Download as pdf or txt
Download as pdf or txt
You are on page 1of 11

JOURNAL OF GEOPHYSICAL RESEARCH, VOL. 111, D22109, doi:10.

1029/2006JD007169, 2006
Click
Here
for
Full
Article

Long-term (105 years) variability in rain erosivity as derived


from 10-min rainfall depth data for Ukkel (Brussels, Belgium):
Implications for assessing soil erosion rates
Gert Verstraeten,1 Jean Poesen,1 Gaston Demarée,2 and Christian Salles3
Received 6 February 2006; revised 9 June 2006; accepted 31 July 2006; published 29 November 2006.

[1] A 10-min rainfall depth time series recorded at Ukkel, Brussels (Belgium) for the
period 1898–2002 was used to calculate a long-term rain erosivity record. The rain
erosivity factor (R factor) of the Revised Universal Soil Loss Equation (RUSLE) for the
period 1898–2002 equals 871 MJ mm ha1 h1 yr1, based on a newly developed rain
intensity–kinetic energy equation (I-KE) for central Belgium. This R value is 26%
larger compared to the R factor based on the I-KE equation recommended in the RUSLE.
No significant monotonic trend in annual R factor could be observed over the entire
period, yet a standard normal homogeneity test showed a significantly higher R factor
(+31%) for the period 1991–2002 compared to the period 1898–1990. Annual variability
in R factor is very high, with a coefficient of variance of 31%. For central Belgium, rain
erosivity is highest in the period May–September, which corresponds well with observed
soil loss rates and the occurrence of muddy floods. Especially the period May–June is
critical with respect to soil erosion. The year-to-year variability in rain erosivity for
May–June shows a different temporal pattern than the annual erosivity. No statistically
significant increase in rain erosivity for May–June was found, and during the last decade
of the twentieth century these values are lower than average. Despite the lack of a
significant trend in annual rain erosivity, average 10-year erosion rates calculated with the
RUSLE have increased by 24–34% from 1903–1912 to 1993–2002 for major crops
grown in central Belgium, solely as a consequence of changing rain erosivity through
time.
Citation: Verstraeten, G., J. Poesen, G. Demarée, and C. Salles (2006), Long-term (105 years) variability in rain erosivity as derived
from 10-min rainfall depth data for Ukkel (Brussels, Belgium): Implications for assessing soil erosion rates, J. Geophys. Res., 111,
D22109, doi:10.1029/2006JD007169.

1. Introduction 1997], and its predecessor the Universal Soil Loss Equation
(USLE) [Wischmeier and Smith, 1978]. The (R)USLE is the
[2] Soil erosion by water following rain is one of the most most widely adopted soil erosion model, certainly for
important soil degradation processes around the world. It is assessing soil erosion at the national or regional scale
well established that the main parameters that control the [e.g., Van der Knijff et al., 2000; Jäger, 1994]. The RUSLE
rate of soil erosion by water are topography, vegetation, soil is an empirical model that was developed in the United
and rainfall [e.g., Morgan, 1995; Toy et al., 2002]. With States and that predicts the long-term annual soil loss
respect to the latter, it is the total kinetic energy of the rain (A) (t ha1 yr1) as the product of six factors representing
that is exerted on the soil that determines how much soil is rain and runoff erosivity (R) (MJ mm ha1 h1 yr1),
being detached or eroded, while the total amount of rain is soil erodibility (K) (t ha h ha1 MJ1 mm1), slope length
important for controlling the fraction of the detached soil (L), slope steepness (S), cover and management practices
that is being transported downslope. The most widely used (C) and supporting conservation practices (P):
parameter that represents the erosional impact of rain on
soils is the rain-runoff erosivity or R factor of the Revised
A ¼ R:K:L:S:C:P ð1Þ
Universal Soil Loss Equation (RUSLE) [Renard et al.,
A major problem when applying the RUSLE for regions
1 outside the United States is a correct interpretation of each
Physical and Regional Geography Research Group, Catholic Uni-
versity Leuven, Leuven, Belgium. of the six parameters from equation (1). This is also the case
2
Royal Meteorological Institute of Belgium, Brussels, Belgium. for the R factor. The relation between rain intensity (I) and
3
Maison des Sciences de l’Eau, Hydrosciences, University of rain kinetic energy (KE) varies from region to region [e.g.,
Montpellier 2, CNRS, IRD, Montpellier, France. Van Dijk et al., 2002]. However, since the relation between I
and KE is not readily available for many regions in the
Copyright 2006 by the American Geophysical Union.
0148-0227/06/2006JD007169$09.00 world, the equation developed in the United States by

D22109 1 of 11
D22109 VERSTRAETEN ET AL.: LONG-TERM VARIABILITY IN RAIN EROSIVITY D22109

Wischmeier and Smith [1958] or Brown and Foster [1987] mainly in the summer period that small-scale flooding
is often being used. This can introduce errors in R factor generated from agricultural land causes problems in local
calculations that can be as large as 10– 20% [Van Dijk et al., villages. However, their analysis also revealed that muddy
2002]. floods, which are related to severe erosion occurring on
[3] For central Belgium, Laurant and Bollinne [1976] arable land, is mainly observed in May – June, and not in
made a first assessment of the rain erosivity from 10-min July –August. In early ‘‘summer,’’ many fields with typical
rainfall data for the reference period 1934 – 1973, using the summer crops like maize, sugar beets, potatoes or chicory
rain-energy relation developed originally for the Universal have a very low vegetation cover and are thus very
Soil Loss Equation [Wischmeier and Smith, 1958]. They vulnerable to erosive rain events. Later on in the summer
calculated a mean annual R of 649 MJ mm ha1 h1 yr1and period, these crops have a well-established vegetation cover
this value has been used for estimating regional-scale that reduces the erosion rate drastically. On the basis of local
erosion rates in Belgium [Van Rompaey et al., 2000, newspaper reports, fire brigade interventions, and inter-
2001]. It can be questioned, however, whether the I-KE views with local people, there appear to be an increased
equation used by Laurant and Bollinne [1976] is appropri- incidence in muddy floods since the 1990s [Coen, 2003].
ate for central Belgium. Furthermore, it can also be argued This increase corresponds well with the observed global
whether the measurement period of 40 years is sufficient for warming trend and could suggest that climate change is the
determining a reliable long-term R factor. For the USLE and main trigger for the higher frequency of soil erosion related
RUSLE, a period of at least 22 years is normally recom- problems. However, increased awareness of soil erosion
mended for calculating R [Renard et al., 1997]. However, with stakeholders (public, farmers and policy makers) in the
this period of 22 years is recommended because very few 1990s in central Belgium also could explain the higher
detailed rainfall data records exist for a longer time period at frequency of reported muddy floods [e.g., Verstraeten et al.,
least not for the USA when the USLE was developed. For 2003]. It is therefore crucial to know whether rain erosivity
the meteorological station of Ukkel (Brussels, Belgium), a has changed during the critical spring period. Hence the
10-min interval rainfall record is available since 1898 objectives of this study are to explore the 10-min rainfall
[Demarée, 2003], which provides a unique opportunity for data set at Ukkel, Brussels, over the reference period 1898–
testing the impact of the length of the observation record 2002 in order (1) to calculate a reliable long-term rain
(10, 22, 40 or 100 years) on the calculated R factor. erosivity factor for central Belgium; (2) to look at the
[4] In the light of climate change, an important issue is impact of different I-KE equations on rain erosivity; (3) to
whether rain erosivity will change as well [e.g., Pruski and test statistically whether there is a temporal evolution in
Nearing, 2002]. In a warming world, it is expected that annual and intra-annual rain erosivity; and (4) to discuss the
changes in extreme precipitation, and thus more erosive implications of changing rain erosivity on soil erosion risk.
precipitation, will be more important than changes in mean
annual precipitation [Moberg and Jones, 2005]. Indeed, 2. Materials and Methods
erosion processes are in particular related to rather infre-
2.1. Calculation of Rain Erosivity
quent but intense rain events and not to annual rainfall. For
central Belgium, for instance, Takken et al. [1999] and [5] At the Royal Meteorological Institute in Ukkel,
Steegen et al. [2000] have shown that the majority of the Brussels (Belgium), 10-min rainfall depth data have
soil losses are due to rain events representing only a fraction been archived after digitizing the daily rainfall charts of a
of the annual rainfall. Tank et al. [2002] found a significant Hellman-Fuess recording rain gauge since 1 January 1898
increase in the mean precipitation amount per wet day in [Demarée, 2003]. We used this data set for the period
northwestern Europe for the period 1946 – 1999, which is 1898– 2002 to calculate the long-term R factor.
confirmed by Moberg and Jones [2005] who used daily [6] The R factor takes into account the kinetic energy of
rainfall records for ca. 80 stations in western and central the rain and the maximum intensity of a rain event for a
Europe for the period 1901– 1999. They could only observe period of 30 min [Renard et al., 1997]. For the calculation
a significant increase in daily precipitation depth in winter, of the R factor, two single rain events can be distinguished
but not in summer. To what extent this influenced annual when there is a dry period of at least 6h in between them. In
rain erosivity is not known. Furthermore, the possible order to obtain reliable measures of the total kinetic energy
impact of changing climate on the distribution of rain of a rain event, every rain event is normally divided into
erosivity throughout the year (i.e., the intra-annual variabil- different increments with constant rain intensity. This can be
ity) remains unknown. Nevertheless, a thorough under- easily read from a traditional analogue pen rain gauge, yet
standing of the intra-annual variability of rain erosivity is because of the digitization scheme used, these data were not
very important to assess the impact on erosion processes. available. Therefore we chose an increment of 10 min, as
Field studies of soil erosion in central Belgium revealed a this is the most detailed information available. The first
striking difference between the summer and the winter increment starts at 1 January 1898 at 00.00 hours, the
period. Vandaele and Poesen [1995] reported that ca. second at 1 January 1998 at 00.10 hours, etc. Rain intensity
60– 70% of total soil loss by water erosion in two small (ir) (mm h1) for every 10-min increment (Dtr = 1/6 h) is
catchments (25 ha each) in central Belgium occurred in late thus:
spring and early summer. For a 250 ha small catchment in
central Belgium, Steegen et al. [2000] measured 75% of DVr DVr
ir ¼ ¼ ð2Þ
the annual sediment export during the summer period, and Dtr 1=6
this was almost completely related to highly intense rain
events. Verstraeten and Poesen [1999] showed that it is where DVr is the rain depth (mm) during the increment.

2 of 11
D22109 VERSTRAETEN ET AL.: LONG-TERM VARIABILITY IN RAIN EROSIVITY D22109

[7] Rain energy per unit depth of rain for each increment Within the RUSLE, rain events with a total rain amount less
(er) (J m1 mm1) is often calculated using the following than 12.7 mm are not considered. However, for Belgium,
relation that was developed for the USA by Brown and Laurant and Bollinne [1976] showed that rain events with a
Foster [1987] and is advised in the RUSLE Handbook rainfall depth in between 1.27 and 12.7 mm have a total
[Renard et al., 1997]: erosivity which is 64% of the erosivity of the rain events
  with a rainfall depth > 12.7 mm. This is in agreement with
er ¼ 29 1  0:72 exp0:05ir ð3Þ data from experimental erosion plots in central Belgium,
Although in most countries rain kinetic energy is usually which indicate that 35% of the annual runoff follows events
calculated in a similar way as in equation (3) (see, e.g., Van with rainfall depths < 12.7 mm [Bollinne and Laurant,
Dijk et al. [2002] for a review), Salles et al. [2002] argued 1983]. Therefore a lower limit of 1.27 mm that was first
that a more appropriate expression to establish an empirical introduced by Laurant and Bollinne [1976] was used in this
law between I and KE is by calculating a time-specific study as well.
kinetic energy (KEtime) (J m2 h1) instead of the widely 2.2. Impact of Rain Erosivity on Soil Erosion
used volume-specific kinetic energy (KEmm) (J m2 mm1).
[9] Changes in annual rain erosivity cannot be linearly
For central Belgium, Salles et al. [1999] developed the
related to changes in soil erosion risk. As mentioned before,
following equation for predicting KEtime:
it is especially in the period May – June that large parts of
KEtime ¼ 12:6i1:22
r ð4Þ cultivated land in central Belgium have a minimal vegeta-
tion cover. An increase in rain erosivity in those months will
for rain intensities calculated with 1-min increments. This thus also generate more soil loss. If the temporal evolution
equation was established using detailed drop size distribu- in intra-annual and interannual rain erosivity is different, the
tion measurements of natural rain that were obtained by an variation in annual rain erosivity will not be entirely
optical spectro pluviometer (OSP) at Leuven for the period synchronous with variations in the rain erosion risk. There-
November 1997 to March 1999 [Salles et al., 1998; Salles fore we also analyzed the impact of temporal variability in
and Poesen, 1999]. This data set comprises 4849 min and rain erosivity on the RUSLE C factor, which expresses the
rain intensities ranged between 1 and 97 mm h1. Data from mean annual erosional susceptibility for different cropping
the OSP were used to calculate kinetic energy using drop- systems by combining the temporal evolution of the soil
size distribution and terminal fall velocity. Since no 1-min loss rate for a given vegetation cover with the temporal
rainfall data are available for a longer time period, a new evolution of the rain erosivity within a year [Renard et al.,
regression equation was calculated, now for rain intensities 1997]:
calculated with clock time 10-min rain increments:
P
m
SLRi Ri
KEtime ¼ 11:12i1:31
r ð5Þ i¼1
C¼ P
m ð9Þ
For further calculation of the rain erosivity, the rain energy Ri
per unit depth of rain is necessary. Therefore KEtime as i¼1

calculated with equation (5) was divided by ir: with SLRi the soil loss ratio for a time period i, which
depends on land cover characteristics, Ri the rain erosivity
KEtime during that period, and m the number of time periods
er ¼ ¼ 11:12i0:31
r ð6Þ
ir considered. Normally a 2-week time period is selected
[Renard et al., 1997]. In order to calculate the impact of
The total storm kinetic energy for a single rain event (E) can changing rain distributions on the C factor, 2-week average
then be calculated multiplying the rain energy per unit depth rain erosivity has been analyzed for 10-year periods (i.e.,
of rain for each increment with the total rain depth during 1903– 1912, . . ., 1993 – 2002). These data were used to
that increment, and summing this up for the entire rain calculate the temporal evolution of the C factor for 4 main
event: crops typically grown in central Belgium [Verstraeten et al.,
X
o 2002]: 3 summer crops (maize, sugar beets and potatoes)
E¼ er DVr ð7Þ and one winter crop (winter wheat). The temporal
r¼1 variability in RUSLE erosion rates for the period 1898–
with o the number of increments for a particular rain event. 2002 is calculated with equation (1), taking into account
[8] The erosion index (EI30) for a given rain event is equation (9). Thereby, it is assumed that the product of L,
calculated by multiplying the total kinetic energy for that S and P is 1 and K equals 0.04 t ha h ha1 MJ1 mm1. This
rain event (E) with the maximum rain intensity recorded assumption corresponds to a silt loam soil, which is typical
within 30 consecutive minutes (I30). The mean annual rain for central Belgium, and a standard erosion plot of 22 m long
and runoff erosivity factor (R) (MJ mm ha1 h1 yr1) is on a slope of 9% where no conservation measures are
finally calculated as the average value of EI30 for m number applied. No data are available on temporal variations in crop
of rain events during n number of years [Renard et al., rotation or farming operations, thus this approach only
1997]: illustrates the impact of changing rain characteristics on the
potential erosion risk, not on the actual erosion rates.
"m #
1X n X j
2.3. Statistical Analysis
R¼ ð EÞk ðI30 Þk ð8Þ
n j¼1 k¼1 [10] In order to test whether there is a significant trend in
j
rain erosivity, a nonparametric Mann-Kendall test (MK)

3 of 11
D22109 VERSTRAETEN ET AL.: LONG-TERM VARIABILITY IN RAIN EROSIVITY D22109

[Mann, 1945; Kendall, 1975] has been performed, which is K = k. The evolution of Sk in time is normally depicted in a
widely used in climatology [Sneyers, 1975] and in hydro- Sk
graph by plotting pffiffiffi over time, with s the standard
meteorological studies [e.g., Helsel and Hirsch, 1992; s n
Birsan et al., 2005]. The test is based on the S statistic deviation. The significance of the break can be tested using
defined as follows: the Buishand R statistic, here denoted as Rb in order to avoid
confusion with the rain erosivity factor of the RUSLE:
n1 X
X n  
S¼ sgn xj  xi ð10Þ ðmax S k  min Sk Þ
Rb ¼ ð16Þ
i¼1 j¼iþ1 s
Rb
If pffiffiffi > critical value, the break is significant. For n = 100,
where xj and xi are sequential data values in years j and i n
with j > i, n the number of observations and sgn() is the sign the critical value equals 1.62 for a 5% significance level.
function: The Buishand range test is more sensitive for breaks in the
middle of the series. For abrupt changes in the beginning
8
> 1 if xj  xi > 0 and at the end of time series, a more appropriate technique
>
>
> is the standard normal homogeneity test or SNHT
  <
sgn xj  xi ¼ 0 if xj  xi ¼ 0 ð11Þ [Alexandersson, 1986; Wijngaard et al., 2003]. In a SNHT,
>
> the T statistic is calculated which compares the mean of the
>
>
: first k years of the record with the mean of the last n-k years:
1 if xj  xi < 0:

If n > 8, the distribution of S can be approximated by a T ðk Þ ¼ kz21 þ ðn  k Þz22 k ¼ 1; . . . :; n ð17Þ


normal distribution. Under the null hypothesis, i.e.,
assuming no trend exists, the mean and variance of S are: where

mS ¼ 0 ð12Þ 1X k
ðxi  xÞ 1 X n
ðxi  xÞ
z1 ¼ and z2 ¼ ð18Þ
k i¼1 s n  k i¼kþ1 s
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi Values of T can be plotted over time. If the maximum value
sS ¼ ðn=18Þðn  1Þð2n þ 5Þ ð13Þ
of T(k) is higher than a critical value, the null hypothesis
(i.e., no break exists) will be rejected. For n = 100 at a
The standardized test statistic Z is calculated by significance level of 5%, the critical value equals 9.15 and
8 for n = 150 it equals 9.32 [Alexandersson and Moberg,
> S1
>
> if S>0 1997].
>
> sS
>
< [12] The Buishand and especially the SNHT are often
Z¼ 0 if S¼0 ð14Þ strongly influenced by outliers. In that case, a nonparamet-
>
> ric Pettitt rank test is preferable [Pettitt, 1979; Vannitsem
>
>
>
> Sþ1 and Demarée, 1991; Wijngaard et al., 2003]. The ranks r1,
: if S < 0:
sS . . .., rn of the X1, . . .., Xn are used to calculate the Pettitt
statistic, which is plotted over time:
For a two-tailed test, the null hypothesis is rejected at
significance level a if jZj > Za/2, where Za/2 is the value of X
k
the standard normal distribution with an exceedance Pk ¼ 2 ri  k ðn þ 1Þ k ¼ 1; . . . :; n ð19Þ
probability a/2. i¼1
[11] The Mann-Kendall test will test the significance of a
monotonic trend. However, rain and rain erosivity may also If a break occurs in year K = k, then the absolute value of
show a more abrupt change at a given time. Therefore a the statistic P is maximal for the year K. The critical value
Buishand range test [Buishand, 1982; Wijngaard et al., for PK equals 677 for n = 100 and a significance level of
2003] was performed to test the homogeneity of the data 5%.
and to detect change points in the time series of rain
erosivity. The null hypothesis assumes that the annual 3. Results
values of a variable are independent and identically distrib-
uted, whereas the alternative hypothesis assumes a stepwise [13] A total of 21 589 rain events could be identified for
shift in the mean. In the Buishand test, the rescaled adjusted the period 1898 – 2002 in Ukkel on 21 087 days with rain.
partial sums are calculated: Off all these rain events, 11 780 could be considered as
being erosive rain events (i.e., P > 1.27 mm). Calculated
X
k rain erosivity values (R) were analyzed for both the annual
Sk ¼ ðxi  xÞ 1in ð15Þ as the intra-annual variability.
i¼1
3.1. Annual Variability of Rain Erosivity
with xi the annual series, i the year from 1 to n and x the [14] The long-term annual R factor using the Brown and
mean value. In a homogeneous series, values for Sk will Foster [1987] equation (equation (3)) equals 689 MJ mm
fluctuate around 0. If, however, a break is present in year K, ha1 h1 yr1, while the R factor calculated using KEtime
then Sk reaches a minimum or maximum value near the year (equation (5)) is 871 MJ mm ha1 h1 yr1, which is 26%

4 of 11
D22109 VERSTRAETEN ET AL.: LONG-TERM VARIABILITY IN RAIN EROSIVITY D22109

Figure 3. Average R factor for the preceding 22 years


(thin line). The bold horizontal black line represents the
Figure 1. Comparison of annual rain erosivity (R) long-term mean R value (i.e., 871 MJ mm ha1 h1 yr1),
calculated with the Brown and Foster [1987] equation whereas the dotted horizontal lines represent the 95%
developed for the United States (equation (3)) and the R confidence interval around the long-term mean for a time
Factor Calculated With the I-KE Equation Developed for period of 22 years (= 1.96 sn0.5, with n = 22 and s2 =
Belgium (equation (5)). population variance for which the variance of the long-term
record was used).
larger. This significant difference between both values of the
long-term annual R factor is more or less similar for every R factors calculated with the Ukkel data set. Only when
single rain event and for every year (Figure 1). Thus the R factor is calculated in the year 1955 from the
changing the way in which the kinetic energy is calculated annual R values of the preceding 22 years, the mean
does not change the relative values of the kinetic energy for value is significantly different from the 105-year average
various time periods. Because equation (5) is based on value at a 5% significance level. For all other 22-year
observations of raindrop-size distributions in central Bel- periods, no significant difference could be found,
gium, we will further analyze the erosivity data that were although the range in mean R values remains quite high
calculated using equation (5). (749– 935 MJ mm ha1 h1 yr1). The analysis based on
[15] Figure 2 shows the evolution of the annual R factor data from Ukkel thus confirms that a 22-year rain erosivity
and the annual precipitation for the period 1898 – 2002. record should be sufficient for calculating the R factor. We
Minimum and maximum annual R equals 232 and also tested whether shorter time records, i.e., a 10-year
1540 MJ mm ha1 h1 yr1, respectively, whereas the record, would be sufficient. Figure 4 shows the mean R
coefficient of variance (CV) is quite large (31%) and illus- values for ten consecutive 10-year periods starting from
trates the need for long-term records in order to obtain a 1903. The mean R value for the period 1993 – 2002 is
reliable mean R factor. Wischmeier and Smith [1978] and significantly larger than the mean R value for all preceding
Renard et al. [1997] therefore recommended a minimum 10-year periods at the 95% confidence level, except for the
period of 22 years for calculating the long-term R factor. periods 1923– 1932 and 1963– 1972, for which the differ-
Figure 3 shows the temporal variation in 22-year average ences are only significant at a 90% confidence level. The
mean R value for the period 1933 –1942 is also significantly
smaller than the corresponding values for the periods 1923–
1932 and 1963 – 1972 at a 95% confidence level. For
calculating a reliable long-term average rain erosivity value,
a 10-year period is thus too short. The analysis of 10-year
rain erosivity (Figure 4) as well as the 9-year running
average of rain depth and rain erosivity depicted in Figure 2
suggests some kind of decadal variability: several periods with
larger then normal rain erosivity alternate with periods of
smaller than average R values. Especially during the most
recent period, the R factor is much higher than the long-term
average: 8 of the 10 last years have an annual R factor above the
mean value for 1898 – 2002. It was tested if this corresponds to
an increasing trend or with an abrupt increase in rain erosivity.
[16] No significant monotonic trend in annual rain ero-
sivity can be detected by the Mann-Kendall test (S = 161;
Z = 0.443 with a corresponding p value = 0.66). The
Buishand range test (Figure 5a) also shows no significant
Figure 2. Annual variability of rain erosivity (R) (MJ
Rb
mm ha1 h1 yr1) and annual precipitation depth (P) (mm) break in the time series of rain erosivity (pffiffiffi = 1.15). The
for the meteorological station of Ukkel, Brussels (Belgium) n
for the period 1898 – 2002. Buishand range test is in particular suited for detecting a

5 of 11
D22109 VERSTRAETEN ET AL.: LONG-TERM VARIABILITY IN RAIN EROSIVITY D22109

Figure 4. Mean annual rain erosivity values (MJ mm ha1 h1 yr1) grouped for 10-year periods
starting with years indicated. Points indicate the mean value for the period, shaded boxes represent the
±95% confidence interval of the mean values (= 1.96 sn0.5, with n = 10 and s2 = population variance for
which the variance of the long-term record was used), and the vertical lines represent the 95% confidence
interval for the individual values.

break in the middle of the time series, and this is illustrated values. Thus most of the statistical tests suggest no trend or
by Figure 5a, which shows that the maximum value of Sk is abrupt increase in rain erosivity over the entire period
found for the year 1959. The SNHT, however, does show an (1898– 2002). Only the SNHT suggest larger rain erosivity
abrupt change of mean annual rain erosivity near the end of since 1991.
the time series. The T statistic is above the critical value for [17] Figure 2 also shows the variability in annual precip-
the years 1989, 1990, 1991 and 2000, with the maximum itation. Also, no significant trend could be detected in
value in 2000 (Figure 5b). This means that a change could annual precipitation over the reference period 1898 – 2002,
be detected in the year 2000. However, the maximum value which confirms the findings of Vaes et al. [2001]. Like
of the T statistic for the final years is strongly influenced by annual rain erosivity, total precipitation depths per year
the very high rain erosivity values for the years 2001 and rather show a pattern of decadal variability. Both patterns
2002, which are the two largest values in the 105-year run more or less parallel (e.g., also higher than average
period. Instead, we selected the 2nd highest T value for precipitation in the period 1960 – 1974), yet the correspon-
fixing the break, i.e., in 1991. The mean value of R for the dence is far from clear. Furthermore, variations in annual R
period 1991 – 2002 is 31% larger compared to the mean are more pronounced than variations in annual P, which is
value for the period 1898 – 1990 (Figure 6). The Pettitt test due to the power equation between precipitation depth and
(Figure 5c), on the other hand, again does not detect a rain erosivity (equations (3) and (5)): CV for annual P is
significant break. However, since the Pettitt test is less only 16% compared to the 31% for annual R. Figure 7
sensitive to outliers, this is not surprising knowing that illustrates the relation between annual precipitation and
especially the final 2 years of the time series have such high annual rain erosivity. The large variability in annual rain

Figure 5. Results of the (a) Buishand range, (b) standard normal homogeneity test (SNHT), and
(c) Pettitt test applied to annual rain erosivity series at the meteorological station of Ukkel
(Brussels, Belgium) for the period 1898 – 2002. Dashed lines give 5% critical values.

6 of 11
D22109 VERSTRAETEN ET AL.: LONG-TERM VARIABILITY IN RAIN EROSIVITY D22109

Figure 6. Change-point result for annual rain erosivity in Ukkel. The break in 1991 is detected from the
standard normal homogeneity test (SNHT). Bold lines indicate the mean values for the periods
1898 –1990 (i.e., 840 MJ mm ha1 h1 yr1) and 1991– 2002 (1105 MJ mm ha1 h1 yr1), whereas the thin
horizontal line represents the mean value for the entire period (i.e., 871 MJ mm ha1 h1 yr1).

erosivity can only partly be explained by variations in of the three variables: Pearson correlation coefficients are
annual precipitation. Therefore the intra-annual variability below 0.1 and not significant.
in rain erosivity needs to be considered as well.
[18] As annual R is increasing during the last decades of 3.2. Intra-annual Variability in Rain Erosivity
the twentieth century, there is a striking parallel with [19] An important aspect of rain erosivity is the temporal
observed records of surface temperature. We therefore also distribution throughout the year. A simple regression
compared the annual variability in R with time series of revealed the following relation between precipitation depth
regional and local temperature and the North Atlantic for every erosive rain event and its corresponding rain
Oscillation Index (NAO). Regional temperature records erosivity (EI30) calculated with equation (5) and equation (7):
for the period 1900 –2002 were retrieved from Hansen et
al. [1999] (updated with data from http://data.giss.nasa.gov/ EI30 ¼ 0:1089:P1:8067 R2 ¼ 0:82 n ¼ 11 780 ð20Þ
gistemp), local temperature data are from the meteorological
station in Ukkel itself, and NAO values are taken from From Figure 8, it is clear that a large scatter exists, which is
Hurrel [1995] (updated with NAO Index Data provided by due to the large variability in rain intensities for rain events
the Climate Analysis Section, NCAR, Boulder, Colorado, with similar precipitation depths. For a given precipitation
USA, available at http://www.cgd.ucar.edu/cas/jhurrell/ depth, rain erosivity can vary by almost 2 orders of
indices.html). However, no correlation was found with any

Figure 8. Relation between event precipitation depth


P (mm) and the corresponding rain erosivity EI30 (MJ mm
Figure 7. Relation between annual precipitation (P) and ha1 h1 yr1) for each erosive rain event (i.e., P > 1.27 mm)
annual rain erosivity (R) for Ukkel, Brussels (1898 –2002). for Ukkel, Brussels (1898– 2002).

7 of 11
D22109 VERSTRAETEN ET AL.: LONG-TERM VARIABILITY IN RAIN EROSIVITY D22109

Table 1. Monthly Parameter Values for the Power Relation Table 2. Statistics of Monthly Rain Erosivity Values (R)
Between Rainfall Depth per Rain Event (P) (mm) and the (MJ mm ha1 h1 yr1) for the Meteorological Station of Ukkel
Corresponding EI30 Erosion Index for the Meteorological Station and for the Period 1898 – 2002
of Ukkel (1898 – 2002) Average
Coefficient Exponent N R Square Monthly R s CV Minimum Maximum
January 0.0974 1.6713 1101 0.84 January 35 28 79 0 127
February 0.1035 1.6399 950 0.83 February 27 27 103 1 168
March 0.1172 1.6233 948 0.81 March 32 25 78 0 115
April 0.1265 1.6569 979 0.81 April 36 29 81 1 146
May 0.1167 1.8857 932 0.82 May 86 93 107 1 493
June 0.1231 1.9099 929 0.83 June 96 88 92 6 410
July 0.1388 1.8455 962 0.82 July 177 163 92 1 869
August 0.1281 1.9713 954 0.85 August 135 135 100 1 728
September 0.1123 1.9189 901 0.84 September 89 87 99 1 548
October 0.1224 1.7244 960 0.84 October 66 67 100 0 428
November 0.1091 1.6721 1048 0.87 November 48 41 86 6 249
December 0.0925 1.7133 1116 0.87 December 43 35 81 2 235

magnitude, which can largely be explained by the timing of


[21] Figure 10 illustrates the annual variability in rain
the rain event. Table 1 and Figure 9 show how the parameter
erosivity for the period May – June. Similar to the total
values for equation (20) vary from one month to another.
annual rain erosivity (Figure 2), a large variability can be
Rain erosivity is much higher in the period May– September
observed, yet, the temporal patterns do not run parallel. The
compared to the period October –April. In the first period,
MK test does not indicate the presence of a linear trend in
the so-called ‘‘summer period,’’ rain is mostly generated
rain erosivity for May – June for 1898 – 2002 (Z = 0.0443).
from intensive convective rain showers. In the ‘‘winter
However, from Figure 10 it is clear that the series are
period,’’ on the other hand, the main rain generating
complicated and this requires subsets to be analyzed for
mechanisms are frontal weather systems related to depres-
identifying abrupt changes [Alexandersson and Moberg,
sions that generate longer lasting but less intensive rain. As
1997]. The Buishand range test and the Pettitt nonparamet-
stated before, it is in the early summer period (May – June)
ric range test indicate an abrupt increase in the May– June
that soil losses and erosion related problems of muddy
floods are most intense. Rb
rain erosivity after 1962 (pffiffiffi = 1.64; PK = 800), but this is
[20] This temporal pattern of rain erosivity is also illus- n
trated when the distribution of the mean monthly rain not confirmed by the SNHT test (T = 7.30). However,
erosivity for the whole period is considered (Table 2). The the period 1898 – 1962 (n = 64) is not homogeneous either.
period May – September (42% of the year) on average Therefore the statistics for homogeneity were repeated
accounts for 68% of the mean annual rain erosivity. July for the times series 1898 – 1962, which revealed a
and August have the largest monthly rain erosivity values. second abrupt change in 1932, when based on all three
Rb
tests (pffiffiffi = 1.71, T = 8.97; PK = 498). For n = 70, critical
n
values are 1.59, 8.80 and 393, respectively [Wijngaard et
al., 2003]. No critical values for n = 64 were calculated
but these will be lower than those for n = 70 and will thus
not change the results. The mean value for the period

Figure 9. Average monthly power equations between


precipitation depth per erosive rain event (P) (mm) and the
corresponding EI30 (MJ mm ha1 h1 yr1) for that rain Figure 10. Annual and 9-year running average rain
event. Parameter values and statistics of these power erosivity as a percentage of the annual rain erosivity for
equations are given in Table 1. the months May and June.

8 of 11
D22109 VERSTRAETEN ET AL.: LONG-TERM VARIABILITY IN RAIN EROSIVITY D22109

Figure 11. Change-point result for May– June rain erosivity (R) in Ukkel. The breaks in 1932 and 1962
are detected from the Buishand range and Pettitt test. Bold lines indicate the mean values for the periods
1898 –1931 (i.e., 195 MJ mm ha1 h1 yr1), 1932– 1961 (111 MJ mm ha1 h1 yr1), and 1962 –2002
(i.e., 225 MJ mm ha1 h1 yr1), whereas the horizontal thin line represents the mean R value for the
entire period (i.e., 182 MJ mm ha1 h1 yr1).

1932 – 1961 (112 MJ mm ha1 h1 yr1) is significantly dif- by a decrease in rain erosivity for May–June. A decrease in
ferent from the mean for the periods 1898 – 1931 and C factor, however, does not mean that soil loss is reduced as
1962 – 2002 (195 and 225 MJ mm ha1 h1 yr1, respec- well. Figures 12c and 12d show the temporal variability in
tively) (Figure 11). No significant differences between the RUSLE erosion rates for a standard slope (see section 2.2).
means for the first and third period exist, however. Annual soil loss rates for the four crops calculated with the
RUSLE run more or less parallel to the variation in R factor,
3.3. Temporal Variability in Potential Soil and deviations can be explained by the varying C values
Erosion Rates through time. For all crops, average 10-year RUSLE soil
[22] Although the RUSLE is intended to predict the long- loss rates have increased by 24– 34% from 1903– 1912 to
term average soil loss rate [Wischmeier and Smith, 1978], it 1993–2002. These variations only take into account the
can be used on shorter timescales as long as the appropriate impact of changing rain erosivity and erosivity distributions
parameter values can be calculated, and if the result is on the RUSLE erosion rates. Other parameters also varied
correctly interpreted as being a mean response for that given through time: the increase of the average field size, for
set of parameter values [Wischmeier, 1976]. For instance, if instance, increased the L factor as well [Van Oost et al.,
one would like to calculate the mean erosion rate for a 10-year 2000]. Furthermore, soil loss rates in central Belgium are
period, one can assess the various parameter values of also dependent on crop rotations, which have changed quite
equation (1) for that 10-year period. Therefore mean poten- a lot during the twentieth century. Maize, for instance, only
tial erosion rates have been calculated for the same 10-year became an important crop since the 1970s. Although no
periods as shown in Figure 4, also taking into account the detailed assessment of the impact of these changes was
impact of changing intra-annual rain erosivity distributions made, it can be safely stated that average soil loss rates by
on the erosional susceptibility of crops as illustrated by water erosion on cropland at the end of the twentieth century
equation (9). Figures 12a and 12b illustrate the impact of were 30–50% larger compared to average soil loss rates at
changing intra-annual erosivity distributions on C. The C the beginning of the same century.
factor for winter wheat is the least sensitive one to the
changing distribution of rain erosivity throughout the year
with decadal means ranging between 93 and 109% of the 4. Conclusions
long-term mean value, while C values for sugar beets vary [23] We used a 10-min rainfall depth time series for a
between 78 and 116%. For potatoes and maize, the varia- 105-year time period (1898– 2002) to calculate a long-term
tions are also quite large. Furthermore, it is clear that the rain erosivity record for Ukkel, Brussels. The rain erosivity
variability in C factor values does not follow the same factor (R factor) of the Revised Universal Soil Loss Equa-
pattern as the variability in annual R, but it rather corre- tion (RUSLE) for the period 1898 – 2002 equals 871 MJ mm
sponds with the temporal variability in rain erosivity for ha1 h1 yr1 but shows a very high annual variability with
May– June (Figure 10). The increase in mean annual rain a CV of 31%. It was shown that rainfall records of 22 years
erosivity during the period 1978– 2002 is counterbalanced or more, as is suggested by the RUSLE, are sufficient to

9 of 11
D22109 VERSTRAETEN ET AL.: LONG-TERM VARIABILITY IN RAIN EROSIVITY D22109

Figure 12. Evolution of 10-year average (a) C factors and (c) RUSLE soil loss rates for four major
crops presently grown in central Belgium, as well as the (b and d) deviation of these 10-year averages
from the long-term mean for the period 1898– 2002. The first 10-year period starts in 1903. For
calculating the RUSLE soil loss rates a K factor of 0.042 and an LS factor of 1 is assumed.

cover this high variability. Shorter records, e.g., 10 years, increased public awareness of soil erosion related problems
however, are not able to accurately predict a long-term or the changes in land use and land use structure have
mean value of rain erosivity. Variations in rain erosivity are caused this increase in reports. Despite the lack of a
only partly explained by variations in annual rainfall be- significant trend in annual rain erosivity, the average
cause of an important intra-annual variability. Although 10-year erosion rates calculated with the RUSLE have
there are temporal patterns in annual rain erosivity, no increased by 24– 34% from 1903– 1912 to 1993 – 2002
significant monotonic trend in annual R factor could be for the major crops grown in central Belgium, solely as a
observed over the entire period. However, for the period consequence of changing rain erosivity through time.
1991 – 2002, R is 31% higher compared to the period 1898 –
1990. This increase corresponds well with the generally References
observed trend in global average temperature, however, so Alexandersson, H. (1986), A homogeneity test applied to precipitation data,
J. Climatol., 6, 661 – 675.
far there is no further indication that global warming or Alexandersson, H., and A. Moberg (1997), Homogenization of Swedish
changes in circulation patterns can be held responsible for temperature data. part 1: Homogeneity test for linear trends, Int. J. Cli-
the higher rain erosivity at the end of twentieth century. matol., 17, 25 – 34.
[24] For central Belgium, rain erosivity is highest in the Birsan, M. V., P. Molnar, P. Burlando, and M. Pfaundler (2005), Streamflow
trends in Switzerland, J. Hydrol., 314, 312 – 329.
period May – September, which corresponds well with ob- Bollinne, A., and A. Laurant (1983), La prevision de l’érosion en Europe
served soil loss rates and the occurrence of muddy floods. Atlantique: Le cas de la zone limoneuse de Belgique, Pedologie, 33,
Especially the period May– June is critical with respect to 117 – 136.
Brown, L. C., and G. R. Foster (1987), Storm erosivity using idealized
soil erosion by water. The year-to-year variability in rain intensity distributions, Trans. ASAE, 30, 379 – 386.
erosivity for May – June shows a different temporal pattern Buishand, T. A. (1982), Some methods for testing homogeneity of rainfall
than the annual erosivity. As for the annual R factor, no records, J. Hydrol., 58, 11 – 27.
Coen, J. (2003), Water- en modderoverlast in Sint-Truiden en Gingelom:
statistically significant increase in rain erosivity for May – Een temporele anayse, M.S. thesis, Dep. of Geogr., Katholieke Univ.
June was found, and during the last decade of the twentieth Leuven, Leuven, Belgium.
century, these values are even lower than average. Thus the Demarée, G. (2003), Le pluviographe centenaire du Plateau d’Uccle: Son
increase in the number of reports on muddy flood events in histoire, ses données et ses applications, Houille Blanche, 4, 95 – 102.
Hansen, J., R. Ruedy, J. Glascoe, and M. Sato (1999), GISS analysis of
central Belgium during the last decades cannot be attributed surface temperature change, J. Geophys. Res., 104, 30,997 – 31,022.
to a climate change effect. We suggest that either the Helsel, D. R., and R. M. Hirsch (Eds.) (1992), Statistical Methods in Water
Resources, 522 pp., Elsevier, New York.

10 of 11
D22109 VERSTRAETEN ET AL.: LONG-TERM VARIABILITY IN RAIN EROSIVITY D22109

Hurrel, J. W. (1995), Decadal trends in the North Atlantic Oscillation: Vandaele, K., and J. Poesen (1995), Spatial and temporal patterns of soil
Regional temperatures and precipitation, Science, 269, 676 – 679. erosion rates in an agricultural catchment, central Belgium, Catena, 25,
Jäger, S. (1994), Modelling regional soil erosion susceptibility using the 213 – 226.
USLE and GIS, in Conserving Soil Resources: European Perspectives, Van der Knijff, J. M., R. J. A. Jones, and L. Montanarella (2000), Soil
edited by R. J. Rickson, pp. 161 – 177, CAB Int., Wallingford, U. K. erosion risk assessment in Europe, EUR 19044 EN, 34 pp., Eur. Soil Bur.,
Kendall, M. G. (Ed.) (1975), Rank Correlation Methods, Oxford Univ. Joint Res. Cent., Ispra, Italy.
Press, New York. van Dijk, A. I. J. M., L. A. Bruijnzeel, and C. J. Rosewell (2002), Rainfall
Laurant, A., and A. Bollinne (1976), L’érosivité des pluies à Uccle (Belgi- intensity-kinetic energy relationships: A critical literature appraisal,
que), Bull. Rech. Agron. Gembloux, 11, 149 – 168. J. Hydrol., 261, 1 – 23.
Mann, H. B. (1945), Nonparametric tests against trends, Econometrica, 13, Vannitsem, S., and G. Demarée (1991), Détection et modélisation des
245 – 259. sécheresses au Sahel: Propositions d’une nouvelle méthodologie, Hydrol.
Moberg, A., and P. D. Jones (2005), Trends in indices for extremes in daily Cont., 6, 155 – 171.
temperature and precipitation in central and western Europe, Int. J. Cli- Van Oost, K., G. Govers, and P. Desmet (2000), Evaluating the effects of
matol., 25, 1149 – 1171. changes in landscape structure on soil erosion by water and tillage, Land-
Morgan, R. P. C., (Ed.) (1995), Soil Erosion and Conservation, 2nd ed., scape Ecol., 15, 577 – 589.
198 pp., Longman, New York. Van Rompaey, A., G. Govers, T. Waumans, K. Van Oost, J. Poesen, and
Pettitt, A. N. (1979), A non-parametric approach to change-point detection, J. Desmet (2000), A regional soil erosion risk map for Flanders (Belgium)
Appl. Stat., 28, 126 – 135. (in Dutch), @WEL, 3, 1 – 6.
Pruski, F. F., and M. A. Nearing (2002), Climate-induced changes in ero- Van Rompaey, A., G. Verstraeten, K. Van Oost, G. Govers, and J. Poesen
sion during the 21st century for eight U.S. locations, Water Resour. Res., (2001), Modelling mean annual sediment yield using a distributed ap-
38(12), 1298, doi:10.1029/2001WR000493. proach, Earth Surf. Processes Landforms, 26, 1221 – 1236.
Renard, K. G., G. R. Foster, G. A. Weesies, D. K. McCool, and D. C. Yoder Verstraeten, G., and J. Poesen (1999), The nature of small-scale flooding,
(Eds.) (1997), Predicting Soil Erosion by Water: A Guide to Conservation muddy floods and retention pond sedimentation in central Belgium, Geo-
Planning With the Revised Universal Soil Loss Equation (RUSLE), Agric. morphology, 29, 275 – 292.
Handbk., vol. 703, U. S. Dep. of Agric., Washington, D. C. Verstraeten, G., A. Van Rompaey, K. Van Oost, G. Govers, J. Poesen, and
Salles, C., and J. Poesen (1999), Performance of an Optical Spectro Pluvi- L. Stalpaert (2002), Thema 2.21 Kwaliteit Bodem: Erosie, in MIRA-T
ometer in measuring basis rain erosivity characteristics, J. Hydrol., 218, 2002, Milieu en Natuurrapport Vlaanderen, Thema’s, pp. 309 – 316,
142 – 156. Vlaamse Milieumaatschappij, Garant, Leuven, Belgium.
Salles, C., J. D. Creutin, and D. Sempere-Torres (1998), The Optical Spec- Verstraeten, G., J. Poesen, G. Govers, K. Gillijns, A. Van Rompaey, and
tro Pluviometer revisited, J. Atmos. Oceanic Technol., 15, 1215 – 1222. K. Van Oost (2003), Integrating science, policy and farmers to reduce soil
Salles, C., J. Poesen, and A. Pissart (1999), Rain erosivity indices and drop loss and sediment delivery in Flanders, Belgium, Environ. Sci. Policy, 6,
size distribution for central Belgium, paper presented at 24th EGS Gen- 95 – 103.
eral Assembly, Eur. Geophys. Soc., The Hague, Netherlands, 22 April Wijngaard, J. B., A. M. G. Klein Tank, and G. P. Können (2003), Homo-
1999. geneity of 20th century European daily temperature and precipitation
Salles, C., J. Poesen, and D. Sempere-Torres (2002), Kinetic energy of rain series, Int. J. Climatol., 23, 679 – 692.
and its functional relationship with intensity, J. Hydrol., 257, 256 – 270. Wischmeier, W. H. (1976), The use and misuse of the universal soil loss
Sneyers, R. (1975), Sur l’analyse statistique des series d’observations, Note equation, J. Soil Water Conserv., 31, 5 – 9.
Tech. 143, 192 pp., Organ. Météorol. Mondiale, Geneva, Switzerland. Wischmeier, W. H., and D. D. Smith (1958), Rainfall energy and its rela-
Steegen, A., G. Govers, J. Nachtergaele, I. Takken, L. Beuselinck, and tionship to soil loss, Soil Sci. Soc. Am. Proc., 33, 131 – 137.
J. Poesen (2000), Sediment export by water from an agricultural catchment Wischmeier, W. H., and D. D. Smith (1978), Predicting Rainfall Erosion
in the Loam Belt of central Belgium, Geomorphology, 33, 25 – 36. Losses: A Guide to Conservation Planning, Agric. Handbk., vol. 537,
Takken, I., L. Beuselinck, J. Nachtergaele, G. Govers, J. Poesen, and U. S. Dep. of Agric., Washington, D. C.
G. Degraer (1999), Spatial evaluation of a physically based distributed
erosion model (LISEM), Catena, 37, 431 – 447. 
Tank, A. M. G. K., et al. (2002), Daily dataset of 20th-century surface air G. Demarée, Royal Meteorological Institute of Belgium, Ringlaan 3,
temperature and precipitation series for the European climate assessment, B-1180 Brussels, Belgium.
Int. J. Climatol., 22, 1441 – 1453. C. Salles, Maison des Sciences de l’Eau, Hydrosciences (UMII/CNRS/
Toy, J. T., G. R. Foster, and K. G. Renard (2002), Soil Erosion. Processes, IRD), F-34095 Montpellier Cedex 5, France.
Prediction, Measurement, and Control, 338 pp., John Wiley, Hoboken, G. Verstraeten and J. Poesen, Physical and Regional Geography Research
N. J. Group, Celestijnenlaan 200E, B-3001 Leuven, Belgium. (gert.verstraeten@
Vaes, G., P. Willems, and J. Berlamont (2001), 100 years of Belgian rain- geo.kuleuven.be)
fall: Are there trends?, Water Sci. Technol., 45, 55 – 61.

11 of 11

You might also like