You are on page 1of 10

Catena 97 (2012) 8594

Contents lists available at SciVerse ScienceDirect

Catena
journal homepage: www.elsevier.com/locate/catena

A robust algorithm for estimating soil erodibility in different climates


L. Borselli a,, D. Torri b, J. Poesen c, P. Iaquinta d
a
Instituto de Geologia/Fac. De Ingenera, Universitad Autonoma de San Luis Potos (UASLP), Av. Dr. Manuel Nava 5, C.P. 78240, San Luis Potos, SLP, Mexico
b
CNR-IRPI, Research Institute for Geo-Hydrological Protection, Via Madonna Alta 126, 06128 Perugia, Italy
c
Division of Geography, KU Leuven, Celestijnenlaan 200E, B-3001 Heverlee, Belgium
d
CNR-IRPI, Research Institute for Geo-Hydrological Protection, Via Cavour 46, 87030 Rende di Cosenza, Italy

a r t i c l e i n f o a b s t r a c t

Article history: The analysis of global soil erodibility data by Salvador Sanchis et al. (2008) showed that there is a signicant
Received 3 November 2011 climate effect on soil erodibility which allows for a split of the data into two subsets, one for prevailing cool
Received in revised form 20 May 2012 conditions and another for prevailing warm conditions (dened using the Kppen climate classication). De-
Accepted 30 May 2012
spite the recognition of this new dichotomous variable, prediction of soil erodibility values remained very
Available online xxxx
poor. This paper presents a new technique for dealing with such a variability by calculating probability den-
Keywords:
sity functions of soil erodibility K values when the user knows a set of textural parameters and the climatic
Soil erosion classication of the site. Finally the user has the possibility to decide, on the basis of local knowledge,
Soil erodibility which K value to use. The procedure has been implemented in a freeware software named KUERY available
Quantile regression for the scientic community. Finally, as an illustration, the methodology is applied to a catchment in south
Probability density function Italy.
KUERY software Crown Copyright 2012 Published by Elsevier B.V. All rights reserved.

1. Introduction up-down direction, hence bare with no cumulative erosion forms


(Wischmeier and Smith, 1978). Data are suggested to be collected
Soil erodibility conceptually represents the way in which soil re- over a 20year period. Obviously, rill erosion, which usually causes in-
acts to erosional agents, hence it is limited to a series of processes tense soil losses, cannot be overwhelmingly important, otherwise soil
(e.g. detachment and transport of soil particles, rain inltration, characteristics, such as texture or other basic characteristics of the
breakdown of soil aggregates, etc.) that occur while soil erosion surface soil horizon, will not be invariant during the observation peri-
takes place and the complex interaction between the erosive agent od (as top soil characteristics may change after intense or prolonged
and soil determines how much and how soil is eroded. Hence soil soil loss).
erodibility depends on the processes that cause erosion. From a prac- Operatively, the USLE denes soil erodibility (K) as the ratio be-
tical point of view, erodibility depends on the way in which we select tween soil loss and the causative rain erosivity (R), when the other
the processes to include in our description (i.e. model), how we de- erosion factors (L: slope length; S: slope gradient; C: crop and man-
ne the mathematics of the chosen processes, how rough is our agement; P: erosion control practice) are unitary or, more generally,
parameterisation and how difcult it is to measure all the relevant as the following ratio:
properties of the soil and of the erosive agents. When we restrict our-
selves to the Universal Soil Loss Equation (Wischmeier and Smith, A
K 1
1978), and its subsequent modications (Foster et al., 2001; Renard RLS
et al., 1997), soil erodibility gets a particular meaning. It represents
the integrated long-term average soil response to rainstorms. Soil This denition implies that all the errors on measurements and all
erodibility represents in a lumped manner the global response of a the inaccuracy in mathematical representations of the factors at the
soil to a large number of hydrological and erosion processes of denominator are errors on K, which acts as a black-box coefcient.
which runoff generation, sheet and rill erosion are the most impor- When the K-data are interpolated in order to generate an alterna-
tant. The soil prole in its entirety determines the soil response to tive to the costly and time-consuming way of measuring K in the eld,
rain erosive power. All this is represented through one single number data are scattered and many aspects remain unanswered. One of the
(i.e. K, the soil erodibility factor ) that is calculated from soil loss mea- best estimates of the soil erodibility factor still remains the
surement in the eld over a standard plot, kept freshly disked in an Wischmeier et al. (1971) formula (or nomograph) which is largely
based on eld-data for medium-textured soils in the USA.
Many additional data that allow an estimation of soil erodibility do
Corresponding author. Tel.: + 524448171039. exist in the literature (e.g. Poesen, 1993; Poesen et al., 1994; Torri et
E-mail address: lborselli@gmail.com (L. Borselli). al., 1997). As data have been measured in conditions often far from

0341-8162/$ see front matter. Crown Copyright 2012 Published by Elsevier B.V. All rights reserved.
doi:10.1016/j.catena.2012.05.012
86 L. Borselli et al. / Catena 97 (2012) 8594

the standard conditions, Torri et al. (1997) and Salvador Sanchis et al. b) the second textural parameter (Sg) is the logarithm of geometric stan-
(2008) have used only part of the data, i.e. those collected in condi- dard deviation of Dg, already dened by Shirazi and Boersma (1984)
tions relatively close to the standard procedure (Torri et al., 1997). and Shirazi et al. (1988); Sg is calculated with the following equation
Despite the reduction, they come from various places in the world s
 q 2
and represent soil response under widely different climatic condi-
Sg f i log10 di di1 Dg 4
tions (Salvador Sanchis et al., 2008). i
The recent re-elaboration of part of the Torri et al. (1997) soil
erodibility global dataset (EGD) by Salvador Sanchis et al. (2008) Borselli et al. (2009) also proposed a simplied equation for Sg if for a
showed that there is a signicant climate effect on soil erodibility given soil only the three main textural components C, L and S are
which allows for a split of the data into two subsets, one with domi- known:
nant cool conditions and another with dominant warm conditions
s
(dened on the basis of Kppen climate classication). Despite the
C 3:5Dg2 L2:0Dg2 S0:5Dg2
recognition of this new dichotomous variable, prediction of soil erod- Sg 5
100
ibility values remains extremely poor. This study aims at developing a
technique for dealing with such a variability, calculating it through
Obviously the simplied Eqs. (3) and (5) make calculations easier but
the frequency distributions of the soil erodibility values observed
are rougher approximations than Eqs. (2) and (4) which were used
for given textural, organic matter and climate data. The user of the al-
for developing the algorithm;
gorithm will nally have the possibility to decide, on the basis of local
c) the percentage of soil organic matter (SOM);
knowledge, which values to use. Finally application of the methodol-
d) the percentage of rock fragment content (Rk).
ogy is illustrated using data from the Esaro catchment (Calabria,
Italy). A full freeware software, implementing the proposed algo- Torri et al. (1997) proposed a technique based on fuzzy sets and
rithms and methodology, is available for users and the scientic numbers, which was implemented into software, to derive the distri-
community. bution of observed K-values. The software operates on a subset of the
whole EGD as stony soils were not included (i.e. soil with rock frag-
2. Algorithm denition ment content Rk >10%). The fuzzy technique used at that time cannot
be considered sufciently general because it was based on a prelimi-
The poor performance of classical statistical techniques, which nary and subjective subdivision of the K-dataset in groups associated
have been applied in previous studies (e.g. Rmkens et al., 1997; to xed ranges of the various input parameters. This limitation can be
Salvador Sanchis et al., 2008; Torri et al., 1997, 2002) for analysing overcome using a continuous interpolation of given quantiles on the
the global soil erodibility dataset (EGD: Poesen et al., 1994; Torri basis of a series of soil parameters.
et al., 1997, 2002) lead to explore alternative ways in order to infer As a preliminary step, the quantile regression techniques (Buchinsky,
the range of uncertainty of K-values associated to every combination 1998; Koenker and Basset, 1978; Lee and Tanaka, 1999; Yu, et al., 2003)
of climate and of input soil parameters. were coded into a specic Object PASCAL language program to obtain
In our Algorithm the soil parameters considered are the following. quantile interpolating algorithms of soil erodibility as a function of vari-
ous parameters (e.g. K vs. Dg; K vs. Sg, K vs. SOM, K vs. Rk).
a) the logarithm of the geometric mean of the particle-size distribu-
The procedure for quantile regression can be dened more rigorously
tion (Dg), used also by Torri et al. (1997, 2002), which can be
as follows. Given a pair of random variables (X,Y), the quantile regression
directly calculated according to Shirazi et al. (1988):
of Y on X can be dened in terms of conditional distribution F(Y | x) of Y
q for any given X=x, for any probability value p[0, 1]. The quantile
Dg f i log10 di di:1 2 regression qp(x) becomes the solution of the conditional distribution
i
equation (Yu, 2002):
 
where fi is the mass fraction of particles in the class with range of F qp xjx p 6
diameters di and di-1 (in mm). If only three main textural compo-
nents of the studied soil, i.e. sand (S), loam (L) and clay (C), are
available, the argument of square root in Eq. (2) can be calculated In other terms, for any given X-value, it is possible to associate a
as shown in Table 1 or using the simplied formula (Borselli et al., quantile in the range 0 b p b 1.0 to every Y-value. This denition has
2009), if only three main textural components of soil: sand (S), been successfully applied in many elds for the last 25 years (Yu et
loam (L) and clay (C) are available. Dg can also be calculated by al., 2003). In our study the two climatic groups of K correspond to
a simplied formula (Borselli et al., 2009): the two different Y-sets while the X set will represent any of the var-
ious soil parameters, e.g. (K|Dg). This allows to dene a Cumulative
3:5C2:0L0:5S
Dg 3 Distribution Function (CDF) of soil erodibility for every value of
100
each soil parameter (e.g. the observed erodibility CDF at Dg = 2.5).
where C, L and S are respectively the percentage of three main tex-
tural classes Clay, Loam and Sand; 2.1. Algorithm description and implementation

According to Koenker and Basset (1978) the optimal quantile re-


gression function qp (x) may be obtained by minimizing the following
Table 1
generalised objective function f(obj):
Constants used for the calculation of parameter Dg, in the case of three basic textural 2 3
components.



p

f obj min 4
t
p yi f i ; xi 1p yi f i ; xi 5
t
7
ifi:yi f i t ;xi g ifi:yi bf i t ;xi g
Textural component di (mm) di 1 (mm) didi 1 (mm2) log10 di di1

CLAY 0.002 0.00005a 0.0000001 3.5 where:


LOAM 0.05 0.002 0.0001 2
SAND 2 0.053 0.1 0.5 i is the index that identies each element of the data set;
a
Conventionally, the lower limit of clay particles is set equal to 0.00005 (mm) xi,yi are the coordinates in X and Y of the ith element of the
(Shirazi et al., 1988). data set
L. Borselli et al. / Catena 97 (2012) 8594 87

p is the chosen regression quantile (e.g.: 0.1, 0.5, 0.75); order to remove possible outliers: all the data-points, plotting
t is a vector that contains the values of the coefcients den- outside the two quartiles and whose distance from the closer
ing the quantile function to be optimized (i.e. coefcients to quartile exceeds twice the local interquartile range, are classi-
be found during the optimization process); ed as outliers and removed.
fi( t, xi) the regression quantile function as dened by t and xi Step 2. The quantile regression, after outlier's removal, is performed
position. for the following quantiles: 0.01, 0.05, 0.1, 0.2, 0.25, 0.3, 0.4,
0.5, 0.6, 0.7, 0.75, 0.8, 0.9, 0.95, 0.99 (note that quantiles
In the case of the EGD, the variance and distribution of observed K- 0.25 and 0.75 are calculated anew).
values varies irregularly and independently from the soil parameter
for which the K-behaviour is examined (see Fig. 1a, b, c, d). Hence lin- The non-linear optimisation process for each single quantile is
ear or monotonic quantile functions (Koenker and Basset, 1978) can- obtained using a differential genetic evolution algorithm (Storn and
not be used because they are not general enough. Instead a non-linear Price, 1997) and adopting all the additional set-up strategies to en-
and non-monotonic function is needed for the quantile regression sure a global optimisation result (Price, 1999; Storn and Price, 1997).
(e.g. kernel function, polynomial or spline; Yu et al., 2003).
In our case we adopted the four-point sliding polynomial tech-
2.2. Soil erodibility subsets and results
nique (Snyder et al., 1991) to build up a non-linear function with
the required properties. The sliding polynomials are specic type of
The global K-data base (EGD) was subdivided in four subsets,
spline functions that preserve the properties of continuity and non-
according to climate (two groups) and to rock fragment content
monotonicity in the explored range of the dataset.
(Rk: two groups: less than and more than 10% rock fragment content
This technique requires the subdivision of the data range of the in-
by mass). The optimisation algorithm was launched for all soil param-
dependent variable X in n intervals (spans) of equal length (n = 7 in
eters. Some of these parameters did not produce any acceptable or
our case). The positioning of the spline's knots is given from a set of
useful result and were consequently removed. The nal explanatory
(xj, yj) coordinates, where j = 1, , n.
variables were slightly different in each group as described below, in
Using sliding polynomial splines the xj coordinate of the spans are
agreement with Salvador Sanchis et al.'s (2008) ndings:
constant values, but the yj values must be obtained by an optimisation
algorithm. In this case the yj coincide with the t shown in Eq. (7).
Group 1) cool climatene soil: Dg, Sg, SOM;
The non-linear optimisation process follows two steps:
Group 2) cool climatestony soil: Dg, Rk > 10%;
Step 1. Quantile functions are optimised and calculated for the 0.25 Group 3) warm climatene soil: Dg, Sg, SOM;
and 0.75 quantiles. A ltering criterion is then established in Group 4) warm climatestony soil: Dg, Rk > 10%.

Fig. 1. Examples of quantile interpolating functions. a, b) Soil erodibility (K) versus the logarithm of the geometric mean particle size (Dg) for cool and warm climates and rock content
b 10%, c, d) Soil erodibility (K) versus the standard deviation of the logarithmic transformation of particle-size distribution (Sg) for cool and warm climates and rock content b10%.
88 L. Borselli et al. / Catena 97 (2012) 8594

In the case of groups 2 and 4 (stony soils), the behaviour is strong- As an example for the Group 1 and 3, the nal CDF distribution
ly dominated by rock fragment content while the effects of SOM and F(K)i for MOT intersection operator that is the normalised product
Sg are not detectable and in these case only Dg and Rk are used. of independent F(K)I for each soil parameter considered (e.g. F(K|
Two examples of the nal results of the optimisation algorithm, Dg)i):
i.e. the quantile functions, are shown in Fig. 2(a, b) the inferred cumu-
h

lative distribution function (CDF) for two Dg-values (2.4 and 0.7) F K i F K jDg i F K jSg F K jSOMi F K i min
and two Sg -values (0.8 and 1.2) (both for cool climate). i
8
When the quantile functions are available, the conditional proba- F K i max F K i min
bility F(K|x) (e.g.: x = Dg or x = Sg) can be obtained for each indepen-
dent variable in the dataset and for each climate group. where:
All the quantile functions derived for the optimisation process on h

the EGD are stored in a special format using the positions of F K i max max F K jDg i ; F K jSg ; F K jSOMi  9
i
optimised knots that dene univocally the sliding polynomial
functions. h

To facilitate the query process and inference from the global F K i min min F K jDg i ; F K jSg ; F K jSOMi  10
i
dataset for an end user, a special software (KUERY: from query
and K) has been developed. This software allows the end user to cal- and the nal CDF distribution for MOT union operator is given by
culate the most probable soil erodibility CDF for the given climate normalising the summation of independent F(K)i for each soil param-
group and the given soil type (dened in terms of Dg, Sg, SOM and Rk). eter considered:
The data processing ow of KUERY follows these steps:
h

F K i F K jDg i F K jSg F K jSOMi 


1) For a given combination of climate group and Dg, Sgand SOM input i
11
values (if soil rock fragment content is Rk b 10%) or of Dg and rock 3
fragment content (if rock fragment content is Rk > 10%), KUERY
interrogates the stored quantile functions for each parameter The resulting CDF represents the most probable values for the soil
and produces an erodibility CDF for each input parameter as illus- erodibility factor corresponding to the given values of soil properties.
trated in Fig. 2(a, b); In the case of intersection, the CDF represents the irreducible disper-
2) The end user selects one of the two available Merging Operator sion of observed data, given the values of the input soil parameters. If,
Types (MOT) for producing one single nal CDF : a) intersection instead, union is chosen then the user obtains the total dispersion of
(the logical operator AND) b) union (the logical operator OR). observed data. In principle, intersection should give the correct

Fig. 3. Probability Density Functions (PDF) produced by KUERY for: a) cool and b) warm
climate groups, and rock content b 10%. Input data are identical for both cases. Note the
Fig. 2. Cumulative distribution function (CDF) of soil erodibility-values (K) corresponding differences in interpolated frequency distributions, despite the fact that input values are
to a) two different Dg values and b) two different Sg values (cool climate case). the same.
L. Borselli et al. / Catena 97 (2012) 8594 89

Fig. 4. Location of the Esaro catchment (115 km) in south Italy with soil map, D g, Sg, SOM. The codes (4.2; 4.4; 4.5; 4.7; 6.3) of the pedological units identify different soil
associations: 4.2 includes Chromic and Vertic Haploxererts; 4.4 is dominated by Entisols; 4.5 includes Typic Haploxerepts and Vertic Haploxerolls; 4.7 includes Typic
Haploxeralfs and Typic Fragixeralfs; 6.3 includes Typic Endoaquents, Vertic Haploxerepts and Vertic Calcixerepts (ARSSA-CAL, 2003).

guess but in case of non-existing intersection a warning is given and inltration rate suggest the selection of a low quantile K-value. Addi-
the use of union operator instead is suggested. tional dispersive behaviour of the soil (ESP > 5% ) may suggest a selec-
KUERY allows for extracting quantiles and relevant statistics as tion of K value corresponding to higher quantile (e.g. 0.9, Borselli et
mean and variance from the nal CDF. The nal probability density al., 2009).
function (PDF) is also calculated by KUERY using a numerical differ-
entiation of the nal CDF. 2.3. Application at watershed level: the Esaro catchment (Calabria, Italy)
Examples of the nal output produced by KUERY are given in
Fig. 3a, b. The Esaro catchment, located in the Calabria region (south Italy),
Knowledge of the distribution of the interpolated K-values at any lies between latitudes 385844 and 390526 and between longi-
given set of proper soil parameters represents an important help for tudes 165752 and 170838. This catchment is approximately
assessing soil erodibility. Actually, information on the distribution of 115 km 2 in size with an average elevation of about 105 m a.s.l. and
soil erodibility, coupled with eld observations or laboratory tests a maximum elevation of 238 m a.s.l.; the average gradient of the
(such as soil aggregate stability assessed using the drop method), slopes is 1.74%.
can help in selecting a soil erodibility value from the distribution The average annual air temperature is about 18 C, the warmest
obtained using KUERY. For example, high aggregate stability or high month reaches an average temperature of ca. 26.5 C in August and,
90 L. Borselli et al. / Catena 97 (2012) 8594

nally, the coldest month is January with an average temperature of the K factor (Kw, in SI units) calculated using the nomograph pro-
about 9.7 C. posed by Wischmeier et al. (1971), but limiting ourselves to the use
The total annual rainfall is about 650 mm, concentrated over 90% of soil textural data and organic matter content only.
in the wet season between October and March. Fig. 6a, b shows the differences k (in %), between the soil erod-
The rounded shape of the catchmentthe Gravelius factor is equal ibility calculated using KUERY for both climatic groups (warm
to 1.65and the radial pattern of drainage network promote the con- (Fig. 6a) and cool (Fig. 6b)) and the soil erodibility computed by the
centration of surface runoff. In fact, although the average runoff dis- classical nomograph method Kw, (Wischmeier et al., 1971). k is cal-
charge is approximately only 0.6 m 3 s 1, frequently occurring culated using the equation:
historical ooding events with peak discharges larger than 100 m3/s,
culminated in the tragic oods of October 14, 1996 (estimated peak KK w
k 100 12
discharge of about 775 m3 s 1). On this last occasion large volumes of Kw
sediment from the slopes were deposited in the town of Crotone.
Marly clay and silty clay Pliocenic marine sediments are the dom-
inant lithologies in this catchment (60% of the area). Land use is 3. Discussion
mainly characterized by cropland (ca. 68%) and by arboreal coverage
(ca. 16%). Soil type distribution in the catchment is shown in Fig. 4. Despite the fact that K-data have been subdivided on the basis of
Soil data come from the local soil service (ARSSA-CAL, 2003) and climate, advances in deterministic prediction of soil erodibility re-
numbers correspond to a code referring to the soil sub Regions (at mains limited. The attempts made by Salvador Sanchis et al. (2008)
scale 1:1,000,000) 4.2 to 6.3 (cartographic units covering several were all negative and even the loose trends found by Torri et al.
soil types). The soil units and dominant soil families represented in (1997) where not conrmed. This paper tries to bridge this lack of
Fig. 4 are described in Table 2. predictability by introducing the probability of nding a value of
According to the Kppen criteria, the local climate is classied as soil erodibility which is in agreement with the measured K-data
Mediterranean climate with hot summers, or, in other words, Csa from soils with similar characteristics. Hence an algorithm (KUERY)
warm group: has been developed which calculates probabilities under the form of
percentiles. KUERY works using as input data climate (dichotomous),
C temperate humid climates with mild winters and air tem- rock fragment content (continuous from 10% up), textural data (log-
peratures between 3 C and 18 C for the coldest month; arithms of ne soil geometric mean particle size and standard devia-
s with a dry summer, where the driest month of summer re- tion) and soil organic matter content.
ceives less than 40 mm, and the month with most rainfall in With this input data a probability distribution of K-values is pro-
winter has at least 3 times the rainfall of the driest summer duced. The user is then invited to opt for some of the values on the
month; basis of available local knowledge or of other data which he/she can
a hot summer. The air temperature of the warmest month is access. Soil aggregate stability, exchangeable sodium content and
above 22 C, at least 4 months have an average above 10 C. electrical conductivity (i.e. dispersivity of the soil) are certainly very
good indicators of soil erodibility. Inltration rate and depth of the
This climatic group corresponds to the warm group according to rst impervious horizon are other parameters that can help selecting
Salvador Sanchis et al. (2008). Here we also present the elaboration a realistic K value.
for the case cool climate in order to exemplify the differences be- An application of KUERY to a real situation is illustrated for a small
tween the two situations. catchment in South Italy.
Applying KUERY gives rise to probability distributions of K-values, The climate of this catchment belongs to the warm climate group.
shown in Fig. 5b and c where soil erodibility for the warm and cool Calculations have also been made for the cool climate group, to show
climate case is shown through its mean value. In both cases (warm the signicant differences in K between the two groups. The differ-
and cool climate) there are soils with and without rock fragments. ences between the two scenarios (Cool and Warm climate, Fig. 5) in-
Hence the example covers the four basic subsets: 2 climate groups dicate that care should be taken when dening the climate zone.
and 2 subgroups each having a different rock fragment content (e.g. Obviously climatic zones, such as rock fragment above/below the
the unit 4.5 in Fig. 4 is characterised by Rk > 10%). Fig. 5a presents 10% thresholds are all sharp discontinuities which add to the

Table 2
Soil cartographic units present in the Esaro catchment (Fig. 4) (from ARSSA-CAL, 2003).

Pedological Description
features
(Soil unit
code)

4.2 Includes at areas, sometimes reclaimed and locally terraced. The substrate are ne, calcareous sediments. Predominant land use: cropland, vineyard.
Dominant soil families: Chromic and Vertic Haploxererts
4.4 Mainly present in the valley bottoms in hilly countryside whose substrate are ne-grained colluvial and alluvial deposits Predominant land use: cropland.
Dominant soil families: Entisols
4.5 This unit includes three orders of Quaternary marine terraces:, located between 40 and 170 m asl The surfaces are terraced, at times, deeply incised by the
drainage network. Along the cuts and slopes of the terraces the transition between the calcarenitic thick crust and the Pliocene clays is visible. Land use:
cropland (including orchards).
Dominant soil families: Typic Haploxerepts and Vertic Haploxerolls
4.7 This unit is mainly distributed parallel to the coastline, usually below 100 m asl.,
along terraces with a pedogenetic substratum which consists of coarse sediment on silty clay Pliocene formation. Land use: cropland, olive grove, and vineyard.
Dominant soil families: Typic Haploxeralfs and Typic Fragixeralfs
6.3 This unit is located on hilly morphology having low to moderately steep slopes over Pliocene silty-clay sediment.
This environment has been affected in recent decades by signicant land use changes: from pasture to grain cultivation in monoculture which triggered soil
degradation and erosion with gullies and local small biancana badlands. Land use: cropland. Dominant soil families: Typic Endoaquents, Vertic Haploxerepts
and Vertic Calcixerepts
L. Borselli et al. / Catena 97 (2012) 8594 91

discontinuity due to soil type borders. Here no sophisticated analysis quickly resulting in a lower soil aggregate stability. Laterization of
is proposed to solve this problem as it is beyond the scope of this soil and the dynamics of sexquioxides are processes typical of warm
paper. climate resulting in a larger soil resistance. Still other processes can
The precise reason behind the differences in soil erodibility due to be suggested and all these observations point to the fact that more re-
climate is not yet known. A possible explanation may be that clay search is needed in order to move forward and beyond the intrinsic
soils tend to have much larger inltration rates in a warm climate limits of soil erodibility denitions such as the one used by the USLE.
where they may have cracks which remain open during the whole The results shown in Fig. 6a, b indicate a strong sensitivity of
year. At the same time soil organic matter will be mineralised more USLE-type erosion models used for the assessment of annual/monthly

Fig. 5. Spatial distribution of soil erodibility (K, in SI units) in the Esaro catchment: a) soil erodibility computed using the nomograph by Wischmeier et al. (1971) (Kw); b) mean
soil erodibility predicted by KUERY for a warm/temperate climate (K Warm; current situation)c) mean soil erodibility predicted by KUERY for a cool/temperate climate (K Cool;
hypothetical scenario.
92 L. Borselli et al. / Catena 97 (2012) 8594

Fig. 5(continued).

soil erosion rate to K values obtained by the KUERY algorithm. This nal user has the choice to select the most appropriate value of K,
conrms the ndings of Borselli et al. (2009) for a similar application depending on additional soil information (Borselli et al., 2009)
in a watershed in central Italy. In the case of Esaro catchment 80% of which was usually not present in detail in previous soil erodibility as-
the area shows signicant large differences (+38%, 59% ) between sessments. The climatic effects on soil erodibility, reported by Sanchis
K-values calculated with KUERY and Kw calculated using the nomo- et al. (2008), are always clearly shown by the KUERY output. A com-
graph (Wischmeier et al., 1971). Under a hypothetical cool climate parison between the new mean soil erodibility values obtained by the
scenario this difference may rise up to +160%. The percentage of dif- KUERY algorithm with the classical ones calculated using a nomo-
ference found between K values will produce an equal difference in graph (Wischmeier et al., 1971) shows large differences. These large
soil erosion rate computed with the K factors in USLE-Type Models differences, i.e. between 59% and +160%, produce an equivalent
(Borselli et al., 2009). percentage of difference in soil loss rates predicted by USLE-type
These results suggest that the procedure for calculating the K fac- soil erosion models. The KUERY algorithm and software is rec-
tor, due to its large effect on soil erosion rate assessments and its en- ommended for a more reliable assessment of soil erodibility worldwide.
vironmental relevance for future climate change scenarios, needs a
complete reappraisal using a large global soil/climate erodibility data-
Acknowledgements
base and new analysis tools.
KUERY algorithm is a promising novel techniques and while waiting
This research was partly funded by the European Commission,
for better soil erodibility datasets to be collected, the KUERY algorithm
Directorate-General of Research, Global Change and Desertication
can help producing K-probability distributions for a range of soil types
Programme, Project No. GOCE-CT-2003-505361 (RECONDES), and
and climate conditions. The last version of the software (KUERY 1.3) is
the European Integrated Project: GOCE 20070370462, Desertication
available as freeware for the scientic community on the website:
Mitigation and Remediation of Landa global approach for local solu-
http://www.lorenzo-borselli.eu/kuery
tions (DESIRE). Regione Calabria contributed through POR Calabria
20002006, asse 1risorse naturali. misura 1.4sistemi insediativi,
4. Conclusions
azione 1.4.cazioni di studio, programmazione, sperimentazione,
monitoraggio, valutazione e informazione nalizzati alla predisposizione
The KUERY algorithm is a new tool allowing a more robust assess-
e gestione di politiche integrate d'intervento di difesa del suolo, lotto
ment of the probability distribution of erodibility values, used in
progettuale n.2, pericolosit legata ai fenomeni di intensa erosione idrica
USLE-type soil erosion models, for soils found worldwide and under
areale e lineare.
different climates. The new algorithm allows a rapid assessment of
soil erodibility (K), by a query to a pre-processed soil database and
climatic groups (Salvador Sanchis et al., 2008; Torri et al., 1997), References
using a specic freeware software (KUERY 1.3). An application of
ARSSA-CAL, 2003. In: Agenzia Regionale per lo Sviluppo e per i Servizi in Agricoltura
this algorithm to a catchment in the Mediterranean demonstrates (Ed.), I Suoli della Calabria, Programma Interregionale Agricoltura-Qualit, Misura,
the potential use of the KUERY algorithm. The new assessment proce- 5. Rubettino Industrie Grache e Editoriali, Soveria Mannelli. 387 pgg.
dure produces a set of parametric and non-parametric statistics of soil Borselli, L., Cassi, P., Salvador Sanchis, P., 2009. Soil Erodibility Assessment for Applica-
tions at Watershed Scale. In: Costantini, Edoardo A.C. (Ed.), Manual of Methods for
erodibility (K) values as well as an estimate of the uncertainty distri- Soil and Land Evaluation. Science Publisher Inc.. ISBN: 978-1-57808-571-2. 600
bution derived from data for similar soils in the global dataset. The pages. November 2009.
L. Borselli et al. / Catena 97 (2012) 8594 93

Fig. 6. Spatial distribution of differences in soil erodibility (k, in SI units, calculated by Eq. (12)) in the Esaro catchment. Soil erodibility had been calculated using different
methods: i.e. Kw (Wischmeier et al., 1971) and mean K by KUERY algorithm. a) difference between Kw (Fig. 5a) and K Warm (Fig. 5b) and b) difference between Kw (Fig. 5a)
and K Cool (Fig. 5c).

Buchinsky, M., 1998. Recent advances in quantile regression models. Journal of Human Poesen, J., 1993. Gully typology and gully control measures in the European loess belt.
Resources 33, 88126. In: Wicherek, S. (Ed.), Farm Land Erosion in Temperate Plains Environment and
Foster, G.R., Yoder, D.C., Weesies, G.A., Toy, T.J., 2001. The Design Philosophy Behind Hills. Elsevier Science Publishers, Amsterdam, pp. 221239.
RUSLE2: Evolution of an Empirical Model. In: Ascough II, J.C., Flanagan, D.C. Poesen, J., Torri, D., Bunte, K., 1994. Effects of rock fragments on soil erosion by water at
(Eds.), Soil Erosion Research for the 21st Century, Proc. Int. Symp. (35 January different spatial scales: a review. Catena 23, 141166.
2001, Honolulu, HI, USA. ASAE, St. Joseph, MI, p. 701P0007. Price, K.V., 1999. An Introduction to differential evolution. In: Corne, D., Dorigo, M., Glover, F.
Koenker, R., Basset, G., 1978. Regression Quantiles. Econometrica 46 (1), 3350. (Eds.), New Ideas in Optimization. Mc Graw-Hill International, London, pp. 79108.
Lee, H., Tanaka, H., 1999. Upper and lower approximation models in interval regression Renard, K.G., Foster, G.R., Weesies, G.A., McCool, D.A., Yoder, D.C., 1997. Predicting soil
using regression quantile techniques. European Journal of Operational Research erosion by water: a guide to conservation planning with the Revised Universal Soil
116, 653666. Loss Equation RUSLE. Agricultural Handbook, 703. USDA-ARS, Washington DC.
94 L. Borselli et al. / Catena 97 (2012) 8594

Rmkens, M.J.M., Young, R.A., Poesen, J.W.A., McCool, D.K., El-Swaify, S.A., Bradford, Storn, R., Price, K., 1997. Differential evolutiona simple and efcient heuristic for
J.M., 1997. Soil Erodibility Factor (K), in: Renard, K.G., Foster, G.R., Weesies, G.A., global optimization over continuous spaces. Journal of Global Optimization 11
McCool, D.K. and Yoder, D.C. (coordinators) Predicting Soil Erosion by Water: A (4), 341359.
Guide to Conservation Planning with the Revised Universal Soil Loss Equation. Ag- Torri, D., Poesen, J., Borselli, L., 1997. Predictability and uncertainty of the soil erodibil-
riculture Handbook Number 703, USDA-ARS, Washington DC, pp. 65100. ity factor using a global dataset. Catena 31 (1/2), 122.
Salvador Sanchis, M.P., Torri, D., Borselli, L., Poesen, J., 2008. Climate Effects on Soil Torri, D., Poesen, J., Borselli, L., 2002. Corrigendum to Predictability and uncertainty of
Erodibility. Earth Surface Processes and Landforms 33 (7), 10821097, http:// the soil erodibility factor using a global dataset. Catena 46, 309310.
dx.doi.org/10.1002/esp.1604. Wischmeier, W.H., Smith, D.D., 1978. Predicting Rainfall Erosion Losses. Agriculture
Shirazi, M.A., Boersma, L., 1984. A unifying quantitative analysis of soil texture. Soil Handbook No. 537. USDA, Washington, D.C. 57 pp.
Science Society of America Journal 48, 142147. Wischmeier, W.H., Johnson, C.B., Cross, B.V., 1971. A soil erodibility nomograph for
Shirazi, M.A., Boersma, L., Hart, W., 1988. A unifying analysis of soil texture: improve- farmland and construction sites. Journal of Soil and Water Conservation 26 (5),
ment of precision and extension of scale. Soil Science Society of America Journal 52, 189194.
181190. Yu, K., 2002. Quantile regression using RJMCMC algorithm. Computational Statistics &
Snyder, W.M., Richard, H., McCuen, R.H., 1991. Numerical Analysis with Sliding Polyno- Data Analysis 40, 303315.
mials. Lighthouse Publications, Mission viejo, California, USA, 561 pp. (ISBN 0- Yu, K., Lu, Z., Stander, J., 2003. Quantile regression: applications and current research
914055-09-7). areas. The Statisticians 52 (3), 331350.

You might also like