ARTICLE IN PRESS

Geoderma xx (2004) xxx – xxx

www.elsevier.com/locate/geoderma

Using the topography of the saprolite upper boundary to improve

the spatial prediction of the soil hydromorphic index
V. Chaplot a,b,*, C. Walter b, P. Curmi b,c, P. Lagacherie d, D. King e
a
IRD-UR049, Ambassade de France, BP 06, Vientiane, Laos
b
USARQ, Unité Science du Sol et d’Agronomie, ENSA-INRA, 65, rue Saint Brieuc, 35042 Rennes, cedex, France
c
ENESAD, ‘‘Milieu Physique et Environnement’’, BP 87999, 26 Bd du Dr Petit jean, 21079 Dijon, cedex, France
d
UMR LISAH, Laboratoire d’étude des Interactions Sol – Agrosystème – Hydrosystème. ENSAM-INRA-IRD,
2 place Pierre Viala, F-34060 Montpellier, cedex 02, France
e
INRA, Science du Sol, Avenue de la Pomme de Pin, B.P. 20619, Ardon, 45166, Olivet cedex, France
Received 11 June 2003; accepted 1 March 2004

Abstract

So far, soil-landscape models have been based on soil surface topographic information only. However, hillslope hydrology
that affects soil distribution is also controlled by sub-surface flow pathways that may not entirely be explained by surface terrain
features. This paper compares the accuracy of a model for predicting the spatial variations of the hydromorphic index (HI) using
the surface topography with that of a model that uses the sub-surface topography. The study was conducted in an agricultural
hillslope of the Armorican Massif (Western France) where two DEMs were generated from observations of the surface
topography and the topography of the saprolite upper boundary. For both DEMs, the best correlations with HI occurred for
altitude, elevation above the stream bank, and compound topographic index. Predictions of HI using the sub-surface topography
greatly decreased prediction errors, especially for intermediary HI values at middleslope position. Finally, the use of subsurface
topography as a novel approach to enhance existing techniques in new applications is discussed.
D 2004 Published by Elsevier B.V.

Keywords: Hydromorphic index; HI; DEM; Modelling; Sub-surface topography

Abbreviations: As, specific monodirectional catchment area; Asm, specific multidirectional catchment area; b, length of an element of
contour; C, carbon; Ch, moist Munsell chroma; CTI, compound topographic index [CTI=ln(As/S)]; CTIm, modified compound topographic
index [CTIm=ln(Asm/DG)]; DEM, digital elevation model; DG, downslope gradient; ES, elevation above the stream bank; F, F value; GPS,
global positioning system; H, moist Munsell hue; HI, hydromorphic index; HI*, estimated hydromorphic index; Kp, profile curvature; Kc,
contour curvature; Kt, tangential curvature; L, upslope length; Lat., latitude; Long., longitude; MAE, mean absolute error; ME, mean error; OC,
organic carbon; P, cumulative thickness of soil horizons with redoximorphic features divided by the total thickness of the soil (the total thickness
of the soil includes the thickness of the loamy horizons and of the loose saprolite to a maximum depth of 3 – m); q, area drainage flux; r,
correlation coefficient; S, slope gradient; Sd, standard deviation; SPI, stream power index [SPI=(As
S)]; SD, soil depth; T, local soil
transmissivity; V, moist Munsell value; Z, elevation above sea level.
* Corresponding author. IRD-UR049, Ambassade de France, BP 06, Vientiane, Laos. Tel.: +856-20-50-77-10; fax: +856-21-41-29-93.
E-mail address: chaplotird@laopdr.com (V. Chaplot).

0016-7061/$ - see front matter D 2004 Published by Elsevier B.V.

doi:10.1016/j.geoderma.2004.03.001

GEODER-02197; No of Pages 12
 ARTICLE IN PRESS
2 V. Chaplot et al. / Geoderma xx (2004) xxx–xxx

1. Introduction compound topographic index (CTI, Moore et al.,

1993) and revised compound topographic index
Topography is recognized as a main controlling (CTIm = ln(As/DG), Mérot et al., 1995); and the terrain
factor of soil distribution in landscapes. During the curvatures (Burt and Butcher, 1986; Thompson et al.,
last century, the effects of such an environmental 2001) could be cited.
factor on soil processes was well documented within So far, modelling approaches for the prediction of
pedogenesis studies (Jenny, 1941). soil spatial distribution assume both lateral surface
Topography may affect soil distribution through and sub-surface flows to be defined by surface topog-
erosion processes such as those linked to interrilling, raphy only. However, when soil infiltration is not
rilling, and sedimentation (e.g. Poesen, 1984; Govers, negligible, sub-surface pathways in particular may
1990; Torri and Poesen, 1992; Fox and Bryan, 1999; be even more affected by the topography of sub-
Chaplot and Le Bissonnais, 2003). One of the other surface discontinuities such as impermeable horizons
possible effects of topography on the soils spatial or saprolite upper limits. In the Armorican Massif,
distribution is through the control of the duration of soils have developed into a loamy material of 1 to 3 m
saturation by water or reducing conditions. Water depth overlying impermeable granitic or schistic sap-
drainage tends to accumulate in favorable locations rolites. Recent soil wetness prediction models showed
such as hillslope hollows and footslopes, inducing O2 that the knowledge of the topography of the saprolite
depletion responsible for the formation of redoximor- upper limit highly enhanced the prediction of the
phic features such as H2S and Fe– Mn concretions due spatial distribution of soil wetness at different depths
to the transformation and translocation of these two (Chaplot and Walter, 2003). The model based on
elements under biochemical reduction (e.g. Vepraskas terrain attributes estimated using sub-surface topogra-
and Wilding, 1983; Thompson and Bell, 1996). These phy decreased the prediction errors of soil surface (0–
features are widely used by world reference bases or 10) wetness by 43% compared to the use of soil
classifications since they are diagnostic criteria for soil surface topography only. In this context, sub-surface
classifications (Soil Survey Staff, 1999; WRB, 1998). topography would also improve the spatial prediction
Local soil saturation occurs in hillslopes when the of soils affected by an excess of water. However,
accumulated water flux, the product of the catchment because soils may have formed under land-forms that
area As and the area drainage flux q, passing across an have been afterwards reshaped over time, their spatial
element of contour of length b, exceeds the product of distribution could be totally uncorrelated with sub-
the local soil transmissivity T and the local surface surface forms that control the actual hydrologic func-
gradient S (O’Loughlin, 1986). tioning of landscapes.
A better knowledge of the spatial distribution of Until now, little work has been done at the hillslope
soils affected by water saturation is crucial since these level on the relation between sub-surface topography
soils: (i) determine the production of surface runoff, and soil distribution. The objective of this study was
which is the driving mechanism for soil erosion, to examine how the use of the topography of the
nutrient or pesticide losses in hillslopes; (ii) are saprolite upper boundary, i.e. the lower limit of the
responsible for floods in downstream areas; (ii) pres- soil cover could improve the quantitative modelling of
ent a nitrogen buffering capacity controlled by micro- the spatial prediction of soil distribution. The study
bial denitrification, and microbial and plant uptakes was conducted in an agricultural hillslope of the
(Haycock and Pinay, 1993). Armorican Massif (Western France) characterized by
Due to the close relationship between topography impermeable granitic saprolites. A continuous index
and soil saturation, some terrain attributes have been of soil hydromorphy (HI) is considered here (Eq. (1),
used to predict the spatial distribution of soils affected Chaplot et al., 2000).
by an excess of water. Among these attributes, the P
elevation above the stream bank (E, Crave and Gas- HI ¼ ð1Þ
V
Ch
cuel-Odoux, 1996); the downslope gradient (DG,
Mérot et al., 1995), which is the ratio between ES With P, the cumulative thickness of the horizons
and the shortest distance to the stream (L); the with redoximorphic features divided by the total
 ARTICLE IN PRESS
V. Chaplot et al. / Geoderma xx (2004) xxx–xxx 3

thickness of the soil (the total thickness of the soil mm and 10.1 jC, respectively. The area was charac-
includes the thickness of the loamy horizons and of terised by relatively moderate agriculture intensifica-
the loose saprolite to a maximum depth of 3 m); V, the tion. From 1995 to 1997, only 65% of the total surface
Munsell value of the 0 – 10 cm surface layer; Ch, the area was under pasture, the remaining surface being
Munsell chroma of the 0– 10 cm surface layer. under corn. The near-stream-bank zones were exclu-
Two soil-landscape models based on multiple-re- sively under pasture.
gression relationships between the soil hydromorphic The Armorican Massif is a complex basement of
index and terrain attributes were considered here. The Proterozoic and Paleozoic ages bedrock. In this
first was generated from terrain attributes of the soil Massif, granites and schists are the two dominant
surface, whereas the second considered attributes substrata, comprising 80% of the region’s surface.
derived from the topography of the saprolite upper Other substrates include mica schists, orthogneisses
boundary. The prediction quality of these two models and sedimentary rocks. Deep mechanical weathering
was evaluated in an entire hillslope by comparing soil of the bedrock occurred during the Variscan orogeny,
observations and estimations from soil-landscape where mylonitization and strong fracturation oc-
modeling. curred. This weathering produced deep saprolites
(>30m), rich in kaolin and mainly impermeable
(Wyns, 1991). During the late Miocene and early
2. Materials and methods Quaternary, saprolites were subjected to erosion, due
to strong cyclic climatic regime changes, from peri-
2.1. Site description glacial to temperate and a succession of tectonic
movements (Van Vliet-Lanoe et al., 1998). Finally,
The study site is an agricultural hillslope located saprolites have been overlaid by eolian deposits
in a granitic watershed (La Roche watershed) of blown during the last 300 ky from west to east of
eastern Armorican Massif (western France, Fig. 1). the Armorican Massif. Afterwards deposits were
La Roche watershed has a surface area of about 7.8 locally reworked on slopes by solifluction, water
km2, a perimeter of 18.9 km and altitudes from 229 erosion and local alluviation (Le Calvez, 1979).
m in its northern part to 44 m at the outlet in the Due to the presence of sub-surface impermeable
South. The length of permanent reaches was 15.3 km, layers, water drainage tends to accumulate in favor-
corresponding to a stream density of 1.96 km km 2. able locations such as hillslope hollows and foot-
The mean slope of the watershed, estimated from a slopes where waterlogged soils are observed
50-m DEM, was 8.2% with a standard deviation of (Chaplot et al., 2000). Well-drained soils character-
4.6%. The main reach, oriented north –south, was ized upper parts of hillslopes.
parallel to the western border of the watershed. Four At the study hillslope, the soil cover has developed
tributaries drain the eastern part along the northeast/ into a loamy material overlying a typical impermeable
southwest axis. At the outlet of the ‘‘La Roche’’ granitic saprolite (Curmi et al., 1998). The depth of
catchment, a mean flow discharge of 0.059 m3 s 1 the soil cover varied from 0.8 to 3 m. Variations in soil
was observed during the 1995– 1997 period (Gri- depth do not seem to be directly linked to the
maldi and Chaplot, 2000). topography of the soil surface. The study hillslope is
The study hillslope is located in the northeastern characterised by a typical succession downslope –
part of the ‘‘La Roche’’ watershed. It is a 150- to upslope of fibric Fluvisols, Albeluvisols, stagnic and
250-m-long hillslope with a mean elevation range gleyic Luvisols, and well drained Luvisols (WRB,
of about 10 m since altitudes varied from 190 to 1998). Fibric Fluvisols showed high Munsell Chroma
203 m (Fig. 2). The topography is relatively Ch, value, V, HI values and exhibited high organic
smooth with downslope gradients from 0% at the matter accumulation (mean of 350 kg C m 2; Chaplot
alluvial plain and hillslope summit to 8 – 10% et al., 2001a). Previous studies have also shown that
middleslope (Fig. 2). loamy horizons of Albeluvisols, stagnic and gleyic
The climate is oceanic with a yearly mean precip- Luvisols are marked by low hydraulic permeability,
itation and temperature for the last 30 years of 898 i.e., average values of 0.007 m day 1 (Curmi et al.,
 ARTICLE IN PRESS
4 V. Chaplot et al. / Geoderma xx (2004) xxx–xxx

Fig. 1. Location of the ‘‘La Roche’’watershed in the Armorican Massif (Western France). Position of the study hillslope and sampling points for
topographic and soil science investigations.
 ARTICLE IN PRESS
V. Chaplot et al. / Geoderma xx (2004) xxx–xxx 5

Fig. 2. Spatial variations of altitude (Z), multidirectional area (Asm), downslope gradient (DG), and modified compound topographic index
(CTIm) over the study hillslope and limited area for soil-landscape models generation and validation.
 ARTICLE IN PRESS
6 V. Chaplot et al. / Geoderma xx (2004) xxx–xxx

1998), whereas underneath saprolites exhibit much downslope gradient (DG) (Mérot et al., 1995), i.e.
lower permeability ( < 0.005 m day 1, Grimaldi and the ratio between ES and the upslope length (L). All
Chaplot, 2000). have a physical significance. The CTIm [CTIm = l-
n(Asm/DG)], a modified CTI with DG in place of S
2.2. Topographic analysis of the surface and the and Asm in place of As which showed higher correla-
sub-surface tions with soil hydromorphy or soil wetness (Chaplot
et al., 2000; Chaplot and Walter, 2003) was used.
At the study hillslope, two DEMs with a 10-m Other terrain attributes include the elevation of data
resolution were generated using topographic informa- points (Z), the profile (Kp), contour (Kc) and, tangen-
tion from a 1:25,000 topographic map showing con- tial (Kt) curvatures (Zevenbergen and Thorne, 1987).
tour lines with a 5-m interval and from a detailed L, Z, ES, S, As, and the curvatures were derived using
topographic survey using a theodolite. The first DEM the GRID function of Arc-Info (ESRI, 1994). We
is a numerical representation of the soil surface estimated Asm using software developed at ENSA
topography, generated from the contour lines with a Rennes (Squividant, 1994) considering the algorithm
5-m interval and 643 additional theodolite points (Fig. of Quinn et al. (1991).
1). The second DEM, a digital elevation model of the
upper boundary of saprolites, was generated over the 2.3. Soil sampling scheme at the study hillslope
hillslope using the set of 643 data points where the
depth to saprolite upper boundary was determined by Sampling points were selected along the central
auger hole. Outside the study hillslope, where no soil area of the study hillslope for models generation and
observations were performed, additional knowledge validation (Fig. 3). Ten transects, perpendicular to the
was needed to generate a DEM. The contour lines of contour lines from the channel network to the top of
the 1:25,000 topographic map were used to derive the the hillslope, were regularly sampled at a 10-m
depth to the saprolite upper boundary by subtracting 1 interval. The distance between transects was also 10
m, the mean depth observed at the study catchment, m. In addition, some intermediary data points were
from the altitude. observed to precisely define the limits between soils.
Both DEMs were generated using the GRID tool At the study area, 141 data points, randomly
of Arc Info 7.1 (ESRI, 1994). The fitting procedure selected from an initial set of 182, constituted the
(‘‘TOPOGRID’’ function of Arc/Info 7.1 Geographic set for models generation. The remaining 41 data
Information System, Hutchinson, 1989) ensures the points were aimed at validating the soil-landscape
smoothest fit to point and contoured elevation data models (Fig. 3).
taking into account specified break lines such as Accurate coordinates (Long., Lat.) of each of the
rivers or ravines. The actual stream bank position data points were derived with a theodolite from a
was used in the drainage enforcement algorithm of single geo-referenced point using a satellite differential
TOPOGRID, which attempted to modify the DEM to global positioning system (GPS). At each auger point,
yield continuous drainage surfaces. At the study the following parameters were estimated every 0.1 m
hillslope, the vertical accuracy of DEMs was 0.3 m. from the soil surface to a depth of at the most 3 m: (i)
Several terrain attributes or indices derived from the moist Munsell hue (H), chroma (Ch) and value (V);
Darcy’s law, originally developed for hydrological (ii) the occurrence of redoximorphic characteristics,
and soil modelling (Beven and Kirkby, 1979) were stagnic, gleyic, oximorphic and/or reductomorphic,
considered here: the compound topographic index and eventually albeluvic tonguing properties; (iii) the
[CTI = ln(As/S)] (e.g. Moore et al., 1993; McKenzie presence of eluvial or clay-enriched horizons. These
and Ryan, 1999); the stream power index morphological observations were computed to esti-
(SPI = As
S) (Moore et al., 1993) with As the specific mate the hydromorphic index, HI, following Eq. (1).
catchment area and S the slope gradient (Moore et al., At the study area, the regular and high sampling
1991). In addition, previous studies in the Armorican density allowed the use of a simple true interpolator.
Massif have shown that the elevation above the stream Therefore, spline interpolation was used to interpolate
bank (ES) (Crave and Gascuel-Odoux, 1996); the HI between observations.
 ARTICLE IN PRESS
V. Chaplot et al. / Geoderma xx (2004) xxx–xxx 7

Fig. 3. Spatial distribution of observed HI within the area for model generation and validation, and position of the 182 data points of the initial
set (A). Interpolated MAE of HI prediction through soil-landscape modeling considering the topography of the soil surface (B) and that of the
saprolite upper boundary (C). Forty-one data points of the validation set are considered.

2.4. Statistical analysis and modelling of the upper boundary of saprolites was investigated
using correlation matrixes and regression techniques.
The relation between terrain attributes estimated Correlation coefficients between the two sets of ter-
using the soil surface topography and the topography rain attributes themselves as well as between these
 ARTICLE IN PRESS
8 V. Chaplot et al. / Geoderma xx (2004) xxx–xxx

two sets and the hydromorphic index, HI are pre- Over the hillslope, slopes are gentle with a
sented. Finally, the relation between HI and terrain mean downslope gradient, DG, of 4.3%, values
attributes estimated using the surface and the sub- ranging from 0% to 22%. Lower DG is shown
surface topography was investigated by using multiple in the valley bottom and at the hillslope summit.
regression analysis for the 141 data points of the Greater DG characterised the stream banks and the
model generation set. The multiple regression method downslope positions near the spring and down-
was used because it is a practical tool furnishing direct stream the study area. At the study area, DG
quantitative results by using easily accessible varia- reached 6.2% at the middleslope position, whereas
bles such as terrain attributes. Multiple regressions the valley bottom and the summit exhibited values
may be further used to directly predict the spatial lower than 4%.
distribution of soils over large areas. The procedure Such an apparent discordance between the surface
used was a stepwise linear regression (Neter et al., and sub-surface topographies also occurred for Asm
1989), which allowed independent variables to be and CTIm as noticed in Fig. 2. For instance, Asm
individually added or deleted from the model at each computed from the sub-surface DEM exhibited a
step of the regression, and therefore to evaluate much-branched system of micro-channels within the
changes in the R-squared value. Only parameters with study area. Furthermore, these channels are more
statistical significance at the 0.01 level were consid- closely connected with those of the northern part of
ered for computing predictive equations and reporting the hillslope. Higher values of CTIm occurred along
results. the valley bottom and at the central depression of the
Statistical parameters for models validation were study area. Mean value at the study hillslope reached
computed using the set of 41 data points: the mean 4.5 m.% 1 on the sub-surface in comparison with 3.9
error ME and the mean absolute error MAE between m.% 1 on the surface.
estimated HI (HI)* and observed value. The matrices of linear correlation between terrain
For both models, the spatial variations of MAE attributes computed at the 182 data points by using
calculated from surface and sub-surface topography the two DEMs are presented in Table 1. For each
using spline interpolation are presented over the attribute, the matrix diagonal shows the correlation
validation area. coefficient between estimations from both DEMs.
Correlation coefficients were very low for DG
(r = 0.17), Kp (r = 0.01) and Kc (r = 0.22). Thus,
3. Results for these attributes, estimations using soil surface
and sub-surface topographies differed greatly. On
3.1. Surface and sub-surface topography at the study the other hand, correlation coefficients were high
hillslope for Z (r = 0.99), ES (r = 0.98), S (r = 0.99), As
(r = 0.80), Asm (r = 0.98) and SPI (r = 0.83) revealing
The study hillslope is located in the western site of few differences between DEMs. Intermediary corre-
the first order reach oriented North – East/South – lation coefficients characterize CTIm (r = 0.64) and
West. Within the hillslope, soil surface topography Kt (r = 0.74).
is relatively gentle (altitudes varied from 191 to 204 In addition, the upper and lower parts of the
m). A small hill with altitudes of around 220 m is matrix of linear correlation coefficients present the
observed in the northern part. In the study area more level of auto-correlation between terrain attributes
closely surveyed, altitudes vary from 193.1 at the estimated from surface and sub-surface topography,
stream bank to 200.2 m at the hillslope summit. A respectively. For both DEMs, correlation coeffi-
depression oriented downslope – upslope could be cients higher than 0.97 were observed between L
observed in the upper part of the hillslope. The and, Z or ES. This is explained by a natural
topography of the saprolite upper boundary showed tendency along hillslopes of an increase in eleva-
similar trends with however a deeper central depres- tion with the increase in distance to the stream
sion stretching out from the stream bank to the hill- bank. Naturally, significant coefficients were also
slope summit (Fig. 2). observed between the physical index CTIm on the
 ARTICLE IN PRESS
V. Chaplot et al. / Geoderma xx (2004) xxx–xxx 9

Table 1
Correlation matrices for (i) soil and topographic attributes estimated using the 10-m DEM of upper boundary of saprolite (values in italics) and
(ii) topographic factors estimated using the 10-m DEM of the soil surface
L SD Z ES DG CTIm S As Asm SPI Kp Kc Kt
L 1:00 0.30 0.99 0.99 0.05  0.54  0.08  0.28  0.64  0.13 0.27 0.14  0.19
SD 0.30 1:00 0.28 0.31 0.12  0.27  0.14  0.20  0.26  0.18 0.20 0.16  0.18
Z 0.98 0.13 0:99 0.99 0.08  0.56  0.09  0.28  0.65  0.14 0.28 0.14  0.21
ES 0.97 0.15 0.98 0:98 0.14  0.61  0.09  0.30  0.67  0.14 0.29 0.16  0.21
DG 0.73 0.06 0.75 0.81 0:17  0.62  0.12 0.06  0.08  0.12 0.22 0.20  0.13
CTIm  0.62  0.22  0.61  0.63  0.88 0:64  0.20 0.39 0.80 0.01  0.37  0.28 0.20
S  0.06  0.13  0.05  0.04 0.22  0.31 0:99  0.26  0.37 0.76  0.01  0.06 0.08
As  0.30  0.32  0.26  0.28  0.36 0.49  0.13 0:80 0.56 0.30  0.42  0.42 0.29
Asm  0.64  0.27  0.62  0.65  0.73 0.82  0.38 0.55 0:98  0.12  0.31  0.20 0.14
SPI  0.16  0.29  0.14  0.15 0.05  0.09 0.74 0.44  0.08 0:83  0.32  0.36 0.28
Kp  0.01  0.35 0.08 0.07 0.18  0.07 0.02 0.01 0.03 0.09 0:01 0.80  0.83
Kc 0.18  0.14 0.24 0.23 0.27  0.16 0.05  0.22  0.14  0.10 0.75 0:22  0.49
Kt 0.15  0.05 0.19 0.20 0.18  0.1  0.04  0.45  0.13  0.37 0.56 0.83 0:74
The following attributes are considered: upslope length from the stream bank (L); soil depth (SD); altitude (Z); elevation above the stream bank
(ES); the ‘‘downslope gradient’’ (DG), which represents the ratio between E and the distance to the stream bank; the revised compound
topographic index (CTIm); the slope gradient (S), the specific monodirectional catchment area (As); the specific multidirectional catchment area
(Asm); the stream power index (SPI); profile curvature (Kp), contour curvature (Kc) and, tangential curvature (Kt). Data set of 149 points.
Correlation coefficients between topographic atttributes estimated using surface and saprolite upper boundary topographies are shown in the
diagonal (underlined values).

one hand and, DG and Asm on the other hand. For Correlation coefficients between the hydromor-
both topographies considered, S and the three phic index, HI and the terrain attributes extracted
curvatures were only slightly correlated with the from surface or sub-surface DEMs are presented
other terrain attributes. Finally, the depth to the in Table 2. Correlation coefficients ranged be-
saprolite upper boundary (SD) significantly corre- tween r = 0.08 for Kt on the surface or Kp on
lated with the terrain attributes L, ES, CTIm, As, the sub-surface to r = 0.80 for Z and DG on the
Asm, SPI, Kp but r values were lower than 0.35. surface and the sub-surface, respectively. On the
surface, greater correlation coefficients were ob-
3.2. Predicting the hydromorphic index HI using served for L (r = 0.77), Z (r = 0.80), ES (r = 0.79),
terrain attributes CTIm (r = 0.62) and, Asm (r = 0.60). On the sub-
surface, greater coefficients similarly occurred for
The spatial distribution of HI within the study L (r = 0.80), DG (r = 0.80), CTIm (r = 0.79) and
hillslope is presented in Fig. 3A. The soils not affected Asm (r = 0.82). The correlation coefficients with L,
by an excess of water (HI = 0) occurred upslope, S, As, SPI, and the three curvatures did not vary
whereas HI values greater than 25 were observed in between the two topographies considered.
the vicinity of the stream. A gradation downslope – Two multiple regression models for HI spatial
upslope was observed with HI contour lines relatively prediction were generated using non-autocorre-
parallel to the stream bank. lated terrain attributes estimated from surface

Table 2
Correlation coefficients r between HI and the terrain attributes estimated from (i) the 10-m DEM of soil surface, and (ii) the 10-m DEM of the
upper boundary of saprolites
L SD Z ES DG CTIm S As Asm SPI Kp Kc Kt
10 m DEM of soil surface  0.77*  0.37*  0.80*  0.79*  0.55* 0.62*  0.10 0.31* 0.60* 0.05  0.29  0.22 0.08
10 m DEM of upper  0.78*  0.38*  0.73*  0.77*  0.80* 0.79*  0.06 0.39* 0.82* 0.14  0.08  0.23  0.22
boundary of saprolites
* Significant at the 0.05 probability level.
 ARTICLE IN PRESS
10 V. Chaplot et al. / Geoderma xx (2004) xxx–xxx

or sub-surface topography (Eqs. (2) and (3), at the hillslope summit and close to the stream bank
respectively). remained unchanged.

ðHIÞsurface ¼ 3:1
LnðAsm =DGÞ  3:4
ES

4. Discussion
ðr2 ¼ 0:80Þ ð2Þ
Improving the estimation of the spatial distribution
ðHIÞsaprolite ¼ 5:7
LnðAsm =DGÞ  3:5
ES of soils within landscapes is one of the most important
ðr2 ¼ 0:87Þ ð3Þ issues in pedometrics. For the most part, predictive
models of the soil spatial distribution have focused on
Both equations included the terrain attributes of the relationship between the soil properties and topog-
elevation difference above the stream bank (ES) and raphy because of the known role relief plays (e.g.
compound topographic index (CTI), a function of the Moore et al., 1993; Skidmore et al., 1991; De Bruin
multidirectional drainage area (Asm) and downslope and Stein, 1998; Chaplot et al., 2000; Thompson et al.,
gradient (DG). Mean error (ME), and mean absolute 2001). Several ways were proposed to increase models
error (MAE), between estimations and observations accuracy including increasing DEM accuracy (Quinn
for the 41 data points of the validation set are presented et al., 1991; Zhang and Montgomery, 1994; Farajall
in Table 3. ME and MAE were 1.5 and 5.3 for the and Vieux, 1995; Thompson et al., 2001); adding
model considering the surface topography and 0.5 and punctual soil observations (e.g. Odeh et al., 1995;
3.9 for that using the topography of the saprolite upper Bourennane et al., 1996; Chaplot et al., 2000) or
boundary. Using regression (5)diminished ME and geologic and land use information (Lagacherie et al.,
MAE by 67% and 26%, respectively. In addition, the 1995, 2001; McKenzie and Ryan, 1999). In this study
mean rank of the model using the sub-surface topog- we examined how the topography of the saprolite upper
raphy at the 41 validation points was 1.19 against 1.76 boundary could enhance the spatial prediction of the
when using the topography of the soil surface. This soil hydromorphic index, HI. This was addressed in an
revealed better prediction accuracies when using the entire hillslope under loamy soils where two DEMs
topography of the saprolite upper boundary. were generated from surface or sub-surface features.
For both regression models, an error analysis was The present study revealed that HI was the most
performed by displaying over the validation area the correlated with the terrain attributes estimated using
spatial variability of the MAE of HI (Fig. 3B,C). For the sub-surface DEM. Indeed, using the topography of
both models, high root mean square errors were the saprolite upper boundary through soil-landscape
encountered in the vicinity of the stream where greater modelling greatly improved HI spatial prediction.
HI values occur. In addition, using the sub-surface One reasonable explanation for this is that the
topography in the modelling process decreased pre- lateral water fluxes in soils along slopes induced by
diction errors at middleslope position, whereas errors the very low permeability of the saprolite ( < 0.005 m
day 1, Grimaldi and Chaplot, 2000) are responsible
Table 3 for the increasing occurrence of soil saturation in the
Mean and standard deviation (Sd) of HI for the whole data and downslope direction (Chaplot and Walter, 2003),
validation sets which in turn was shown here to directly affect the
n Mean Sd F P level ME MAE Mean formation of redoximorphic features in soils. At the
rank study hillslope, lateral fluxes affecting both soil satu-
Observed HI 182 17.3 16.9 ration and its consequences on soil organization, i.e.
(HI)surface 41 15.6 14.1 419 * 1.5 5.3 1.76 distribution along slopes, are mainly driven by the
(HI)saprolite 41 14.6 14.0 584 * 0.5 3.9 1.19
topography of the saprolite upper boundary.
Prediction errors of HI for soil-landscape models using the surface
Terrain attributes such as the compound topograph-
topography and the topography of the saprolite upper boundary:
correlation coefficient, r; mean error (ME); mean absolute error ic index, CTI and elevation above the stream bank, ES
(MAE) and mean rank between estimations and observations. were shown to better explain the processes leading to
* Significant at the 0.001 probability levels. soil hydromorphy. This is probably due to the already
 ARTICLE IN PRESS
V. Chaplot et al. / Geoderma xx (2004) xxx–xxx 11

demonstrated fact they have a physical basis (Beven ences between predicted and observed values may be
and Kirkby, 1979; Moore et al., 1993; Crave and caused by measurement error in HI itself. Indeed,
Gascuel-Odoux, 1996; Chaplot and Walter, 2003). especially in the case of soils greatly affected by an
Modelling soil distribution using the topography of excess of water (HI>50), slight differences in the
a sub-surface feature may be extended to other areas attribution of Munsell value or chroma may result in
where water pathways affect soil properties and where high differences in HI. This issue would have to be
a non-parallelism between surface and subsurface further investigated. Another explanation for the pres-
topographies occurs. This is likely to be the case for ence of remaining prediction errors may be due to
numerous natural systems. However, generating a continuing lack of models to fully integrate, through
sub-surface DEM would be very costly, especially environmental factors, processes leading to the soil
when using auger hole for the characterisation of the hydromorphy. This statement was partly confirmed
sub-surface topography. Despite DEM of surface by the presence of systematic over-estimations of HI.
topography being now widely available, numerical It shows that predictive models may still be improved
representations of the topography of sub-surface dis- by a better understanding of how soil properties and
continuities are not easily accessible over large areas. their spatial distribution depend on external environ-
No convenient databases are yet available for repre- mental factors.
senting the continuously varying topography of sub-
surface discontinuities of the Earth. At larger scales
References
(i.e. on areas from 1 to hundreds of hectares), such
information may be more easily accessed using for
Beven, K.J., Kirkby, M.J., 1979. A physically based variable con-
instance geophysical techniques. Geophysical meth- tributing area model of basin hydrologie. Hydrological Sciences
ods such as GPR may be relevant for the estimation of Bulletin 24, 43 – 69.
the spatial variations of sub-surface features including Bourennane, H., King, D., Chery, P., Bruand, A., 1996. Improving
the depth to saprolite upper boundary (e.g. Dominic et the kriging of a soil variable using slope gradient as external
al., 1995; Lapen et al., 1996). At the same study site drift. European Journal of Soil Science 47, 473 – 483.
Bourennane, H., King, D., Couturier, A., 2000. Comparison of
Chaplot et al. (2001b) showed the interesting potential kriging with external drift and simple linear regression for pre-
of the Radio Magneto Tellurique (RMT) in predicting dicting soil horizon thickness with different sample densities.
the spatial variations of the depth to the saprolite Geoderma 97, 255 – 271.
upper boundary. However, these techniques up to Burt, T.P., Butcher, D.P., 1986. Development of topographic indices
for use in semi-distributed hillslope runoff models. In: Slay-
now may not only lack relevancy but also need a
maker, O., Balteanu, D. (Eds.), Geomorphology and Land Man-
great amount of fieldwork and data processing. For agement. Gebruder Borntraeger, Berlin, pp. 1 – 19.
these reasons, geophysical methods cannot be directly Chaplot, V., Le Bissonnais, Y., 2003. Runoff features for interrill
applied over large areas for the prediction of the sub- erosion at different rainfall intensities, slope lengths and gra-
surface topography. On the other hand, a promising dients in an agricultural loessial hillslope. Soil Science Society
alternative is to use quantitative soil-landscape mod- of American Journal 67 (3), 844 – 851.
Chaplot, V., Walter, C., 2003. Sub-surface topography to enhance
elling designed for the prediction of the depth to time and space distribution of soil water content. Hydrological
saprolite upper boundary (SD) itself. Such models Processes 17 (13), 2567 – 2580.
relating SD to more accessible soil-forming factors Chaplot, V., Walter, C., Curmi, P., 2000. Improving soil hydromor-
such as surface topography or geology are already phy prediction according to DEM resolution and available ped-
available at other sites (e.g. Gessler et al., 1995; ological data. Geoderma 97, 405 – 422.
Chaplot, V., Bernoux, M., Walter, C., Curmi, P., Herpin, U., 2001a.
Bourennane et al., 2000). Such a question of the Soil carbon storage prediction in temperate hydromorphic soils
practical application of this technique through the by using morphologic index and Digital Elevation Model. Soil
acquisition of subsurface topography in landscapes Science 166 (1), 48 – 60.
on a routine basis should be further addressed. Chaplot, V., Walter, C., Curmi, P., Hollier-Larousse, A., 2001b.
Finally, even though SLMs benefit from the better Mapping field-scale hydromorphic horizons using Radio-MT
electrical resistivity. Geoderma 102, 61 – 74.
knowledge of the controlling factors of the spatial Crave, A., Gascuel-Odoux, C., 1996. The influence of topography
distribution of soil properties, differences between on time and space distribution of soil surface water content.
observed and predicted HI remain. Some of the differ- Hydrological Processes 11, 203 – 210.
 ARTICLE IN PRESS
12 V. Chaplot et al. / Geoderma xx (2004) xxx–xxx

Curmi, P., Durand, P., Gascuel-Odoux, C., Mérot, P., Walter, C., attribute prediction using terrain analysis. Soil Science Society
Taha, A., 1998. Hydromorphic soils, hydrology and water qual- of America Journal 57, 443 – 452.
ity: spatial distribution and functional modeling at different Neter, J., Wasserman, W., Kutner, M.H., 1989. Applied Linear
scales. Nutrient Cycling in Agroecosystems 50, 127 – 142. Regression Models, 2nd ed. Richard D. Irwin, Homewood.
De Bruin, S., Stein, A., 1998. Soil-landscape modeling using fuzzy Odeh, I.O.A., McBratney, A.B., Chittleborough, D.J., 1995. Further
c-means clustering of attribute data derived from a Digital Ele- results on prediction of soil properties from terrain attributes—
vation Model DEM. Geoderma 83, 17 – 33. heterotropy cokriging and regression-kriging. Geoderma 67,
Dominic, D.F., Egan, K., Carney, C., Wolfe, P.J., Boardman, M.R., 215 – 226.
1995. Delineation of shallow stratigraphy using ground penetrat- O’Loughlin, E.M., 1986. Prediction of surface saturation zones in
ing radar. Journal of Applied Geophysics 33, 167 – 175. natural catchments by topographic analysis. Water Resource
ESRI, 1994. Understanding, G.I.S. The ARC/INFO Method, ESRI, Research 22, 794 – 804.
380 New York Steet, Redlands, CA USA 92273. 300 pp. Poesen, J., 1984. The influence of slope angle on infiltration rate
Farajall, N.S., Vieux, B.E., 1995. Capturing the essential spatial and Hortonian overland flow volume. Zeitschrift Geomorphol-
variability in distributed hydrological modeling infiltration ogy 49, 117 – 131.
parameters. Hydrological Processes 9, 55 – 68. Quinn, P., Beven, K., Chevallier, P., Planchon, O., 1991. The pre-
Fox, D.M., Bryan, R.B., 1999. The relationship of soil loss by diction of hillslope flow paths for distributed hydrological mod-
interrill erosion to slope gradient. Catena 38, 211 – 222. elling using digital terrain models. Hydrological Processes 5,
Gessler, P.E., Moore, I.D., McKenzie, N.J., Ryan, P.J., 1995. Soil- 59 – 80.
landscape modelling and the spatial prediction of soil attributes. Skidmore, A.K., Ryan, P.J., Dawes, W., Short, D., O’Loughlin,
International Journal of Geographical Information Systems 9, E.M., 1991. Use of an expert system to map forest soil from
421 – 432. a geographical information system. Int. J. Geographical Infor-
Govers, G., 1990. A field study on topographical and topsoil effects mation Systems 5, 431 – 445.
on runoff generation. Catena 18, 91 – 111. Soil Survey Staff, 1999. Soil Taxonomy. A Basic System of Soil
Grimaldi, C., Chaplot, V., 2000. Nitrate removal in headstreams. Classification for Making and Interpreting Soil Surveys, 2nd ed.
Influence of exchanges between stream and near-bank environ- USDA, Washington, DC.
ment. Water Air and Soil Pollution 124, 95 – 112. Squividant, H., 1994. MNTsurf: Logiciel de Traitement des Mod-
Haycock, N.E., Pinay, G., 1993. Groundwater nitrate dynamics in eles Numeriques de Terrain. Doc. ENSAR, Rennes. 36 pp.
grass and poplar vegetated riparian buffer strips during the win- Thompson, J.A., Bell, J.C., 1996. Color index for identifying hydric
ter. Journal of Environmental Quality 22, 273 – 278. conditions for seasonally saturated Mollisols in Minnesota. Soil
Hutchinson, M.F., 1989. A new procedure for gridding elevation Science Society of American Journal 60, 1979 – 1988.
and stream line data with automatic removal of spurious pits. Thompson, J., Bell, J., Butler, C., 2001. Digital elevation model
Journal of Hydrology 106, 211 – 232. resolution: effects on terrain attribute calculation and quantita-
Jenny, H., 1941. Factors of Soil Formation: A System of Quantita- tive soil-landscape modeling. Geoderma 100, 67 – 89.
tive Pedology. McGraw Hill, New York. Torri, D., Poesen, J., 1992. The effect of soil surface slope on
Lagacherie, P., Legros, J.P., Burrough, P.A., 1995. A soil survey raindrop detachment. In erosion, transport and depositional pro-
procedure using the knowledge of soil pattern established on a cesses. Catena 19, 561 – 577.
previously mapped reference area. Geoderma 65, 283 – 301. Van Vliet-Lanoe, B., Laurent, M., Hallegouet, B., Margerel, J.P.,
Lagacherie, P., Robbez-Masson, J.M., Nguyen-The, N., Barthes, Chauvel, J.J., Michel, Y., Moguedet, G., Trautman, F., Vauthier,
J.P., 2001. Mapping of reference area representativity using a S., 1998. The mio-pliocene of the armorican massive. New data.
mathematical soilscape distance. Geoderma 101, 105 – 118. C. R. Acad. Sci. 326, 333 – 340.
Lapen, D.R., Moorman, B.J., Price, J.S., 1996. Using ground pen- Vepraskas, M.J., Wilding, L.P., 1983. Aquic moisture regimes in
etrating radar to delineate subsurface features along a wetland soils with and without low chroma colors. Soil Science Society
catena. Soil Science Society of America Journal 60, 923 – 931. of America Journal 47, 280 – 285.
Le Calvez, L., 1979. Genèse des formations limoneuses de Bre- WRB, 1998. World Reference Base for Soil Resources. Internatio-
tagne centrale. Essai de modélisation. MS thesis. Univ. of Ren- nal Society of Soil Science, International Soil Reference and
nes, Rennes. Information Centre, Spaargaren, Rome.
McKenzie, N.J., Ryan, P.J., 1999. Spatial prediction of soil proper- Wyns, R., 1991. L’utilisation des paléosurfaces continentales en
ties using environmental correlation. Geoderma 89, 67 – 94. cartographie thématique probabiliste. Géologie de la France 3,
Mérot, P., Ezzahar, B., Walter, C., Aurousseau, P., 1995. Mapping 3 – 9.
waterlogging of soils using digital terrain models. Hydrological Zevenbergen, L.W., Thorne, C.R., 1987. Quantitative analysis of
Processes 9, 27 – 34. land surface topography. Earth Surface Processes and Land-
Moore, I.D., Grayson, R.B., Ladson, A.R., 1991. Digital terrain forms 12, 47 – 56.
modeling: A review of hydrological, geomorphological, and Zhang, W., Montgomery, D.R., 1994. Digital elevation model grid
biological applications. Hydrol. Processes 5, 3 – 30. size, landscape representation, and hydrological simulations.
Moore, I.D., Gessler, P.E., Nielsen, G.A., Peterson, G.A., 1993. Soil Water Resources Research 30, 1019 – 1028.