were available for five stations (c.f., Section 3.3), neither the comprehensive demands on the precise calculation requiring long-term precipitation data of single storm events nor on the spatially explicit determination of R factors based on a sufficient number of rainfall gauging stations could be met. Thus, Manuscript 3 presents a method that overcomes these restrictions by
(i) approximating the R factor based on erosivity indices and regressions functions, and
(ii) regionalizing the R factor using the elevation as a spatially explicit covariable interacting with the rainfall erosivity.
To overcome the restriction of data scarcity in terms of temporal resolution, many approaches have been developed to calculate the R factor for specific regions based on erosivity indices, such as the Modified Fournier Index (Fmod), and based on regression functions (e.g., LAL, 1976; ARNOLDUS,
1977; KAVIAN ET AL., 2011; ALIPOUR ET AL., 2012). The regressions are mostly based on annual precipitation Pa (e.g., VAN DER KNIJFF ET AL., 1999), precipitation per month Pm (FU ET AL., 2005),
precipitation per day Pd (YU and ROSEWELL, 1998; MEN ET AL., 2008), and on single rainfall events or
precipitation per hour Pe (e.g., MANNAERTS and GABRIELS, 2000). Those alternative approaches can
be very effective according to their correlation with the soil loss due to their regional or scale- dependent determination approach (RENARD ET AL., 1997). They further show that monthly and daily precipitation data already offer a useful and valid approximation of rainfall erosivity when used for the region of origin (e.g., STOCKING and ELWELL, 1976, ARNOLDUS,1977; LO ET AL., 1985; RENARD and
FREIMUND, 1994).
In total, 15 regression functions were tested in Manuscript 3 to approximate the R factors. These regressions incorporate the (i) MAP (ROOSE, 1976; BOLINNE ET AL., 1980; ROGLER and SCHWERTMANN, 1981; LO ET AL., 1985; VAN DER KNIJFF ET AL., 1999), (ii) precipitation per day Pd
(ELSENBEER ET AL., 1993; BAGARELLO and D’ASARO, 1994; YU and ROSEWELL, 1998; MANNAERTS
and GABRIELS, 2000; DE SANTOS LOUREIRO and DE AZEVEDO COUTINHO, 2001; SHAMSHAD ET AL., 2008; MEN ET AL., 2008) and (iii) Fmod (ARNOLDUS, 1977); BOLINNE ET AL., 1980; DVWK,1990;
MEN ET AL., 2008). They all fulfill the requirement of a sufficient discussion in literature in terms of units and can be calculated with the available data. For evaluating their power, the resulting R factors were compared with reference data based on high-resolution rainfall data according to WISCHMEIER
and SMITH (1978) for study sites in the TGA (SHI ET AL., 2004; LIU and LUO, 2005; WU ET AL., 2011b, SHI ET AL., 2012) that are directly comparable to the Xiangxi catchment in terms physio- geographic settings.
All available climate stations used in the study presented in Manuscript 3 belong to the low to medium mountain range between 134 and 950 m a.s.l. With 1,132 mm, Yichang station exhibits the highest mean annual precipitation (MAP). MAP at Badong, Xingshan, and Zigui are 1,074, 991, and 992 mm, respectively. Shennongjia shows the lowest MAP of all stations with 961 mm (c.f., Section 3.3).
The resulting R factors are highly variable and range from the lowest of approximately 26 to the highest of approximately 434,491 MJ mm ha-1 h-1 a-1 (Table 7). Except the R factors derived from the regressions by MEN ET AL. (2008), all other R factors were considered as not applicable to the Xiangxi catchment as discussed in Manuscript 3. They were excluded from further analyses as they do not adequately reveal the rainfall erosivity for the sub-humid climate and the orographic effect on the precipitation in the Xiangxi catchment. For instance, the R factors calculated according to ROOSE
(1976) are distinctly higher (8,149-9,225 MJ mm ha-1 h-1 a-1) than the evaluation data. The regression by Roose (1976) was originally developed for tropical West Africa with two rainy seasons. Thus, the relation between the R factor and the total annual precipitation is considered as not valid for mountainous regions like the TGA and other transition areas with unimodal annual rainfall patterns. Additionally, R factors based on the regressions by BOLINNE ET AL. (1980) and LO ET AL. (1985) can be interpreted as extreme outliers compared to all other R factors (Table 7).
The regression by MEN ET AL. (2008) integrates Fmod, the maximum daily precipitation of days
with more than 12 mm and the mean, annual summed precipitation from days with more than 12 mm precipitation. The threshold value of 12 mm is close to the threshold value of 12.7 mm as suggested by WISCHMEIER and SMITH (1978) and was reported as practical for separating erosive and nonerosive storms for the Yellow River basin in China (XIE ET AL., 2002). Thus, Fmod combines the precipitation
totals of all single erosive storm events within one year and correlates with MAP. Consequently, the analyses were continued with the R factors calculated according to MEN ET AL. (2008) as this method fits best to the objectives regarding sparse data considering the precipitation characteristics in the Xiangxi catchment.
Given the causal relation between precipitation and elevation in the Xiangxi catchment (WU ET AL., 2006; c.f., Section 3.2.2), the rainfall-dependent R factors are also assumed to increase with the elevation. Thus, R factors were spatially regionalized based on elevation bands taking into account (1) Ra in each elevation band and (2) the increase of Ra within each elevation band. Based on the linear,
available climate stations, Ra factors for each elevation band were derived according to the division
method by WU ET AL. (2006).
As explained in Manuscript 3, Ra progressively increases from about 1,986 MJ mm ha -1
h-1 a-1 at the outlet to 7,547 MJ mm ha-1 h-1 a-1 at the highest elevation in the Xiangxi catchment. Based on the information of Ra at the upper and lower limit of each elevation band, the increase in rainfall erosivity
per 100 meters was calculated according to the division method by WU ET AL. (2006) as shown in Manuscript 3. This allowed for the elevation-based, spatially explicit regionalization of Ra factors by
assigning R factors to each pixel in the DEM (Table 1) according to their increase within the elevation band using GIS techniques.
Table 7 Approximated R factors (MJ mm ha-1 h-1 a-1) for the Xiangxi catchment calculated from 1971 to 2000 based on regression equations.
Type of approximation/Author(s) Climate station
Badong Shennongjia Xingshan Yichang Zigui
R factors based on regression equations using the mean annual precipitation MAP
ROOSE (1976) 9,225.2 8,149.1 8,483.7 9,684.5 8,482.3
BOLINNE ET AL. (1980) 44.3 41.0 42.0 42.0 42.0
ROGLER and SCHWERTMANN (1981) 882.0 777.1 809.7 809.7 809.6
LO ET AL. (1985) 415,709.4 371,702.0 385,388.1 434,490.9 385,327.8
VAN DER KNIJFF (1999) 1,409.2 1,244.8 1,296.0 1,479.4 1,295.7
R factors based on regression equations using precipitation per day Pd
ELSENBEER ET AL. (1993) 5,093.2 3,130.7 4,091.7 5,575.1 4,168.6
BAGARELLO and D’ASARO (1994) 1,853.0 1,231.6 1,546.9 2,001.4 1,562.3
YU and ROSEWELL (1998) 4,895.2 3,285.0 4,121.2 5,400.6 4,142.0
MANNAERTS and GABRIELS (2000) 4,784.9 3,345.5 4,009.0 5,276.4 4,079.4
DE SANTOS LOUREIRO and DE 2,977.5 1,972.9 2,505.2 3,179.7 2,520.6
SHAMSHAD ET AL. (2008) 5,417.9 4,266.3 4,908.6 5,791.3 4,838.4
R factors based on regression equations using the Modified Fournier Index Fmod
ARNOLDUS (1977) 4,898.5 3,919.1 4,215.2 5,573.7 4,223.9
BOLINNE ET AL. (1980) 32.5 25.7 27.8 36.9 27.9
DVWK (1990) 1,075.7 948.5 988.0 1,158.5 989.1
MEN ET AL. (2008) 2,157.9 1,821.4 1,981.3 2,579.3 1,983.7
With approximately 530 MJ mm ha-1 h-1 a-1, the highest increase in Ra per 100 m can be
observed within the elevation band from 500 to 800 m a.s.l. This is due to the high increase of Pa from
1,200 to 1,600 mm in the humid northern subtropics in the Xiangxi catchment with intensive convective precipitation (HE ET AL., 2003). The lowest increase in Ra (~85 MJ mm ha
-1
h-1 a-1) per 100 m can be observed for altitudes above 1,200 m a.s.l. in the mid- and northern temperate zone of the Xiangxi catchment. A close link between MAP and the elevation is typical for climates with intensive convective precipitation such as in the Xiangxi catchment (HE ET AL., 2003). Subsequently, precipitation is mainly erosive with high rainfall intensities. Comparable ranges of R factors from elevation-based regionalization were given by GOOVAERTS (1999) for a study area in Portugal with a R factor range of 600 MJ mm ha-1 h-1 a-1 per 100 m increase in elevation, which is comparable with an
increase of Ra of approximately 530 MJ mm ha -1
h-1 a-1 per 100 m for the elevation band from 500 to 800 m a.s.l. in the Xiangxi catchment.
In the Xiangxi catchment, the long-term (1971-2000) Ra amounts to 5,222
MJ mm ha-1 h-1 a-1, generally showing enormous rainfall erosivity. The rainfall erosivity pattern clearly reflects the altitudinal layers in the Xiangxi catchment due to the regionalization based on elevation bands (Figure 13). A narrow strip of comparably low rainfall erosivity characterizes the steep topography of the lower reaches of the Xiangxi River belonging to the Mid Subtropics below 500 m a.s.l. According to the global assessment of land vulnerability to water erosion by BATJES (1996), R factors > 1,250 MJ mm ha-1 h-1 a-1, however, already account for high rainfall erosivity. This, in fact, means high to extreme rainfall erosivity to a hundred per cent for the Xiangxi catchment as the lowest rainfall erosivity at the outlet is R = 1,986 MJ mm ha-1 h-1 a-1 (Figure 13).
Figure 13 Spatial distribution of R factors in the Xiangxi catchment based on mean annual R factors according to MEN ET AL. (2008). R factors are in SI unit (MJ mm ha-1 h-1 a-1). Note: Incorrect figure
published in Manuscript 3. The above figure shows the correct color ramp and labeling.
Finally, Manuscript 3 shows that, considering the superimposed role of the climate regime for the rainfall erosivity, the approximation of R factors using the regression by MEN ET AL. (2008) is
adequate for the conditions in the Xiangxi catchment. The results are in agreement with findings from other studies using higher temporal resolution rainfall data under comparable climate conditions (SHI ET AL., 2004; LIU and LUO,2005; WU ET AL., 2011b; SHI ET AL.; 2012). Due to the low spatial density
of rainfall data, common regionalization methods, such as regression kriging as used by MEUSBERGER ET AL. (2012) for 71 gauging stations or inverses distance weighting based on 121 gauging stations as reported by ALIPOUR ET AL. (2012), could not be applied. Yet, under the given situation of data scarcity in terms of spatial resolution, the spatial regionalization of R factors also offers optimal results. According to WEILGUNI (2006), R factors from spatially sparse data can be regionalized using a spatial explicit feature as long as they interact with each other. Since the Xiangxi catchment causes a vertically zoned climate, the elevation can be used as an additional feature. Already BIEGER ET AL. (2012) used the altitudinal pattern in the Xiangxi catchment as benefit and sufficiently adjusted the low annual precipitation and volume of discharge through a regionalization using elevation bands. Thus, the division method by WU ET AL. (2006) serves as superior advantage in spatially regionalizing the R factors. The applied approach is able to overcome the limited data and to address the highly mountainous topography in the Xiangxi catchment.
4.4 DEGRADATION OF CULTIVATED BENCH TERRACES IN THE THREE GORGES AREA - FIELD MAPPING AND DATA MINING
(Manuscript 4, published in ECOLOGICAL INDICATORS 34, pp. 478-493, November 2013, DOI: 10.1016/j.ecolind.2013.06.010)
Much is known about terrace degradation due to agricultural abandonment (c.f., Section 1.1.3). However, until now no attention was paid to terrace degradation in areas experiencing agricultural intensification and rapid ecosystem changes. Since the widely terraced reservoir of the TGA currently belongs to the most dynamic large-scale anthropogenic influenced regions in the world (c.f., Section 1.2.2), rapid land use changes are expected to likewise impact terracing (Figure 9). In order to minimize soil erosion, e.g., through well-adapted land management and conservation practices, information on soil loss and its spatial and temporal variability is required. This of course, also includes the effect of terraces on soil erosion, especially, since bench terraces are widely-applied in the TGA (SHI ET AL., 2012).