Prof Ni-Bin Chang talked about the urban growth model to be adopted in the "Flood impact assessment in mega cities under urban sprawl and climate change" project.
Pests of soyabean_Binomics_IdentificationDr.UPR.pdf
Urban Growth Model
1. Flood Impact Assessment in Mega Cities under
Urban Sprawl and Climate Change (Part I)
Ni-Bin Chang, Ph.D., P.E.
Director, Stormwater Management Academy
University of Central Florida
July 6, 2015
2. Current Relevant Research At UCF
• “Coupling Risk and Resilience Assessment for Networked Sustainable
Drainage Systems in a Coastal City under Climate Change Impact”
funded by NOAA Florida Sea Grant.
• “Developing a Sustainable Hong Kong through Low Impact
Development: from Science to Innovation Policy.” funded by Hong Kong
Research Council.
• “Flood impact assessment in mega cities under urban sprawl and
climate change” submitted to British Council, Global Innovation
Initiatives Grant Program.
3. Historical satellite
imagery analysis
Urban growth
model
Detailed hydraulic
model
Fast hydraulic
model
Urban extents
Impervious areas
Future land cover
classification
Detailed hydraulic
model
Fast hydraulic
model
Future climate
scenarios
Future flood impact
Adaptation
strategies
Future urban
extents
Land cover
classification
Future impervious
areas
Flood impact
Current climate
scenarios
Correlation analysis
between r modelling
results at regional
and city scales
Historical social-
economic data
Trend analysis
using Big Data
Future social-
economic state
AI for pattern
recognition
Trend
analysis
using Big
Data
The Framework for Analysing Future Flood Impact under
Urban Growth and Climate Change in Mega Cities
5. Four Types of Flooding in Coastal Cities : Three Mega Cities
• Coastal flooding : It affects areas along the ocean, bays, rivers,
streams, or estuaries of tidal influence of tidal influence and storm
surge.
• Tidal flooding: Sea level fluctuates daily due to gravitational forces
and the orbital cycle of moon, sun and earth. Flooding from high tide
in low lying area is an issue.
• Riverine flooding: Flooding occurs when freshwater rivers and
streams exceed local flow capacity and water spills over their banks.
• Inland flooding: Flash floods can be caused by short-term, high-
density rainfall, often associated with sudden thunder storms,
hurricanes, or large scale storms.
6. Planning Framework for Vulnerability Assessment
Source: Climate Risks and Adaptation in Asian Coastal Mega Cities, World Bank, 2010
7. Types of Urban Growth Models
• Land Use Transportation Models – Top down models : They
dealing with location and interaction, transport and the urban
economy, represented at a level of abstraction involving
administrative rather than physical subdivisions of the city.
• Cellular Automata Models (CA) – Bottom up models : They
dealing with urban growth sprawl, land development and land
cover, represented at finer spatial scales defined by or
detecting physical morphology, do not deal with explicit
transportation; dynamic in time.
8. Types of Urban Growth Models
• Land Cover Models (LUCC) : They simulate vegetation cover,
ecosystem properties, agriculture, as well as some urban
dynamics.
• Agent‐Based Models (ABM) : They are a generic style of
representation for individual‐based dynamics processes, such
as movement of individuals and objects.
9. Three Generalizations of Urban Structure
• Upper Left: Burgess'
Concentric Zone Model;
• Upper Right: Hoyt's Sector
Model;
• Bottom Left: Harris and
Ullman Multiple Nuclei
Model.
Sources: Graphic repared by Department of
Geography and Earth Sciences, University of
North Carolina at Charlotte.
Beijing New York
Longdon
10. The Cellular Automata Approach: Urban Growth and
Complexity Theory
• These CA models have found
favour in rapidly growing
systems which are characterised
by urban sprawl, like Phoenix,
Las Vegas, Taipei and Beijing.
• They have been quite
inappropriately applied to
non‐rapid growth cities where
the focus is on redistribution.
1
(A)
(B)
11. Logistic Cellular Automata (CA) Models for Urban Planning
• Such models view cities as complex systems based on the
principle of self-organization.
• Cell, state, neighborhood, and the transition rule are the primary
components in CA models.
• The state variation of a cell depends on its previous state and
those of its neighbors.
• The change of state for each cell is controlled by a set of
transitional rules (functions) that are assessed at each time step.
• Transitional functions can be either deterministic or stochastic,
and time is in discrete steps.
12. Formulation of a Logistic CA Model
• A reference set of cells, usually a raster grid of pixels covering an urban
area;
• A set of states associated with the cells at any given time, which can be
in the detailed land uses such as {urban, forest, agricultural, wildland,
wetlands, water};
• A set of rules that govern state changes over time;
• An update mechanism, in which rules are applied to the state at one
time period to yield the states of the same cells in the next time period;
and
• An initial condition of the framework is required and a boundary
condiiton may be present.
13. Assumptions of Traditional Logistic CA Models
for Urban Planning
• The underlying plane is homogeneous Cells don't have intrinsic
properties.
• Transition rules must be uniform, and they must apply to every cell,
state, and neighborhood.
• Every change in state must be local, which in turn implies that there is
no action-at-a-distance effect.
• All these features operate uniformly and universally (i.e. each cell is
an automata).
14. Cellular Automata (CA) Models for Urban Planning:
the 1990s and before
• Before the 1990s, CA models mainly have two featurers:
1. Land use allocation and other geographic factors with the dynamic approach of the
CA model are considered in this period.
2. Lacking of software to deal with extensive lattices.
• In the 1990s, Dynamic Urban Evolutionary Modeling (DUEM):
• Developed with the application of other software such as GIS to demonstrate the
hypothesis.
• To maximize the use of GIS approach to visualize urban simulations.
• Language: C++, CDL
• Xie/Batty– Ypsilanti/London, US/UK–DUEM
Batty et al., 1999
15. Case Studies in the 1990s
• In 1992, Dublin was selected as a
case study urban city that is simulated
30 years from 1968 to 1998 using a
GIS-based CA software prototype.
• Calibration is included by means
of the fractal dimension and the
comparison matrix methods.
• The simulation results are
relatively accurate.
Jose et al., 2003
16. The Fractal Dimension and Urban Growth
In fact in mathematics, a function is scaling if it can be shown to be scalable under a
simple transformation – i.e. if we can scale a distance by multiplying it by 2 and the
function does not change qualitatively, then this is scaling – so power laws – functions
like f(y)=x‐1 scale because if we multiply by 2, say, we get f(2y)= 2x‐1 =2‐1x‐1~f(y)
Berlin, 1875 Berlin, 1920 Berlin, 1945
Hern, 2008
17. Cellular Automata (CA) Models for Urban Planning:
at the end of the 1990s
• Cellular automata models were used to simulate urban dynamics
through GIS-based approach.
• AUGH model (generalised urban automata with the help on-line) and other
GIS-based models were developed around this time.
• Calibration and prediction results were achieved.
• The model was expected to simulate the urban growth process and provide
long-term predictions for urban planning.
• Language: C, PERL
• Data: historical digital maps
• Testing region: Marseilles region (Meaille & Wald, 1990), Cincinnati (White &
Engelen, 1993), the Bay Area (Clarke et al., 1997), the Washington/Baltimore
corridor (Clarke & Gaydos, 1998), Guandong (Yeh & Li, 1998), and Guanzhou
(Wu, 1998),
18. • At the beginning of 2000’s, different types of computer languages and
tools were applied to build CA models, such as C language, Java,
matlab, and so forth.
• Models:
• CLUE (Conversion of Land Use and its Effects) Model – University of
Amsterdam, The Netehrlands
• SLEUTH Model – UC Santa Babara, USA
• ANN - SLEUTH CA Model – UC Santa Babara, USA
• Metronamica Model – Research Institute for Knowledge Systems (RIKS), The
Netehrlands
• JCASim Model – Technical University Braunschweig, Germany
Cellular Automata (CA) Models for Urban Planning:
the 2000s
19. Overview of the CLUE Modelling Procedure
The model is sub-divided
into two distinct modules,
namely a non-spatial
demand module and a
spatially explicit allocation
procedure.
Curtesy of Peter Verburg
20. Illustration of the translation of a hypothetical land use
change sequence into a land use conversion matrix
Overview of the Information Flow in the CLUE-S
Model
Curtesy of Peter Verburg
Two sets of parameters are needed to characterize
the individual land use types: conversion elasticities
and land use transition sequences.
21. Flow Chart of the Allocation Module
of the CLUE-S Model
22. How Does SLEUTH Simulate Urban Growth and Land
Cover Change?
• Coefficients : Five coefficient, or parameter, values effect how the growth
rules are applied. These values are calibrated by comparing simulated land
cover change to a study area's historical data.
• Growth rules : SLEUTH begins with a set of inital conditions which is the
input data configuration of the landscape. A set of decision, or growth,
rules is then applied to the data to simulate urban driven land cover
change.
• Self modification: The coefficients do not necessarily remain static
throughout an application. In response to rapid or depressed growth rates,
the coefficients may increase or decrease to further encourage growth rate
trends.
23. The Existing Coefficients of SLEUTH Model
• The calibration process is automated, so SLEUTH “learns” the best set for
any given application from the data (slope, land use, exclusion, urban
extent, transportation, hillshade).
• The parameters were chosen after extensive testing by trial and error. They
include
• parameters that control the random likelihood of any pixel turning urban
(dispersion),
• the likelihood of cells starting their own independent growth trajectory (breed),
• the regular outward expansion of existing urban areas and infill (spread),
• the degree of resistance of urbanization to growing up steep slopes (slope) and
• the attraction of new development toward roads (road gravity).
24. • Markov-CA model:
• Goal:
• Analyze temporal change and spatial distribution of land use influenced by the natural
and socioeconomic factors
• forecast the future land use changes
• Approach: GIS
• Calibration: included
• Parameters: agriculture land, forestland areas, and upward trend in built-up
areas
• Restriction: the land use dynamics changes of the social and environmental
interactions among people are not considered in this model.
Cellular Automata (CA) Models for Urban Planning:
from 2010 to the Present
25. Case Studies in 2010 and after
• Markov-CA model
• Case study city:
• Saga, Japan
• Fangshan, a district of Beijing, China
Guan et al., 2011
26. Cellular Automata (CA) Models for Urban Planning:
from 2010 to the Present
• AIS-Based CA model:
• Self-adaptive CA model (an artificial immune system)
• Goal: simulate the rural-urban land conversion
• Parameters are allowed to be self-modified
• Can be used to retrieve the changing urban dynamic evolution rules over
time.
• Data: Landsat TM satellite image from 1995 to 2012
• Case study city: Guangzhou, China
• Comparison between the AIS-based model and a Logistic CA model: The
results indicate that the AIS-based CA model can perform better.
• Advantage:
• Perform better and higher is precision in simulating urban growth
• The simulated spatial pattern is more close to the real development situation.He et al., 2015
27. Case Studies in 2010 and after: AIS-based Model
Urban evolution process of Guangzhou
city during the period 1990-2012
Simulation results of Guangzhou city during
the period 1990-2012 with the AIs-based CA
He et al., 2015
28. Case Studies in 2010 and after: Urban Growth in Beijing City
• This study applied the model to assess the general urban
development plan entitled "disperse polycentric urban
development plan" of Beijing City and found that the plan failed
to meet its objectives.
29. CA-based Urban Growth Model in Beijing: 1975-1997
Source: Chen Jin, Gong Peng, He Chunyang, Luo Wei, Tamura Masayuki, and Shi Peijun,
Assessment of the Urban Development Plan of Beijing by Using a CA-Based Urban
Growth Model, Photogrammetric Engineering & Remote Sensing, October 2002, 1063-
1071.
30. CA-based Urban Growth Model in Beijing: 1975-1997
• The transitional function is the core of CA models.
• There are two groups of factors in the transitional function.
• The first group includes local factors, such as interactions between adjacent
land uses.
• The second group includes broad-scale factors such as regional interactions
based on transportation networks.
• The following modifications to the formal CA framework to reflect the
realistic situation:
• External land demand control
• Transition potential from non-urban land to urban land based on land
suitability and neighborhood effect
• Definition of neighborhood effect was relaxed to involve the more distant
influence of neighbors
31. Unique Features
• An adaptive Monte-Carlo method
was used to automate the
calibration of factor weights used
in the CA transitional rules.
• This study used one scene of
Landsat MSS imagery from 1975
and three scenes of Landsat TM
imageryfiom 1984, 1991, and
1997 to classify the land-use
patterns.
32. Unique Features
• Constrained Condition: w𝑚
𝑘=1 k=100
• Objective function: Max F(w,, w2, ..., wm,)
where wk > 0, and F is a fitness function between simulation results
and the actual situation
• The objective is to find optimal weights so that a fitness index reaches
its maximum. This inverse problem can be solved using an adaptive
Monte Carlo method
34. Tietenberg Model
• According to Tietenberg (1992), land
resource can be treated as a depletable,
non-recyclable resource.
• Its demand and supply are influenced by
price.
• Thus, the optimal allocation of land
resources is to maximize the net benefit.
• The maximum net benefit can be obtained
when the marginal benefit function is
equal to the marginal cost function.
36. From Theory to Practice
• Because the marginal benefit falls as land consumption or land
consumption per capita increases, the marginal benefit function in
year t can be given by assuming the land demand curve is linear and
stable over time (Tietenberg, 1992).
• Population and economic growth driven by the development of the
tertiary industry and infrastructure construction propelled
urbanization as a whole.
• Factors such as traffic condition, distance to central city, slope, and so
on determined the spatial distribution of urban growth.
37. Tietenberg Model
• To execute the Tietenberg Model, increased population in the future
is needed.
• By using the Logistic regression, based on the population data in the
history, the increasing curve of population can be calculated as
follows
38. Modelling Structural Change in Spatial System Dynamics
• System dynamics (SD) is an effective approach for helping reveal the
temporal behavior of complex systems.
• This is especially true for models on structural change (e.g. LULC modeling).
• A Python program is proposed to tightly couple SD software to a
Geographic Information System (GIS).
• The comparison of spatial and non-spatial simulations emphasizes the
importance of considering spatio-temporal feedbacks.
• Practical applications of structural change models in agriculture and
disaster management are proposed in a spatial system dynamics (SSD)
environment.
Neuwirth et al., 2014
42. Modifications of Traditional Logistic CA Models
for Urban Planning
• CA models are often relaxed to adapt to real problems at hand.
• Common relaxations include
• adopting heterogeneous underlying planes;
• extending the immediate neighborhood definition from a Moore or
Neumann neighborhood to a larger extent;
• incorporating action-at-a-distance effects, or broad-scale factors, etc.
• Use of Adaptive Monte Carlo Simualtion or ANN/AIS model to determine
the paratemetrs.
• These modified CA models are easy for integration with GIS and remote
sensing algorithms also facilitates their implementation.
• The structure dynamic change in a spatial system dynamic environment
was developed.
43. Unsolved Issues and Problems
• Almost all variance captured and measured in Monte Carlo simulation
is contained in the first few iterations, and that increasing the number
of iterations quickly has diminishing returns in terms of model fit.
• Modelers lack of attention in spatial modeling to the idiosyncrasies of
pseudo-random number generators - the lack of repetitive cycling in
the random numbers, and the ability to replicate sequences across
computational platforms.
• Memory effect - the persistence is both of type (i.e. which land use
transition changed to which) and time, since changes are spatially
autocorrelated in time and space
44. Unsolved Issues and Problems
• The fourth behavior type of SLUETH simulated is “road gravity”, in which
new growth is attracted to and allowed to travel along the road network. It
would be of interest to determine is the value changes over time, over
space, or with transportation technology.
• The remaining constants in SLEUTH all determine how the model
implements self-modification. Self-modification is macro-scale behavior. It
lacks sensitivity when tuning them one by one in sequence.
• Load balance in parallele computing when more models/tools need to be
integrated
• Big data analytics may need to be in place in support of the urban growth
model.
47. Multi-temporal Change Detection of Land Use Using Remote
Sensing
• The location of the study area and the corresponding SPOT-5
images in 2003 and 2007.
Ground truth Database
Training Dataset Testing Dataset
PL-ELM Classifier
Feature Extraction
Field Trips
LULC Class Definition
The PL-ELM Classifier
1
2 2
3 3
4 4
Major experimental steps:
1. Extract multiple features from the original remote sensing images;
2. Construct the training data set and testing data set based on the ground truth data base;
3. Train the classifier with the training data set, and test its performance with the testing data set;
4. Classify the full scale image of the study area using the PL-ELM classifier.
50. Novelty of This Study
• Strict CA are models whose rules work on neighbourhoods defined by
nearest neighbours and exhibit emergence – i.e. their operation is
local giving rise to global pattern.
• Neighbourhoods can be wider or they could be formed as fields – like
interaction fields around a cell - like interaction fields around a cell.
• Cells are irregular and not necessarily spatially adjacent.
• Structure dynamic changes may be explored by using Stella.
51. Modeling the Spatial Trasition Rules by Gravity Theory
• According to the gravity theory proposed by Newton in 1687, the
attraction Fik between two objects i and k can be briefly formulated
by their masses and the distance between them and expressed as
thefollowing equation:
in which Mi and Mk are the mass of object i and k, respectively; Dik is the
distance between object i and object k; G stands for gravity factor.
52. Modelling the Spatial Trasition Rules by Gravity Theory
• It is determined by the distance (Dij) between jth cell of major land use
change (Aj) and and ith cell tat may be influenced by the changeover
time.
• Following the gravity theory, this study assumes that the decay rate of
a crowd due to such a land use change follows the Inverse Square
Law.
• Concerning the diffidence among various types of land use changes,
the preference are clustered into four groups.
Gij = f (Dij, Aj)
53. UGM Calibration and Validation Using Remote Sensing
Hindcasting
Nowcasting
Forecasting
Yeh and Li, 2002
Current Trend
Managed Growth
Ecologically Sustainable
54. How Do Low Impact Development (LID) Technologies Come to
Help?
# Introduce a spatially-explicit
approach to assist landscape
architects, urban planners, and
water managers in identifying
priority sites for LID.
# Examine the current flood
proofing facilities to public utility
department in identifying priority
sites in response to sea level rise,
storm surge, and storm tides.
55. Risk & Resilience: A Systems Approach for Water Security
• Sustainable stormwater management mimics nature by integrating
management of stormwater runoff into the surrounding terrain, using
systems like landscaped medians, swales and interchange areas to store
and treat runoff.
57. Science Questions
• How can neighbourhood interactions and inherent constraining and
enhancing factors for urban development be extracted and related to
actual changes in land use patterns?
• How can scenarios of planned and unplanned growth be created and
used for evaluating policy options?
• How to connect data driven model with knowledge drivenr model to
closely capture the spatial an dtemporal dynamics?
58. Challenges in Synergistic Research
• Integration between socioeconomical development, smart
growth, and urban growth model for different mega-cities.
• Integration between the CA-based urban growth model
(UCF) and the CA-based flood impact assessment model
(Exeter).