I'm usually simply amused when other bloggers have trouble 'admitting' that Microsoft has done something really well. If you read slashdot, you know the type of posts i'm referring to. they generally take this form - "micro$oft sux! gates' empire is pure evil... but I have to admit, Halo really rawks." you can feel their pain, the inner turmoil they must grapple with as they enjoy the fruits of the borg.

After reading this post from Mickey at the Digital Earth Blog reacting to my post on the beautiful VE imagery of Niagara Falls, I nearly passed it by. But then I realized it was a good opportunity to present some facts on what goes into modeling the Virtual Earth. Mickey said:

...I have to admit, it’s quite stunning... As you’ll see, there is no comparison. I have to imagine that Microsoft has done a bit of hand-tweaking to make the falls look so good, while Google obviously has not. If they wanted, Microsoft could tweak little areas all over the world to create fake examples of how much better their product is. That being said, this isn’t even close. Virtual Earths imagery looks better, the altitude mesh is much more precise, and they have 3D buildings all around the area

Mickey makes it sound as if we cork the bat or have some other unfair advantage over our competitors in the field, and that perhaps we are falsifying data to make it look better. Niagara Falls is beautiful in the real world and shouldn't need a Botox injection to look good in the Virtual. But there are shortcomings in the sensing process used to model the Earth that need to be addressed.

A bit of "hand tweaking"? If you think applications like Virtual Earth and Google Earth are simply fed raw data from remote sensing systems and manage to present the experiences they do you've been watching Minority Report too many times :-) Tweaking doesn't even come close to doing justice to the enormous production process to get data from sensor to web. It's understandable that most of us are completely unaware of this process; there would be a lot more vegetarians if we all had to kill what we eat. In the same way we are insulated from the gory details of the journey from farm to grocer, we don't need to understand how to process imagery and data models of the Earth's surface if all we want to do is cruise around in Virtual Earth and enjoy the experience. To appreciate this beautiful image of Niagara Falls we don't need to be aware of the image processing that went into creating it:

1. UltraCam is flown and captures the imagery with a prescribed set of overlap
2. Pre ortho processing - ingest formatting, pan-sharpening (mutispectral brought to panchromatic resolution), radiometric normalization (images brought to similar spectral range)
3. Orthorectification (surface and camera distortions removed from imagery and images are geographically referenced to a coordinate system)
4. Orthomosaic (individual ortho images seamed and color balanced together)
5. Final radiometric look up table (final spectral adjustment applied to entire mosaic)

The Virtual Earth team employs lots of people whose job is to make the sensor data collected represent the Earth as best as possible where automation breaks down. Rob Waterman is one of these guys. I asked Rob what goes into creating elevation models of the Earth's surface:

My goal is to make the data represent real life as best I can. Specifically for Niagara I used three different input DEM sources and some well placed breaklines etc. to generate a single improved DEM for the area.  There is nothing “fake” about it.  I combined the best of all the data sources available to me to come up with the most realistic representation of the area (including ortho imagery for breakline placement and many oblique photos to help determine the most accurate/representative layout of the area).  I think we are here to generate the most accurate/realistic Virtual Earth we can and that means using the all the data available in an attempt to arrive at the “truth”.

The VE team has done an amazing job of automating much of this process over the past year. But until problems like extracting DEM accurately from water are perfected, wizards like Rob are going to continue to 'tweak' our sensor data to create a true representation of the Earth. There's nothing fake about what we do - these manipulations are intended to present the most accurate Earth model possible.