**Developing a Landscape Design Approach for the Sustainable Land Management of Hill Country Farms in New Zealand**

#### **Duy X. Tran \*, Diane Pearson, Alan Palmer and David Gray**

School of Agriculture and Environment, College of Sciences, Massey University, Palmerston North 4414, New Zealand; D.Pearson@massey.ac.nz (D.P.); A.S.Palmer@massey.ac.nz (A.P.); D.I.Gray@massey.ac.nz (D.G.)

**\*** Correspondence: D.Tran@massey.ac.nz; Tel.: +64-27-270-4426

Received: 11 May 2020; Accepted: 31 May 2020; Published: 3 June 2020

**Abstract:** Landscape modification associated with agricultural intensification has brought considerable challenges for the sustainable development of New Zealand hill country farms. Addressing these challenges requires an appropriate approach to support farmers and design a better landscape that can have beneficial environmental outcomes whilst ensuring continued profitability. In this paper we suggest using geodesign and theories drawn from landscape ecology to plan and design multifunctional landscapes that offer improved sustainability for hill country farm systems and landscapes in New Zealand. This approach suggests that better decisions can be made by considering the major landscape services that are, and could be, provided by the landscapes in which these farm systems are situated. These important services should be included in future landscape design of hill country by creating a patterning and configuration of landscape features that actively maintains or restores important landscape functioning. This will help to improve landscape health and promote landscape resilience in the face of climate change. Through illustrating the potential of this type of approach for wider adoption we believe that the proposed conceptual framework offers a valuable reference for sustainable farm system design that can make an important contribution to advancing environmental management globally as well as in New Zealand.

**Keywords:** multifunctional landscapes; landscape services; geodesign; landscape ecology; agricultural landscape planning

#### **1. Introduction**

The green revolution in agriculture that occurred during the second half of the 20th century has greatly contributed to increased global food and fiber production, which has enabled a rapidly growing world population to be fed [1]. In order to increase productivity, agricultural intensification has taken the form of an increase in single crop cultivation and chemical and mechanical inputs [2]. This has led to negative impacts on the environment, evident through a loss of biodiversity and a decline in soil and water quality [3]. In response to the resultant environmental issues and the need to feed a growing population, agriculture needs to evolve from a production paradigm that has focused primarily on productivity and profitability to a more sustainable paradigm that focuses on how to ensure productivity can support human needs whilst also preserving important land resources and environmental integrity [4]. Recently, society and the market have initiated a shift from a focus on agricultural productivity and intensification to a focus on sustainable farming (with an emphasis towards efficiency, sustainability and resilience) [5]. New Zealand (NZ) is a good example of an agriculturally-focused nation that faces sustainable production challenges. It has achieved great

improvements in agricultural productivity and product quality over recent decades [6] but progress has come with significant environmental costs [7].

Although New Zealand is accredited as one of the more sustainable countries in the world and was ranked 11th globally in 2019 for sustainable development [8], its agricultural sector is facing a number of significant issues, such as soil degradation [9], water pollution [10], greenhouse gas (GHGs) emissions [11,12] and soil erosion [13]. Moreover, the possible impact of climate change (e.g., increased flood risk, storm damage and drought severity) is also a crucial threat to agricultural production [14]. To respond to these environmental issues farmers are now faced with a situation of having to operate farm systems that are productive and profitable as well as being sustainable with limited impacts on the environment [15]. This is a major challenge facing NZ farmers, as agricultural production could potentially become increasingly constrained by environmental regulations [16] as governments also respond to growing environmental concerns.

The environmental challenges facing future farming systems are likely to strongly impact upon NZ hill country farming. This is because environmental issues compound already high concerns for these farms, which are associated with the contemporary impacts of increasing production costs, market volatility, climate change, highly variable topography and climatic conditions, and more dispersed and isolated families and communities [17]. This means that future hill country farming systems will need to improve its profitability and build resilience in order to be able to adapt to a changing climate whilst reducing its impacts on the environment. To do this, farmers will need good support systems to help with land use decision-making. However, current land use planning and management approaches that support farm and landscape decision-making in NZ reveal several limitations, such as lack of data and model transparency, insufficient collaboration capability among researchers, policy-makers and other end users, and are limited in terms of the communication of modeling results to end-users [18]. Additionally, some land and environmental planning tools are not simple to implement, as farmers are overwhelmed with information and the process required to develop the land and environmental plans [19]. Consequently, these limitations will reduce the effectiveness of land and environmental planning strategies. Therefore, the development of an effective landscape design approach will be central to helping farmers develop profitable and sustainable farming systems in the future.

The multiple objectives of sustainable agriculture require a multifunctional agricultural landscape that promotes agricultural production whilst ensuring environmental standards are met [20], and landscape ecology can have an important role to play in this [21]. Developing a multi-functional agriculture landscape that provides multiple landscape services (i.e., ecosystem services) for society in addition to the service of food and fiber production [22] has become a key focus for sustainable agricultural research and policy-making, and this has been widely discussed internationally [23–26]. However, there is a gap between theory and practice [27], and transferring the concept of creating multifunctional landscapes into the practice of landscape planning and management has proved to be challenging [28]. The reason for this is that agricultural landscape planning needs to be implemented for a specific geographical region that is strongly associated with local knowledge [29]. This needs the planning process to involve the considerations of local people and therefore requires participation and collaboration of the main stakeholders [30]. Often this does not happen and as a result, local people (or "people of the place") may not agree or may not be able to afford the future landscape scenarios proposed by landscape planners [31], so it is critical that the relevant different stakeholder groups can actively contribute to designing the future landscape by bringing their knowledge and aspirations to the table [32]. It is important that effective landscape planning and scenario development involves an iterative collaborative process and that a design-driven perspective is taken [33].

Recently, geodesign has emerged as an efficient instrument for the implementation of sustainable landscape planning [34]. Geodesign integrates geospatial technologies and scientific methods (e.g., geospatial science, environmental science) to inform spatial decision-making based on the knowledge and information obtained from spatial data [35]. By integrating multiple layers of geographic information and spatial analysis models, geodesign enables the identification and development of a future landscape that has an appropriate spatial pattern or configuration of landscape features [36]. This also enables the rapid generation of future landscape scenarios for a study area, the ability to visualize change scenarios, and the assessment of the impacts of future landscape designs on multiple landscape functions and services [37]. In addition, visualization tools and iterative quantitative modeling used in geodesign can promote collaboration between participants, as they enable stakeholders to enter into the discussion and express their opinions and aspirations as part of the design procedure [38]. Among the geodesign frameworks that have been published, the operation framework developed by [31] has been disseminated to a wide range of landscape and environmental design situations [39]. This framework considers landscape design as an iterative process in which the collaboration among the group of people involved in the design process (which includes design professionals, the people of the place, information technologists and geographic scientists) is an integral part of the design procedure, and the relevant stakeholders play a central role in all of the design stages [40].

The adaptation of the framework outlined by Steinitz offers a potential solution to guide farm system decision-making for the creation of multifunctional landscapes. This paper develops these ideas by proposing a landscape design approach for the sustainable land use planning and management of hill country farms in NZ. The approach developed utilizes geodesign and the concepts of landscape function and services as informed by landscape ecology. The specific objectives of this paper are to: (i) define the major challenges facing current and future agriculture in the NZ hill country that need to be considered in future farm landscape planning; and (ii) design a framework that can assist in the creation of multifunctional landscapes for sustainable agricultural production. In doing this, the paper highlights the benefits of integrating geodesign into multifunctional landscape planning for the creation of multifunctional farm landscapes in NZ. This research offers a valuable reference for sustainable farm system design that can make an important contribution to advancing environmental management globally as well as in NZ.

#### **2. Multifunctional Landscape and Geodesign**

#### *2.1. Multifunctional Landscapes and its Application in Agricultural Landscape Planning*

A multifunctional landscape is seen as being one capable of providing a wide range of landscape services (i.e., ecosystem services) covering three main areas relevant to landscape management, i.e., ecological, cultural and production functions [41]. Natural and semi-natural landscapes are considered as multifunctional landscapes because they provide a variety of goods and services to people, such as food and fiber, climate regulation and water purification [42]. However, multifunctional landscapes of the past have been transformed into more simple landscapes (e.g., single-function landscapes), which have a dominant land use type (e.g., croplands). This is because land managers and decision-makers have focused on increasing agricultural productivity rather than considering the benefits that can be provided by a multifunctional landscape [43]. The transformation of a natural landscape into an agricultural landscape, especially one that is farmed intensively, leads to landscape simplification. This occurs as diverse stands of native vegetation are cleared and replaced with a monoculture, resulting in a loss of biodiversity and a reduction in landscape functions and services [44]. Many studies have demonstrated the negative effects of landscape simplification, such as an increase of insecticide use [45], loss of habitats [46] and a reduction in biological control [47]. As such, developing a multifunctional landscape is increasingly being recognized as offering an appropriate solution for solving the issues and challenges that have arisen from agricultural intensification (i.e., landscape simplification) [20].

A landscape ecological approach based on the concept of the multifunctional landscape has been widely applied in sustainable agricultural landscape planning [48–51]. In the European Union (EU), multifunctional agriculture is significantly encouraged, as it is a key concept of the Common Agricultural Policy for the EU countries [52]. This concept is also applied in many developed countries, like the United States of America, Canada and Australia (as cited in [48]). The overall goal of agricultural landscape planning that is based on the concept of the multifunctional landscape is to develop future or alternative landscapes that can enhance and increase the multifunctionality of the current landscape, in order to achieve a better balance between agricultural production and other landscape services [53].

A landscape services approach has been applied in order to examine a wide range of issues in NZ, such as biological control [54], biodiversity [55] and land use planning and management [56–58]. However, some limitations have been identified, such as the obstacles associated with incorporating the landscape services concept into agricultural land use decision-making and the lack of participation and contribution of farmers in the creation of a future multifunctional landscape [59]. Another important limitation is the inadequacy of the link between landscape service supply and demand. For instance, there is a lack of research that assesses the imbalance between landscape service supply and non-market demand in a spatially explicit manner (e.g., where and to what extent in the landscape are certain services generated by agro-ecosystems needed to maintain desirable environmental conditions) [59]. In addition, current research involves limited measurements of landscape services (e.g., biodiversity) other than production services (e.g., food and fiber) across small areas (e.g., farm scale) [60]. Therefore, the ability to fully integrate multiple landscape services into land use planning and the implementation of a collaborative planning process will provide a greater opportunity to address these gaps.

In this research we have used the terms landscape functions and landscape services instead of ecosystem functions and ecosystem services. Although these concepts are often used as synonyms, it is advocated that the use of the terms landscape functions and landscape services is more appropriate, as these terms are more attractive to people outside the ecological sciences and may be more related to local people [61]. In addition, landscape functions and services are more appropriate to landscape planning, which is strongly associated with human involvement, whereas ecosystems are often perceived as merely natural and semi-natural systems [62].

#### *2.2. Geodesign*

Geodesign is defined as "a design and planning method which tightly couples the creation of design proposals with impact simulations informed by geographic contexts, systems thinking, and digital technology" [37] (p. 29).

Geodesign often involves collaboration among essential groups (e.g., the design experts, geographical information system (GIS) scientists, information technologists and the stakeholders) to develop and decide sustainable scenarios for the future landscape of their area [31]. These groups comprise a geodesign team, and collaborate based on a set of questions and methods, typically within a framework that consists of six key questions [63]:


Six models are employed to answer each of the six questions, ranging from the description of the study area to the decision on a desired future landscape. The process presented in the framework is an integrated and continuous procedure, because the outcome of each phase serves as an input for the subsequent phase, and all the stages of the design (understand study area, specify methods and perform study) are incorporated into one unified system.

Recently, geodesign has emerged as an innovative design approach, developed to provide alternative scenarios for future landscapes, based on a rich knowledge base about the environment [35]. Geodesign has been extensively applied to different landscape planning and management case studies, such as urban development [64–66], environmental management [67,68], and sustainable agricultural

land use [69–71]. This approach is also flexible in terms of the scale of application (e.g., a street, a farm, small town, catchment and regional scales) [72,73]. Various examples of geodesign applications were discussed at the Geodesign Summit in 2019 [74]. In the case of agricultural landscape planning, a typical example of the application of geodesign is illustrated through the use of the approach to increase food production and biofuel commodities and improve water quality and habitat performance in the Seven Mile Creek watershed, Minnesota, United States [75]. At the farm scale, another example is a geodesign project that utilizes 3D modeling and geospatial analysis to design strategies for climate change mitigation on a farm in Iowa, United States. This project applies geodesign for real-time scenario development and interactively evaluating alternative farm design [76].

In New Zealand, GIS tools and techniques have been widely applied to solve environmental problems [77–81], but the tools and approaches that link design and GIS have not been readily available [82] and there is a limited number of applications that follow the geodesign framework to solve problems in landscape planning, especially at the farm scale. For instance, only one previous paper was identified that applied geodesign to plan a route for visitor access across a farm in NZ [83]. Meanwhile, there is an absence of geodesign applications that focus on developing a multifunctional agricultural landscape. Hence, research that utilizes geodesign procedures in an agricultural landscape, especially at the farm scale, has the potential to contribute to environmental management studies in NZ but has not yet been fully explored.

#### *2.3. The Benefits of Integrating Geodesign into Multifunctional Landscape Planning*

Geodesign offers an efficient solution to implement the adaptive design of multifunctional landscape planning. It is an effective approach because it can (1) promote collaborative and adaptive landscape design among different stakeholders, (2) advance landscape multifunctionality in agricultural landscape planning and (3) enable the implementation of the landscape design problem on a large scale. One key advantage of geodesign compared to traditional landscape planning approaches is that it allows for collaboration among researchers, policy-makers, and other end users, because it divides the landscape planning into different processes (with six distinct phases) and allows the participants involved to provide feedback and suggestions at any step in the process [84]. With the latest geospatial technologies (e.g., WebGIS application, human–computer interaction tools), participants can directly interact with both the data and the analysis procedure. This is considered an efficient way to initiate discussion among different stakeholders about alternative futures or visions for the new landscape [85]. In addition, a geodesign framework includes a decision model [63] so this can make the application of landscape planning more adaptive and practical. It supposes that decision-makers may agree with or oppose the proposed change, so the decision model that includes a negotiation process (e.g., discussion) and method (e.g., Delphi method) will be able to effectively build consensus among decision-makers and other stakeholders, as well as able to suggest necessary modifications to the proposed changes or the development of new adapted plans [40]. Additionally, alternative landscape plans are not always going to provide a first and ultimate fix, so decision-makers can iteratively discover the trade-offs and synergies inherent in different design scenarios until a final decision is achieved [27]. In the case of NZ, where agricultural land is under private ownership and farmers are the final decision-makers, the inclusion of a decision model in landscape planning is critical because it increases the role of farmers in the landscape design process. This can potentially facilitate the approval by private landowners of proposed landscape change and therefore make the implementation of future landscape change more feasible [86].

Compared to other landscape design methods and techniques, geodesign has a great potential to break new grounds in the design industry, as it is based on advanced geospatial technologies [87]. State-of-the-art remote sensing, image processing, and GPS tools and techniques enable the collection and processing of large amounts of biophysical data in high spatial and temporal resolution. This means that geodesign can be implemented at various scales [88]. This is an asset in the case of NZ hill country, where geospatial data, and especially data for farm scale application, is poor. For instance, it is

common that there is a lack of detailed land use land cover (LULC) data at the farm scale, so in this case high-resolution remotely sensed data can be used to produce necessary LULC information. In addition, a wide range of tools, techniques, and models that have arisen from GIS, geospatial information, spatial statistics and computer programming can be incorporated into one spatially informed planning platform so as to allow comprehensive landscape design issues to be resolved (as it is a multidisciplinary or transdisciplinary problem) and to provide a more efficient communication mechanism for the modeling processes and results [89]. Geodesign can also integrate different kinds of environmental and socio-economic models to quantitatively and spatially measure the cost and benefit of implementing alternative land use scenarios [90]. The outcome from each geodesign question, such as landscape structure and pattern, environmental sensitivity and risk, and future landscape scenarios, are presented in a meaningful and intuitive visualization (e.g., dynamic map, table and graph) so as to provide better assistance for decision-makers. Once the farmers can see the environmental issues on their farms and measure how much they must invest and can benefit from the future landscape, they will be more confident to make a decision.

In order to effectively co-design future multifunctional landscapes, non-technical people (i.e., farmers) may require an understanding of the basic landscape concepts, such as different socio-economic and ecological landscape functions and services [34]. Through collaboration with other participants, farmers can receive support from technical people (e.g., scientists) to acquire the necessary knowledge. More importantly, geodesign employs GIS models, tools and applications to incorporate numerous layers of geospatial information and transfer the key multifunctional landscape concepts into realistic visualization forms (e.g., map, graph) [91], as well as to develop future landscape scenarios, visualize them and analyze the impacts of the different proposed landscapes on multiple landscape services [37]. This may encourage farmers to pay attention to not only commerce and food production but also the role of the non-trade functions of agricultural landscapes. In addition, the adaptive design capability of geodesign enables farmers' priorities to be considered, as their preferences or requirements can be set in the land change model and this can subsequently increase the ability to reach a consensus between farmers and other stakeholders on future multifunctional landscape scenarios.

#### **3. The Case Study**

#### *3.1. Introduction to New Zealand Hill Country and its Environmnetal Challenges*

New Zealand hill country is defined as land with slopes above 15◦ and located below an altitude of 1000 m above sea level [92]. This landscape type covers a variety of land class types, climatic conditions, geology, and topography properties [93]. The hill country landscape is a mixture of steeplands, rolling land and flat land [94] (Figure 1).

**Figure 1.** Hill country landscape: (**a**) earth flow; (**b**) slump/earth flow; (**c**) steep slopes > 25◦; (**d**) flat topped ridges; (**e**) hilly slopes 15–25◦. Photographed by Duy X. Tran in 2019.

Most of the hill country is classified as land use capability classes (LUC1) 5–7, which are suitable for pastoral grazing, tree crops or production forestry [95]. Other LUC (e.g., classes 3, 4 and 8) often occupy a small proportion of hill country land. Overall, approximately 10 million hectares of NZ's total land area is classified as hill country (approximately 37.5% of the NZ land surface), with the majority located in the North Island (6.3 million hectares or 23.5% of NZ's total area) [96]. Approximately half of the hill country land (5 million hectares or 18% of NZ's total area) is allocated to pastoral farmland used for sheep and cattle farming [97]. It has been reported that sheep and cattle farms, the bulk of which are located on hill country, also own some 25% of the total native vegetation remaining in NZ [98]. This significant proportion of native vegetation plays an important role in carbon sequestration and biodiversity conservation [99].

In recent years, hill country farms have become increasingly concerned about environmental issues [100]. For instance, Beef and Lamb NZ, an industry organization representing NZ's sheep and beef farmers, has defined four pillars for an environment strategy (created in 2018) for sheep and cattle farms. These include working towards cleaner freshwater, healthy and productive soils, thriving biodiversity and reduced emissions in order to achieve the goal of being carbon neutral by 2050 [98]. However, several environmental problems and the negative effects of climate change are challenging the sustainable development of this type of farming [17,101].

Understanding the major environmental challenges facing hill country farming is vital to ensure that good planning for future landscape and farm systems is made for the future. In the following section, the five major issues that need to be considered prior to landscape planning in order to make progress towards a more sustainable future for hill country farming are examined in the discussion below. These are land use change and deforestation, soil erosion, climate change, agricultural intensification and change in consumers preferences.

Large areas of native forests and shrubland on the steep erodible terrain of NZ hill country were cleared for pastoral farming by the European settlers [102]. Although limited deforestation has occurred since the 1980s, the response to historic deforestation and land clearing is still affecting the current landscape and environment [103]. The negative impact of deforestation has been reflected in a significant increase in soil erosion [104]. Over the last three decades, reforestation and regenerating of native vegetation has been increasingly implemented on hill country [105] to reduce sediment loss from steep slopes into river channels [106] and to increase the capacity for climate change mitigation and adaptation [107]. Plantation forestry has a number of positive effects on the environment, such as a reduction in soil erosion and flooding, an increase in carbon sequestration and a reduction in the GHGs emissions, and it has also reduced pressure on native forests for timber [108]. For instance, a report on erosion-prone hill country (for the period of 1997 to 2002) reported that the area prone to soil erosion had been reduced by 36,000 hectares (3% of the total erosion-prone area) due to the planting of exotic forest or through reversion to native shrublands [104]. However, removal of forest cover at harvest on steeplands can result in significant environmental impacts, such as landslides, debris flows and significant impacts on water quality due to sediment loss into waterways [102].

Over the period of 1990–2015, the total area of hill country sheep and beef farms decreased by approximately 1.3 million ha [100]. This is because the more productive land was converted to dairy farming or higher-value horticultural crops [97] whilst the steeper, less productive land, which is more vulnerable to erosion and generates lower financial returns [109], was converted to an alternative land use, such as forestry, manuka<sup>2</sup> for honey production or retirement and a return to native vegetation [17]. Recently, carbon farming, which is a conversion from pasture to forest, is emerging as an alternative to sheep and beef farming in hill country due to the dramatic increase in the price of carbon credits, and this conversion can bring high economic profit if this occurs in eligible areas (the land areas where

<sup>1</sup> LUC class 1 is flat highly productive land and LUC class 8 is very steep unproductive land.

<sup>2</sup> Manuka honey is a monofloral honey produced from the nectar of the manuka, a native tree (Leptospermum scoparium) that grows in New Zealand and parts of Australia.

there has been a net land use conversion to new forests since 1 January 1990) [110]. Therefore, it is important that relevant scientific information (e.g., mapping of suitable areas for alternative land use options) is available so as to allow landowners to make appropriate decisions [111].

In the NZ hill country, soil erosion is a critical issue that contributes to land degradation [112]. The hill country has a high level of both natural and human-induced erosion [113] due to the amalgamation of coarse-textured soils, high slope terrain, high precipitation and agricultural intensification [114]. Soil erosion presents a significant problem to the practices of current pastoral land, and it is especially severe on hill country, which has substantial areas of steep slopes and erodible rocks (e.g., soft rock) [115], especially in combination with high rainfall and high-intensity rainstorms [104]. It is estimated that 192 million tons of soil are lost every year because of erosion and 44% of this takes place on grassland [116]. Soil erosion does not only represent a reduction in NZ's natural resources, but it also results in a decline in soil productivity and a reduction in water quality [113]. In relation to the economic cost, the effects of soil erosion on hill country can be on-site (e.g., a reduction in productivity) and off-site (e.g., an increase in flood damage in downstream regions) [95]. The cost of erosion control and mitigation has often surpassed the value of the production that can be obtained from that land [117], and an increase in vegetation cover (e.g., regenerating native trees, tree planting and reforestation) has been described as being the most efficient solution for this problem [118]. For instance, it is argued that the reforestation of unstable and degraded land can not only effectively control current erosion problems, but also preclude the formation of new forms of erosion [106]. For these reasons, soil erosion control is important in land use planning and management in hill country. Characterizing the detailed spatiotemporal pattern of soil erosion and the capability of landscape options to reduce this environmental problem are central to managing this issue.

Climate change is recognized as one of the significant challenges facing agricultural development in NZ hill country [109], as the country's land-based economy is profoundly reliant on climatic conditions for the growth of pasture and crops [119]. Increased frequency of intense rainfall events is a threat to soil erosion, predominantly on hill country steepland [113]. The expected increase in drought frequency and intensity in some drier regions may severely affect the water supply, agricultural production and magnitude of wildfire risk [120–123]. Climate change may also directly affect pastoral production, because the seasonal variation of pasture growth is influenced by rising temperatures, CO2 fertilization and changes in rainfall patterns [119]. Thus, climate change may result in greater variation in sheep and cattle growth and productivity [124]. Adaptation solutions have been developed to reduce risks and build resilience to climate change impacts in NZ. Some of the major adaptation strategies put emphasis on a long-term perspective and suggest an integration of climate change adaptation into the decision-making process [125].

The impacts of climate change on hill country farming may also be off-site and long-term [126]. For example, climate-concerned international consumers or markets might result in an increased demand for the outputs from production that has low GHGs emissions [127], which will mean that NZ agricultural production will have to change accordingly to maintain their market share. Considerable effort has been made by both the public and private sector to determine climate change mitigation solutions in NZ, and central to this is to reduce the GHGs emissions caused by agricultural production [128]. For instance, in the agricultural sector it is suggested that changes to land use and pasture management will be key solutions for reducing GHGs emissions along with other strategies (e.g., innovation in animal genetics and breeding) [129]. It is therefore suggested that multiple land use options (e.g., pasture, forestry, horticulture) need to be considered in relevant areas of the hill country and the integration of climate change scenarios needs to be made into future land use plans for more comprehensive land use planning and management models capable of addressing issues related to climate change.

Intensive pastoral farming in hill country increased rapidly from the late 1940s to early 1980s. This was due to the increasing demand and rising prices for meat and wool products on the world market [130]. It was also supported by government subsidies for land development, as well as the

emergence of new technological developments (e.g., aerial topdressing—application of aircraft for fertilizers spreading and pasture seeding) [131]. Intensive farming during this period was reflected in a re-clearance of a substantial area of native vegetation that was planted in pasture grass for meat and wool production, an extensive application of fertilizers and agrichemicals, and a high stocking rate [132]. Agricultural intensification and inappropriate agricultural practices in the hill country have resulted in negative impacts on the environment. This includes an increase in soil erosion on steeplands where native bush and shrubs were cleared for pasture, a decrease in biodiversity [133], an increase of nutrient leaching [134], a reduction in water quality [135] and a reduction of future carbon stocks [136].

Since 1984, hill country farming has undergone a dramatic reduction in sheep numbers, as more productive pastoral land was converted to other land use types, and farmers also reduced the stocking rate [130]. Recently, sustainable practices such as organic farming have also been increasingly implemented on some NZ hill country farms [137]. These sorts of changes have resulted in both productivity improvement and better environmental outcomes [138,139]. However, despite these successes, some hill country farms have been managed intensively to improve economic profitability and unsuitable agricultural practices are still happening [140,141]. For example, farmers tend to eliminate the reinvading bush, shrubs and exotic weeds in some high-altitude farms, or marginal land is not fenced off, and this limits the restoration of native forest, which can cause problems associated with soil erosion as well as reducing future carbon stocks [136].

With increasing concerns about the environmental impacts of agricultural intensification and the need to mitigate the impacts of climate change, it is necessary to promote a wider uptake of more sustainable agricultural practices in the hill country [131,142]. Several studies have shown that applying appropriate farming practices, such as developing shelterbelts and hedges, using native plants, or riparian plantings can significantly enhance the provision of landscape services (e.g., increase biodiversity, pest control, water purification) [143–146]. Moreover, by applying appropriate land management decisions it is possible to increase farm productivity whist reducing the impacts on the environment [147]. For instance, using soil data, topographic maps and spatial analysis can help to determine optimum fertilizer application to the appropriate areas and assist in the reduction of nitrate runoff [148]. Making informed decisions requires good land use planning and management tools, which can provide detailed land use and environmental information at the farm scale.

Meat and fiber from NZ hill country farms are well recognized on the world market because they are safe, nutritious and grass-fed [109]. However, international consumers are increasingly becoming aware of environmental issues that arise from intensive agricultural production and are requesting more eco-friendly agricultural products or products that respect environmental standards [149,150]. Therefore, the way food is produced (i.e., considering factors such as environmental impact, animal welfare and carbon footprint) is becoming an important focus of consumer preference that now needs to be considered alongside the more traditional values associated with high quality [109]. Subsequently, environmental and sustainability standards are being added to the traditional quality and health standard requirements for produce. As a result of changes in consumers' preferences, NZ hill country farmers are required to adopt more sustainable farming systems that take into account the impact of their practices on the environment [151,152]. Adopting more sustainable farming practices will not only improve the environmental health of NZ hill country; it also presents an opportunity for farmers to capitalize on the growing market for environmentally-friendly products. The utilization of effective tools for land use planning and appropriate resource allocation will contribute to solving many of the issues faced by NZ's hill country.

#### *3.2. Tools and Approaches for Supporting Sustainable Land Use Planning used in New Zealand*

Government organizations, research institutions and the private sector have developed a wide range of land use models and tools to help to address some of the impacts associated with land use issues and environmental concerns in NZ [153] as well as supporting farm and landscape decision-making in hill country [154]. Various types of models have enabled the user to deal with specific environmental concerns, such as carbon sequestration [155], greenhouse gas emissions [12], soil erosion [156], nutrient loss [157] or water use [158]. There are also various applications to help farmers deal with the issues of farm production: AgInform [159], BiomeBGC [160], MitAgator [161] and Farmax [162]. There are more complex land use models (e.g., Agent-Based Rural Land Use New Zealand (ARLUNZ) [163], New Zealand Forest and Agriculture Regional Model (NZ-FARM) [164], Waikato Integrated Scenario Explorer (WISE) [101]), which can take into account different factors, such as land use information, socio-economic conditions and environmental parameters (climate, water quality and biodiversity) to provide projected outcomes for land use and environmental, economic and demographic indicators.There are also Whole Farm Plans (WFP), which are a long-established land management tool that is being widely used across NZ to deal with both economic considerations and environmental constraints on farming systems [165]. Recently, the Land and Environment Plan (LEP) was developed by Beef and Lamb NZ to support sheep and beef farmers to have a better understanding about the land and environmental issues that exist on their farms so that they can develop a land use and environment plan to manage these issues [154].

Land use and environmental planning tools and models have contributed significantly to agricultural development as well as supporting farmers in decision-making to address sustainability issues in NZ [18]. However, several improvements are required to increase the effectiveness of the model outcomes. A review conducted by Motu Economic and Public Policy Research in 2018 [18] pointed out some gaps that NZ land use modeling needs to take into account in order to improve its usability. These include increasing the reliability of the data and increasing model transparency, improving collaboration capability among researchers, policy-makers and other end-users, enhancing the communication associated with the model results to stakeholders and enabling a climate change mitigation framework in the land use planning process [18].

Of upmost importance for improving model reliability is the use of data with a better spatial and temporal resolution. It was conceded that NZ lacks good GIS data when compared to many other developed countries [18]. Using data that are too generalized means that it is not possible to achieve accurate analysis, especially at the finer scales (e.g., farm and paddock) [166], as it will fail to capture the variability present at the a farm scale in relation to factors such as variations in slope, soil types, soil fertility and effective rainfall [97]. Therefore, it is important to consider acquiring better data at a high resolution so that land use optimization models can adequately represent the environmental and ecosystem services variability within small farm-scale areas [59]. It is also important to have an appropriate amount of time-series data to enable trends in environmental issues to be examined over time [167]. This is critical for predicting change to the future environment and is an important basis on which to develop long-term land use and environmental planning.

It is also important that land use planning takes into account the collaboration between different stakeholders so that they can be involved in the planning process [18]. Farm system research has evolved to recognize that there needs to be a shift towards more trans-disciplinary approaches to farm system management, which require collaboration and integration of knowledge and ideas between different people, disciplines and methods [168]. A framework that allows the collaboration among researchers, policy-makers and users will enable them to easily and actively be involved in the planning process and develop a comprehensive land use plan that satisfies multiple objectives (i.e., socio-cultural, economic and environment issues).

A land use planning framework needs to enable the integration of different models and tools to better solve different aspects of land use planning. Various tools and applications have been developed to deal with a wide range of the land use and environmental issues in NZ, and these continue to receive support and investment from the government, research institutions and the private sector [169]. However, the integration of different models into a single framework to solve interdisciplinary questions has been limited in NZ [169]. Hence, future land use models need to consider the synergies between different models and techniques so that they can be utilized to solve real world problems.

It is necessary to improve the communication of both the modeling processes and the outcomes from this process. Some land use and environmental planning tools are not simple to implement, as they require farmers to prepare and enter a large and complicated set of data into the model. Such models may also use several complex spatial analysis processes (e.g., map overlay, multicriteria analysis) to define environmental issues on a farm, which are often difficult to interpret [170]. In fact, land and environmental planning is a spatially complex problem, since it requires the integration of a wide range of geographic information (e.g., soil, land use types, climate variables) to define issues and allocate and plan resource use. Without an appropriate spatial support system, the process is intimidating for farmers, as they are overwhelmed with information [19]. An adaptive spatial-based decision support system incorporating spatial analysis tools and techniques would provide models with the capability to capture, store, manipulate, analyze, manage and visualize land resources and environmental data and information [171]. This would make the results more transparent to the various decision-makers through the use of different forms of visualization (such as interactive maps, graphs and reports).

#### *3.3. Why Have Multifunctional Landscapes on Hill Country Farms?*

Landscape simplification is significant in the hill country landscapes in NZ, as there has been extensive conversion of the natural vegetation to pastoral land associated with the expansion of agriculture since European settlement [172] (Figure 2). The area under pasture has increased rapidly from less than 70,000 hectares in 1861 to 1.4 million hectares in 1881, 4.5 million hectares in 1901, and 7.7 million hectares in 2016 [173,174]. The conversion of natural ecosystems (e.g., forest, shrubs) to pasture has led to a degradation of landscape functions in the sense that provisioning services (e.g., grazing production) are dominant and increasing, whereas regulating services are weak and declining. In other words, the human need to produce food has eroded the capacity of the ecosystems to produce other essential services (e.g., regulating services) [175]. The negative impacts of landscape intensification on hill country are well documented, such as the impacts on the provision of freshwater [176], soil and plant biodiversity [177,178] or soil biogeochemical cycling of nutrients [179].

**Figure 2.** Example of NZ hill country landscapes: high simplification with low regulating services (**left**); low simplification with high regulating services (**right**). Photographed by Duy X. Tran in 2019.

It is suggested that the issues that originate from landscape simplification due to agricultural intensification could only be solved by taking into account the redesign of agricultural landscapes [180]. The goal of the approach suggested in this paper is to redesign (or plan) the agricultural landscape to achieve a better balance between ecological, cultural and production functions [180]. The cultural and production functions reflect the capability of the landscape to produce goods and services that support human demand from a socio-economic perspective [181]. Whereas maintaining and improving the ecological functions of the landscape is thought to increase biodiversity and landscape connectivity, which has important conservation and landscape resilience implications, including the ability to adapt to climate change and disturbance [182–185]. The creation of this kind of landscape is expected to be an effective solution to solve the problems related to landscape simplification in NZ hill country farms. The justification for this is that a multifunctional agricultural landscape that is made up of a mosaic of natural habitat areas and agricultural production areas could help to maximize the balance of ecological and socio-economic demands and minimize the conflicts between them [186]. This allows the landscape to provide multiple services and achieve multiple objectives (both agricultural production demand and environmental standards) [51,187]. By diversifying farming activities, farmers can secure various income sources whilst at the same time promoting the cultural and natural heritage [188]. For instance, a sustainable multifunctional agricultural landscape may provide the option to develop agritourism or environmental education. Consequently, this contributes to an added income for farmers and increases public interest in the social and environmental values that the farms bring to the community. However, the challenge comes in determining how to implement the multifunctional landscape approach as a practical application to develop a sustainable agricultural landscape where different land use and land cover types (e.g., wetland pasture, forest, and horticulture) co-exist and the land use pattern is appropriate to maintain and promote sufficient heterogeneity so that different landscape functions work properly [189,190].

#### **4. A Conceptual Framework that Combines Multifunctional Landscapes and Geodesign Concepts for Sustainable Agricultural Landscape Planning**

In this paper, we propose a conceptual framework for sustainable agricultural landscape planning (Figure 3) that integrates the concept of multifunctional landscapes with a geodesign approach. It also draws on several studies that have focused on developing a framework for landscape planning [87,187,191]. The geodesign processes in this framework follows the approach outlined by Steinitz [31], which comprises six phases. These phases are: (1) Landscape description, (2) Landscape process; (3) Landscape evaluation; (4) Future landscape scenarios development; (5) Impact assessment of alternative landscape scenarios; and (6) Decision-making. Within this framework, the basic concepts of a multifunctional landscape and a landscape services approach can be fully integrated.

#### *4.1. Landscape Description*

The landscape description phase is used to describe a general picture of the study area. The first task is to define an appropriate boundary for the study area. It is suggested to consider both the social and ecological boundaries (i.e., boundaries that cover both the ecological and socio-political/cultural functions of the landscape) when defining the boundary for the study area [192,193]. The ecological boundary of the study area may be determined based on ecological processes or biophysical constraints (e.g., land management unit, catchment or sub-catchment boundaries) [194]. The cultural functions of the landscape sometimes may not align with the boundary of the ecological functions, so it is recommended to work with the "people of the place" to properly define an appropriate boundary [192]. In NZ, a catchment group is a community network of farmers who operate in a particular catchment. They are increasingly committed to tackling environmental issues and responding to a long-term sustainable development plan for the catchment [195,196]. Working with such groups offers the potential to assist in developing a relevant cultural boundary.

Once the study area boundary is defined, the next step is collecting necessary physical and socio-economic data, especially data for characterizing landscape services and environmental issues (e.g., soils, topography, LULC, climate). In the case of NZ hill country, the lack of data is a limiting factor for analysis. To navigate around this requires an integration of multiple data sources that may come from the government, research institutions, remote sensing and field surveys. In fine-scale applications, such as those undertaken at the farm and paddock scales, information provided by farmers (e.g., stocking unit, grazing rotation) is an important source of data. The integration of local and global data to model landscape services is therefore a valuable option to address data deficiencies in remote and data-poor areas [197]. In addition, data are normally archived in different formats, standards and scales, so data standardization is an important step to make sure multiple data layers can be appropriately integrated and used.

**Figure 3.** A conceptual framework for a multifunctional landscape-based geodesign for sustainable landscape planning, adapted from Steinitz [31].

A representation model (e.g., a raster-based 2D data model) is used to organize and visualize data collected for the study area through space and over time. For example, maps visualize LULC types of a farm or rainfall and temperature patterns in a catchment from 20–30 years ago to the present. This gives a general understanding of the landscape (from the past to the present) and provides necessary input for the other stages of the framework. Data resolution and availability will affect all other processes of landscape design, as the difference in the resolution and level of data accuracy in the input process could lead to completely different results. For instance, small landscape features (e.g., small plots of shrubs or ponds) play an important role in a farm, such as providing biodiversity, water resources and shade for stocks. However, these features are often eliminated in the low-resolution data (e.g., LULC at the catchment or smaller scale), so landscape services provided by these features may not be quantified when using such coarser data.

#### *4.2. Landscape Process*/*Operation*

The landscape process phase aims to define key processes in the study area that include both physical/ecological drivers and socio-economic drivers. The first step is spatially and quantitatively characterizing major landscape functions, services, and values. This provides insights into the landscape operation in which important landscape characteristics are examined. An example of major landscape functions or services supply and their indicators on a hill country landscape in NZ is presented in Table 1. It is important that landscape service supply is estimated in monetary units so that the overall benefit that a landscape provided can be easily measured. Various economic valuation methods have been used for estimating the value of landscape services, such as market prices, replacement cost and provision cost [57]. For instance, the market price method can be applied directly to convert several landscape services (e.g., pasture and timber production, carbon sequestration) to appropriate monetary units. Many indirect use services (e.g., drought mitigation, flood mitigation, nutrient retention) may require using provision cost or replacement cost methods to transfer their qualities to monetary value. Additionally, the economic value of landscape aesthetics in an area can be evaluated by estimating people's willingness to pay for visiting heritage or tourist sites distributed in the landscape.


**Table 1.** Example of landscape services in the hill country New Zealand.

Sources: adapted and revised from [57–60,206,207].

After that, the spatial interaction between the provision of landscape services and landscape simplification and LULC dynamics are analyzed to determine how these processes are linked to each other. A substantial number of studies have stated that the provision of landscape services has been significantly affected by LULC dynamics [198–203]. Quantifying these relationships will be a key to transferring a multifunctional landscape design to a future land use plan. Landscape indicators

that reflect the landscape simplification (i.e., agricultural intensification) well (e.g., the proportion of cropland and semi-natural land obtained from LULC data [45]), the variations in landscape services provision (e.g., landscape services change index [204] or multifunctionality index [205]) and spatial regression analysis will be used to characterize the spatial interactions between the change in LULC and variations in landscape services.

Quantifying and mapping landscape services can help farmers recognize and understand the multiple values of their farms. This is an advantage compared to using land cover information, as many landscape services may not be directly quantified by using land cover data alone [208]. Understanding major landscape processes and the interaction between them is the key basis for designing a sustainable multifunctional landscape.

#### *4.3. Landscape Evaluation*

The landscape evaluation phase seeks to assess whether the landscape is working well or not [28], in other words, assessing the overall quality of the landscape [209]. In a multifunctional landscape this can be understood as assessing the quality of goods and services that a landscape provides to humans and the environment. To determine landscape quality, an evaluation model that utilizes comprehensive indicators will be used to evaluate the attractiveness, vulnerability and sustainability of the study area. Attractiveness refers to the advantages that landscapes may have for a specific land use purpose or for socio-economic activities (e.g., suitable soil and climate conditions for fruit production). The vulnerability relates to characteristics that negatively contribute to socio-economic development or the environment (e.g., impacts of extreme climate and steep slopes on agricultural production, or negative effects of agricultural intensification on water quality and biodiversity). Sustainability reflects the landscape's capacity for steadily supplying long-term landscape services that are critical for maintaining human and environmental well-being (e.g., a landscape that has different functions and services that co-exist and balance) [210].

Landscape assessment indicators, which can be of various types, including single (e.g., GHG emissions mitigation index), multiple (e.g., a combined-index integrating several parameters, such as soil erosion control, carbon sequestration and drought mitigation), static (the sustainable threshold being classified into a fixed category) and dynamic (the sustainable threshold being subjected to the dynamic interaction between indicators) [87,211], and come from various sources (e.g., expert consultant, environmentalist, empirical analysis, law and regulation) [31], could be used to assess past and present situations of a study site, monitor the design process and compare design alternatives [87]. Hence, choosing appropriate indicators is important for the success of a landscape design project. Suitable landscape indicators should satisfy several requirements, such as the capability to reflect a wide range of landscape services to analyze the trade-offs between landscape service provision and land use change options [212], providing reliable, detailed, understandable, comparable and spatially explicit information to support decision-making [213], and providing cost-effective indicators by utilizing available data or employing low-cost generated data and models [214].

Landscape evaluation models also need to reside within the geographical context in the sense that assessment indicators should recognize and align with existing legitimized environmental strategy and policy and reflect major landscape processes in the study area. For example, in the case study of hill country in NZ, water quality, soil erosion control, drought mitigation, pasture productivity and GHGs emission mitigation could be used as some of the indicators for landscape sustainability assessment.

#### *4.4. Future Landscape Scenarios Development*

Based on the results achieved from the landscape evaluation process, change models will be used to define a series of alternative future scenarios for the proposed multifunctional landscapes. In this stage, stakeholders can follow the scenarios developed by scientists or propose their scenarios (a user-defined plan) for the future landscape. Alternative scenarios for future landscape design can be implemented by applying the following procedure:

First, the information on landscape process (characters, services and values) as well as major socio-economic drivers and environment issues are used to define how the landscape should be changed. Determining the expected future landscape is based on several assumptions, such as the preferences of local people, the landscape functions or services that the future landscape will be capable of providing, and the implications of policies and regulations [215]. In agricultural landscapes, the design goal for future landscapes is mainly based on the level of agricultural intensification (or landscape simplification) [216]. Landscapes that have been highly simplified may need to be redesigned in order to restore integrity between provisioning, supporting, regulating and cultural services, whereas the likely design goal for less simplified landscapes is to increase provisioning services while maintaining current levels of other services [180]. Climate change scenarios can be integrated in this step to measure how the changes in climate variability can affect the landscape operation through the interaction with landscape functions.

Afterwards, a design strategy that could take an offensive approach (where the design goal is utilizing the advantageous or attractive landscape characteristics to develop a future landscape), or a defensive approach (where the development of a future landscape is based on one that avoids vulnerability or risks), or a combination of these approaches, will be used to create a specific change model to simulate future change for the landscape [31]. There are different methods of designing for landscape change, such as rule-based, optimized, and agent-based approaches (see [31] (pp. 56–59) for further details). Among these, the use of multi-criteria decision-making (MCA) can be an efficient method to propose future landscape scenarios in the study area, as the creation of a future landscape can be regarded as a complex MCA process [217]. Each land use scenario or option often requires multiple objectives (e.g., erosion control, carbon sequestration, pasture productivity, GHGs emission) and the final decision will be a compromise between the interests of the different stakeholders involved in the design process. The results from these approaches are maps showing the future landscape with the distribution and pattern of different LULC types. Associated with each LULC map will be the provision of landscape services and landscape multifunctionality maps. For each scenario and stage, different alternatives can be created and reassessed iteratively until consensus is achieved.

#### *4.5. Impact Assessment*

In the impact assessment of alternative landscape options, the criteria and indicators used in landscape evaluation will be applied to assess the positive and negative impacts (benefits, risks and sustainability) of the future landscape. In a geodesign project, an environmental impact assessment is often implemented to characterize the consequences of the proposed change. In the context of developing a multifunctional agricultural landscape, the impact assessment is related to quantifying the costs and benefits (including both socio-economic and environmental costs) of recovering landscape functions or re-designing the landscape to increase landscape diversification (or landscape multifunctionality). The results of this stage include maps and statistical data showing the cost–benefit ratio of each alternative landscape option. For instance, associated with each land use scenario will be maps showing landscape services provision and value of carbon sequestration, GHGs emissions, erosion control, drought mitigation and pasture productivity, as well as the total benefit (value) of that scenario. This includes the cost to implement such a landscape (e.g., loss of pastoral area, fencing cost, tree planting cost). This will be critical for the decision-making stage.

#### *4.6. Decision-Making*

In the last phase, the scenario analysis and group discussion will be conducted with the public, experts and stakeholders. The results of the future landscape scenario development and impact analysis will be utilized for discussion, and this will from a basis for making the final decision. According to Steinitz [31], participants in the geodesign process might give different answers, including "Yes", "Maybe" and "No", in response to proposed scenarios. If decision-makers agree with one of the proposed plans, the next stage is to develop the implementation plan. In case stakeholders are not sure

about their decision, further study or analysis is needed to provide more information to help them decide. Sometimes decision-makers may not approve the designed landscape. If this is the case it is necessary to get comments and feedback on why this is so. This will be valuable information to integrate into the landscape project in the future.

The proposed framework in this research inherits the major advantages exhibited by a geodesign approach. These include the fact that it can be a continuous procedure, a multidisciplinary or transdisciplinary approach, and a participatory collaborative planning technique. Moreover, this framework integrates concepts drawn from landscape ecological theory (such as incorporating information on landscape functions and services, landscape simplification and landscape pattern). This means that the theory provides the scientific context to informed and collaborative decision support processes for farm systems that are faced with the need to change in response to environmental pressures and market influences.

#### **5. Conclusions**

This paper reviewed the major challenges facing NZ hill country farms and proposed an approach for sustainable agricultural landscape planning. The significant issues facing hill country farming include land use changes and deforestation, soil erosion, agricultural intensification, climate change and the impacts of changes in consumers' preferences. These challenges are considerations for farmers striving towards the long-term sustainable development of NZ's hill country. Currently, landscape simplification associated with agricultural intensification is a significant feature of hill country farms. This may reduce the landscape's capacity to mitigate and adapt to the environmental challenges and climate change effects. Therefore, we have suggested that designing a more sustainable multifunctional landscape is a possible solution to tackle the issues facing NZ hill country. The development of multifunctional agricultural landscapes can contribute towards innovative future farming systems that can deal with emerging environmental issues [218]. In addition, the design of multifunctional landscapes can improve their resilience to change and disturbance [219], which will be crucial for ongoing sustainability in NZ hill country.

This is one of the first studies to propose a geodesign framework for sustainable multifunctional agricultural landscape planning in NZ. By integrating a multifunctional landscape approach in a geodesign context we offer a solution to address some of the implementation problems that have restricted uptake. Considering landscape planning in a design-driven perspective, geodesign embraces collaborative planning (among different stakeholders) as the key to landscape design. It also enables the incorporation of stakeholder values and aspirations as a central element to this process. By dividing the landscape design process into different phases and utilizing geospatial technologies (e.g., human–computer interaction), geodesign allows important stakeholders to be effectively involved and contribute to the planning process. In addition, geodesign enables the use of multiple sources of relevant spatial and temporal resolution data for landscape planning, especially in large-scale applications, as well as being better at dealing with different aspects of land use planning.

The proposed framework in this paper considers the major concepts associated with a multifunctional landscape approach, including landscape functions and services, landscape supply and demand, the value of landscape services, sustainable landscape indicators, spatial patterns and interactions. This facilitates a comprehensive implementation of the multifunctional landscape approach in land use planning and management. A landscape ecological approach has been talked about conceptually for landscape sustainability but has not been widely applied practically in NZ. Therefore, the comprehensive integration of a landscape services approach in landscape planning offers a solution to address some of the limitations faced by current land use planning and management practices in NZ [59,60]. The proposed approach and associated framework can provide a scientific basis towards the development of a future commercial land and environmental planning tool. This will hopefully give farmers and rural professionals more options to conduct useful land use planning at the farm scale.

We believe that the proposed conceptual framework of an integrated landscape ecological (the scientific theory behind a multifunctional landscape concept) and geodesign approach will be a valuable reference for future work about agricultural landscape planning. Ideas around creating multifunctional farm landscapes have been discussed [24,220,221], the role that geodesign can play in future planning has been explored [34,210] and frameworks for developing sustainable landscape based on an integration of geodesign and landscape ecology have been proposed [87,222]. However, there is a lack of a detailed framework that can demonstrate how concepts associated with the generation of multifunctional landscapes can be incorporated into a geodesign process to create a planning tool at the farm scale. Hence, the approach proposed in our paper, which covers a comprehensive description of a type of geodesign process applied to the management of a multifunctional agricultural landscape, will significantly contribute to environmental management studies and illustrate the potential of this type of approach for global application.

Although the framework proposed in this paper demonstrates a comprehensive approach for agricultural landscape planning that can be applied to NZ hill country farms, we acknowledge that future work needs to consider and investigate the issue regarding the financial resources required to support the farmers to overcome their economic concerns associated with changes in land use. Farmers may recognize and be motivated by the great value that extra landscape services can provide and agree with a proposed landscape design, but a barrier to implementation of this design might be the lack of the long-term support that is needed to enable them to be able to afford the cost of implementation and to follow the suggested revised land use and environmental plan. For instance, increasing native woody vegetation on a farm provides a great range of landscape services, but it may potentially affect economic profit in the short term due to the fact that it would decrease land available for grazing and has a low growth rate [223], and thus have less earning capacity in its early life stages. A solution to this is for policy-makers in NZ to consider payment for landscape services. In many countries a wide range of regulating and supporting services are estimated in terms of economic value, and the farmers (i.e., landowners) are able to get a payment for these services [224–226]. Currently, farmers in NZ can only receive payment for carbon sequestration services, so there are no strong incentives to encourage farmers to implement a land use plan that promotes multiple landscape services on their farm. An approach such as the one outlined in this paper can help to demonstrate a proof of concept to policy-makers so that they recognize the greater environmental value that farmers can provide by designing future landscapes for multifunctionality and landscape services and therefore build financial support into future policy-making.

**Author Contributions:** All authors contributed to the research. D.X.T. conceptualized the main idea of the study. D.X.T. and D.P. designed the structure of the paper. D.X.T., D.P., A.P. and D.G. contributed to the writing of the manuscript. D.P., A.P. and D.G. supervised the final paper content and edited the writing. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research received no external funding.

**Acknowledgments:** Acknowledgement is given to Massey University for supporting this research. The authors would like to acknowledge the valuable comments of three anonymous reviewers that assisted with the finalization of this manuscript.

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **References**


© 2020 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

### *Article* **Using the Ecosystem Services Concept to Assess Transformation of Agricultural Landscapes in the European Alps**

**Uta Schirpke 1,2, Erich Tasser 2, Georg Leitinger <sup>1</sup> and Ulrike Tappeiner 1,2,\***


**\*** Correspondence: ulrike.tappeiner@uibk.ac.at

**Abstract:** Mountain farming sustains human well-being by providing various ecosystem services (ES). In the last decades, socio-economic developments have led to worldwide changes in land-use/cover (LULC), but the related effects on ES have not been fully explored. This study aimed at assessing the impacts of the transformation of agricultural land on ES in the European Alps. We mapped 19 ES within the agriculturally used areas in the year 2000 and analyzed LULC changes by 2018. We compared eight regions with a similar development, regarding social–ecological characteristics, to outline contrasting trends. Our results indicate that the ES decreased most strongly in regions with a massive abandonment of mountain grassland, while ES in the 'traditional agricultural region' remained the most stable. In regions with an intensification of agriculture, together with urban sprawl, ES had the lowest values. Across all regions, a shift from ES that are typically associated with mountain farming towards forest-related ES occurred, due to forest regrowth. By relating differing trends in ES to social–ecological developments, we can discuss our findings regarding new landscapes and farming systems across the European Alps. Our quantitative and spatially explicit findings provide a valuable basis for policy development, from the regional to the international/EU level, and for adopting sustainable management strategies.

**Keywords:** social–ecological system; mountain region; spatial analysis; land-use change; farming

#### **1. Introduction**

The IPBES (Intergovernmental Platform on Biodiversity and Ecosystem Services) conceptual framework names three interactions between human societies and the non-human world: nature, nature's benefits to humans, and a good quality of life. To value NCPs (nature's contributions to people; defined here as any positive contribution or benefit, and occasionally negative contributions, losses, or degradations, that humans receive from nature), the concept of ecosystem services (ES) is often used. Since, NCPs are consistent with the original use of the term ES in the Millennium Ecosystem Assessment [1], we define ES as the contributions ecosystems make to human well-being, including the goods and benefits that people subsequently derive from them. The IPBES further reinforces the need for initiatives at the science–society interface, aiming at sustainable futures in the light of global change [2]. Our study applies the ES concept to value the transformation of landscapes in this context, contributing to a possible sustainable adaptation of land-use/cover changes (LULC). Here, we focus on agricultural landscapes, as they are particularly affected by global change, with wide-ranging consequences for society [3–5]. Agricultural ecosystems contribute to a variety of ES, such as food and fodder provision, soil conservation, erosion protection, climate regulation, habitat provision, aesthetics, and recreation [6–8]. In particular, organic or traditional farming systems provide high levels of multiple ES, while conventional farming systems are focused on food production [9–13]. In mountain regions, small-scale farming systems and sustainable management practices have been developed

**Citation:** Schirpke, U.; Tasser, E.; Leitinger, G.; Tappeiner, U. Using the Ecosystem Services Concept to Assess Transformation of Agricultural Landscapes in the European Alps. *Land* **2022**, *11*, 49. https://doi.org/10.3390/land11010049

Academic Editor: Monika Kopecká

Received: 17 November 2021 Accepted: 28 December 2021 Published: 30 December 2021

**Publisher's Note:** MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

**Copyright:** © 2021 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ 4.0/).

over centuries to cope with the challenging topographic and climatic conditions [14]. This has shaped appealing mountain landscapes, which are rich in biodiversity and provide many ES to local people, tourists, and adjacent lowland populations [15–24]. However, institutional and political drivers, socio-economic shifts, urbanization, and technical developments have reduced the competitiveness of these marginal areas and induced a massive abandonment of alpine pastures and meadows in European mountain regions during the last century [25–32]. At the same time, less steep areas in the valley bottoms with a favorable climate and easy access have been intensified, often managed by larger and more specialized farms [33,34]. Such changes have led to still ongoing transformations in agricultural landscapes, with implications for biodiversity and manifold ES [22,23,35–37]. For example, the intensification of agricultural land causes a decline in water quality, because of higher nutrient input, and a reduction in pollination, due to the use of pesticides and a habitat loss [6,38,39]. In addition to a decline in forage provision, the abandonment of alpine pastures and meadows leads to a loss of many cultural ES. In contrast, the provision of timber and non-wood products, and the regulation of the climate and protection from hazards increase due to forest regrowth [33,36]. Hence, previous conditions and past processes not only show an impact on current landscape patterns and functions, but can also determine, to a great extent, future pathways of landscape change [40].

Such developments require the attention of decision-makers and land managers, to foster a sustainable development of mountain regions and maintain high levels of multiple ES provision [41–44]. There is growing evidence that the concept of ES, acknowledging the human benefits obtained from the interaction with ecosystems, provides a valuable basis to support landscape planning and management, in various ways [41,45]. This may include raising the awareness of stakeholders, developing management strategies, and taking decisions [46]. In particular, ES maps can be supportive for identifying developments of ES over space and time [36,38,47]. They can be used for revealing synergies and tradeoffs among multiple ES [24,48,49], and consequently, for setting priorities in land-use decisions; for example, how intensive agricultural use maximizes provisioning ES, while reducing other ES [50,51]. Maps can also help to identify the spatial separation between farming activities and consumers, which is responsible for trading agricultural products globally [19,52,53]. On this basis, decision-makers can develop nature-based solutions, such as promoting dietary shifts, to strengthen the consumption of local products.

However, quantitative and spatially explicit information on the impacts from the transformation of agricultural land is often not available [54,55]. One reason is that studies on LULC in agricultural landscapes are often not sufficiently linked to the concept of ES [51]. Although there is an increasing number of studies dealing with ES in mountain regions in general, many studies have not considered changes in ES over time [56] or did not specifically analyze agricultural landscapes [51]. Regarding spatial coverage, most studies concentrated on the local level, e.g., [28,57–59], or, if carried out at regional or national level, largely neglected social–ecological differences within and across regions [21]. Consequently, national or even regional policies fail to consider diverging local developments, which occur due to the high complexity of large mountain ranges, such as the European Alps, that include a high variety of climatic, topographic, socio-cultural, and political conditions [14]. Furthermore, many studies focused on a limited number of ES [60]. Data on ES that are not directly linked to land-use/land-cover (LULC) or that are more difficult to assess (e.g., many cultural ES) are largely lacking. Therefore, ES are still rarely integrated into policies and decision-making [60].

To contribute to a more comprehensive understanding of recent developments in the European Alps, this study was aimed at assessing the impacts of the transformation of agricultural landscapes on 19 ES. By differentiating eight regions with distinct social–ecological characteristics, our findings illustrate contrasting developments in ES and highlight diverse pathways for agricultural landscapes in mountain regions.

#### **2. Materials and Methods**

#### *2.1. Study Area*

The European Alps are the highest mountain chain in Europe and contain a large variety of landscapes, species, and cultures. They count about 14 million inhabitants and stretch over eight different countries, including parts of France, Switzerland, Italy, Austria, Germany, and Slovenia, as well as the countries Liechtenstein and Monaco. The Alps are a hot spot of biodiversity, and the complex topography influences the natural distribution of soil, the typology of land, and habitat variety. About 49% of the area is covered by forest, followed by agricultural land (27%), high mountain landscapes with shrubs, natural grasslands and rocks (19%), artificial surfaces (3.7%), and water (1.1%).

Due to the high variety of social–ecological conditions across the Alps, the analyses of this study are based on eight regions with different economic and social structures or environmental situations [61]. These regions were identified by Tappeiner et al. [61] through cluster analysis (Ward method, squared Euclidian distance), based on 21 indicators that reflect the three pillars of sustainability in equal measure (Table S1). The classification refers to data between 2000 and 2008, as well as between 1990 and 2002 for change indicators. An updated classification is currently not available. The eight regions (Figure 1) can be summarized as follows:


**Figure 1.** Location of the European Alps in Europe (small map), and the eight social–ecological regions in the study area, as identified by Tappeiner et al. [61] (large map). Authors own elaboration.

#### *2.2. Analysis Steps*

We analyzed changes in agricultural land between 2000 and 2018 in the European Alps for the eight above described social–ecological regions, based on LULC distribution, and related these developments to 19 ES. The ES mapping and impact analysis in this study comprised the following three steps (Figure 2):


3. Impact analysis: To identify differing trends in LULC and ES across the eight regions with differing social–ecological characteristics, we spatially overlaid the raster maps (aggregated LULC, ES values) with the eight regions (Figure 1). We calculated areaweighted mean values for each region in 2000 and 2018, which were used to map and evaluate changes in LULC and ES values.

**Figure 2.** Conceptual steps for examining trends in ES values across different regions with distinct social–ecological characteristics in the European Alps, only considering areas that were agriculturally used in 2000. Authors own elaboration.

#### **3. Results**

#### *3.1. LULC Changes*

The composition of agricultural land in 2000 varied across the eight regions (Figure 3). The residential municipalities, employment hubs, the standard Alpine region, and the traditional agricultural region comprised mostly intensively used LULC types, such as crop cultivation, permanent cultures, and fertilized grassland, whereas unfertilized grassland prevailed in the other four regions. LULC changes between 2000 and 2018 generally consisted in the abandonment of fertilized and unfertilized grassland, and in an increase in settlement area, forest, and abandoned grassland. Change rates, however, greatly differed across the eight regions (Figure 3). The smallest changes occurred in the traditional agricultural region, while the largest changes happened in rural retreats and forgotten rural areas. Residential municipalities and employment hubs had the largest increases in settlement areas, while forest increased above average in forgotten rural areas and rural retreats. Agriculturally used grasslands were frequently abandoned, especially in the latter two regions, but also in the tourist centers and in the dynamic rural areas, resulting in forested areas or succession stages towards forest, such as dwarf-shrub habitats and bushland. In addition, crop cultivation and permanent cultures slightly increased around the main settlement regions, mainly in residential municipalities and standard Alpine regions.

#### *3.2. Changes in ES Values*

Considering only agricultural LULC types (i.e., crop cultivation, permanent culture, fertilized grassland, unfertilized grassland) that were present in 2000, ES values varied across the eight regions (Figures 4 and A1). The lowest ES values occurred in the economically prosperous employment hubs, including the suburbanization region (residential municipalities), mainly due to less ecosystems with high ES supply and below-average values for cultural and regulating ES. In contrast, regions with a high increase in forest or a high share of unfertilized grassland, including alpine pastures and traditional agro-forestry

systems, had the highest ES values; in particular, wood production (P5), occurrence of mushrooms and wild berries (P4), protection against natural hazards (R1), availability of usable water (R2), preservation of valuable habitats and species (R3, R4, and R5), positive impact on climate (R9), opportunities for leisure activities (C1), aesthetic experiences (C4), and cultural heritage (C5). Regions with a high proportion of intensive agricultural land (i.e., crop cultivation, permanent crops, and fertilized grassland) had lower ES values for regulating and cultural ES.

**Figure 3.** LULC distribution across eight social–ecological regions in the European Alps in 2000 (**left**) and 2018 (**center**), as well as LULC changes between 2000 and 2018 (**right**), only referring to the area that was covered in 2000 by agricultural LULC types (i.e., crop cultivation, permanent culture, fertilized grassland, and unfertilized grassland). Wetlands, rivers, lakes are summarized as 'wetlands & water'. Authors own elaboration.

**Figure 4.** ES values in 2000 (**left**), 2018 (**center**), and change in ES value between 2000 and 2018 (**right**). Authors own elaboration.

ES values between 2000 and 2018 mostly declined, especially, regulating and cultural ES, mainly due to LULC changes of agricultural land towards other LULC types, including abandoned land, forest, and settlement areas. On the other hand, provisioning ES increased except for fodder production (P1), but the changes in ES values varied across the eight regions (Figures 4 and 5). Corresponding to the small LULC changes, the smallest changes in ES values occurred in the traditional agricultural region. Changes in employment hubs and residential municipalities were also below average, but there was a further decline in provisioning ES, due to the increasing urban sprawl. Rural retreats had a particularly strong decrease in many cultural and regulating ES values, with the exception of the positive effect on the climate (R9), due to an increase of forests and abandoned land (including heathlands, transitional woodlands, and shrub) on former agricultural land; however, provisioning ES also increased above average, apart from fodder production (P1) and agricultural food production (P2). Across all ES, positive trends only prevailed over negative ones in the dynamic rural areas and the traditional agricultural regions. In spatial terms, the greatest changes occurred in the Southern Alps in Italy and Slovenia, and the Western Alps were more affected by changes than the Eastern Alps.

**Figure 5.** Changes in ES values between 2000 and 2018 across the European Alps. Only agricultural LULC types (crop cultivation, permanent culture, fertilized grassland, unfertilized grassland) that were present in 2000 were considered. Thus, these maps illustrate the changes in agricultural land to other LULC types (including abandoned land, forest, and settlement areas). Authors own elaboration.

#### **4. Discussion**

#### *4.1. Current Trends in Alpine Agricultural Landscapes*

European land management has not been evolving unidirectionally, following predefined trajectories, but rather as path-dependent processes affected by technological, institutional, economic, and social drivers, including sudden events [31]. This is also true for the European Alps [30,31]. Since the beginning of the 20th century, the number of farms in most Alpine regions has decreased by more than 50%, and the share of the population employed in agriculture has decreased from about 70% to less than 5% [34], while employment has strongly increased in the secondary, and later in the tertiary, sectors [34,65]. Due to unfavorable growing conditions, such as short growing seasons, steep slopes, and small property parcels, which necessitate expensive management practices, while having low productivity, mountain farming cannot compete in national and international markets [31,32,59]. Today, many farmers generate their main income outside their own farm, e.g., in business parks, industrial facilities, shopping centers, and tourism, and the share of part-time farmers is about 70% in the Alps [61,66]. Therefore, the agricultural area decreased on average by about 20%, and in some areas up to even 70% [23]. Our results indicate that this trend is still ongoing. The abandoned grassland areas are currently subject to a natural succession process towards site-typical climax vegetation (forest up to the natural timberline, with dwarf shrubs and alpine grassland above) with impacts on ecosystem structure and processes [67]. Our results also indicate that land use has been intensified in favorable locations, mainly through conversion to permanent crops or transformation of agroecosystems into urban or suburban areas. The extent to which these trends will continue depends not only on socio-economic drivers, but increasingly also on climate change. At high altitudes, climate warming will lead to a rise in the timberline, from 300 m (at +2◦K) to 800 m (at +5◦K), resulting in a decrease in alpine grassland [67]. However, the temperature increase will cause only a marginal expansion of forested area in 84.3% of Alpine municipalities, because they do not have areas in the alpine and nival belts [68]. Although climate change does have impacts on land use below 2000 m a.s.l., economic impacts override climate effects [59,69]. As a result of temperature increase and regionally lower precipitation, agricultural use will shift from grassland toward arable farming and permanent crops at lower elevations, whereas grassland farming will intensify at higher elevations [12,59].

Our results indicate that these transformations of agricultural use can jeopardize ES provision and may simultaneously aggravate associated disservices, such as increased leaching of soil nutrients or pests [6,70]. Many ES have declined in recent decades on land formerly used for agriculture, due to the intensification of use or urban sprawl, resulting in LULC types that produce fewer regulating and cultural ES, in particular (see also [25,50]). In addition, some provisioning ES, such as food and fodder production, decreased. In contrast, if forest growth occurs on formerly used grassland (above all in the Italian and Slovenian Alpine regions), timber production will increase, but provision of drinking water (i.e., streamflow) could decrease [71].

#### *4.2. Implications for Management and Decision-Making*

Our results show that ES values are reduced in most of the selected regions, but with different expressions when divided into provisioning, regulating, and cultural ES. This suggests that increases in ES value can be achieved through targeted regional planning, which also conserves landscape and species diversity, as well as powerful bundles of ES [72]. Moreover, abandoned land can contribute to sustainable land use transitions, providing opportunities to nature-based solutions based on biodiversity, cultural, and regulating services [73]. For management and decision-making, a respective framing must be set to comprehensively evaluate the impacts of agricultural strategies (i.e., on environment and economy). Here, the ES concept should be better integrated into existing frameworks such as the sustainable rural livelihood framework [65]. Moreover, a stronger focus on transdisciplinary research, including the development of adaptive pathways would enable

stakeholders to translate ES changes into a tangible local or regional agricultural strategy [74]. To highlight the interdependence of different economic sectors and the need for collective action at the local/regional level, to successfully tackle future challenges, the resilience of ES needs to be addressed, in an ecosystem-based approach, in order to duly incorporate the steadily increasing knowledge of changing ecosystem functions and ecosystem processes due to climate change [57,75]. This requires a clear commitment to basic research in the field of global change and the use of promising scientific approaches, especially in topographically complex areas, in order to have results available quickly at the landscape level [76,77].

To complete the picture, an appropriate framework must consider the historical development of agricultural strategies, and socio-economic and landscape developments, which means that 'history' must be part of future strategies. Results such as those shown in our study can form the basis and at the same time the starting point for future development paths, which are also increasingly taken up scientifically in landscape ecology, e.g., [40]. However, shifting to more resilient pathways, i.e., developing innovative and adaptive pathways that can mitigate the negative effects of global change on ES [78], can pose significant challenges, especially if land use decisions are predominantly based on agricultural market values. Farmers and decision makers seem to be 'locked-in' to their production-oriented view [40], disregarding the importance of land-use change in promoting other values such as greenhouse gas emissions and sequestration or recreational use and biodiversity [72].

#### *4.3. Methodological Considerations*

In this study, we applied a simple approach for mapping and quantifying ES values based on LULC maps, which is often applied to generate comprehensive information suitable for decision-making, because it sufficiently accounts for underlying mechanisms and directly illustrates possible impacts from LULC changes [79,80]. While this approach is easily replicable, the results contain some uncertainties that need to be considered. One issue regards the LULC types that were used to map ES values, as we differentiated only four types of agricultural use. These are linked to different levels of fertilizer use and have distinct ecological functions and differing species composition [50]. Further levels of fertilization of grassland or differently stocked pastures [12,13], as well as specific types of annual and permanent crops [55,81] could not be distinguished, due to lacking spatial information at a cross-national level. A further refinement could also include a distinction between conventional and organic farming systems for annual crops or permanent cultures [9–11].

There are also limitations with regard to the underlying databases. An updated version of the classification of the eight Alpine regions was not yet available and the classification of some municipalities may have changed, as socio-economic indicators, especially, are less stable than environmental conditions. This may have greater effects on municipalities at low to medium elevations compared to municipalities located higher [82]. Nevertheless, future studies should reclassify the Alpine regions using recent data, which would be particularly important when predicting future agricultural landscapes. Another uncertainty is related to the LULC maps, which originates from methodological issues during the interpretation of different remote sensing data over time, for generating the CLC [83]. To reduce mapping uncertainties, we used the newer versions of CLC. However, in this relatively short time period, only the immediately visible changes from an intensification of use are reflected, while long-term effects such as forest regrowth on abandoned grassland can only be captured over longer monitoring periods [36,67]. Over such short periods as in our study, only transitional stages to forest (e.g., heathlands, transitional woodland, shrub) could be considered. In future studies, the results may be improved by differentiating ES values between young and mature forests. Basing our analysis on earlier time steps with a greater extent of agriculturally used land would have revealed greater transformations of agricultural landscapes and related impacts on ES [36,77,84].

Furthermore, it has to be noted that our results represent the potential ES supply weighted by socio-cultural preferences, that is, the capacity of ecosystems to provide ES independently of their actual use [80]. However, many studies indicated spatial mismatches between ES supply and ES demand, i.e., the demand exceeds the supply at the local or regional level, requiring the transfer of goods or the movement of people [19,52,85,86]. Such dynamics need to be taken into account in the development of sustainable management strategies, and our results should, therefore, be complemented with spatial information on ES demand [20,24].

#### **5. Conclusions**

By applying the concept of ES, the consequences for society can be assessed in a comprehensive way, highlighting both the direct impacts on agricultural production and the associated effects on regulating and cultural ES. Our results reveal that the agricultural area in the Alpine region is under massive pressure, as up to 30% of agricultural land in some regions has been abandoned or converted to other uses within the last two decades, despite the efforts made within the framework of the Common Agriculture Policy (CAP) of the European Union (EU). Consequently, ES values mostly declined between 2000 and 2018, especially, regulating and cultural ES, while some forest-related provisioning ES have increased. Our results also indicate that LULC change rates and, hence, changes in ES greatly differed across regions with different social-ecological characteristics. The smallest changes occurred in the traditional agricultural region, while rural retreats and forgotten rural areas were affected by the largest changes.

Such quantitative and spatially explicit information on impacts from the transformation of agricultural land can be used as an information basis for developing sustainable management strategies and for evaluating underlying policies such the CAP. The frequent abandonment of mountain grassland, providing an above-average number of ES, also emphasizes the importance of the Green Deal in the EU, which should be an impulse for an agricultural and food transition. The Green Deal's target of 25% ecologically valuable farmland in agriculture is one of the central and most important targets, and, therefore, particular attention should be paid to the maintenance of mountain grassland in the European Alps. Finally, to support decision-making in adopting tangible local or regional strategies that can maintain cultural landscapes and multiple ES, greater efforts should be put into transdisciplinary research, allowing for the development of adaptive pathways, depending on the historical development of agricultural use, and socio-economic and landscape developments.

**Supplementary Materials:** The following are available online at https://www.mdpi.com/article/ 10.3390/land11010049/s1, Table S1: Indicators selected for cluster analysis; Table S2: LULC types aggregated from CLC classes; Table S3: ES values (ES supply weighted with socio-cultural preferences) for different LULC types.

**Author Contributions:** Conceptualization, U.S. and E.T.; methodology, U.S. and E.T.; formal analysis, U.S. and E.T.; writing—original draft preparation, U.S., E.T., G.L. and U.T. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research received no external funding.

**Institutional Review Board Statement:** Not applicable.

**Informed Consent Statement:** Not applicable.

**Data Availability Statement:** All data used for analysis are publicly available and data sources are indicated. Further information is included in the Supplementary Material.

**Conflicts of Interest:** The authors declare no conflict of interest.

**Figure A1.** Distribution of ES values in 2018 across the European Alps. Only agricultural LULC types (crop cultivation, permanent culture, fertilized grassland, unfertilized grassland) that were present in 2000 were considered. Thus, these maps illustrate the changes in agricultural land to other LULC types (other agricultural types, abandoned land, forest, settlement areas). Authors own elaboration.

#### **References**


### *Article* **Emergent Properties of Land Systems: Nonlinear Dynamics of Scottish Farming Systems from 1867 to 2020**

**Richard Aspinall 1,\*, Michele Staiano <sup>2</sup> and Diane Pearson <sup>3</sup>**


**Abstract:** Dynamics of arable and pastoral farming systems in Scotland over the period 1867–2020 are documented using time series analysis methods, including for nonlinear dynamical systems. Results show arable and pastoral farming, at a national scale, are dynamic over a range of timescales, with medium- and short-term dynamics associated with endogenous system forces and exogenous factors, respectively. Medium-term dynamics provide evidence of endogenous systems-level feedbacks between farming sectors responding to change in world and national cereal prices as an economic driver, and act to dampen impacts of exogenous shocks and events (weather, disease). Regime shifts are identified in national cereal prices. Results show change and dynamics as emergent properties of system interactions. Changes in dynamics and strength of endogenous dampening over the duration of the study are associated with dynamical changes from major governmental policy decisions that altered the boundary conditions for interdependencies of arable and pastoral farming.

**Keywords:** land system dynamics; emergent properties; time series analysis; nonlinear dynamics; Recurrence Plots; Scotland

#### **1. Introduction**

Much research in land systems science has focused on process–response (cause and effect) relationships of changes in land use and land cover with a variety of drivers of change as causal factors [1–4]. Many of these studies have focused on dynamics defined by change resulting from either land conversion (changes in type of cover and/or use) or land modification (land use intensification, land degradation, land abandonment) using snapshots in time and simple differencing between dates to elucidate patterns in observed changes. Observed changes are hypothesized to be caused by a variety of drivers and processes. In a whole systems context, however, dynamics are the set of behaviours exhibited as a result of the interactions of the elements that define land as a system [5]. The nature of land systems as complex systems with dynamic and emergent behaviours is recognized in, for example, land use transitions and the causal roles of endogenous forces and exogenous factors [6] and in calls for identification of regime shifts [7,8]. To date, however, despite recognizing land systems as inherently coupled systems, relatively few studies of land systems beyond agent-based models [9] have attempted to interpret dynamics as a function of the structures, interactions, and feedback mechanisms that define land as a coupled system. Additionally, as Turner and colleagues note, despite wide recognition of land as an exemplar of coupled human-environment systems [10], these explanations typically invoke one of the human or environment subsystem explanations in more detail, and few are rooted in the interactions of human and environment systems [1].

There are three main limitations on the current description and understanding of dynamics in land systems. First, the short time spans of studies provide a limited set

**Citation:** Aspinall, R.; Staiano, M.; Pearson, D. Emergent Properties of Land Systems: Nonlinear Dynamics of Scottish Farming Systems from 1867 to 2020. *Land* **2021**, *10*, 1172. https://doi.org/10.3390/land10111172

Academic Editor: Marta Debolini

Received: 20 September 2021 Accepted: 28 October 2021 Published: 1 November 2021

**Publisher's Note:** MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

**Copyright:** © 2021 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ 4.0/).

of system states recording change, which can allow changes to be mis-characterized [8]. Second, drivers are treated as constants over the short time span of interest, with little attempt to describe change in drivers over time or feedbacks from the system to the drivers; this can preclude identification of regime shifts and other dynamic system responses, particularly if change occurs as a punctuated equilibrium process [7,8] and may lead to misidentification of mechanisms generating change [11]. Third, in the context of whole systems approaches to land, land systems are not only dynamic, but also dynamical, in that their state can change over time even in the absence of changes in use or cover. This has implications for studies of process–response relationships in land change, since change may be a result of dynamical responses that are endogenous to the land system, in addition to being caused by drivers and other exogenous forcing variables.

The dynamic and dynamical nature of land systems are central to understanding land as a complex, coupled human-environment system. Land conversion and modification are undoubtedly important, and central to the scientific and applied needs for understanding land system change and its impacts [12–14]; understanding dynamics and dynamical changes in land systems is essential for describing functional behaviours of the whole land system. In dynamic and dynamical systems with many feedbacks and interactions the linear and sequential distinction between cause and effect becomes weak since each variable is both a cause and an effect. In this context, explanation of dynamics is based on system functioning via interaction of structures through uncovering endogenous forces and measurement of system-level responses to exogenous factors. This is related to the problem of equifinality, in which there are multiple plausible explanations for an outcome, a phenomenon well known in environmental science [15], and with an importance recognized for policy advice and developing models for socio-ecological systems [9].

Description of the dynamics of a system requires data that describe the structures, funds, flows (inputs, outputs, changes in funds) over time, as well as frameworks that organize the funds, flows and feedbacks into a system, and mathematical and other kinds of models that encapsulate functional dynamics of the system. Erb et al., and Kuemmerle et al., describe potential inputs, outputs, and structures for use to describe land intensity [16,17], responding to knowledge gaps that limit understanding and characterization of dynamics and patterns of land use intensity, but their conceptual framing is limited in description of the structure of the land system itself, and no attempt is made to incorporate feedbacks or land system funds beyond the biophysical structures they identify. Rahim et al., (2017) describe a causal loop model for analysis of supply and demand in Malaysian rice production as a complex system but have yet to quantify the model [18]. Elements defining land systems include state descriptors and system drivers, such as the structures and funds that comprise the system, but also the interactions of these elements, associated with connections between structures and funds through flows and feedbacks.

The aim of this study is to analyse long-, medium- and short-term patterns in land system changes, to understand the dynamics associated with interaction of systems at these different scales. Few studies have attempted formally to characterize multi-scale dynamics or analyse long time-series of land system data. The study also demonstrates some of the techniques available for this type of analysis, using a case study for Scotland.

This paper analyses dynamics of farming systems in Scotland, using data describing farming at a national (Scotland-wide) scale from the last third of the 19th century until 2020. The record of farming in Scotland over this period is well known from studies contemporary with the changes observed (see, for example, annual publications of the Transactions of the Royal Highland and Agricultural Society (1790–1969) and Scottish Agricultural Economics (1950–1960), and reviews and audits of the history [19–25]. The difference here is that the analysis develops from the perspective of farming as a land system and spans the full period from 1867 to 2020 within a single quantitative analysis. Time series analysis, including methods for nonlinear dynamical systems, is used (i) to examine dynamics of the system over the full timespan and (ii) to characterize internal feedbacks and coupling of farming as a system at the national scale. These analyses reveal

some system characteristics and behaviours associated with the evolution of the system itself. The results also identify regime shifts using analytical and graphical tools for study of nonlinear dynamical systems.

Analysis is based on the contention that the time series of data recording the history of land use for farming in Scotland from 1867 to 2020 contain a record of the impacts of all long-, medium-, and short-term dynamics associated with both endogenous system forces and exogenous factors that have influenced the land system. Just as spatial patterns embed all the processes, from many spatial and temporal scales, that are involved in the production of landscape patterns (Dobson, 1990, 1992), so the temporal record of land systems contains an embedded record of both the effects of processes acting over long-, medium- and short- timescales and system responses. Because of this, the temporal scale at which a land system is studied should be made explicit, as the factors needed for explanation of changes and dynamics will vary with the time scale of interest. The case study shows that arable and pastoral farming, at a national scale, are dynamic over a range of timescales, but that throughout much of the timespan of the study the system has maintained a pattern of changes consistent with endogenous systems-level feedbacks between sectors that act to dampen the impacts of exogenous drivers. Changes in these system dynamics over the timespan are associated with policy changes that altered the interaction of arable and pastoral farming.

The rest of this paper is structured as follows:

	- (a) reports the results of analysis of time series data to identify trends, cycles, and random elements at long-, medium-, and short timescales. The shortterm component in the data is treated mathematically and operationally as statistical noise, but in practice reflects the impacts of real exogenous shocks, and other perturbations at specific times during the period of interest. This analysis shows the capacity of time series analyses to reveal the variety of long-, medium-, and short-term patterns recorded within the data, and ways in which system variables interact when viewed over various time spans.
	- (b) analyses time series data for cereal prices, area planted with cereals, and number of sheep using methods from nonlinear dynamics. This analysis is based on understanding the interactions of arable and pastoral farming at a national scale in Scotland over the 19th and 20th Century. Arable and pastoral farming typically are treated as separate land uses and receive separate economic treatment as relatively distinct farming sectors in contemporary studies; this reflects increasing specialization in farming associated with intensification and modernization [26], and land cover and land management differences. However, this separation has not always been the case, and the interaction of pastoral and arable farming has long been widely recognized [21,27–29].

#### **2. Terminology, Assumptions, and Organization: Pre-Analytical Definitions**

In the discussion that follows, land system implicitly refers to a coupled humanenvironment system. The description of the system and definition of system elements are central to analysis; this forms a necessary and fundamental pre-analytical stage for subsequent data collection and analysis. An underpinning conceptual model for land systems has been described elsewhere [5], using funds and flows to define the system structures and their interactions. This conceptual representation of the land system describes a series of sub-systems that are associated with both driving factors, that operate as system processes, and different types of capital (human, social, financial, physical/manufactured, and natural). A suite of human and social factors that influence individual and group decisions and decision-making is also included within a decision-making subsystem. Funds are linked by flows, as fluxes and changes in funds. The conceptual model in Aspinall and Staiano (2017) does not quantify the funds and flows, although it does indicate some of the time scales over which the fluxes and changes in the system elements operate, from days, months, and seasons to years and decades, and longer. The model has been used to underpin an accounting approach for analysis of supply of provisioning services and the dynamics of agricultural land use in Scotland between 1940 and 2016 [26].

The timespan considered is 154 years (1867 to 2020, inclusive). The time step is 1 year. The geographic unit used is the aggregate national land use in arable (cereals) and pastoral (sheep) farming in Scotland. In the example here we are primarily interested in exploring the nature of change in the land system and in the ways in which structural measures of the system as well as drivers have changed over the long-, medium-, and short-term, using data for the period from the last third of the 19th Century to 2020. The long run and annual time step allow us to measure long-, medium-, and short-term patterns within these data.

Hierarchy theory is helpful in conceptual organization of hypotheses about scale (Wu, 2013), including the interaction of different time scales, and the use of time series analyses. Allen and Starr (1982) define hierarchies as a process-oriented framework, and Allen (2009) lists some general principles for ordering levels in ecological hierarchies. These include:


This hierarchical organization of process embeds and defines relationships between processes at different levels, based on timescale of processes from fast to slow (short to long), with dynamics of slower processes at higher levels appearing as a constant at lower levels, and dynamics of faster processes at lower levels appearing as noise at higher levels. Hierarchy theory provides, therefore, a coherent conceptual architecture for addressing complexity, ordering levels by rate of processes, and defining coupling of system components across and within timescales such that they can be decomposed for description, analysis, and understanding (Wu, 2013). Nonlinear dynamics also offers potential, particularly in the interaction of fast and slow processes [30] and understanding the consequences of managing resources based on one over the other [31], leading to fragility in system resilience [32]. A hierarchical structure of process dynamics from fast to slow is embedded in the conceptualization and analysis of the system dynamics for farming used to inform interpretation.

#### **3. Methods for Addressing Dynamics**

Dynamics are changes or motion in systems that reflect the nature and interactions of system elements, including system states, feedbacks, and evolution. As such, dynamics are characterized in temporal changes in system drivers and states and in the operation of interactions between system elements. Techniques from time series analysis and for analysis of nonlinear dynamical systems are used to describe, extract, and understand observed dynamical behaviours and patterns from noisy time series. This section outlines the methods used in the context of some of the time-varying characteristics of system funds and drivers; the methods are then used to describe and understand the dynamics of arable and pastoral system behaviours.

Dynamics are examined using two sets of methods:


#### *3.1. Time Series Analysis*

Time series analysis is described in a number of standard texts [33,34] and implemented in numerous statistical and mathematical packages. Representation of change in the time domain is straightforward, using plots of data against time. Lag plots and decomposition of time series are used to identify long-, medium-, and short-term patterns in the time series, specifically identifying long-, medium-, and short-term patterns using detrending, smoothing, and calculation of residuals respectively. Least-squares regression is used to model exponential growth and other long-term changes in data over the complete timespan of the data. Deviations from these trends are modelled with smoothing splines to describe cycles over medium-term time scales. The residuals from the trends and cycles describe short-term variation. Results are reported both as absolute values and, for comparison between variables, as normalized values using z scores. The separate patterns and values of long-, medium-, and short-term components (trends, cycles, and residuals) for multiple variables can be analysed further to assess possible influences and feedbacks between variables at different timescales.

To identify and describe trends and cycles in time series of prices, and to link prices in Scotland to global prices, we use the Christiano-Fitzgerald Filter, a bandpass filter designed to identify patterns in data that lie within a specific range of frequencies [35,36]. The Christiano-Fitzgerald Filter has been used to identify long-term trends, cycles, and boom/bust episodes for world price data for commodities [37]. The filter characterizes time series as the sum of periodic functions, using a bandpass to accommodate trends without restrictions on the distribution of the underlying data. The method allows filtered series to be extracted for the duration of the full span of time in the data, without discarding data from the beginning or end of the series, as observations from the beginning of the period can be filtered with future values and from the end with past values. In the use here, the filter identifies cycles in prices data, allowing comparison of data for Scotland with world prices for commodities.

#### *3.2. Recurrence Plots and Recurrence Quantification Analysis*

Further analysis of the time series data is carried out using Recurrence Plots (RP) and Recurrence Quantification Analysis (RQA). RP and RQA are nonlinear dynamics methods for analysis and visualization of time series data [38,39]. Analysis is based on phase space reconstruction [40,41], a method for discovery of deterministic structure present in real-world dynamical systems using time series data of a single variable [42–44]. The approaches can be applied to coupled variables through cross and joint recurrence plots [45,46]. Detailed descriptions of the approach can be found in literature from physics and mathematics [38,45,47] and complexity science [39]; and examples of applications found in a variety of disciplines including economics [48,49], ecology [50–52], psychology [53,54], epidemiology [55], atmospheric science [56–58], and geosciences [59–62].

Using time series of values for structural or state variables as basic building blocks, we define *X*, a vector set of elements describing system structure and fund at a series of discrete time intervals, to represent the land system. Change in elements of this set over time (*t*) formally can be represented with a standard equation describing change over time:

$$\frac{dX}{dt} = f(X) \tag{1}$$

Continuous and discrete representations of change are unified by time-scale formulation [63].

Data are analysed by plotting values of the time series for *X* against lagged values of *X*, a standard procedure for time series that reveals changes from time (*t* − *k*) to time (*t*) over the period of the lag (*k*), and that graphically identifies the patterns of changes. We use this method to identify annual changes of magnitude greater or less than 2.0 SDs from the probability density function of all observed differences, using a lag of 1 year to represent annual decisions in the farming calendar over time.

Specializing Equation (1) to derive the matrix of first order time derivatives of *X* for all lags across the discrete time series defines a new space and a Jacobian matrix:

$$\mathbf{J} = \{ \mathbf{x}(t), \; \mathbf{x}(t - \Delta t), \; \mathbf{x}(t - 2\Delta t), \dots, \mathbf{x}(t - n\Delta t) \}\tag{2}$$

This matrix describes vectors of delay space coordinates that estimate the original phase space generating the dynamics of *X* [40,41]. The eigenvalues that can be calculated from the Jacobian matrix are local Lyapunov exponents, used in diagnostic analysis of chaotic systems [64], including in geomorphology [65] and ecology [51].

In practice, the matrix of delay space coordinates is calculated for time lags up to the embedding dimension and each column is a vector of coordinates <sup>→</sup> *xt*

$$\overrightarrow{\mathbf{x}}\_{l} = \{ \mathbf{x}(t), \; \mathbf{x}(t-\Delta t), \; \mathbf{x}(t-2\Delta t), \dots, \mathbf{x}(t-(d\_{\mathbf{c}}-1)\Delta t) \}\tag{3}$$

where:

Δ*<sup>t</sup>* is the time delay or lag between data

*dε* is the embedding dimension or dimension of the space required to recover the dynamics.

The embedding dimension is estimated for a time series using the method proposed by Cao [66].

The delay space matrix is a representation of the phase space and used for phase space reconstruction. Recurrence plots [38] and Recurrence Quantification Analysis [47] are used to evaluate the dynamics of the time series from the delay space coordinates. These methods are robust, RP and RQA being independent of limiting constraints, such as data set size, non-stationarity, and assumptions about underlying statistical distributions of data. RP and RQA can also identify thresholds in datasets, and have been proposed as a nonlinear time series analysis method for detection of environmental thresholds [50]. In the context of land systems, RP and RQA offer potential for both characterizing and identifying complex dynamics and identification of thresholds and regime shifts.

A Recurrence Plot is a graphical tool for interpretation of delay space. The plot is based on computation of a matrix of distances R between the vectors of reconstructed points in the delay space, identifying when transitions in the delay space revisit the same value:

$$\mathbb{R}\_{i\dot{j}} = \left\{ \begin{array}{c} 1 \ \stackrel{\rightarrow}{\mathbf{x}}\_{i} \approx \stackrel{\rightarrow}{\mathbf{x}}\_{\dot{j}} \\ 0 \ \stackrel{\rightarrow}{\mathbf{x}}\_{i} \approx \stackrel{\rightarrow}{\mathbf{x}}\_{\dot{j}} \end{array} \middle| \begin{array}{c} i,j = 1, \ldots, N \end{array} \right. \tag{4}$$

where *N* is the number of considered states and <sup>→</sup> *xi* <sup>≈</sup> <sup>→</sup> *xj* means equivalence up to a distance r, a radius threshold identifying proximity in the delay space. The RP is hence sensitive to the value of r. A value that is too small will result in a sparse RP with little to no information. Similarly, a value that is too large will fill the RP, again providing little to no information. A number of criteria guide selection of r [45]:


To generate RPs for each of the variables in this study we set the initial value of r to the minimum of (a) 10% of the maximum value of the phase space diameter and (b) 5σ where σ is initially estimated from the standard deviation of the annual changes at lag *t* = 1 over the duration of the time series. We then iterate from this value of r to reduce the point density in the RP, examining changes in the RP as r changes. The value of r used here for each variable is the smallest that retains pattern in the RP.

The RP is a square, symmetrical plot of the *<sup>N</sup>* <sup>×</sup> *<sup>N</sup>* matrix <sup>R</sup>. In the analysis, *<sup>N</sup>* is the number of time points under study. Values from Equation (4) are plotted, 1 being coloured black, 0 being white. Black points highlight the recurrences of the dynamical system (defined by the radius r), the patterns in the RP giving insight into periodic structures and clustering properties within the data that do not show up in the original time series. The main diagonal is the identity line. RPs reveal structures in the data which can be single dots, diagonal, horizontal, and vertical lines, and blocks. Infrequent states are represented as isolated dots. Diagonal lines are the result of the system visiting the same region of state space several times. Horizontal and vertical lines represent periods when the system remains in the same state for a while; the lengths of lines represent the time the system is in the state. A threshold, or regime shift, will appear as a two (or more) separate square areas along the diagonal. White noise produces a uniformly distributed structure, and periodic oscillations produce a regular pattern within the RP. White bands are caused by abrupt changes in the dynamics and by extreme events, facilitating identification of extreme and rare events [45].

Analysis of the structures in the RP use methods from Recurrence Quantification Analysis [39,47] providing several measures indicative of the dynamics [45]. We use (i) recurrence rate, which is the percentage of points in the RP and indicates the probability that a specific state recurs; (ii) laminarity, which is the percentage of recurrence points that form vertical lines and is a measure of the presence of laminar phases in the system, and (iii) entropy, the Shannon entropy of the probability distribution of diagonal line lengths, indicative of the deterministic coupling of the system. RQA can be applied to a single RP for the full time series and to sliding windows traversing the time series. The RQA values for sliding windows of different sizes are computed to build up a picture of the dynamical properties over time between 1867 and 2020 across different time periods within the timespan of the data. A RP for the full time series is constructed to identify any regime shifts.

#### **4. Data**

The data used are time series of annual records describing the farming system in Scotland from 1867 to 2020. Specifically, data are


Additional data include time series for environment (weather data describing rainfall and temperature), and events (disease outbreaks, wars, introduction of legislation, trade agreements). These data are distilled to identify key events relevant to explanation of patterns and residuals in the analysis of the time series (see Figure 1).

A condition for use for all data is that each time series is required to be complete, with no missing data values within the timespan covered. Data describing structural elements of Scottish agriculture are collated as time series compiled from the Annual Agricultural (June) Censuses of Scotland which have been published over the last 154 years (Transactions of the Royal Highland and Agricultural Society of Scotland (1867–1910), Board of Agriculture for Scotland (1911–1927), Department of Agriculture for Scotland (1928–1958), Department of Agriculture and Fisheries for Scotland (1959–89), Scottish Office Agriculture and Fisheries Department (1990–1995), The Scottish Office Agriculture, Environment and Fisheries Department (1996–1999), Scottish Executive Rural Affairs Department (2000), Scottish Executive Environment and Rural Affairs Department (2001–2007), Scottish Government Rural and Environment Research and Analysis Directorate (2008–2010), Scottish Government Rural Payments and Inspections Directorate (2011), Scottish Government Environment and Forestry Directorate (2012–2013), Scottish Government Directorate for Environment and Forestry (2014–2017), Scottish Government Rural and Environmental Science & Analytical Services Division (2018–2020)—see, for example, [71–77].

The Agricultural (June) Census started in 1866, but many of the entries for 1866 are officially considered to be unrepresentative because of incomplete returns [23], and 1867 is used as the start for the time series. The total planted area in cereals is used to represent the arable sector; the number of sheep are used to represent the pastoral sector [22]. Together, wheat, barley, and oats are over 99% of the area of cereals grown in Scotland and have not been less than 97% in the period since 1867. In 2018, cereals contributed about 14% of the annual value of Scottish agriculture (about £430 million); sheep were a further 7% of the total value of agriculture (£236 m) [78].

Financial data on prices of cereals are from multiple sources. Wheat, barley, and oats prices for Scotland are from the annual reports on Agricultural Statistics (1912–1978), Economic Reports on Scottish Agriculture (1980–2020), Scottish Agricultural Economics (1950–1959), and records of prices from Fiars Markets around Scotland from the Transactions of the Royal Highland and Agricultural Society (1790–1969), themselves a continuation of a longer series of records from Fiars Markets from 1550–1780 [79]. All prices data are converted to pounds sterling per tonne (£/tonne), from a variety of source price and weight units (viz. Scots and English pounds, shillings, and pence (£/s/d), GB Pounds after decimalization in 1971, and weight (e.g., boll, bushel, cwt, ton, tonne) in use at the time of original data collection. The prices and their trends for the three cereals over time are similar. Barley price is used for price of cereals since barley is the cereal grown over the largest area since the mid-1960s; oats were the major cereal crop by area until the mid-1960s [22].

Potential links between these data as descriptors of structure and drivers of Scottish farming are derived from the literature describing farming in Scotland [21,22], the UK [29], and from the conceptual model of land systems described by Aspinall and Staiano (2017). We focus on cereal area (km2), numbers of sheep (millions), and prices of cereals (barley £/tonne). There are no single or sets of equations that relate these variables, since relationships between them would not only have to deal with their different metrics and scaling, but are also linked within a social and environmental system with limited potential for description with invariant or deterministic mathematical functions. Although marginal cost and other models for agriculture attempt to inform decisions e.g., [80,81], these are developed for specific times and circumstances, and there are no existing universal models describing the relationships between prices, crop areas, and livestock numbers that could be considered to apply over the timespan of our study. Even with this, however, some hypotheses about the general nature and direction of relationships can be formulated to describe the general process we examine, based on principles and market signals resulting from interactions of supply, demand, and price.

Traditionally, farming systems across the UK and its component countries have combined arable and pastoral activities within many of the different environments farmed [27]. Historically, grass was grown as part of crop and land use rotations in Scotland, helping rest and fertilize land prior to the next cycle of crop production [21]. Rotation grass grazing also serves as a low maintenance land use during periods when pressures reduce the capacity or potential for cereal farming [21,22]. Bowers and Cheshire compare sheep numbers with wheat price for England and Wales from 1893 to 1940, using five year means for the variables and a lag of three years for sheep numbers, revealing a negative linear relationship between wheat price and sheep numbers [29]. Our hypotheses about the cereal, grassland, and sheep farming system in Scotland is similar to Bowers and Cheshire's argument. An increase in price of a commodity produced by farming (barley price), itself linked to increased demand or insufficient supply (or both) for the commodity, might be expected to result in an increase in effort to produce that commodity (increased area for

cereals). In turn, interdependence between parts of the system means that this increase in production will be associated with a reduction in land for other uses and a decline in livestock (sheep). Conversely, as prices or demand fall, and supply increases, the area of cereals will decrease, and other land uses (sheep farming) will increase. Both cereals and sheep are, therefore, expected to show a correlation with barley price, and cereal area with sheep numbers. Additionally, trends in barley price in Scotland are expected to be aligned with global trends in prices.

In land system terms, this set of relationships and feedbacks links economic/financial, arable (cereal) and pastoral land uses, and livestock (sheep) components of the land system. Fortunately, even in the absence of mathematical models for the relationships and feedbacks, it is possible to interpret the patterns in time-series data against the expectations of these general hypotheses about the directions of relationships and feedbacks. In addition to the hypothesized relationships between prices, cereal area, and sheep numbers, which reflect endogenous relationships within the farming system, there are a variety of exogenous variables that can also be described with time series data. For some, such as plant and animal diseases, war, and legislation and regulation the variable is binary. For others, such as weather variables, the data provide a continuous measure that can be interpreted using time series analysis to separate trend, cycles, and extremes (as statistical 'noise'). These exogenous events present shocks to the system that must be accommodated, depending on the specific nature of the event, over time scales from short- to longer terms.

#### **5. Results**

Figure 1a shows plots of the price of barley (£/t), number of sheep (millions), and area of cereals (km2) for Scotland from 1867–2020. The plot is scaled with natural logarithms. The plots show that area of cereals and number of sheep, as funds, and barley price, as a driver, vary with superficially different patterns. As noted above, these data contain a record of changes at many temporal scales over the last century and a half.

Figure 1b shows some major legislative, trade, commodity price, weather, war, and disease events from 1867–2020. These provide not only boundary conditions for farming but also events that coincide with many of the more extreme values identified in the time series data. For example, government policy during and after the two world wars, and membership of the EEC/EU Common Agricultural Policy (CAP) provide system boundary conditions that shape agriculture in the short and medium terms. The 1947 Agriculture Act and associated support mechanisms (price guarantees, deficiency payments, marketing boards, investment in R&D etc.) set the context for farming from 1947 until the UK formerly joined the EEC CAP in 1973. Similarly, the influence of events such as wars, disease outbreaks, extreme weather, and financial crashes were short-term shocks that have clear signals in the short-term results ('noise') in the time series.

#### *5.1. National and World Crop Prices*

Long-, medium-, and short-term trends, cycles and patterns are identified in each of the cereal price datasets for Scotland, and in global prices data for cereals [37]. Prices for wheat, barley, and oats are shown in Figure 2a. Analysis of trend, cycles, and noise in data for prices of wheat, barley, and oats in Scotland shows that all have similar long-, medium-, and short-term trends. The long-term trend is exponential growth for each of the prices of wheat, barley, and oats (Figure 2a), the exponents in the equations being quite similar for the three (Table 1). Deviations from the long-term trend are modelled with a smoothing spline, revealing four main cycles over the 154-year period (Figure 2b). The residuals after removal of the long-term trend show high prices in 1918, 1942, the early 1980s, mid 1990s, 2007/8, 2012, and 2018, and low prices in 1972, 2005, 2009, and 2014–16 (Figure 2c). The concentration of high and low residuals in the post-1972 period is unsurprising given the increase in absolute values of prices due to inflation, but fluctuations are also latterly related to payments from CAP being in Euros, prices thus being subject to exchange rate variations in addition to inflation [82].

**Figure 1.** (**a**) Price of barley (£/t), number of sheep (millions), and area of cereals (km2) for Scotland from 1867–2020. The plot is scaled with natural logarithms. (**b**) selected major legislative, trade, commodity price, weather, war, and disease events from 1867–2020.

Applying the Christiano-Fitzgerald filter to cereal prices in Scotland and world cereal prices from datasets used by Jacks [37,83] shows that the medium-term cycles revealed in cereal prices in Scotland are synchronized with cycles in prices for grain crops (wheat, barley, corn, and rye) for world data (Figure 2d).

**Figure 2.** Long-, medium-, and short-term variation in prices for wheat, barley, and oats in Scotland (1867–2020). (**a**) data and long-term trends, (**b**) deviations from the long-term trend and smoothing spline model, (**c**) residuals after removal of the long- and medium-term trends. (**d**) comparison of cycles for prices of wheat, barley, and oats in Scotland with grain crops (wheat, barley, corn, and rye) in world data from Jacks [37,83]. Note: the prices shown in Figure 2a,b, and c are prices for the year. In Figure 2d the cycles comparing prices in Scotland with world prices are based on these prices adjusted for inflation and indexed to a specific year so that they match the indexing for prices for the world dataset used (see Jacks [83]). Price of barley in the year in question, not adjusted for inflation, is used in the analysis of price with other data describing the farming system, since this is the price data available for each individual year when decisions are being made within the farming system.


**Table 1.** Regression models for prices of cereals (1867–2020).

#### *5.2. Lag Plots and Recurrence Plots*

Lag plots and RPs for the three variables are shown in Figure 3. Lag plots of barley price (Figure 3a) show a decrease in price of more than 2 SDs in the year-on-year difference in 1920 and increases in 1919, 1940, 1942, 1973, and 2007. The 1920 decrease reflects adjustment after increased prices during the first world war [21]; the increases in 1940 and 1942 reflect UK government decisions to guarantee prices for farmers during the second world war. The 1973 increase is associated with both UK accession to CAP and the world oil price crisis [84], and the 2007 increase with the world food and financial crises [85]. The RP for barley price (Figure 3b) shows clear evidence of regime shifts, with areas of black along the diagonal of the plot and virtually no recurrence points outside those boxes. The regime shift in the 1970s is clear in the time series plot (Figure 1), but the RP shows there was a further shift starting in the 1950s during post war recovery and lasting into the 1970s. The white bars coinciding with the world wars indicate extreme variability in barley price; high variability since 1973 is also revealed in the absence of recurrence points.

The lag plot for area of cereals (Figure 3c) shows that 1919 was a decrease in area planted of greater than 2SDs of the long-term annual changes, while 1918, 1940, 1941, 1942, and 1993 were increases in area of more than 2SDs. The decline in 1919 represents a return to pre-war farming practices after the focus on cereals during the First War [19,21], offsetting the 1918 increase. The 1940–1942 increases represent the intense and sustained efforts to increase crop production during the Second World War [20]. The increase in area in 1993 coincided with the introduction of set-aside [86]. Set-aside was a policy to reduce the area of cereals, but payments under the scheme were based on the registered cultivated area. The RP for cereals shows long term cycles in the area, with variability during the wars. The RP also shows increased cycles in the period since 1973.

The lag plot for sheep (Figure 3e) shows that most year-on-year changes in the numbers of sheep are relatively small. Three years, 1941, 1947, and 2001, were decreases of greater than 2 SDs, while 1948 is an increase of more than 2 SDs from the previous year. Of these, 1941 represents government policy to reduce the sheep flock to allow crop production to be increased during the Second World War [21]; the number of sheep was reduced by over 1 million in 1941, and total sheep numbers were reduced by about 25% from pre-war levels during the years of the war [20]. February and March 1947 were extremely cold and snowy (see below), the timing additionally coinciding with lambing that led to high sheep mortality (almost 1 million sheep fewer in 1947 than in 1946). In 2001 an outbreak of foot-and-mouth led to 962,000 sheep being culled in Scotland to control the disease [87]. In 1948 over 700,000 sheep were added to the Scottish total through efforts to recover from the 1947 winter. The RP for sheep shows very clear evidence of cycles in the number of sheep, with regular pattern of recurrences spaces about 30 years apart, three cycles being evident since 1950. Numbers were more stable in the latter part of the 19th century.

**Figure 3.** Lag plots and Recurrence Plots for barely price, cereal area, and sheep numbers in Scotland (1867–2020). (**a**) lag plot and (**b**) RP for barley price, (**c**) lag plot, and (**d**) RP for cereal area, and (**e**) lag plot, and (**f**) RP for sheep numbers.

#### *5.3. Time Series Analysis and Recurrence Quantification Analysis*

Figures 4–6 show the time series analysis results and RQA results for barley price, area of cereals, and number of sheep respectively.

**Figure 4.** Time series analysis results and RQA results for barley price in Scotland (1867–2020): (**a**) data and long-term trend, (**b**) deviation from long-term trend and smoothing spline, (**c**) residuals from long- and medium-term trends, (**d**) recurrence rate, (**e**) laminarity, and (**f**) entropy from RQA using sliding windows of 10–30 years duration in 2-year increments.

> The time series analysis for barley price has been described above but is shown in Figure 4a–c for comparison with the results of the RQA for barley price. In the RQA for barley price (Figure 4d–f) the recurrence rate (Figure 4d), laminarity (Figure 4e), and entropy (Figure 4f) all show that barley price was relatively stable until the first world war, from the mid-1920s and through the 1930s, and from the mid-1950s to about 1970. The low values of recurrence rate, laminarity, and entropy since 1973 reflect increasing variability and volatility in price.

> The total area planted with cereals in Scotland has varied between 3900 km2 and 6000 km<sup>2</sup> over the period from 1867 to 2020, with major changes in both the cereals planted and yields. The long-term trend is an annual decline in area planted of 0.14%, accumulating to a total of about 19% over the 154-year period (Figure 5a). The yearly difference between annual data and the long-term trend ranges from −1000 to +1000 km2, and shows four cycles superimposed on the long-term trend, with greater areas planted in the 1870s and 1880s, during the two world wars, and again in the 1980s (Figure 5b). Negative deviations from the long-term trend are in the 1920s and 1930s, mid-1950s to mid-1960s, and mid-1990s and late 2000s. The residuals, after removing the long-term trend and medium-term cycles, are high in 1918 and 1942, and low in 1939, 1993 and 1994, and 2006 and 2007 (Figure 5c), similar to the results of the lag plot (Figure 3c). The recurrence rate (Figure 5d), laminarity (Figure 5e), and entropy (Figure 5f) show that cereal area changes gradually for most of the 154 years, although was more dynamic during the two wars, and also in the early-mid 1990s, coinciding with the onset of the policy and practice of set-aside.

**Figure 5.** Time series analysis results and RQA results for cereal area in Scotland (1867–2020): (**a**) data and long-term trend, (**b**) deviation from long-term trend and smoothing spline, (**c**) residuals from long- and medium-term trends, (**d**) recurrence rate, (**e**) laminarity, and (**f**) entropy from RQA using sliding windows of 10–30 years duration in 2-year increments.

**Figure 6.** Time series analysis results and RQA results for sheep numbers in Scotland (1867–2020): (**a**) data and long-term trend, (**b**) deviation from long-term trend and smoothing spline, (**c**) residuals from long- and medium-term trends, (**d**) recurrence rate, (**e**) laminarity, and (**f**) entropy from RQA using sliding windows of 10–30 years duration in 2-year increments.

Sheep numbers in Scotland have varied between 6 and 10 million over the period from 1867 to 2020. The long-term trend is of increasing numbers at a rate of 0.1% per year, accumulating to a total of about 17% over the 154-year period (Figure 6a). The yearly difference between annual data and the long-term trend ranges from about −1.5 million to +1.5 million and shows five cycles (Figure 6b) superimposed on the long-term trend, with maxima in the late 1890s, 1930s, 1960s, and late 1980s and early 1990s, and minima in the 1880s, 1919/20, late 1940s, 1970s, and 2010s. The residuals, after removing the long-term trend and the medium-term cycles show peaks in 1937–1939, 1950, 1998, and 1999, and lows in 1947 and 2001 (Figure 6c). RQA results for sheep numbers show that although the recurrence rate is low for much of the 154 years (Figure 6d), the laminarity shows only two periods of extreme change (Figure 6f), these being during the 1940s, corresponding to the second world war and subsequent high mortality of sheep in the cold spring of 1947 [21], and in the early 2000s, coinciding with the outbreak of foot and mouth disease that reduced sheep flocks [87].

#### *5.4. Dynamics from Interdependencies in the System*

Interdependencies among the three data series are assessed from correspondence between the time series of medium-term changes with long-term trends and short-term noise removed. The medium-term patterns of variation for barley price, cereal area, and sheep numbers, expressed as time series and as x-y plots in Figures 7–9; all variables are normalized with their mean and standard deviation to account for differences in scaling between the variables. The correlation coefficients r and percent r2 for the pairs of variables for 1867–1947, 1947–1972, and 1973 to 2020 are shown in Table 2. All coefficients except two (marked by n.s. in the table) are significant at *p* ≤ 0.001. The signs of the correlations correspond to expectations for associations between the variables.

The sequencing of cycles is of interest since this indicates their timings relative to one another and is indicative of the influences we posit in our general model (see above). Cycles for barley price and cereal area are synchronized and in phase until the late 1940s after which they become less synchronized (Figure 7a). This can also be seen in the x-y plot (Figure 7b), where data for 1867–1947 are tightly clustered along a line with positive slope, and data from 1960 onwards, and particularly from the early 1970s have a different trajectory in the x-y space. The change in trajectory in 1992, coinciding with the introduction of set-aside and lasting until 2012, is particularly evident.

The medium-term trends for cereal area and sheep number are also synchronized, but with a lag that places the peaks for sheep at the minima for cereal area, and vice versa for the period from 1867 to the early 1970s (Figure 8a). After the early 1970s this synchronization weakens (Figure 8a). The x-y plot of the medium-term trends (Figure 8b) shows the switch in emphasis to sheep from cereals during the 1920s and the agricultural depression of that period, the increase in area of cereals and decline in sheep numbers between the mid-1930s and particularly from 1939–1944. The general negative association between cereal area and sheep numbers from 1867–1972 contrasts with the positive association in the trajectory after 1973 when sheep and cereals became decoupled under CAP.

Medium-term trends and cycles in barley price and sheep numbers are also synchronized, although there is a lag between the two cycles (Figure 9a), as is expected from the associations already described (Figures 7 and 8). The x-y plot of these trends shows close association between 1867 and the 1950s, before the trajectory of the data changes to a peak for price and sheep numbers between 1985 and 1992 (Figure 9b). This reflects the decoupling of sheep numbers and cereal prices.

These associations are also apparent in the correlation coefficients (Table 2). Between 1867–1946 and 1947–1972, barley price has a positive correlation with cereal area and negative with sheep numbers (Table 2). After 1973, the correlation coefficient between cycles for cereal area and sheep number changes to +0.764, from negative correlations prior to 1973. The correlations between barley price and both cereal area and sheep numbers are not significant for the period 1973–2020. In summary, the different trajectories of each of the plots in Figures 7–9 for the 1973–2020 period compared with 1867–1946 and 1947–1972 are marked, the close associations between the variables in 1867–1946 and 1947–1972 being shown by a trace over time that clusters along a positive or negative line through the x-y space (Figures 7–9) and the 1973–2020 trace departing from these patterns.

**Figure 7.** Medium-term patterns of variation for barley price and cereal area in Scotland (1867–2020) as (**a**) time series and (**b**) phase plots. Note: the variables are normalized to account for differences in magnitude.

**Figure 8.** Medium-term patterns of variation for cereal area and sheep numbers in Scotland (1867–2020) as (**a**) time series and (**b**) phase plots. Note: the variables are normalized to account for differences in magnitude.

**Figure 9.** Medium-term patterns of variation for barley price and sheep numbers in Scotland (1867–2020) as (**a**) time series and (**b**) phase plots. Note: the variables are normalized to account for differences in magnitude.


**Table 2.** Correlation coefficients (upper right quadrant) and r<sup>2</sup> (lower left quadrant) between mediumterm cycles for 1867–1946, 1947–1972, and 1973–2020.

#### **6. Discussion**

#### *6.1. Dynamics of Scottish Farming Systems*

Trends and cycles over different timespans and timescales identified within the data using time series analysis, as well as RP and RQA, characterize long-, medium-, and shortterm dynamics of cereal and sheep farming and cereal prices. Irregular cycles are evident in each of barley price, cereal area, and sheep numbers, the cycles being synchronized with each other but with phase shifts. The period of these cycles is between 15 and 40 years. Cycles for cereal area and barley price are synchronized and in phase up to the early 1970s (Table 2); both barley price and cereal area are negatively correlated with sheep numbers until 1972 (Table 2), particularly under the policies that operated from 1947–1972. The changes in synchronization and correlations following 1973 reflect decoupling of arable and sheep farming sectors under the provisions of the EU CAP. The long-term trends and patterns of cycles, as well as the year-to-year variability superimposed on the long- and medium-term trends, for farming, reveal the multiscale nature of temporal variation in changes to farming systems. The RP and RQA also help to identify regime shifts. The RP (Figure 3b) and RQA (Figure 4) for Barley price shows clear evidence of regime shifts, with one regime over the period from 1867 to the late-1930s (interrupted by World War One), and two further shifts in about 1950 and 1970; since 1970 the price has been highly volatile. Regime shifts are not apparent for cereal area and sheep numbers (Figures 3, 5 and 6).

Results of analysis of changes in Scottish farming over a century and a half show the signal of endogenous system dynamics. Domestic cereal prices are linked to changes in world prices (Figure 2d), and to national and international policies and events (Figure 1b), but behind the influences of these exogenous factors, there is evidence from the period from 1867 to 1972 for the dampening influence of endogenous dynamics associated with the (loose) coupling of components of Scottish farming systems. From 1973–2020 system feedback and interaction at a national aggregate scale has been weakened as sheep and cereal farming have been decoupled. The dampening feedback provided resilience to Scottish farming as long-term trends and medium-term changes in the world and domestic economies, and short-term events influenced farming. The importance of system dynamics for description and explanation of changes in system funds, and the presence of longterm trends and medium-term cycles also challenges analysis of changes based on data that cover only a restricted timespan. The results show higher level interdependencies between arable and pastoral sectors, dependencies that have themselves changed during the course of the twentieth century as boundary conditions are changed by events, policies (e.g., the Agriculture Act of 1947, the UK's accession to the EEC/EU CAP in 1973), that are important for understanding both arable and pastoral farming, development of policy, and land management. The lessons from the period studied, despite much of it being historic remain important for strategic decisions about policy regarding farming, land management, and farming livelihoods in Scotland. The results show ways in which dynamic behaviours of farming systems have evolved as policy context has changed. The extent to which the dynamics have become decoupled and less resilient with modernization of farming raises concerns for land use in future especially as new policy is developed following the UK's departure from the EU and CAP in 2020.

#### *6.2. Land System Dynamics and Time Series Analyses*

The results and analyses also highlight the variety of ways in which exogenous drivers and endogenous interactions of state variables within the system can influence land system change and dynamics across these time scales. Medium-term trends, revealed as cycles in the data here, are particularly important in this analysis. It is important to note that cycles (in both drivers and funds of the system) are not cycles in a strict mathematical sense, and they are not required to be regular, to have fixed periods or magnitude of oscillation, or to be predictable [83]. Rather, they are medium-term patterns of deviations from long-term trends, with short-term noise filtered out. The focus on cycles is based on the expectation that they reflect behaviours that result from interaction of system factors over the mediumterm, and, as such, cycles are of particular interest in characterizing and understanding system dynamics. The interactions and feedbacks of system components over time result in a statistical tendency for cycles, as (irregular) waves, to be found in the data, with values increasing and decreasing as feedbacks propagate through the system (with characteristic, but variable, time scales). Regime shifts are apparent in the RP for barley price, with cycles in the RPs for cereal area and sheep number (Figure 3d,f) and by recurrence rate, laminarity and entropy in the RQA (Figures 4d–f, 5d–f and 6d–f), as well as in the medium-term patterns of the time series (Figures 4b, 5b and 6b). Interpreting these cycles provides insights into system changes (Figures 7–9). Long-term trends represent slow dynamics and secular changes. Short-term changes represent impacts of events, and year-to-year stochastic variability, as well as a range of uncertainties, including inherent uncertainty of environmental and social systems, measurement errors (statistical uncertainty), short-term decision-making (partial controllability of complex systems), and structural uncertainty (the inability to describe the system fully) [88].

The use of time series and nonlinear dynamical systems methods is guided by hypotheses about the nature of farming as a coupled land system integrating human- and environment- drivers through farmer choices and decisions, manifest at the aggregate national and regional scales. Results are informative on the nature of dynamics of farming systems, the relation between dynamics and both endogenous feedbacks and exogenous noise, the influence of different timescales in establishing explanations based on potential processes and drivers, and on the impacts of drivers at multiple scales from farm to international trade, finance, and legislation. Together, the methods reveal aspects of the dynamic nature of drivers that underpin land system change, evolution, and dynamics, as well as the specific nature of dynamics in land systems themselves. The examples also elucidate some fundamental principles and mechanisms for studying land systems as complex coupled human-environment systems; the approaches have application to study and explanation of both dynamics and change.

The short-, medium-, and long-term trends and process relationships embedded within time series' data offer potential for study of not only change in land systems, but also temporal and cross-scale dynamics in system function, leading to improved understanding of coupling between human and environment systems, evolution in land systems over time, and influence and response to changes in land system drivers. The analysis uses a long data series, necessary for identifying long- and medium-term patterns. A snapshot in time cannot reveal these dynamics, and consideration of too short a time span can lead to misinterpretation of change and dynamics, for example by focusing only on increase or decrease [25].

If a central tenet for study of land systems, as exemplars of coupled systems, is that they are dynamic systems because of the functional interactions between the human and environment subsystems, then the dynamics of both system drivers and dynamical behaviours of land systems themselves (based on the interactions and coupling of human and environment) is as much a part of land dynamics as changes in the structures of land systems. Dynamics of both land system structures and drivers are also necessarily embedded in the pervasive impacts of spatial, temporal, and organizational scales, and in both the hierarchical complexity and contingent history of land systems within societal and environmental change more generally. Even in the absence of major categorical conversion in type or intensity of change, land systems operate with complex dynamics, and they require to be understood as dynamical complex systems.

The variety of dynamics represented by the patterns in the data used in this case study emphasizes the need for explicit pre-analytical hypotheses to be constructed about relationships between land system dynamics and changes with potential trends and changes in system drivers. The results also emphasize why hypotheses need to be explicit about the time scale, or scales, of interest, since long-, medium-, and short- time scale patterns are contained within the observed data, and all may be relevant to understanding the variety of land system dynamics. If we accept that time series data represent and reflect all the processes from all scales involved in their formation, then these data potentially provide a source of insight into multi-scale consequences of the actions of drivers. In this context, time series analysis provides a set of mechanisms for distinguishing these temporal patterns at various time scales.

In summary, in systems terms the analysis of the historical record of changes in cereal area and price, and sheep numbers in Scotland reveal a complex pattern of interdependencies and coupling over time and at different scales, combining endogenous system dynamics with short-term variability associated with stochastic events, within a broader set of higher-level interdependencies and boundary conditions for the system. The long time-period of the study also shows that the embedded system dynamics can make farming relatively resilient to changes in policy, exogenous shocks (such as weather events or disease outbreaks), or regime changes and thresholds (as seen here in prices). The whole systems perspective is one that is seldom considered by short-term or sectoral approaches to farming. Although many of the results are not new, the long-term, whole systems perspective shows the evolution of land use in Scottish farming as a dynamic and dynamical system, hence demonstrating that this kind of approach is suitable for study and interpretation using a single analysis. The contribution of time series analysis and the tools of NLDS (RP, RQA) in land systems science is also evident. The long time series of data, and the impacts of historical contingency over the timespan of the study, combined with the coupling and complexity of system-level relationships, weakens the chances that steady-state latent structures would emerge by means of classical modelling. Instead, analysis with time series analysis and methods from NLDS allows exploration of discontinuities (if any) in the system dynamics, allowing abrupt changes and extreme values to be identified, that would be difficult or impossible to capture in steady-state global models. Time series analysis and NLDS also enable exploration of system dynamics at hierarchically nested time scales, moving beyond use of classical models in describing phenomena over short periods that are of little relevance over the longer duration of land use history, as captured in the data used in the case study. As such, the results offer a challenge to the land systems community to address timescales and dynamics explicitly, while demonstrating some approaches and methodologies for achieving this. Authors should discuss the results and how they can be interpreted in perspective of previous studies and of the working hypotheses. The findings and their implications should be discussed in the broadest context possible.

#### *6.3. Directions for Future Research*

Further research is needed into dynamics of land systems based on system interdependencies, interactions, and feedbacks. As noted in the Introduction, studies of system dynamics based in system structures and coupling are rare within land system science. Methods such as time series analysis, for analysis of non-linear dynamic systems, and models for exploring input-output within non-linear dynamic systems such as NARMAX [89], offer tools with potential for use by the land systems community. This study shows the importance of long time-series of data for capturing dynamics, the dynamical behaviours found for Scotland being evident over medium-term cycles of about 30 years duration. The consequences of policy changes for dynamic behaviours are also apparent from this case study, producing dynamical shifts in system coupling, and showing the importance of comparative studies across different socio-political, economic, and other contexts that provide the boundary conditions for land use decisions. Finally, this case study uses aggregate national data. The extent to which this is representative of the behaviours and experience of individual farm or other land use units requires further research.

#### **7. Conclusions**

Time series analysis, including methods from analysis of non-linear dynamical systems, are used to separate long-, medium- and short-term dynamics encapsulated within a historical record of land system states for farming in Scotland over the period from 1867 to 2020. The results show that cereal prices in Scotland follow similar trends and cycles to those shown in global prices, and that the dynamics of both the area under cereal cultivation and sheep numbers are linked to the dynamics of barley prices, as well as to each other, particularly for the period from 1867–1972. The relationships revealed in the medium-term trends are weaker since 1973 as prices, and cereal and sheep farming have become decoupled under modernization associated with policies in the EEC/EU CAP and as prices have become more volatile. These medium-term cycles in the data represent the endogenous dynamics of the farming system itself, operating within boundary conditions set by the policy environment. Short-term variability in the data reflect year-to-year variability associated with weather, disease, and other events.

Our results characterize dynamics from internal feedbacks and coupling of farming as a system at the national scale, reveal some system characteristics and behaviours associated with the dynamical evolution of farming as a system, and identify some regime shifts over the full 154-year timespan of the census. Specifically, the results reveal (i) consequences of several exogenous factors as events that had an impact on system states, (ii) show that arable and pastoral farming, at a national scale, are dynamically related over a range of timescales and coupled to global trends, and (iii) that throughout much of the timespan of the study the system has maintained a pattern of changes consistent with endogenous systems-level feedbacks between sectors that act to dampen the impacts of exogenous factors. Changes in system dynamics over the timespan are also associated with policy changes that altered the interaction of arable and pastoral farming.

The analysis is based on the contention that the time series of system states recording the history of land use contain an embedded record of the impacts of long-, medium-, and short-term dynamics associated with both endogenous system forces and exogenous factors that have influenced the land system. Because of this, both the underlying systems framework structuring the land system and the temporal scales at which a land system is studied should be made explicit, as the information needed for explanation of changes and dynamics will vary with the system structure and the time scales of interest. The use of time series analysis and methods from non-linear dynamics forces explicit attention to system structure, time scales, and the multi-scale behaviours of land systems. This demonstration of interdependencies between the prices, and arable and pastoral systems in Scotland shows that farming land use in Scotland has functioned as a complex system and was particularly resilient as a coupled arable-pastoral system prior to 1973, displaying characteristic behaviours of endogenous variables within a nonlinear dynamical system with noise-dampening feedbacks. The cases study illustrates a more general problem. Because of the dominance of studies of land conversion and modification, the prevalence of studies of short timespan [90], and the requirement for long time series of data to support time series and NLDS analyses, there are, correspondingly, still few exemplars or results of

analytical approaches applied to land system dynamics found in the land systems literature, beyond those based on change detection. More are needed. Further studies of land systems could usefully attempt to identify emergent properties and behaviours of land systems, developing analyses focusing on dynamics in long-term time-series data, complementing analyses based on spatial snapshots over short time spans.

**Author Contributions:** Conceptualization, R.A., M.S. and D.P.; Data curation, R.A.; Formal analysis, R.A. and M.S.; Investigation, R.A., M.S. and D.P.; Methodology, R.A., M.S. and D.P.; Software, R.A. and M.S.; Validation, R.A., M.S. and D.P.; Writing—original draft, R.A., M.S. and D.P.; Writing review & editing, R.A., M.S. and D.P. All authors have read and agreed to the published version of the manuscript.

**Funding:** R.A. acknowledges financial support from University of Naples for collaboration with the STAD group based at University of Naples Federico II. M.S. acknowledges financial support from the European Union's Horizon 2020 research and innovation programme under grant agreement No 689669. D.P. and R.A. acknowledge financial support from Massey University International Visitor Research Fund. This research received no specific grant from any funding agency in the public, commercial, or not-for-profit sectors.

**Acknowledgments:** This work reflects the authors' views only.

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **References**

