Four typical agricultural fields were selected: (i) organic and (ii) conventional greenhouses for vegetable crops, (iii) an open crop field with rotating seasonal crops, and (iv) a citrus orchard. All sites were located in the agricultural area of Israel's coastal plain. The porewater samples were collected by vadose zone-monitoring systems (VMSs) that have been operating at these sites continuously for more than 9 years. The VMS includes a porewater sampler that is permanently installed in the unsaturated zone under the cultivated fields. Accordingly, variations in the chemical characteristics of the soil porewater may be detected continuously at the same spot in the subsurface over many years. A detailed description of the VMSs at each site can be found in Dahan et al. (2014), Turkeltaub et al. (2014, 2015, 2016), and in Sect. S1 in the Supplement. Additional information on the research site locations, crop types, and irrigation and fertilization regimes can be found in Sect. S2. The porewater-sampling ports at each site are distributed at various depths, ranging from 1 to 21 m (Table S3 in the Supplement). In this study, soil water samples were collected in four sampling campaigns: (i) August 2015, (ii) September 2015, (iii) January 2017, and (iv) February 2017. Note that the VMS sampling ports are permanently installed at the site and therefore enable repeat sampling from the exact locations for many years, while the agricultural activity on land surface remains undisturbed.

Samples were analyzed for nitrate concentration with a Dionex ICS-5000 ion chromatograph and the Analytik Jena TOC, DOC, TN, DN multi N/C 2100S TOC/TN analyzer for DOC and total nitrogen (TN) concentration. Spectral analyses of the samples were performed with a Thermo Scientific Evolution 201/220 Desktop laboratory spectrophotometer. Double-distilled water (DDW) was used as a reference/baseline for the analyses. The samples were held in a standard 5 mL quartz cuvette with an optical path of 10 mm and were scanned over a broad spectrum of 190–1000 nm. The analytical procedure for UV spectral analysis of nitrate concentration in porewater samples usually requires colloid filtration, dilution, and sometimes spiking with the target constituent or supplementary reagents. However, since the purpose of this study was to develop an analytical protocol that enables in situ measurement of nitrate concentration through spectral analyses of the soil porewater, the samples were analyzed without any additional preparation (i.e., dilution or filtration). The porewater samples were then examined for absorption at a few specific wavelengths that have been previously suggested for direct nitrate measurement in untreated soil water: (i) 302 nm (Moo et al., 2016), (ii) 235 nm (Shaw et al., 2014; Tuli et al., 2009), and (iii) where the absorbance used for calibration equals the absorbance at 220 nm after subtraction of twice the absorbance at 275 nm (hereafter 220/275 nm) (Armstrong, 1963). An additional measurement at 220 nm, as suggested by Michael et al. (2017), was also carried out, but there was no significant difference in absorption characteristics compared to the 220/275 nm method. Therefore, the data from this test are not presented.

In order to validate our suggested method's resistance to measurement drift, which may occur in response to changes in the solution chemical matrix, a second spectral analysis was performed. This analysis was carried out in a Spark 10M multimode microplate reader spectrophotometer at wavelengths of 200 to 1000 nm. Absorbance was defined by the Lambert–Beer equation (Eq. 1): 1absorbance=-log⁡10II0, where I is the light intensity after passing through the examined solution, and I0 is the light intensity after passing through a reference sample (blank).

The accuracy of the suggested method was determined by fitting a linear regression model to the absorbance and the nitrate concentration (measured by ion chromatography) data. The model fit, coefficient of determination (R2), and its corresponding P values were obtained using the fitlm function in MATLAB.

In order to enable continuous in situ measurement of nitrate concentration in the soil, a monitoring concept was developed in which the spectral absorption of the soil porewater is measured in an optical flow cell (Fig. 2) (a patent is pending on the methodology described in this article). The optical setup consists of a UV lamp and UV–VIS spectrometer, designed to measure transmission and absorbance between 190 and 850 nm. A special feature in SpectroWiz (StellarNet software) was used to prevent possible measurement drift. A StellarNet SL3 deuterium light source was used as continuous-wave UV light source. The spectrometer and UV lamp were connected to a flow cell using optical fibers and collimating lenses. The optical flow cell was connected at one end to a customized suction cup, which enables continuous sampling of the soil porewater under a low flow rate (a few milliliters per hour). At the other end, the flow cell was connected to a sampling cell. Charging the sampling cell with low pressure draws a continuous flux of porewater from the soil through the optical flow cell to the sampling cell. The system is designed to function under a small dead volume (4–6 mL) by reducing the suction cup's inner volume and using small-diameter tubing (inner diameter 1.6 mm). Porewater solution that flows from the suction cup through the optical cell accumulates in the sampling cell, and it is used later to determine nitrate concentration by standard laboratory procedure.

The monitoring system for continuous measurement of nitrate concentration in the soil was tested in two sets of column experiments. The first was conducted to test the ability of the optical setup to measure nitrate concentration in the soil under controlled conditions. In this experiment, 18 L of clean (low organic matter) sandy loam was packed in a 50 cm long column. Two identical customized suction cups and one water-content sensor (TDT, Acclima) were placed at a depth of 22 cm in the soil column. One of the suction cups was connected to the flow cell and the other directly to its sampling cell (Fig. 2). The column was irrigated daily with 1 L of fresh tap water (equivalent to about 14 mm), where one of the irrigation cycles was enriched with 1000 ppm nitrate (as KNO3). In this experiment, nitrate concentration of the soil porewater was measured continuously using absorption spectroscopy technique in the optical flow cell and compared to the concentration in the porewater samples that were accumulated in the two sampling cells and in the column drainage. The second experiment was conducted using agricultural soils in three soil columns packed with fine sandy loam, dark clay soil, and fine sandy loam mixed with 10 % commercial compost, respectively. The experiments were conducted in all three columns under similar irrigation, fertilization, and monitoring setups (Table S4). The irrigation regimes in the column experiments were designed to ensure unsaturated conditions, similar to agricultural soils (immediate drainage and no flooding conditions). Water content in the column experiment varied between 15 % and 16.5 % in the sandy loam, which is equivalent to a water potential of 850 to 950 mbar. To ensure continuous water flux from the soil to the optical sensor a pressure between 600 and 800 mbar (absolute values) was applied to the suction cups.

Nitrate concentration plotted against absorbance at the selected wavelengths for all the porewater samples had shown inconsistencies between the nitrate concentration to the absorbance values (Fig. 3). At 302 nm (Fig. 3a), a reasonable correlation between the absorbance and nitrate concentration was obtained for the open crop field (R2=0.99) and conventional greenhouse (R2=0.95), whereas poor correlations were obtained for the other two fields: organic greenhouse (R2=0.39) and citrus orchard (R2=0.49). Partial improvement was achieved at 235 nm (Fig. 3b), with R2 values of 0.97, 0.91, and 0.98 for the organic greenhouse, open field crop, and conventional greenhouse, respectively. However, a poor correlation was obtained for water samples from the orchard (R2=0.71). Moreover, a close inspection of the absorbance of water samples from the open crop field showed a strong shift in absorbance values at nitrate concentrations exceeding 1000 ppm. This phenomenon was observed in repeat analyses of additional water samples (Fig. S5 in the Supplement). With the 220/275 nm method (Fig. 3c), poor correlations between absorbance values and nitrate concentration were observed at most sites (R2=0.39, 0.09, 0.75 for organic greenhouse, open field crop, and conventional greenhouse, respectively); however, for the orchard site, the correlation was improved compared to the other methods, reaching R2=0.9. Note that one of the porewater samples from the organic greenhouse (from 13.3 m below the surface with 171.36 ppm nitrate) did not meet the requirements of the 220/275 nm absorbance ratio and is therefore not included in Fig. 3c. None of the methods based on specified wavelengths seemed robust enough for direct analysis of untreated soil water obtained from various fields with different soils.

Several reasons could account for the observed mismatch between absorbance values and nitrate concentration at the various sites. At short wavelengths, such as 220 nm, absorbance is typically very high (Fig. 1); therefore, the measurement is very sensitive to low nitrate concentrations. At high nitrate concentrations, however, absorption saturation occurs, and the absorbance is no longer indicative of increased concentrations. Accordingly, in agricultural soils, where nitrate concentration may vary from tens to thousands of parts per million, as demonstrated in the water samples obtained from sites used for this research, the shorter wavelengths are less applicable for direct analysis (i.e., the samples need to be diluted). This explains the low correlation found for 220/275 nm and the low sensitivity to high concentration at 235 nm. The 300 nm region is typically characterized by low absorption rates for nitrate (Fig. 1), thereby reducing the potential for signal saturation. As such, it is more ideal for measuring nitrate at high concentrations. Our measurements at 302 nm were insensitive to the low nitrate concentrations (49.7–75.4 ppm) at the orchard site. Furthermore, significant mismatch was observed for the organic greenhouse, even though the nitrate concentration at this site was relatively high, ranging from 171 to 520 ppm (Fig. 3a). This mismatch was expressed as increasing absorption values, regardless of the nitrate concentration. The main reason for the increased absorption could be attributed to signal masking as a result of the presence of DOC, which is commonly found in agricultural soil porewater (Jones and Willett, 2006; Kalbitz et al., 2000). Nevertheless, a closer look at the absorption pattern showed that different sites may have appropriate calibration curve for nitrate concentrations at different wavelengths, which implies the possibility of adopting a unique wavelength for each site.

The absorption spectrum of porewater samples obtained from various depths under the organic greenhouse showed the highest absorbance for samples from cells located at a depth of 1.3 m (Fig. 4a), despite having the lowest nitrate concentration in the sample batch (Fig. 4b). Although the high absorbance values might be attributed to the presence of DOC, these water samples did not have the highest DOC concentration. On the other hand, the water sample at a depth of 13.3 m, which did have the highest DOC concentration of the current batch (Fig. 4b), showed the lowest absorbance value (Fig. 4a). This peculiar behavior was found consistently in subsequent sampling campaigns (Fig. S6). Thus, it could be deduced then that the DOC absorption characteristics are not impacted solely by the overall DOC concentration but also influenced by the specific characteristics of the various organic compounds composing the overall DOC. Accordingly, different soils at different sites could potentially be characterized by different organic compounds in their specific DOC “soup”, which could therefore have its own typical absorption spectrum.

The attempts to measure nitrate concentration at a specific wavelength (302, 235, and 220/275 nm) showed inconsistencies between the absorption characteristics and nitrate concentration, attributed to absorption saturation and the presence of DOC. However, DOC concentration was not always correlated with absorbance. As a result, a new approach was adopted to better assess the effect of nitrate and DOC concentrations on the absorption spectra. In this approach, the coefficient of determination (R2) between a set of nitrate/DOC concentration vectors and their corresponding absorbance vectors was calculated for the entire spectrum (Fig. 5, Table 1 and Fig. S7).

The coefficients of determination (R2) vs. wavelength, for both nitrate and DOC concentrations, are shown in Fig. 5a for the open crop field and Fig. 5b for the citrus orchard samples. The R2 values for nitrate in the crop field show an increase at 225 nm, reaching a plateau (R2>0.99) between 235 and 250 nm. They then decreased to a minimum value of 0.57 at 264 nm and rose again to a second, high-value plateau (>0.9) between 290 and 320 nm. However, the R2 pattern for the DOC concentrations in the crop field differed from that for nitrate. In some sections (220–235 and 225–360 nm), the trends were positively correlated, whereas in others (250–325 nm) they were either negatively correlated or not correlated (Fig. 5a). Unlike the case of the open crop field, where two distinct high R2 value plateaus were visible, analysis of the citrus orchard R2 values showed only a narrow area with high R2 values between the wavelengths of 220 and 230 nm. Here, the high R2 values (>0.8) were only reached at 220–235 nm, whereas for the rest of the spectrum, the correlation was very poor (<0.4) (Fig. 5b). On the other hand, at this site, R2 values for the DOC remained very low (<0.3) over the entire spectrum. A similar R2 vs. wavelength analysis was carried out for the other fields, and the trend in R2 for each field seemed to show unique behavior (Fig. S7).

The wavelength regions with high R2 values showed a higher correlation between the targeted chemical concentration (nitrate or DOC) and absorbance values. Thus, absorbance values in those areas had greater potential for measuring the targeted constituent's concentration. For example, in the open crop field, the areas of the two distinct high R2 plateaus (Fig. 5a) hold high potential for measuring nitrate concentrations in soil porewater collected from that field. Between the two sections of high correlation to nitrate concentration, at around 267 nm, absorbance values were correlated with DOC concentration, meaning that this area of the spectrum is expected to have a high DOC masking effect. These characteristics were unique to the open crop field. In the citrus orchard (Fig. 5b), for example, the data series associated with nitrate concentration presents high potential for estimating nitrate concentration at wavelengths between 220 and 230 nm. Moreover, the low R2 values for the DOC curve suggest that the DOC chemical composition in the citrus orchard porewater samples does not have much effect on the UV absorbance absorption spectrum over a greater section of the spectrum.

Although DOC concentration in porewater at the different sites in this study was rather similar (Table S8), the DOC impact on the absorption spectrum was very different at each site. It was assumed that these variations are due to the composition of the various organic molecules making up the DOC in the different fields. DOC is a general term folding thousands of different organic molecules within it. Accordingly, the specific chemical composition of the DOC may be affected by various factors, such as differences in soil type, crop type, differences in the applied fertilizers, and local climate (Kalbitz et al., 2000). For example, regardless of the proximity between DOC concentration values in the citrus orchard and the open crop field, the presence of DOC did not cause similar interference in the spectral analyses of the porewater at those sites. In fact, in the crop field, where DOC concentrations were slightly lower than those in the citrus orchard, the presence of DOC had a much higher impact on the absorption spectra of the porewater samples taken from the crop field compared to samples taken from the citrus orchard. Nevertheless, every field site can be characterized by wavelength regions that have greater potential for measuring nitrate concentration and those that might be more susceptible to interference by DOC or other constituents in the solution (Fig. 5). This phenomenon opens the way to a new concept, whereby a wavelength can be determined that is uniquely suited to measuring nitrate in each field while avoiding possible interference related to other natural water constituents, such as DOC.

The observed variations in the coefficients of determination for nitrate and DOC concentrations at different wavelengths (Fig. 5) led to the adoption of an innovative strategy for analyzing nitrate concentration by absorbance spectroscopy. The new analytical procedure was designed to overcome the measurement inconsistencies associated with estimations of nitrate concentration using absorbance spectroscopy methods with a fixed wavelength (Fig. 3).

A two-step procedure was used to determine the optimal wavelength for nitrate concentration measurements in soil porewater samples at specific sites. The first step consisted of creating a set of candidate wavelengths that show high potential for measuring nitrate concentration. This was achieved by plotting the R2 values of absorbance intensities of known nitrate concentrations vs. wavelength (Fig. 5). The candidate wavelengths were then screened to satisfy two requirements:

R2 test. An initial screening of the wavelength range was performed by setting a threshold value that is within 98 % of the maximum R2 value in the tested batch (Fig. 6). Wavelengths showing R2 values below that threshold were rejected, while the wavelengths displaying R2 values above the threshold were used to form a set of candidate wavelengths for a site-specific calibration equation. In this example, Rmax2=0.9953, so the R2 threshold value was set to R98%2=0.9753.

The site-specific optimal wavelength was determined by combining the R2 and σ2 values for each wavelength; the square root of the sum of the two criteria's values (Eq. 2) was calculated for those wavelengths that have R2 values above the set threshold. Figure 6 shows that, at a wavelength of 238 nm, a peak point on the curve emerges, indicating that it is the most suitable wavelength for spectral analysis of nitrate concentration for this particular site (open crop field). 2Combinedcriteria=R2+σ2 Application of this procedure to determine the optimal wavelengths for all fields used in this study enabled establishing a specific calibration curve for each site. Plotting the nitrate concentration as obtained by ion chromatograph against absorbance values at multiple wavelengths (organic greenhouse at 231 nm and R2=0.99, open crop field at 238 nm and R2=0.99, conventional greenhouse at 234 nm and R2=0.99, and citrus orchard at 223 nm and R2=0.98) showed very high correlations. In this case, each of the fields was successfully assigned to an individual calibration curve, generated by the most suitable wavelength for that specific site. Figure 6 shows information for the open crop field station; further information for the two-step procedure's application to the other field stations is presented in Sect. S9. Note that the poorly correlated data in Fig. 3 and the highly correlated data in Fig. 8 were produced from same absorption spectra of the same water samples. The only difference is that the data in Fig. 3 were created by application of fixed wavelengths of known methods, whereas the highly correlated data in Fig. 8 were created on the basis of an analytical procedure that searches for a site-specific optimal wavelength.

The robustness of the suggested monitoring concept is primarily dependent on the temporal stability of the site-specific calibration equations, as it gained from the previously described calibration procedure. There are two main reasons for calibration drift: (i) drift in the optical apparatus due to light source degradation or intensity fluctuations and (ii) changes in the porewater solution matrix chemical composition, which might lead to absorbance-signal masking or other interference patterns in the spectral analyses.

The data collected from August 2015 samples were used as input for the site-specific algorithm. As the algorithm output, a calibration equation at different wavelengths was obtained for each field site. The stability of these calibration equations had been tested on samples from additional sampling campaigns later in 2015, and in 2017, where results from standard laboratory analyses (observed nitrate concentrations) were plotted in reference to the result of the calibration equation, obtained in August 2015 (predicted nitrate concentration). Figure 9 shows a good correlation between the predicted and observed values with general R2>0.9. It is therefore suggested that the initial calibration equation which was determined by the spectral analytical procedure 2 years earlier (2015) was still valid for nitrate concentration estimations, regardless of the changes in agricultural activity between growing seasons. It may therefore be deduced that establishment of a site-specific calibration curve that is based on the adoption of a site-specific wavelength can be used for long-duration monitoring of nitrate in soil porewater, as long as stability of the UV light source is maintained.