The Réal Collobrier hydrological observatory in south-eastern France,
managed by Irstea since 1966, constitutes a benchmark site for regional
hydro-climatology. Because of the dense network of stream gauges and
rain gauges available, this site provides a unique opportunity to evaluate
long-term hydro-meteorological Mediterranean trends. The main catchment (70 km
The Mediterranean climate is known for its warm, dry summers and high-precipitation events, mainly occurring during autumn (Drobinski et al., 2014). Hence climate variability is higher in this area than in neighbouring regions. In the context of the ongoing climate change, several studies have attempted to detect trends in the hydrological variables. Chaouche et al. (2010) studied trends in the western part of the Mediterranean region over the 1970–2006 period. They mainly found an annual increase in temperature and potential evapotranspiration (PET), while no significant trends were detected for annual precipitation. Several studies focused on the evolution of precipitation extremes over the region (Tramblay et al., 2013; Ribes et al., 2018; Pujol et al., 2007). They concluded that there had been a significant intensification in extreme precipitation events at the regional scale, while at the local scale the evolution was highly variable. Hertig and Tramblay (2017) found a widespread increase in meteorological droughts in the whole Mediterranean basin for the period 1970–2000. Lespinas et al. (2010) suggested that groundwater evolution has a significant impact on drought trends and dynamics in the region.
In the 1960s, the French Ministry of Agriculture decided to create several
experimental catchments in order to improve our knowledge of the
hydrological cycle and its impact on agriculture under various French
climates. One of these watersheds is the Réal Collobrier Catchment,
situated in south-eastern France, 50 km from the sea, at elevations ranging
from 70 to 780 m a.s.l. (Lavabre, 1989; Folton et al., 2012). The whole
watershed (area 70 km
The objectives of this paper are to document hydrological trends over this catchment, including extreme precipitation and droughts. The study also aims to better understand drought mechanisms through precipitation and flow measurements associated with hydrological modelling. First, the physical characteristics of the catchment, the associated dataset and the methods and model used are described. Second, annual and monthly precipitation trends are analysed. Finally, we discuss the 50-year evolution of the main hydrological characteristics and changes in the hydrological processes occurring in the watershed, in relation to the larger Mediterranean context.
The Réal Collobrier research catchment is located in south-eastern France,
at the western end of the Maures mountain range on the Mediterranean coast.
Its surface is 70 km
The Réal Collobrier watershed is representative of the geological formations of the crystalline Provence of the Maures, composed of metamorphic and granitic massifs. The main hydrographic axis is aligned with the large Collobrières east–west fault and the tributaries of the Réal Collobrier are oriented NW–SE, perpendicular to the direction of the rocks. We observe different metamorphic facies, from west to east: micaschists, amphibolites, phyllads and gneiss (Martin, 1972).
The terrains of the crystalline basement of the Maures are characterized by an alteration of the basement with a coarse and thin texture (1–2 m). Permeability is generally quite poor and groundwater supplies are extremely limited. Aquifers in this crystalline zone supply little to the Réal Collobrier tributaries that dry up rapidly in summer.
Forest cover largely dominates the hillside areas. The upstream
sub-catchments therefore have a densely wooded vegetation cover composed of
most of the calcifuge Mediterranean species: cork oaks, holm oaks, heather,
arbutus trees, cistus and chestnut plantations. Scrubland occupies land with
a thin soil cover. Soil thickness ranges from thin skeletal ranker soil in
the small Rimbaud watershed (to the east) to alluvial soil several metres
thick in the low valleys. The urbanization of the catchment is low, with the
exception of the small town of Collobrières (1200 inhabitants in the
1960s, 2000 inhabitants today). The downstream part of the watershed
(downstream of Collobrières) is occupied by a large agricultural plain
planted mainly with non-irrigated vineyards. Because of the weak anthropic
influence, the hydrological cycle of the watershed is very close to the
natural one. The catchment is characterized by a typical Mediterranean
climate with dry summers and high precipitation events, mainly during autumn
(September to December). Due to the orography, the mean precipitation (1055 mm yr
Map of the whole catchment and the five sub-catchments.
The analyses were conducted at the scale of the whole catchment (outlet: Pont de Fer) and five sub-catchments. A total of 6 hydrological stations and 14 rain gauges were used. Available data were calculated at a daily time step from January 1968 to December 2017. The equipment was slightly modified over the period analysed in order to adapt to the evolution of metrological material. Precipitation was measured by tipping-bucket rain gauges uniformly distributed over the entire catchment; their spatial distribution is shown in Fig. 1. The data were recorded on stable sites in terms of measurement protocols and observation conditions (approved equipment, no displacement of rain gauges and preservation of site conditions). To satisfy this condition, two rain gauges were removed from the study (out of 16 measured rainfall sites, only 14 were used). The time series were subject to prior descriptive statistical reviews and contain less than 20 % missing data. The trend tests were applied on 14 rain gauges, but the results of the trend tests were very different from one rain gauge to another. According to Cantat (2004), many direct and indirect rainfall factors induce a certain heterogeneity in rainfall series. He concluded that the spatial and temporal fragmentation of the rainfall series affects the identification of long-term rainfall trends, and that series with too many gaps make reconstruction and analysis difficult. In order to have an overall idea of the rainfall trends in the Réal Collobrier catchment, the tests were therefore applied to the catchment averaged rainfall, at the risk of concealing the trend (if it exists) in the different rainfall signals. According to Louvet et al. (2011) and given the strong density of our rain gauge network, the interpolation method has no impact on the trend diagnostics. Spatial interpolation of the rainfall over the catchment was done with the inverse distance weighted method (IDW), widely used in rainfall interpolation (Yang et al., 2015). IDW is based on the function of the inverse distances and assumes that each input point has a local influence that diminishes with distance (Eckstein, 1989).
The catchment runoffs were computed from water depth measurements with hydrometric equipment (a float). The gauging sections were stable over the period, as they were built in a calibrated structure in order to not be subject to change (erosion, seasonal variations in the growth of aquatic plants). The gauging sections were equipped with triangular or rectangular sharp-crested weirs to reduce uncertainties in the relation of river level and river flow.
Main characteristics of watersheds.
This section describes the main indices used to characterize the hydrologic regime of the catchment. These indices are time series on which statistical tests were applied to determine trends.
We used hydroclimatic indices recommended by the ETCCDI/CRD Climate Change
Indices (Peterson et al., 2001), derived from daily rainfall. These indices
can be separated into four categories: threshold indices, duration indices,
absolute indices and standardized indices. Table 2 provides the detailed
description of these indices. In this table “
Besides analysis of the annual extremes, drought phenomena can be analysed
at the monthly time step through the standardized precipitation index (SPI)
(inter alia McKee et al., 1993; Hayes et al., 2011; Lee et al., 2017;
Javanmard et al., 2017). The SPI is calculated from a continuous daily
precipitation time series accumulated over
Precipitation indices used in this study.
The stationarity analysis of hydrological regimes investigated the evolution of indices calculated from daily discharges, on the six catchments of the Réal Collobrier. These indices (see Table 3) can be divided into five categories: absolute, threshold, duration, low flows and standardized indices.
Hydrological indices used to characterize the hydrological cycle.
The absolute indices used were the annual maximum flow, the mean annual flows and the mean monthly flows. These indices were complemented by other indices to characterize the various aspects of hydrological drought events based on daily discharge.
The characterization of a drought episode usually depends on the definition of a threshold which characterizes the drought duration, volume deficit or similar indices (Yevjevich, 1967; Smathkin and Watkins 1997; Hisdal et al., 2004). This threshold is most often derived from the flow duration curve (FDC, Tallaksen and Van Lanen, 2004), which characterizes catchment runoff variability. The flow duration curve relates the streamflow magnitude (vertical axis) with the exceedance frequency (horizontal axis). The classical method (Beard, 1943) uses the entire series available to build a single curve. However, Vogel and Fennessey (1995a, b) proposed to estimate a median flow duration curve based on yearly curves calculated from the observed series. This method is less sensitive to extreme events and is considered more representative of the mean catchment behaviour. The choice of the threshold (a percentile of the curve) results from a compromise that allows the characterization of the low-flow severity. Then, periods with values below the defined threshold are defined as periods of hydrological drought. The most commonly used percentiles range from 0.7th percentile (Q70) to 0.9th percentile (Q90) (Gregor, 2013). In the present study, we used the 0.8-percentile (Q80), a value used in many drought studies (Laaha et al., 2017).
The low-flow indices were derived using this threshold. Figure 2 shows the
methodology for evaluating the hydrological drought indices. These indices
are as follows:
The annual volume deficit relative to the low-flow threshold (called
“DEF”) (Giuntoli et al., 2013) is the volume between the threshold and
the flows below the threshold. The number of days per year where the flow is below the threshold (low-flow
duration called “LFD”). The seasonality of low flow was studied using the concept of the centre of
mass introduced by Stewart et al. (2005). The
Definition of hydrological drought indices.
The last index to characterize low-flow is the base flow index (BFI). We
used this index to characterize the stationarity of the low-flow sensitivity
of the watersheds. The BFI varies between 0 and 1. A high value of BFI
reflects a high groundwater contribution to discharge, while low values are
typical for catchments with a low influence of groundwater on the total
discharge. It was developed in the UK by the Institute of Hydrology (1980)
and represents the ratio between the base flow (BFlow) and total discharge.
To calculate the base flow, there are number of methods available (Brodie
and Hostetler, 2005). For the basic hydrological separation of the base flow
from the total runoff, the most often used method is the automated
time-based separation, named “local minimum method”, proposed by the
Institute of Hydrology (1980) and detailed by Gustard and Tallaksen (2008).
This method was applied using the code developed by the European Drought
Center (Hisdal et al., 2004). The technique consists in subdividing
the time series into intervals of “
When we compare the two separation processes, the “local minimum method” (Bflow_L) has a less fine smoothing than the filter method (Bflow_F).
The Standardized Streamflow Index (SSI) uses the same principle as the SPI, aggregating streamflow data over the given accumulation periods (Vincente-Serrano et al., 2012; Lorenzo-Lacruz et al., 2013). However, there is no widely adapted probability distribution to fit streamflow data in order to calculate the SSI (Barker et al., 2016). We therefore verified the goodness of fit of the gamma distribution to empirical data for the Réal Collobrier discharge series.
Trend detection tests that indicate possible non-stationarity were applied to all indices. We used two non-parametric trend tests which require only independent data, while parametric trend tests require the assumption of independent and normally distributed data. The first-order autocorrelation coefficient of the time series was calculated for each series and a test applied to detect autocorrelations significantly different from zero (function “acf” from the R package STATS; Venables and Ripley, 2002).
The Mann–Kendall test (MK test) (Mann, 1945; Kendall, 1975) is a rank-based
non-parametric test used to detect significant trends in climatic variables.
It is widely applied to hydro-meteorological time series in different
regions around the world (Douglas et al., 2000; Yue at al., 2002; Tabari et
al., 2011; Jhajharia et al., 2012; Gocic and Trajkovic, 2013; De la Casa and
Ovando, 2016; Gao et al., 2017). This test, based on the ranks of
observations rather than their values, is not affected by the distribution
of the data and it is less sensitive to outliers. The robustness of the test
was validated by several comparison tests conducted by Yue and Wang (2004).
The
The Mann–Kendall test performs poorly when the autocorrelation is high (Yue et al., 2002; Yue and Wang, 2004; Hamed, 2008). Tied values are present in samples of annual streamflow, of the standardized precipitation and of standardized flows. To compute the MK statistic in correlated series, Hamed and Rao (1998) therefore modified the Mann–Kendall test by variance correction. This modified MK test was applied to the data series for which a significant autocorrelation was detected.
The sequential version of the Mann–Kendall test statistic (Sneyres, 1990)
applied on time series detects approximate potential trend turning points in
long-term series (Kumar et al., 2016). Gerstengarbe and Werner (1999) or
Bisai et al. (2014) described the successive steps to be applied in order to
accept or reject the null hypothesis (H0: sample under investigation shows
no beginning of a new trend). To apply this test, we used the seqMK function
of the PHENO R Package (Sneyres, 1990), with a
The Mann–Kendall test is associated with the calculation of Sen's slope or
the Theil–Sen estimator (Theil, 1950; Sen, 1968). This method has been
largely used in order to identify the slope of trend line in a hydrological
time series. Sen's slope (
The estimate of the intercept is then computed by the method recommended by Helsel and Hirsh (2002) using Sen's slope and the median of the variables (Conover, 1980).
Annual rainfall and flow on Pont de Fer catchment.
The analysis of Table 1 and Fig. 1 allows an initial comparison between
catchments. The Pont de Fer watershed is the main catchment, fed by the
other sub-catchments. For the analysis we compared catchments individually
or by groups: e.g. the Rimbaud and Vaubarnier catchments (two small neighbouring
catchments, about 1.5 km
Figure 3 shows the interannual evolution of the mean annual rainfall and the
mean annual discharge at the main outlet of the catchment (Pont de Fer). The
average precipitation is 993 mm yr
The two smaller catchments have very different behaviours. The Rimbaud catchment is characterized by a runoff coefficient of 0.57, due to its special soil characteristics: the massive nature of the rocks (gneiss), the thinness of the surface formation and the morphometric characteristics promote a dynamic hydrological response to rainfall. There is a high contrast between this sub-catchment and the surrounding Vaubarnier sub-catchment (runoff coefficient of 0.38). The Vaubarnier sub-catchment is characterized by a strong inertia due to its geological structure that plays a dominant role in the runoff production. An ensemble of juxtaposed and independent aquifers with a high retention capacity plays a strictly capacitive role, a situation which favours high losses through evapotranspiration from vegetation. This catchment can be characterized by inter-annual regulations. The runoff coefficients of the other three sub-catchments (Maurets, Valescure and Malière) are equivalent, on the order of 30 % to 36 %.
The seasonal rainfall cycle of the catchments is typical of a Mediterranean
catchment (Fig. 4). In general, for all catchments, the June–August period
is very dry, with rainfall of less than 40 mm month
If we compare the smallest catchments, the very particular hydrological behaviour of the Rimbaud sub-catchment described above is clearly marked; its bare soils (gneiss) generate the highest runoff, except in summer when it is the lowest. During this period, the Vaubarnier sub-catchment has the most advantageous restitution capacity with the highest runoff, in relation with its hydrogeological characteristics already mentioned above.
If we compare the contiguous sub-catchments of Maurets and Valescure, which have similar morphometric characteristics, the Maurets sub-catchment is in a sheltered situation and receives slightly less rain than the Valescure sub-catchment. The excess rainfall and lithological conditions determine the differences in hydrological behaviour between the two catchments. Valescure has a higher flow linked to higher rainfall and its lithology is more suitable for flow (gneiss and micaschist). The Maurets sub-catchment composed of phyllads upstream promotes a slightly more effective low-water support than the Valescure sub-catchment.
The Malière sub-catchment of 12.3 km
Mean annual cycle of rainfall at Pont de Fer from 1968 to 2017 (histogram) and mean annual cycle of runoff for the six catchments for the same period (curves).
At the annual scale, the precipitation trend (Sen's estimator) is
For annual maximum
The indices based on daily precipitation above and below the thresholds show a drying trend over the whole 1968–2017 period (Table 4). The wet days' indices (WDx) showed a significant decreasing trend for the thresholds ranging from 2 to 20 mm. The results of the sequential Mann–Kendall test statistic (not shown) showed a breakpoint in 1980. Before this date the indices were stable and afterwards a significant decrease until 2007 was detected. In the case of the dry spell length (DSLx), a significant increasing trend was observed for the 5 and 20 mm thresholds. Before 1975, the index decreased, then a meaningful long-term trend can be identified until 2007, after which it decreased again. No significant trend could be identified from the other indices characterizing the duration or the intensity of rainfall (CWD or CDD). These indices appear to be too sensitive to sampling, since a rainstorm can disturb the calculation of consecutive dry days.
Sen's slope of linear trends and the value of the Mann–Kendall test in
duration and threshold indices of rainfall. Symbols represent significance
level
The Mann–Kendall test and Sen's slope estimator for 1, 3, 6, 12 and 24
monthly SPI showed significant trends only for SPI-1 and SPI-3. For example,
for SPI-1 the slope is
SPI-12 at Pont de Fer.
Yearly, seasonal and low flows of Réal Collobrier's catchments were analysed with the hydrological indices defined in Sect. 3.2.
We investigated the existence of trends in mean annual, maximum annual
(Amax1d) and mean monthly flows. For annual mean flows, all the detected
trends were negative (ranging from
At the seasonal scale, all the catchments presented significant negative trends for spring flows, especially for March and April. Moreover, only the Vaubarnier sub-catchment presented a significant decreasing trend in September and October.
There are differences, especially in the autumn months (November–December). The flow decreases in at least one of these months in the two catchments with slow hydrological processes (Vaubarnier and Maurets), while in the other catchments the flows increase. However, the trends are not significant. The null values of Sen's slope observed for the summer months (July–August) are influenced by the null flows in these months, except for the Vaubarnier catchment which has the highest runoff in summer.
Sen's slope of linear trends and the value of the Mann–Kendall test in
monthly (mm month
Table 6 shows the indices computed in order to detect trends in low flows and their duration. The following appear to be true:
The low-flow period starts earlier for all watersheds, except the Malière
sub-catchment, and this change is significant only in the Rimbaud
sub-catchment. The amplitude of this change is on average 19 d over the
50-year period observed for this catchment. This is particularly the case
for the Rimbaud catchment, whose functioning is strongly linked to rainfall.
The decrease in spring rainfall then leads to earlier low flows. The end of the low-flow period occurs later for all catchments; this shift
is significant for Pont de Fer and Valescure catchments. For five of the six
catchments, the dry period starts earlier and finishes later in the year.
Overall, the low-flow period is longer and the low-water period is centred
on a later date, representing a shift of 1 to 3 weeks over the observation
period. Trends in durations are significant for the two smaller catchments. The
volume deficit of drought (DEF) increases for all catchments and
significantly only for Vaubarnier. The Vaubarnier sub-watershed is the most
strongly affected in terms of low-flow deficit, followed by the Maurets
sub-watershed. These are the two catchments that strongly support the low-flow period (high values of the mean BFI) and are therefore more sensitive
during low-flow periods. Catchments such as Rimbaud, with null flows in
summer, are less impacted in terms of deficit although their period of low water increases, because of low retention capacity. Although the BFI values
are calculated using two different methods, their trend analysis shows
identical results (Table 6). The base flow (BFlow) shows a downwards trend
for all catchments, and is significant for Pont de Fer and Rimbaud
catchments. The low values of mean BFI reflect a low storage capacity (e.g. 0.21 for the
Rimbaud sub-catchment) and higher values reflect a greater storage capacity
(0.367 for the Vaubarnier sub-catchment). The results are mixed, however: a
decreasing trend is observed for three catchments and an increasing trend
for the other three. Only one positive trend is significant (Malière).
Overall, the results show an increase in volume deficits associated with an extension of the low-flow period. The catchments with the smallest area appear to be the most significantly affected.
Sen's slope of linear trends and the value of the Mann–Kendall test in
indices of low flows for each catchment. Symbols represent significance
level
The results of the Mann–Kendall test and Sen's slope estimation for 1, 3, 6, 9, 12 and 24 monthly SFI over the whole period are presented in Table 7.
The Rimbaud and Vaubarnier sub-catchments show a significant decreasing trend for all timescales, except for SFI24 for Vaubarnier. Again, the trends are significant only for the small catchments. In the other catchments, while the trends are also decreasing, they are not significant, except for Pont de Fer for the 1- and 3-month timescales. Significant 1-month deficits (SFI-1) are doubtless related to the March rainfall deficit observed, which also translates into a deficit in March and April flows in most catchments. The sequential Mann–Kendall test statistic detects significant change points in SFI calculated on periods above 6 months. The test indicates a change point in 1980 for all catchments, except the Vaubarnier sub-catchment with a change point in 1989.
Sen's slope of linear trends and the value of the Mann–Kendall test in SFI
indices for each catchment. Symbols represent significance level
The monthly flows show a significant decreasing trend for spring flows for six catchments, in March and April. The Maurets sub-catchment is the only catchment showing a significant decrease in monthly flows for 6 months from February to June (the high-flow period). This persistent decrease in these seasonal flows explains why the only significant decrease in annual flows was observed for this catchment. In addition, the Vaubarnier catchment is the only one to present a significant decrease in September and October flows. For this catchment, the decrease in these monthly flows is confirmed by the significant increase in the low-flow volume deficit and the low-flow duration. Among the other catchments, two of them (Pont de Fer and Rimbaud) show a significant decrease in base flow (BFlow), associated with a significant lengthening of the low-flow period. In this section we try to understand the reasons for this change. These trends and their variability within the whole catchment can be explained by several factors that are detailed below.
A decrease in flow rates is often related to a decrease in rainfall.
However, annual rainfall has not exhibited any trend, as reported in
Chaouche et al. (2010), who did not detect any significant trends in annual
precipitation in the Mediterranean region during 1970–2006. Only a
significant downward trend in March precipitation, with a decrease in wet
days (thresholds between 2 and 20 mm d
Evapotranspiration is a key flux controlling the surface water balance. It is basically determined by the potential evapotranspiration (PET), which represents the climatic demand, and the upper soil wetness, which represents the actual amount of water that can be mobilized. In order to evaluate the climatic demand, temperatures from the nearest Météo-France climatological station in the reference homogenized French climatic network (Le Cannet des Maures) from 1959 to 2011 and the monthly PET calculated by the Penman–Monteith (Allen et al., 1998) formula for the Toulon station from 1959 to 2014 were studied (Table 8).
Sen's slope of linear trends and value of Mann–Kendall test in PET
at Toulon and
At the annual scale, the mean temperature shows significant changes with
To combine PET and soil wetness, the GR4J model (Perrin et al., 2003) was applied over the whole catchment (Pont de Fer) from 1968 to 2015. Tendency tests were applied to the annual time series of the mean reservoir levels that represent the amount of water in the upper soil and showed a decrease, with a particularly low level in 1989 and 2007. In addition, outputs from a reanalysis of soil wetness by the SAFRAN–ISBA–MODCOU model (Habets et al., 2008; Vidal et al., 2010) were tested, showing the same significant decrease in soil wetness. Hence, it is clear that the increase in evaporation demand, combined with a rainfall decrease in March, tends to dry the upper soil, reducing the water supply to the different catchments.
There are common trends between the catchments, such as a greater severity of the low-flow period, but the response differs according to the sub-catchment.
Firstly, the size of the watershed seems to be an explanatory factor for these differences. For example, the smaller the catchment the more sensitive it is to climate variation. This is the case for the Rimbaud and Vaubarnier catchments, although their sensitivity to climate variation is totally different. Secondly, the physiographic characteristics of the catchments (topography, geology, lithology, etc.) imply different sensitivities of the catchments to climatic changes. Thirdly, the water storage capacity of watersheds seems to be an important explanatory factor of the different impacts of climatic variations on flows. The greater the storage capacity of the catchment or the slower the dynamics of the catchment, the less the low flows will be affected by a decrease in rainfall and an increase in temperature. This is explained by the decrease in underground flows and their contribution to base flows. This is the case of the two catchments of Maurets and Vaubarnier, which have the lowest dynamics and are characterized by a higher infiltration capacity.
In order to give more details about the interpretation of trend analyses on
SPI and SFI, a homoscedasticity test was realized by the Bartlett's test
(1937). This test is used to test whether different samples have equal variances.
We verified the homogeneity of variances for SPI and SFI accumulated on 1
month to 24 months (with a
As for the previous results on the SFI, the results of the homoscedasticity test on SFI depend essentially on the hydrological functioning. So, the catchments characterized by a hydrological functioning mainly controlled by precipitation (Rimbaud, Valescure) show a significant increase in SFI variability, for SFI accumulated on scale of 1 month to 6 months. This suggests that these watersheds are more influenced by decreasing precipitation, creating more of a gap between high- and low-water periods. For the catchments characterized by a lower dynamic (Vaubarnier and Mauret), the test indicates a significant decrease in SFI variability for longer accumulation (SFI12 and SFI24). This is probably related to the presence of longer periods of drought over the period P2, reducing the variability of mean flows.
Trends in flows may also be due to changes in land use. As many studies have shown, the evolution of land use over the years can affect the regime of rivers (Hibbert, 1967; Hewlett, 1982; Andréassian, 2004; Brown et al., 2005; Ssegane et al., 2013). These disturbances are generally caused by human actions, such as urbanization, deforestation and agriculture, or by natural causes such as fires. This point was disregarded in our explanations because the Réal Collobrier watershed is a forest catchment area that has been weakly impacted by human activity for the last 50 years. The only urban area (the village of Collobrières) has not extended much and is located downstream of the six sub-catchments studied. Only the Rimbaud catchment was burnt by forest fires in 1990 (80 % of its surface). Studies on the impact of this fire on the flows showed that the flood dynamics were impacted only during a period of 4 years after the fire (Folton et al., 2015; Lavabre et al., 1998, 2000). No impact was observed on the annual flows or low flows in this catchment. These arguments suggest that the physiographic and land cover characteristics of the watershed have remained relatively stable over the last 50 years and cannot be considered an explanatory factor for the flow trends.
The analysis of the 1968–2017 period of the Réal Collobrier catchment data shows a decrease in rainfall in March and a decrease in rainfall duration.
With the observed increase in temperatures and PET, climatic factors show a tendency to decrease the amount of water available for watersheds. This is observed globally in all of the sub-watersheds studied, especially in spring and during the low-flow period.
Hydrological factors modulate the climatic trends among the sub-catchments. Depending on their hydrological functioning, the impact of a climatic change leads to responses that can be highly contrasted. Flows respond not only to the magnitude of changes in rainfall and temperature but also to the timing of these changes. Thus, significant reductions observed in spring rainfall lead to changes in spring, summer or autumn flows, depending on catchment characteristics. For example, there is a direct link between same-season climate and flow changes in the Rimbaud catchment where spring flows show a clear relationship with spring precipitation and spring to summer temperature. This is definitely due to the low storage capacity of this catchment, associated with little “hydrological memory”. Spring flow decreases with precipitation but also with increased temperature associated with a higher evapotranspiration demand. In this catchment, the increase in summer temperature (and potential evapotranspiration) is not associated with a decrease in summer flow (because of large periods of null-flow periods over all the 50-year series), but with a significant lengthening of the low-flow period.
This study shows also that the catchments that usually support low flows are the most sensitive to climatic disturbances. Unlike the Rimbaud catchment, the Vaubarnier catchment reflects a large storage capacity (the higher mean BFI) that delays the climate to low-flow response. So, significant changes in spring precipitation, combined with increased spring and summer temperature and evapotranspiration demand, have influence on significant decreases in spring, summer and autumn flows. As a result, the length of the severe continuous low-flow periods increases significantly, and the volume deficit of drought increases as well.
The Maurets catchment, which also has a slow hydrological process, shows a relatively important range of change in monthly flows, with a significant decrease from February to June. The Valescure catchment, close to the previous one, shows contrast. The significant decrease concerns only the flows for the month of April. And on the Pont de Fer catchment, the outlet of the study area, the trends in the hydrological indices are more smoothed, except for the significant decrease in base flow.
The fine-scale hydrological information obtained thanks to the Réal Collobrier research catchment has enabled us to observe that at this fine scale the impact of a climatic variation in the hydrology is much more variable than the variability imposed by the climate. The rainfall–runoff relationship is complex, and multiple nonlinearities (between flow and climatic variables such as rainfall or temperature) make it difficult to understand and to quantify the influence of the detected change in climate on the water resource. However, this appears to be specific to small watersheds. This has to be considered as an important result too as it underlines the complex functioning at small scales. The response of a particular watershed to climatic change integrates not only the climate input, but also changes in storage, as well as change and variability in the hydrology of the watershed.
The reliable evaluation of trends requires the availability of data in
catchments that are near-natural and unregulated, contain long records
lengths, are active gauged catchments, and have good-quality data. The
Réal Collobrier research catchment is part of the French reference
hydrological observatory (
The measures in the catchments will be continued and enriched in the future according to need. These measures will be completed by modelling to better document climate-related drivers of change. Concerning hydrological functioning and land use changes, reforestation experimentation on the Malière catchment planned in the future will be analysed.
The data on Real Collobier catchment are accessible to everyone as long as the users register on the application and observe the terms of use on Data Base BDOH. The Hydrology Observatory Database (Base de Données des Observatoires en Hydrologie, BDOH) aims at managing, banking and providing hydrological and biochemical data from long-term observatories managed by Irstea or where Irstea is strongly involved.
These data can be downloaded freely for non-commercial purposes at
TM collected the data. FN and LP conceived and designed the analysis. FN performed the calculations and analysed the results. AP and ME contributed to the interpretation of the results. FN, ME, AP wrote and prepared the paper with contribution from all co-authors.
The authors declare that they have no conflict of interest.
This article is part of the special issue “Hydrological cycle in the Mediterranean (ACP/AMT/GMD/HESS/NHESS/OS inter-journal SI)”. It is not associated with a conference.
Data from the SAFRAN–ISBA–MODCOU were provided by François Besson (Météo-France, Direction de la climatologie et des services climatiques). Data from the Cannet des Maures Station (temperature) and Toulon (PET) were provided by Météo-France. We also would like to acknowledge the review of two anonymous reviewers for valuable suggestions.
This paper was edited by Giuseppe Tito Aronica and reviewed by two anonymous referees.