As I posted on my mobile I realise thr question went in the title instead. I Will try to formulate it in whole below instead. From what it seems I cant edit the title either. Atleast not from a mobile browser.
This question might not have a simple answer and I have read about it in a vast number of articles and forums. All with a different conclusion. Anyway, I have elevation data in an evenly spaced 2x2 m asciitab. I want this data interpolated to a surface. Which of the methods will achieve this rask with best accuracy?
From what I gather kriging works best for scattered data points with an uneven spacing, but can not pass through points, giving it a small error marigin in the actual location of the data points.
Nearest neighbour is supposed to work well with evenly spaced data but can not extrapolate beyond the max and min bounds of the dataset.
In some studies something called anudem has been mentioned as best and also idw has been recommended as it handles steep slopes such as cliff/medge.
Natural neighbour has also been recommended but seems more suited for data with dense points in some areas and more scattered in others.
I do not know what conclusion to draw from all theese contradictions. If somone has a recommended report or knowledge to share in this question I would be thankful!
David, there is no "one" interpolator that is perfect for all data, that is why there are so many of them.
You match the data you have or can get with your goals. If no interpolator seems to fit, collect more data or qualify your interpretation in light of the limitations of your analysis.
Consider "kriging".... which one are you talking about?, there are many, just a quick reference link from the online help
Clear as mud? It should be. If you are worried about that illusive "accuracy" thing you might want to look at interpolators that enable you to get an assessment of the quality of what it did. If you want a nice looking DEM, then look at those that produce a smooth result. If your terrain is highly varied with localized differences, then look for those that put more emphasis on local variations. If you can't collect more data, then work within the limitations of the conclusions that can be drawn from it.
As for references... everyone has a favourite, some are obtuse yet well explained, some are well explained but narrowly focused. Dr Google will provide the main links, but experiment with what you have or generate artificial surfaces of know characteristics for testing ( this missive was written to that end diamond_square... surface generation )
Thank you for a thorugh response. As I have been away on a bussiness trip I'm a little late in my reply.
I hear what you are saying regarding interpolation methods and it is clear that no one interpolator is best in any given situation.
But surely, there must be a general rule of thumb for some situations in a specific setup.
Consider the case of a relatively ordinary terrain with no vast local variations. A grid with processed laserscanned elevation data of high accuracy and with a 2x2 m spacing.
One can expect a high spatial correlation between adjacent points and no extreme changes between two points (unless there is a wall of a cliff, excavation, tunnel etc).
There must be one or a few interpolators to prefer over others in terms of quality. The goal is ofcourse to produce a surface with best resemblece to reality.