1. Introduction
Water shortages, especially for agriculture, are prevalent in countries in arid and semiarid regions and they have become more and more severe in these regions in recent years due to a lack of water resources and abrupt climatic changes [
1]. In the world’s arid regions, the agriculture sector depends on groundwater as a primary water source for irrigation. Generally, groundwater is used to irrigate 30–40% of irrigated land [
2]. However, in recent decades, various human activities have resulted in groundwater pollution drastically and caused a periodic change in their quality. Therefore, regular assessment of groundwater quality is necessary to identify its suitability for irrigation purposes where the productivity and quality of crops depend highly on soil characteristics and quality of irrigation water.
Several important water quality indicators are frequently applied to assess the quality and suitability of groundwater for irrigation. Generally, the water quality index (WQI), which was initially proposed by Brown et al. [
3] and is based on the primary parameters of water quality, namely major cations and ions, heavy metals, temperature, and pH, is frequently used to classify the quality of water for drinking purposes. While, the irrigation water quality indicators (IWQIs), namely percentage sodium (%Na), sodium absorption ratio (SAR), soluble sodium percentage (SSP), total dissolved solids (TDS), potential salinity (PS), total hardness, and Kelly’s index (KI), which are based on the major cations (K
+, Ca
2+, and Mg
2+) and ions (Na
+, Cl
−, and SO
42−), are frequently used to assess the suitability of water for irrigation purposes [
4,
5,
6,
7,
8]. When the values of WQI and IWQIs are outside of standard requirements, there are direct and indirect negative effects on soil characteristics and plant growth, which ultimately results in significant reductions in crop production without contributing to solve the water shortage problems in the agriculture sector. Therefore, accurate and efficient assessment of these indicators can improve the management and effective use of groundwater.
The standard approach for assessing different water quality indicators, which includes collecting samples from the field at fixed points and then analyzing the samples in the laboratory and comparing the data with a reference standard following protocols, is practicable and highly accurate. However, the methods used for assessing water quality should not only accurately reflect spatial variations in water quality, but they should also conveniently monitor water quality levels in an inexpensive and fast manner to provide a real-time estimation of spatial and temporal variations in water quality on a large scale, which is very important for the comprehensive assessment and management of groundwater quality. For instance, although the WQI can be assessed using a simple mathematical instrument, which transforms the large quantity of water characterization data into a single value that represents the water quality level and reflects overall water quality levels [
6,
9,
10,
11], the WQI does not indicate spatial and temporal variations in water quality [
12]. In addition, although different geometrical diagrams and mathematical models are frequently applied in groundwater chemistry assessment, they are generally costly and time-inefficient, tedious, and require significant input and model assumptions, which limit estimations to only being reliable at regional or smaller scales [
13,
14]. Due to the limitations of standard methods for monitoring and managing groundwater quality in real-time and on a large scale, there is an urgent need to apply reliable, practical, rapid, and economical monitoring tools, which can be deployed easily and help decision-makers to assess key indicators relevant to groundwater quality in a comprehensive manner.
With the rapid developments in space information and increasing use of computer applications, different airborne, satellite, or proximal remote sensing techniques have been demonstrated to be cost-effective and applicable on a large scale for integrative assessment of several IWQIs [
12,
15,
16,
17,
18,
19]. The concept behind these techniques is that the different sensors of these tools can detect changes in the optical properties of the water surface at various wavelengths. The optical properties of the water surface are significantly interlinked with the changes that take place in the physical, chemical, biological, and hydrological characteristics of the water. Therefore, the spectral signatures reflected from the water surface can be exploited directly or indirectly to assess different water quality indicators, such as temperature, pH, salinity, total suspended solids (TSS), chemical and biological oxygen demand, total phosphorus, ammonia nitrogen (NH
3-N), and dissolved organic carbon [
18,
19,
20,
21,
22,
23,
24,
25,
26,
27].
Space-borne optical remote sensing systems, such as satellites and airborne remote sensing systems, have a wide spatial coverage and are commonly used to detect different water quality indicators. However, the spectral resolution of these systems is insufficient and results in uncertainty in extracted water quality data. This is because ~90–98% of the signals obtained by these systems are from the surface of the water and atmosphere, and only 2–10% of the signal is from the water components. This results in complexity in the data of the optical properties of the water surface, which makes the assessment of water quality using these systems inaccurate [
19]. A ground-based hyperspectral system can be a useful tool to overcome the limitation of external interference in water quality assessment. This system can obtain high-resolution spectral information and ensure the accuracy of spectral inversion because of the proximity between the optical sensors and the target. In addition, the data of this system has the advantages of a large amount of information, a large number of bands, and strong quantitative inversion flexibility [
27]. Generally, in passive mode, this system uses sunlight, and in active mode, it uses its own light to detect the spectral signatures reflected from the water surface in the visible (VIS), near-infrared (NIR), and shortwave-infrared (SWIR) parts of the spectrum [
20,
28]. Therefore, this system could be used to more effectively and efficiently monitor spatial and temporal variations that take place in surface groundwater quality based on the substantial changes that take place in the reflectance signature from the water surface in the three parts of the spectrum.
Several studies have found a close relationship between different water quality characteristics and spectral reflectance from the water surface at specific wavebands, especially in the VIS and NIR bands. For example, Gitelson et al. [
15] concluded that the spectral reflectance in the 700–900 nm range is the optimal spectrum range for estimating the concentration of TSS using remote sensing. Vinciková et al. [
22] also reported that the spectral reflectance ratio in the red edge (714 nm) and red (650 nm) regions provide the best estimate for chlorophyll concentrations, while simple spectral reflectance in NIR wavebands, especially at 806 nm, is well correlated with the TSS of water (r
2 = 0.89). A close relationship between colored dissolved organic matter and spectral reflectance in the blue and green regions of the spectrum was also reported by Miller et al. [
29]. In addition, spectral reflectance in the blue (450–510 nm) and green (500–600 nm) wavebands are sensitive to changes in concentrations of total phosphorus in water [
18,
30,
31,
32]. Whereas the majority of researchers have used a space-borne optical remote sensing system to assess water quality indicators, there has been limited investigation of the potential of using a high-resolution spectroradiometer technique for assessing these parameters in situ, especially for groundwater. In situ hyperspectral measurements are useful for monitoring water quality in a reliable manner and could be used to estimate water quality parameters in the water column by estimating the water quality of samples collected from above and below the water surface. This is a very important step for comprehensive assessment and management of groundwater, which is the primary source of irrigation water in arid regions. Such in situ hyperspectral measurements could also be used to calibrate and validate the data acquired by space-borne optical remote sensing systems.
As groundwater quality is affected by several factors, the analysis of hyperspectral reflectance data using an appropriate statistical method is a critical step in unraveling the relationship between these data and specific water quality parameters. Generally, most previous studies have focused on investigating relationships between measured water quality indicators and specific spectral reflectance indices (SRIs). However, these SRIs focus on only 2–3 specific wavelengths. Therefore, instead of formulating individual specific SRIs, a wide range of wavelengths within the full VIS-SWIR spectrum could be applied on an empirical basis to fit the best model for estimating the traits of interest. Several multivariate analyses, such as partial least square regression (PLSR) models, are typically used to create a reliable linear relationship between a set of independent variables, such as full-spectrum ranges and SRIs, and response variables, which are often measured parameters. These analyses consider a set of SRIs as a single independent index and create a more flexible model for indirect estimation of measured parameters when the number of SRIs exceeds the number of measured parameters substantially [
12,
27,
33,
34]. Generally, hyperspectral data allow various multivariate analyses to consider the full VIS-SWIR spectrum and various SRIs. Thus far, few reports evaluate the performance of PLSR models for indirectly estimating IWQIs of groundwater. There is only one study that focuses on predicting the WQI using a model of support vector regression based on SRIs [
12]. Therefore, we propose that developing reliable models using PLSR based on different SRIs and full-spectrum ranges will help in the precise assessment and management of groundwater quality for arid and semiarid regions.
The primary objectives of this study were to (1) evaluate the quality of groundwater for irrigation purposes using different physiochemical parameters, IWQIs, and hydrochemical facies in two distinct regions of Egypt, (2) identify the most sensitive wavelengths in the VIS-NIR spectrum that correspond to the measured IWQIs using different contour maps, (3) utilize these effective wavelengths for building specific SRIs, (4) examine the potential of these new SRIs and published SRIs for indirectly estimating IWQIs, and (5) compare the performance of SRIs and different PLSR models, which are based on SRIs or full-spectrum ranges (302–1148 nm), for estimating IWQIs.
5. Conclusions
In this study, the groundwater quality of the El Fayoum depression in the Western Desert and the Central Nile Delta was assessed traditionally using different physiochemical parameters, IWQIs, and hydrochemical facies in conjunction with non-destructive high-throughput passive sensing as a rapid and cost-effective assessment tool. Spectral reflectance water surface data were used to develop different SRIs, and the performance of these SRIs and different PLSR models based on SRIs or the full-spectrum range were compared for their assessment of IWQIs. The results of the study provided several conclusions. The groundwater of the WD showed more substantial variation in physiochemical parameters and IWQIs than the CND. According to the IWQ, %Na, SAR, SSP, PS, and KI, approximately 6.7%, 26.7%, 60.0%, 46.7%, 6.7%, and 53.3% of groundwater samples in the WD are suitable for irrigation purposes and classified low restriction, good, excellent, good, excellent to good, and suitable, respectively, indicating that this water type is only suitable for well-drained soils. However, almost all (85–100%) groundwater samples of the CND are very suitable for irrigation. Based on hydrochemical facies, the major groundwater of the WD and CND were Na-Cl and Ca-HCO3, respectively, based on Piper’s diagram. Based on Chadha’s plot, most of the samples of the WD and CND found in field 2 (alkali metals (Na+ + K+) exceed alkaline earth metals (Ca2+ + Mg2+) and field 1 (alkaline earth metals (Ca2+ + Mg2+) exceed alkali metals (Na+ + K+), respectively. The most developed and published SRIs, which were developed as a ratio between combination of UV/VIS, UV/NIR, VIS/VIS, VIS/NIR, and NIR/NIR, exhibited moderate, weak, and moderate to strong correlations with IWQIs and several physiochemical parameters in the WD, CND, and across both regions, respectively. The PLSR models based on all SRIs provided a more accurate estimation of different IWQIs in both the Cal. and Val. datasets than those PLSR models that were based on the full-spectrum ranges (302–1148 nm), and both PLSR models were better than the individual SRIs at estimating IWQIs. In addition, the PLSR models that were based on SRIs provided a good relationship between calculated and predicted values for all parameters of IWQIs. The results of this study, which has rarely been conducted in irrigated arid and semiarid regions, provide useful insights for future analyses on the assessment and management of groundwater quality in these regions, especially when using a high-throughput remote sensing tool.