Transcript GIS Programs

th
19 Advanced
Summer School in
Regional Science
Overview of advanced techniques
in ArcGIS data manipulation
Merging raster data with vector

Zonal statistics
– Consider reading elevation into Dutch
Municipalities
– Now we can identify the Dutch cities most at
risk from rising sea levels due to global
warming
– Join zonal statistics, select by attributes
Cutting the raster data down to size
Map of Dutch municipalities would be more
attractive if elevation raster were smaller
 Use Toolbox – Clip to trim raster

– Loads more quickly as well
Raster Data

Creating rasters through interpolation
– Interpolating from Points
• Inverse distance weighted
• Spline
• Kriging
– Interpolation from polygons is also possible – see this
later in the program

Consider an example using the Netherlands
zipcode data
–
–
–
–
Join poly data to point data by attributes
Interpolate manufacturing share
Join point data to poly spatially
Compare interpolations
Raster Interpolation

Given data at selected points
– Most natural if these are samples from some
process that is continuously distributed
• Economic activity
• Pollution levels
– Construct a raster surface to approximate
using these data
• Value at each location should depend on the values
of nearby points
• Closer points should matter more
– Simplest – average weighted by inverse
distance
Raster Interpolation
Spatial Analyst can be used to construct an
IDW raster approximation
 Several paramters to set

– Exponent to specify distance decay
– Search radius (fixed distance, variable points)
– Search radius (variable distance, fixed points)
Raster Interpolation: Kriging

Kriging provides a more sophisticated model of
spatial dependence for interpolation
 All interpolation approaches use some form of the
relation:
N
Zˆ  s0    i  Z  si 
i 1
– location where an approximate value is to be
calculated
– locations with known values
– Weights
• IDW weights depend only on a power of distance
• Kriging weights depend on the structure of spatial covariance
Raster Interpolation: Kriging

Kriging takes points with known values and
estimates the “semi-variogram” as a function of
distance
– This is a scaled spatial covariance:


2
1 
   E Z  si   Z  s j  

2 
– Kriging makes some assumptions about how this
covariance depends on distance
Raster interpolations

How do these interpolation techniques
compare?
– IDW and Kriging capture some of the structure
– The surface can be averaged over a region to
provide an alternative measure
– Zonal statistics again!
Rasters to measure distance

Raster data can be employed to measure
distance and cost of travel
– We started this process yesterday
– Continue the analysis of distance

Spatial Analyst has several distance tools
– Straight line
– Cost weighted
– Min distance
#
#
#
#
#
#
#
#
#
#
#
#
#
#
#
#
#
#
#
#
#
#
#
#
#
#
#
#
#
#
#
#
#
#
#
#
#
#
#
#
#
#
#
#
#
#
#
#
#
Rasters to measure distance
First step is to generate raster to represent
the cost of traversing a pixel
 Several possibilities

– Use elevation – implies that traveler tries to
remain at lowest elevation (like water!)
– Use slope – implies that traveler tries to
minimize the amount of climbing and
descending
– Use a transport network – cheapter to travel
along major roads
– Use a combination of these
• Raster calculator can be used to combine different
sources of cost
Rasters to measure distance

Analysis of minimum distance path
– Identifies roadway sections that might carry
less traffic
– Generate a contour map of costs
Analysis of remotely sensed data

Modeling changes in urban land use
 Theory suggests these changes should depend
on several key variables
–
–
–
–
Population
Income
Transportation costs
Opportunity cost of urban land use (agricultural
productivity)
– Policy variables

Income measurement is a big problem for some
countries
 Strategy: use remotely sensed data to estimate
Night Lights Data

DMSP/OLS
– Began in 1978
– Approx 2.5
KM resolution
– Problems
• Diffusion or
“bloom”
• Lighting
technology
• Sensitivity to
density
• Instrument
saturation
Night light data

Night light levels might reasonably be
related to several variables
– Income
– Per capita income
– Latitude
– Density of population
– Global connectedness
– Capital intensity of production

Explore to see what we can learn
Night light data
First – download data from NOAA
 Second – clip to European region
 Third – form composite image

– Load different years into different colors
Night Light Data

Next – analyze
–
–
–
–
–
–
Dissolve NUTS3 data set as desired
Calculate GDP and Population for areas
Calculate log GDP
Use zonal statistics to calculate total light levels
Calculate log light level
Use graph to plot scatterplot
GDP as a function of Light
15
10
5
0
0
5
10
15
20
Now you can try this for smaller areas
Try this for NUTS 1 (or NUTS 2) areas
 Try this for a particular country
 Suggest some hypotheses about this
relationship
