Contemporary regional spatial plan-making in the Netherlands is characterized as a complex process wherein multiple actors, with different levels of interests and demands, try to commonly develop a coherent and comprehensive set of future plan scenarios. The construction of the set of spatial plan scenarios is the core activity of each regional spatial planning process and is often unique and tailored to the specific context and policy objectives formulated for a plan area. Modern collaborative scenario construction is complex due to a variety of participating actors, as public planners, domain experts and non-experts as interest groups and landowners. The level of participation of the non-expert group varies from process to process, but for effective spatial scenarios it is important to ergonomically construct, surprising and plausible scenarios with vivid, proximate and concrete content. The last decades, many attempts have been undertaken to support plan scenario development with digital systems, with strong emphasis on the analytical capabilities of computers. Little attention, however is given to the development of intuitive sketch and design tools and methods, that support the interactive process of large-scale collaborative multi-level plan design, by visualizing and modeling comprehensive landscape scenarios down to the level of cadastral lots. Therefore, the main objective of this research is to develop and evaluate a method, that generates plausible landscape configurations by using user-defined landscape typologies, as a digital support tool for participatory spatial plan-making. To enable the effective design and modeling of vivid and plausible future spatial scenarios, there is a need for a method which supports the two main steps of plan scenario construction in Simlandscape. Simlandscape introduces a rich set of instruments and procedures in order to construct a diverse and coherent scenario set that supports communication and social learning and that facilitate a better informed decision-making process. The central notion in Simlandscape is that actual transformation of the landscape takes place at the ownership lot level. Through construction of strategic spatial scenarios down to the level of individual or clustered lots, comprehensive qualitative and quantitative evaluation becomes possible. Design instruments are proposed, that are intuitive in supporting the funneling creative design process from abstract and general sketches to specific and detailed economic function allocation and landscape layout modeling. The latter activity is supported by the definition and allocation of landscape lot typologies with (non-spatial) attributes. The first step in plan scenario construction in Simlandscape consists of the distribution and allocation of landscape lot typologies to lot geometries. This step poses a complex problem, which can be manually as well as automatically be solved, but is not the core of this research. The second step, assumes that a landscape lot typology is allocated to a lot geometry, and contains generation of a plausible landscape configuration, based on the attributes of the landscape lot typology. This step can also be done manually, but is very time-consuming for a total plan area involved. Therefore, automatic generation of a plausible landscape configuration, based on the properties of the allocated landscape lot typology is important and central subject in this research. The automatic generation of landscape configuration is part of the research field called ‘generative modeling’. In chapter 2, the most of the established existing generative approaches in generative landscape modeling are reviewed for their applicability and relevance as the base for the method to generate plausible landscape configurations from landscape lot typologies. In spatial planning literature, four important more or less distinct fields of research are identified which offer directly or indirectly approaches for developing a generative method: 1) procedural modeling, 2) spatial multi-objective optimization modeling, 3) cellular automata and 4) multi-agent systems. The approaches to generate landscape configurations provide several points of departure. Unfortunately, none of the current approaches is directly applicable for the addressed objective in this research. Procedural modeling techniques as shape or landscape grammars are able to produce, or support the creation of detailed, appealing and realistic landscape visualizations. Due to this level of detail of modeling, the process of inference to identify relevant objects and mutual relations in reality, is complex due to the large number of objects and relations to be modeled. Moreover, the ambiguous character of the relations between objects provides large difficulties in identifying objective and generic rules. Spatial multi-objective optimization modeling in spatial planning problems, as linear integer programming, genetic algorithms and simulated annealing, have a strong theoretical base and are applied frequently in spatial planning literature to provide ‘the most favourable’ landscape and plan layout in terms of minimal development costs. More recently, also general spatial shape objectives are included in the multi-objective functions devised. The research objectives in these studies however, are often restricted to a level of layout planning which is less detailed than the objective stated in this research. A direct consequence is that shape objectives are in general terms of compactness and solely defined at the land-use class level. Furthermore, the number of land-uses to be allocated and the site to be modelled is kept relatively small. These features are enough to provide a proof of principle, but not to deal with realistic planning challenges. Cellular automata and multi-agent systems provide robust frameworks to realistically model subject and object interactions in space and time. However, the non-deterministic behavior and outcomes of the model runs make them less suitable to generate plausible landscape configurations as defined in this research. Chapter 3 describes the (development of the) landscape generator, that is compatible with the regional plan scenario development approach identified in Simlandscape. The landscape generator uses landscape types as building blocks of plan scenarios. A landscape typology describes a proposed future spatial development and contains spatial and (non)spatial (descriptive) attributes. A 2D reference image indirectly provides objective compositional and configurational characteristics of the proposed development. In essence, users allocate a landscape typology to a cadastral lot typology and based on this information, the landscape generator produces a comprehensive landscape configuration. The landscape generator is developed as a multi-objective heuristic optimization modeling approach. In this approach a sequentially updated multi-objective function is optimized for a two-dimensional allocation site. It is assumed that the site is homogeneous in physical characteristics (e.g. height, soil etc.). The multi-objective function is compiled from an available library of single spatial attributes. These spatial attributes and their target values are retrieved from the compositional and configurational characteristics present in the reference image of the landscape typology. Examples from the available spatial attributes are the number of landscape component instances, the relative size of each component or each component instance, compactness and shape of component instances and direct adjacency between two different landscape components. In a hypothetical case study, the capabilities and behavior of the landscape generator are demonstrated. In the case study, the landscape generator generates a variety of landscape configurations for a hypothetical allocation site (20x20 cells) and a rural forest estate as allocated landscape typology. The reference image of the rural forest estate provides detailed information for the compilation of the multi-objective function. The landscape generator contains probabilistic elements (e.g. random starting situation, near-random cell swap), which results in different output, each time it is run with identical input settings. The landscape generator is capable of producing a range of landscape configurations for a variety of situations. A unique situation is defined by the allocation of one landscape typology to one allocation site. Theoretically, since the method is based on the objective measurement of spatial characteristics present in a reference image, each user-defined typology can be used for a selected allocation site. The landscape typologies cannot be allocated to every imaginable dimensioned allocation site, but are bounded by the spatial extent which specifies a valid spatial extent. At the heart of the method lies the compilation of the multi-objective function. Ideally, this compilation can be executed completely objectively and without user-interaction, as the reference image of the landscape typology provides the required information. In the current prototype version of the landscape generator, however, the compilation process is partly (and in advance) controlled by the modeler. The modeler needs to specify which of the available spatial attributes to include, in which sequence to optimize them and what attribute target values to specify. Surely, the modeler is informed by statistics calculated for the reference image. An important task is to define consistent guidelines for the compilation of the multi-objective function from each landscape typology, irrespective of the properties of a valid allocation site. In this research, the modeler has been able to define specific guidelines for each landscape typology. In the current state of the method, a continuous assessment, through iterative testing, needs to be made by the modeler, about which compilation is sufficient in producing plausible configurations and which compilation process produces solutions within reasonable computation times. In chapter 4, a method is presented to obtain insight in the usability of the landscape generator. The produced landscape configurations are extensively evaluated in an extensive internet-based validation experiment. For a broad variety of different situations, landscape configurations are generated by the landscape generator for realistically dimensioned and enclosed sites. The configurations are compared with professional hand-drawn configurations, by a large group of planning professionals. The subjects are provided an interactive, user-friendly web-based inquiry, in which they are requested to (graphically) rank order a random selection out of a total set of landscape configurations (hand-made or computer-generated), from ‘most to least plausible’. The population is not informed about the difference in production process of each landscape configuration. In the experiment a distinction is made between subjective and objective plausibility, representing design quality aspects and representativeness of the landscape typology respectively. Eight different situations (three subjective and five objective) are assessed by the group of respondents and analyzed with a modified version of an approved statistical method, known as ‘the law of comparative judgement’. In addition, to indicate points of interest for further improvements of the methodology, implicit and explicit dimensions of evaluation used by the respondents for each of the objective assessments are identified. The implicit dimensions are identified using linear regression analysis, with single spatial metric properties of the configurations as explanatory variables. To identify explicit dimensions of evaluation the respondents are asked for two of the earlier presented situations, to select five pre-defined used dimensions of evaluation. The current experiment setup provides a robust method as well as reliable results about the capability of the landscape generator to produce plausible landscape configurations. With its modern interactive web interface, its well-balanced data scheme (randomness, several situations) and the use of approved statistical methods, the experiment finds a balance between maximum effective information retrieval and an acceptable level of user workload. In chapter 5, the results of the validation experiment are presented and in chapter 6 these results are analyzed. For each of the three assignments of the design quality test, it is concluded that the whole set of computer-generated configurations is not of comparable design quality as the whole set of professional configurations. Several individual computer-generated landscape configurations have comparable design quality as the professional configurations. The landscape generator is able to produce configurations with landscape components which are with respect to its individual area, shape and relative adjacency plausible. The overall structure is, however, often perceived as near-random. In some situations this is regarded plausible, while in other situations it is regarded implausible. The results of the four analyzed assignments of the representativeness test show a more favorable view on the capabilities of the landscape generator. In half of the cases, the whole set of computer-generated configurations are considered comparable in representativeness to professional onfigurations. In the other half, several individual computer-generated are considered of comparable representativeness. The representativeness test is most important in plausibility validation of the landscape generator, as the primary objective of the research implies that each actor (with different levels of design experience) should be able to provide her development idea (described in the landscape typology) as a comprehensive visualization in an integrated plan scenario. In the initial planning phases of application of the landscape generator, it is more important to obtain a first impression of the impacts (visual and analytical) of a plan scenario than a completely well-modeled and calculated landscape design. Possible non-professional design choices in a landscape typology can be reflected in the generated landscape configurations. Analysis to dimensions of evaluation gives insight into possible explanations for the plausibility ordering of the subjects.A distinction is made between explicit and implicit dimensions of evaluation. Explicit dimensions are directly assembled in the experiment and provide perceived dimensions of evaluations. The implicit dimensions, identified with linear regression analysis are however uncertain in its reliability and ideally should be assembled in relation to explicit dimensions. Results of the linear regression analysis can direct future research with different approaches. First, the attribute target values in the current compilation can be re-specified. Second, non-used but available spatial attributes can be added to the multi-objective function. Third, new spatial attributes may be developed to be included in the optimization process. In light of the main objective in this research, it is important to define consistent guidelines for generating landscape typologies for different situations. In this research, a start is made to identify important choices with respect to the minimal selection of spatial attributes, the influence of its sequence and feasible attribute target value specification. The experiment results further provide detailed directions for improvements of the landscape generator. Other recommendations put forward in this research are related to: 1) the modification of the current heuristic approach (for performance improvement and local trapping avoidance purposes) by hybridization with existing heuristic approaches as simulated annealing and evolutionary algorithms, 2) full-automatic translation from the main characteristics of a landscape typology into the compilation of the multi-objective optimization function; this translation should be as generic as possible and the resulting configurations should be thoroughly validated for plausibility for a variety of possible representative situations (i.e. combination of proposed landscape typology with typical influential allocation site characteristics), 3) extending, if possible, the current library of available spatial attributes with functions that describe more overall organizational properties of landscape typologies or investigation of (parallel or sequential) optimization at different scale levels, 4) the inclusion or extension with representative infrastructure generation and 5) the increase in the effectiveness of the validation experiment by standardizing the acquisition of professional configurations (e.g. designing materials, formats and conditions and automation of conversion to images used in the inquiry) and 6) increase in the reliability of the validation experiment by separating the different parts of the experiment according prioritisation of experiment objectives.
|Qualification||Doctor of Philosophy|
|Award date||27 Apr 2011|
|Place of Publication||Eindhoven|
|Publication status||Published - 2011|