Regional soil erosion risk mapping in Lebanon

Geomorphology 82 (2006) 347 – 359 www.elsevier.com/locate/geomorph

Regional soil erosion risk mapping in Lebanon Rania Bou Kheir a,⁎, Olivier Cerdan b , Chadi Abdallah a a

Lebanese National Council for Scientific Research/Remote Sensing Center, P.O. Box: 11-8281, Beirut, Lebanon b BRGM, B.P. 6009, 3 avenue Claude Guillemin, 45060 Orléans, France Received 26 April 2005; received in revised form 12 April 2006; accepted 19 May 2006 Available online 7 July 2006

Abstract Soil erosion by water is one of the major causes of land degradation in Lebanon. The problem has not yet been treated in detail although it affects vast areas. This study elaborates a model for mapping soil erosion risk in a representative area of Lebanon at a scale of 1:100,000 using a spatial database and GIS. First, three basic maps were derived: (1) runoff potential obtained from mean annual precipitation, soil-water retention capacity and soil/rock infiltration capacity; (2) landscape sensitivity based on vegetal cover, drainage density and slope; and (3) erodibility of rock and soil. Then two thematic maps were derived: potential sensitivity to erosion obtained from the runoff potential and landscape sensitivity maps, and erosion risk based on the potential erosion and erodibility maps. The risk map corresponds well to field observations on the occurrence of rills and gullies. The model used seems to be applicable to other areas of Lebanon, constituting a tool for soil conservation planning and sustainable management. © 2006 Elsevier B.V. All rights reserved. Keywords: Water–soil erosion; Erosion modelling; Land degradation; GIS; Risk assessment; Lebanon

1. Introduction Soil erosion is a serious geo-environmental issue causing land degradation in sub-humid to arid Mediterranean countries including Lebanon. It causes damages to vulnerable agricultural lands having shallow soils with low organic matter content, water pollution by soil particles and chemicals, and mudflows which may affect urban areas (Poesen and Hooke, 1997). FAO (1986) indicates that erosion rates in the Lebanese mountainous areas reach 50–70 tons ha− 1 year− 1, which far exceed the rate of pedogenesis under the Mediterranean climate. Some agricultural areas have already declined due to soil erosion. It is necessary to ⁎ Corresponding author. Tel.: +961 4 409845/6; fax: +961 4 409847. E-mail address: [email protected] (R. Bou Kheir). 0169-555X/$ - see front matter © 2006 Elsevier B.V. All rights reserved. doi:10.1016/j.geomorph.2006.05.012

establish soil conservation measures which can reduce land degradation and assure a sustainable management of soil resources. The implementation of effective soil conservation measures has to be preceded by a spatially distributed erosion risk assessment (Moussa et al., 2002; Souchère et al., 2005). Previous studies on soil resources in Lebanon have dealt with soil classification and distribution (Darwish and Zurayk, 1997; Darwish et al., 2002) as well as some aspects related to land degradation (Khawlie et al., 2002). However, quantitative studies on erosion processes in Lebanon have been scarce. Although general situations of soil-water erosion in Lebanon have been described (Ryan, 1983; Khawlie, 1991; Zurayk, 1994; Bou Kheir et al., 2001, 2003), mapping erosion risks at a regional scale (e.g., 1:100,000) has not been performed. At a regional scale, large spatial variabilities of

348

R. Bou Kheir et al. / Geomorphology 82 (2006) 347–359

landscape characteristics such as land use, topography, soil and climate inhibit the application of complex models established from localised measurements at small experimental plots (Kirkby et al., 1996; Renschler and Harbor, 2002). Therefore, proper soil erosion mapping based on advanced information techniques including remote sensing (RS) and Geographic Information Systems (GIS) is needed. In Mediterranean Europe, empirical soil erosion models such as the CORINE (CORINE, 1992) and more deterministic models such as PESERA (Gobin and Govers, 2001) have already been used at a regional scale. The CORINE model is based on the USLE (Wischmeier and Smith, 1978) which was not originally developed for regional applications, and it tends to overestimate the effect of slope (Wischmeier and Smith, 1978; Gobin et al., 2003). The PESERA model was elaborated using European data and is thus mostly valid for European soils (Le Bissonnais et al., 2005), but its applicability to other areas is uncertain. It seems necessary to develop a “cognitive” soil erosion model for Lebanon based on factors influencing soil-water erosion. In this context, the objective of this study is to provide a soil erosion risk map of a representative region

of Lebanon and to evaluate its accuracy. Available data from existing maps, satellite images and local expert knowledge are combined based on qualitative decision rules and hierarchical organization of effective parameters in order to define homogeneous response units in terms of the severity of erosion risk. 2. The study area The study area (955 km2; Fig. 1) corresponds to 9% of the total area of Lebanon. It represents the environmental diversity of the country in terms of geology, soil, hydrography, land cover and climate. It extends from west to east Lebanon with three major landform zones: coastal (< 100 m altitude), the Lebanon mountainous chains (100 to > 1500 m) and Bekaa Valley (500–1500 m). Geology of the study area comprises mainly of sedimentary rocks (Jurassic, Cretaceous, Tertiary and Quaternary) with minor occurrence of basaltic rocks (Dubertret, 1945). Thirty-two soil units occur in the study area according to a 1:200,000 soil map (Gèze, 1956). The majority of rain and snow (75–85%) falls between November and April, and the rest

Fig. 1. Location of the study area within Lebanon.

R. Bou Kheir et al. / Geomorphology 82 (2006) 347–359

349

corresponds to autumn storms and spring showers (CAL, 1977).

(erodibility) to produce the final thematic map “erosion risk”.

3. Modelling approach

4. Map construction

The developed “cognitive model” uses qualitative decision rules and hierarchical organization of effective parameters based on the knowledge of experts. This type of approach has already been applied successfully to soil erosion modelling at various scales when detailed input data are unavailable (Harris and Boardman, 1990; Cerdan et al., 2002a,b; Le Bissonnais et al., 2002). The model depends on various maps established from satellite images (Landsat TM) and data of basic environmental parameters such as geology, soil, land cover, slope, drainage density and annual rainfall (Bou Kheir, 2002; Bou Kheir et al., 2004). These parameters are coded based on their sensitivity to water erosion, and some of them are validated at 200 field sites having various characteristics of land cover, lithology and soil. Geographic locations of these sites were determined using a GPS with a precision of ca. 10 m. The combined effects of these parameters on soil are evaluated to produce homogeneous response units called “Erosion Response Units (ERU)”. The mapping of erosion risk is realized in several steps (Fig. 2), where some base maps are used to produce factorial maps, which in turn produce some derived/thematic maps, and finally the risk map. As shown in Fig. 2, the first two derived maps (landscape sensitivity and runoff potential) are combined to define the thematic map “potential sensitivity to erosion”. The thematic map is combined with the third derived map

4.1. Runoff potential Runoff potential is defined as the annual quantity of water which may erode soil. It was calculated from the overlay of mean annual precipitation, soil-water retention capacity and soil/rock infiltration capacity (Fig. 2). The mean annual precipitation data used have 12 classes from 500 to > 1600 mm with an interval of 100 mm (Bou Kheir, 2002). The available rainfall map at a scale of 1:200,000 (Plassard, 1971) shows annual precipitation from 500 to > 1400 mm, and it suggests the important role of topography in determining precipitation. Rainfall increases from the coastal plain until the summits of the Lebanon Mountains and then decreases towards Bekaa Valley. Therefore, we expanded the limits of the precipitation data by Plassard (1971) based on altitude. We digitized the isohyets curves of the rainfall map and compared them with the altitudinal classes (Fig. 1). This resulted in a few additional rainfall classes, up to > 1600 mm. Soil-water retention capacity (SWR) was derived by applying certain classes to soil units shown in the soil map by Gèze (1956) (Table 1). Principal soil characteristics affecting SWR are clay content, stoniness, organic matter content and depth. A soil containing a large quantity of clay and organic matter have high SWR, while high stoniness and low depth lead to low SWR. We determined the three qualitative classes of SWR

Fig. 2. Flow diagram of the modelling approach to elaborate the soil erosion risk map.

350

R. Bou Kheir et al. / Geomorphology 82 (2006) 347–359

Table 1 Classes of soil-water retention, soil infiltration and soil erodibility Soil number and type

Soil-water Soil Erodibility retention infiltration

(1) Coastal sands Low (2) Consolidated dunes Low (3) Rock fall and mass Medium movement deposits (4) Coastal soils Low (5) Discontinuous red soils High (6) Continuous red soils High (7) Brownish soils High (8) Yellow mountainous soils Low (9) Sandy soils Medium (10) Greyish soils (on basalts) High (11) Fluvial alluviums Medium (12) Dejection cones Low (13) Alluviums associated High with red soils (14) Alluviums associated Medium with clear chestnut soils (15) Alluviums associated Medium with dark chestnut soils (16) Mixed soils on marl High and limestone (17) Mixed soils on marl, High limestone and sandstone (18) Mixed soils on marl, High limestone and basalt (19) Rendzinas High (20) Black or grey soils High (21) Dark chestnut soils High (22) Light chestnut soils Medium (23) Coastal sands– Low continuous terra rossa (24) Coastal sands–black Low or grey soils (25) continuous terra rossa– High stones and residual blocks (26) Continuous terra Medium rossa–alluviums associated with light chestnut soils (27) Continuous terra High rossa–rendzinas (28) Yellow mountainous Low soils–rendzinas Medium (29) Alluviums associated with clear chestnut soils– alluviums associated with black or grey soils (30) Light chestnut soils– High black or grey soils (31) Light and dark Medium chestnut soils (32) Yellow mountainous Low soils–rendzinas–sandy soils

High High Medium

High High High

High Low Low Low Medium High Medium High High Medium

High Low Low Medium Medium High Medium High High High

High

High

High

High

Medium

Medium

Medium

High

Medium

Medium

Medium Medium Medium Medium High

Medium Medium High High High

High

High

Low

Low

Medium

High

Low

Low

Medium

Medium

High

High

Medium

High

Medium

High

Medium

High

from field data using the equation by Baize and Girard (1995): SWRðmillimeters of water=meters of soilÞ ¼ tdf

ð1Þ

where t is the texture parameter, d is soil depth (m) and f is the coarse fragment content parameter with five values: 1.0 for < 5% content, 0.8 for 5–35%, 0.6 for 35– 65%, 0.4 for 65–95% and 0.2 for > 95%. Soil samples were collected from the 200 sites; t was estimated based on some published data specifying typical values of this parameter using particle components (sand %, clay % and silt %) (Baize and Girard, 1995). The percentages of these particles were determined in the laboratory using the Bouyoucous method (Ryan et al., 1996). d was measured through a sounding by an auger at each site, and f was determined by visual observations. The calculated values of SWR for the 200 sites were grouped into three classes: (1) high retention (150–200 mm/m), (2) medium retention (100–150 mm/m), and (3) low retention (< 100 mm/m). An error matrix was then established between the modelling results and the reference data provided by Gèze (1956) (Table 2). It indicates a high overall accuracy of 94% (188/200). The user's accuracy, the percentage of sites belonging to a model class correctly corresponds to the reference data is 91–96%, and the producer's accuracy, the percentage of sites belonging to a reference class correctly classified by the model is 88–95%. Rainfall infiltration was determined by combining the maps of infiltration established from soil data (Gèze, 1956; Table 1) and geology (Dubertret, 1945; Table 3) based on certain influence rules (Tables 4 and 5). Deep, sandy soils containing large quantities of stones absorb precipitation quickly, whereas others inhibit infiltration, thereby promoting runoff and accelerating water erosion. Rock infiltration capacity depends on lithology,

Table 2 Validation of the soil-water retention map

Pu = user's precision; Ee = excess error (commission); Pp = producer's precision; Ed = deficit error (omission); Pt = total precision; Grey background shows correct classification.

R. Bou Kheir et al. / Geomorphology 82 (2006) 347–359

351

Table 3 Classes of rock infiltration and erodibility Geologic stage

Lithology

Rock infiltration

Rock erodibility

Quaternary

Alluvium Silts Red soils of decalcification of ramleh Mobile dunes Fixed dunes Brownish soils Cones Gravels Mudflows Rock fall deposits Marl, marly limestone Limestone, marly limestone Marly limestone Chalky and siliceous limestone Marly limestone, limestone Marl Marly limestone Limestone, marly limestone, dolomitic limestone Marly limestone, shale, marl Limestone, dolomitic limestone Clastic limestone, marl, clayey sandstone Basaltic tuff Quartztic calcareous sandstone, clay, silts Basaltic tuff Clastic oolithic limestone, marl Dolomite, limestone, dolomitic limestone Basalt

Low Medium Low High High Low Medium Medium Low Medium Low Medium Medium Medium Medium Low Medium Medium Low High Medium Low Medium Low Medium High Medium

High High High High High High High High High High High Medium Medium High High High Medium Low High Low Medium High High High Medium Low High

Mio-Pliocene Miocene Eocene Upper Cretaceous Middle Cretaceous

Senonian Turonian Cenomanian Albian Upper Aptian Lower Aptian Basalt of Aptian Neocomian–Barremian Basalt of Neocomian–Barremian Portlandian Kimmeridgian Basalt of Kimmeridgian

Lower Cretaceous

Upper Jurassic Middle Jurassic

fracture distribution and karst development. We provided a criterion for creating rock infiltration map based on previous studies (Davis and Deweist, 1966; FAO, 1967; Beydoun, 1977; Abbud and Aker, 1986; Khair et al., 1992). Five infiltration classes are distinguished based on infiltration capacities of soils and rocks: (1) very high, (2) high, (3) medium, (4) low and (5) very low (Table 5). Class 1 (very high infiltration) indicates that rainfall infiltrates very quickly in soil and rock. This condition allows frequent change in soil-water retention. Class 2 (high infiltration) indicates that rainfall quickly infiltrates soil and increases soil-water retention, which will decrease as time goes on because water infiltrates

Table 4 Influence rule for the elaboration of rainfall infiltration map

quickly also in rocks. Class 3 (medium infiltration) indicates that rainfall infiltrates into soil and moves slowly through rock. Class 4 (low infiltration) indicates that rainfall will infiltrate into soil only with a rate at which the infiltration is similar to the retention capacity. The resultant runoff will be inversely proportional to the retention capacity. Class 5 (very low infiltration) covers two situations: (1) water does not infiltrate into soil but underlying rock is permeable, and (2) both soil and rock are impermeable. The mean annual precipitation (P) and the aforementioned infiltration classes were then used to calculate the runoff potential as follows. If soil is impermeable (infiltration class 5), the infiltration of water into rock is not considered since water does not reach them. Then, runoff potential (W) is given by W ¼ P−SWR

Rock infiltration

Soil infiltration High (1)

Medium (2)

Low (3)

High (1) Medium (2) Low (3)

1 3 4

2 3 4

5 5 5

(1–5) = number of classes; 1–very high; 2–high; 3–medium; 4–low; 5–very low.

ð2Þ

If soil is more or less permeable (infiltration classes 1 to 4), the rock infiltration ratio (z: see Table 5) needs to be considered: W ¼ Pð1−zÞ−SWR

ð3Þ

352

R. Bou Kheir et al. / Geomorphology 82 (2006) 347–359

Table 5 Interpretation of water infiltration classes Class of infiltration

1

2

3

Soil Rock z Ratio infiltration (Inf)/SWR

H H 45% Inf ≫ SWR

M H 45% Inf > SWR

H M 30% Inf > SWR

4 M M 30% Inf > SWR

H L 15% Inf = SWR

5 M L 15% Inf = SWR

L H 0% Inf < SWR

L M 0% Inf < SWR

L L 0% Inf = 0

z = percentage of water infiltrated into rocks; Inf = infiltration; SWR = soil-water retention. H–high infiltration; M–medium infiltration; L–low infiltration.

SWR was represented by the highest value for each of the three classes: 200 mm for high retention, 150 mm for medium retention and 100 mm for low retention. The highest value indicates the maximum quantity of rain water that can be retained in a given soil. 4.2. Landscape sensitivity Landscape sensitivity is assumed to be inversely proportional to vegetal cover but directly proportional to slope and drainage density. The types of land cover in the study area, shown in a 1:50,000 map by FAO (1990), were divided into three classes: (1) high vegetal cover (> 95%) mainly with oak trees with persisting leaves (Quercus calliprinos); (2) medium vegetal cover (35–95%) including grasslands, shrub lands, degraded coniferous, forests as well as plantations of lemons or bananas; and (3) low to nil vegetal cover (0–35%) including horticulture/agricultural lands, olive yards, vineyards, deciduous fruit trees, bare rocks, badlands and urban areas. The classes based on FAO (1990) at the 200 sites completely agree with field observations. The use of the only three classes limits the detailed Table 6 Classification of landscape sensitivity

4.3. Erodibility

Slope class

Vegetal cover class

Drainage density class

Landscape sensitivity class

1 (gentle)

1 (high) 2 (medium) 3 (low)

2 (medium)

1

any any 1 or 2 3 2 (medium) 3 (high) 1 (low) 2 or 3 1 2 3 Any Any Any

1 (null) 1 (null) 2 (very low) 3 (low) 2 (very low) 3 (low) 3 (low) 4 (medium) 4 (medium) 5 (high) 6 (very high) 4 (medium) 5 (high) 6 (very high)

2 3

3 (steep)

1 2 3

assessment of the effects of land use on erosion. However, the application of the relatively simple method seems to be suitable for currently available data. Drainage density reflects erosion and runoff processes. Three classes of drainage density were distinguished from a 1:50,000 drainage map using the structural classification approach called OASIS (Bou Kheir et al., 2004): (1) low, (2) medium and (3) high. Three classes of slope (S) were distinguished from a digital elevation model (DEM) with a resolution of 50 m: (1) gentle (< 10°), (2) medium (10°–18°) and (3) steep (> 18°). The three maps described above were combined using GIS (ArcView) to create a map of landscape sensitivity to erosion with six classes (Table 6). Areas with gentle slopes, few drainage lines, and dense vegetation cover were assigned low landscape sensitivity, and areas with steep slopes, dense drainage lines, and sparse vegetation cover were assigned high sensitivity. We assumed that slope and vegetation are major controlling factors of erosion, while drainage density plays a subordinate role (Wischmeier and Smith, 1978).

A simple decision rule was defined to obtain the erodibility of land with three classes (low–1, medium–2 and high–3) from rock/soil erodibility classes shown in Tables 1 and 3 (Table 7). The rule reflects the availability of erodible soil. The quantity of movable Table 7 Decision rule of land erodibility Erodibility of rocks

Erodibility of soils Low (1)

Medium (2)

High (3)

Low (1) Medium (2) High (3)

Low Low Medium

Low Medium High

Medium High High

R. Bou Kheir et al. / Geomorphology 82 (2006) 347–359

353

soil particles will be small when soils are thin and rocks are very hard and weathered very slowly. The quantity of available soil particles will be large when soils are thick and rocks are friable and easily weathered.

Table 9 Decision rule of erosion risk

4.4. Thematic maps

Potential erosion

Potential erosion was defined from interaction between the landscape sensitivity and runoff potential based on a decision rule in Table 8. The rule was determined by our expert judgements and can be summarized as follows:

⁎Class 1 is applied to urban areas. 1–null, 2–very low, 3–low, 4–medium, 5–high, 6–very high.

• Potential erosion is null if the landscape sensitivity is null or if the eroding rains are very low (0–250 mm) irrespective of the classes of the landscape sensitivity. • For the other cases, erosion is potentially very high (class 6) if the eroding rain or the landscape sensitivity is very high, except for the very low landscape sensitivity as it corresponds to flat areas. • For the other cases, the potential sensitivity to erosion for low eroding rain (250–500 mm) is the same for terrain with a very low to low landscape sensitivity and increases by one class for terrain with a medium to high runoff potential. This reflects the non-linear effect of vegetation cover with a certain threshold. • For medium eroding rains (500–750 mm), the potential erosion simply increases by one class with the increase in landscape sensitivity class. • For highly eroding rains (750–1000 mm), the potential erosion increases by one class for landscapes with very low, low, and medium sensitivity, but is the same for landscapes in the medium to high landscape sensitivity classes, because vegetation cover is still present even when slopes are steep. Then, the erosion risk was classified into six groups from the combination of the potential erosion and Table 8 Decision rule of potential erosion Potential erosion

Runoff potential

Landscape sensitivity

1 (0–250 mm) 2 (250–500 mm) 3 (500–750 mm) 4 (750–1000 mm) 5 (>1000 mm)

1 (null)

2

3

4

5

6 (very high)

1 1 1 1 1

1 2 2 3 4

1 2 3 4 6

1 3 4 5 6

1 4 5 5 6

1 6 6 6 6

(1–6) = number of classes; 1–null, 2–very low, 3–low, 4–medium, 5– high, 6–very high.

Erosion risk

Erodibility

1 (null) 2 3 4 5 6 ( very high)

1 (low)

2

3 (high)

2 2 2 4 4 4

2 3 4 5 5 6

3 3 5 6 6 6

erodibility (Table 9). The rule in the table also takes into account the non-linear effect of erodibility on erosion risk. The effect is accentuated for medium classes of potential erosion, but weak for the lowest or highest classes. This is coherent with the fact that erodibility will have less influence when parameters like rainfall, vegetation cover or slope angle have extreme values. The risks are null in urban areas (class 1; not shown in Table 9) since the soil cover is absent. 5. Results 5.1. Runoff potential The high to very high runoff potential (Fig. 3) characterises a large part of the studied region (49%), whereas 24% of the region is covered by the very low to low runoff potential. This indicates a widespread possibility of terrain degradation by erosion if there are materials susceptible to erosion. 5.2. Landscape sensitivity In the produced landscape sensitivity map with six classes (Fig. 4), class 4 (medium sensitivity) covers the largest area (43%), while class 5 (high sensitivity) has the largest number of polygons, indicating a highly dispersed distribution. Classes 1 and 2 (null sensitivity and very low sensitivity) mostly occur in the coastal plain and Bekaa Valley. 5.3. Erodibility In the produced erodibility map (Fig. 5), class 1 (low erodibility) is often associated with “terra rossa” on Kimmeridgian and Cenomanian bedrocks of the Lebanon Mountains and covers the largest area of the studied region (36%). Classes 2 and 3 have nearly equal areas:

354

R. Bou Kheir et al. / Geomorphology 82 (2006) 347–359

Fig. 3. Runoff potential map of the study area.

Fig. 4. Landscape sensitivity map of the study area.

R. Bou Kheir et al. / Geomorphology 82 (2006) 347–359

Fig. 5. Erodibility map of the study area.

Fig. 6. Potential erosion map of the study area.

355

356

R. Bou Kheir et al. / Geomorphology 82 (2006) 347–359

class 2 (medium erodibility; 29%) occurs mainly in Bekaa Valley, and class 3 (high erodibility; 28%) often corresponds to sandy soils.

corresponds to terra rossa on Jurassic and Cenomanian bedrocks covered by oak trees. 6. Discussion

5.4. Potential erosion 6.1. Evaluation of the erosion risk map On the produced potential erosion map (Fig. 6), class 5 (high potential) covers the largest part of the study area (29%), mainly in the coastal zones and in the Lebanon Mountainous chains. It is followed by class 4 (medium potential; 21%), and then by class 6 (very high potential; 16%) which are also found mainly in the Lebanon Mountains. Class 1 (null potential; 14%) disperses in the Lebanon Mountains and Bekaa Valley. 5.5. Erosion risk In the final erosion risk map (Fig. 7), 34.5% of the study area show high to very high erosion risk (classes 5 and 6), while 25.5% show very low to low erosion risk (classes 2 and 3), and 33% have medium risk (class 4). The rest (urban area; class 1) is found mostly in Beirut City and its suburbs. Classes 2 and 3 are mostly located in the coastal plain and Bekaa Valley, while classes 4–6 mainly occur in the Lebanon Mountains. Class 4 often

The validation of the constructed erosion risk map is crucial for this study. However, quantitative measurements of the soil erosion rates and sediment yields have rarely been made in Lebanon due to the lack of monitoring systems. Therefore, we made detailed field measurements of the distribution of small rills and gullies at the 200 sites in the study area, and the results were compared with the erosion risk map. Rills and gullies are the most distinct signs of soil erosion in the study area. Hence, their number was counted and the mean volumes (width, depth and length) were calculated at each field site. Six classes of erosion were distinguished based on the measurement results, i.e., (1) null, (2) very low (15– 30 m3), (3) low (30–45 m3), (4) medium (45–60 m3), (5) high (60–75 m3) and (6) very high (75–90 m3). This classification system is coherent with the six classes of the erosion risk map. A confusion matrix was established

Fig. 7. Erosion risk map of the study area.

R. Bou Kheir et al. / Geomorphology 82 (2006) 347–359

357

Table 10 Error matrix for the modeled erosion risk and field observation on rills and gullies

Pu = user's precision, Ee = excess error (commission), Pp = producer's precision, Ed = deficit error (omission), Pt = total precision. Grey background shows correct classification, Σ = total number of correctly modeled sites.

between the measured erosion classes and the modelled erosion risk classes. The matrix indicates a good overall accuracy of 70% (Table 10). However, some classes contain more errors than others do. The user's accuracy ranges from 50% to 84%, and the producer's accuracy ranges from 23% to 100%. Modelling often overestimate risks while underestimation is rare; this can be considered as a positive point for management planning considerations because the possibility of overlooking actual risk decreases. 6.2. Advantages and problems of the model Our model has defined a map of erosion risk with the six classes for a representative region of Lebanon. Such a map was unavailable in Lebanon, although at this stage the map depends on qualitative and relatively subjective procedures. The map represents the result of modelling from available knowledge and data and can meet scientific needs of researchers as well as socioeconomic demands of decision makers. This model can be extrapolated to the whole country if the functional capacities of GIS are used, because they allow model integration with additional basic and factorial data and code modification in order to analyse the data. At the same time, the intermediate maps of the landscape sensitivity, runoff potential and erodibility can be updated and evaluated to assure precision before computing the final map of erosion risk. This is an important future research topic because it is preferable to introduce more objective procedures to define these maps. The two maps of landscape sensitivity and erodibility are also more dependent on time than the runoff potential map, because the most sensitive factor of the two maps is vegetal cover, which may change drastically due to human activities.

Our consideration of a new factor in the model, i.e., rock infiltration capacity, has allowed a better adaptation to local conditions. However, we encountered many difficulties: (1) the acquisition of data necessary for the model, in particular, the determination of the intensity, duration and frequency of storms; (2) low precision of some old available data such as the 50 m DEM which may not be sufficient to estimate local slopes, and the soil map established by Gèze (1956) with relatively rough evaluations of soils; and (3) the paucity of data on the temporal aspect in the established erosion model, as is usually the case for erosion susceptibility mapping. For example, a rock like limestone of Mount Lebanon can give a high quantity of erodible materials, but they are transported slowly after each rain. In contrast, soils on sandstones are transported with a lower frequency but a larger quantity for each time. A similar difference in relation to lithology can occur when we take into account the penetration of water in rocks. In addition, it may be necessary to incorporate seasonal differences in storm characteristics with the model. 7. Conclusions The established model, for the first time, enabled mapping of erosion risk in a large region of Lebanon at a scale of 1:100,000, based on nine factors issued from available data and expert knowledge. This model may be easily extrapolated to the whole country if the functional capacities of GIS are used. The intermediate maps of landscape sensitivity, runoff potential, erodibility and erosion potential have also been provided through data analysis. The risks are defined with six classes to fit the need of land managers and decision makers. A comparison of the erosion map with observed signs of erosion (rills and gullies) proved the robustness

358

R. Bou Kheir et al. / Geomorphology 82 (2006) 347–359

of the model used. Although the chosen scale of the risk map (1:100,000) seems to be sufficient for estimating the zones susceptible to water erosion to consider strategies for land protection and targeting policy implementation, the map can be improved for more localized erosion assessment when more detailed data sets become available. Acknowledgments This research is a part of a project (2003–2006) of the Remote Sensing Center, the Lebanese National Council for Scientific Research (LNCSR), in cooperation with the “Bureau de Recherches Géologiques et Minières (BRGM-France)” and was funded by “Agence Universitaire de la Francophonie (AUF)”. We acknowledge all responsibilities of LNCSR and AUF, in particular Dr. Mouin Hamzé, the General Secretary of LNCSR. References Abbud, M., Aker, N., 1986. The study of the aquiferous formations of Lebanon through the chemistry of their typical springs. Lebanese Science Bulletin 2 (2), 5–22. Baize, D., Girard, M.-C., 1995. Soil Referential. INRA-AFES, Orléans, France. Beydoun, Z., 1977. Petroleum prospects of Lebanon re-evaluation. American Association of Petroleum Geological Bulletin 61, 43–64. Bou Kheir, R., 2002. Study of soil-water erosion risk by remote sensing and GIS: application to a representative region of Lebanon. PhD thesis, Agronomical National Institute Paris–Grignon, France. 261 pp. Bou Kheir, R., Shaban, A., Girard, M.-C., Khawlie, M., 2001. Impact of human activities on soil-water erosion in the mountainous coastal region of Lebanon. Sécheresse 12 (3), 157–165. Bou Kheir, R., Shaban, A., Girard, M.-C., Khawlie, M., Abdallah, C., 2003. Characterization of karst in Lebanon: sensitivity to soilwater erosion. Sécheresse 14 (4), 1–9. Bou Kheir, R., Girard, M.-C., Khawlie, M., 2004. Use of a structural classification OASIS for the mapping of landscape units in a representative region of Lebanon. Canadian Journal of Remote Sensing 30, 617–630. CAL, 1977. Climatic Atlas of Lebanon. Tome 1. Meteorological Service, Ministry of Public Affairs and Transports, Lebanon. 45 pp. Cerdan, O., Souchère, V., Lecomte, V., Couturier, A., Le Bissonnais, Y., 2002a. Incorporating soil surface crusting processes in an expert-based runoff and erosion model STREAM (Sealing Transfer Runoff Erosion Agricultural Modification). Catena 46, 189–205. Cerdan, O., Le Bissonnais, Y., Couturier, A., Saby, N., 2002b. Modelling interrill erosion in small cultivated catchments. Hydrological Processes 16, 3215–3226. CORINE, 1992. Corine Soil Erosion Risk and Important Land Resources in the Southern Regions of the European Community, Publication EUR 13233 EN, Luxembourg. 97 pp.

Darwish, T., Zurayk, R., 1997. Distribution and nature of red Mediterranean soils in Lebanon along an altitudinal sequence. Catena 28, 191–202. Darwish, T., Khawlie, M., Jomaa, M., Awad, M., Abou Daher, M., Zdruli, P., 2002. A survey to upgrade information for soil mapping and management in Lebanon. Options Méditerranéennes. Séries A: Mediterranean Seminars 50, 57–71. Davis, S., Deweist, R., 1966. Hydrogeology. John Willey and Sons, Chichester. 463 pp. Dubertret, L., 1945. Geological map of Lebanon at a scale of 1:50,000. Sheets of Beirut and Zahlé. Republic of Lebanon, Ministry of Public Affairs, Beirut, Lebanon. FAO, 1967. Hydrogeologic Map of Lebanon at a Scale of 1:200,000. United Nations. FAO, 1986. Conservation and management of soils in the countries under development. Bulletin Pédologique 33, 1–98. FAO, 1990. Land Cover Map of Lebanon at a Scale of 1:50,000. Sheets of Beirut and Zahle. Gèze, B., 1956. Soil Map of Lebanon at a Scale of 1:200,000 Explicative Note. Republic of Lebanon, Ministry of Agriculture. 52 pp. Gobin, A., Govers, G., 2001. PESERA (Pan-European Soil Erosion Assessment). EC Contract No. QLK5-CT-1999-01323, First Interim Report. 69 pp. Gobin, A., Govers, G., Jones, R., Kirkby, M., Kosmas, C., 2003. Assessment and Reporting on Soil Erosion. European Environment Agency, Technical report 94, Copenhagen. 103 pp. Harris, M.T., Boardman, J., 1990. A rule-based expert system approach to predicting waterborne soil erosion. In: Boardman, J., Foster, D.L., Dearing, J.A. (Eds.), Soil Erosion on Agricultural Land. John Wiley & Sons, Chichester, pp. 401–412. Khair, K., Aker, N., Zahrudine, K., 1992. Characterization and Subdivision of Hydrogeological Units of Lebanon. American University of Beirut. 19 pp. Khawlie, M., 1991. Balancing natural resources within environmental systems in Lebanon. Proc. Man and the Environment in Lebanon. InUNESCO, Beirut, Lebanon, pp. 51–84. Khawlie, M., Awad, M., Shaban, A., Bou Kheir, R., Abdallah, C., 2002. Remotely sensed data in the coastal eastern Mediterranean rugged mountainous areas: resolving difficulties for environmental protection. ISPRS Journal of Photogrammetry and Remote Sensing 57, 13–23. Kirkby, M.J., Imeson, A.C., Bergkamp, G., Cammeraat, L.H., 1996. Scaling up processes and models from the field plot to the watershed and regional areas. Journal of Soil and Water Conservation 51, 391–396. Le Bissonnais, Y., Montier, C., Jamagne, M., Daroussin, J., King, D., 2002. Mapping erosion risk for cultivated soil in France. Catena 46, 207–220. Le Bissonnais, Y., Jamagne, M., Lambert, J.J., Le Bas, C., Daroussin, J., King, D., Cerdan, O., Léonard, J., Bresson, L.M., Jones, R., 2005. Pan-European soil crusting and erodibility assessment from the European Soil Geographical Database using pedotransfer rules. Advances in Environmental Monitoring and Modelling 2, 1–15. Moussa, R., Voltz, M., Andrieux, P., 2002. Effects of the spatial organization of agricultural management on the hydrological behaviour of a farmed catchment during flood events. Hydrological Processes 16, 393–412. Plassard, J., 1971. Pluviometric Map of Lebanon at a Scale of 1:200,000. Republic of Lebanon. Ministry of Public Affairs. General Direction of Civil Aviation.

R. Bou Kheir et al. / Geomorphology 82 (2006) 347–359 Poesen, J.W.A., Hooke, J.M., 1997. Erosion, flooding and channel management in Mediterranean environments of southern Europe. Progress in Physical Geography 21, 157–199. Renschler, C.S., Harbor, J., 2002. Soil erosion assessment tools from point to regional scales—the role of geomorphologists in land management research and implementation. Geomorphology 47, 189–209. Ryan, J., 1983. Soil conservation in Lebanon. Journal of Soil and Water Conservation 38, 404–406. Ryan, J., Garabet, S., Harmsen, K., Rashid, A., 1996. A soil and plant analysis manual adapted for the West Asia and North Africa

359

region. International Center for Agricultural Research in the Dry Areas (ICARDA), Aleppo, Syria. 140 pp. Souchère, V., Cerdan, O., Dubreuil, N., Le Bissonnais, Y., King, C., 2005. Modelling the impact of agri-environmental scenarios on overland flow in a cultivated catchment (Normandy, France). Catena 61, 229–240. Wischmeier, W.H., Smith, D.D., 1978. Predicting Rainfall Erosion Losses. A Guide to Conservation Planning. In: USDA, vol. 537. Agriculture Research Service, Washington. 58 pp. Zurayk, R., 1994. Rehabilitating the ancient terraced lands of Lebanon. Journal of Soil and Water Conservation 49, 106–113.