The workflow presented in this section is not necessary to repeat for every year. The techniques presented in this section were performed by Hensleigh (2013) to determine appropriate raster resolutions to perform subsequent raster analysis in. However the technique presented in Applications - Determining Sample Spacing which serves as a proxy for sample spacing may be helpful to perform after each survey to ensure that surveys are being formed with consistent sample spacing from year to year.
Support & Raster Resolution
The cell resolution of a raster controls the amount of information included in an analysis and defines computation times of calculations. The surveying method is the main factor that regulates raster quality and in turn cell resolution through its control over point quality, point density, and point distribution (Heritage et al 2009).
Support is the concept that the size, shape, orientation, and distribution of data determine how it can be used (Huevelink 1998). Cell resolutions should be chosen that are fit-for-purpose to how the data will be used. It is commonly (incorrectly) assumed that cell resolution should never exceed the sample spacing. The cell resolution needs to be high enough to topographically define features that are resolved in the raw data. This is obviously related to sample spacing, but it is more closely related to sampling methods. For example, a ground-based feature-stratified sampling scheme that surveys grade breaks can topographically resolve features at a much finer cell resolution, then the sample resolution. By contrast, more uniformly distributed systematic sampling schemes (like those in MBES, TLS and ALS), will only pick up features if the sample resolution happens to land on those features and therefore should generally should not use cell resolutions that greatly exceed the sample resolution.
In general when using high-resolution systematic sampling schemes like MBES, data sampled at smaller intervals over a uniform spacing and distribution will provide a more accurate representation of the surface and provide a greater amount of support. For example data sampled over intervals of 1 ft. can be used to identify objects 1 ft. or greater but not any smaller, implying that the raster resolution of analysis should be no less than 1 ft. By determining the sampling resolution, the distribution and spacing of a survey are identified providing a metric for the amount of support that is provided by the survey. By knowing the amount of support provided by the survey a more informed determination of the appropriate raster resolution for use analysis can be made.
After analyzing point quality, density and distribution, considerations should be made for the level of accuracy required in the surface model and the size of the study area. It is important to recognize the level of accuracy that is required in a surface model and ultimately the use of the surface model. As cell resolution increases the total number of cells increases exponentially, the number of cells in a raster will greatly impact computation times.
Determining Sample Spacing
This video shows how to calculate the average sample spacing using ArcGIS. It is recommended that this is periodically performed for every survey so that an understanding of the average sample spacing is understood per survey and can be used to determine whether data acquisition is consistent over time and/or whether that should result in any changes to the raster resolutions used. Background information for the tool demonstrated in this video can be found by clicking here
Information Loss Sensitivity Analysis - Comparing Different Resolutions
ToPCAT, the Topographic Point Cloud Analysis Toolkit, was used to decimate point clouds to different resolutions for comparison purposes. Each decimated point cloud,10, 7, 5, 4, 3, 2, 1, .5, .25 ft. resolutions, was used to create a subsequent DEM, slope, and curvature raster. By comparing the mean and variance of curvature raster values between resolution sizes an F-statistic can be computed and a probability as to if the difference between the two resolutions is statistically significant.
Results of this analysis are presented in Figure 1, analysis of these plots reveal that moving from 3 foot to 2 foot, 2 foot to 1 foot, and 1 foot to 6 inch resolution are all statistically significant. The most statistically significant increase between cell resolutions is when moving from 2 foot to 1 foot resolution. This analysis also reveals that little information is gained when moving from 6 inch to 3 inch resolution.
Based on this information sensitivity loss analyses the best resolutions are 2 feet, 1 foot and 6 inches. Considering the large area this study covers and the many computations that will be performed in latter steps of analysis 2 foot resolution are recommended. Although some information is lost by choosing a coarser resolution, much is gained in the ability to quickly perform computations and visualize the data.
2013, Hensleigh J. Geomorphic Change Detection Using Mulit-Beam SONAR. Masters Thesis, Utah State University, Logan, UT. 110 p.