The present study was carried out in Alicante, SE Spain (38.386654N, 0.520001W), a semi-arid region where the average annual precipitation is 330 mm, and the reference evapotranspiration, ET ₀ , range is 1200-1400 mm with a mean summer temperature of 24.4ºC (standard deviation of 1.7ºC) and a mean winter temperature of 12.5ºC (standard deviation of 1.5ºC). For the field experiment, a 9 × 5 m plot was set up. The plot was located over an unconfined saline quaternary aquifer of detrital origin, which overlies impervious silty-clay materials dating back to the Cretaceous Period.

On the top surface, a mixture of turf grass species (St. Agustine grass, Stenotaphrum secundatum, and ray grass, Lolium perenne) was cropped to mimic current agricultural landscape management. For plot irrigation, brackish groundwater from the aquifer is currently desalted in a reverse osmosis plant. Subsequently, it is stored in a pond, where it is blended with raw groundwater. The rate of raw groundwater and desalted water mixture may range between 1:19 and 1:4 (brackish:desalted water) for winter and summer, respectively (Valdes-Abellan, 2013). Irrigation is finally carried out with the blended water. The plot is irrigated by micro-sprinklers, and this system provides a nearly uniform water input over the field experiment surface (Dechmi et al., 2010). This system is commonly used in this type of land cover.

Two different types of monitoring strategies adopted along a vertical profile in the vadose zone were applied in the experimental plot: non-automatic instruments, vertically installed from the surface and spatially distributed; and automatic devices (connected to a datalogger), horizontally installed from a lateral trench. The data obtained from each monitoring strategy were independently analysed and soil parameters were obtained to be compared. The monitoring process took place during 20 consecutive months.

The plot design included vertical boreholes drilled at different depths to install the non-automatic instrumentation, as well as a trench of 1.5 m in depth and 0.8 m in width, which was dug on one lateral side of the plot to install the horizontal automatic sensors (Fig. 1). After the in situ installation, the lateral trench wall was covered with a plastic film along its vertical extension, and was finally refilled with soil material to establish a condition of absence of flux along this plot side.

Manual instrumentation consisted in two Jet Fill tensiometers for suction measurements at 30, 45, 60, 90 and 120 cm (Soil Moisture®), and two polycarbonate access tubes of 44 mm i.d. and 1.9 m in length for soil moisture measurements with a TRIME FM TDR portable probe (IMKO®). The probe was calibrated in the laboratory following the manufacturer’s recommendations. Specific calibration for the investigated soil was carried out by the comparison of volumetric water content (θ) measured by the sensors and by the gravimetric method according to Laurent et al. (2001, 2005) and to Varble & Chavez (2011). Tensiometers and polycarbonate access tubes were vertically installed from the top surface by hand auger equipment (Eijkelkamp®). Vertical installation was carefully performed to prevent the preferential flow associated with the devices.

Automatic data acquisition comprised a set of seven real-time monitoring devices controlled by a datalogger (Campbell Scientific®). Instrumentation consisted in: five 5TE sensors (Decagon Devices®) installed at depths of 20, 40, 60, 90 and 120 cm to take measurements of soil water content, temperature and electrical conductivity; two MPS sensors (Decagon Devices®) for the soil suction measurements at depths of 20 and 60 cm. Sensors were horizontally placed into soil media at a 40-50 cm horizontal distance from the lateral trench side (Fig. 1) in order to minimise boundary effects. Finally, the device installation holes were refilled with local soil material to reduce soil disturbance (Rothe et al., 1997; Ghezzehei, 2008). All the connecting wires were protected with a 16-mm flexible plastic tube to avoid instrumentation damage (Fig. 1).

For the present research, data acquisition and recording started on 1 September, 2011 (day 1) and ended on 30 April, 2013 (day 608). Instrumentation was installed 8 (non-automatic) and 5 (automatic) weeks prior to beginning the monitored period in order to avoid any uncertainties associated with the equipment settlement process. For the non-automatic tensiometers, two repeated measurements were taken 3 times per week at each depth and the average value considered. Manual measurements of θ were calculated by averaging the TDR probe readings carried out twice a week (two measurements per depth in each access tube). Automatic measurements were obtained at time steps of 1-hour without replication.

The time series of precipitation, air temperature, relative humidity, wind speed and direction, barometric pressure and solar radiation were obtained from a meteorological station located 950 m from the experimental plot. All meteorological data, captured every 30 minutes, were provided by the Laboratory of Climatology (University of Alicante).

In order to determine the soil physical properties, five undisturbed soil cores were collected at different depths from three drilled boreholes. In two of them, soil samples were taken at depths of 30, 45, 60, 90 and 120 cm, whereas in the third at 20, 40, 60, 90 and 120 cm.

Soil bulk density (Grossman & Reinsch, 2002), volumetric water content (Topp & Ferré, 2002), particle size distribution (Gee & Or, 2002), saturated hydraulic conductivity (Reynolds & Elrick, 2002) and soil water retention curves (SWRC) with a ceramic plate extractor (Dane & Hopmans, 2002) were obtained at the laboratory. The unsaturated soil hydraulic parameters were: i) model fitting from the SWRC-fit code (Seki, 2007) based on the van Genuchten-Mualem constitutive relationships (Mualem, 1976; van Genuchten, 1980); ii) from Rosetta (Schaap et al., 2001) based on the soil textural fractions. Finally, soil mineralogy was determined by X-Ray diffractometry (Brown & Brindley, 1980); Jones, 1991.

From the analysis of the soil samples from the drilled boreholes, the soil parameters estimation in the laboratory and the lateral trench inspection, three soil layers (Fig. 1) were defined in the plot. Due to the thickness of layer 1 (6 cm), the retention curve determination in the laboratory was not possible, and the soil parameters had to be obtained from textural fractions through Rosetta. For layers 2 and 3 (Fig. 1), the soil parameters were determined from laboratory retention curves and the SWRC-fit code.

According to X-Ray mineralogy, the three layers presented high calcite and quartz contents (more than 85%). Clay mineral content was practically negligible, except in the first layer where illite accounted for up to 9.7% (Table 1).

Table 2 summarises the van Genuchten-Mualem parameters K _s, α, n, θ _s and θ _r , estimated by the different procedures and for the three defined layers: a) laboratory determinations, and using Rosetta and the SWRC-FIT code; b) by inverse modelling and data from non-automatic devices; c) by inverse modelling and data from automatic devices.

The atmospheric boundary conditions, potential transpiration (T _p), potential evaporation (E _p), daily precipitation (P) and irrigation (I), recorded and computed throughout the monitored period are shown in Fig. 2. Precipitation occurred mainly in spring and autumn, and was unevenly distributed in a few intensive events, while irrigation dose was applied following climatic conditions and plant demand. T _p presented a more accentuated annual cycle according to soil water availability than E _p, which showed a smoother daily value throughout the monitored period.

Fig. 3 illustrates volumetric water content θ measured at all the monitored depths with both the automatic and non-automatic devices. Due to technical problems, some gaps exist in the time series record. The measured pressure head h displayed a similar behaviour for both the automatic and non-automatic devices (Fig. 4).

Field data quality was assessed by visual inspection and the values not considered representative were not taken into account for modelling purposes. The significant difference between the number of measurements collected from each device type is noteworthy. Automatic data acquisition provided 3,820 daily values, whereas less than 1,000 were collected in the more time-consuming non-automatic, or classical, monitoring devices for the same period.

Fig. 5 shows the observed and simulated θ data for the automatic and non-automatic systems. A good agreement is observed between the experimental and modelled data; simulation captured the general trend, except for the detected extreme values, as expected. Fig. 6 shows the h modelled results and field measurements for both data acquisition types. The general trends, and even some extreme events, were well predicted by the model, which were also reflected by the statistics, as presented below.

Despite the good agreement obtained between the two modelled results, the flow through the lower boundary accounted for 711 mm and 774 mm from the automatic and non-automatic data, respectively, for the period from 1 September, 2011 (day 1) to 31 August, 2012 (day 366). Considering that this flow constitutes the bottom drainage to the aquifer (i.e., deep percolation), the observed difference of 8.1% could be important when attempting to assess water recharge in arid or semi-arid areas.

Regarding water availability for plants, similar trend conclusions were drawn applying the model by using different experimental data sets (Fig. 7). Nevertheless, modelling by using automatic devices data resulted in a slight overestimation of water availability if compared to the other case. This highlights the fact that high frequency measurements in the field are not necessary to obtain reliable outputs for irrigation planning definition.

Fig. 8 shows the SWRCs at the depths of 20 cm (layer 2) and 60 cm (layer 3) obtained from the laboratory experiments (drying curve), as well as the corresponding estimated on the basis of inverse modelling, using data from manual and automatic devices.

Table 3 shows RMSE, EF and R ² evaluated for h and θ at the monitored depths for both the devices (calibration period), as well as RMSE and R ² obtained for the validation period. The values of RMSE and EF for θ do not clearly allow the identification of the best data acquisition system at the different depths or when all the data are considered. The non-automatic EF results gave negative values for two sensors at 90 cm and 120 cm depths (Table 3), indicating that the mean experimental value was a better predictor than simulation. This might be due to the low variability of the measurements for both data acquisition systems at these depths. If considering h at both depths, the RMSE and EF values indicated better adjustment for the manual acquisition data.

R ²for the combined water content and pressure head data estimated by Hydrus gave similar results for both the non-automatic (calibration: 0.81; validation: 0.94) and automatic (calibration: 0.98; validation: 0.98) devices, respectively; the different numbers of experimental data used in each model, which were considerably lower for the non-automatic devices could explain the observed differences.

Data analysis shows that water content tends to take an almost constant value, or a field capacity threshold, immediately after water input, which indicates the high hydraulic diffusivity of soil. Similar trends between the volumetric water content measurements from both types of devices were observed, except for the manual measurements at depths of 45 cm and 60 cm, which presented a flat behaviour from day 350. For the non-automatic equipment, the θ data were averaged from the two TDRs observations (TDR1 and TDR2) as horizontal homogeneity was assumed over the experimental plot. Minor differences in the volumetric water content measurements between the non-automatic (TDR1 and TDR2) and automatic sensors measurements at similar depths were observed. The differences did not follow a clear pattern among all the studied depths. They can be related to soil heterogeneity, the installation process and sensor limitations in the measured domain. Regarding the installation processes, lack of contact between soil and the polycarbonate access tube (non-automatic, TDR1 and TDR2) may lead to the presence of soil voids along the tube (West & Truss, 2006). As regards the sensors, the different soil volume controlled by the devices (3,100 and 147 cm³ in non-automatic and automatic devices, respectively), which may contain large-sized particles, may also constitute a source of measurement uncertainty. In addition, changes in temperature and electric conductivity may have produced some alternation in the 5TE sensor readings as observed by Rosenbaum et al. (2010, 2011) for sensors installed at shallow soil depths.

As a general behaviour, changes in the water input on the upper boundary were first detected by the manual devices after starting irrigation following a drier winter period. The lag between the automatic and non-automatic readings can be attributed to the vertical layout of the non-automatic tensiometers since presence of a preferential flow along tensiometer length should not be ruled out. Moreover, automatic tensiometers’ (MPS-1) accuracy exhibited a non-linear and uncertainty increased with h. The associated measurement of uncertainty within the -100 to -600 cm range was lower than -100 cm, and could even be higher beyond this range (Malazian et al., 2011).

The main differences among the three soil parameterisation methods lie in the obtained θ _s, α and n values. Although differences of K _s have been generally observed in other research studies, in this research significant differences among the three methods were not observed according to the coefficient of variation values of 0.13, 0.13 and 0.24 for layers 1, 2, 3 respectively and K _s values reported in Table 2.

Based on the obtained results, it can be stated that the range of parameter values covered in the field experiments was not wide enough to obtain reliable soil water retention curves and, consequently, reliable soil parameters. However, Wollschläger et al. (2009) concluded that θ and h evaluated with in situ time series, along with inverse modelling, were suitable for determining soil hydraulic characteristics. Moreover, the SWRCs estimated from the field data, unlike those data obtained from the laboratory, were generally determined under transient conditions, which confers uncertainty to the final results, as well as differences of soil volume between the soil cores analysed in the laboratory and the volume considered by the sensors.

Regarding K _s, and by presenting a range of several orders of magnitude, similar values were obtained for all the layers and parameterisation approaches. The parameters obtained with Rosetta for layer 1 showed that pedotransfer functions can be a useful tool when laboratory determinations are not provided. However, in order to reduce uncertainty with soil hydraulic parameters, pedotransfer functions should be combined with other currently applied parameter determination techniques.

Soil parameterisations through inverse flow modelling from both data types gave similar results despite the data length record and variability. Accordingly, both the applied data acquisition methodologies proved useful and were equally accurate for assessing in situ water flux dynamics.

Field monitoring for irrigation dose demand is crucial for avoiding plant stress and water resources sustainability in order to prevent water losses by deep infiltration or further groundwater pollution consequences. The need for efficient irrigation management implies agricultural system knowledge, and also the appropriate technologies and methods to be applied for accurate soil media measurements. The results obtained from this study provide some insights into soil and vadose zone monitoring strategies to assess water dynamics and soil hydraulic properties.

Field research and hydraulic parameters monitoring offer the complexity of profiling values in soil depths. One of the main limitations of the present work is certainly our inability to measure effective properties to describe the overall system; used devices for hydraulic parameters monitoring only provide information for a small soil volume to describe the average properties of the complete profile.

The observations made with the vertical-installed instruments need to be carefully analysed as the installation process can lead to preferential flow paths, and to lack of contact between the device wall and soil media, which can distort the observed values, which would not likely to be detected by our sensors if only one monitoring strategy type was considered. It is not clear whether conclusive measurements can be taken despite reasonable efforts made, especially when it is necessary to quantify extreme heterogeneity phenomena, such as preferential flow.

The laboratory determinations of soil hydraulic properties better fit the experimental measurements (from both the automatic and non-automatic sensors) than the soil hydraulic properties inferred from inverse modelling fit the laboratory results. It should be noticed that if previous information on soil characteristics from laboratory determinations are not considered, unrealistic soil hydraulic parameters from modelling approaches may be obtained.

Hydrus simulations were able to mimic the responses of a field-scale instrumented plot with the automatic and manual data acquisition systems. The statistical results and visual inspection showed a good agreement between the predicted and the observed values in θ and h, even with a major difference in the number of observations associated to the different strategies of data collection.

Nevertheless, the θ and h trends were better predicted than extreme episodes, a fact that has also been observed by other researchers. The goodness-of-fit values showed slight differences between the modelled parameters from the manual and automatic experimental data. The fit of the θ data was slightly better than for the h data. This fact may be related to the equilibrium time with soil required by the pressure head devices to obtain accurate measurements.

From the obtained results, it is stated that selection between automatic and manual devices has to be based on the particular research objective. Time-consuming data acquisition (longer with manual equipment) and high equipment acquisition costs (for the automatic devices) both appear to be the most important aspects when finally selecting instrumentation. From an irrigation management viewpoint, no significant differences are highlighted between the two investigated strategies. However, soil hydraulic properties may be better assessed through automatic acquisition as data variability and accuracy are lower than by the non-automatic type. Moreover, the differences in the bottom drainage value obtained by considering the two different data acquisition systems highlight the variability of this assessment to similar experimental data.