecological planning should be ‘based on the need for pragmatism in a complex world’ (p. 312).

Landscape ecology best exemplified the non-equilibrium paradigm at this time. Landscape ecology was described by its presenter Carl Troll (1971) as differing from ‘ecosystem ecology’ in that it expands from within ecosystems to the ‘complex causal relationships’ between ecosystems, which are expressed regionally in certain patterns at various orders of magnitude. The structures and mechanisms of CAS, such as complexity network, hierarchical structure, and self-organisation, became premises and models in landscape ecology (e.g. Naveh & Lieberman, 2013; Wu & Loucks, 1995) (Figure 4). Landscape ecology has also been integrated into ecological planning, which complemented earlier ecological planning with a chorological approach to facilitate horizontal connectivity at multiple nested scales (Ahern, 1999). The connectivity and its resultant configurations were often understood through patterns, e.g. the well-known ‘patch-corridor-matrix’ model (Forman & Godron, 1986), which provided landscape planners with a viable method to foster ecological complexity (North & Waldheim, 2013).

Figure 4. Zev Naveh and Arthur S. Lieberman’s illustration of landscape ecological systems with a hierarchical interactive structure of complex adaptive systems. Adapted based on Naveh and Lieberman (2013, p. 75).

Another advancement in ecological planning was the emphasis on across-scale coordination. Since the early 1990s, the European Union has launched a series of pan-European ecological network projects, and most European countries have included ecological networks or greenways in schemes and legislations encompassing multiple scales. The planning at the general level has mainly been top-down by national and regional governments, with connections to the local levels where implementations take place (Jongman et al., 2004). From a North American perspective, Jack Ahern (1995) also pointed out that different scales of greenways should be interrelated—‘a higher order greenway implies that the lower order greenways exist or will be formed, for greenways cannot be implemented and managed at such vast spatial scales’.

Scales were less of a focus in earlier landscape planning and design. For example, Geddes’s incremental improvement alone would be hardly effective at a regional scale. This can be demonstrated by the fact that most early-20th-century Regional Planning practices (e.g. Geddes’s Tel Aviv planning, Benton MacKaye’s Appalachia Trail planning, and the American Regional Planning Association’s Tennessee Valley planning) had deviated, in varying degrees, from Regional Planning’s original vision of landscape evolution and were similar to modernist master planning (Hall, 2014, p. 186).

Patterns, semantics, and postmodern thinking

There is a strong relationship between complexity science and postmodernism (Cilliers, 2002). Complexity science revealed the inherent uncertainty and unpredictability nature of complex systems. Although patterns in CASs have shared features, they are also locally specific, exhibiting remarkable variation from one site to the next (Braithwaite, Churruca, Long, Ellis, & Herkes, 2018). Likewise, philosophies that postmodernists frequently refer to (such as schools of phenomenology, hermeneutics, and post-structuralism), to different degrees, dispute absolute objectivity, subject-object distinctions, and universal knowledge that are characteristics of modernity or positive science, whereas they emphasise individuals, connections, and contexts (Popper, 1972). Like ecological planners, some landscape theorists and practitioners influenced by postmodernism also used patterns to understand and facilitate the complex connections in landscapes.

Spirn sees landscapes as a language that conveys meaning through interweaving patterns in a context. Instead of looking for prototypical patterns or general procedures for landscape analysis, which was typical at that time, she reads the patterns of specific landscapes—first zeroing into details, letting them work on the feelings and mind, and seeking connections before understanding the whole (Spirn, 1998, p. 4). Spirn (1988) cited chaos theory, where complex patterns at the macro level are not created from the top down, but result from the interplay of several simple processes at the local level. This prompted the realisation that landscape patterns result from a similar interplay of multiple natural and cultural processes, which ‘vary in response to the specific context of environment and culture and to the idiosyncrasies of individuals’ (Figure 5). While acknowledging the new science as a basis, Spirn (1988) considered such notion as ‘a new aesthetic’ that ‘builds on a rich history of antecedents [in landscape architecture]’. Spirn also integrates bottom-up approaches in practice, as exemplified by the multi-phase project of the Mill Creek neighbourhood of West Philadelphia that began in 1987 and continues in the present. Proposed in the first phase (1987–1991), the West Philadelphia Landscape Plan (WPLP) was a framework for local actions that ‘combine[s] a comprehensive, top-down approach with a grassroots, bottom-up approach to urban planning and design’ (Spirn, 2005). In the second phase (1994–2002), Spirn and her students taught local youth to read the landscape, and they collaborated to design and restore the community (Spirn, 2005).

Figure 5. Anne Whiston Spirn (1988) used drawings (top) of patterns from chaotic dynamics, which result when several rhythms interact, produced by James A. Yorke (Gleick, 1987, pp. 294–295) as an analogy of the formation of landscape patterns by the interplay of natural and cultural processes. In the photo (bottom) taken by Spirn, the arc of poplars reflects the meandering of the river, and the crazy quilt pattern of allotments is fashioned by individuals’ gardening.

Corner also believes that landscapes are texts with meanings. He countered the ‘tyranny’ of positivism in landscape architecture that pursues universal knowledge and instead suggested a strategy grounded in the ‘hermeneutics’ (Corner, 1991)—a postmodern epistemology that sees knowledge as relative, contextual, and evolutionary, and works through rhetoric and metaphor to join seemingly disparate knowledge and excavate new relationships and meanings. Corner (1997) also viewed ecology as a rhetorical representation of how things are ‘deeply bound into dynamic, complex, and indeterminate networks’, which ‘has been further reinforced by new scientific findings of nonlinearity, complexity, and chaos dynamics’. In practice methods, Corner (1999) considered mapping in landscape architecture as a hermeneutic process that inspires creative design through discovering and establishing new and meaningful relationships (Figure 6).

Figure 6. James Corner and Alex MacLean’s mapping of the Very Large Array Radio Telescope in Magdalena, New Mexico, US, connects the physical landscape with a sense of the cosmic order in a self-similar hierarchical structure (Corner & MacLean, 1996, p. 171).

In this section, complexity notions in landscape architecture have become more profound. However, a gap existed between theory and practice at this time. Although landscape ecological and postmodern theories facilitated landscape architects’ understanding of the structure and process of landscapes as complex systems, there were still limited methods in planning and design to handle such complexity. As Zube (1986) noted, landscape theories were often ‘belief systems’ in the form of spatial or functional relationship models that dictated ‘what ought to be’, rather than ‘exploratory systems’ asking ‘what is’ and ‘why is it the way it is’ (let alone how to be it). As a result, some practitioners might adopt methodologies based on incomplete or incorrect understandings of theories, such as considering spatial patterns as an instrument for creating optimised master plans rather than fostering dynamic interactions. Attempts at bottom-up approaches confronted challenges in reconciling the diverse and potentially incompatible perspectives from different parties.

Complexity notions as an ‘explorative system’ (the 1990s to present)

With the advancement of technology and the expansion of the scope and complexity of ecological issues, landscape architects have been seeking better collaboration between analytical science and creative design. Complexity notions have been more widely accepted and practiced, not only in understanding and creating complex landscape structures and dynamics but also in the creative and resilient processes of landscape planning, design, and management.

Scenario-feedback in research through design process

There was a time when landscape planning pursued rational and replicable procedures to find the optimal solutions, where ‘any man, assembling the same evidence, would come to the same conclusion’ (McHarg, 1969, p. 35). However, identifying the best solutions for problems in complex, nonlinear systems is mathematically impossible (Wu, 2013). Additionally, because uncertainty is prevalent in coupled human-natural systems, the outcomes of any interventions would be unpredictable (Moallemi, Kwakkel, de Haan, & Bryan, 2020). Since the 1960s, urban planning scholars have suggested planning processes that include multiple scenarios and feedback loops, drawing analogies to cybernetics and decision-making theories used in complex system control (Hall, 2002, p. 249).

In the 1990s, landscape scholars also proposed planning or design processes characterised by scenario-feedback loops (e.g. Steiner, 2008, p. 11; Ahern, 1999). Carl Steinitz proposed a six-step iterative process in 1990, which is now referred to as a ‘framework for Geodesign’ (Steinitz, 2012). Frequently citing Herbert Simon’s (1996) definition of design that ‘everyone designs who devises courses of action aimed at changing existing situations into preferred ones’ (p. 111), Steinitz (2008) expounded that there are two strategies in design thinking: one is anticipatory by designing the future and asking by what actions might it be achieved; the other is exploratory by proposing alternatives and asking what future scenarios might result from them (Figure 7). The latter is especially applicable to complex landscape projects, where future assumptions and requirements are usually uncertain (Steinitz, 2012).

Figure 7. Based on Herbert A. Simon’s definition of ‘design’, Carl Steinitz (2008) described two ways of design thinking: the anticipatory and the exploratory. Reproduced by the authors.

In the past decade, scenario-feedback processes have been increasingly adopted in the research method for landscape architecture called ‘research through design’ (Lenzholzer, Duchhart, & Koh, 2013; Lenzholzer & Cortesão, 2018). Traditional methods that integrate landscape design and scientific research can be described as ‘research for design’, in which design is anticipatory with research results as a goal. In contrast, research through design starts with designing multiple options (scenarios), because of the uncertainty of the future, before evaluating which options can lead to more desirable outcomes (Moosavi, 2022; Nijhuis & de Vries, 2019). Stakeholders’ knowledge and demands gathered through surveys or workshops can be involved in setting goals, co-designing scenarios, and simulating and evaluating the potential outcomes (J. Liu, Zhang, & Nikita, 2021; Tieskens, Shaw, Haer, Schulp, & Verburg, 2017). In this way, research through design integrates science with the design culture where creativity and negotiations are central (Cortesão & Lenzholzer, 2022; Maher, Maher, Mann, & McAlpine, 2018). Positioned in specific project contexts, the methods and new knowledge generated in research through design are more pragmatic and transferable to practitioners (Prochner & Godin, 2022).

An example of research through design is offered by a recent study in Nansha, China, a traditional polder landscape undergoing rapid urbanisation (J. Liu, Zhang, Xia, Zheng, & Chen, 2022). Faced with multiple assumptions that may lead to a desirable future and the fact that their actual effects are uncertain, this study did not try to find ‘the optimal’ solution based on scientific analysis; instead, it used alternative ‘design’ proposals that reflect different stakeholder’s assumptions—local willingness-based, historical landscape-based, and zoning plan-based scenarios—to conduct experiments (Figure 8). An agent-based model was employed to simulate future landscape changes derived from the hehaviours and interactions of different local actors under different scenarios. The simulated outcomes of different scenarios were evaluated and compared by stakeholder-provided and weighted indicators, which offered feedback for design.

Figure 8. Alternative design proposals reflecting different stakeholders’ assumptions were developed, simulated, and evaluated to explore desirable urbanisation strategies for a cultural landscape (Liu et al., 2022).

Learning by doing for resilient landscapes

While the structure of CAS inspired landscape planning and design in the previous phase, the process of CAS has become increasingly important in this phase. Addressing the prominent complexity and vulnerability of human-dominated or urban ecosystems, resilience theory has attracted the attention of ecologists and designers (e.g. Forman, 2014; Lister, 2007, 2018; Pickett & Grove, 2009). Resilience refers to a system’s ability to absorb disturbance without altering its basic structure and function (Holling, 2001). Drawing from resilience theory, CAS often exhibits recurring dynamics, moving through ‘adaptive cycles’ (involving four phases of exploitation, conservation, release, and reorganization), with cross-scale interactions and feedback (Holling, 2001).

In integrating new ecological insights associated with CAS and resilience with the design realm, ‘metaphors’ that build on the familiarity of both disciplines have proven a powerful tool (Pickett, Cadenasso, & Grove, 2004). Because metaphors of nature order have long been essential in landscape architecture, some landscape architects have called for ecology to serve as a metaphorical tool for understanding and creating complex relationships, structures, and processes in landscapes (e.g. Corner, 1997; Spirn, 2013; Waldheim, 2016, p. 6). Meanwhile, it has also been stressed that metaphors, instead of being too rhetorical, must be founded upon scientific and evidence-based models of CAS and resilience for planning and design tactics (Lister, 2018; J. Liu & Zhang, 2018; Pickett et al., 2013).

As resilience theory shows that ecosystems have many unpredictable evolution paths and no single optimal equilibrium state, ecologists and designers have suggested that landscape design and management be a continuous learning process through creating and testing alternatives and modifying them based on feedback (e.g. Ahern, 2011; Lister, 2007; Markolf, Chester, Helmrich, & Shannon, 2021; Reed & Lister, 2014). Instead of evaluating alternative scenarios with computer models, practitioners prefer a ‘learning-by-doing’ manner, which conducts ‘designed experiments’ with ‘safe-to-fail’ rather than ‘fail-safe’ design options through implementation in the real world (Ahern, 2011; Felson, Bradford, & Terway, 2013; Kato & Ahern, 2008; Lister, 2007; Moosavi, 2022). Unlike the traditional approach using experiments as a means to study a system, designed experiments analyse while shaping the system, making it a learning and adaptation process (Figure 9) (Felson et al., 2013)

Figure 9. Traditional research approaches first analyse a system and then alter it, while designed experiments analyse and alter a system at the same time, making it a feedback-learning and adaptation process (Felson et al., 2013). Reproduced by the authors.

An example of design experiments is the Hart-Miller Island project in Baltimore, United States, a novel ecosystem built on an artificial island (Mahan Rykiel Associates, 2020). The team first designed a landform stratifying the site’s geohydrology, which accommodates diverse species adaptive to different field conditions, aiming to support ‘complex emergent ecologies’. To develop an effective seeding rate for the implementation of the design, the team conducted evidence-based experiments with amended and unamended soils collected from the site and alternative seed mixes from wet to dry specialist seedlings. The experiment findings were translated into a custom seed mix and seeding plan for the original landform design, which is reproducible across the site. In this project, experiments and design alternated as a feedback-learning process.

The Jiangyangfan Ecological Park in Hangzhou, China, built in a valley filled with dredged mud, provides an example of learning by doing under uncertainty. Apart from areas intentionally left out for natural succession, the designers experimented with various previously established and newly introduced plant species competing under supervision in implementation (Figure 10). When the park construction was finished in 2010, the experimental plantation resulted in astonishingly high biodiversity, especially in the number of butterfly species, which had not been expected by the designer (Wang & Lin, 2011). However, this was only the beginning of a continuous change and a long-term monitoring and adaptation process (Quan, Huang, & Fan, 2015; Quan, Zhang, & Gao, 2021). The monitoring showed that the mud level gradually sank during 2009–2021, changing many terrestrial ecosystems into aquatic ecosystems. Consequently, the originally dominant Salix mesnyi was gradually replaced by reeds and tickseed sunflowers (Quan et al., 2021), resulting in an increasing number of birds; the number of butterflies decreases due to the new food chain, forming a new temporal equilibrium (Y. Liu, 2018).

Figure 10. During the construction process of the Jiangyangfan Ecological Park, the designers experimented with various previously established and newly introduced plant species competing under supervision. The park keeps changing and being monitored and managed after construction (Atelier DYJG).

Spirn’s WPLP project mentioned in the previous section is also a salient example of long-term continuous learning. Lasting over 30 years till the present, three successive phases have fostered support for a plan produced in 1987–1991. Numerous demonstration projects have been launched to test and refine the plan based on lessons from successes and failures. The current phase (2017–present) is focussed on revising the unanticipated consequences of gentrification derived from environmental improvements in low-income communities (Spirn, 2021).

To summarise, the properties of CAS, such as self-organisation and resilience, have become critical goals for landscape design and management. Meanwhile, uncertainty and feedback-learning mechanisms have been considered significant characteristics of design thinking and process. Although most practitioners may still be unfamiliar with complexity science, the new science has been better integrated with design culture. Emerging cases have shown that research and design addressing place-specific challenges have propelled complexity science’s pragmatic understanding and application. Complexity notions in landscape architecture have been evolving from a ‘belief system’ of authorities into an ‘exploratory system’ for creative practice.

Conclusions and discussions

Complexity notions are essential to landscape architecture

This review shows that complexity notions in landscape architecture have been a long-standing and increasingly significant subject rather than a passing fad. Such notions are primarily derived from the particular needs of landscape architecture research and practice, though they often draw on analogies from complexity science. Complexity notions in landscape architecture embody scientific understandings of the structure and mechanisms of complex systems, as well as an increasingly profound understanding of the landscape itself in terms of ontology (what a landscape is and how it forms and operates), epistemology (how we can understand landscapes), and methodology (how we can change or design them) (Table 1).

Table 1. Impacts of changing complexity notions on the ontology, epistemology, and methodology of landscape research and practice in response to increasingly complex issues.

Phases	Issues faced	Complexity notions in landscape architecture
		Ontology	Epistemology/methodology
Emergence of complexity notions (early 20th century)	Urban expansion in regional contexts	Regional landscape is an evolving whole underpinned by local human–nature adaptations	Surveying local human–nature interactions before planning
Complexity notions in ‘belief systems’ (mid-20th century–1990s)	Ecological degradation and fragmentation Loss of sense of place	Landscape is a complex adaptive system with natural and cultural agents interacting in a context Landscape has recurrent patterns with nested hierarchical structure	Seeking horizontal connectivity through multi-scale network patterns Excavating and fostering the relationships and meanings in landscapes
Complexity notions as ‘exploratory systems’ (1990s–present)	Expanding complexity and uncertainty of landscape change Demanding better collaboration between science and design	Highlighting the indeterminant and resilient dynamic nature of landscapes	Design is a decision-making process with scenarios and feedback learning Using design to conduct experiments in the real world

Given the unprecedented challenge of uncertainty and the need for resilience in today’s landscapes, practitioners might benefit from a scientific understanding of the structure and processes of CAS, generally, and resilience, specifically (Lister, 2018). However, Spirn (1997) has noted that landscape architects tend to accord higher status to ideas in other disciplines without examining and reconciling their beliefs and traditions and ignore pertinent works in landscape architecture, which often places disparate methods in ‘hostile juxtaposition’. To avoid this, complexity science must be understood in the context of landscape theories and applied to handle the real needs in landscape practice. In this way, the integration of theories and methods from complexity science can be an inheritance and advancement rather than a deviation from the long-standing complexity notions in landscape architecture.

Complexity notions are an exploratory system rather than a tyrannical dogma

Many discourses have used the term ‘new paradigm’ when referring to complexity notions in landscape architecture. It is worth noting that paradigms are not tyrannical dogmas but are competing and ever-changing (Kuhn, 1962). In this sense, paradigms are more open-ended ‘exploratory systems’ than ‘belief systems’. Providing a dynamic and holistic alternative to static and reductionist views of understanding and designing landscapes, complexity notions serve as a common ground that enables and stimulates dialogues between different theories and methods, such as hermeneutic mapping, feedback learning, resilient and adaptive design and management, and the integration of bottom-up and top-down interventions. However, this paper has primarily focussed on the physical/ecological aspect of the landscape. Applications of complexity notions to the cultural/human aspects of the landscape need to be examined in the future.

Meanwhile, embracing complexity notions does not mean rejecting the reductionist, deterministic, and top-down paradigm that is nonetheless pragmatic and useful in practice. In most cases, the two paradigms complement one another. Top-down reductionism helps designers identify aesthetical and applicable spatial patterns and manage the design processes step by step. Bottom-up synthesis and iterations are needed between particular scales or phases so that new possibilities can emerge and evolve adaptively.

Feedback learning become a new paradigm for landscape research and practice

Landscape changes are highly uncertain due to the nonlinear dynamics of complex systems, especially in today’s ecological and socioeconomic context. Meanwhile, planning and design must deal with a wide range of objectives and demands, leading to various possible strategies with unpredictable impacts. Furthermore, it is impossible to identify the best solution in theory under such uncertainty or to conduct empirical scientific experiments at the landscape’s spatiotemporal scale (Wu, 2013). Therefore, landscape practice has to rely on incomplete knowledge.

Learning by doing provides a solution to the above challenges. In such processes, the search for multiple options becomes a key to effective decision-making. Identifying these options requires creativity and imagination that are essential to design expertise, as well as inclusive public participation to ensure that decisions are made based on adequate information and to embrace place-specific sensitivity and social equity. Based on the range of options, computer simulations and implementation experiments can help delineate the impacts of different options and enable continuous learning and adjustment. While decision-making theory based on complexity science informs planning and design, exploratory landscape practice also promotes the development of decision-making methods. The incorporation of adaptive learning processes based on scenario exploration and evaluation is becoming a new consensus in future landscape research and practice.

Acknowledgements

We are grateful to prof. Qing Lin at Beijing Forestry University for her guidance on our previous work that inspired this current paper. We would also like to thank the editor and the anonymous reviewers for their feedback.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Additional information

Funding

This work was supported by the National Natural Science Foundation of China under [Grant 52108051] and Science and Technology Projects in Guangzhou under [Grant 2023A04J2007].

Notes on contributors

Jingyi Liu

Jingyi Liu is a lecturer at South China Agricultural University, China. He holds a Ph.D. degree from Beijing Forestry University. His research interests focus on the history and design theory of landscape architecture and empirical research on landscape change through modelling techniques.

Menghan Zhang

Menghan Zhang is a Ph.D. candidate at Beijing Forestry University, China. She also holds a master’s degree from Columbia University, US. Her research interests focus on the historical and empirical research on regional landscape and rural landscape.