Hi! I am a complete beginner in GIS and taking a class right now, and we are doing some assignments in ArcGIS - ArcMap.
In one of them we were given a shapefile with heightpoints. These had been manually collected using gps, and they had been chosen to best represent the variations of the landscape.
Next we were to build a TIN-model and a Grid (with 3D Analyst toolbox -Interpolation / Natural Neighbor, Z vaule field = SPOT, Output Cell Size = 10 meter) from these points.
And finally to interpolate contour lines with Surface Contour - Contour interval = 10 meter.
Now here comes the question: Which of these methods give the most accurate contour lines, and why?
I have a hard time getting my head aroud this. I have written in my report that the Grid-based contours are more accurate because the Grid has alot more interpolated points to use when drawing the contours. Or is this just making the lines more smooth, not more accurate.
In this example file we are using, the grid cell size set to 10 meters, gives a smaller grid than the TIN-triangles, even when the original data points are close together.
But the grid based contours are interpolated twice? First from the data points to all the centers of all the cells, and then to the contour lines.
And the TIN-based contours are interpolated straight from the data points to new points on the lines that form the edges of the triangles. Are they maybe more accurate but less smooth??
Any help to clear my confusion would be appreciated.
Whether the TIN interpolation (straight lines with a constant slope between points) or some raster interpolation method is more accurate is unknown. The density of your points, the weights you give them, and the raster interpolation method will all influence the resulting surface model in different ways. You could assess the quality of your surfaces using root mean square error (http://en.wikipedia.org/wiki/Root-mean-square_deviation). In order to do that you would need a statistically sound number of surveyed point locations (not your GPS points) with z values to calculate RMSE.
