Rethinking Tiled Web Map for Raster Analysis

Without a doubt the tiled web map scheme has revolutionized web mapping. It has standardized the generation, rendering and display of map based imagery at preset scales purely for the purposes of visualizing.

As great as it is, I don’t believe as a GIS Analyst, that enough thought was put into the development of the standard.

My Issue

The primary issue I have is, the many decimal place pixel sizes. ie. level 17: 1.19432856695588 m/pixel.

I have a passion for raster data and performing raster analytics including focal and zonal analysis where the extent and the raster cell size when overlaying two rasters must be exact. If the maximum and minimum X and Y extents of the two rasters, and the pixel sizes don’t match exactly it will cause an error in the industry standard tools I am using (ArcGIS, FME, QGIS, GDAL, R).

The Problem

The tile scheme is based on the web mercator projection with a radius of 6378137.0 metres, making for a circumference of 40075016.6855785 metres. This map is projected onto the current 256x256 pixel based image tile resulting in a pixel size of 156543.033928041 metres/pixel at level zero, and 0.59716428347794 metres/pixel at level 18.

My Idea

Round the pixel size! ie. 0.59716428347794 metres/pixel at level 18 to an even 0.5 metres/pixel, 1.0 metres/pixel at level 17, and continue to 131,072 metres/pixel at level zero.

Replace the 256x256 pixel based image tile with a 306x306 pixel based image tile, which is configurable in LeafLet. The 306 pixel image would result in a circumference of the equator of 40,108,032 metres, which is ~33016 metre longer than the current web mercator circumference. The additional circumference would result in an overlap of tiled imagery.

306 pixels * 131,072 m/pixel = 40,108,032 m circumference

By increasing the radius, and circumference, of the ellipsoid to 6383391.55048792 metre (zero flattening) this overlap would be removed.

SPHEROID[“SPHERICAL MERCATOR RASTER”,6383391.55048792,0]

From a cartographic perspective the new ellipsoid also gets you closer to the published mapping scales in web mapping applications at 96 dpi:

• Level 0 (published 1:500M) from 1:554,678,932 to 1:495,389,245
• Level 10 (published 1:500k) from 1:541,678 to 1:483,779

What do you think?

As I mentioned, I love raster and performing raster analytics at scale using tiles as the task separator, and the tiled web map scheme is an awesome format to follow.

I’d love to hear from other GIS Analysts. Do you use the tiled web map scheme for raster analysis, what successes and failures have you had, do you feel my pain, and do you like my idea?

Thanks, Mike

original DEM

aspect kernel analysis (Stage, 2007): slope * elevation² * cos(direction)

NB. This is my first and hopefully not my last blog post.