Handling Outliers in Soil Resistivity Modelling

Modified on Tue, 13 Jun, 2023 at 12:09 PM

SafeGrid Earthing Software can perform automatic multilayer soil modelling calculations from field measurements. If you would like more information, please take a look at this video about how to use SafeGrid.


Soil electrical resistivity (herein referred to as ‘soil resistivity’ for brevity) is crucial when designing an earthing system. Soil resistivity is not fixed and varies in multiple layers. The Wenner test method is commonly used to measure soil resistivity and involves driving four electrode pins into the ground surface, then injecting current and measuring voltage at different (but equal) electrode spacings which provide a measured profile of soil resistivity with depth (related to spacing).


During the earthing design process, a multilayer soil resistivity model is derived based on the measurements by optimally fitting a curve with the measurements. Occasionally there may be errors in some of the soil resistivity data of the measurement set. The problem with the bad (erroneous) measurements is they can cause the optimal model to be skewed. These measurements may be identified by an engineer (simply plotting the measurements may highlight any potential erroneous measurements or outlier data), and if they do not makeup too large a proportion of the measurement set (this would lead to the conclusion the entire measurement set is bad) they can be excluded during the model fitting to derive an accurate soil resistivity model from the measurements.

 

An example soil model calculation

Consider the Wenner method soil resistivity data measured at a particular location, as shown in Table 1. A total of 12 measurements (R1 to R12) were taken at electrode spacings from 1 m up to 50 m, and soil resistivity at the corresponding spacings is recorded in Ω.m. Figure 1 shows the soil resistivity versus electrode spacing graph. It is observed that one of the values, R7 is substantially higher than the other values (in statistical terms is an ‘outlier’).


Note: When the differences between a single soil resistivity measurement (or a small portion of measurements relative to the total number of measurements) and the rest are substantial, they may be treated with suspicion.


Table 1: Wenner method soil resistivity data

 

Spacing (m)

Resistivity (Ω.m)

R1

1

35.49

R2

2

37.23

R3

5

46.54

R4

10

16.17

R5

15

17.15

R6

20

26.13

R7

25

466.62

R8

30

33.24

R9

35

59.69

R10

40

57.68

R11

45

40.68

R12

50

48.55

 

A picture containing text, line, diagram, plot

Description automatically generated

Figure 1. Soil resistivity versus electrode spacing plot.

 


Obtaining the final soil model

In this case, it is assumed the final soil resistivity model will be a four (4) layer model. All the soil resistivity measurements from Table 1 are entered into the Wenner field measurements table within the software.


Note: The newest version of SafeGrid will automatically determine the optimal number of soil layers to match with the data. It also allows field measurements to be entered using the Schlumberger method.


Upon pressing Calculate, three items are plotted; 1. The actual field measurements (marked by crosses), 2. The apparent resistivity curve (smooth black line), and 3. The soil model (stepped blue curve). The objective of the calculation optimisation is to ensure that the apparent resistivity curve fits well with the actual field measurements, and this goodness of fit is indicated by the calculated root mean square error (RMSE) displayed with the model.


Note: A good soil resistivity model will have an RMSE value below 15 %.

 

Results

When the soil model is calculated using all the measurement data points from Table 1 (including R7), the apparent resistivity model must be better aligned with the measured data points. Hence the RMSE for the model is very high at around 82 %. Figure 2 below shows the calculated soil model when R7 is excluded from the analysis. In this case, the model has an RMSE of 17 %, which could be better but still much improved. Therefore, in this case, the 4-layer soil model is a reasonably good fit for the measurements.


A screenshot of a computer

Description automatically generated

Figure 2. Four-layer soil model with outlier R7 excluded. 

Was this article helpful?

That’s Great!

Thank you for your feedback

Sorry! We couldn't be helpful

Thank you for your feedback

Let us know how can we improve this article!

Select at least one of the reasons
CAPTCHA verification is required.

Feedback sent

We appreciate your effort and will try to fix the article