Recent changes in climate, hydrology and sediment load in the Wadi Abd, Algeria (1970–2010)

Abstract. Here we investigate the changes of temperature, precipitation, river runoff and sediment transport in the Wadi Abd in northwest Algeria over a time series of 40 hydrological years (1970–2010). Temperature increased and precipitation decreased with the reduction in rainfall being relatively higher during the rainy season. A shift towards an earlier onset of first rains during summer was also found with cascading effects on hydrology (hydrological regimes, vegetation, etc.) and thus on erosion and sediment yield. During the 1980s, the flow regime shifted from perennial to intermittent with an amplification of the variations of discharge and a modification of the sediment regime with higher and more irregular suspended particulate flux. Sediment flux was shown to almost double every decade from the 1970s to the 2000s. The sediment regime shifted from two equivalent seasons of sediment yield (spring and fall) to a single major season regime. In the 2000s, autumn produced over 4 times more sediment than spring. The enhanced scatter of the C–Q pairs denotes an increase of hysteresis phenomena in the Wadi Abd that is probably related to the change in the hydrologic regime. At the end of the period, due to irregularity of the discharge, the ability of a rating curve to derive suspended sediment concentration from river discharge was poor.


Introduction
Fluvial and estuarine suspended sediment fluxes are changing dramatically under the combined effects of anthropogenic activities and climate change.On a global scale, re-cent changes showed a trend towards increasing land erosion and decreasing fluxes to coastal waters (Walling and Fang, 2003;Vörösmarty et al., 2003;Wang et al., 2006).The sediment flux trapped in regulated basins with reservoirs is higher than 50 % (Vörösmarty et al., 2003).Locally, it can reach more than 60 % after the impoundment of one single dam, like on the Red River (Vinh et al., 2014), and more than 80 % on rivers with many dams (86 % on the Yellow River, Wang et al., 2007; > 95 % on the Ebro River, Durand et al., 2002).Other engineering activities (meander cutoffs, river-training structures, bank revetments, soil erosion controls) also significantly affect sediment fluxes and can participate in the shift from a transport-limited system to a supply-limited system, like on the Missouri-Mississipi River system (Meade and Moody, 2010).
Climate change, through increasing temperatures and evaporation, tends to accelerate the water cycle and modify hydrologic regimes (Bates et al., 2008).Precipitation intensities and the frequency of extreme events are projected to increase under climate change, leading to more frequent flood events of higher magnitude that will, in turn, affect patterns of erosion and deposition within river basins (Tucker and Slingerland, 1997;Pruski and Nearing, 2002;Tockner and Stanford, 2002;Coulthard et al., 2012).Recent studies focused on the impact of climate change on sediment transport (e.g.Gomez et al., 2009;Hancock, 2009;Walling, 2009;Hancock and Coulthard, 2011;Knight and Harrison, 2013;Lu et al., 2013).Syvitski (2003) showed on an example that sediment transport may increase due to the increasing discharge or decrease because of the enhanced tempera-ture.Studies have compared trends in hydrological and sediment time series to land use changes (Wang et al., 2007;Memariam et al., 2012;Gao et al., 2012).Climate projections are consistent regarding warming and the acceleration of the water cycle (IPCC, 2013), however, they remain to be defined on sediment transport where projections show a high uncertainty (Shrestha et al., 2013;Lu et al., 2013).This is in part due to the fact that climate affects many factors controlling sediment yield, such as surface moisture availability, weathering processes and rates, and the nature of riparian vegetation (Nanson et al., 2002).
While sediment transport is well documented in perennial rivers in humid or temperate climates, its study in semiarid areas is still fragmentary due to the difficulty of sampling during flashfloods.Amongst the factors favouring erosion (slope, nature of rocks, relief, climate, human activities), climate is recognized to be the main factor in semi-arid Mediterranean areas of Algeria which experience short and intense rain episodes, high evaporating power of wind, prolonged droughts and freezing and thawing cycles (Touaibia, 2010;Houyou et al., 2014).Erosion is extremely active and the average concentration is at least 1 order of magnitude higher than the global average (Achite and Ouillon, 2007).One of the main impacts of this high erosion is the rapid silting up of reservoirs (up to 2-5 % per year, Kassoul et al., 1997;Remini et al., 2009;Touaibia, 2010) with important consequences on water resource management in a region where 85 % of rain evaporates (Benhamiche et al., 2014).The high temporal variability and recent changes in forcings mean that it is necessary to study sediment dynamics in such environments over time periods of several decades in order to document and understand the changes in sediment regime.Achite and Ouillon (2007 hereafter referred as AO2007) analysed sediment transport changes in the Wadi Abd, an Algerian wadi over a 22-year period (1973-1995).Here we extend this analysis of sediment transport changes to cover a 40-year period .The hydrologic gauging station is located upstream from a dam and is not affected by any major management.This river sub-basin is also particularly suitable for such study because its hydrologic regime was shown to have drastically changed between the 1970s and the 1980s.Precipitation decreased and became more irregular and the flow regime shifted from perennial to intermittent with 26 % dry days, on average, in 1990-1995.Variations of discharge were amplified, and a modified sediment regime occurred with a higher and more irregular suspended particulate flux, that was 4.7 times higher over the period 1985-1995than over 1973-1985. AO2007. AO2007, showing the advantage of working with over 22 years of measurement, however, stressed the difficulty of defining a reference period, and the need to extend the study over a longer period of time.The objectives of this additional study are to (1) describe precipitation, discharge and sediment flux variability of the Wadi Abd basin over a 40-year period; (2) detect the shift, if any, in temperature, runoff and sediment yield; (3) determine the relation-ship between sediment load and runoff over the last 40 years; (4) detect when a shift occurred in the runoff-sediment load relationship; (5) analyse the possible causes of the change in flow regime and its consequences on suspended sediment discharge; and (6) assess the use of rating curves and the physical signification of their parameters when a river is experiencing a transition and shifts from a perennial regime to an intermittent regime.
2 Study area: the Wadi Abd

General information
The Wadi Abd, located in northwest Algeria, is a tributary of the Wadi Cheliff, the principal river of Algeria (Fig. 1).The length of the Wadi Abd's main stream is 118 km, the basin area is 2480 km 2 and the drainage density is 3.70 km km −2 (Fig. 2a).The Wadi Abd supplies the downstream Sidi Mohamed Benaouda (SMB) reservoir which has a basin area of 4900 km 2 .The Wadi Abd catchment area is formed of erodible sedimentary rocks from the Upper Jurassic (45.9 % of its surface), Middle Jurassic (20.2 %) and Pliocene (7.4 %) (Fig. 2b).Soft bottom sedimentary deposits from the Quaternary cover 13 % of the basin along the wadi (Tecsult International, 2004).
The climate is Mediterranean and is characterized by a dry season from April to August/September, and a wet season from September to March.The hydraulic deficit is very high.Annual precipitation is 264 mm, on average, while the mean potential evapotranspiration over the SMB basin is 1525 mm (Tecsult International, 2004).
The watershed mainly consists of steep slopes (Fig. 2c) with very sparse vegetation or bare soil (Fig. 2d).The main land use is natural environment (73 %; 17 % of forests +56 % of scrub and bare steppe soils), cultivated lands cover about 26 % and cities 0.4 %.Seven hill reservoirs were built in the Wadi Abd basin from 1986 to 2004 for agriculture (irrigation and livestock watering) or for fire-fighting measures.Their total cumulated capacity is 0.88 hm 3 , representing 2.3 % of the yearly averaged discharge at Ain Hamara station.These small reservoirs are now silted up to 70 % of their volume.
In 2008, 123 000 inhabitants were living in the Wadi Abd basin (average density: 49 inhabitants km −2 ), 44 % of them living in the city of Takhmaret.The Wadi Abd is thus little influenced by human activities, in view of its extensive surface that is subject to severe natural erosion.
In the plain, sheet (interrill) and rill erosion dominates (Fig. 3b and f).Gully erosion is mainly restricted to the mountainous regions of Frenda and Tiaret in the north (Figs. 3c,d and 2c), while some mid-slope areas are gullying (Fig. 3a and e).

Data
Long-term series of temperature measured at three stations in Algeria were extracted from CRUTEM4 (Jones et al., 2012;Osborn and Jones, 2014).These stations are located at Chlef (36.20 • N, 1.    -1951- -2011), Miliana (36.30 ), Miliana (36.30 -1856-2011).Annual average temperatures were calculated for each station from the 12 monthly averages.The 20 missing monthly data (out of 480) at Chlef, the nearest station from the Wadi Abd, were extrapolated from the monthly temperatures measured at Miliana and Dar El Beida using the relationships between the monthly average temperatures at Chlef and Miliana, and Chlef and Dar El Beida (Fig. 4).The resulting estimates of temperature at Chlef on seasonal and yearly scales allowed us to estimate changes by decade over the period 1970-2010.
Rainfall and hydrometric records were provided by the National Agency of Hydraulic Resources (ANRH).Time series of rainfall data are available at six stations within the basin (see Fig. 2a): S1 Ain Kermes (altitude: 1162 m), S2 Rosfa (960 m), S3 Sidi Youcef (1100 m), S4 Tiricine (1070 m), S5 Takhmaret (655 m) and S6 Ain Hamara (288 m).There were 9076 coincident instantaneous measurements of water discharge (namely Q, in m 3 s −1 ) and suspended sediment concentrations (C, in g L −1 ) recorded at the Ain Hamara gauging station between September 1970 and August 2010.Water depths were measured continuously and a calibration between water level and discharge was regularly performed from velocity profiles.During flow measurements, water was manually sampled once or twice using a 1 L dip at the edge of the wadi.The number of samples was adapted to the flow regime.During baseflow samples were collected every other day, whereas during floods samples were collected at higher rates (up to one every 30 min).Water samples were filtered on preweighed Whatman glass fiber filters (GFF), oven dried at 105 • C for 24 h, and weighed again to determine the concentration.This method, used by ANRH at all hydrologic stations in Algeria, underestimates the suspended load as compared to its value averaged over the cross section under low turbulence (i.e. at low flow) since water is sampling near the surface (Touat, 1989).During floods, which transport most of the sediment load, turbulence is high enough to homogenize suspension load.While this underestimation may slightly affect the budget, it does not severely affect the time variability of suspended matter which is analysed in this paper.From 9076 coincident instantaneous data measured during 1213 days, average arithmetic values were calculated per day so as to obtain 1213 pairs of "mean daily" (C, Q) values.The resulting "mean daily Q" differs from the (true) daily discharge obtained from the averaging of 24 h of continuous instant Q.
The Atlantic Multidecadal Oscillation (AMO) index is an index of North Atlantic temperatures.The monthly unsmoothed values used in this study were calculated by NOAA, Earth System Research Laboratory, Physical Sciences Division/ESRL/PSD1 (http://www.esrl.noaa.gov/psd/data/timeseries/AMO/).

Trends
The analysis of trends was conducted following a method fully described by Stahl et al. (2010) and Déry et al. (2005) for river runoff.The Kendall-Theil robust line furnishes a linear equation from a time series of n measurements such as where t is time (year), y denotes the hydrological parameter (precipitation, river discharge and sediment discharge) and m is the magnitude of the trend over this period.m is calculated as the median of all slopes m k of consecutive pairs of values: where This slope is often referred to as the Sen slope (Sen, 1968).The significance of this trend at a level p was calculated following Ziegler et al. (2003).

Rating curves
C and Q measurements were used to define rating curves that estimate C from measured values of Q, according to a common approach (e.g.Walling, 1977a;Asselman, 2000;El Mahi et al., 2012;Tebbi et al., 2012;Louamri et al., 2013).
The most suitable model is a power law of the type C = a Q b for which the coefficients (a, b) determined empirically account for the effectiveness of erosion and transport.The rating curve established from the 1213 daily averages of C and Q data enabled the estimation of for the whole period 1970-2010 from the measured daily Q values.
Considering the change in hydrologic regime during the study period, we wondered if the estimate of C and Q s per subperiods such as decades would be more appropriate.We therefore applied the four rating curves established for the 4 decades to the time series of daily Q to obtain daily C and then daily Q s .This method (B) enabled us to compare the estimated solid discharge with the value provided by the global relationship established from 40 years of data (method A).The average error for daily Q s values was 51 % using method A and 42.1 % using method B. However, the cumulative flux of suspended matter over the 1213 days for which daily data are available was overestimated by 3.1 % using method A while it was underestimated by 5 % using method B. A comparison of the estimates by these two methods showed that method B is not reliable at high discharge during the last decade because of an increase in scattering of the C, Q pairs.The relationship obtained over the last decade (2000)(2001)(2002)(2003)(2004)(2005)(2006)(2007)(2008)(2009)(2010) led to an underestimation of Q s of 23 % over the 314 days for which daily C and Q are known.In contrast, the global algorithm from method A led to an underestimation of the same cumulated Q s by only 3.5 % over the same period.The relationship established over 40 years was therefore used for this study.
It should be noted that although method A provides some daily solid discharges from the 1213 daily Q values with a high error (the average error being 51 %), it enabled the reconstruction of good trends of Q s values over more than 7 orders of magnitude (Fig. 5).The temporal variability of the coefficients a and b of the rating curves calculated over years or decades will be discussed in light of the variability of the forcings and their consequences on sediment transport in order to better understand their physical meaning.

Average loads
In order to analyse the temporal variability of suspended sediment flux, we use the average concentration resulting from the ratio between the solid and the liquid flow rate, denoted as SPM * , which can be defined for any integration period (day, month, season, year).

Study of breaks: double-mass curve
Double-mass curves were used to determine long-term trends and changes in the hydrosedimentary regime (Searcy and Hardison, 1960;Walling, 1977bWalling, , 2006)).

Interannual variations of temperature, precipitation, river discharge and flow regime
The statistics of hydrological parameters at Ain Hamara gauging station over the period 1970-2010 are reported in Table 1.

Temperature
Temperature in northern Algeria at the three stations of Chlef, Miliana and Dar El Beida increased from the 1970s onwards (Fig. 6), with higher values at Chlef than at Dar El Beida and Miliana.On average, temperature at Chlef increased by 1.17 • C from the 1970s to the 2000s (Table 2).
The increase was 0.87 • C between the 1970s and the 1980s which is more than 4 times the difference between the 1980s and the 1990s (+0.18 • C) and the 1990s and the 2000s (+0.12 • C).As has been shown on a global scale, the decade of the 2000s was the warmest (IPCC, 2013).
Mean annual precipitation (P ) was 264 mm, with a coefficient of variation (CV) of 27 % between 1970-1971 and 2009-2010.The interannual variations of P (Fig. 7) showed trends towards a decrease of rainfall (−1.86 mm yr −1 , on average, over 40 years, p < 0.05).P decreased by 15 % (from 310 to 264 mm) between the 1970s and 2000s (Table 2).A more precise analysis shows that rainfall greatly decreased from the 1970s to the next decade (from 310 to 231 mm, −25 %), then slightly increased in the 2 following decades (see Table 2).
The average precipitation over the six rainfall gauging stations within the basin was 273 mm yr −1 , with consistent variations as compared to the Ain Hamara station.Five out of six stations show a decrease in precipitation between 1970-1985 and 1985-2010, the average deficit being equal to 3.7 %.

River discharge and flow regime
The annual discharge was 1.18 m 3 s −1 , on average, and exhibited a higher interannual variability (CV = 44.4%) than annual precipitation (Table 1).Yearly values showed a trend towards an increase of river flow (+11.3L s −1 yr −1 , on average, over 40 years, p < 0.01; Fig. 7), with decreasing decadal values between the 1970s and the 1980s, then increasing values afterwards, similar to P (Table 2).
Detailed analysis of daily river discharge shows that the river was perennial in the 1970s and then became intermittent during the 1980s (Fig. 8).The driest year occurred in 1993-1994 with 117 days of the fully dry river.In Fig. 8, very low river discharges (around 0.01 m 3 s −1 ) were not considered as days of dry river.
From the 1970s to 2000s, when Q averaged over 10 years increased by 25 %, the wet discharge Q w (i.e. the yearly average discharge of the days of running river) increased by more than 35 % (Table 2).Two indicators of intra-annual discharge variability are shown in Fig. 7: Q 98 , the 98th percentile of annual flows calculated from daily discharge and the standard deviation of daily discharge within each year (σ Q ).Q 98 increased by a factor 3.2 between 1970-1980 and 2000-2010 (Table 2).Q 98 is a good indicator of changes in sediment transport as it occurs during the highest flood events that occur each year.

Rating curve
The rating curve obtained from 1213 pairs of daily averages gave the following: Of the variations of C, 43 % are explained by those of Q.
The rating curve obtained between Q and Q s shows a much higher coefficient of determination (r 2 = 0.831) but is biased since Q s = C × Q.Nevertheless, both relationships give estimates of Q s values from Q with less than 1 % difference which is less than the uncertainty of Q s .

Variability of mean annual load SPM *
The average value of SPM * over the period 1970-2010 is 12.3 g L −1 , with annual values comprised between 2.5 and 50.2 g L −1 (Tables 1 and 2).Their interannual variation was smaller than that of solid discharge because annual SPM * is the ratio of the annual Q s to the annual Q (which increased less than Q s ).

Analysis of break points
The double-mass plot enabled us to identify changes in the sediment response of the stream (Fig. 9).A major break occurred in 1985-1986.A secondary break was noticed in 1991-1992, but the entire period 1985-1986/2009-2010 may be considered as a single period (with the relationship "cum Q s " = 0.021 × "cum Q" − 9.417, r 2 = 0.989).The period 1985-1986/1991-1992 may thus be considered as a transient event towards a new regime.
The response of sediment flow to various constraints differs clearly from that of discharge from the year 1985-1986 onwards.This break corresponds to the first year of dry river over a long period in summer (49 days).This initiates a phase of intermittent flow regime.The averaged parameters for the two periods 1970-1985 and 1985-2010 were added to the tables, in addition to average values throughout the full study period and values for decades to illustrate the dynamics of the hydrological and hydrosedimentary change.
Yearly SPM * and yearly σ Q still showed a strong linearity over 40-year period (r 2 = 0.956, Fig. 10a).A higher correlation was obtained between yearly Q s or SSY, the specific sediment yield, and yearly σ Q (r 2 = 0.991, Fig. 10b).In conclusion, for this river, the yearly solid discharge is more closely dependent on the discharge variability than on discharge values.

Variation of the seasonality of climatic and hydrological parameters
The yearly values of temperature at Chlef generally increased but the monthly averages showed high discrepancies.Temperature from March to November increased with a maximum of increase in June (+3.30presented in Fig. 11.The monthly values of P , Q and Q s per decade over 40 years also clearly illustrate the absolute changes in intensity and in seasonality of the river regime (Fig. 12).The main conclusions of the analysis of P , Q and Q s seasonal variations are the following: -Rainfall decreased in spring and increased in autumn.
Precipitation in autumn increased from 22 to 30 % at the expense of spring rains (decreasing from 41 to 29 %).
For the decade 2000-2010 precipitation was the same in autumn and in spring (78 mm) while for the decade 1970-1980 spring rainfall was 87 % higher than in fall (see Table 3 and Fig. 11a).
-Average monthly rainfall from six weather stations in the river basin for 1970-1985and 1985-2010 (Fig. 13 (Fig. 13) illustrates the changes.Two marked seasons typical of a Mediterranean climate are present (a dry season and a rainy season) but the following changes are observable: (1) differences between seasons decrease, as indicated by the CV of monthly rainfall from 57.3 in 1970-1985 to 45.9 % in 1985-2000 (there is a decrease of spring rains (March-May) and at the beginning of the cold season (November-December), as well as the strengthening of rain in the warm season (July-October) and in winter (January-February)); (2) advancement of the rainy season as evidenced by precipitation in September and October; (3) spreading of the rainy season over 9 months (September-May) for 1985-2010 from previously 7 or 8 months (from October or November onwards); (4) increased regularity of rainy season precipitation.
-Proportionally, flow decreased from winter to summer and increased dramatically in autumn from just over a quarter (27.3 %) of the flow delivered during 1970-1980 to more than one half (52.5 %) during 2000-2010 (Table 3 and Fig. 11b).Flow decreased in summer and the river became dry for much of the summer.Over the last decade, it is striking to see the difference between the average flow rates in fall and spring: the fall rate is almost 3 times that of spring with almost the same rainfall.This trend is evident over the 40-year period (Fig. 11b).-These results point towards a change in runoff as defined by the ratio Q/P .Considering the whole basin area, the river discharge at the Ain Hamara station averaged over 40 years corresponds to a water depth of 15 mm yr −1 , while the average precipitation is 264 mm yr −1 .For comparison, on average, 85 % of rain in this region evaporates and the remaining 15 % runs into surface waters or infiltrates underground storage (Sari, 2009, quoted by Benhamiche et al., 2014).On the Wadi Abd, Q/P averages 5.7 %.We calculated the value of Q/P averaged over 3 consecutive years and over 3 consecutive months (centred) and then took the average per decade (Fig. 14).It appears that the Q/P ratio remains constant during the months from December to April (around 4.4 %, on average), it increased slightly in November and May during the decade 2000-2010 and it increased significantly from September to November.In other words, runoff increased, rain decreased slightly and the temperature (and therefore ETP) increased.As a consequence, infiltration will decrease and the water level in the aquifers will be lowered.) -In absolute values, solid discharge has been increasing in all seasons over 4 decades, but more so in the fall than in the other seasons (Table 3 and Fig. 12c).During autumn, it more than doubled from one decade to another.During the other seasons, it doubled or tripled between the 1970s and 2000s (see Table 2).While during the 1970s the Wadi Abd had two major periods of roughly equivalent sediment discharge in the fall and spring, suspended sediment loads were greater in the autumn during the 2000s (> 70 % of the yearly discharge).
The Wadi shifted from a regime with two equivalent seasons of sediment production to a regime with one dominant season in the 2000s.Autumn produced over 4 times more sediment than spring in the 2000s (Table 3, Fig. 11c).This phenomenon does not seem to be due to some exceptional floods because the trend is observable over 4 consecutive decades (Fig. 11c).6 • C between 1977-1979 and 2000-2006 was noted by Dahmani and Meddi (2009) for the Wadi Fekan basin in west Algeria and Bakreti et al. (2013) also showed a significant trend of increasing temperature in spring by 0.0183 • C per year in the Tafna basin in west Algeria over the same period.However, temperature has not increased as rapidly over the 20th century (Fig. 6) and as mentioned by IPCC (2013), "trends based on short records are very sensitive to the beginning and end dates and do not in general reflect long-term climate trends."The longest available time series of temperature in Algeria was measured at Dar El Beida near Algiers.At this station, average temperature increased by 0.62 • C between 1850-1900(29 yearly values available) and 2003-2012 (Fig. 6) (Fig. 6), while it increased between 1880 and 2012 by 0.85 • C globally (IPCC, 2013).

Hydrol
A global trend towards increasing temperatures and increasing dryness in Algeria from the 1970s onwards has already been described (Meddi and Meddi, 2009).Over the period 1923-2006, north Algeria experienced an alternation of wet periods (1923)(1924)(1925)(1926)(1927)(1928)(1929)(1930)(1931)(1932)(1933)(1934)(1935)(1936)(1937)(1938)(1939) and dry periods (1939-1946 and from 1974 onwards) (Benhamiche et al., 2014).Over 70 years in the Wadi Fekan, Dahmani and Meddi (2009) showed that the period 1943-1960 was rather wet, that 1960-1975 was average, and that the period 1975 onwards (up to the end of their data set in 2004) was dry and of an exceptional long duration.Using three different statistical tests (Pettitt, Lee-Heghinian and Hubert), Meddi and Meddi (2007) shown that a shift was observed between 1973 and 1980 over most of the rain gauges in Algeria.In northwest Algeria, a shift was noticed in 1973 in winter rainfall and between 1974 and 1980 in spring rainfall, both of these shifts being responsible for the yearly rainfall deficit (Meddi and Talia, 2008).From the rainfall data set at the Ain Hamara station between 1968 and 2007, Hallouz et al. (2013) showed that the break in annual rainfall occurred in 1976 and calculated a deficit of 19 % between 1968-1976 (304 mm yr −1 ) and 1976-2007 (247 mm yr −1 ).At the stations Ponteba and Rechaiga, near to the Abd basin, the trends of decreasing total precipitation and of increasing mean length of dry spells were amongst the five highest in the Maghreb area over the 22 stations considered by Tramblay et al. (2013, see their Fig. 8).
As a consequence of the decrease of rainfall after the 1970s break which was observed in most basins of western Algeria, river discharges were generally seen to decrease as well.Meddi and Hubert (2003) showed that the decrease in river discharge varied between −37 and −70 % from eastern Algeria to western Algeria.In the Mecta basin in northwest Algeria, runoff was estimated to be 28-36 % lower in 197628-36 % lower in -200228-36 % lower in as compared to 194928-36 % lower in -197628-36 % lower in (Meddi et al., 2009)).In the Tafna basin, also in northwest Algeria, Ghenim and Megnounif (2013a, b) showed that the decrease in precipitation after the break point was, on average, 29 % over the whole basin (especially in winter and spring) and was accompanied by a decrease of 60 % in river flow.
In contrast, the Wadi Abd behaved differently in that river discharge increased.The counterintuitive increase of runoff with decreasing rainfall has also been observed in the Sahel and is referred to as "the Sahelian paradox" (see Mahé and Paturel, 2009;Mahé et al., 2012).A closer look at the seasonal variations of the different parameters shows that Q decreased in winter and spring but that Q/P increased in autumn when rainfall increased.Overall Q increased.The decrease of rainfall in spring and its low level in summer may have led to a change in vegetation cover which would in turn decrease infiltration.However, although studying the vegetation dynamics of the basin goes beyond the scope of this study, this aspect could be investigated in the future using satellite data, for example.

What is the influence of large-scale circulation indices?
Changes in precipitation are derived from atmosphericoceanic signals (Milliman et al., 2008;Giuntoli et al., 2013).Low-frequency fluctuations related to climate change are modulated with higher-frequency interannual fluctuations, such as ENSO (El Niño-Southern Oscillation), NAO (North Atlantic Oscillation), AMO (Atlantic Multidecadal Oscillation) or MO (Mediterranean Oscillation).Tramblay et al. (2013) showed that the precipitation amounts and the number of dry days over the Maghreb were significantly correlated with the MO and NAO patterns.MO and NAO showed positive trends from the 1970s onwards which could explain the trend towards decreasing frontal conditions over the Mediterranean basin and thus increasing droughts.
Interannual influence by the Austral oscillation ENSO over Algeria was shown to be higher in northwest Algeria on the highest discharges than on the average discharge.The maximum Q seems to be smaller during El Niño and higher during La Niña in northwest Algeria (Ward et al., 2014).The frequency of extreme rainfall events shows the highest correlation with the Mediterranean Oscillation index in Algiers and with the Southern Oscillation index in Oran (Taibi et al., 2014).
In this study, no significant correlation was established between a series of hydrological parameters in the Wadi Abd and the Southern Oscillation index.The average of AMO per hydrologic year was calculated from its monthly values.AMO has increased from 1970s to the 2000s, with negative values until 1993-1994, then positive values thereafter (except in 1996-1997).Its decadal average was −0.25 in the 1970s, −0.12 in the 1980s, 0.0 in the 1990s and 0.18 in the 2000s.AMO and the discharge variability of the Wadi Abd within the year increased coincidently.The yearly AMO values have a coefficient of determination of 0.226 when correlated with the standard deviation of daily river discharges within the year, a proxy for the variability of daily discharge.However, this information does not allow us to conclude that the Atlantic Multidecadal Oscillation is responsible for hydrological changes in the Wadi Abd basin.

Break point in 1985-1986: change of flow regime
The several weeks of dry river for the first time in 1985-1986 (49 days) can be considered as a threshold effect, which marks the start of a new flow regime.The appearance of a dry regime is a break, a fully nonlinear phenomenon.It has strong consequences for water infiltration and groundwater recharge, on the seasonality, intensity and type of floods, and in turn, on erosion and sediment transport.The year 1985 is also a pivotal year for recent climate change as evidenced by the rapid increase in global mean air temperature anomaly from that year until 1993 (Fig. 1 in Lockwood and Fröhlich, 2007).The hypothesis of a temporary warming caused by dust emitted during the eruption of Mount Pinatubo had been advanced to explain the warming since 1985, but climate scientists later recognized that the temperature anomaly has been increasing since 1993, reaching about 0.6 • C by 2007 compared to the global average temperature calculated for the period 1951-1980(Lockwood and Fröhlich, 2007).
This threshold is coincident with hydrological shifts in the Tafna basin in northwest Algeria.Bakreti et al. (2013) analysed the baseflow and baseflow index of five of the Tafna's sub-basins between 1976 and 2006 and found ruptures of the baseflow index between 1984 and 1990 depending on the sub-basin, in 1984, 1985 and 1990 in the mountains, and in 1985 and 1986 in the plain.These changes in flow regimes in the Tafna basin were likely caused by shifts in rainfall in the late 1970s in the Mounts of Tlemcen and early 1980s in the plains (Ghenim and Megnounif, 2013a).

Shift of the onset of the first summer flood
The analysis of the time series of daily flows enables the determination of the start of the first summer flood.The average daily flow per decade suddenly increases the day at which the first summer flood occurred, at least once in the decade.By observing these decadal averaged daily flows, there is no ambiguity on the start of the earliest flood by decade: -In 1970-1980, the first flood starts on 6 September with an average 4-day discharge (6-9 September) of 1.59 m 3 s −1 , while it was, on average, 0.58 m 3 s −1 over the 4 previous days.
-In 2000-2010, the first flood of summer starts on 8 August with an average 4-day discharge (8-11 August) of 2.03 m 3 s −1 , while it was, on average, 0.03 m 3 s −1 from 4 to 7 August.
During the 2000s, the first flood in summer started close to 1 month before that of the 1970s and the magnitude was 27 % higher.It can be asked if this trend was observable over the 40-year period or only between 2 specific decades.The analysis of mean dates and discharges of the first flood in the late dry season gave the following results for the intermediate decades: -In 1980-1990, the first flood started, on average, on 31 August with a 4-day average discharge (31 August-3 September) of 2.69 m 3 s −1 , while the average rate over the 4 previous days was 0.13 m 3 s −1 .
-In 1990-2000, the first flood started, on average, on 22 August with a 4-day average discharge (22-25 August) of 7.67 m 3 s −1 , while the average rate over the 4 previous days was 0.33 m 3 s −1 .The existence of a precursor peak on 17 August, which was not observed in previous decades, was also observed.
It therefore appears that the date of the first flood advanced by about 10 days each decade over the previous 40 years.
The shift in the onset of the first flood in summer probably has important consequences on flow and erosion rates.7.2 Relationships between several parameters and sediment yield

Temperature and sediment yield
The curve showing annual suspended load versus the global air temperature anomaly (base period 1951-1980) calculated by hydrological year from monthly data provided by NOAA (Hansen et al., 2010;GISTEMP Team, 2015) shows a correlation between the sediment yield and ongoing climate change (r 2 = 0.388, Fig. 15).

Precipitation and sediment yield
Many authors studied the variations of sediment load per unit of catchment area against annual rainfall (e.g.Summerfield and Hulton, 1994) or effective rainfall (e.g.Langbein and Schumm, 1958).On the Wadi Abd, annual rainfall fell sharply between the 1970s and the 1980s, then slightly increased over the following decades.Meanwhile, yearly sediment concentration and suspended sediment discharge have increased.The comparison of their respective variations shows a lack of correlation between precipitation and annual sediment yield (r 2 < 0.1, regardless of the type of regression considered).Regarding the relationship between precipitation and erosion, if there are correlations between their spatial variations reported in the literature (though with a strong scatter, see Riebe et al., 2001), our study shows that the temporal variations of precipitation and sediment yield are not correlated in the Wadi Abd.This may be due to the change of flow regime within the study period.

Runoff and sediment yield
Although runoff was noted to have a limited impact on the distribution of sediment yield at regional or global scales by Aalto et al. (2006), Syvitski and Milliman (2007) and Vanmaercke et al. (2014), the temporal variability in precipita-tion, runoff (or discharge) and consecutive vegetation cover was shown to be locally the main impact on fluvial sediment load (see Vanmaercke et al., 2014, p. 360).On the Wadi Abd, the yearly suspended sediment load was highly correlated with discharge (Q mean or its highest percentiles) and to its intra-annual fluctuation (Fig. 10).Although the river regime shift clearly impacted several parameters, the relationship between yearly sediment load and discharge variability did not change over the 40-year study period.

On the use of double-mass curves to determine the climate change and anthropogenic influences
Double-mass curves are often used to determine the impact of developments such as dams on sediment discharge (e.g.Lu et al., 2013).Our findings warn about extrapolations that could be wrongly made to quantify the impact of a development by extending the double-mass curves.Indeed, this study shows that the double-mass curve can change its slope (here increasing) when the flow regime change is driven by seasonal temporal variation in precipitation and runoff that is not linked to any specific anthropogenic activity (such as a dam impoundment) within the basin.
thus b varies almost like 1/a (Asselman, 2000).Many papers discuss the physical meaning of the rating parameters a and b (see AO2007) and try to connect their values to physiographical characteristics, vegetation cover or hydrometeorological forcing.The river's regime change is accompanied by a change in the (a, b) pairs of rating curves defined for multi-year periods such that a increases and b decreases (Table 2) Equation ( 5a) is very similar to that presented by Iadanza and Napolitano (2006) for the Tiber River after the construction of a dam (b = − 0.3815 ln a + 0.7794, r 2 = 0.992).Before the construction of this dam, another relationship corresponded to more than 3 times higher sediment yields.Asselman (2000) has suggested interpreting the regression lines in a ln a − b graph as different sediment transport regimes.On the Wadi Abd, the change in sediment transport regime is not evident from the yearly (a, b) values but it becomes clearly observable when considering a and b values averaged over moving periods of several years.The best correlations were obtained for running averages over 15 years named a 15 and b 15 (N = 25, from 1970-1985 to 1995-2010, see Fig. 16).The available data set does not allow us to determine if results obtained from averaging over longer periods would perform best.
The time evolution of the moving average pair (a 15 , b 15 ) clearly shows a first relationship with the values dominated by the pre-1985 regime (8 values from 1970-1985 to 1977-1991), another one for the values predominantly after 1990 (12 values from 1983-1997 to 1995-2010), both with a 15 increasing and b 15 decreasing, and a transitional regime centred on the period 1985-1990 (Fig. 16).During the transition period centred over 1985-1990, b 15 was almost constant (between 0.72 and 0.74) while a 15 was increasing from 2.01 to 2.34.During the period 1985-1991, the yearly values of b varied very little (between 0.653 and 0.672) while yearly a increased significantly from 1.81 in 1985-1986 to 3.23 in 1990-1991.Higher a and lower b values are in the literature typical of highly arid river basins, such as the ephemeral Nahal Eshtemoa in Israel, where a = 16.98 and b = 0.43 (Alexandrov et al., 2003).
As the break points were coincident, it is possible to analyse the change (of a 15 and b 15 ) in terms of shift of hydrological regime.However, if the new hydrological regime was immediate from 1985 onwards, the change in the C-Q relationship was only evidenced in the Wadi Abd at midterm, considering 15-year average values.

Parameters that explain a (or b)
The coefficient of determination between a and SSY is low at the annual scale but higher when we consider the moving averages of a and SSY over 15 years.The SSY explained 95.2 % of the variance in the interannual scale (Fig. 17 showed a lower correlation with the SSY (r 2 = 0.853) than a 15 did.
In summary, the moving average of a is strongly correlated to specific sediment yield over the same moving period of 15 years, and the moving average of b can be deduced from a using the relationship which is given in Fig. 16 as a function of flow regime, either perennial or intermittent.

Validity range of rating curves
The estimation of sediment yield from flow measurements and a rating curve is still acceptable throughout the study period (Fig. 5).However, the pairs (C, Q) become increasingly scattered with time around the best-fit curve, as attested by the decrease of the coefficient of determination from one decade to another (Table 2).
Intermittent flows induce a stronger dependency of river behaviour on antecedent wetness (Beven, 2002) and antecedent weathering, i.e. a strong dependency on memory through threshold and hysteresis effects.With increasing memory effects, coincident values of C and Q become less dependent on each other and the rating curves less suitable to model their relation.The study of sediment dynamics in the Wadi Abd in the future will thus likely require a more appropriate method than rating curves, such as the study of each individual flood, like Megnounif et al. ( 2013) did in the Wadi Sebdou.This finding may have consequences on water management as well.When dealing with rating curves, water discharge must be recorded at frequent intervals, although measurements of concentration can be sparser.When rating curves cannot be applied, river discharge and sediment concentration should be both frequently and simultaneously measured.

Conclusions
In response to climate change which resulted in an increase in temperature of around 1.1 • C between the 1970s and 2000s at Chlef, rainfall moved forward during the late warm season and the watershed of Wadi Abd experienced a significant change in flow regime and an increased variability at both the inter-annual and intra-annual levels.These changes ultimately led to a dramatic and continuous increase in sediment load over 4 decades (on average, 84 % more every decade as compared to the previous one).
The main result of our analysis is the shift of the onset of the first summer flood that occurred 1 month earlier in the 2000s than in the 1970s.This shift is likely responsible for the cascading effects on the hydrological regime of the Wadi Abd.In particular, earlier floods during the warmer season have higher evaporation which limits the groundwater storage.A parallel study of seasonal changes in vegetation cover is needed to provide additional information.
The increase in erosion of the watershed (coefficient a) is accompanied by a decrease in the coefficient b.The traditional rating curves approach, which was applicable when the river was perennial, is now less adapted to model the behaviour of the river (Table 2).This could be explained by a more pronounced hysteresis phenomenon, which is consistent with the change of hydrological regime in the dry season thereby limiting the utility of rating curves to model C-Q relationships.
The rapid change in sediment regime which is instantaneously driven by the changing flow regime should be distinguished from the slow change in the concentration-flow relationship.The change in flow regime can be precisely dated in May-July 1986 (with 49 consecutive dry days), while the change in the C-Q relationship requires averaging over several years of a, b and specific sediment yield to become evident.Such inertial effect may be attributed to the time for the basin soil properties (such as humidity) or vegetation to adapt to the new climate conditions.It likely depends, amongst other factors, on underground water storage, and thus on basin lithology and infiltration history.Over the Wadi Abd basin, the time needed for the flow regime to change after the dryness settlement in early 1970s (see Fig. 6) is estimated as being around 15 years in this study.
The present analysis only includes hydrological parameters.Management programs that were conducted to fight erosion in Algeria from the 1960s until the 1990s by reforesting and setting up banks over cultivated marl and clay areas proved to be little or not efficient (Touaibia, 2010).Human activities may have influenced the hydrological regime change and increased erosion, in particular through firewood cutting during economically difficult periods (1990s), however the shift was shown to occur earlier.The lack of data on land use and land cover changes over 40 years does not allow us to isolate the factors directly related to climate change from those related to other anthropogenic activities.How-ever, the small population, the low coverage of pasture (see Fig. 2d), cultivated areas and vegetation (43 %) in the basin and the small volume of reservoirs (nominally 2.3 % of the annual discharge, and silted up to 70 %) make us think that in this system the effects of climate change dominate anthropogenic effects.The quantification of forcing changes on sediment sources (raindrop erosion, sheet erosion, rill erosion, gully erosion and stream channel erosion) may be investigated in situ (e.g.Poesen et al., 2003) and/or estimated using a numerical model of the hydrologic and sedimentological functioning of the basin, such as WEPP (Nearing et al., 1989), EUROSEM (Morgan et al., 1998) or SWAT (Neitsch et al., 2011).Such a model could help us to test hypotheses and quantify or at least estimate the effects of different forcing changes (temperature, runoff, vegetation, etc.) in future studies.
It is important to emphasize that it is impossible to define long-term hydrological averages in the context of a changing flow regime.The example of the Wadi Abd shows that the difficulty is challenging with regard to sediment transport in suspension, since changes of flux cannot be counted as a fraction but can reach an order of magnitude.
Changes in flow regime in relation to climate change can be investigated using climate models.Das et al. (2013), using 16 climate projections, showed that more intense floods of a return period 2-50 years should occur in the Sierra Nevada, regardless of the rainfall variation.The recent changes in the Wadi Abd show that extreme events with increasing variability already occur in the basin.Over Algeria, an increase of 1-2 • C in temperature could induce a reduction of 10 % in precipitation before the end of the 21st century (Benhamiche et al., 2014) with unknown consequences on erosion and sediment transport.Lu et al. (2013) calculated the impact on sediment loads of every 1 % change in precipitation or river discharge in large Chinese rivers.Such a calculation has no meaning in our basin since the rainfall and discharge were not monotonic (severe decrease in the 1970s then slight increase during 30 years) while the sediment loads have always increased.The difficulty of forecasting climate changedriven impacts on sediment yield due to nonlinear effects has been underlined by geomorphologists (see Goudie, 2006;Jerolmack and Paola, 2010;Coulthard et al., 2012;Knight and Harrison, 2013).The present study illustrates that the change in flow regime induced a fully nonlinear effect between river discharge and sediment yield.This needs be considered in forecasts especially in small river basins in semiarid areas.
Changes in erosion and sediment transport under new climate constraints will induce changes from the middle to long term that decision-makers must integrate into water resources management, habitat status, agricultural adaptation (O'Neal et al., 2005), landscape evolution (Temme and Veldkamp, 2009) as well as in many other environmental adaptations (Ouillon, 1998).We thus encourage the local adaptation of sampling strategies and measurements to take into account changing in flow regimes.Furthermore, due to the uncertainty of water resources and erosion in the Maghreb (Taabni and El Jihad, 2012) and in the Mediterranean basin (Nunes et al., 2008), we also encourage the development of studies on long-term sediment transport in north African basins, in connection with changes in forcing factors.

Figure 1 .
Figure 1.Location of the Wadi Abd sub-basin within the Mina and Cheliff basins, and the other main basins of Algeria.

Figure 2 .
Figure 2. The Wadi Abd catchment area.(a) displays rain and hydrometric stations including HS1 at Takhmaret and HS2 at Ain Hamara, (b) geology, (c) slopes from the digital elevation model of north Algeria, (d) vegetation cover from Landsat ETM+ data of 2009.

Figure 5 .
Figure 5.Comparison between estimates of Q s obtained from Q and the global rating curve, and measured Q s .

Figure 6 .
Figure 6.Interannual variations of mean yearly temperature (calculated from September to August monthly temperatures) at three stations in northern Algeria: Dar El Beida, Miliana and Chlef (from measurements of CRUTEM4 only, extrapolated values are not shown).

Figure 7 .Figure 8 .
Figure 7. Interannual variations of annual precipitation, water discharge and sediment yield at Ain Hamara station.

Figure 9 .
Figure 9. Double-mass plot of sediment yield versus water flow.

Figure 10 .
Figure 10.Yearly average of related sediment load parameters vs. intra-annual variability of daily river discharge, characterized by their annual standard deviation.(a) displays SPM * , (b) specific sediment yield.

Figure 11 .
Figure 11.Trends of the seasonal indexes of precipitation (a), discharge (b) and sediment discharge (c) in the Wadi Abd basin.

Figure 12 .
Figure 12.Monthly values of precipitation (a), Q (b) and Q s (c) averaged over decades in the Wadi Abd basin.

Figure 14 .
Figure 14.Monthly values of the ratio Q/P averaged over decades.

Figure 15 .
Figure 15.Variations of SPM * against the global mean temperature anomaly.

7. 4
Physical meaning of rating parameters a and b 7.4.1 Interannual variation of (a, b) Since C = a Q b , with b = 0, then C(1) = a. a thus represents the sediment concentration when the river discharge is 1 m 3 s −1 , and b reflects the sensitivity of concentration to discharge variation.The general formula ln C = ln (a Q b ) provides:

Figure 16 .
Figure 16.Relationship between the rating curves parameters averaged over 15 years.

Figure 17 .
Figure 17.Relationship between the rating curve parameter a averaged over 15 years and the averaged values of specific sediment yield over 15 years.

Table 1 .
General statistics on the yearly averages of hydrologic parameters from the Wadi Abd at Ain Hamara gauging station over the period 1970-2010 (note: T at Chlef was estimated from measurements at Dar El Beida and Miliana for 20 months over 480 months). 1

Table 2 .
General statistics on the hydrologic parameters (averages) from the Wadi Abd at Ain Hamara gauging station per decade and for the entire period from 1970 to 2010 (note: T at Chlef was estimated from measurements at Dar El Beida and/or Miliana for 20 months with missing values over 480 months).Italic numbers refer either to the full 40-year period or to the two subperiods of perennial regime

Hydrol. Earth Syst. Sci., 20, 1355-1372, 2016 www
• C, on average, between the 1970s and 2000s), it remained quite constant in December and February and decreased by 0.98 • C in January over the same period.Considering the average values per season, winter values (December-February) decreased by 0.33 • C between the 1970s and the 2000s, while spring values (March-May) increased by 1.66 • C, summer values (June-August) by 2.22 • C and fall values (September-November) by 1.29 • C.In summary, annual temperature differences increased with minimum temperatures down slightly and maximum temperatures rising sharply.The increase was most marked in July-August.Averaged seasonal values of P , Q and Q s for each decade are given in absolute values and in percent of the yearly values in Table3.The seasonal relative contribution of P , Q and Q s centred and averaged over 9 consecutive years are .hydrol-earth-syst-sci.net/20/1355/2016/

Table 3 .
Variation of precipitation, water discharge and sediment yield averaged per season over each decade.