Daily evaporation is manually measured using a 1 m² Colorado-type evaporation pan, which is labelled Ep in Fig. 1. Every day around 08:00 a.m. UTC, an operator measures the volume of water that needs to be added or removed to return to a reference level in the tank of the evaporation pan.

Daily evaporation amount (in mm) is then evaluated as the difference between the volume of water measured each day in the evaporation pan and the daily rainfall measured locally in Pm (see Fig. 1). Measurement accuracy is estimated at 0.5 mm.

By default, measurements are qualified with a RAW source code and a F3 quality code. On the infrequent days when the operator has been unable to take the measurement, a value of NA is indicated with an IND source code and an IND quality code. The daily evaporation value for the day following a period with NA is indicated with a RAW source code and a NF quality code, and a comment is added to clarify that it is a cumulative value.

The measurement of rainfall began in 1994 using a tipping bucket rain gauge (tipping threshold: 0.5 mm), and a totalisator rain gauge, which are labelled I2 and Pm, respectively, in Fig. 1. The measuring equipment was subsequently supplemented in 2003 by a series of 2 additional tipping bucket rain gauges (I1 and I3, with tipping threshold of 0.5 and 0.2 mm, respectively) and a series of 7 additional totalisator rain gauges within the catchment area (i.e., P1; P2; P5; P6; P7; P12 and P13). The model is “SPIEA, FR” for totalisator rain gauges and “PRECIS MECANIQUE #3029, FR” for tipping bucket rain gauges. All the rain gauges have a receiving cone surface of 400 cm². Data for tipping bucket rain gauges are stored in a logger system that records tipping times. Measures of standard totalisator rain gauges are made daily at around 09:00 a.m. local time (i.e., 08:00 a.m. UTC) by an operator, providing a daily rainfall measurement with an accuracy of 0.25 mm up to 10 mm and 0.5 mm for larger daily amounts. All tipping bucket rain gauges are checked frequently and cleaned if necessary to prevent clogging.

By default, all daily rainfall amounts recorded by the operator are qualified with a RAW source code and considered as reliable (F3 quality code). A checking procedure consisting of visually comparing the amount of daily rainfall between all the other rain gauges enables anomalies, such as reading or data entry errors, to be identified manually. Data detected as anomalous is given an NF quality code. The daily rainfall value for the periods when a totalisator rain gauge is not operational due to vandalism, for example, is indicated as NA, with an IND quality code. On the infrequent days when the operator has been unable to take the measurement, a value of NA is indicated with an IND quality code. The daily rainfall amount for the day following a period with NA is indicated with a RAW source code and a NF quality code, and a comment indicating that the value corresponds to a cumulative value is added.

By default, all high-frequency rainfall records are qualified with a RAW source code and a F3 quality code. The measurements corresponding to the calibration tests are first removed from the time series. A two-step checking procedure is then applied to identify periods of failure. The first step consists in comparing the daily accumulated values derived from the rain gauge tipping buckets with the daily rainfall amounts derived from all rain gauges. The second step consists in comparing the 5 min rainfall intensities between the tipping bucket rain gauges to identify failures such as partial or complete clogging. For periods of complete failure, 2 consecutive records with NA as values and IND as source and quality codes are entered in the measurement file to the start and end of the failure. For periods of malfunction, the raw recordings are kept and the NF quality code is assigned. Note that every malfunction identified by the operator during his recurring visits is also recorded in the field notebook, enabling particular attention to be paid to the quality of the data concerned.

A high-frequency homogenized time series and a daily homogenized time series of rainfall measurements over the entire 1994–2023 period are provided in addition to the previous individual daily or high-frequency rainfall time series.

The homogenized daily rainfall time series (from 08:00 a.m. to 08:00 a.m. UTC the following day) is produced using the average of all reliable daily rainfall records provided from the eight totalisator rain gauges and calculated from the three tipping-bucket rain gauges. On the rare days when the operator was unable to take the measurement, the cumulative daily rainfall recorded on the totalisator rain gauge on the day following the NA period was first distributed over the days without recording using the daily rainfall distribution derived from tipping bucket rain gauges. The daily homogenized time series is qualified with a CALC source code and a F3 quality code, except for the period prior to 1 September 2004 for which an IND quality code is assigned because there were not enough rain gauges to guarantee reliable daily rainfall value.

The homogenized rainfall time series is compiled from the three rainfall tipping bucket rain gauges, retaining only reliable values, and applying the following order of priority from 2003 onwards: I1 > I2 > I3. This choice is based on the fact that the I1 rain gauge is slightly more centrally located in the catchment area than the two others. The homogenized time series is qualified with a RAW source code and a F3 quality code, except for the period prior to 1 September 2004 for which an IND quality code is assigned because there was only one operational tipping bucket rain gauge in the catchment.

Each hydrosedimentary station has been designed to measure water levels at two points, one suitable for low flows (LQ) and the other for high flows (HQ), as illustrated in Fig. 2. For the PAR station, the gauging system is a Venturi flume for HQ and a V-notch weir for LQ. For the MBV station, the gauging system is a rectangular flume for HQ and a V-notch weir for LQ. For the GBV station, the gauging system is a rectangular flume for HQ and a smaller rectangular flume for LQ. Water levels for each station are recorded using digital piezoresistive pressure transducers (CS451, Campbell Scientific, UK) at these two measurement points. Measurements are taken with a time step of 1 min and recorded when height variations exceed 5 mm or systematically every 30 min. Water level sensors are accurate to within a few millimetres. Each measuring point is equipped with a limnimetric scale, i.e., a metal ruler, so that an operator can manually read the water level on a daily basis, or even more frequently in the event of flooding.

The time series of raw water levels are first compared with the scales readings in order to identify any biases. Biases are mainly induced by changes in the vertical position of the digital sensors during cleaning operations, for example. A piecewise bias-correction is then made to the raw water levels to bring them into line with the scale readings. When necessary, corrections are generally less than one or two centimetres.

Flow discharges are then derived from bias-corrected water levels using a rating curve specifically elaborated for each monitoring point. The rating curves were all based on information supplied by the manufacturer (e.g., for the Venturi flume) or theoretical laws (e.g., for the V-notch weir), then adjusted locally by episodic discharge measurements taken during floods. The latter were carried out on several occasions in order to cover the full range of possible flow rates. They were carried out by recording the time required to fill a container of known volume for low flow rates, and by measuring the velocity profile across cross-sections using a portable FLO-MATE electromagnetic current meter (Model 2000, Marsh-McBirney, USA) for high flow rates. The water discharge time series for each hydrosedimentary station was compiled from a combination of the LQ and HQ discharge time series, using the most accurate measurement source at each instant.

By default, the origin and quality of the water discharge time series for each hydrosedimentary station are set to LoC and F3. If there are gaps at the measurement point suitable for low discharges (LQ), it may be necessary to use the HQ discharge values, but in this case the quality code has been changed from F3 to F2 for the PAR station (because the Venturi flume provides an accurate discharge rate even for low values); and from F3 to F1 for the MBV or GBV stations. If there are gaps at the HQ measurement point, a gap period is inserted (NA value at the start and end) for the corresponding periods. In some cases, discharge values exist for the appropriate discharge range but they are derived from water levels of altered quality. This happens, for example, when there are heavy deposits of sediment at the point where the water level is measured. These periods of flow obstruction are detected through reports made by the operator during their regular site visits and through visual analysis of water level fluctuations. If this is the case for LQ discharges, either the code source was changed from F3 to F2 or from F3 to F1 depending on the estimated level of distortion, or the HQ discharge values were used and the quality code was modified according to the rule mentioned above. If this is the case for HQ discharges, either the quality code was changed from F3 to F2 or from F3 to F1 depending on the estimated level of distortion. Where distortions were considered too great, a gap period has been introduced.

The water level in the outlet reservoir, i.e., at the LAKE station, is monitored using a digital piezoresistive pressure transducer (CS451, Campbell Scientific, UK). Raw measurements are made every 5 min and they are recorded when height variations exceed 10 mm or systematically every 60 min. The outlet reservoir is equipped with a limnimetric scale so that an operator can manually read the water level on a daily basis, or even more frequently in the event of flooding. The spillway has been designed so that the overflow discharge can be accurately estimated from the water level using a rating table. This table was made up of two theoretical curves: one for the V section, adjusted locally by flow control measurements, and one for the water level over the V section of the spillway. The periods during which the bottom gate is open are recorded manually, enabling the flow rate discharged to be estimated from the water level using a theoretical curve adjusted locally by flow control measurements. Lastly, the depth/volume and depth/surface curves of the reservoir are established on the basis of topo-bathymetric surveys carried out every 2 to 5 years.

As with the hydrosedimentary stations, the time series of raw water levels are first compared with the scale readings and a piecewise bias correction is then applied to the raw water levels to bring them into line with the scale readings. Then daily runoff entering into the reservoir (Q_LAKE) could be derived using an hydrologic budget as described in Albergel et al. (1998). Basically, the flow into the lake is evaluated as the variation in the stock of water in the reservoir, to which is added the volume of water leaving through overflow, the bottom gate and evaporation, and from which is subtracted the volume of rainwater falling directly onto the water body. The hydrological balance was established on a daily time step, with the value for day D corresponding to the daily runoff entering the reservoir between day D at 08:00 a.m. UTC and day D+1 at 08:00 a.m. UTC, in order to be in line with the operator's field surveys.

By default, Q_LAKE is qualified with a CALC source code and a F3 quality code, except for the period with overflow or bottom discharge for which a F2 quality code is assigned because the corresponding terms in the hydrological budget are estimated less accurately. When there are gaps in the time series of water levels (e.g., sensor failure) and overflow or bottom discharge occurs, the daily runoff entering the reservoir is set to NA, with an IND code source and an IND quality code. The daily time series of outflow from the reservoir via the spillway and/or bottom gate, labelled Q_LAKE_Outflow, is also provided. By default, it has a CALC source code and a F2 quality code. In the LAKE_Vol time series, we also shared the evolution of the reservoir's water storage volume, established from the 14 bathymetric surveys conducted between 1994 and 2023. By default, it has a CALC source code and a F2 quality code.

A gap-filled time series of daily runoff depth is proposed for the lake station over the entire 1994–2023 period. Gap-filling was realized using the GR5J model and the procedure available in the baseflow R package (Coron et al., 2017; Pelletier and Andréassian, 2020). The daily gap-filled time series is qualified with the same source code and quality code as the water discharge values for the day in question. The gap-filled values are qualified with a CALC source code and an IND quality code. A specific comment is also added to indicate the values derived from the GR5J gap-filling procedure.

Manual and automatic water samples are used to evaluate suspended sediment concentrations (SSC) in runoff water at each hydrosedimentary station. Automatic samples are taken using a 24-vial sequential sampler (SIGMA 900P or AS950, Hach-Lange, USA), the sampling of which is controlled by the flow rate. The capacity of each vial is either 350 mL for glass vials or 500 mL for plastic vials. They are supplemented by manual sampling of approximately 500 mL when an operator is present at the site during flooding. Samples were transported after each flood to the laboratory for oven drying (48 h at 105 °C, without prior decanting or filtration) and weighing.

SSC is calculated as the ratio of the dry weight to the sample volume. The measurement uncertainty associated with the SSC is estimated at 5 % maximum.

The SSC values are all qualified by default with a RAW source code and a F3 quality code. The quality code was modified in NF for a few values deemed unreliable, in particular very high concentrations at the end of major floods when it is suspected that the sample taken is affected by the deposit of sediment at the strainer. An additional comment indicates for each SSC value whether the value comes from a manual or automatic sample.

Figure 5a depicts the interannual variability of the rainfall regime in the Kamech catchment. Annual rainfall amounts range from a minimum of 395 mm yr⁻¹ to a maximum of 1036 mm yr⁻¹, resulting in a range of 642 mm. The standard deviation is approximately 150 mm around an interannual mean of 618 mm. Annual runoff rates range from 2.2 to 275.5 mm, with a standard deviation of 84.5 mm around a mean close to 108 mm yr⁻¹ at the LAKE Station. The annual runoff coefficient, defined as the ratio of annual discharge to annual precipitation, exhibits considerable variability, with a standard deviation of 10.8 %, an interannual mean around 16 %, and a range from 0.5 % to 38 %.

Figure 5b shows the intra-annual variations of runoff discharge and suspended sediment concentration (SSC) at GBV hydrometric station. For runoff, maximum rates are observed from December to March, with intermediate values in November. For SSC, maximum values are observed in September and October, and then quickly decrease, with intermediate values in November. As a result, there is a significant seasonal lag between runoff rates and suspended sediment concentrations, which greatly affects sediment yield dynamics in the catchment area.

The analysis of the daily discharge time series at the catchment outlet (i.e., Q_LAKE) shows a mean of 258 d with zero-flow per year over the period 1994–2023 (Fig. 6). Notably, discharge from the catchment is highly intermittent, as it only occurs 30 % of the time. Figure 4 also indicates the number of dry periods, a dry period being defined as consecutive days without flow discharge at the catchment outlet. The number of dry periods per year ranged from 15 to over 50, highlighting the very frequent changes between periods of flow and no flow.

Analysis of the cross-correlation between precipitation and discharge enables a rapid estimate of the lag time (tlag), which is a measure of how quickly a stream responds to runoff-producing rainfall (Sultan et al., 2022). According to these authors, the lag time corresponds to about two-thirds of the time of concentration. Figure 7 shows the cross-correlogram drawn up for the 2005–2023 period using the precipitation and discharge time series at the GBV station at 5 min time steps. It highlights narrow peaks with very low time lag ranging from 15 min in autumn to 50 min in spring. These results show that the Kamech catchment is highly reactive, with sub-hourly lag time.

A series of base flow separation methods (Table 3) was applied to the daily discharge entering the outlet reservoir (Q-LAKE time series) to derive the baseflow index, which represents the percentage of slow or delayed components to the streamflow. Five hydrograph separation methods come from the R “grwat” package (Rets et al., 2022). They were applied with the default parameters proposed in the package. A sixth one comes from the baseflow package (Pelletier and Andréassian, 2020), and is considered more impartial as it integrates a way to estimate the parameters entering in the flow separation procedure. As shown in Table 3 below, the contribution of baseflow varies depending on the algorithm applied from 3 % to 14 %, confirming the very high reactivity of the Kamech catchment, and therefore a presumed very low contribution of groundwater to the streamflow.

In this paper we characterized the duration curves in the Kamech catchment during the 18-year period between 2005 and 2023 using the gap-filled daily runoff depth and specific sediment yield at the GBV station (Fig. 8). The results show that a runoff depth over 10 mm d⁻¹ occurred less than 1 % of the time, which equates to 42 d during the 18-year period or an average of 2.3 d yr⁻¹. The average number of annual days decreases to less than 1 when considering a threshold of 20 mm d⁻¹, whereas one day exceeded 60 mm during the considered period. During the 18-year period, only 44 d exceeded 0.5 t ha⁻¹, 30 d exceeded 1 t ha⁻¹, 12 d exceeded 2 t ha⁻¹ and one day exceeded 10 t ha⁻¹. Results also show that 75 % of flow and sediment yield occurred during only 302 d, and 28 d, respectively, representing 4.6 %, and 0.4 % of the period considered.

The hydro-erosive regime in Kamech catchment is clearly non-perennial, it can be classified as ephemeral according to the classification proposed by Busch et al. (2020). Indeed, our preliminary analysis shows that streamflow occurs during a very short percentage of time and almost exclusively in response to precipitation events. The very significant compression in time of the runoff and the sediment yield in the catchment emphasised the importance of accurately and comprehensively documenting the response of extreme events, which are rare but account for most of the runoff and sediment fluxes in a Mediterranean context as already shown by Gonzalez-Hidalgo et al. (2007).