Van Vliet J., de Groot H.L.F., Rietveld P., Verburg P.H. Manifestations and underlying drivers of agricultural land use change in Europe. Van Vuuren D.P., Stehfest E., Gernaat D.E.H.J., Doelman J.C., van den Berg M., Harmsen M., de Boer H.S., Bouwman L.F., Daioglou V., Edelenbosch. A rapid process of change is taking place in European landscapes today (Estel et al., 2015; Pedroli et al. First of all a landscape inherited from a period when agriculture was the dominant function underlying. Is often a main driver of land use changes (Primdahl & Swaffield, 2010) (see e.g., Figure 3(a)). In land use science. Various meta-studies have been conducted, which synthesize deforestation and agricultural land use change processes, while other. Van Vliet, J., H.L.F. Rietveld, and P.H. Manifestations and underlying drivers of agricultural land use change in Europe.
- Manifestations And Underlying Drivers Of Agricultural Land Use Change In Europe
- Monish Jose
- Leibniz University Of Hanover
Correspondence to: Institute of Life Sciences, Scuola Superiore Sant’Anna, via Santa Cecilia 3, 5612 Pisa, Italy. E-mail: a Conference presentation: SIA XLIII Congress, Pisa, 2014. Abstract Since the 1960s, research has dealt with agricultural intensification (AI) as a solution to ensure global food security. Recently, sustainable intensification (SI) has increasingly been used to describe those agricultural and farming systems that ensure adequate ecosystem service provision.
Studies differ in terms of the application scales and methodologies, thus we aim to summarize the main findings from the literature on how AI and SI are assessed, from the farm to global levels. Our literature review is based on 7865 papers selected from the Web of Science database and analysed using CorText software. A further selection of 105 relevant papers was used for an in-depth full-text analysis on: i) farming systems studied; ii) related ecosystem services; iii) indicators of intensity; and iv) temporal and spatial scales of analysis. Through this two-step analysis we were able to highlight three main research gaps in the AI research indicators. Firstly, the farming systems analysed for assessing AI are often quite simplified or monoculture-oriented, and they do not take the diversity and complex organisation of farming systems into account. Secondly, these studies mainly focus on northern countries or developing countries, whereas there is a gap of knowledge in Mediterranean areas, which are the areas with a high complexity of farming systems and diversity in ecosystem services.
Finally, AI is mostly assessed through nitrogen inputs and economic yield, which are used the most both at very local and global levels. Intermediate regional or local levels, which are relevant for policy implementation and local planning, are often neglected. Introduction Since the escalation of agricultural commodity prices, agricultural intensification (AI) has increased its importance due to the debate on global food security (Buckwell, ). This debate started in the 1960s, when Boserup used intensification to explain higher levels of agricultural productivity associated with higher population densities in agriculture. Turner and Doolittle defined AI as the amount of output per unit area and per unit time. However, noting that output data were scarce, they developed a proxy scale of AI based on two input variables: frequency of cropping and the use of agricultural technology.
Since then, several authors have exploited this theory together with others, and the terminology has been modified over the years often becoming quite ambiguous. Likewise, some publications have attempted to clarify the difference between agricultural intensification and intensity (Shriar,; Kleijn et al.,; Dietrich et al., ) however they are still often used indifferently. We believe that there is no univocal definition for the different terms used, such as agricultural intensity, land use intensity (LUI) or AI, and therefore we begin by trying to differentiate between them. In general, agricultural intensity is defined as the ratio of inputs and outputs within an agricultural system, i.e., in terms of yield per land area and per input unit (Turner and Doolittle,; Shriar,; Herzog et al., ) or alternatively as the sum of different categories of input costs and the total usable agricultural area of the farm (Teillard et al., ).
Therefore either output-oriented (production) or input-oriented (utilisation) measures can be used to describe agricultural intensity. Many studies tackling the environmental impacts of agricultural intensity have focused on a single component, such as nitrogen input (Kleijn et al.,; Fumagalli et al.,; Temme and Verburg,; Overmars et al., ) or pesticides (Geiger et al.,; Jepson et al., ). Others have used proxy indicators of agricultural intensity, such as yield or profitability (Caraveli,; Stoate et al.,; Schneider et al.,; Kuemmerle et al.,; Niedertscheider and Erb, ) or the relative amount of arable fields (Schneider et al.,; Tuck et al., ).
In regional studies, land use intensity is referred to the area required to produce one unit of output or the yield per time and area unit (Lambin et al., ). In dynamic terms, we also found explicit definitions for the term AI or LUI changes ( i.e., intensification) as the resulting process of land use changes over time or the changes in yields and land productivity (Shriar, ). In fact, recently all approaches have been spatially-explicit and based on the analysis of the changes in land uses (Lambin et al., ).
Main results of these works confirmed that most of land use changes in Europe occurred along gradients of management intensity (Rounsevell et al.,; Ewert et al.,; Foley,; Rounsevell et al.,; Kleijn et al.,; Renting et al.,; Foley et al.,; Lambin and Meyfroidt,; Erb,; Rounsevell et al.,; Allan et al., ). Likewise, intensification is usually measured as a resulting process from land use changes ( e.g., abandonment, fragmentation or urban sprawl) and conversely, intensity and intensification affect land use changes through of different pathways as the land pressure in terms of profitability. In other words, AI can be related to land scarcity or to its high cost, where land is available, combined with the supply of ecosystem services (Byerlee et al., ). This has also triggered others assumptions, such as defining intensification as the replacement of heterogeneity in habitat structure, in time and space, with homogeneity (Benton et al., ). Tscharntke et al.
defined it as the conversion of complex natural ecosystems into simplified managed ecosystems with a high resource use and a generally higher input and output. In addition to defining AI or LUI, it is also necessary to create an application framework for the concept, as well as to analyse how it should be measured, and which indicators should be used and under what conditions. In fact, identifying the degree of intensification may also increase the knowledge of the impact of land use change on ecosystem services across different landscapes. Unlike a natural system, a system managed through intensification is able to produce abundant food, however reducing other ecosystem services (Foley, ), would thus be important in understanding how ecosystems are altered by agricultural intensification (Matson,; Snapp et al., ).
Indeed, there is a general agreement on the idea that AI reduces the quantity and quality of the services that ecosystems provide, including the loss of biodiversity (Allan et al.,; Egorov et al.,; Overmars et al.,; Tuck et al., ) and the water and soil quality (Foley,; Stoate et al.,; Snapp et al.,; Zhang et al.,; Tittonell and Giller,; Williams and Hedlund, ). Therefore an overall vision that includes all the indicators in the literature, thus increasing the empirical basis, is a pertinent and realistic tool to measure agricultural performance and monitor progress in order to produce detailed knowledge of the intensity of agricultural land use (Erb et al.,; Temme and Verburg,; Rousevell et al., ). In this study, AI is measured as a result of farming practices at any given time based on indicators and proxies. The review will therefore answer the following two research questions: What are the effects of AI and how are they measured? Which agricultural land uses and systems are linked to AI?
In order to compare the various ways of measuring the AI/LUI as well as the approaches and aims, a bibliometric analysis was carried out to place both these topics in the relevant scientific contexts ( e.g., land use science, food planning and ecosystem services) and to make an analysis on a farmland to global scale. A subsequent qualitative and statistical analysis on the full texts of selected papers highlighted the interactions among variables indirectly related to AI to identify major research gaps and recommendations. In order to systematically review the obtained papers’ database and highlight their main relevant concepts, we applied an analysis grid containing the following criteria: i) the main declared topic related to AI research; ii) the literature definition used for identifying AI concept; iii) and the methods applied to evaluate AI. Materials and methods This paper involves two main steps. Research papers were selected in both steps according to the PRISMA (preferred reporting items for systematic reviews and meta-analysis) flowchart (Liberati et al., ). In the first step, we provided an overview of how the concept of AI has been used in the literature. We reviewed international studies on AI from the first publications in 1975 until now, and from the international bibliographic database Web of Science (WoS).
The papers obtained were analysed using the bibliometric software CorText (Tancoigne et al., ). Software CorText is a platform dedicated to the cleaning and treatment of large textual corpuses with the aim of synthesising and analysing big data, whether structured or unstructured (IFRIS, ). In the second step, starting from the whole database and also considering different bibliographic databases ( Scopus, WoS and Google Scholar), we selected various relevant papers. The full texts of these papers were then analysed in depth in order to answer our research questions. Bibliometric analysis The bibliometric analysis was performed using the CorText software (IFRIS, ), which enabled us to upload data sets and run the different analytical process in order to perform lexical analysis and mapping the structure and the dynamics of the corpus. The common procedure is specified as follows: i) calculation of the frequency of occurrence of each term; ii) normalisation of these occurrence and co-occurrence measurements as proximity measurements to link the nodes (Tancoigne et al., ). CorText manager recommends choosing direct measures for heterogeneous network like chi-squared test, which only takes into account the raw co-occurrence number between two nodes.
The dataset is analysed through a lexical extraction of the title and abstract from the selected papers, which supply key information on the most co-common topics. Data collection was carried out using keywords including various synonyms and combinations of the concept.
A systematic search in the WoS database (main keywords: agricultural intensification OR land use intensity OR agricultural intensity), combined with others sustainability OR ecosystem services OR land use modelling ; Timespan: 1975-2014; Search language: English, yielded 7865 publications that were retrieved and imported as a corpus to the CorText Manager Software for data analysis. The resulting map based on a keywords analysis identified the most relevant terms, the dynamics across the links and nodes, the interactions between them and the distribution along with their weight.
Cluster analysis highlighted the aggregation of the most frequently used terms. Nodes and links between clusters captured the information flows between those aggregations (IFRIS, ). Selection of papers for the full-text analysis Publications were included in the selection if they provided enough evidence on the way of leading the AI within our context. Firstly, a filter was applied on approximately 300 publications by screening the title and abstract, and secondly by full text on around 160, thus yielding the final selection. The study was based on the concept that agricultural intensity is defined as the result of farming practices at any given time, and intensification is considered as the total process rather than one condition at any particular time.
We filtered the publications that met the following criteria: the relevant scientific disciplines (studies related to landscape agronomy, landscape planning and land use science); the presence of a quantitative approach (studies had to quantify the changes in agricultural land use or the indicator values); assessments of ecosystem services (studies had to quantify the land use intensity as a proxy for the assessment of ecosystem services delivered by agricultural systems); and the spatial level of the analysis (analyses beyond the field gate, so from the farm level to a global level). The selection for the full-text analysis resulted in a final corpus of 105 papers. Full-text analysis From our 105 papers sample, the items searched for were: i) the case study location and the spatial level of the analysis; ii) the methods applied and the presence of thresholds in the evaluation of the sustainability of AI/LUI. We also collected data on the ecosystem services, the farming systems or land uses in which AI was analysed and the indicators taken into consideration. The set of variables (ecosystem services, farming systems and indicators) were coded and consisted of binary and discrete variables within the database. These variables were calculated through frequency and descriptive analyses and points of significance for both ecosystem services and indicators at different scales.
Statistical methods enhanced the understanding of the data and were used for comparisons. We first hypothesised that there are several indicators driving the AI that depend on initial conditions, such as geographical context and farming system (Rounsevell et al.,; Geiger et al.,; Overmars et al., ). Based on the literature reviewed, we further hypothesised that AI needs to be measured with several indicators, since a single indicator is not able to assess AI ( e.g., nitrogen or pesticides applications (Herzog et al.,; Overmars et al., ); ratio of livestock numbers (computed as livestock units) (Rounsevell et al., ); net primary production (Erb et al.,; Krausmann et al., ).
We also wanted to define sustainability thresholds on the specific indicators under any context (Paracchini et al., ). On the other hand, we relied on the premise that an intensive system or cropland, which, in turn, is explicitly managed to maintain other ecosystem services, may be able to support a broader portfolio (Foley,; Bommarco et al., ). Bibliometric research The semantic maps represent the results of the network analysis, which identified the main topics around AI. They are composed of clusters, which represent the most different co-occurrence topics regarding AI.
The most frequent topics are those related to land use intensity (yellow cluster) and those on land use change and climate change (red cluster). In fact, according to, the most frequent are represented by the bigger dimension of the cluster because of the dimension of the circles is proportional to the number of retrievals and nodes are linked according to different types of proximity (measure: chi-squared test). The publications found in these clusters marginally tackle the question of the sustainability of the intensification of farming systems and the importance of analysing it from a regional approach. Other smaller clusters are more specific ( e.g., biodiversity or soil loss), but supply quantitative information on intensity indicators and intensification in terms of particular ecosystem services. Thus, an overview of the research field was obtained. The over-time analysis was of particular interest in revealing the different applications of AI over the last few decades knowing when the most frequent topics were used. Unlike the previous it was only analysed with the co-occurrence of the terms, in this analysis both years and terms were taken into consideration resulting a mapping of co-occurrence of the main common topics placed in terms of time (from 1975 to 2014).
In fact, between 1996 and 2007, research in this field was related to a greater proportion of topics such as changes in land use, the sustainable development of farming systems, and soil and water quality. Between 2010 and 2014, the density of nodes and links is less on issues related to land use, land cover or land use intensity as proxies to measure AI, whereas the main ecosystem services are linked to soil and water quality.
These are therefore the emerging research areas, and underline the permanence over time of the interest in water and soil resources. In depth analysis on full-text papers Of the 105 selected papers, 50 were reviews and 55 were research articles containing 225 case studies distributed across all continents. The European Union constituted 33% of these studies and applied case studies (77%), while the rest focused on Asia, the Philippines, Africa, South America and the USA. In the European papers, the case studies represented in the sample were highly distributed in Germany (8%), UK (8%) and Netherlands (7%) followed in a lower tendency by Mediterranean case studies like Spain (7%), France (6%) and Italy (6%). The low number of cases may be influenced by the lower degree of intensive farming in south-eastern Europe (Caraveli, ). The case studies are mostly located at continental, national or administrative levels. Very few apply to agro-ecoregions or natural regions and none of them concerned the Mediterranean basin at this level, whereas Mediterranean case studies are mainly targeted at national (Caraveli, ) or very local levels (Serra et al.,; Salvati and Tombolini, ).
Studying past dynamics to predict future trends of agricultural intensification Several analyses describing the global trends in production yields have been carried out on AI (Matson,; Cassman,; Rudel et al., ). There is increasing interest in quantifying agricultural inputs related to land productivity at a local scale, considering variables that can be easily retrieved from interviews with farmer (Herzog et al.,; Reidsma et al.,; Armengot et al.,; Dietrich et al.,; Gaudino et al., ). Many studies have explored the statistical relationships between crop yield and land use (Rudel et al.,; Ewers et al., ) and in the last few years there has been an increasing interest in predicting future changes in land-use intensity (Lambin et al., ). Using a scenario analysis, a dynamic and spatially explicit land-use change model was presented for the analysis of land use in small regions at a fine spatial resolution (Verburg et al.,; Lima et al., ). Modelling techniques have been developed for predicting future land use, thus supporting decision-making on important issues such as climate variability (Rounsevell et al.,; Rounsevell et al.,; Audsley et al., ). Several studies have used such models to spatially and explicitly quantify the trade-off between productivity, cropland use and intensity of land use (Ewers et al.,; Barretto, ). Recently, a quantitative method called Intensity Analysis has been used to characterise patterns of change at different levels and over several time intervals, and to explore the processes and drivers of change (Aldwaik and Pontius,; Huang et al., ).
Few indicator thresholds of agricultural intensification Each indicator performs differently according to the geographical/environmental/socio/economic contexts in which it is measured. Defining a threshold maximises the benefits obtained from a given parcel of land in a sustainable way over a long period of time (Paracchini et al., ). Thresholds also reveal the spatial variability, or whether a system is sustainable in any context. In our review sample there are few papers where thresholds are defined (13%), including thresholds always using the same indicators.
Some studies propose future scenarios with intensity thresholds based on nitrogen input, as was the case of Temme and Verburg. They defined low intensity as being from 0 to 100 kg-N input/ha, medium intensity up to 250 kg-N input/ha, and high intensity higher than 250 kg N-input/ha. Similarly nitrogen input rates were classified into three classes: low (150 kg/ha), based on the relevance for biodiversity (Overmars et al., ).
Others have defined thresholds based on spent inputs per ha, defining intensive systems as above 250 €/ha (Reidsma et al., ) or 350 €/ha of inputs (Audsley et al., ). From these works, able to assess whether an indicator increased or decreased over time or if it has different values at different locations, but considering our sample of case studies, we would not able to ascertain whether a farming system was sustainable or not. The relationship between agricultural intensity and ecosystem services In our sample, an average of three indicators is used to evaluate the intensity of a given agricultural area. Often, the total nitrogen input is used as an indicator to ensure a strong link to biodiversity (Temme and Verburg, ). This was also confirmed in our sample where a larger number of indicators is used directly on one or a few specific ecosystem services. The studies are also not focused on assessing multiple ecosystem services and most are targeted on a low range of such services. We found that the ecosystem services considered in 44% of our sample are only analysed in cropland systems, which in many cases means an intensive monoculture.
Furthermore in others cases (16% of the studies), they are analysed globally on land without specifically defining the targeted system of assessment (Verburg et al.,; Niedertscheider and Erb, ). Many studies do not consider the large diversity of other crops or other kinds of farming systems that are also critically important sources of food (Stoate et al., ). For instance, Temme and Verburg proposed combining European databases to build land-use intensity maps using separate methodologies for arable land and grassland.
A few papers analyse AI in more complex agricultural systems or heterogeneous/mix farming systems in local areas. The assessment of agricultural intensity is made at different levels The results show a weak trend in the spatial context of the AI changes, as these dynamics are driven by a wide range of indicators operating at different scales (Letourneau et al., ). Shows the relation between the scale at which an analysis is conducted and the number of indicators used. There were no significant differences between indicator frequencies and study scale. We were also not able to find a clear trend for the indicators used in a specific spatial scale. Significant differences were found in the frequency of ecosystem services, which yielded clearer trends among study scales.
They are more prevalent in studies whose analysis is conducted at a continent level or at different levels and conversely, less prevalent in studies at national and regional levels. The farming systems or land use systems analysed in agricultural intensity oriented studies. We observed that the analysis of agricultural intensification/intensity is conducted on large-scale land use or crop land, in particular monocultural crop systems, whereas there is a lack of studies focused on more complex and heterogeneous systems, e.g., polycultural systems or periurban farming systems. This gap is also reflected by the level of analysis at which these studies are generally performed: global and farm scales are the most common, missing out local and regional studies.
These kinds of regional and territorial studies are less easy to perform because of the need for data and methods, as underlined by Benoit et al. or Boiffin et al.
The methods needed are linked to the upscaling of field/plot research on ecosystem services provided by different agricultural practices (Kragt and Robertson,; Nieto-Romero et al., ) and are affected by the difficult assessment of the spatial distribution of cropping and farming systems at a regional level (Leenhardt et al., ). Geographical distribution of the case studies. We highlighted that the study of agricultural intensity or intensification in Mediterranean agricultural systems is not being sufficiently addressed. Most research is based on central/northern Europe, whereas the Mediterranean environment would be an interesting case study due to its diversity and also its vulnerability due to the biophysical, climatic and structural conditions (Caraveli, ).
In recent decades, a major driver of land use changes in these areas has been urban sprawl, i.e., low-density expansion of large urban areas mainly into the surrounding agricultural or natural areas (EEA, ). These areas characterised by extensive systems are, however, threatened by the changes in the intensity of farming (Caraveli,; Stoate et al., ). Indicators and thresholds used to assess agricultural intensity. Indicators are driven differently according to the context and the location in which they are measured. It would therefore be useful to include a larger number of case studies across different regions in order to assess a broader range (Rousevell et al., ). The studies we sampled show the importance of a major focus on the spatial context in land use intensity changes, as such changes are driven by a wide range of indicators operating at different scales (Letourneau et al., ). Although some studies discuss these approaches, and despite the importance of local studies on farming systems and environmental changes, the rate and magnitude of agricultural intensification have been quantified globally so that, the final outcome in these challenges is not enough.
Because a few thresholds have been defined in the literature, we are able to measure the variability of intensity indicators in a given area or for a given time span per area unit, but this is not enough to measure how intensive or sustainable a system is. To overcome this problem, some authors (Castoldi and Bechini, ) proposed defining thresholds with local stakeholders in order to take into account the local preferences for a given ecosystem service or a given indicator. It is clear that because of the increasing world population, we must continue to increase agricultural production (Bommarco et al., ) and therefore we need to understand how and under which conditions agro-ecosystems are altered by agriculture (Matson,; Snapp et al., ).
Regional and territorial case studies on complex agricultural systems could offer a solution to increasing our knowledge on how to measure and assess agricultural intensity. More multi-scale trials linking the plot/field with the territorial levels should be provided in order to evaluate the introduction of innovative and sustainable farming systems. A meta-analysis on case study results would support a further generalisation of local research findings. Finally, the social acceptability of AI or LUI should be tested with local stakeholders. Summary of methodological processes and approaches in the different publications.
Indicators of agricultural intensity. Reference Summary of the identified indicators at agricultural intensity Identified indicators Region Discipline Herzog et al.,; Snapp et al.,; Zhang et al.,; Gaudino et al.,; Jepson et al., Mineral fertilisers (NPK) Organic fertilisers Pesticides EU, Greece, China, West Africa Ecology Agronomy Environment Armengot et al.,; Fumagalli et al.,; Lu et al.,; Erb et al.,; Allan et al.,; Egorov et al., Technologies/labour intensity Mechanical weed control regime Mowing frequency (no.
The global land system is facing unprecedented pressures from growing human populations and climatic change. Understanding the effects these pressures may have is necessary to designing land management strategies that ensure food security, ecosystem service provision and successful climate mitigation and adaptation. However, the number of complex, interacting effects involved makes any complete understanding very difficult to achieve. Nevertheless, the recent development of integrated modelling frameworks allows for the exploration of the co-development of human and natural systems under scenarios of global change, potentially illuminating the main drivers and processes in future land system change. Here, we use one such integrated modelling framework (the CLIMSAVE Integrated Assessment Platform) to investigate the range of projected outcomes in the European land system across climatic and socio-economic scenarios for the 2050s. We find substantial consistency in locations and types of change even under the most divergent conditions, with results suggesting that climate change alone will lead to a contraction in the agricultural and forest area within Europe, particularly in southern Europe.
This is partly offset by the introduction of socioeconomic changes that change both the demand for agricultural production, through changing food demand and net imports, and the efficiency of agricultural production. Simulated extensification and abandonment in the Mediterranean region is driven by future decreases in the relative profitability of the agricultural sector in southern Europe, owing to decreased productivity as a consequence of increased heat and drought stress and reduced irrigation water availability. The very low likelihood (. 1. Introduction Humans have been changing the European landscape for millennia in response to their requirements for the many benefits or ecosystem services arising from the natural environment and its constituent resources. Developments in social systems, new technologies and crops, growing populations and economies have all had dramatic effects (, ).
Climatic changes have also had substantial impacts, both on the landscape and on human societies, driving a complex pattern of inter-related environmental changes (, ). Now, as the pace of socio-economic and climatic change continues to quicken, their consequences for the land system are commensurately greater and more uncertain. Climate change is likely to have impacts through changes in precipitation, temperature, CO 2 concentrations and sea level rise, affecting the suitability of land for different crops (, ), tree species , habitats (e.g. ) and forms of management (e.g. Irrigation - ). Meanwhile, human activities will further modify the European landscape across scales, as populations, dietary preferences, trading patterns and management practices all change (, ). Previous research suggests that these non-climatic pressures may be more important drivers of land use change than climate change (, ).
However, the diversity of social, economic, political and technical factors involved mean that future demands for living space and natural resources are hard to predict. This is exemplified by the breadth of potential socio-economic storylines used in impact assessments, from the global-to-regional Special Report on Emissions Scenarios (SRES; ) and Shared Socioeconomic Pathways (SSP; ) to the continental-to-national storylines developed by stakeholders (e.g., ). Further uncertainty arises from differences between methods of analysis and modelling that emphasise distinct processes or sectors. These often give divergent projections even under identical climatic and socio-economic scenario conditions, suggesting that the identification and representation of major land change drivers requires significant improvement. In particular, the discrete sectoral nature of many models precludes consideration of the numerous cross-sectoral interactions that influence land use distributions. Examples include changes in urban extent or coastal flood defence policy affecting agricultural land availability , and changes in population and water consumption affecting availability of water for irrigation. Omitting such interactions can lead to substantial over- or under-estimation of climate impacts and direct and indirect consequences for land use (e.g., ).
One important outcome of uncertainty-focused model applications has been the identification of areas where high levels of uncertainty imply that land use is especially vulnerable to change. However, far less has been discovered about the areas where land use is robust to climatic and socio-economic pressures, and therefore the conditions which allow for the maintenance of food supplies, livelihoods, biodiversity and other ecosystem services. For this purpose, comprehensive scenarios and integrated modelling frameworks are particularly valuable because they allow for fuller, more realistic representations of the land system. Previous examples include studies focusing on future crop yields (, ), agricultural land use change , broader land and/or energy usage (, ), inter-sectoral climate impacts (, ), and policies for the sustainable development of land use systems (, ). Many of these studies involve integrated modelling of European land use; a particularly interesting case due to the research attention it has received in support of a coherent political system that attempts to influence land use outcomes across scales.
While Europe therefore provides a setting where advances in understanding of future land use change should be both possible and of practical value, this potential has not yet been fully realised. In particular, there has been a lack of assessment of the extent of certainty across land use categories under climatic and socio-economic changes acting directly and indirectly through representative cross-sectoral interactions. This paper addresses this gap using an integrated multi-sectoral modelling platform, the CLIMSAVE Integrated Assessment Platform (IAP), which incorporates a broader range of drivers and cross-sectoral processes than previous integrated models, and so allows for less strongly conditional projections of future land use change. We address two research questions:. The CLIMSAVE IA Platform The CLIMSAVE IA Platform (IAP) is an interactive, exploratory, web-based tool for simulating climate change impacts and vulnerabilities on a range of sectors (, ). The Platform integrates a suite of models of urban development, water resources , coasts , agriculture and forests , and biodiversity to simulate the spatial effects of different climatic and socio-economic scenarios across Europe.
The IAP has been applied widely in climate change impact (, ), adaptation and vulnerability assessments, in robust policy analysis and has been tested extensively through model sensitivity and uncertainty analyses (, ). The Platform operates at a spatial resolution of 10 arcmin × 10 arcmin (approximately 16 km × 16 km in Europe) grid cells, although multiple soil types are represented in each grid cell, and covers two thirty-year timeslices (2020s and 2050s).
Climate scenarios The climate change scenarios within the IA Platform are based on combinations of the IPCC emissions scenarios (A1b, A2, B1 or B2), three climate sensitivities (low, medium or high) and five global climate models (GCMs). The five GCMs (MPEH5, CSMK3, HadGEM, GFCM21 and IPCM4) were chosen from the CMIP3 database using an objective method to represent as much uncertainty as possible due to between-GCM differences (see for further details).
Projections of Europe-wide area-average temperature change across these climate models and scenarios range from 1.1 to 4.9 °C in winter and from 1.0 to 3.6 °C in summer in the 2050s. Projections for precipitation change range from increases of between 1.1 and 12.5% in winter and decreases of between 2.0 and 29.5% in summer. The pattern of temperature and precipitation changes differs according to the GCM (see Online Resource 2 of ). Although we acknowledge that there are more recent scenarios available than those in the IAP, the European area-average changes across these scenarios cover at least the 25th to 75th percentile range of the European changes in summer and winter precipitation and temperature change to 2065 for the CMIP5 global models for the RCP2.6 to RCP8.5 scenarios. Socio-economic scenarios The IAP contains four European socio-economic scenarios that were developed by stakeholders in a series of professionally-facilitated participatory workshops (see ). In the first and second workshops, the objectively selected stakeholder group developed and iterated qualitative socio-economic stories and dynamics according to the two drivers that they considered most important and uncertain: “ effective vs ineffective solutions by innovation” and “ gradual vs roller-coaster economic development”. Models A meta-modelling approach based on computationally efficient or reduced-form models that emulate the performance of more complex models was used to facilitate greater complexity of model linkages within the IA Platform and a relatively fast run time.
Although the CLIMSAVE IAP includes a large number of interlinked models , this section briefly describes those models which indirectly or directly affect spatial land allocation. For further details on the models, see, and papers cited. Urban expansion: The Regional Urban Growth (RUG) metamodel consists of a look-up table of percentage of artificial surfaces per grid cell (between zero and almost 100%) derived from running the RUG model (based on ) with all possible combinations of input values (population, GDP, household preference for proximity to green space versus social amenities, attractiveness of the coast (scenic value versus flood risk) and strictness of the planning regulations to limit sprawl).
Development in urban and rural areas is given first priority in the allocation of land;. Flooding: The Coastal Fluvial Flood (CFFlood) meta-model is a simplified process-based model that identifies the area at risk of flooding based on topography, relative sea-level rise or change in peak river flow and the estimated Standard of Protection of flood defences. The probability of flood inundation constrains the allocation of land for agriculture, with land with a 10% and 50% annual probability of flooding being unsuitable for intensive agriculture and extensive agriculture, respectively, according to. Water: The WaterGAP (WGMM) meta-model uses 3D response surfaces to reproduce WaterGAP3 runs at a 5′ × 5′ resolution for about 100 spatial units (single large river basins or clusters of smaller, neighbouring river basins with similar hydro-geographic properties). The difference between simulated water availability and projected non-agricultural water consumption determines the maximum water available for agricultural irrigation in each spatial unit;.
Forest: MetaGOTILWA + is an artificial neural network (ANN) that emulates GOTILWA +. The ANN was trained on GOTILWA results for 889 grid cells across Europe, and simulates average timber yields for a range of deciduous and coniferous tree species under different management regimes and soil characteristics;. Crops: The crop yield metamodels use ANNs to predict the average yield of a range of annual and permanent crops under rainfed and irrigated conditions.
They have each been trained and validated on simulated outputs across Europe from the daily ROIMPEL model for winter and spring wheat, barley and oilseed rape, potatoes, maize, sunflower, soya, cotton, grass and olives. The training datasets were sampled from 150,000 model data points to adequately cover the range of soil and climate predictors and the predictands. Rural land allocation: The SFARMOD meta-model allocates available land across Europe based on profit and other constraints (urban land use, irrigation availability; food and timber demand). It uses a series of regression equations to simulate the behaviour of the full SFARMOD-LP model, a mechanistic farm-based optimising linear programming model of long-term strategic land use. The metamodel was fitted to SFARMOD-LP outputs from 20,000 randomly selected sets of input data that fully cover the current and future parameter input space. The regression is broken into steps that estimate first the percentage of the area of each crop in each grid cell, then the costs of dairy cows (concentrates), then the fixed costs of labour and machinery, from which gross margins, net income and profit is derived. Up to 10 iterations adjust crop and livestock prices to meet the demand for food within Europe, which is a function of population, imports, food preferences and bioenergy.
Where the resulting profit is above a threshold (set at €350/ha) land is deemed to be used for intensive agriculture (either arable or dairy agriculture). Otherwise the profit is re-calculated without the arable crops to represent extensive agriculture (sheep and beef) and compared with the profit from managed forests (based on the annual equivalent profit of a total Net Present Value over the life of the forest). If the resulting profit is greater than a second threshold (set at €150/ha) then this land is used for whichever of managed forest or extensive agriculture has the greatest profit.
Otherwise the land is not used for productive purposes – it is assumed to be unmanaged forest if the Net Primary Productivity of unmanaged forests is positive and greater than the grass yield of extensive grass, else unmanaged land. All metamodels were satisfactorily validated against either baseline observations or the validated outputs of the full model (see for the validation of each model; and the sensitivity analysis of the linked models within the IA Platform in ). Model runs and analysis The CLIMSAVE IA Platform was run for 300 scenarios for the 2050s timeslice to explore the effects of climate change and socio-economic change uncertainties on European land use change. The scenario combinations can be categorised into two scenario groups:.
Climate change only - Climate scenarios for every combination of the 4 emissions scenarios, three levels of climate sensitivity and five GCMs, combined with baseline socio-economics (60 runs);. C limate and socio-economic change - Climate scenarios (60 runs above) combined with each of the four socio-economic scenarios (240 runs). Four indicators from the IAP were analysed: 1) area of intensive agriculture; 2) area of extensive agriculture; 3) forest area; and 4) area of unmanaged land. In all analyses, it was assumed that each scenario was equally probable i.e. No prior assumptions were made regarding the likelihood of a particular magnitude of climate or socio-economic change.
Grid based results for each of the 60 and 240 simulations were averaged to provide maps of the multi-scenario means for climate change only and climate and socio-economic change simulations in the 2050s, which were compared against the simulated baseline distribution to assess whether there is spatial coherency within the two scenario groups despite the diverse climate and socio-economic scenarios. Secondly, we investigated whether there are areas in Europe in which individual land uses have future spatial certainty in the direction of change in their spatial extent. There is no common definition of what constitutes important land use change, with recent land cover change across Europe ranging from 0.5% per annum in Portugal and Slovakia and scenario studies projecting changes in the spatial extent of individual European land use classes of up to 5% by 2020 (e.g.
) and 8% by 2050 (e.g. ) compared to 2000. We adopted a threshold of 5%, but tested the robustness of results across a range of thresholds from 0.1% to 25%. The probability of changes in extent of each land use within each grid cell of 5% of the simulated baseline (using 1961–90 climate and 2010 socio-economics) was calculated and classified according to the likelihood scale of to identify those cells in which an increase or decrease in land use extent is “likely” (66–90% probability) or “very likely” ( 90% probability).
Cells in which no change in the spatial extent of each individual land use greater than ± 5% was “very likely” were also identified. Finally, the probability that none of the four land uses change in extent by 5% was calculated for each grid square to identify areas where there is spatial certainty that all land uses will remain largely unchanged in extent. These analyses were then repeated across the range of thresholds above. Multi-scenario average spatial land use allocation Given the divergent nature of the socioeconomic scenarios and differing spatial patterns of temperature and precipitation change from the climate models, the simulated baseline distribution of each land use was compared with the multi-scenario mean of the 60 simulations with climate change only and the 240 simulations with climate and socio-economic change ( and Figs.
It is apparent that there is a spatial coherency in the simulated future distribution across Europe for each land use; i.e. Averaging the individual grid-level values resulting from the diverse range of input scenarios has not led to a quasi-random distribution of land use allocation. Focusing on intensive agriculture, the effect of climate change alone (with baseline economics meaning that there is no change in the food demand or imports) leads to a projected northward migration of intensive agriculture , particularly into northern UK and Finland.
This is also associated with reduced intensive agricultural areas in southern Europe (especially Spain and Italy) due to a combination of increased heat stress and reduced availability of irrigation water. These two factors reduce the simulated relative competitiveness of Mediterranean agriculture in contributing towards meeting the demand for European agricultural production, so that the required demand for production can be met by a smaller agricultural area focused in a band across central and north-western Europe due to yield increases. Comparison of the simulated baseline distribution of intensive agricultural land with the multi-scenario mean of the 60 simulations with climate change only and the 240 simulations with climate and socio-economic change for the 2050s. The introduction of socioeconomic change affects modelled European agricultural land use requirements by changing both the demand for agricultural production, through changing food demand (due to changes in population, wealth and dietary preferences) and net imports (arising from changes in Europe's relationship with the rest of the world), and the efficiency of agricultural production (through scenario changes in mechanisation, yield development and crop breeding and irrigation efficiency). The multi-scenario mean in shows the most intensive areas of agriculture (red areas) in the same locations as the climate-only multi-scenario mean (as these are the most profitable production areas) but expands the production area into regions that had lost competitiveness under climate change such as southern France and the Baltic states. Model certainty in the direction of land use change across Europe and Fig. SM5 shows the certainty in the modelled direction of change in the spatial extent of each land use class across all of the scenario combinations, expressed as the percentage of runs in which the land use class changes in extent by 5% (compared to the baseline simulation) in a given grid cell.
The sensitivity analysis using change thresholds of between 0.1 and 25% within each grid cell (Fig. SM1 in the Supplementary material) shows that increasing the change threshold inevitably leads to an increasing spatial extent in the area with certain “no change” and decreasing extent of areas with both uncertain change and certain (increasing and decreasing) change. However, the overall percentage in each certainty class is not fundamentally changed, demonstrating the robustness of the results. Multi-scenario certainty in the direction of modelled land use change (of at least 5% within a grid cell) for the 2050s for (left) climate change only and (right) climate and socioeconomic change. With climate change alone, there are areas with significant confidence in the direction of change for all land uses, with forest area decreasing in 90% of simulations in significant areas of all regions of Europe; unmanaged land increasing in southern Europe and Scandinavia at the expense of intensive and extensive agricultural land; extensive agricultural land increasing through a band across the centre of Europe at the expense of intensive agricultural land; and intensive agriculture increasing in parts of northern Europe at the expense of forest. There are also areas with confidence of no change – for example, intensive agriculture is unchanged within much of central England, northern France and the Benelux countries; whilst forest is little changed in large areas of Scandinavia, Spain and central Europe. However, there is a major decrease in the confidence of the direction of change and a large expansion of the areas with uncertain change (i.e.
90% of simulations) confidence in the decrease of intensive agricultural land or increase in unmanaged land. There are only areas within Scandinavia, Italy, France and Hungary in which there is at least a 66% agreement of an increase in unmanaged land. The areas with certainty of no change (change of. Model certainty in stable land use patterns across Europe showed that there are areas across Europe in which there are high levels of certainty in the direction of change (either increases or decreases) in the extent of individual land use classes due to climate and socioeconomic change. However, it also showed extensive areas across Europe with either certainty of little change or uncertainty in the magnitude or direction of change.
Therefore shows the percentage of simulations in which all four land use classes within a given grid cell change by less than ± 5% from the baseline proportions, classed according to the likelihood scale of. Approximately 20% of the simulated cells across Europe are very likely (90–99% probability) or virtually certain (99–100% probability) to maintain their baseline land use proportions despite the effects of climate change on land suitability and crop and timber yields. These stable cells include significant areas within the UK, northern France, northern Spain, Germany and Scandinavia, and account for 50% of the areas of three countries: The Netherlands (88%), Republic of Ireland (54%) and Norway (51%). Likelihood of baseline land cover proportions remaining unchanged (± 5%) across modelled grids under 2050s climate and socioeconomic change. The introduction of the socio-economic scenarios decreases the extent of areas very likely or virtually certain to retain baseline land use proportions from 20% to about 12% of the simulated grid cells. Only the Netherlands (54%) and Republic of Ireland (50%) now have 50% of grid cells in this category.
At a national scale, the socio-economic scenarios lead to a reduction in areas of certainty in all countries, with the exception of the Czech Republic and Austria where there is an increase of 1% and 5%, respectively. There are also smaller areas where the certainty of maintaining the current land use distribution increases; for example eastern England, northern Romania and the Po valley in Italy. However, whilst socio-economic scenarios reduce overall land use certainty in Europe, the percentage of cells in which baseline land use proportions are exceptionally unlikely (0–1% probability) to remain unchanged also decreases, from 42% to 20%, with associated increases in cells that are very unlikely (1–10% probability) or unlikely (10–33% probability) to remain unchanged.
4. Discussion Changes in global land systems in the coming decades will be strongly influenced by a range of interacting climatic and socio-economic factors. This complexity makes projections of future change hard to achieve, and reliant on integrated modelling approaches that respect the dynamics that occur within and between individual sectors. As a result, assessments of future conditions, made largely in the absence of well-developed integrated approaches of this kind, have so far focused on areas of uncertainty. However, recent methodological advances allow for more confident exploration of the converse; areas that, with some level of certainty, appear robust to external drivers of change and internal complexity. Here, we applied a cross-sectoral European modelling platform spanning a wide range of climatic and socio-economic conditions, which therefore covers or exceeds the uncertainty space identified in previous studies of future land use change (e.g., ).
Manifestations And Underlying Drivers Of Agricultural Land Use Change In Europe
Uncertainty in future climate conditions was addressed through sixty climate change scenarios that spanned at least the 25th to 75th percentile range of the European changes in summer and winter precipitation and temperature change to 2065 for the CMIP5 global models for the RCP2.6 to RCP8.5 scenarios, and included differing spatial patterns of change. Furthermore the four socio-economic scenarios included contrasting directions of change in key scenario inputs such as population, GDP, spatial planning policy, societal behaviour (including dietary preferences for meat and water consumption), crop breeding and agronomic improvement that influence land-take for development, food demand, irrigation water availability and agricultural productivity. Given the extent of uncertainty space investigated here, an entirely divergent set of projected future conditions might be expected. However, we instead find substantial consistencies across results. Indeed, the strong spatial patterns within the multi-scenario means of individual land use distributions in the 2050s ( and Figs. SM2–4) suggest a considerable degree of predictability in future land use at the European scale. In particular, the analysis presented here suggests that there are areas of Europe with high levels of certainty in two aspects of future land use change – the direction of change for individual land uses and the probability of maintaining the current land use distribution.
Monish Jose
These findings are consistent with who found significant overlap in the probability density functions of the grid-based indicators of food production per capita, land use intensity index and land use diversity index for the four contrasting CLIMSAVE scenarios. This consistency is especially notable because the analysis of included model uncertainty that we do not directly measure here; a potentially large source of uncertainty (e.g. Alexander et al., 2016). Furthermore, some of our broad and specific findings are consistent with those of other integrated and stand-alone models. For example, the magnitudes of change in the agricultural sector are in line with the (albeit wide) range found by the inter-model comparison of Alexander et al.
(2016), while our finding that some areas of Europe appear to be robust to a wide range of scenario conditions also agrees with the studies of, and. These earlier studies found substantial areas of stability and/or predictability using more narrowly-based or discrete modelling approaches at European and global scales (which included the effects, for instance, of GDP and population change).
These earlier findings also have some similar geographical characteristics, particularly in the locations of agricultural intensification in northern and western Europe and agricultural extensification, diversification or abandonment in parts of southern and eastern Europe (see e.g., Fig. These suggest that error propagation in the IAP's cross-sectoral models is not amplifying the impact of individual, sub-model uncertainty, supporting the conclusions of and. The present study represents a valuable advance, therefore, because it encompasses a greater number of socio-economic drivers of change and of sectoral and cross-sectoral processes (also known to be large potential sources of uncertainty; ), thereby allowing for the identification of particular characteristics that lead to certainty in land use futures. Some of the most informative and novel findings presented here relate to the ways in which specific drivers affect, or fail to affect, land use patterns. For instance, we found that climate change was associated with a number of key simulated changes.
Leibniz University Of Hanover
A decrease in forest area was likely (66–100% probability) in many regions of Europe due to large increases in tree growth simulated by metaGOTILWA + under climate change. This suggests a direction of change and degree of certainty not identified by previous studies (e.g. ), which may result from differences in underlying assumptions about climate impacts on tree growth and socio-economic impacts on wood demand. In our simulations, climate change also drove agricultural changes that appear to reinforce the current trends of intensification of agriculture in northern and western Europe and extensification and abandonment in the Mediterranean region , as shown by the likely (66–100% probability) decreases in intensive and extensive agriculture and likely increase in unmanaged or very low intensity agriculture in many parts of southern Europe. The combination of increased heat stress due to higher temperatures, increased summer drought stress in rainfed systems and reduced availability of irrigation water in the region all contribute to reducing profitability and competitiveness of Mediterranean agriculture compared to that in central and northwestern Europe.
This suggests that established findings about the impacts of climate change on European agriculture (e.g., ) are largely robust to socio-economic change and cross-sectoral interactions, as also indicated by earlier integrated modelling studies (e.g., ). We also find that socio-economic drivers of land use change play a crucial role, sometimes even dominating over climatic drivers. For instance, the introduction of socio-economic scenarios partly offsets the loss of agricultural land in southern Europe leading to a (small) increased likelihood of stable land systems in very unstable areas (as shown by grids shifting from “exceptionally unlikely” to remain unchanged to “unlikely or “very unlikely” – ). This arises as large increases in European food demand in some scenarios (due to population growth, increased affluence and reduced net imports) allow more marginal areas to continue to contribute profitably to production.
This is also true of areas such as the Po valley in Italy with its deep water-retentive soils. Nevertheless, the total proportion of grid cells which are very likely (90–99% probability) or virtually certain (99–100% probability) to maintain baseline land use proportions decrease from 20% of grid cells under climate change alone to just 12% with both climate and socio-economic change as a consequence of the major differences in food and timber demand, agricultural productivity and food trade across the divergent socio-economic scenarios, which together determine the requirements for land. These stable areas are mostly in north-western Europe and Scandinavia, while many parts of the Mediterranean have a high likelihood of changing land use distributions (with the main exception of the Pyrenees and Cantabrian mountain chains in northern and north-eastern Spain respectively). In fact, extensification or abandonment in currently marginal areas, as agronomic conditions become less favourable through increased summer drought stress, reduced availability of irrigation water and decreasing profitability, give a high likelihood of land use change in much of Spain, southern France and Italy. Elsewhere (e.g. Scandinavia), climatically-driven increases in timber yields act alongside increasing demands for agricultural land to substantially reduce the projected area of forestry, especially where timber prices also fall.
It is apparent from these results that some specific factors promote or impede land use change in general. Particularly influential were population growth rates, dietary preferences, yields and imports (, ), with each having direct consequences for the extent of agricultural land required. 5. Conclusions The CLIMSAVE IAP has been used to simulate the spatial distribution of four broad land use classes (intensive agriculture, extensive agriculture, very low intensity/unmanaged land and forest) across Europe in the 2050s under a broad range of climatic and socioeconomic conditions. Sixty climate scenarios spanned the equivalent of at least the 25th to 75th percentile range of changes in summer and winter precipitation and temperature change to 2065 within RCP2.8 to RCP8.5. These were combined with baseline socio-economics (climate change only) and four socioeconomic scenarios (climate and socioeconomic change) to understand the certainty in the direction of change of future land use allocation and the certainty in maintaining an unchanged proportion of land use classes into the future.
Results suggested that climate change alone will lead to a contraction in the agricultural and forest area within Europe, particularly in southern Europe, which is partly offset by socioeconomic changes in both the demand for agricultural production, through changing food demand and net imports, and the efficiency of agricultural production. Whilst this modelling shows significant areas in northern and western Europe in which it is likely ( 66% probability) that current land use proportions will remain largely unchanged, the results reinforce current trends and previous integrated model findings of intensification of agriculture in northern and western Europe and extensification and abandonment in the Mediterranean region. These changes are driven by decreases in the relative profitability of the agricultural sector in southern Europe, owing to decreased productivity as a consequence of increased heat and drought stress and reduced irrigation water availability. The very low likelihood (. The research leading to these results has received funding from the Biotechnology and Biological Sciences Research Council (BBSRC) grant no BB/K010301/1 and BB/N00485X/1 (as part of the MACSUR knowledge hub within FACCE Joint Programming Initiative for Agriculture, Climate Change, and Food Security) and the European Commission Seventh Framework Programme under Grant Agreement No. 244031 (The CLIMSAVE Project; Climate change integrated assessment methodology for cross-sectoral adaptation and vulnerability in Europe; ). CLIMSAVE is an endorsed project of the Global Land Project of the IGBP.
The authors would like to thank all CLIMSAVE partners for their contributions to the IAP. The CLIMSAVE IAP model outputs can be accessed by contacting researchdata@cranfield.ac.uk.
Christensen, J.H., K. Krishna Kumar, E. Aldrian, S.-I. Cavalcanti, M.
De Castro, W. Kossin, N.-C. Renwick, D.B. Stephenson, S.-P. Zhou, 2013: Climate Phenomena and their Relevance for Future Regional Climate Change Supplementary Material.
In: Climate Change 2013: The Physical Science Basis. Contribution of Working Group I to the Fifth Assessment Report of the Intergovernmental Panel on Climate Change Stocker, T.F., D. Midgley, eds.
Available from and www.ipcc.ch.