Importing large raster datasets

Importing raster datasets in the range of 10 GB or less can usually be done with the default Map Import settings. The general documentation on importing maps can be found here.

However, if the dataset you are importing is significantly larger - 20 GB or more - the settings become more important to get a good result without using too much processing time and file size. This article will go more in-depth on the different parameters and how they affect the import process and result.

Image Quality

Image quality in this context refers to how similar the output result is to the input data.

Relevant Map Import parameters:

Resampling

This is the method to use when creating downsampled versions of the main dataset (also called pyramids or overviews). The default resampling method is average. The following table shows a benchmark comparing the different methods when used on a typical 700 MB dataset. The column "Difference vs. baseline" shows percentage difference in execution time from average

Resampling methods benchmark

Method	Average time (seconds)	Difference vs. baseline	Sample tile
average	42.45	0%
nearest	38.50	-9%
bilinear	57.66	36%
cubic	61.80	46%
cubicspline	60.64	43%
lanczos	66.82	57%
gauss	54.28	28%

Full article here: Comparing resampling methods in massivegeopackage

Tile format

The best choice of tile format depends on the source data.

• Auto uses JPEG on opaque tiles and PNG on tiles where transparency is needed. Reccomended for:
  • Imagery from satellites, aeroplanes or drones
  • Topographical maps with elevation shading or other smooth gradient effects.
• PNG uses 32 bit RGBA PNG format on all tiles. This is a lossless format, which means that the output will be identical to the input (as long as no resampling is neccessary). This comes at the cost of a very large file size. Reccomended for:
  • Datasets where lossless quality is needed, with no regard for file size.
• PNG8 calculates a color table with maximum 256 different colors for each tile. If the input data has discrete gradients and a small number of unique colors, this is the best option. Reccomended for:
  • Topograhical maps with no shading,
  • Thematic datasets such as land classification.

Quality

The quality parameter controls how aggresively JPEG tiles are compressed. It is expressed as a value between 10 and 100 where

• Higher values = less compression, larger files, better image quality.
• Lower values = more compression, smaller files, more visible artifacts.

The relationship between file size and visual quality is highly nonlinear. Using Quality=100 is not a lossless conversion, but it will result in a very large dataset. Decreasing the number will retain most of the visible quality while giving substantial size reductions. The default value of 75 gives a good balance between visual quality and file size.

Visual quality can be objectively measured as Structural Similarity (SSIM), which is a comparison of the compressed image to the original. We have done a benchmark of several compression algorithms, comparing SSIM to file size at various quality measures. Notice how the curve starts to flatten out at around 60-75 quality. Increasing quality from 85 to 95 will almost double file size with minimal increase in visual quality. The full article can be found at Comparing compression in different formats

CPU use

If the input dataset consists of multiple files, Maria Map Maker will process each file in parallel, before subsequently merging them together into a single file. This allows the program to more efficiently utilize a multi-core CPU. The parameter CPU use controls the number of logical processors to use. The following options are available:

• Light (25% of logical processors)
• Moderate (50% of logical processors)
• Aggressive (75% of logical processors)
• Extreme (100% of logical processors)

The default is Moderate, which is usually a good choice. More parallel processes do not automatically result in higher performance. Beyond a certain point, processes compete for shared resources including disk read/write throughput, memory bandwidth and CPU caches.