Skip to content

Geostatistical contour plots on site maps

← Visualization


Many of our users want to estimate the extent and concentration of a contaminant plume using SmartData.  To aid in this, we offer an immediate geostatistical contouring option for any uploaded data.   Individual analytes, or direct push detector data can be contoured, and each overlaid to examine the spatial correlation.

Click the Layers button on the specific type of data you'd like to generate contours for.  Select the Contours option in the control panel to activate the contour map, then adjust the slider to increase and decrease the spatial autocorrelation parameter.   By default it is set at 10 meters, meaning that the values found at any particular station location will affect the immediate surroundings out to 10 meters, then the impact will greatly drop off until there is no impact at longer distances.



Interpolation

A lambda interpolation function is applied to every point in a grid pattern in the study area.  Point very close to an actual measurement are greatly influenced by that measurement, and points farther away in distance are less impacted by a measurement. 

The expected spatial autocorrelation function is

where the lambda parameter determines the level of local smoothing between actual measured samples.    Several autocorrelation functions are shown below, with more and less smoothing represented. 

 

Once this grid is completely calculated, the surface is contoured using the CONREC routine as described here: http://paulbourke.net/papers/conrec/

A contour then gives a line which connects points which all are estimated to have nearly the same value being contoured.


Recommendations

Users should be careful not to simply create a contour plot that 'looks good' or shows the maximum value where the samples were taken.  By setting the slider too low, it results in overfitting the data.  The lack of smoothing makes it seem that there is only contamination where the samples were taken.   But setting the slider too high results in over smoothing, which masks the variation throughout the site, and overestimates the extent of the plume.   Our research team is constantly working on methods to measure the uncertainty associated with these models, and we will be reporting out that uncertainty to you in upcoming versions of SmartData so you can make the best informed decisions about future sampling and remediation actions.  Also please note that when creating geostatistical contour plots for analytes, it is important to make sure that all units are the same in your lab data upload (i.e. do not enter concentrations in ug/L and mg/L for the same analyte).





Feedback and Knowledge Base

(thinking…)