1. INTRODUCTION

2. TYPES OF MODEL

3. TECHNOLOGY FOR MODELING

4. MULTICRITERIA METHODS

5. ACCURACY AND VALIDITY

An overworked term

data model

a template for data
relational, object-oriented, coverage, shapefile
concerned with form, how the world looks

represented by the database

model

a representation of some real-world process
concerned with process, how the world works

represented by software

spatial modeling

manipulation of geographic information in multiple steps
steps may represent stages in some complex analysis

calculation of indicators

steps may represent time

dynamic model
iterative analysis

looping

geocomputation

analog or digital?

analog

use of a scale model
or some analogous process

Varignon frame
electrical flows to simulate water flows

digital

process represented in 0s and 1s
program in C
GIS script in VBA, Python

scale of a digital model?

spatial, temporal resolution
define what is left out of the model
leave uncertainty about model predictions

ideally, the scale of the model should match the minimum resolution of important system processes

models are often built at coarser scales in order to

reduce model complexity to a managable level

match the scale of available data

scaling data down for use in a finer-scale model is statistically problematic

model components may operate over more than one scale

data may also be entered at multiple scales

model must run faster than the real world

positive or normative?

positive: emulates real-world processes

normative: represents human ideals, desires

Why model?

to support some design process

to allow the user to experiment with a replica

to investigate what-if scenarios
to examine dynamic outcomes

evacuation scenarios

Oakland hills fire

footprint

Tom Cova's analysis

Church's simulations

the Paramics polyline problem

analysis

static, one point in time
search for patterns, anomalies
generating hypotheses
revealing what would otherwise be invisible

modeling

multiple stages
perhaps different points in time
implementing ideas and hypotheses
experimenting with policy options, scenarios

better than experimenting on the real thing

Is a model spatial?

four possible tests

invariance test

spatially explicit models (SEM) are not invariant under relocation of the objects of study

representation test

SEM include representation of location in their implementations

formulation test

concepts such as location or distance appear directly in the model

in algebraic expressions or behavioral rules

outcome test

spatial structures of inputs and outputs are different

it modifies the landscape on which it operates

Static models and indicators

combining GIS layers through overlay

e.g., using ModelBuilder

Universal Soil Loss Equation

A = R x K x LS x C x P

DRASTIC model of groundwater vulnerability

karst groundwater protection model

Rhonda Pfaff and Alan Glennon

online tutorial

ModelBuilder screen

results screen

Canadian case study

landslide risk in Quebec

marine clays

La Baie, Quebec

10km by 6km area

pixels 5m by 5m

1,954,836 pixels

73 landslides in past 40 years

independent variables determining risk

geology

12 lithologic groups

forest cover

binary

elevation

slope angle

aspect

two sets of locations

landslide, no landslide

discriminant function analysis

find the combination of the independent variables that best discriminates

linear combination

a1 x1 + a2 x2 + ...

is it possible to model every individual element in the system?

every molecule of groundwater?
every person in a crowd?

autonomous agent models

crowd behavior

the Hajj

panic and massive loss of life

the Notting Hill carnival

orderly
breakdown

Cellular models

on a raster

each cell in one of a number of states
rules of state transition at each timestep

based on states of cell and neighbors
the Game of Life

Keith Clarke's urban growth model

Vector or raster?

vector modeling

less computationally intensive

more difficult to program outside a GIS

more accurate data representation

raster modeling

higher computational and data storage demands

computationally easier to subdivide and aggregate parcels

easier to program outside a GIS

difficulty representing topological relationships inside GIS

rasterization introduces more errors

Graphic interface

ModelBuilder

access to all ArcGIS functions
looping

ArcGIS

Visual Basic for Applications
Perl
Python
JScript

ArcScripts

Model coupling

linking model software to GIS

e.g. Paramics and GIS

loose coupling

exchanging files

close coupling

common files, common interface

Multiple factors affect decisions

three in the Pfaff and Glennon case

slope > 5%
land use = cropping
distance from stream < 300m

simple binary decision

overlay three layers

how to assign weights to each factor?

stakeholders may disagree on weights

MCDM = multicriteria decision making

devised by Thomas Saaty

each stakeholder compares each pair of factors

assigns comparative weights

e.g., slope 7 times as important as land use
e.g., distance from stream 1/2 as important as slope

forming a complete matrix
 

	slope	land use	distance from stream
slope		7	2
land use	1/7		1/3
distance from stream	1/2	3

analyze to obtain consensus weights

example application

How to know if the model is correct?

can its results be trusted?

results from a computer are often trusted implicitly

how to calibrate the model?

to make its results match observation

e.g., to make the results of the Clarke model match previous growth patterns in SB

a model is never more than an approximation to reality

but how good/bad is the approximation?

important to provide measures of confidence in results

varying the inputs to observe effects on outputs

some inputs affect outputs more than others
these are the inputs that most need to be correct

examining the impacts of input errors on outputs

mostly by simulation