Evaluating the Blue Marble next generation dataset
After making the well known global color image of the earth in 1km resolution in 2002 which was widely used for earth visualization purposes and which i used for my earth renders as well Reto Stöckli and his colleagues made available a new improved version with 500m nominal resolution and i had a quick look at the new data to see how much it is improved.
The most striking improvement about the new data is that separate images are available for every month. This way the seasonal variations, esp. snow cover, can be found accurately represented in the images. Significant effort has been put into generating those images, the usual monthly composite images available from the MODIS satellites contain cloud cover and other faults. Details on the techniques used for generating the images can be found on the Author's site.
Low resolution JPEG compressed versions of all the monthly images can be found on the NASA site:
The images are available in three variants: The plain reflectance image, with relief shading and with relief as well as underwater shading. The full size data is available upon request or on other sites, for example:
If you consider retrieving the full size version bear in mind that it is quite large. The compressed version is more than 2GB which uncompresses to more than 10GB. To use the full image you need programs capable of dealing with large files and able to deal with images without loading them into memory as a whole. The GDAL library and the tools it comes with are very useful for this.
My evaluation is based only on the 500m July version without relief shading and the 2km versions for the other months. Below you can see a part of the 2002 version and the new version in reduced size for comparison.
|Blue Marble (2002)|
|Blue Marble next generation (2004-07)|
The most obvious difference are the less saturated colors in the new image. This is a huge advantage. The 2002 version had the color values scaled in a way resulting in clipping - this is especially obvious in regions with snow cover but also occurs elsewhere resulting in unrealistic colors for example in desert regions.
The images contain color information for the land surface as well as coastal sea. The deep ocean areas are in uniform color. This is different from the 2002 version where complete ocean coloring was available at least in the low resolution version (see image on the right). The ocean color is not blended to the coasts so there is a visible line in most parts.
There are no sea ice and clouds layers available either but it is possible to use those from 2002 in combination with the new images of course.
The 2002 image contained significant cloud cover in some equatorial regions. This is much improved in the new version but still not perfect. The images below show an area in Ecuador (south america) - most of the white in the 2004 image isn't snow cover but clouds - probably falsely classified as snow.
The new images have a double nominal resolution - 500m at the equator or 0.00416667 degrees per pixel. The gain in detail at least proportional to this, maybe even better. Below you can see a comparison - the 2004 sample is in original size, the old one scaled up to this size.
There is however still room for improvement. The MODIS satellites generate data with quite strongly varying resolution but in most parts it is quite a bit better. Below on the left you can find the same area in a raw MODIS image. The original image is larger and with non-uniform resolution, see the link, but scaled down the detail level is still better. The blur in the Blue Marble image probably results from combining a lot of different images for cloud freeness.
If even more detailed image data is needed other data sources could be used. Such images are rarely available with enough coverage to generate full global images for every month. On the right you can see the same area on a Landsat image.
Using the images
Renders using the new data sets can be found on a separate page.