|
I have about 500,000 points collected in a linear grid pattern over a 1 km2 area. The points were collected as part of a metal detection survey. Each point has X,Y and Coil Response mV (min -675.38 max 6104.25) values. What type of interpolation should be used to visualize the results? |
|||||
|
|
Your figures show that the point-to-point spacing is about 1 meter. That likely could be close to or less than the resolution (especially if you're penetrating significantly below depth). Thus, almost any form of interpolation will work fine and your task is to minimize the effort. If the data are truly regularly spaced, then a good fast way is to format the data as an ASCII grid export file and open it in your GIS: the software will automatically interpolate in order to display or resample the data. (In ArcGIS you can choose among three methods--nearest neighbor, bilinear, and cubic convolution--with the latter two giving smooth interpolation.) A slower but almost as simple way is to open the data as a point layer, align a grid specification with the layer's extent and spacing so that each point falls into a single cell, and convert the data to raster format. (You might have to rotate the coordinate system to achieve a good alignment, though.) The slowest (and probably most unsatisfactory) methods will be any of the GIS's interpolate-to-raster methods, such as inverse distance, natural neighbor, Voronoi polygon, or--God forbid--any form of Kriging (listed approximately in increasing order of computation time and your effort). If your spacing is a bit irregular, choose a fast method. If you can control the inverse distance parameter, choose as small a value as you are able in order to avoid peak-like artifacts. |
||||
|
|
|
This may not be the right answer, but just my own experience. I had to produce a heat-map style surface of radiometric readings for a site. These survey points were not consistent, but fairly good coverage of the whole site. I tried several interpolation methods, but Inverse Distance Weighting worked best for me. I worked out the max distance between 2 survey points, and then used this as the search radius. This produced a good looking result that I was satisfied was giving a realistic visualisation of the spread of radiometric readings for the site. |
||||
|
|
|
I assume you're trying to produce some sort of heat map, and the question really is what scale parameters to use. The answer is likely to be strongly dependent on what exactly you're looking for; depending on the application, there are likely to be thresholds of interest, and you'd want to highlight those. Without more info on what you're trying to use the data for, we're kinda guessing - applications range from archaeological site exploration to military detection of ... stuff. |
|||||||
|
|
What is the variability of your data? For visualization you might have to aggregate rather than interpolate your date if its variability is high. |
||||
|
|
|
This is -exactly- the problem that kriging was designed for. You have easily enough data to estimate the variogram, and you likely have a polynomial trend surface; so, universal kriging would be appropriate to the problem. The issue is the processing time for universal kriging. One very simple way to deal with this problem rather than any statistical interpolation method is to just use the deterministic interpolation built into the terrain dataset in arcgis. Create an x,y,z point dataset, where the z value is your Coil Response mV. Load this in as the mass points in your terrain dataset and simply load the set into a map document ArcMap will take care of the on the fly IDW to visualize the dataset at different map scales. Also, you might want to check out this video series: The 2nd video in the series discusses how to choose the correct neighborhood radius for IDW using Moran's I. |
||||
|
|
