Scaling the land use system : a modelling approach with case studies for Central America

<em>Introduction</em>There is a growing demand for quantitative information on actual land use/land cover and their future changes in space and time. Particularly during the last decade, land use and land cover change have become important issues. Besides local and direct effects like loss of biodiversity through deforestation or soil degradation through unsustainable land use, increasing importance is given to the global impact of more indirect (future) effects like greenhouse gas emissions and carbon fixation.The land use system is highly complex and this notion of complexity has consequences for the way the system should be described. Theories applied in this thesis draw from ecology, a field of science that has acquired ample experience in describing the complexity of the ecosystem. The land use system was considered to be functionally complex, <em>i.e.</em> land use patterns are governed by a broad variety of potential spatial determinants. International agreements initiated by <em>e.g.</em> the United Nations, World Bank or GATT influence regional and national policies that trickle down to local decision-makers and that will ultimately affect farmers' decision to change land use. A second property of the land use system was its structural complexity, that is, which variables will emerge as important spatial determinants of land use depends largely on the adopted (spatial) scale of analysis. Finding ways to translate information among scales is one of the fundamental challenges faced by researchers in land use analysis. Models are crucial tools to help understand the dynamics and complexity of a land use system. The CLUE (Conversion of Land Use and its Effects) modelling framework was selected as the appropriate means to model land use change. It was one of the few examples of land use models that take into account both structural and functional complexity.<em>Objectives</em>The main objectives of this study were to analyse the present land use patterns and to project possible future pathways of the dynamics of those patterns for a number of hierarchically nested cases. Following the structure of the CLUE modelling framework, three sub-objectives were formulated:<OL><LI>To identify biogeophysical and socioeconomic determinants of land use patterns at different spatial extents and resolutions, by means of an empirical analysis of actual land use systems.</LI><LI>To analyse demand for the major agricultural commodities at the national or regional level and to develop a limited number of scenarios that explore plausible future pathways of land use change.</LI><LI>To spatially explicit allocate national changes as projected in those scenarios, using the information of the spatial analysis of the land use system.</LI></OL><em>Study area</em>All case studies concerned (parts of) Central America. The region has specific characteristics that made it very suitable for the research described in this thesis. Firstly, biophysical, climatic, and socioeconomic gradients are steep over small distances, which induces a strong variation in land use over relatively small areas. High mountain ranges with steep slopes, shallow soils, low temperatures, and low population densities could be near to flat, fertile, and hot lowlands that were densely populated. Secondly, the area included six countries that display large economic and political differences. Honduras and Nicaragua rank among the poorest in the Western Hemisphere, while Costa Rica is one of the more stable and robust democracies of Latin America based on a strong, export-led, development. Finally, most countries collected extensive data sets. Detailed agricultural censuses exist for Honduras and Costa Rica, and because of the presence of a research centre of Wageningen University, an outstanding data set exists for the Atlantic Zone of Costa Rica. As a result, Central America offered the unique possibility to construct a hierarchically nested range of case studies to systematically analyse land use dynamics that are influenced by a variety of factors over a range of spatial scales. A total of four case studies were modelled: the Atlantic Zone of Costa Rica ('Atlantic Zone'), Costa Rica, Honduras, and Central America.<em>Statistical analysis</em>Within this thesis, the quantification of the relationships between land use and their spatial determinants for the Atlantic Zone (Chapter 2), Honduras (Chapter 3), and all countries of Central America (Chapter 4) were discussed. Except for the Atlantic Zone, agricultural censuses were always the main source of spatially explicit land use data. Despite the huge amount of information, little use has been made of this potentially valuable data source. All data were transformed to a raster. Instead of using uniform grid cells with one dominant land use and <em>e.g.</em> one type of soil, sub-grid information (using percentages) was present for most of the spatial determinants and each land use. Spatial resolution ranged from 2 × 2 km for the Atlantic Zone to 75 × 75 km for Central America. Spatial extent varied between 5,000 km <sup>2</SUP>and 500,000 km <sup>2</SUP>. For all cases, the influence of spatial scale was quantified by executing the statistical analysis at a range of spatial resolutions or spatial extents. Specific results varied from case to case, but the following general conclusions applied:<UL><LI>The patterns of land uses that produce basic foodstuffs (maize, beans, rice, pasture) were very well explained, whereas distributions of crops that are export-oriented (bananas, coffee, sugar cane) were less well captured.</LI><LI>Socioeconomic, soil-related, climatic and sometimes (Atlantic Zone) political factors were necessary to describe land use patterns. Omission of whichever of those groups of variables from the analysis would impoverish results, which proved that the land use system is functionally complex.</LI><LI>Relationships were relatively stable over time. Although total quantities as well as patterns changed considerably in all cases, composition of explaining variables remained relatively constant.</LI><LI>Relationships were not very sensitive for changes in spatial resolution. In none of the cases, large changes in composition of explaining variables were observed with changing spatial resolution.</LI><LI>Description of land use patterns proved very sensitive to changes in spatial extent as demonstrated by varying extent between national and regional (Chapter 4). The importance of national borders suggested that the search for scale dependencies should focus on administrative or political levels rather than on biophysically defined levels, as is proposed by ecosystem theory. This would call for a separate land use system theory.</LI><LI>Spatial determinants of land use change were sometimes counterintuitive and contradicted prevailing hypotheses on causes of land use change. A variety of examples substantiated the plea to start an analysis at coarser scales with empirical methodologies.</LI></UL><em>Scenario development</em>The near future area changes of land use at the national or regional (Atlantic Zone) level were studied by developing a limited number of plausible scenarios. The time period for which scenarios were developed varied from case to case, but usually started in the early 1990s and finished in 2005 or 2010. Scenarios were developed for a number of commodities that translated into a smaller number of land uses. Scenarios were divided into two types. In the demand-controlled scenarios, only the total area to be allocated was variable. Changes in land use depended on (macro-)economic, demographic, and crop specific factors at national level. Most important factors included population growth, income growth, export/import development, and yield. All variants of the developed base scenario (market liberalisation; market protection; 1%, 3%, and 5% GDP growth) and the sustainable scenario belonged to this type. In the allocation-controlled scenarios, land use changes also depended on spatially specific allocation conditions. The developed park protection scenario, where deforestation within national parks was inhibited, was an example of this type. The natural hazard scenario described in Chapter 5 was an example of a realistic scenario that combined changes in demand with location-specific alterations.<em>Spatially explicit modelling</em>Resulting maps from the allocation module of the CLUE modelling framework were discussed for the Atlantic Zone (Chapter 2), and Honduras and Central America (Chapter 5). The allocation module combined the spatially explicit information obtained from the multi-scale empirical analyses and the non-spatial area development from the scenario studies. Results from the various cases, mostly maps of hot-spots of change, demonstrated the feasibility of the application of the CLUE model at different spatial scales. Satisfactory results were obtained for the Atlantic Zone as well as for Central America as a whole. Results of the natural hazard scenario for Honduras separately and for Central America indicated the likeliness of the effects of a hurricane on land use patterns, though initially strong, to largely disappear within a period of 10 year. Concepts from ecology were used to illustrate the modelled behaviour of the land use system. CLUE thus proved to be able to mimic spatially explicit land use changes for a number of diverging scenarios. The effects of protecting national parks, macroeconomic changes, as well as an extreme weather event could be evaluated.<em>Validation</em>In spite of the large number of models and in spite of a general agreement that validation should be an essential part of any model, the majority lacked a validity check, often because of data problems. All case studies in this thesis were validated, also because for all cases independent data from two different points in time was available. Statistical relationships between land use and their spatial determinants were established for the older of the two data sets. Subsequently, the CLUE allocation module was run, starting at the oldest year until the most recent year. Validation was quantified by statistically comparing the modelled and actual land use changes. For the Atlantic Zone, Honduras and Costa Rica, the allocation module of the CLUE model was successfully validated, yielding satisfactory coefficients of determination for the relationships between actual and modelled land use patterns. Besides, for Honduras and Costa Rica a multi-scale validation was executed, as it should accompany multi-scale analysis and multi-scale modelling. Results improved strongly, and exponentially, with a coarsening of spatial resolution. Validations demonstrated that the CLUE modelling framework could reproduce changes as they took place in Central America between the 1970s and 1990s.<em>Stakeholders</em>Three possible groups of users of the CLUE model were identified, namely other modellers, farmers, and national or regional policy makers. The development of the CLUE modelling framework was initiated by a demand from the global modelling community. Eventually, a number of land use models could be used together in a sequence, forming a so-called modelling tool-box. Results of a coarse scale model could help focus other more detailed models, which could then be employed to model local processes. In its present form, the CLUE model would have little value at the farm level. The model, however, was constructed such that neither the application at a more detailed level, nor the replacement of statistical equations by any other kind of (process-based) rules would be problematic. Policy makers at various organisational levels are a group of potential stakeholders that could be addressed with the model in its present form. The Central America case study presented in Chapter 5 was an example of how this could function in practice. The mutually beneficial collaboration between various international institutes showed how a model like CLUE, designed primarily to fulfil scientific needs, was used by other stakeholders. The incorporation of a knowledge-broker, an intermediary that links the people who use knowledge (policy makers) and those who create it (scientists), provided the setting for the successful communication.