DTMs from multiple datasets

Production of an integrated digital terrain model from multiple datasets of different quality [1]

Overview

• Describes production of a DTM from multiple input sets
• Claim that result is better than any of the parts
• Assumes that multiple DTMs are available for study area
• Implementation not ready for commercial use

DEMs

• Digital Elevation Models
• Rasterized, contain elevation data for each cell
• Elevation can be discrete (assumed to be uniform throughout cell, changes on cell boundary) or continuous

DTMs

• Digital Terrain Models
• Assumed to be continuous
• Contains elevation data and other features that define surface [1]
  • Slope
  • Curvature
  • Gradient

Techniques used

Simultaneous interpolation

• As name implies, interpolate all sources at once
• Least squares interpolation
  • Predict new location from surrounding values
• Weight different data sources depending on quality
  • Use characteristic features from different sources
    • Each feature can come from only one source
  • Often poorly distributed
  • Use "automatic eliminiation and densification" [1]
    • Remove points that are too far from the surface being predicted
• Need to find good weight distribution
  • Based on strength of correlation among points

Geomorphologic enhancement

• Not well-defined; assume audience will know particulars
• DTM statistically usually too smooth -- use these enhancements to closer approximate reality
• Use averages, take quality into consideration
  • Find proper weights to reflect quality
  • Apply best geomorphological characteristics
• Go from lowest quality to highest quality for best results
• Error goes down as more sources considered
• Apply geomorphological enhancements at end
  • Don't really explain why
• Unique grid size for all sources imposed on final result

Steps involved

Pre-processing

• Visual evaluation, eliminate gross errors
  • Cut hole in layer and look at other layers
• Statistical elimination of bad points
  • need high-quality DTM to do this
• Use updated reference points
  • Tell how accurate data is
  • Can give inaccurate picture -- most reference points are in areas with high accuracy already

Processing

• Mosaicing
• Reference points
  • Also used to improve final result
  • Elevation only; no geomorphological information
  • Only considered if within threshold
• Temporal differences
  • Correct absolutely: remove outdated info and replace with something else
    • Used for man-made changes (roads, quarries)
  • Correct relatively: apply difference surface to bring up to date
    • Used for natural changes (volcanoes, landslides)

Evaluation

• Give applications
  • Produce enhanced DTMs (better than any of the inputs)
  • Update DTMs based on temporal changes
• Good test case (Slovenia has interesting terrain and a wide variety of available data sources)
• Drawbacks possible when integrating data from a variety of sources:
  • Lowest common denominator
  • Inconsistencies in accuracy across DTM
• Possible solutions:
  • Multi-resolution approach (referenced but not described)
  • Produce multiple DTMs depending on area
• Describes methods of removing gross errors, but requires reference with "high and known quality" -- thus several DTMs can make a DTM better than any of the components, but at least one must be realtively good

References

• [1] Podobnikar, T., Production of integrated digital terrain model from multiple datasets of different quality, International Journal of Geographical Information Science, 19:1, January 2005, pp. 69-89.

Slovenia Links

• http://europa.eu.int/abc/maps/index_en.htm -- map of the EU with Slovenia
• http://www.gumbopages.com/festivaltours/ljubljana.html -- a picture of Ljubljana, the capital of Slovenia
• http://www.bled.si/slo/narava/otok/jezero.jpg -- sample picture of lake and mountains

-- HenryMcEuen - 03 May 2005