Disc Golf Course Design

by Steve West
Steve West Disc Golf
About Me
Contact Me
Course Maps
Site Map
What is Disc Golf?
Disc Golf Badges
Top 32 Places
Geography of Disc Golf
Maps of Courses in US
Clusters of Courses
Most Courses in X Miles
Map Data Points
Play All MN Courses
500 Holes in a Day
Tools for Course Designer
Testimonials
Gallery

When looking at the map of disc golf courses, one notices that they come in clumps.  This leads to questions such as: Where are the most courses? and How should Regions be defined? 


I've been looking for a way to mechanically identify these clumps.  The traditional method of measuring X number of courses within Y miles only identifies the nearly circular clumps. But most clumps are not circular.  There is a concentration all up and down the left coast.  Florida  The west shore of Lake Michigan is concave. has a ring around it.


I found what is called density-based clustering.  It doesn't rely on any particular shape to define a clump.  Rather, it looks for "densely coursed" regions.  If a course has a certain number of baskets within a specified radius, that course can "seed" a cluster.  That course and any courses within the radius are in the cluster.  If any of the courses in the radius also have a certain number of courses within the specified radius, then the courses around that course are also in the cluster.  It's a chain reaction.


So, the cluster grows in any direction, as long as it can find nearby courses that are in densely coursed areas.


Unfortunately, one must set the threshold for density by picking a number of holes and a radius.  For number of holes, I chose 72.  This seemed like enough to make an area a "good" place for disc golf.  Also, some courses list 72 holes, so it seemed like this should be a minimum.


I tried several different radii.  An extremely small radius produces few clusters because there aren't that many really densely packed areas.  At a five mile radius, I found the following clusters.


15 courses, 213 holes: Lockport, IL 60441

10 courses, 150 holes: Bowling Green, Kentucky 42103

8 courses, 117 holes: Moline, Illinois 61265

6 courses, 81 holes: Red Fork, OK 74107

5 courses, 81 holes: High Bridge, WI 54846

5 courses, 84 holes: Spokane, Washington 99224

4 courses, 72 holes: Tulsa, Oklahoma 74115

4 courses, 72 holes: Crown Point, Indiana 46307

4 courses, 72 holes: Emporia, KS 66801

4 courses, 72 holes: Huntsville, Alabama 35805

4 courses, 72 holes: Rocky Mount, NC 27804

3 courses, 74 holes: Live Oak, TX 78233

3 courses, 90 holes: Worcester, MA 01602

2 courses, 108 holes: Ludington, MI 49431

2 courses, 72 holes: Hanover, Pennsylvania 17331


For a longer radius, the number of clusters is small, because the size of the clusters keeps growing and growing.  For example, here are the clusters based on 72 holes within 30 miles.




The "natural" radius seems to be about 17 miles.  At this radius, the number of clusters is maximized.  I ran a 15 mile radius and found 98 clusters.  These clusters have an average of 11 courses and 168 holes.  The top ten are:


69 courses 875 holes: Lombard, IL 60148

60 courses 771 holes: Maple Grove, MN 55311

37 courses 589 holes: Monroe, OH 45050

38 courses 567 holes: Irving, TX 75038

30 courses 482 holes: Novi, MI 48375

25 courses 382 holes: Austin, TX 78746

26 courses 356 holes: Crocker, IA 50226

27 courses 343 holes: Portland, OR 97202

23 courses 313 holes: Milford, PA 18951

17 courses 310 holes: Monmouth, ME 04259


Here is a map.  There is also a cluster in Anchorage, AK.




I'm not completely satisfied with this method.  I don't like the requirement that a radius and a threshold should be specified in advance.  Also, it seems like clusters should be allowed to be more spread-out in areas with few courses.  For example, all the courses in Colorado look like a cluster to me, but so do the courses around Dallas - even though the scale is quite different.  I'd like a method that recognizes a drop-off in density and defines the edge of a cluster that way.  I've defined a method that does this, and doesn't result in circular clusters.  But, it would take eons to run.  So, unless someone comes up with a real good reason to define clusters, I think I'll not program that.