Introduction
Gully erosion is likely the most serious form of land
degradation (Poesen et al., 2003). Gullies form because they are an
energy-efficient way for runoff to travel from uplands to valley bottoms
(Gyssels and Poesen, 2003; Simon et al., 2011). Gullies can contribute more
than 90 % of catchment sediment yield (Tebebu et al., 2010; Simon et al.,
2011; Zegeye et al., 2014). They
have also been found to damage structures and transport routes (Nyssen et
al., 2004b; Valentin et al., 2005).
Gullying is a threshold-dependent process controlled by a wide range of
factors (Valentin et al., 2005), including rainfall and flowing water, soil
properties, and drainage area. Capra et al. (2009) and Campo et al. (2013)
found that most of the gully erosion took place during heavy rainfall events
– i.e., storm events were one of the drivers for gully erosion. The
mechanical actions of the flowing water can result in a rapid mass movement
in the gullies by undercutting of the banks (see Fig. 1; Lanckriet et al.,
2015). When these mechanical actions at the gully head exceed the cohesive
strength of soil, erosion proceeds upslope through a headward cutting gully
(Muñoz-Robels et al., 2010). Interactions between such processes are
important as hydraulic erosion promotes bank collapse, which then modifies
subsequent hydraulic erosion (Thorne, 1990; Avni, 2005). Similarly, gully
formation is initiated with the occurrence of convergent shallow subsurface
flow that leads to seepage-induced erosion of surface soils, gully heads, and
sidewalls (Fig. 1f; Vanmaercke et al., 2016; Tilahun et al., 2013a) as well
as sliding (Fig. 1d). Soil saturation by a rising water table decreases the
soil shear strength (Poesen, 1993; Langendoen and Simon, 2008) and therefore
destabilizes banks (Simon et al., 2000; Langendoen et al., 2013). Active
gully networks are therefore predominantly found in the saturated valley
bottomlands (Tebebu et al., 2010; Steenhuis et al., 2014), and the deepest
and the most spectacular gullies occur in the bottom of the watershed, where,
in subhumid monsoonal and wetter climates, the soil becomes saturated
starting around the middle of the rain phase and then remain saturated until
the end of the rain phase (Tebebu et al., 2014).
Examples of gully expansion controlled by (a, b) bank
geometry (height and slope), (c) tension cracks, (d)
landslide, (e) soil pipes, and (f) saturated Vertisols
(gully development on conservation ditches, narrow ditch upstream of gully
headcut) in the subhumid Debre Mawi (a, b, d, f), Mota (c),
and Geregera (e) watersheds (pictures taken in 2013).
![]()
Soil properties and soil types also play a role in gully formation and
expansion. For example, Vertisols, heavy clay soils with a high proportion of
swelling clays (IUSS Working Group WRB, 2015), form deep, wide cracks from the
surface downward when they dry out (Fig. 1c) and are prone to the development
of pipes (Fig. 1e) that can collapse and thereby turn into large rills or
gullies (Valentin et al., 2005; Frankl et al., 2014). This may be one of the
reasons that most severe gully areas are often associated with Vertisols
(Valentin et al., 2005; Tebebu et al., 2014; Frankl et al., 2014). Similarly,
in pasture bottom lands, piping often leads to development of permanent
gullies (Jones, 1987; Zegeye et al., 2014). These pipes are part of gully
networks, and during the rain phase the infiltrating rainfall discharges
through the pipes, which increases the lower soil horizon's vulnerability to
erosion.
The drainage area at the gully head is one of the parameters explaining
linear, areal, and volumetric gully headcut retreat (Vandekerckhove et al.,
2003; Frankl et al., 2012, 2013b; Vanmaercke et al., 2016).
Runoff-contributing drainage area can be used as a surrogate for runoff,
especially if it is assumed that the rainfall amount is equal for all
drainage areas and that surface conditions and land use are also very similar
(Oostwoud Wijdenes and Bryan, 2001). Frankl et al. (2012) reported for the
semiarid Tigray region in northern Ethiopia that, among all environmental
characteristics in a catchment, only the drainage area had a strong positive
association with gully headcut retreat (hereafter, headcut retreat refers to
the longitudinal gully growth and bank failure refers to cross-sectional
gully growth). Similarly, when data from stable and unstable subcatchments
were combined, were stream heads and drainage area of gully (Muñoz-Robles
et al., 2010). Nyssen et al. (2002) claimed that increasing the drainage area
of the gully head enhances gully development. Additionally, long-term retreat
rates often show negative-exponential trends with runoff-contributing area of
the gully head, which could be explained by the declining runoff-contributing
area of the gully head as it moves upslope (Begin et al., 1980).
Most gully erosion studies in Ethiopia have been carried out in the semiarid
Tigray region in northern Ethiopia (e.g., Billi and Dramis, 2003; Nyssen et
al., 2006, 2008; Tamene et al., 2006; Frankl et al., 2011, 2012, 2013a, b,
c). In this region, rehabilitation of gully erosion has been relatively
successful (soil loss has been decreased to a range of 1–6 t ha-1) by
using check dams and upland soil and water conservation (SWC) measures
(Nyssen et al., 2004a, 2006, 2009). Using repeat photographs from 2006 to
2009, Frankl et al. (2011, 2013c) found that about 25 % of the assessed
gully sections were stabilized as a result of siltation behind check dams.
However, such physical structures have been ineffective in controlling gully
erosion in the (sub)humid Ethiopian Highlands, where gullies are formed in
saturated Vertisol areas and where water often bypasses the check dams
(Dagnew et al., 2015). Importantly, the amount of interflow and surface flow
in the humid region is different from that in the arid and semiarid regions
(Bayabil et al., 2010; Engda et al., 2011; Tilahun et al., 2013a, b;
Steenhuis et al., 2014). Conservation structures that are effective in
preventing gullying by overland flow in semiarid regions (Nyssen et al.,
2004a, 2006) may not be effective in the humid Ethiopian Highlands regions,
where interflow elevates groundwater tables in the valley bottom that promote
gully formation and expansion (Tebebu et al., 2010).
Thus, there is a clear need for a better understanding of gully erosion
processes and factors in the subhumid Ethiopian Highlands in order to design
effective gully control or rehabilitation measures. The objectives of this
study were therefore (1) to understand gully erosion processes in the subhumid
Ethiopian Highlands and identify factors controlling these processes for
effective conservation practices and (2) to provide quantitative estimates
of the severity of gully erosion in the subhumid Debre Mawi watershed.
Results
Gully expansion rates at the watershed scale (2005–2013)
The expansion of the gully network in the Debre Mawi watershed significantly
impacted the landscape (Table 1, Fig. S2 in the Supplement). Based on the
digitized aerial images of 2005 and 2013 we found that the total length of
the gully network increased from 8.7 km in 2005 to 26 km in 2013 (Table 1,
Table S1 in the Supplement). The surface area taken up by the gully expanded
from 4.5 ± 0.17 ha in 2005 to 20.4 ± 0.4 ha (or 3 % of the
watershed) in 2013, or equivalent to 2 ha yr-1. Using Eq. (4), this
represents a soil loss of about 0.80 ± 0.013 Mt, which is
equivalent to 127 t ha-1 yr-1 (Tables 1 and 2).
Expansion rates of the 13 gullies (2005–2014)
In this section we discuss the 13 gullies (G1–G13) monitored in more detail.
They have a combined watershed area of 200 ± 9.4 ha (Table 3).
Measurements were used from both aerial imagery (up to 23 March 2013) and
manual measurement (2013–2014 rain phases) (Table 2). The surface area of
the 13 gullies was 0.7 ± 0.05 ha in 2005 and expanded to a total
of 3.8 ± 0.26 ha in 2014. The corresponding soil loss from these
gullies between 2005 and 2014 was estimated at 156 ± 9 t
(Table 2). This is equivalent to 78 t ha-1 yr-1 (ranging from 7
to 350 t ha-1 yr-1 with standard deviation of
90 t ha-1 yr-1 for the individual gullies). During the last 2
years of the study (2013–2014), the land area lost due to the expansion of
the 13 gullies was 0.17 ± 0.014 ha (Table 2), which is about
10 000 t of soil (of which about 60 % or 47 t ha-1 occurred in
2013) and is equivalent to 25 ± 0.8 t ha-1 yr-1. The soil
loss of the individual gullies ranged from 14 ± 3 t
(1.5 ± 0.3 t ha-1 yr-1) for G12 to 5445 ± 804 t
(205 ± 52 t ha-1 yr-1) for G6 (Table 2). In 2014, the
longitudinal growth of most gullies (G1, G2, G3, G6, G7, G10, G12, and G13)
was significantly reduced resulting in less annual soil loss (Table 3).
List of soil and gully topographic factors for the 13 gully heads
in the Debre Mawi watershed, as well as observed gully head erosion during the 2013
and 2014 rain phases (between July and September). BD is bulk density and DA
is drainage area.
Gully
Min. water table
Clay content
Mean bulk
Headcut
Drainage
Linear headcut
Area retreat (m2)
Volumetric retreat (m3)
Soil loss (t)
name
depth (m)
(%)
density (g cm-3)
depth (m)
area (ha)
retreat (m)
2013–2014
2013–2014
2013–2014
Measured
Error
Measured
Error
2013
2014
Measured
Error
Measured
Error
Measured
Error
G1
1.50
58
1.26
0.02
3.9
12.8
1.44
0.4
0
3.3
0.46
9
1.5
11
2.0
G2
1.22
53
1.19
0.02
2.2
13
1.46
2.2
0.5
10.9
1.32
32
5.3
38.5
6.4
G3
0.02
55
1.14
0.05
1.4
41.6
4.59
36
0
22.5
3.17
146
25.3
167
29.7
G4
0.59
59
1.17
0.11
2
1.7
0.19
7
5
15.7
1.57
61
7.9
72
11.3
G5
0.05
60
1.16
0.04
4.8
68
7.38
10
3
101.5
10.16
1087
167
1260
199.4
G6
0.08
67
1.22
0.18
4.6
13.3
1.49
12
0
182.0
25.66
413
71.5
504
114.7
G7
1.36
59
1.15
0.05
1.4
0.7
0.08
0.2
0
0.7
0.09
1
0.1
0.9
0.2
G8
0.07
56
1.19
0.09
3.3
17.4
1.95
24.4
7
108.9
12.10
237
34.5
281
46.1
G9
1.20
59
1.19
0.09
3.4
6.8
0.77
3.8
1.65
21.2
2.46
73
9.5
87
13.2
G10
1.44
55
1.25
0.11
2.5
6.5
0.73
0.7
0
2.7
0.38
6
1.0
7.5
1.5
G11
0.45
66
1.23
0.07
4.2
9.2
1.03
6.2
1.4
123.4
14.17
356
55.6
437
72.2
G12
1.38
66
1.22
0.06
1.9
4.1
0.46
0.07
0
5.0
0.71
3
0.6
4
0.7
G13
1.25
60
1.14
0.05
1.3
4.8
0.54
0.04
0.8
10.1
1.18
3
0.5
2.8
0.6
Total/ave.
0.82
59.5
1.19
0.30
2.84
200
9.4
103
19
608
33.6
2427
195
2873
248
The observed expansion of the 13 study gullies in the Debre Mawi
watershed (see Fig. 2 for gully location): (a) cumulative headcut
retreat and rainfall during the 2013 rain phase, (b) increase in
gully surface area and volume during the 2013 and 2014 rain phases, and
(c) increase in the combined gully surface area and the total
summer rainfall (RF) between 2011 and 2014.
![]()
The recorded precipitation during the 2013 rain phase (44 days of rainfall)
was 917 mm, and in 2014 (31 days of rainfall) it was 1107 mm (Fig. 4c). The gully
headcut retreat in 2013 ranged from 0.04 to 36 m, with a combined total of
103 m (Fig. 4a, Table 3), whereas the total retreat in 2014 ranged from 0 to
7 m, with a combined total of 19 m (Table 3). Over these two monsoon
seasons (2013–2014), about 608 ± 33 m2 of cultivated land was
consumed by only the longitudinal headcut retreat of the 13 gullies. This is
equivalent to 36 % of the increase in total surface area (both
longitudinal and lateral retreat of the entire gully) of the 13 gullies
during 2013–2014, and about 1.5 % of the total surface area of the 13
gullies since their formation up to 2014. During 2013–2014, the soil loss
solely due to headcut migration equalled 2875 ± 248 t (Table 3),
which represented 30 % of the total soil loss from the 13 gullies in the
same period presented in Table 2. The total 2-year soil loss caused by the
linear headcut retreat of the individual gullies varied from 0.9 (G7) to 1260 t (G5) (Fig. 4b, Table 3). During 2013, only 6 of the 13 gullies (G3, G4,
G5, G6, G8, and G11) actively expanded, with lateral retreat (widening) varying
from 3 to 11 m and a headward retreat varying from 6 to 36 m, while the
other gullies remained fairly stable. The headcuts of gullies G3 and G8
migrated the farthest, 36 and 24 m, respectively, but only during the 2013
rain phase. However, because of the relatively shallow headcut depth (1.4 m)
and narrow width (2.6 m) of gully G3, its headcut migration contributed
little (only 7 %) to the total soil loss of the 13 gullies (Fig. 4b). As
shown in Fig. 4b, the four largest gullies (G5, G6, G8, and G11) were
responsible for about 94 % of the total soil loss from the 13 gullies.
The relationships between the lateral and longitudinal retreat and the
associated volumetric soil losses are discussed in Sect. 4.
Factors controlling gully headcut retreat and their relationships
with gully dimensions (2013)
The linear headcut retreat of the gullies varied by more than an order of
magnitude and was not related to geographic location. This variation should,
therefore, be explained by other factors, including groundwater elevation,
soil physical properties (texture, bulk density, and porosity), gully head
height, and drainage area (Table 3).
Power-type and linear regression equations of the longitudinal
headcut retreat, L, and the volumetric gully expansion, V, with the
controlling factors, X, listed in the first column: L or V = aXb. The
goodness of the fit is represented by the coefficient of determination,
R2.
Controlling factors (X)
L=aXb
For linear regression
V=aXb
For linear regression
equation of L with (X)
equation of V with (X)
a
b
R2
R2
p value
a
b
R2
R2
p value
Water table
0.73
1.11
0.62
0.64
0.001
12.7
1.09
0.60
0.49
0.009
Drainage area
0.21
1.047
0.33
0.28
0.06
2.32
1.26
0.47
0.67
0.0007
Headcut depth
0.25
2.16
0.21
0.0004
0.9
1.57
3.24
0.47
0.49
0.007
Clay content
110.7
-1.05
0.023
0.05
0.46
0.000
4.04
0.018
0.08
0.35
Bulk density
5.73
-6.2
0.008
0.13
0.23
26.8
1.28
0.0004
0.02
0.67
The correlation between the observed change in linear gully headcut retreat
(RL) and the precipitation recorded during the day of the gully
head retreat occurrence varied between -0.23 and 0.88. Some of the big
gullies such as G5, G6, and G11 showed strong correlation
(RL,G5 = 0.88, p = 0.009 and
RL,G6 = 0.84, p = 0.017), whereas gullies with the
greatest linear retreat (LG3 = 36 m and
LG8 = 24 m; Fig. 4a) showed weak relationships
(RL,G3 = 0.27, p = 0.55 and
RL,G8 = 0.34, p = 0.37). The fairly low correlation
coefficient is likely caused by the time delay between daily rainfall and
saturation of the soil surrounding the gully (Tebebu et al., 2010; Tilahun et
al., 2013b). Saturation of the gully banks is principally responsible for
destabilizing the gully head (Tebebu et al., 2010). Due to such slow
saturation processes, the daily precipitation and gully head retreat on the
same day are not correlated well. This does not mean that precipitation was
not related to retreat since the largest retreat rates were observed on
13 August 2013, after the maximum recorded daily rainfall (94 mm) on
7 August 2013, with little or no rainfall within this period (Fig. 4a).
The combined linear retreat (daily and cumulative) of the 13 gully heads in
2013 was compared with three different rainfall amounts: daily rainfall
recorded during the retreat events, cumulative rainfall between gully head
retreat events, and the cumulative rainfall since the beginning of the rain
phase. The combined linear headcut retreat showed a moderate relationship
with daily rainfall (RL = 0.76, p = 0.13) but fairly
strong relationship with cumulative rainfall between retreat events
(RL = 0.91, p = 0.01). Note that the relationship with
daily rainfall was relatively high due to the retreat that occurred on
13 August 2013 during the largest rainfall event as discussed above. When
this rainfall was excluded from the analysis, the correlation was reduced to
RL = 0.035 (p = 0.95). The combined cumulative linear
retreat was highly correlated with the cumulative rainfall since the
beginning of the rain phase (RL = 0.99, p = 0.0001).
This clearly indicates that cumulative rainfall, and thus gradual wetting and
saturation of the soil, is more important to headcut retreat than the wetting
and surface runoff from daily rainfall or individual storms.
The drainage area for the studied gullies varied from 0.7 (± 0.1) to 68
(± 7) ha with an average value of 15.4 ha and standard deviation of
18.9 ha (Table 3). In order to understand whether drainage area is related
to both linear and volumetric retreat of the gully in 2013, simple linear
regression models (Eqs. 10 and 11) and power-law relationships (Eqs. 12
and 13) between drainage area (DA, in ha) and cumulative headcut retreat
length (L, in m) and increase in gully volume (V in m3) were
developed. Since rainfall in 2014 was less erosive and small gullies did not
retreat, we did not use regression relationships for the 2014 rain phase.
L=0.3DA+3.15,(R2=0.28,p=0.06)V=11.7DA-13.2,(R2=0.67,p=0.0007)
Fitting a power-law relationship between both the linear (L) and volumetric
(V) gully retreat and drainage area yielded
L=0.21(DA)1.047,(R2=0.33),V=2.32(DA)1.26,(R2=0.48).
The predicted L and V using Eqs. (10) and (11) were compared linearly
with the measured L and V. The goodness-of-fit parameters for the length
L and volume V were RL2 = 0.28 (p = 0.06),
NSEL = 0.11, and PBIASL = 52 % and
RV2 = 0.69 (p ≪ 0.01), NSEV = 0.47,
and PBIASV = 49 %, respectively. Similarly, the predicted
L and V using power-type regression equations (Eqs. 12 and 13) were
compared with the measured L and V. The goodness-of-fit parameters were
RL2 = 0.33, NSEL = -0.36, and
PBIASL = 98 % and RV2 = 0.48,
NSEV = 0.48, and PBIASV = 49 %. For both
the linear and power-type fitting, the R2 and NSE of the volumetric gully
retreat were larger than those for the gully linear retreat.
Comparison of minimum groundwater table depth, gully headcut depth,
and the average groundwater table (WT) fluctuation between morning and night for the 13
study gullies in the Debre Mawi watershed, Ethiopia, for the 2013 rain phase.
![]()
Figure 5 shows the water table rose above the gully bottom for all 13 gullies
during the rain phase, which indicates mostly saturated gully head and bank
soils. In this study, the water table measurements were carried out twice a
day (i.e., in the morning and evening). The groundwater table fluctuated
between these readings (Fig. 5), but the variation was not significant
(p = 0.98). The water table decreased between morning and evening
readings on average by 0.7 cm with a standard deviation of 4.0 cm. The
greatest fluctuations were observed at G2 (Fig. 5). The power-type regression
model between the minimum water table depth during the rain phase (ranging
from 0.02 m at G3 to 1.5 m at G1) and the linear retreat and volumetric
expansion of the 13 gullies had fairly high coefficients of determination
with length (RL2 = 0.62) and volume (RV2 = 0.60) (Table 4).
By fitting a simple linear regression, the volumetric gully expansion was
significantly related to the height of the gully headcut (RV2 = 0.49, p = 0.007). However, the linear retreat of the gully was not
explained by the headcut height (RL2 = 0.0004,
p = 0.9). The reason is likely the fact that gully G3, which had large
linear retreat but small headcut height, affected the analysis. When this
gully is excluded from the analysis, the RL2 for the linear
and power relationship between the gully linear retreat and gully head height
increased from 0.0004 to 0.26 (p = 0.09) and from 0.21 to 0.52,
respectively. In this case, the gully height fairly well explained the linear
retreat. Note that gully heads of lower height are relatively more stable than
those with greater heights as the factor of safety for stability is
approximately inversely proportional to gully head height. An equivalent
increase in gully head stability can be obtained by regrading the gully head
to a lower slope.
The major soil texture for all gully banks was clay-sized (53 to 67 % with
standard deviation of 4.5 %), and an average bulk density of
1.2 ± 0.3 g cm-3 (Table 3). The gully head retreat rates were only weakly
correlated with the texture (Table 4), probably because of the limited
variation in soil texture. Linear regression and power-type regression of
clay content with the volumetric and linear headcut retreat were therefore
not significant (Table 4).
Discussion
Effects of gully erosion on agricultural lands
Gully expansion affects the economic feasibility of soil conservation
measures in reducing the amount of land available to farm. In 2013, the net
gully area in the Debre Mawi watershed was 3 % of the watershed area. If
additional strips of 1 m width on each side of the gully area is not
cultivated, the total area taken up by gullies becomes 5 % of the total
watershed area.
Most gullies in the watershed were not stable and impaired more than 16 ha of agricultural land from 2005 to 2013. Gully expansion in the Debre
Mawi watershed is not evenly distributed because the upper slopes of the
watershed (about 50 % of the watershed area) reduces gully formation
mainly because it does not saturate (Tilahun et al., 2013b; Steenhuis et al.,
2014; Tebebu et al., 2015). Gully expansion therefore affects mostly the
bottomlands, where soils become saturated around the beginning of July
(Tilahun et al., 2013b). A loss of 2 ha of productive farmland per year is
considerable for any farmer, but even more significant in a region with
smallholder farmers. As farmers' land holding in the Ethiopian Highlands is
about 1 ha of land per household (Sonneveld and Keyzer, 2003), the
land loss observed between 2005 and 2013 could have provided farmland for 16
farming households in the watershed.
The rate of soil erosion (2005–2014) due to gully expansion in the
subhumid Debre Mawi watershed (127 t ha-1 yr-1) is more than 5
times as much as the upland erosion reported by Tebebu et al. (2010) and
Zegeye et al. (2010) in this watershed. The soil loss relative to the
change in gully surface area is about 4000 t ha-1 yr-1 or
400 kg m-2 yr-1, which is more than 2-fold the rate reported by Daba et
al. (2003) for semiarid eastern Ethiopia over a 30-year period. One of the
reasons for the difference is that the gullies in the subhumid Debre Mawi
watershed are much deeper than in the semiarid area studied by Daba et
al. (2003). Another reason is that the soils are more often saturated in a humid
climate than in semiarid areas. Upland soil and water conservation
practices are not effective for areas with gullies because sediment
concentration in the runoff increased greatly, effectively negating any
positive effect of upstream practices (Zegeye et al., 2015).
Relationships of the
volumetric gully headcut erosion (V) with the headward migration length
(L) and the lateral erosion (W) during the 2013 rain phase for the 13
gullies in the Debre Mawi watershed. The observed gully volumes were fitted
as functions of L or W using power-law and linear regression models.
V–W and V–L refer to V as a function of W or L,
respectively.
Relationship
Power-law relationship (V = axb)
Predicted versus measured
a
b
R2
R2
NSE
PBIAS
V–W
0.65
2.15
0.63
0.63
0.89
-21
V–L
18.3
0.91
0.83
0.83
-0.004
-32.4
Linear regression (V=mx+n)
Predicted versus measured
m
n
R2
p
R2
NSE
PBIAS
V–W
33
-115
0.87
6.2 × 10-6
0.89
0.88
11
V–L
6.9
112
0.07
0.36
0.08
0.08
0
x represents W or L; a, b, m, and n are constants.
The relationship between gully headcut dimensions and their controlling
factors
Table 5 lists the goodness-of-fit parameters Eqs. (5–7) of the power-law and
linear regression relations between the change in gully volumetric headcut
erosion (V), the top width retreat or lateral expansion (W), and the linear
headward migration of the headcut (L). Both the power and linear regression
analyses (Table 5) show that the volumetric gully expansion (V) was strongly
related to top width retreat (p < 0.01), whereas no significant
relation was found between V and L (p = 0.36). Similar results were
obtained using a power-law relationship (Table 5). Additionally, as shown in
Table 1, the relative change between 2005 and 2013 in net gully area
(350 %) is more than 2-fold the relative change in length (199 %).
This indicates that sideways or lateral gully retreat is a more important
mechanism of soil loss and gully expansion than linear migration of gully
headcuts. Note that the R2 value can sometimes be misleading, as
indicated by the other goodness-of-fit parameters assessed. For example, the
V–L power-law relationship for all 13 gullies has an R2 = 0.83,
which indicates a good fit between gully volume and linear gully extension.
However, based on NSE and PBIAS (Table 5), the relationship between gully
volumetric expansion predicted by the V–L equation (V = 18.3 L0.91) and
the observed volumetric expansion is not in the acceptable range (NSE = -0.004,
PBIAS = -32.4). This indicates that assessing the quality of fits
between gully expansion parameters cannot solely be done based on R2,
and that good fits also require other measures like NSE and PBIAS to be in
the acceptable range.
Both the linear (Eqs. 10 and 11) and power-type (Eqs. 12 and 13) regression
relationships indicated that drainage area predicted the volumetric gully
erosion (V) better than the linear headward migration (L) of the gully
headcut. This suggests that the larger the drainage area, the greater the
lateral expansion is by collapsing gully banks, and hence the greater the
sediment production. Studies in the semiarid Ethiopian Highlands with
relatively shallow soils over bedrock have indicated that drainage area
(which was not significantly related in the Debre Mawi catchment with deep
soils) was a major controlling factor of gully head retreat (Poesen et al.,
2003; Frankl et al., 2012).
Similar relationships were also developed by Vandekerckhove et al. (2003) for
semiarid southeastern Spain (V = 0.069 DA0.38, R2 = 0.51) and
Frankl et al. (2012) for the semiarid Tigray region in northern Ethiopia
(V = 0.53 DA0.31, R2 = 0.27). Note that, in the Debre Mawi
watershed, gully volume expansion is stronger related to drainage area than
in southeastern Spain and the Tigray region, as the power-law exponent is about
4 times greater in the Debre Mawi watershed: the larger the exponent, the
greater the increase in V per unit increase in drainage area (Frankl et al.,
2013b). The ratio of the volumetric expansion relations for Debre Mawi and the
Tigray region is 4.4 DA0.95, which shows a near linear increase in this
ratio with drainage area. For a gully draining 10 ha of land in the Debre
Mawi watershed, the volumetric expansion is on average almost 40 times
greater than that of a gully draining the same area in the semiarid Tigray
region studied by Frankl et al. (2012). The greater retreat rates in Debre
Mawi are caused by the rainfall amounts during the rain phase exceeding
potential evaporation with excess water saturating the valley bottoms (see
Sect. 3.3). In addition, the Vertisol soils are up to 10 m deep overlying
the bedrock. This combined with high groundwater tables indicates the
potential for erosion is greater in the Debre Mawi watershed compared with
the drier semiarid regions of Tigray and southeastern Spain, where soils are
also thinner. These findings are in accordance with Frankl et al. (2013b) in that
the establishment of relationships like Eqs. (12) and (13) is necessarily
region-specific and only representative of similar environmental settings
with respect to climate, topography, lithology, soil, and vegetation.
Examples of gully expansion in the Debre Mawi watershed. Top four
photos: expansion of gullies G8 and G11 during the 2013 rain phase and the
trees which fell into the gullies from upstream. Bottom images: expansion of
gullies G6 and G11 between 2005 and 2013.
![]()
Viable gully erosion control measures for the subhumid Ethiopian
Highlands
Gully erosion can rapidly change landscapes as can be seen for instance for
G6, G8, and G11 in Fig. 6. Gully G6 has expanded laterally into cultivated
land through erosion of the right bank (west bank), whereas lateral erosion
of the left bank, located on grassland, was rather limited. This decreased
gully expansion on the grassed bank may be due to the effect of the grasses
either in terms of increasing the topsoil shear strength (De Baets et al.,
2008) or drying out the soil through evapotranspiration (Pollen and Simon,
2005) and thereby reducing soil saturation. Also, grasses could modify
overland flow and infiltration patterns and therefore affect subsurface
drainage. Gully G11 was surrounded by cultivated land on both sides and hence
expanded laterally through erosion of both left (south) and right (north)
banks. In 2012 the land adjacent to the left bank was planted with eucalyptus
trees to halt erosion. In 2013, erosion of this left (south) bank was
significantly reduced, and gully development then occurred through extension
in northeastern direction and lateral expansion in northern direction instead
(see Fig. 6). The lesson learned from these two gullies (G6 and G11) is that
vegetation may reduce gully expansion by increasing soil shear strength
through their roots, slowing down the storm runoff, and trapping sediments,
which was also observed by, among others, Gyssels and Poesen (2003) and De
Baets et al. (2006). Therefore, planting suitable species on the gully face
and around the boundary may reduce or slow down bank failure and
water-induced erosion, especially for fairly deep gullies. In contrast to the
above explanation, although both banks of G8 were surrounded by grasses
(Fig. 6), the gully head migrated uphill by about 25 m in 2 months. The
reasons for this could be that (1) both banks were steep and deep enough for
gravity-induced bank failure and that (2) the surrounding soil was highly
saturated (Table 3) and bank layers near the bottom were more erodible than
the overlying layer, causing a preferential retreat that undercut the bank
and consequent cantilever failures (Fig. S3 in the Supplement).
Our monitoring data also contained valuable information regarding the
effectiveness of soil and water conservation measures such as soil bunds that
were extensively installed across the upper portion of the catchment since
2012. Dagnew et al. (2015), in the same watershed, reported that soil bunds
reduced runoff by 60 %, sediment concentration by 36 %, and sediment
load by 80 %, which resulted in a significant reduction of runoff volume
and sediment loads in the first 2 years of implementation. However, a
reduction of downslope sediment concentration was not significant due to the
presence of large gullies near the watershed outlet. Further, the SWC
measures (soil bunds), aimed to reduce the development of rills and gullies
in the area, were implemented on saturated Vertisol areas and have rather
led to gully initiation and development (see Fig. 1f; Steenhuis et al., 2014;
Dagnew et al., 2015). These soil and water conservation measures appear to be
ineffective on these locations as they cannot reduce or stop upslope gully
headcut migration, which requires alternative, structural measures.
Similarly, diversion waterways have been tested in the watershed to arrest
gully heads but have produced new gully branches (Zegeye et al., 2014). Our
data therefore support the findings of Dagnew et al. (2015), which indicate
that the extensive implementation of soil and water conservation measures on
periodically saturated Vertisol areas may have exacerbated rather than
mitigated gully formation and expansion.
In the bottomlands of the watershed with Vertisols dominant, gully formation
was severe due to alternate swelling and shrinking of expanding clays
resulting in deep cracks in the dry season (Fig. 1c). As was previously
observed by Frankl et al. (2012), the shrink–swell behavior of Vertisols
eventually developed into pipes (Fig. 1d) and contributed to gully
development. Though pipes contribute to gully formation, we observed in this
study that they are also important to drain excess subsurface water near the
gully banks, thereby potentially mitigating gully expansion. For example,
soil pipes in the heads of gullies G7 and G13 drained the elevated groundwater table, resulting in only minor headcut retreat (Table 3). This implies
that gully expansion rates could be reduced by controlling the water table
and therefore the pore-water pressures in the gully head (Zegeye et al.,
2016). For example, drop pipes are a common practice in the United States
(Field Office Technical Guide standard 587; NRCS, 2015) to control
groundwater and surface water level to halt erosion of gully heads up to 15 m
in height, but they can be very costly (> USD 50 000 each).
Moreover, most of the Debre Mawi watershed gullies are deep gullies (up to 7 m)
that are therefore susceptible to gravity-induced bank collapse.
Regrading the gully head and bank slopes decreases their weight and reduces
the probability of bank failure (Langendoen et al., 2014; Zegeye et al.,
2016).
In the semiarid Tigray region of northern Ethiopia, Frankl et al. (2012)
recommended the application of a subsurface geo-membrane (vertical dam) at
the gully head to increase groundwater levels and subsequently decrease soil
cracking and soil piping. However, this may not be effective in the (sub)humid region of the Ethiopian Highlands as we have shown that elevated
groundwater tables increases the rate of gully expansion (Table 4, Fig. 1f).
Therefore, gully mitigation measures in the subhumid Ethiopian Highlands and
similar climate types should aim to reduce soil water content.
Conclusions
Field observations in the Debre Mawi watershed indicate that permanent
valley-bottom gully drainage networks and in particular gully widening and
headcut retreat are important erosion processes severely impacting the
productive farmlands.
Gully mapping and monitoring indicated that the continued gully incision,
lateral expansion, and headward extension are governed by the collapse of the
gully head and sidewalls, and the subsequent removal of the failed materials
by flowing water (Figs. 1 and 6). About 5 % of the watershed area has
been impaired by the expanding gully network. The gully expansion rate at the
watershed scale between 2005 and 2013 was 127 t ha-1 yr-1
(Table 1). The headcut migration of the 13 gullies during the 2013 rain phase
varied significantly from 0.04 to 36 m yr-1 (Table 3 and Fig. 4).
Understanding the controlling factors of gully head migration and lateral
expansion of gullies is crucial to design appropriate gully control measures.
Retreat rates depended most strongly on groundwater table elevation
(Table 4). The elevated water tables saturate the soils surrounding the
gullies, thereby reducing the soil erosion resistance. An elevated groundwater
table may also lead to seepage-induced erosion. Additionally, the gully head
depth and the drainage area, which is representative of surface runoff
magnitude, were other factors controlling gully erosion in the Debre Mawi
watershed (Table 4). Therefore, conservation practices that address these
parameters may be most effective.
The lateral retreat for deep gullies contributes the most to the volumetric
gully erosion in the Debre Mawi watershed (Table 5). Therefore, regrading the
gully head and bank slopes could reduce the occurrence of gravity-induced
bank collapse for deep gullies. Studies need to be designed to evaluate the
effects of controlling groundwater movement, for example by subsurface
drainage, on the stability of Vertisols. Vegetation may play a vital role in
reducing soil water and increasing soil shear strength. The planting of an
assemblage of suitable, native plant species (both herbaceous and woody) is
being tested in the watershed.