Carré & Jacobson RC Ispra - IES 1/17 Soil profile distance measures and classifications A B Carré & Jacobson D(A;B ) Application to an Australian soil dataset Pedometrics 2007, 27-29/08/2007
Jan 03, 2016
Carré & Jacobson
JRC Ispra - IES 1/17
Soil profile distance measures and classifications
A B
Carré & Jacobson
D(A;B)
Application to an Australian soil dataset
Pedometrics 2007, 27-29/08/2007
Carré & Jacobson
JRC Ispra - IES 2/17
Objectives
To make a quantitative tool for soil profile classification
Which provides different types of distances according to the soil classification purpose
Which is able to make supervised classification (principle of soil taxonomies) or unsupervised classification
To test the quantitative tool relatively to
Soil taxonomy purpose
Available Water Capacity prediction
OSACA presentation
Application
Focusing on
The input content of the classification (soil against terron)
The number of classes (level of taxonomic detail)
The classes robusteness (sensitivity of the distances)
Carré & Jacobson
JRC Ispra - IES 3/17
OSACA principles
Soil observations Horizon classification
horizons
User requirements
Range of horizon classes
Range of solum classes
Min. # Horizon
centroids Calculation of
distances
Calculation of new
centroidsDintra/Dinter ratio
Min+1. # Horizon
centroids
Max. # Horizon
centroids
Horizon types
Solum classification
Min. # solum
centroids
Calculation of
distances Calculation
of new centroids
Dintra/Dinter ratio
Min+1 # solum
centroids
Max # solum
centroids
Soil types
optimization
Carré & Jacobson
JRC Ispra - IES 4/17
OSACA principles
Soil observations
Soil types
D(obs,r
ef)
1 0.7 1.30.1 0.3 B 0.1
2
3
4
5
A B C D REF
2.5
3.0
0.2
1.20.1
0.4
0.8
0.6
0.61.5
0.1
0.1
0.0
0.1
1.2
1.9
A
C
B
B
Result table
dmin
0.1
0.1
0.1
0.1
Soil observations
Soil types
If user has references, only one iteration
Carré & Jacobson
JRC Ispra - IES 5/17
OSACA distances
Horizon distances between horizon i and horizon k having n soil variables
Euclidian distance
Manhattan distance
Carré & Jacobson
JRC Ispra - IES 6/17
OSACA distances
Solum distances
D(x,A)
D(x,B)
D(y,B)
D(y,C)
D(z,C)
D(?,C)
x
y
z
0 cm
30 cm
80 cm
100 cm
Observation
0 cm
20 cm
60 cm
120 cm
A
B
C
Reference
Pedological distance
+
+
+
+
+
D
(# D)
Carré & Jacobson
JRC Ispra - IES 7/17
OSACA distances
Solum distances
D(x,A)
D(x,B)
D(y,B)
D(y,C)
D(z,C)
D(?,C)
x
y
z
0 cm
30 cm
80 cm
100 cm
Observation
0 cm
20 cm
60 cm
120 cm
A
B
C
Reference
Utilitarian distance D*e
Max depth
+
+
+
+
+
Carré & Jacobson
JRC Ispra - IES 8/17
OSACA distances
Solum distances
D(x,A)
D(x,B)
D(y,B)
D(y,C)
D(z,C)
x
y
z
0 cm
30 cm
80 cm
100 cm
Observation
0 cm
20 cm
60 cm
120 cm
A
B
C
Reference
Joint distance
x
y
z
0
0.3
0.8
1
0
0.16
0.5
1
A
B
C
+
+
+
+
D
(# D)
Carré & Jacobson
JRC Ispra - IES 9/17
ApplicationEdgeroi area
341 soil profiles
Altitude
Slope
Aspect
Plan curvature
Profile curvature
CTIDEM 25m
Landsat ETM 7c
LS Panchromatic
NDVI
Clay Index (5/7)
RS 30m
K
U
Th
2m
SPOT 4
NDVI
RS 20m
Soil variables
9 order names/ 20 subgroup names
AWC
N
Carré & Jacobson
JRC Ispra - IES 10/17
Application
Sand (%)/ Silt (%)/ Clay (%)
pHH20
CaCO3
EC
Munsell Colour (Hue, Value, Chroma)
Solum depth
Profile description
0 cm
10 cm
20 cm
30 cm
40 cm
70 cm
80 cm
120 cm
130 cm
C
R, G, B (0- 255)
Illuminant C two degrees observer (C 1931)http://WalkillColor.com
Standardization of soil variables (320 soil profiles)
0 cm
10 cm
20 cm
30 cm
40 cm
70 cm
80 cm
120 cm
130 cm
s.depth
Ca,Mg,Na
Carré & Jacobson
JRC Ispra - IES 11/17
Tests
Testing OSACA classifications against soil taxonomy
1st question: Do we have enough data to speak about pedogenesis?
Texture
pHH20
CaCO3
EC
Munsell Colour
Solum depth
Ca,Mg,Na
C
Pool of soil variables
Sufficient pool of
variables
Rudo Teno Verto Kuro Sodo Chromo Calcaro Dermo Kando
Clay>35% 0 0 1 0 0 0 0 1 0
throughout
Cracks & 0 0 1 0 0 0 0 0 0
slickenside
Texture contrast 0 0 0 1 1 1 0 0 0
Sodic subsoil 0 0 1 0 1 0 0 0 0
B2 pH<5.5 0 1 1 1 0 0 0 1 1
B2 pH>5.5 0 1 0 0 1 1 1 1 1
Calcareous 0 0 0 0 1 0 1 1 0
Lack texture 1 1 1 0 0 0 1 1 1
contrast
Structured B2 0 1 1 1 1 1 1 1 0
Massive B2 0 1 0 0 0 0 0 0 1
Rudiment B 1 0 0 0 0 0 0 0 0
Isbell et al. (1997) from Minasny & McBratney (to be published)
Lack in DB
Carré & Jacobson
JRC Ispra - IES 12/17
Tests
Testing OSACA classifications against soil taxonomy
Testing OSACA results against the 20 taxonomy suborders
13 Soil variables
OSACA run OSACA run
15 Landscape attributes
PCA
13 PC Landscape attributes
18-25 Horizon classes
15-25 soil classes
Pedological distance
Utilitarian distance
Joint distance
Terron classes
35-45 soil classes
15-25 terron classes
35-45 terron classes
19 classes
20 classes
20 classes
45 classes
45 classes
43 classes
R= 40%
R= 29%
R= 30%
R= 51%
R= 62%
R= 49%
18 classes
19 classes
24 classes
45 classes
44 classes
43 classes
R= 38%
R= 29%
R= 33%
R= 51%
R= 50%
R= 48%
R= 41% R= 52% R= 38% R= 52%
R= 31% R= 51% R= 34% R= 49%
R= 30% R= 63% R= 30% R= 52%confusions (dmin+ 10%)
Allocation comparison
² test
R²(U) likelihood-ratio
Carré & Jacobson
JRC Ispra - IES 13/17
Tests
Testing OSACA classifications against Available Water Capacity (AWC)
AWC1m predicted by Minasny (2007) from:
Sand, Clay, BD
Sand, Org C Bulk Density (BD)
Field Capacity (FC) & Permanent Wilting Point (PWP)
AWC FC & PWPhor
1m
AWCi= K + tdit + it=1
C
Pedological distance
Utilitarian distance
Joint distance
19 classes
20 classes
20 classes
45 classes
45 classes
43 classes
R²adj= 77%
R²adj= 57%
R²adj= 57%
R²adj= 82%
R²adj= 70%
R²adj= 71%
18 classes
19 classes
24 classes
45 classes
44 classes
43 classes
R²adj= 72%
R²adj= 58%
R²adj= 58%
R²adj= 83%
R²adj= 69%
R²adj= 69%
SOIL TERRON
R²adj prediction
Carré & Jacobson
JRC Ispra - IES 14/17
Conclusions
OSACA is a good tool for transforming soil observations into quantitative classes. As a WebApplication, it is easy to use and as an open source, it can be modified.
The quantitative soil and terron classes formed by OSACA are significatively correlated with soil taxonomy (if enough soil variables to describe the pedogenesis) and secondary soil variables.
For dealing with soil taxonomy, the soil classes are better predictors than the terron classes. Joint distance and pedological distances give better correlations between soil classes and Australian soil subgroups.
The distances can be afterwards used for mapping purposes and for deriving uncertainties associated to predictions (DSM purposes).
For dealing with secondary soil variable, the pedological distance is the best predictor. With low and high number of classes, soil seems to be the best predictor. The terron does not increase so much the prediction.
The allocation confusion due to close distances changes 1% of the correlations.
Carré & Jacobson
JRC Ispra - IES 15/17
Enter your observations
Enter your references (if you have some)
! Soon downloadable at http://eusoils.jrc.it !
Carré & Jacobson
JRC Ispra - IES 16/17
Enter the number of horizon classes
Enter the number of soil classes you want
Carré & Jacobson
JRC Ispra - IES 17/17
Get the results (tables)