Algorithms for GIS

Space Filling Curves

Laura Toma

Bowdoin College
 Space filling curves

A map from an interval to a square

 Space filling curves

https://mathsbyagirl.wordpress.com/tag/curve/

A map from an interval to a square

Z-order space filling curves

• Assume 2D points with integer coordinates on k bits

p = (x1x2x3…xk, y1y2y3…yk)

• Define the Z-index of a point

Z_index : {0,..,2k-1} x {0,..,2k-1}—> {0,..,22k-1}
Z_index(p) = x1y1x2y2…xkyk
 k=1 bit

p Z_index(p)
(0,0) 0

(0,1) (1,1) (0,1) 1

(1,0) 2
(1,1) 3
x
(0,0) (1,0)
 k=1 bit

p Z_index(p)
(0,0) 0
(0,1) 1
1 3
(1,0) 2
(1,1) 3
x
0 2
 k=1 bit

p Z_index(p)
(0,0) 0
(0,1) 1
1 3
(1,0) 2
(1,1) 3
x
0 2
 p Z_index(p)
(00,00) 0000=0
k=2 bits (00,01) 0001=1
(00,10) 0100=4
Find the Z-order!
(00,11) 0101=5
(01,00)
(01,01)
y (01,10)
(01,11)
(10,00)
(10,01)
(10,10)
(10,11)
(11,00)
(11,01)
x
(11,10)
(11,11)
 k=2 bits

x
 k=3 bits

Find the Z-order!

x
Computing the Z-index
• Consider an x-coordinate x1x2x3 in the square [0,…8)
• x1=0 means the point will reside in the first half
• x1=1 means the point will reside in the second half

x
x=0** x=1**
Computing the Z-index
• Consider an y-coordinate y1y2y3 in the square [0,…8)
• y1=0 means the point will reside in the first half
• y1=1 means the point will reside in the second half

y=1**

y=0**

x
Computing the Z-index
• Consider an y-coordinate y1y2y3 in the square [0,…8)
• y1=0 means the point will reside in the first half
• y1=1 means the point will reside in the second half

x=0** x=1**
y=1** y=1**

x=0** x=1**
y=0** y=0**

x
Computing the Z-index
• Consider an y-coordinate y1y2y3 in the square [0,…8)
• y1=0 means the point will reside in the first half
• y1=1 means the point will reside in the second half

y y

x=0** x=1** Z=01** Z=11**

y=1** y=1**

x=0** x=1**
y=0** y=0**
Z=00** Z=10**
x x
Computing the Z-index
• Consider an y-coordinate y1y2y3 in the square [0,…8)
• y1=0 means the point will reside in the first half
• y1=1 means the point will reside in the second half

y y

Z=01** Z=11**

Z=00** Z=10**
x x
Computing the Z-index
• Consider an y-coordinate y1y2y3 in the square [0,…8)
• y1=0 means the point will reside in the first half
• y1=1 means the point will reside in the second half

y y

Z=01** Z=11**

Z=00** Z=10**
x x
Computing the Z-index
• Consider an y-coordinate y1y2y3 in the square [0,…8)
• y1=0 means the point will reside in the first half
• y1=1 means the point will reside in the second half

y y

Z=01** Z=11**

Z=00** Z=10**
x x
Computing the Z-index
• Consider an y-coordinate y1y2y3 in the square [0,…8)
• y1=0 means the point will reside in the first half
• y1=1 means the point will reside in the second half

y y

Z=01** Z=11**

Z=00** Z=10**
x x
Computing the Z-index
• Consider an y-coordinate y1y2y3 in the square [0,…8)
• y1=0 means the point will reside in the first half
• y1=1 means the point will reside in the second half

y y

Z=01** Z=11**

Z=00** Z=10**
x x
Z-order space filling curves

• Other Z-orders can be obtained similarly

y
y

• Can be extended to work with decimal numbers in [0,1)

• make values positive (add smallest value)
• divide all values by max value
• ==> now we got values in [0,1)p=(.1100, .0101)
 Z-order and quadtrees
visit quadrants recursively in this order: NW, NE, SW, SE
 Z-order and quadtrees
visit quadrants recursively in this order: NW, NE, SW, SE
 Z-order and quadtrees
visit quadrants recursively in this order: NW, NE, SW, SE
 Z-order and quadtrees
visit quadrants recursively in this order: NW, NE, SW, SE
 Z-order and quadtrees
visit quadrants recursively in this order: NW, NE, SW, SE

• At the limit, it will reach all points in the square ==> space filling curve
 Z-order and quadtrees
visit quadrants recursively in this order: NW, NE, SW, SE

Where is the very first

point visited?

• At the limit, it will reach all points in the square ==> space filling curve
 Z-order and quadtrees
visit quadrants recursively in this order: NW, NE, SW, SE

first point in this order

last point in this order

• At the limit, it will reach all points in the square ==> space filling curve
 Z-order and quadtrees
visit quadrants recursively in this order: NW, NE, SW, SE

first point in this order

last point in this order

• At the limit, it will reach all points in the square ==> space filling curve
 Z-order and quadtrees
visit quadrants recursively in this order: NW, NE, SW, SE

first point in this order

last point in this order

• Every point in the square will be visited by this curve

• 2D ==> 1D
 Z-order and quadtrees
visit quadrants recursively in this order: NW, NE, SW, SE

first point in this order

last point in this order

• We visit quadrant 1 before we visit quadrant 2:

==> All points in quadrant 1 comes before all points in quadrant 2
 Z-order and quadtrees
visit quadrants recursively in this order: NW, NE, SW, SE

first point in this order

last point in this order

• We visit quadrant 1 before we visit quadrant 2:

==> All points in quadrant 1 comes before all points in quadrant 2
 Z-order and quadtrees
visit quadrants recursively in this order: NW, NE, SW, SE

first point in this order

last point in this order

 Z-order and quadtrees
visit quadrants recursively in this order: NW, NE, SW, SE

first point in this order

last point in this order

 Z-order and quadtrees
visit quadrants recursively in this order: NW, NE, SW, SE

first point in this order

last point in this order

 Z-order and quadtrees
visit quadrants recursively in this order: NW, NE, SW, SE

first point in this order

last point in this order

• and so on…..
 Z-order and quadtrees
visit quadrants recursively in this order: NW, NE, SW, SE

first point in this order

last point in this order

• and so on…..
 Z-order and quadtrees
visit quadrants recursively in this order: NW, NE, SW, SE

first point in this order

last point in this order

• Every canonical square corresponds to an interval of the z-order curve

 Z-order and quadtrees
visit quadrants recursively in this order: NW, NE, SW, SE

first point in this order

last point in this order

• Two canonical squares are non-intersecting, or one included in the other

 Z-order and quadtrees
visit quadrants recursively in this order: NW, NE, SW, SE

Zindex(a) Zindex(b)

• If Zindex(a) < Zindex(b), we say that a < b

• Any two points a,b can be compared
Computing the Z-index

Z_index : R2 —> R
For simplicity assume points with integer coordinates on k bits
• What is the largest integer representable on k bits?

Z-index(p)
Space-filling curves
• Z-order curves are a special type of space-filling curves
• First SFC were described by Peano and Hilbert

Peano curve
Hilbert curve
Spatial locality

• Big-Oh analysis does not have the final word

• Two algorithms that have the same big-Oh can differ a lot in performance
depending on their cache efficiency
• To analyze and fine tune the algorithm we need to look at the performance
across all levels of the memory hierarchy
http://classconnection.s3.amazonaws.com/149/flashcards/3088149/png/memory_hierarchy1367201501848.png
http://users.informatik.uni-halle.de/~jopsi/dinf104/memory_hierarchy.gif
• At all levels, data is organized and moved in blocks/pages
• Each level acts as a “cache” for the next level: stores most recently used
blocks
• Applications that access data that’s stored in a “recent” block will find it in
cache
• 1ns vs 100ns <—- SIGNIFICANT!
Spatial locality

Spatial applications usually have spatial locality in their access to data, i.e. they
are likely to access together points that are close to each other in space

We would like points “close” in 2D to be stored “close” to each other in the data structure
Spatial locality

Spatial applications usually have spatial locality in their access to data, i.e. they
are likely to access together points that are close to each other in space

We would like points “close” in 2D to be stored “close” to each other in the data structure
Spatial locality

Spatial applications usually have spatial locality in their access to data, i.e. they
are likely to access together points that are close to each other in space

We would like points “close” in 2D to be stored “close” to each other in the data structure
Spatial locality

Spatial applications usually have spatial locality in their access to data, i.e. they
are likely to access together points that are close to each other in space

We would like points “close” in 2D to be stored “close” to each other in the data structure
Spatial locality

Spatial applications usually have spatial locality in their access to data, i.e. they
are likely to access together points that are close to each other in space

We would like points “close” in 2D to be stored “close” to each other in the data structure
Grid default layout: row-major order

…
Grid default layout: row-major order

• Does this layout have good spatial locality?

• How far are these points in the array?

Grid default layout: row-major order

• Does this layout have good spatial locality?

• How far are these points in the array?

Spatial locality

grid in default row-major order grid stored in Z-order

… …
Spatial locality

grid in default row-major order grid stored in Z-order

… …
Spatial locality

grid in default row-major order grid stored in Z-order

… …
Spatial locality

grid in default row-major order grid stored in Z-order

… …
Spatial locality

grid in default row-major order grid stored in Z-order

… …
Spatial locality

grid in default row-major order grid stored in Z-order

… …
Spatial locality

grid in default row-major order grid stored in Z-order

… …
Spatial locality

grid in default row-major order grid stored in Z-order

… …
Spatial locality

Arranging data in order of a space-filling curve improves spatial locality

• points that are close together in space, will be stored close to each other

grid in default row-major order grid stored in Z-order

… …
 Spatial locality
• Arranging data in order of a space-filling curve improves spatial locality
• points that are close together in space, will be stored close to each
other
• data will be in the same blocks as previous data
• data will be found in cache
• improvements at all levels of the memory hierarchy

• Hilbert curve has better locality than z-order, but slower to compute
• Z-order used with Strassen’s algorithm —> speedups (2002)
SFC in art

Don Relyea, artist futurist and tehnologist

• http://www.donrelyea.com/site2015/space-filling-curve-art-2004-2014-wide-format/