Transcript GeoInfo_33_engl - mtc

Spatial Interpolators to generate
Population Density Surfaces in the
Brazilian Amazon: problems and
perspectives
Silvana Amaral
Antonio Miguel V. Monteiro
Gilberto Câmara
José A. Quintanilha
Introduction

Brazilian Amazonia – 5 million km2, 4 million of forest

Deforestation rate 15.787 km2/year

Environment x Life quality

Urban Population 1970 – 35.5%, 2000 - 70%

Health, education and urban equipments - precarious

Planning – consider the human dimension

POPULATION – subject and object of the
transformations ?
GEOINFO – Dez/2002
Introduction

Geographic phenomena – computing representation
models to socio-economic data






Individual
Area
Continuous phenomena in space
Area– discrete region phenomena, homogenous unit
Unit – arbitrary as the census sector – do NOT
represent the spatial distribution of the variable.
Modifiable Area Unit Problem (MAUP) – temporal
series???
GEOINFO – Dez/2002
Introduction

Surface Models – alternatives to Area restrictions




Demographic Density – continuous phenomenon
Objective: to estimate distribution in detail (as better as possible)
Advantage: manipulation and analysis - Area independent
Data storage and accessibility in Global Database

Census Data – Municipal boundaries or census sector

Land use and coverage evolution in Amazonia



Territorial divisions
Regular grid for spatial models
Population pressure – Population density gradient
GEOINFO – Dez/2002
Introduction

Objective – discuss the principal spatial
interpolation techniques used to represent
Population at density surfaces and indicate
the more suitable methods to represent
population in the Amazonia Region.
GEOINFO – Dez/2002
To represent Population in Amazonia…

Data availability

Census Data (10 years)

Inter-census – counting based on sampling


Statistic estimates – PNAD – UF, metropolitan region,
for urban population in the N region
only
Spatial Reference


Municipal limits – up to 2000 census, (analogical maps),
official territorial limit (IBGE) – municipal
2000 census – digital census sector (just to the urban area –
mun. > 25,000 inhabitants)
GEOINFO – Dez/2002
To represent Population in Amazonia…

Census Zone



Surveyed area - 1 month:
350 rural residences
250 urban
Amazonia – vast areas and
heterogeneous
Alta Floresta d’Oeste (RO)


165 km2 and regular boundaries –
settlements
435 km2 in forested areas
GEOINFO – Dez/2002
To represent Population in Amazonia…

Region Heterogeneity

Municipal Dimension: Raposa (MA) - 64 km2,
Altamira (PA) – 160,000 km2

Municipal Area: Average = 6,770 km2, Stand. Dev.=14,000 km2

RO – 52 municipios – average area of 4,600 km2

AM - 62 municipios – average area of 25,800 km2

Municipal area influences the census zone dimension
GEOINFO – Dez/2002
To represent Population in Amazonia…

Process complexity -> spatial distribution

Rondônia: migrants, INCRA settlements, urban nuclei along the
road axis and population at rural zone.

Amazonas: lower urban nuclei density, concentrated in Manaus.

Tendencies:



Dispersion from metropolis,
Increasing relative participation of cities up to 100,000 inhab.
Population growing at 20,000 inhab. nuclei

Dispersal population at rural zone and along river sides

Forest continuous – demographic emptiness
GEOINFO – Dez/2002
Population Models


Human Dispersion:
Important at
regional projects LBA and LUCC
More frequent
representation:
Thematic Maps
GEOINFO – Dez/2002
Population Models



Demographic
Density instead of
Total Population
2000
Visualization:
Intervals and criteria
Highlight: Densely
populated regions
and Demographic
emptiness
GEOINFO – Dez/2002
Population Models

Surface Interpolation Techniques - “Models” –
two groups:

Considering only one variable – POPULATION:


Area Weighted, Kriging, Tobler Pycnophylatic, Martin’s
Population Centroids
Considering auxiliary variables, human presence
indicators:

Dasimetric method, Intelligent Interpolators and variants
GEOINFO – Dez/2002
“Univariate” Population Models

Area Weighted



Population Density proportional to the intersection
between original zones and grid cells.
Sharp limits in the boundaries and constant values inside
the units.
Error increases with:



more clustered distribution,
smaller destiny regions compared to the origin regions
At the Amazonia region –> raster representation of the
Population Density (previous map)
GEOINFO – Dez/2002
“Univariate” Population Models

Kriging



Interpolation for spatial random process. It estimates
the occurrence of an event in a certain place based on
the occurrence in other places.
The variable values are dependent of the distance
between them, a function describes this spatial
distribution.
Using Municipal centres as sample points, taking the
demographic density (log) –> a gaussian function can
model the population spatial distribution
GEOINFO – Dez/2002
Spatial Representation - “Univariate”

Kriging
Imprecision for
modeling
Population
volume
Manaus ->
Pará
Empty areas
RO
 Synoptic vision
 General
Tendency
GEOINFO – Dez/2002
“Univariate” Population Models

Tobler Pycnophylatic



Based on the Geometric
centroids of the census unit
Smooth surface ~ “average
filter”
Weighted by the centroid distance, concentric demographic
density function
Population value for the entirely surface (there is NO zeros)
 Consider the adjacent values and maintain the Population
volume
GEOINFO – Dez/2002
“Univariate” Population Models

Tobler Pycnophylatic





Ex: Global Demography
Project, 9km grid, 1994.
Manaus ->
Municipal Data
Pará
Homogeneous region,
diffuse boundaries
RO – smaller municipios,
interpolator effect.
Better results – smaller
units (census zone) and
high populated areas.
GEOINFO – Dez/2002
RO
“Univariate” Population Models
Based on Kernel




Martin’s Centroids Weighted
Census mapping - UK
Adaptive Kernel: point density
define the populated area
extension
Distance decay function:



Weight for each cell – redistribute
the total counting
Function shape – affects the
distribution of the population over
areas
Rebuild the distribution geography, maintaining areas without
population at the final surface.
GEOINFO – Dez/2002
“Univariate” Population Models

Kernel – 2000




Municipal centres centroids
Gradient at high
populated areas
Demographic
emptiness preserved
Better results:
additional centroids
(districts and RS
images), and smaller
units and densely
populated regions
GEOINFO – Dez/2002
“Multivariate” Population Models



Auxiliary variables - human
presence indicators - to distribute
population
Dasimetric Method – Remote
Sensing classified images –
weights to disaggregate
Intelligent Interpolators: Spatial
information from other sources to
guide the interpolation


A weighted surface map the original
data on the final surface
Predictors variables x interest variables
GEOINFO – Dez/2002
Land use categories
High housing
Low housing
Industry
Open space
Probabilities by raster cell detail
Weights
10
5
1
1
Probability
No intervals
n total weights of
zone
Zonal data to microdata
1483
Data element
100
50
10
Data element
“Multivariate” Population Models

Intelligent Interpolators :



Ex: LandScan –1km grid,
1995
Population Model: land
use, roads proximity,
night-time lights =>
probability coefficients
Population at risk:
information for emergency
response for natural
disasters or anthropogenic
GEOINFO – Dez/2002
“Multivariate” Population Models

Intelligent Interpolators - Variants:

Clever SIM – besides the auxiliary variables, neural
network to:




understand the relations between predictors variables and
population
generate the surface.
Crucial: variable selection and interactions – ”model”
Availability and quality of the auxiliary data ->
responsible for the final density surface precision
GEOINFO – Dez/2002
Perspectives

Density Surfaces in Amazonia:




Interpolator Methods – characteristics e restrictions
Adaptive Approach – based on scale of analysis and phenomena
complexity
Scaling Top-Down
Amazonia Legal:



“Multivariate” models : heterogeneities
“Univariate” Models: Tobler – related to the sampling unit;
Martin – additional centroids; Kriging – general tendencies
=>OK
Kriging including barriers (further)
GEOINFO – Dez/2002
Perspectives
Macro-zones: Spatial-Temporal Subdivision:
I. Oriental and South Amazonia: “deforestation arc”

Martin’s Centroids Weighted– villages, districts, night-time lights
II. Central Amazonia : Pará, new axis region

“Multivariate” Model - intelligent Interpolators

Scenarios Analyze as BR-163 paving
III. Occidental Amazonia : “Nature rhythm”

“Multivariate” Model – Disaggregating by land use (e.g.)
GEOINFO – Dez/2002
Finally
Scale – Census Zones


Tobler Pycnophylatic or Martin’s Centroids Weighted
The interpolation procedure should be defined according to
the analysis of land use and settlement process in the region
– different characteristics considering capital, frontier,
ranching, etc.
To be continued:

Define and execute an experimental procedure to generate
population density surface for the Amazonia region, following
the approach proposed, with data validation and analysis of
results.
GEOINFO – Dez/2002
Some results
Population Density Surface - Kriging
GEOINFO – Dez/2002
Some results
Population Density Surface - Kriging
GEOINFO – Dez/2002