Evolving Cities

Geocomputation in Territorial Planning

Lidia Diappi

• Publisher: Routledge
• Year: 2017
• Print ISBN: 9780754641940
• eBook ISBN: 9781351937962
• Topic: Computer Science
• Subtopic: Certification Guides in Computer Science
• Index: Computer Science

Geocomputation has come of age. The whirlwind of change experienced in Geographical Information Science (GIS) - developments in IT, and new data gathering and earth observing technologies - has taken GIS beyond mere data and towards its analysis, modeling, and use in problem solving. Geocomputation is now at the dynamic edge of this revolution. Bringing together the leading researchers in geocomputation, this volume provides an up-to-date overview of the development of new artificial intelligence principles and technologies (NN, CA, Multi-agent Systems and Evolutionary Algorithms) used for the analysis, development and evaluation of urban planning policies and programmes. Charting the new approaches to data-processing, the book provides pointers on how to harness these technologies, advancing the knowledge level of planning by multiplying the information capacity of GIS, and offering a new approach to territorial modeling and micro-scale descriptions of socio-economic, behavioural and micro-spatial theories of urban processes and land use change.

Chapter 1

Introduction

Lidia Diappi

We seek a learning system that turns expert behavior into fuzzy rules. We seek a system which learns fuzzy rules from experience.

Bart Kosko

Emergent Territorial Phenomena

There have been a number of significant changes affecting the field of territorial planning which have become increasingly evident over recent decades.

The present work identifies some of the main changes.

• The first distinctive feature is the growing dispersion of settlements. This phenomenon, described as diffused city (Secchi, 2001), metapole (Ascher, 1995) or urban glue (Hall, 1999) does not so much involve a spreading outwards of large cities, but rather a gradual urbanisation of the territory between cities. Understanding the factors underlying these new locational impulses and finding their hidden aspects is an important challenge inspiring various contributions to this book, particularly in the second part.
• The increasing mobility within this new urban morphology has occurred along routes which follow a process more closely resembling percolation in a porous fragmented fabric (Secchi, 2001) than the orderly flows channelled into road networks, as traditional transport models interpret.
• The new settlement processes and greater mobility can also be described in terms of increasing globalisation. But globalisation also results in greater emphasis on the local, since the process is driven by differences. The diffused city and globalisation induce a mutually supportive process of standardisation and differentiation. Standardisation occurs as one tends to find the same type of economic actors, services and opportunities in every city and country. Differentiation occurs because what drives the circulation of people, goods and ideas is ultimately the difference between territories.
• The increasing autonomy of individuals means there is a proliferation in choices and behaviours, even though they may be expressed within increasingly socialised and complex systems. The flexibility in working hours, the increasing autonomy of family members, new technologies, from mobile phones to portable PCs, enable ever increasing individual management of time and space.

This does not only mean that urban planning practice has had to address new priorities. It has also had to examine new techniques for analysis and evaluation, together with procedures for monitoring and control. In a situation marked by growing uncertainty, forecasts are inevitably subject to increasing limitations. The conception and implementation of design and management principles in planning therefore require much better understanding of phenomena than in the past, with the factors affecting the evolution of the city and society being identified. New hypotheses, new models and a new scale for examining phenomena are needed. This can be effectively achieved by making use of the large databases in high-dimensional space which are now available. It is necessary to once again observe, investigate behaviours, classify and create models starting from observations on an individual level. This book provides a clear demonstration of how this can be done, by integrating GIS (Geographical Information Systems) databases with the powerful tools offered by the new processing abilities of the Geocomputation approach.

What is Geocomputation?

This book aims to show the potential for building knowledge of territorial phenomena using the new tools of Geocomputation (Openshaw and Abrahart, 2000; Fischer and Leung, 2001; Longley et al., 2001).

Geocomputation (GC) is an emerging paradigm which has the potential to dramatically improve the effectiveness of urban studies through the use of computational intelligence technologies (CIT).

The most widely accepted definitions of GC emphasise the role of new technologies in processing large databases. For Stan Openshaw (2000), father of this neologism, GC is “an approach based on high performance computing to solve currently unsolvable or even unknown problems”.

Longley defines GC as “...(the) application of computationally intensive approaches to the problem of physical and human geography. In some important respects the term GC is synonymous with geographic information science, although it has often put greater emphasis upon the use of high performance computers” (Longley et al., 2001).

For Helen Couclelis GC is “...the eclectic application of computational methods and tools to solve geographical problems and to explain geographical phenomena” (Couclelis, 1998). A more specific definition is given by Fischer and Leung who, in addition to the greater computational efficiency and “fuzziness” provided by computational intelligence technologies, state that GC can:

improve the quality of research results by utilizing computationally intensive procedures to reduce the number of assumptions and remove the simplifications imposed by computational constraints that are no longer relevant (Fischer and Leung, 2001, p. 5).

In our view Geocomputing includes approaches to human reasoning that try to make use of the human tolerance to incompleteness, uncertainty, imprecision and fuzziness in decision, making processes. In addition to neural networks and adaptive fuzzy systems, it also incorporates evolutionary computation, cellular automata, expert systems and probabilistic reasoning. Geocomputing is especially concerned with combinations of these methodologies and introduces the spatial dimension developed and structured by GIS into soft computing techniques (Zadeh, 1965).

Many innovative features characterise this approach:

• The first concerns identification of the rules. The well established body of theories and models developed since the beginning of the 1960s is based on explicit rule formulation of assumptions derived deductively from theories. The goal of learning is to formulate explicit rules (propositions, hypotheses, etc.) which generalise in a very succinct manner. Powerful mechanisms, with considerable innate knowledge of a domain, formulate general hypothetical rules by analysing particular cases and then formulate explicit generalisations. The Geocomputation approach is completely different, since it assumes that information processing should itself be able to find out the rules through learning.
• The second major difference concerns the scale of description, i.e. the level of resolution of the system. The micro-scale description, which characterises the GC approach, with agents representing individual decision units, is suitable for articulating micro-spatial, socio-economic assumptions and other well formed behavioural theories of urban processes, including land use change. This is because many GC tools use parallel distributed computing, which is particularly suited for describing interactions between subjects. This permits the generation of a new, socially-based type of knowledge which can greatly increase effectiveness in analysis, simulation and planning.
• Finally Geocomputation offers opportunities to reconstruct missing information, because it is based on “...substituting a vast amount of computation as a substitute for missing knowledge or theory and even to augment intelligence” (Openshaw, 2000). Geocomputation is something more than a set of methods and techniques for sophisticated data processing. It is, to some extent, a different approach for understanding phenomena compared to the traditional model building approach.

About this Book

The present volume aims to make a scientific contribution to the Geocomputation approach in urban planning, with a specific focus on the development of Distributed Artificial Intelligence principles and techniques (Werner, 1996) as a support to planning. Neural Networks (NN), Multi-Agent Systems (MAS) and Evolutionary Algorithms (EA), in particular, allow the knowledge level to be increased by multiplying the information capacity of the GIS and offering a new approach to territorial modelling.

PART I: The Spatial Investigation Capabilities of Neural Networks

The first part of the book is devoted to the investigative potential of neural networks. The most prominent feature of NNs is their ability to learn from examples. Using so-called learning algorithms they solve problems by processing a set of training data. Some types of neural network are comparable to statistical regression or discriminant models. However they do not explicitly make assumptions on the distribution of their training data, or on the relationship between their input and output variables. Another basic question refers to the stored knowledge that gives rise to a specified pattern of activation. In parallel distributed processing (PDP) models, the patterns themselves are not stored. Instead, what is stored are the connection strengths between units that allow these patterns to be recreated.

From a statistical point of view, neural networks are non-parametric models, and for some it can be shown that they are universal function approximators. There is a drawback of NNs, which can pose a problem for some applications. In general it cannot be proved that a NN works as expected. Due to its distributed nature, the solution that a NN has learned cannot be expressed explicitly. A neural network learns, but a user cannot learn from the network. For the user it is simply a black box.

The strong points of neural networks are their learning capabilities and their distributed structure which allows for highly parallel software or hardware implementations.

The second chapter by Silvio Griguolo shows the power of neural networks as pattern recognisers. Neural networks are universally recognised as efficient classifiers for multi-dimensional problems involving pattern recognition of massive quantities of data for remotely-sensed imagery.

There may be a range of different objectives such as land cover recognition based on a set of radiometric bands; eco-climatic zoning or crop monitoring based on time series of images, mostly representing some kind of Vegetation Index; or the singling out of specific objects of interest, like roads, building, etc., based on the relationships between the features of pixels and the characteristics of their neighbourhoods.

The approach can be supervised or unsupervised, depending on the problem: neural networks are available for both cases. It is commonly acknowledged that, thanks to their independence from a particular data model, they quite often perform better than statistical classifiers.

In the specific case of land cover classification, it is well known that radiometric information alone is insufficient to achieve soundly reliable recognition. The use of suitable ancillary information is necessary to solve dubious cases, and this is commonly done by defining a specific set of rules that are applied as a post classification stage. It is easy to show that this way of proceeding has some drawbacks that cannot be overlooked.

The paper presents a method of land cover recognition that operates via a supervised neural network and uses the available ancillary information during the assignment process itself and not as a separate step. The approach is operational and is illustrated by means of an application described in detail.

Chapter 3, by Manfred Fischer and Katerina Hlavackova-Schindler, presents two new approaches using neural networks and statistical optimisation to solve the parameter estimation problem, one of the main issues in neural spatial interaction modelling. Current practice is dominated by gradient-based local minimization techniques. They efficiently find local minima and work best in unimodal mmimization problems, but can fail in multimodal problems. Global search procedures provide an alternative optimisation scheme which allows an escape from local minima.

This contribution presents two global optimisation methods, Differential Evolution and Alopex. Differential Evolution was introduced as an efficient direct search method for optimising real-valued multi-modal objective functions. Alopex, the second alternative of an appropriate global search method for spatial interaction modelling applications, was introduced in its original version as early as 1973 by Tzanakou and Hart. Little is known about the behaviour of either global search procedure in real world applications. Both methods were successfully tested on Austrian inter-regional telecommunication traffic data by the authors.

This work evaluates both methods for robustness when applied in the neural spatial interaction context with respect to common benchmark models and measured in terms of in-sample and out-of-sample performance. A benchmark comparison of both methods for robustness and generalisation performance, when applied on a neural network model against backpropagation of conjugate gradients and measured by their in-sample and out-of-sample performance, is based on Austrian inter-regional telecommunication traffic data.

Chapter 4, written by Lidia Diappi, Massimo Buscema and Michela Ottanà, addresses the problem of evaluating the complex facets of urban sustainability in Italian cities.

The question to be posed is the following: is it possible to evaluate urban sustainability in cities with different contexts, with major dissimilarities in terms of environmental conditions, social welfare and economic indicators, as well as dynamics of growth and decline? Since the cities differ with respect to institutional, historical, cultural and economic variables, there is no uniform scale for measuring sustainability and the process of urban ranking is rather arbitrary. With this in mind, suitable indicators to investigate the relationships between the different properties of cities and the various patterns of development have resulted in a better understanding of the influence of different attributes. Since sustainability should be defined as a positive co-evolution of social, economic and environmental systems, the complex interactions among the phenomena give rise to positive and neg...

Table of contents

• Cover Page
• Title Page
• Copyright Page
• Contents
• List of Figures
• List of Tables
• List of Contributors
• Foreword
• 1 Introduction
• PART I THE SPATIAL INVESTIGATION CAPABILITIES OF NEURAL NETWORKS
• PART II LAND USE DYNAMICS THROUGH ARTIFICIAL INTELLIGENCE TOOLS
• PART III MULTI-AGENT SYSTEMS: INTERACTIONS AMONG ACTORS AND THEIR BEHAVIOURS
• Index