Interpolation

[jayeetachakraborty1]

Here is the map with sampling locations. The locations are scattered yet clustered.  The concentration of contaminants have been measured against each locations and the variation in concentration is quiet high. Considering the distribution pattern of sampling wells, would it be wise to pull a spatial distribution map ?  Would it be better to use natural neighbor ? Me and my professor had a different opinion. Since the distribution of the points are not uniform, I used Krging to pull a  map. But my professor advised me to use NN method since the data points were clustered. I had a difficulty to extrapolate the map beyond the convex hull using NN method. Kriging worked better in terms of both extrapolation and interpolate. Is there any alternative? Please advice. I tried Geostatistical wizard to look at RMS value for different interpolation methods and IDW had the lowest. But the distribution map produced bulls eye effect and my professor asked me to avoid IDW for any geologic map.

Attached is the map.

[DanPatterson_Retired]

That has to be one of those classic point patterns that really shouldn't be interpolated beyond the bounds of their individual clusters.

The blue area is apparently some formation? of some kind? and you had to use the only data available to you?  Would that be a fair guess? 

What is wrong with dealing with analysing the three clusters separately, perhaps using a small buffer around each cluster's convex hull.  What do each of the clusters show?  Are they hugely similar? different?  Are there any distinct directional trends?

What is the proposition that the intervening space between the clusters shares properties or differences with each other?

What is the need to interpolate between the clusters?

More importantly, what is the need to extrapolate beyond the boundaries.

Just some food for thought and discussion, before you get into a 'which is the best interpolator' discussion

[jayeetachakraborty1]

The blue area is the spatial extent of the Carrizo-Wilcox aquifer. The purpose of interpolation (and extrapolation) is to look at the variation of concentration of the contaminants along the aquifer trend. The three clusters represent locations in the north, central and south of the aquifer trend. It was not feasible for me and my team to sample groundwater wells throughout the aquifer trend. So we sampled just 48 wells representing the three regions of the aquifer. Would it be a good idea to pull a spatial distribution map along the trend ? Else, should I pull three maps around the three clusters. Having said that, are there any minimum sample points for kriging or NN interpolation methods?

[SteveLynch]

1) Have you tried to decluster the data when using kriging in Geostatistical Analyst?

2) how different are the values within each cluster?

3) please also show a map of the interpolated surface and a screenshot of the semivariogram

[jayeetachakraborty1]

Attached are the maps and semivariograms. I did extrapolate over a larger area and the values are somewhat unreliable.

[SteveLynch]

• Have you looked at the Standard error of prediction map?
• Did you look at the cross validation page?
• are you transforming the data?
• have you tried
  
  • other kriging models?
  • different semivariogram models?
• parameter optimization?

[DanPatterson_Retired]

sorry the additions didn't clarify anything for me, perhaps slynch-esristaff‌ see something