Sabine Grunwald

Profile Cone Penetrometer (PCP) Applications

Collaborators
Time
Funding Source
Study Area
PCP Specifications
Data / Sampling Design
Objectives A
Analyses A
Results A
Conclusions / Summary A
Objectives B
Analyses B
Results B
Conclusions / Summary B
Objectives C
Analyses C
Results C
Conclusions / Summary C

Collaborators

S. Grunwald, K. McSweeney, B. Lowery

(Department of Soil Science, University of Wisconsin-Madison)

Data collection: D.J. Rooney, G. Hart, and A. I. Malik

Time

08/1997 - 03/2000

Funding Source

Department of Soil Science, University of Wisconsin-Madison

Study Area

A 2.73-ha site located on the Agricultural Research Station West Madison, University

of Wisconsin-Madison, in southern Wisconsin. Soils were formed in reworked loess overlying

glacial till. The site was mapped as fine-loamy, mixed, mesic Typic Argiudolls. Land use was

alfalfa (Medicago sativa) for the past three years. Mean annual precipitation is 720-mm.

The soil moisture regime is udic.

PCP Specifications

Profile cone penetrometers measure penetration resistance expressed as cone index (kPa).

The PCP system components included a PCP probe, a hydraulic truck- mounted push system

(Gidding #9HD; Fort Collins, CO)1, a load cell (1360-kg capacity; Omegadyne LC 101;

Sunbury, OH), a depth transducer (Unimeasure HX-EP; Corvallis, OR) and data aquisition system.

PCP 1: 60 degree cone angle, 2-cm diameter surface area, no sleeve friction measurement,

and datalogger (Campbell Scientific 21X; Logan, UT).

PCP 2: 30 degree cone angle, tip and sleeve friction measurements and real-time data

aquisistion and analysis system (Applied Research Associates, Inc., South Royalton, Vermont).

PCP 1	PCP 2
Real-time data aquisition system and GPS	Continuous cone index curve

Data / Sampling Design

Data set 1: grid sampling (10-m spacing); 273 penetrations; 21 soil core (4.3-cm diameter) samples

randomly taken at grid locations; soil cores were analyzed by horizons for texture, bulk density, and

water content. Soil texture was analyzed by the UW Soil and Plant Analysis Laboratory (Madison, WI)

based on the hydrometer method. Soil water content was derived by collecting known volumes of

samples and oven drying them to obtain volumetric values. Bulk density was determined by oven drying

a known volume of soil and presented on dry weight basis.

Data set 2: targeted sampling design based on an entropy analysis of topographic attributes; 77 penetrations

conducted with PCP probe 1 and 2; 77 soil cores analyzed in 5-cm depth increments for bulk density and

water content.

A Trimble 4600 LS differential global positioning system (GPS) (Trimble, 1996; Sunnyvale, CA) was used

to georeference sampling locations and to develop a digital elevation model (DEM) for the study site.

Objectives A

To use PCP and soil property data to distinguish among soil materials (reworked loess and glacial till)

Analyses A

Steps (data set 1):

(1) We grouped cone index profiles (dense data set) using a hierarchical cluster analysis

(2) We used landform element classes as supplementary information to define groups

(3) Experts defined characteristics of soil materials

(4) We calculated mean and standard deviation for soil attributes (sparse data set) and identified soil

materials

(5) We combined cone index and soil property data - transfer of soil material information from the sparse

soil property data set to the dense cone index data set

(6) We created a 3D soil layer model showing the spatial distribution of soil materials. 2D ordinary kriging

was used to generate surfaces of layers and volumes of layers were created with linear interpolation in

the vertical direction between the surfaces (EVS-PRO software)

Results A

Grunwald S., B. Lowery, D.J. Rooney, and K. McSweeney. 2001. Profile cone penetrometer

data used to distinguish between soil materials. Soil & Tillage Research 62: 27-40.

Grunwald S., B. Lowery, M.K. Clayton, K. McSweeney, G.L. Hart, and D.J. Rooney. 1999.

Using a penetrometer to produce a 3-D Model of soil patterns. ASA-CSSA-SSSA Annual

Meeting, Salt Lake City, Oct. 31- Nov. 4, 1999, Div. S-5, pp. 271.

Rooney D.J., J.M. Norman, S. L. Lieberman, and S. Grunwald. 1999. Characterizing soil

properties in 3-D using multiple sensor penetrometer technology. ASA-CSSA-SSSA Annual

Meeting, Salt Lake City, Oct. 31 – Nov. 4, 1999, Div. S-1, pp. 185.

Grunwald S., K. McSweeney, B. Lowery, and D.J. Rooney. 1998. Continuous description

of soil attributes on a landscape in southern Wisconsin. ASA-CSSA-SSSA Annual Meeting,

Baltimore, Oct. 18-22, Div. S-5, pp. 253.

Mean and standard deviation for clay, silt and sand content, bulk density (r_b) and soil

volumetric water content (q) for layers identified by the hierarchical cluster analysis.

Cluster B was further classified using landform element information into cluster B1 and B2.

Table 1. Mean and standard deviation of soil attributes.

	Attribute	Cluster A	Cluster B	Cluster B1	Cluster B2	Cluster C	Cluster D
1	Clay (%)	13.5†; 0.9‡	21.7; 1.4	20.5; 1.2	11.8; 0.6	21.5; 1.4	22.5; 0.6
	Silt (%)	63.0; 1.9	59.5; 2.0	58.6; 1.4	60.3; 1.7	58.3; 1.4	65.5; 1.5
	Sand (%)	22.1; 2.0	21.7; 1.9	20.6; 1.4	22.7; 1.8	21.2; 2.4	10.1; 1.0
	r_b (Mg m^-3)	1.30; 0.12	1.39; 0.07	1.45; 0.05	1.32; 0.11	1.60; 0.10	1.71; 0.10
	q (m³m^-3)	0.19; 0.012	0.188; 0.006	0.184; 0.006	0.192; 0.006	0.188; 0.007	0.194; 0.005
2	Clay (%)	13.3; 1.0	21.6; 1.9	20.7; 1.1	22.5; 0.7	26.0; 0.9	24.3; 0.7
	Silt (%)	62.1; 1.8	62.5; 2.0	61.9; 1.1	63.1; 0.9	67.2; 1.1	62.2; 1.5
	Sand (%)	36.3; 7.9	15.7; 6.1	20.3; 1.5	11.1; 1.2	12.1; 1.6	11.2; 1.7
	r_b (Mg m^-3)	1.38; 0.02	1.43; 0.09	1.50; 0.07	1.36; 0.05	1.38; 0.13	1.37; 0.05
	q (m³m^-3)	0.198; 0.003	0.197; 0.006	0.200; 0.011	0.194; 0.012	0.197; 0.005	0.199; 0.004
3	Clay (%)	9.1; 1.2	20.0; 15.1	10.3; 0.7	30.0; 0.5	-	-
	Silt (%)	9.1; 1.0	36.3; 28.5	10.4; 1.2	62.1; 0.7	-	-
	Sand (%)	80.5; 9.1	43.9; 35.0	79.0; 1.3	8.8; 0.9	-	-
	r_b (Mg m^-3)	1.71; 0.05	1.58; 0.11	1.65; 0.07	1.51; 0.10	-	-
	q (m³m^-3)	0.127; 0.006	0.169; 0.049	0.121; 0.012	0.217; 0.004	-	-

† Mean, ‡ SD

Reworked loess

Glacial till

3D soil-layer model showing the spatial distribution of soil materials

(top layer, reworked loess, and glacial till).

Conclusion / Summary A

In this study, we used a heuristic approach using a working hypothesis about the spatial distribution

of soil materials. The objective o this study was to distinguish among soil materials found in a glaciated

landscape in southern Wisconsin. We used a dense data set of cone penetrations, a sparse set of

laboratory measured soil properties, and landform element classes to accomplish our goal. Thus, we

combined rapid sampling techniques with cost-intensive and labor intensive procedures. This method

yielded good results mapping contrasting soil materials such as reworked loess and glacial till. Results

indicated that combined information consisting of cone index, soil properties, and geographic location

improved soil material mapping.

Objectives B

(a) To describe continuously the spatial distribution of soil materials and layers in three space

dimensions.

(b) To evaluate two different approaches to create 3D soil layer models utilizing cone index profiles

and soil property data

Analyses B

Steps:

(1) Crisp hierarchical clustering / vertical point inflection method (data set 1)

(2) Fuzzy k-mean classification / defuzzification (data set 1)

(3) Validation procedure (data set 2): the root mean square error (RMSE) was used to evaluate the

accuracy of predictions and the coefficient of efficiency (E) was used to evaluate the fit of

measured vs. predicted values

(4)

We created 3D soil layer models showing the spatial distribution of soil materials. 2D ordinary kriging

was used to generate surfaces of layers and volumes of layers were created with linear interpolation in

the vertical direction between the surfaces (EVS-PRO software)

Results B

Grunwald S., K. McSweeney, D.J. Rooney, and B. Lowery. 2001. Soil layer models created

with profile cone penetrometer data. Geoderma 103: 181-201.

Grunwald S., D.J. Rooney, K. McSweeney, and B. Lowery. 1999. Application of a profile

cone penetrometer to distinguish between soil materials. Proceedings 3^rd Conference of the

Working Group on Pedometrics of the International Union of Soil Science (IUSS),

Sept. 27-29, 1999, Sydney, Australia.

The RMSE and E coefficient for different layers and methods:

Layers	Crisp method		Fuzzy method
	RMSE	E	RMSE	E
1	3.81	0.96	3.67	0.97
2	4.18	0.90	15.69	0.51
3	4.98	0.86	17.33	0.40

3D soil layer models created with two different methods

Conclusions / Summary B

Results indicate that the crisp method overall produced better predictions than the fuzzy method.

The crisp method yielded excellent results predicting the spatial variability of changes in layer

depth and / or soil material across the 2.73-ha site in southern Wisconsin. An advantage of using

cone index data is rapid and non-invasive data collection. These data, combined with limited soil

coring are ideal to create 3D soil layer models.

Objectives C

(a) To develop pedotransfer functions, which describe the relationship between cone index and

soil texture,
bulk density, and
water content

(b) To evaluate the sensitivity of parameters used in these functions

Analyses C

Steps (data set 1):

(1) We derived cone index layer profiles (CILP) using horizontal hierarchical clustering and a

vertical point inflection method

(2) Within each CILP group, cone index was regressed with soil physical properties to develop

pedotransfer functions

(3) Pedotransfer functions were evaluated using the coefficient of determination (R²)

(4) We conducted a sensitivity analysis to evaluate the relative importance of different parameters

in the regression models

Results C

Grunwald S., D.J. Rooney, K. McSweeney, and B. Lowery. 2001. Development of pedotransfer

functions for a profile cone penetrometer. Geoderma, 100: 25-47.

Table 2. Coefficients for multiple linear regression equations for prediction of cone index (CI) in kPa.

Data sets	M	Intercept	Depth (cm)	Sand (%)	Silt (%)	Clay (%)	r_b (Mg m^-3)	q (m³ m^-3)	R²	Eq.
G	S	818.45	13.15						0.345	1.1
	S	-2,318.77	10.39				2,348.36		0.409	1.2
	S	-1,434.77	12.88			-25.61	2,031.77		0.446	1.3
	S	-3,615.65	13.52			-32.74	2,704.26	69.61	0.468	1.4
	E	-3,371.53	12.52		-12.54	-27.48	2,614.97	101.29	0.480	1.5
CILP1	S	1,029.54		14.75					0.445	1.6
	E	4,992.29	3.75	1.16		-39.80	-1,203.12	-80.87	0.621	1.7
CILP2	S	1,820.82				-30.57			0.331	1.8
	S	1,082.59	37.81			-46.14			0.422	1.9
	E	5,853.12	48.16	86.76	15.04		-1,483.21	-288.05	0.760	2.0
CILP3	S	852.58	8.53						0.510	2.1
	S	-820.41	6.88				1,284.51		0.655	2.2
	E	-3,209.55	6.49		15.82	-1.87	2,410.55	-4.78	0.700	2.3
CILP4	S	-18,182.50				735.50			0.455	2.4
	S	-23,509.00				574.02		482.03	0.601	2.5
	S	-19,755.80				606.14	-2,107.70	411.32	0.622	2.6
	E	23,395.32	56.29			527.97	-12,414.9	-1,110.04	0.630	2.7
CILP5	S	532.47	33.21						0.851	2.8
	S	-4679.62	30.32				2,883.32	69.20	0.900	2.9
	E	-5,777.01	27.54		29.03	-90.66	3,775.17	117.09	0.980	3.0

G: global data set

CLIP1 to CILP5: cone index layer profiles

M: Methods, S: stepwise; E: enter

Example Eq. 1.2: CI = -2,318.77 + 10.39 * depth + 2,348.36 * r_b

Spatial distribution of CILP in 3D geographic space

Pedotransfer functions using the global data set showed large sensitivities for bulk

density and soil water content. Similar results were calculated for all other CILPs,

except CILP4 where clay content showed large sensitivity.

Conclusions / Summary C

The R² for pedotransfer funcitons ranged from 0.345 up to 0.98. Depth, bulk density,

clay content and water content showed largest predictability for cone index using the

global data set. Textural variables had strong predictive power in the top layers, CILP1

and CILP2. In CILP4, clay content along with bulk density and soil water content were

variables with large predictive power. In contrast, the predictive power of bulk density

and soil water content was strong in layers CILP3 and CILP5, whereas soil textural

characteristics showed low predictability of cone index.

The sensitivity analysis calculated the relative importance of variables using a specific

pedotransfer function to predict cone index. Sensitivity values have merit for any user

who might apply one of the pedotransfer functions. Parameters with largest sensitivity

have the greatest impact on model output and therefore should be collected/measured

carefully.

Penetrometer measurements are useful to describe gradual and continuous variation

of cone index and associated soil properties within fields and within soil map units.

Besides detailed measurements of soil physical and hydraulic properties, the collection

of cone index facilitates to improve our knowledge of the within-field spatial variability

of soil properties. This is the basis for effective site-specific farming which is environ-

mentally sound and economically. The costs for measurements of texture, bulk density

and water content with high precision would be higher when compared with the

approach presented in this paper. A drawback of using pedotransfer functions is the

uncertainty associated with estimates.

We believe that pedotransfer functions for cone penetrometers would make these

devices widely applicable. Measurements of cone index are rapid and the costs are modest

when compared to traditional soil sampling of soil physical properties. We conclude that

this technique shows promise to improve fine-scale sampling of soil physical properties.

Considering the limited nature of this study, these results strongly indicate the feasibility of

this approach.