Interpolation methods for spatial distribution of groundwater mapping electrical conductivity Authors: Saeed Salehi, Reza Barati, Mozareza Baghani, Saeed Sakhdari, Mohsen Maghrebi Published: 2024 ABSTRACT This study was carried out to develop a conceptual framework for determining the best interpolation method which mainly is employed to calculate the variability maps of electrical conductivity (EC) in neighboring regions. The considered case study is parts of the Khorasan Razavi province, Iran (including five aquifers Kashmar, Fariman, Doruneh, Sarakhs and Joveyn). In the first step, the empirical variogram (semi-variogram) was computed for the study area. The methods of the variability of a variable with spatial or temporal distance were considered to measure the semi-variogram function. In the next step, the best variogram model (e.g. spherical, exponential or Gaussian) was considered in the Geographic Information System (GIS) environment and f for the Environmental Sciences (GS+) software. By plotting the semi-variogram in GS+ program based on different method as Global Polynomial Interpolation (GPI), Inverse distance weighing (IDW), Radial basis function (RBF), Kriging method, Global Polynomial Interpolation (GPI), Local Polynomial Interpolation (LPI), the best variogram model fitted to spatial structure of the EC. Finally, by considering the acceptable range for different parameters which impact on EC and evaluating their impacts by scaling, the best interpolation method has been selected for that area for employing their neighborhood basin. Result indicated that the precipitation located within the range of 140 to 180 mm, RBI has the priority. This process is continued for all 14 parameters and eventually one method gets the most points.Supplementary InformationThe online version contains supplementary material available at 10.1038/s41598-024-81893-y. Introduction Recent anthropogenic activity like industrialization or land use change pattern threat agricultural area via water detrition. This challenge is much more important in groundwater resources due to its sustainability and participation in agricultural applications1–6. Groundwater is always considered as the one of accessible and cheapest sources to provide the requirements of the water, especially in arid and semiarid regions. These facts show the importance of providing an accurate picture of the status of groundwater resources for future planning. The proper exploitation can perform an important role in sustainable development of the society. However, over exploitation of the groundwater can cause irreversible damage to this valuable resource which can impact directly on the quality and quantity of the water. Overexploitation of the groundwater, pouring the municipal, husbandry and industrial wastewaters into plains has contaminated these valuable resources7–11. The population growth in recent decades causes a lot of hostile conditions in arid regions like raising the EC through the agricultural areas. For the proper management of water resources, it is important to know the spatial variation of qualitative parameters of the groundwater like EC. In the past for analyzing of variability of groundwater parameters, classic statistical approaches were applied which were very time consuming and costly. Nowadays, geo-statistics is used as an efficient technique to study spatial distribution of qualitative parameters12. Also, it is supposed that some novel and new solutions were introduced based on the numerical models to assess the alluvia aquifer by using the genetic DRASTIC model13. Geo-statistics is introduced as a process which the value of a quantity in the considered point is estimated by employing the value of the same quantity in another given point14–18. It should be noted that, recently, artificial intelligence technologies have been widely used in the area of environmental pollution controls [e.g.19–22. However, most of these technologies suffer from time-consuming validation, black-box modeling without considering physical concepts beyond the considered problem, failing to achieve the desired accuracy, etc., as discussed by23. Using the interpolation methods is always known as an effective approach to make relation between field data like groundwater level and the level of the electrical conductivity (EC). Hu et al.24 plotted the map of the groundwater contaminated by nitrate which was generated by employing the indicator Kriging (IK) and the ordinary Kriging (OK) methods. Their results indicated that the IK method is useful to evaluate the risk of the nitrate in groundwater and the OK method could not correctly reflect the concentration of nitrate through the groundwater. They also found that the regions with waste water irrigation were featured with high level EC. Measurement of the nitrate concentration suggested that the high level of EC in regions was associated with high values of nitrate concentration. Finally, they suggested that the assessment of the threat of the nitrate pollution could be useful to manage the groundwater resource. Finding the consequences of the human activities on the underground pollution is one of important issues that can be studied by employing the interpolation techniques. Goovaerts et al.25 compared multi-Gaussian and indicator Kriging for modeling the spatial probabilistically distribution of arsenic concentrations in groundwater of southeast Michigan. Ta et al.26 by employing the geospatial analysis of spatiotemporal variability of groundwater level analyzed the impacts of the war on the pollution of the groundwater in real basin. They expounded that the Kriging interpolation technique can estimate the directions of the groundwater with good agreement compared by the reality. They presented that the fluctuations of the flow direction can be predicted based on the Kriging method. Adhikary et al.27used ordinary Kriging and indicator Kriging methods for spatial analysis of bicarbonate, calcium, chloride, EC, magnesium, nitrate, sodium, and sulphate at Najafgarh, NCT of Delhi. The thematic maps of all the groundwater quality parameters indicated an increasing trend of pollution from the northern and western part of the study area towards the southern and eastern part. Kazemi and Hosseini28 estimated the spatial distribution of the six heavy metals such as Arsenic, Cadmium, Copper, Mercury, Plumbum, and Zinc through the sediment of the Caspian Sea in Volga delta. The results of the Ordinary Kriging method (OK) showed that the spatial variability of the pollution of the six heavy metals which were fluctuated between maximum and minimum values. Seyedmohammadi et al.29 focused on the determination of the most suitable interpolation method for spatial analysis of groundwater EC in central regions of Guilan province. They used inverse distance weighting (IDW), global polynomial interpolation (GPI), local polynomial interpolation (LPI), radial basis function (RBF) and ordinary Kriging (OK) methods for estimation of groundwater EC, and they find that the best estimator OK method that consists of several performance evaluation criteria. Kumar30 studied spatial variation of the fluoride and nitrate parameters of groundwater resources. He observed a definite correlation with the geological formations and water quality parameters, which indicated importance of geo-statistics in groundwater management. Rawat et al.31 developed a new approach to define a patio-temporal function (NDDI) for groundwater quality parameters (e.g., PH, EC, milligram/ litter of the TH, Ca, and Mg materials). NDDI is defined as a compact index which is used for the analysis of water pollution states for spatial distribution in different time durations. It has the capability of conversion of parameter concentration between − 1 and + 1. They depicted the contour plots of the groundwater quality parameters during the different seasons. The frequency distribution and inter correlation of analysis information of the data bank was coherence. Karami et al.32 used geostatistical methods for spatial analysis of seven main quality parameters including total dissolved solids, sodium adsorption ratio, EC, sodium, total hardness, chloride and sulfate. Their results showed that Kriging method is more accurate than the traditional interpolation methods. Safarbeiranvnd et al.33 evaluated variations of water quality parameters of the Central Plain Aquifer using Kriging and IDW methods. They find that the effect of the structural conditions on aquifer properties such as transmissivity and flow direction were important factors on the spatial variation of water quality parameters. Maroufpoor et al.34 applied geostatistical methods and hybrid intelligent models to predict the spatial distribution of chlorine (Cl), EC, and sodium absorption ratio (SAR) parameters of groundwater. Among Kriging, inverse distance weighting (IDW), and radial basis function (RBF) methods, the Kriging was found more accurate based on the employed comparisons among the interpolated values and field data with same coordination. Bon et al.35assessed the pollution index of the shallow quaternary aquifer of the Lake Chad Basin using a geostatistical model of EC. A spatial heterogeneity of EC was observed where it is possible to distinguish six groups EC. Folorunso36 studied the spatial salinity flow of surface water into coastal groundwater in Okun-Ajah and its environ. Azlaoui et al.37evaluated the suitability of groundwater for agriculture and drinking in the Ain Oussera plains. They created a spatial distribution map for EC using IDW. Bhunia and Shit38 developed a geostatistical-based water quality index to analyze the spatial variation of contamination zone in Surguja district of Chhattisgarh, India. Their results indicated that north and north-east of the district have low groundwater quality and the north-west and central part of the district have very good groundwater quality. Ghachoui et al.39 assessed the groundwater status of the Souk El Arbaa by collecting 14 samples in 2022 from different locations and analyzing their physicochemical characteristics. They used IDW tool spatial distribution mapping. Jamal et al.40 evaluated spatial distribution and hydrogeochemical of tubewell water of six unions of Kaligonj upazila in Satkhira district using IDW. The background of the present study shows that the use of geo-statistics methods is unavoidable in the analysis of spatial variation of groundwater quality as the decision making is based upon discrete observation sampling points. Tam and Wong41 findings illustrated that the geostatistical methods are appropriate for an area that depends on the type of variable and regional factors and could not be generalized to other regions. To the best of knowledge of the authors, there is no available practical framework to select appropriate geo-statistics methods for a given aquifer with specific hydrological and hydrogeological features. These facts motivated the authors to develop a conceptual framework to select the best interpolation method for spatial distribution of groundwater EC. The assumptions used in this study are, predicting trend of qualitative variation in an aquifer in the coming years and taking protective and managerial action before the crisis is possible, appropriate areas for exploitation of groundwater for drinking and irrigation purposes can be determined, and geo-statistics methods are accurate enough for analysis of spatial variation of qualitative parameters of groundwater. The following tasks were accomplished as variation of EC was studied in five aquifers (including Kashmar, Fariman, Doruneh, Sarakhs and Joveyn) of Khorasan Razavi province, Iran, spatial and temporal variation of parameters in these five regions with seven different geo-statistics methods were analyzed and the best method for interpolation of EC was selected in each region based hydrologic and meteorological data, examination of trend of variations in each region based on the physics of the region (hydrology and hydrogeology), developing a conceptual framework between the best interpolation method and hydrologic and hydrogeological factors in each region. Since we need a great deal of geographic and qualitative data for finding the best interpolation method, GIS was used as a powerful tool to create spatial variability maps. The considered interpolation methods are IDW (inverse distance weighing), RBF (radial basis function), GPI (global polynomial interpolation), LPI (local polynomial interpolation), Kriging and Cokriging. Interpolation methods in GIS are divided into two groups of deterministic and geo-statistics. In deterministic methods, mathematical functions are used, but in geostatistical methods statistical methods are used; however, in to the considered functions involved the main part of this solution. The interpolation methods are included IDW (inverse distance weighing), RBF (radial basis function), GPI (global polynomial interpolation), LPI (local polynomial interpolation) are deterministic and Kriging and Cokriging which are classified as the geo-statistics methods43–50. There are many sources of uncertainty in the selection of interpolation methods for different parameters including groundwater EC, and these probably lead to some issues in spatial distribution assessment. For example, different interpolation methods for spatial distribution of a specific parameter are suitable in different regions. In addition, different interpolation methods show better performance in a specific region in different seasons. These facts motivated the authors to discover other ways to improve spatial distribution mapping, including linking between interpolation methods selection and hydraulic, hydrologic and meteorological characteristics of considered regions. To the best of the authors’ knowledge, this is the first study to analyze this issue, and the first to propose an algorithm for choosing the best interpolation method for spatial distribution of groundwater mapping electrical conductivity by considering physical criteria instead of only using mathematical concepts. This research aims to reduce uncertainty associated with spatial interpolation methods. As following, the general conceptions of the employed interpolation method were discussed. Initially, the empirical variogram (semi-variogram) was computed for the study area based on the EC values. Due to the best variogram model (spherical, exponential or Gaussian) by considering the GS+, some different interpolation methods including: IDW, RBF, OK, SK, UK, GPI and LPI were applied in GIS to find out the best of them based on the “cross validation” technique. Secondly, by using the selected method, the variability EC maps of the study area were plotted. Then, hydraulic, hydrologic and meteorological characteristics of considered studied regions were analyzed and, and by discussing achieved results a conceptual framework to select the best interpolation method for spatial distribution of groundwater EC will be proposed. Finally, the proposed algorithm can be used to rank the interpolation methods for other new area (e.g., some region which were located near the studied basin). Most of the tools used in GIS are recurrent and time consuming. Thus, for easier solution of a problem in GIS, instead of using conventional methods, the programmability potential of that can be used. One of the programing languages that is compatible with GIS the most is Python. In this way, instead of producing different layers and entering data into GIS, the process of problem solving and getting results is programmed and part of the program is allocated to entering data which is EC. The benefit of this work is that programing the EC of the groundwater based on the environmental, morphologic and hydraulic parameters can ease the calculation and estimate of this parameter for different basin. The present study, the variations of EC were defined as several important field data such as the annual precipitation, temperature index, evaporation from the evaporation pan, flood and discharge, average height of the study area, SAR (sodium absorption ratio) and EC of the wells of the study area, percentage of clay, percentage of the sand and silt, bedrock inclination and the basin slope, hydraulic conductivity, storage coefficient of the basin and also the types of the land usage. Finally, result of this study indicated that the proposed algorithm and considering the hydrologic and meteorological data can proposed the suitable method of interpolation to attain the EC values nearby the studied basin. Inverse distance weighing (IDW) It is a deterministic method suggested which is based on the assumption that as distance between the position of the observation and estimate increases, its influence on the interpolated value decreases. Let r = (x, y), be the position of a given point, the IDW formula would be as following:1\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$F(r) \Sigma_{k=1}^{n} w(r_k)(f(r_k))$$\end{document} where F (r) is interpolated value at, n is the number of observations, f (rk) is value observed at location rk= (xk, yk) at station i and w (rk) w (rk) is the weight of that station obtained from the following relation:2\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$W(r_k)= \frac{d_k(r)^{-p}}{\Sigma_{K=1}^{n}d_{k}(r)^{-p}}$$\end{document} where dk(r) is Euclidean distance between r and rk obtained from \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$d_{k} (r) = \sqrt {(x - x_{k} )^{2} + (y - y_{k} )^{2} }$$\end{document} , p = sole model parameter that indicates the rate of lessening of influence of observations with distance. Value of pcan have an enormous impact on the result of interpolation. More details on weights have been offered by51–53. Radial basis function (RBF) An important group of interpolation methods which fall under the category of RBFs were offered by54,55. As following:3\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$F(r)=\Sigma^{n}_{k=1}\lambda_{k}\phi (|| r-r_{k}||)$$\end{document} where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\varnothing\:$$\end{document} is definite positive RBF, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:r\parallel-{r}_{k}\parallel$$\end{document} is the Euclidean norm and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\lambda\:}_{k}$$\end{document} is the set of unknown weights obtained as following:4\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$F(r_k)=f(r_k)$$\end{document} Combining the Eqs. (3) and (4) yields a system of linear equations as following:5\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\phi A = \varphi$$\end{document} where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\varnothing\:$$\end{document} is a n×n matrix called the interpolation matrix and A = λk and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\varnothing\:=\left[{f}_{k}\right]$$\end{document} are respectively, n×1 columns of (unknown) weights and observed vales. The RBF interpolation (obtained by solution of Eq. (5) for A) is dependent on the basis of function \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\varnothing\:$$\end{document} chosen (choices are usually, inverse multiquadric, multilog, natural cubic spline and thin plate spline) which are dependent, on the Euclidean distance between points r and rk and the priori specification of R as a shape parameter. More details can be found in56–58. Kriging method The theoretical basis for the method was put forward by the French mathematician Georges Matheron in 1960, based on the Master’s dissertation of Danie G. Krige, who was a mining engineer in South Africa. Krige was trying to estimate the most likely distribution of gold based on the samples obtained from boreholes. Kriging is an interpolation method like IDW, in which the weights, unlike IDW are determined from spatial and statistical relations obtained from the plot of empirical semi-variogram defined by:6\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\hat{r}(h) = \Sigma_{k=1}^{n(h)}[f(r_k)-f(r_k+h)]^{2}$$\end{document} where n (h) is the number of sample pairs f(rk), f(rk+h), which are separated by distance h. \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\widehat{\gamma\:}\left(h\right)$$\end{document}is an empirical function that is fitted to a specific functional form (variogram model) and some of its most commonly used models are linear variogram59, exponential60, rational quadratic60, wave60, logarithmic61, pentaspherical62and cubic62. The assumption of the ordinary Kriging (OK) is that in the local neighborhood of each observation point, the mean is constant63. Resulting in Eq. 1 with the condition:7\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Sigma_{k=1}^{n} W (r_k)=1$$\end{document} To make sure that the estimate is uniformly unbiased when making the variance of estimated error minimum. Global polynomial interpolation (GPI) In this method a mathematical function is used for fitting a smooth surface to sample points. The GPI surface gradually changes from region to region, over the favorite area, so the global trend in the data is captured. In comparison with IDW, in GPI, calculation of the predictions takes place using the whole dataset instead of using the measured points within neighborhoods. A single global polynomial doesn’t fit a surface with a varying shape very well and multiple polynomial planes would do a better job64. Local polynomial interpolation (LPI) Unlike GPI, LPI fits the local polynomial and uses the points only within the given neighborhood, rather than the whole data. Then the neighborhoods overlap and the surface value at the center of neighborhood is set to be the predicted value. GPI is mostly suitable for determining long-range trends in dataset, while LPI can be used for generating surfaces that capture the short-range variation64. Study areas and data This study aims to analyze the salinity of groundwater reflected by groundwater EC in five regions of Khorasan Razavi Province65–67 including Kashmar, Fariman, Doruneh, Sarakhs and Joveyn. Figure 1 shows the locations of the considered aquifers. Several aquifers with different hydrological and hydrogeological features have been considered to develop a conceptual framework for selecting the best interpolation method for spatial distribution of groundwater EC. Fig. 1The location of five aquifers of the study areas. The location of five aquifers of the study areas. The EC data of five aquifers was obtained from the dataset provided by Regional Water Company of Khorasan Razavi in 2008, 2013 and 2018. EC is an indicator of ability of water to pass electricity through. Distilled or pure water can’t pass electricity. It is the ions in water that let the electricity pass. So, there is a relation between EC and concentration of dissolved solids. The EC of pure water in 25 degrees Celsius is 0.056 µmho/cm and in surface water and groundwater water it may vary from 90 to several thousand µmho/cm. First of all, the data with wrong x, y coordinates located outside the study areas must be removed. At some points, there are even more than one sample. And since in ArcGIS, only one value can be assigned to each point, the average of the EC samples has to be calculated. The final data was saved in an Excel sheet with 3 columns, xutm, yutm, EC as a Unicode Text (.txt) that could be imported into ArcGIS and used for interpolation. In order to normalizing the data, EC data of each region was separately explored statistically by using SPSS. For using data in ArcGIS, the data must be normalized and if the data is not normal, it must be normalized using different methods. Table 1 represents the statistical parameters of data such as, mean, skewness, kurtosis, maximum and minimum of values, which were calculated using SPSS. Table 1Results of statistical analysis summary of EC data for five aquifers.YearAquiferMinimumMaximumMeanStad. devSkewnessKurtosis2008Joveyn22012,78024512777.81.464.90Sarakhs180080003998615.80.620-0.38Doruneh58610,3507253.41996.7-0.943.89Fariman48442001742.8831.45192.619Kashmar36319598080.9357.11.293.952013Joveyn35014,8003061.93125.61.410.92Sarakhs70065,0004017.74416.611.41156.4Doruneh69014,180758.62567.9-0.413.93Fariman39089501848.11088.71.853.99Kashmar39577101113.2981.53.5114.812018Joveyn50076001398.31014.23.5017.18Sarakhs30013,0003778.12257.11.553.05Doruneh245011,1007855.01994.0-0.620.46Fariman22080272114.51119.21.635.88Kashmar43022,7502183.33908.43.139.16 Table 1Results of statistical analysis summary of EC data for five aquifers.YearAquiferMinimumMaximumMeanStad. devSkewnessKurtosis2008Joveyn22012,78024512777.81.464.90Sarakhs180080003998615.80.620-0.38Doruneh58610,3507253.41996.7-0.943.89Fariman48442001742.8831.45192.619Kashmar36319598080.9357.11.293.952013Joveyn35014,8003061.93125.61.410.92Sarakhs70065,0004017.74416.611.41156.4Doruneh69014,180758.62567.9-0.413.93Fariman39089501848.11088.71.853.99Kashmar39577101113.2981.53.5114.812018Joveyn50076001398.31014.23.5017.18Sarakhs30013,0003778.12257.11.553.05Doruneh245011,1007855.01994.0-0.620.46Fariman22080272114.51119.21.635.88Kashmar43022,7502183.33908.43.139.16 Results of statistical analysis summary of EC data for five aquifers. Methodology and method Q-Q plots In the next step, the Q-Q plots were drawn in ArcGIS. We use these plots to determine if the data is normal. If the data is distributed about the line y = x, the data (EC) distribution is normal. If the line y = x cannot be fitted through the data, it must be normalized. Using not normalized date in GIS would lead to great error in interpolation. In this study, data of three regions of Kashmar, Joveyn and Fariman was normalized, while two other regions followed normal distribution and left intact. It must be noted that ArcGIS produces the results based on the real input data and doesn’t involve the logarithm taking into the results. In fact, the interpolation takes place on the normalized data but the results are presented in real scale. In Fig. 2, the normalized groundwater EC data for the study area is presented, as it can be seen, the data has good agreement with normalized data. Fig. 2Normalizing data: (a) Joveyn 2008, (b) Sarakhs 2008, (c) Doruneh 2008, (d) Fariman 2008, (e) Kashmar 2008, (f) Joveyn 2013, (g) Sarakhs 2013, (h) Doruneh 2013, (i) Fariman 2013, (j) Kashmar 2013, (k) Joveyn 2018, (l) Sarakhs 2018, (m) Doruneh 2018, (n) Fariman 2018, (o) Kashmar 2018 Normalizing data: (a) Joveyn 2008, (b) Sarakhs 2008, (c) Doruneh 2008, (d) Fariman 2008, (e) Kashmar 2008, (f) Joveyn 2013, (g) Sarakhs 2013, (h) Doruneh 2013, (i) Fariman 2013, (j) Kashmar 2013, (k) Joveyn 2018, (l) Sarakhs 2018, (m) Doruneh 2018, (n) Fariman 2018, (o) Kashmar 2018 Finding the best variogram model By plotting the semi-variogram in GS+ program, the best variogram model fitted to spatial structure of the parameter under analysis (i.e. EC) is determined with least Residual Sum of Squares (RSS) and maximum coefficient of determination R2. Figure 3 presents the results of analysis of semi-variogram for normalized data, the values and elements of a variogram could be obtained for each given region and year. These values are presented in Table 2 for considered regions. Fig. 3semi variance and the best fitted model for EC in 2008, 2013 and 2018 for five regions: (a) Joveyn (spherical), (b) Sarakhs (exponential) (c) Doruneh (linear) (e) Fariman (spherical) (f) Kashmar (spherical). semi variance and the best fitted model for EC in 2008, 2013 and 2018 for five regions: (a) Joveyn (spherical), (b) Sarakhs (exponential) (c) Doruneh (linear) (e) Fariman (spherical) (f) Kashmar (spherical). Table 2Specifications of semi-variogram fitted to EC at 2008.RegionyearRSS R 2 C/C0 + CrangeC0 + C(sill)C0(nugget effect)Joveyn20080.061.51682,6000.950.700.78Sarakhs20080.00360.4530.57626,4700.317800.13490Doruneh20086.697e + 110.9911.00030,45021,120,00010,000Fariman20081.291e + 100.970.8313,1001,207,000198,000Kashmar20083.113e-0.40.960.9285600.180.01Joveyn20130.1720.900.9784,7001.410.03Sarakhs20130.010.280.4439,1000.4480.22Doruneh20134.22e140.5980.9981,10041,300,00010,000Fariman20136.63e-30.9790.9881,1000.920.0180Kashmar20134.97e-30.9290.99711,5800.350.001Joveyn20181.460.0080.0111266.70.520.52Sarakhs20180.070.120.5064,5000.510.25Doruneh20181.072e + 130.4100.67514,5504,770,000154,000Fariman20180.660.93144,5003.010.001Kashmar20184.66e-30.870.9892300.210.003 Table 2Specifications of semi-variogram fitted to EC at 2008.RegionyearRSS R 2 C/C0 + CrangeC0 + C(sill)C0(nugget effect)Joveyn20080.061.51682,6000.950.700.78Sarakhs20080.00360.4530.57626,4700.317800.13490Doruneh20086.697e + 110.9911.00030,45021,120,00010,000Fariman20081.291e + 100.970.8313,1001,207,000198,000Kashmar20083.113e-0.40.960.9285600.180.01Joveyn20130.1720.900.9784,7001.410.03Sarakhs20130.010.280.4439,1000.4480.22Doruneh20134.22e140.5980.9981,10041,300,00010,000Fariman20136.63e-30.9790.9881,1000.920.0180Kashmar20134.97e-30.9290.99711,5800.350.001Joveyn20181.460.0080.0111266.70.520.52Sarakhs20180.070.120.5064,5000.510.25Doruneh20181.072e + 130.4100.67514,5504,770,000154,000Fariman20180.660.93144,5003.010.001Kashmar20184.66e-30.870.9892300.210.003 Specifications of semi-variogram fitted to EC at 2008. Validation of interpolation methods To validate the interpolation methods, the “cross validation” technique was used. In this way, from among the dataset a sample point is removed, then using all interpolation methods, the value is estimated at that point. The estimates are compared with the real observed value and the closest value to the sample is assumed to be the result of the best interpolation method15. In this study, the parameters used for comparing the estimates and sample are: the ratio of the nugget effect to range, coefficient of determination of variogram model (r2) and RMSE (Root Mean Square Error). The best interpolation method for EC in 2008, 2013 and 2018 Where the best variogram model was chosen, the EC data was input into ArcMap and interpolation was done using: simple kriging (SK), ordinary Kriging (OK), universal Kriging (UK), Inverse distance weighing (IDW), Global polynomial interpolator (GPI), Local polynomial interpolator (LPI) and Radial based function (RBF). Then using the “cross validation” technique and RMSE (lower RMSE indicates a smaller error) the best interpolation approach was determined for each region. The results are presented in Table 3. The best interpolation method for five regions were presented in Table 4 based on the maximum value of RMSE in Table 3. As it can be seen, for different years, different methods have been chosen and it might be because of the different number of samples with different locations in the same area. So, if there was the same number of samples with the same coordinates, sure there was more similarity between the results. Table 3Results of the best interpolation method for EC for study areas in 2008, 2013 and 2018.Study areaRMSE calculated for each interpolation methodYearSKOKUKIDWGPILPIRBFDoruneh20082189.82342.213676.9225931382761.17922.6Sarakhs20084162.54197.84258.84114.64171.24137.04169.4Joveyn20081275.31262.71008.01103.62587.21033.61059.8Fariman2008280.8256.2255.2257.7432.1483.5259.1Kashmar2008629.9616.9625.6652.4784.9631.1642.4Doruneh20131686.71734.01715.21771.02542.42716.61560.6Sarakhs20131103.21039.51845.711301401168.31053.8Joveyn201312211.91206.21105.11162.22278.31049.21156Fariman2013445.2415.81735.4429591.8425.2425.2Kashmar2013224.4221.4225.1234313.9224.2229.3Doruneh2018210320241779.12209.12324.22060.12042.2Sarakhs20181216.11360.31360.21410.11788.21327.31399.2Joveyn20181153.21330.11331.11154.21243.21168.41153.2Fariman2018448349331.1307.2568.1292.2358.2Kashmar2018811.1742.2745.1835.31102.21101.2795 Table 3Results of the best interpolation method for EC for study areas in 2008, 2013 and 2018.Study areaRMSE calculated for each interpolation methodYearSKOKUKIDWGPILPIRBFDoruneh20082189.82342.213676.9225931382761.17922.6Sarakhs20084162.54197.84258.84114.64171.24137.04169.4Joveyn20081275.31262.71008.01103.62587.21033.61059.8Fariman2008280.8256.2255.2257.7432.1483.5259.1Kashmar2008629.9616.9625.6652.4784.9631.1642.4Doruneh20131686.71734.01715.21771.02542.42716.61560.6Sarakhs20131103.21039.51845.711301401168.31053.8Joveyn201312211.91206.21105.11162.22278.31049.21156Fariman2013445.2415.81735.4429591.8425.2425.2Kashmar2013224.4221.4225.1234313.9224.2229.3Doruneh2018210320241779.12209.12324.22060.12042.2Sarakhs20181216.11360.31360.21410.11788.21327.31399.2Joveyn20181153.21330.11331.11154.21243.21168.41153.2Fariman2018448349331.1307.2568.1292.2358.2Kashmar2018811.1742.2745.1835.31102.21101.2795 Results of the best interpolation method for EC for study areas in 2008, 2013 and 2018. Result and discussion Suitable interpolation methods Based on the results of Table 4, the variability maps are produced as following. Table 4The best interpolation method for five regions in 3 years.year200820132018DorunehRBFSKUKSarakhsOKIDWLPIJoveynLPIUKSKFarimanOKUKGPIKashmarOKOKOK Table 4The best interpolation method for five regions in 3 years.year200820132018DorunehRBFSKUKSarakhsOKIDWLPIJoveynLPIUKSKFarimanOKUKGPIKashmarOKOKOK The best interpolation method for five regions in 3 years. Doruneh basin Figure 4 shows the EC variability maps of Doruneh in 2008, 2013 and 2018. The highest level of EC can be seen in the west and north east of this city and the lowest level in the south east and the center. Considering the mountains in the north of this sub basin, it seems that in regions with high altitude the EC reach higher values. Also, the maximum salinity of the area has decreased over the years, but the expanse of the saline soils has increased over the time. It can be deduced that the salinity is being transferred from regions with saline soils to regions with no salinity, which could be affected by the precipitation. Fig. 4EC map of Doruneh (µmho/cm) (a) Calculated by RBF (2008), (b) calculated by SK (2013), and (c) calculated by UK (2018). EC map of Doruneh (µmho/cm) (a) Calculated by RBF (2008), (b) calculated by SK (2013), and (c) calculated by UK (2018). Sarakhs basin The highest EC level was observed in the north and the lowest in the east and center in 2008 (see Fig. 5a). Comparisons in 2013 and 2018 (see Fig. 5b and c) illustrated that the salinity decreased in the north and became more uniform. Since 2013 to 2018, the peak of salinity has been transferred from the northern to the central part, while the minimum level remains in the west. The maximum and minimum of EC level in 2013 are related to the northern and central parts. In 2018, the maximum EC is related to central and the minimum EC to the eastern regions. Considering the maps, it is concluded that over the time, the EC is increasing in the region. Since Dousti Dam and a river are located near this area, there is a very high hydraulic gradient in the region which could have a great impact on the EC of the water. Hydraulic gradient of the river bed, leaches the salts and the EC is decreased. Fig. 5EC map of Sarakhs (µmho/cm). (a) Calculated by OK in 2008, (b) calculated by IDW in 2013, and (c) calculated by LPI, in 2018. EC map of Sarakhs (µmho/cm). (a) Calculated by OK in 2008, (b) calculated by IDW in 2013, and (c) calculated by LPI, in 2018. Fariman basin In 2008, the maximum EC is found in the south east of the city and the minimum was located in the North West, (see Fig. 6). Also, in 2013, the maximum EC was still in the south east and the minim was constantly located in the North West and center of the city. In 2018, the western, south western and central regions of this city, had the highest level of EC and in north east the minimum EC has been reported. In total, it can be deduced that the EC has decreased over the study period in Fariman region. Considering the dominant slope of the region which is from the North West to the south east, the groundwater washes the mixed salt into the soil and increases the concentration of them over the time. Fig. 6EC map of Fariman (µmho/cm) (a) Calculated by OK in 2008, (b) calculated by IDW in 2013, and (c) calculated by LPI, in 2018. EC map of Fariman (µmho/cm) (a) Calculated by OK in 2008, (b) calculated by IDW in 2013, and (c) calculated by LPI, in 2018. Joveyn basin In 2008, the maximum EC is seen in the west and east of this city and the central, northern and southern regions enjoyed the minimum EC (Fig. 7). Furthermore, the maximum EC was still related to the east and expanding to the center in 2013. In 2018, the maximum EC is in the west and the minimum is still in the south, center and north. Considering the results of the 10-year study period (2008–2018) shows that, the maximum EC in the region has decreased, but the value of the minimum EC has been increased significantly. Also, the expanse of the regions with highest EC has reduced. To explain the happened status for EC, as it can be seen in Fig. 7, Joveyn is located between two mountain ranges, one in the north and the other in the west. Each one with its specific water yield can have impact on the EC of the region. The lowland located in the north east, can be a place where the ions carried by groundwater can accumulate and increase the salinity of the water. Fig. 7EC map of Joveyn (µmho/cm) (a) Calculated by OK in 2008, (b) calculated by IDW in 2013, (c) calculated by LPI in 2018. EC map of Joveyn (µmho/cm) (a) Calculated by OK in 2008, (b) calculated by IDW in 2013, (c) calculated by LPI in 2018. Kashmar Basin In 2008, the maximum EC was located in the south of this regiom and the north of this area were allocated to the minimum EC. Also, the maximum and minimum EC were the same as 2008 in2013. As it can be seen in Fig. 8c, the maximum EC was switched to the south and the minimum was in the north and North West, constantly. By considering the corresponding sub-plots (see Fig. 8), it can be inferred that the EC was increased over the time in 2008 to 2018. Furthermore, the reduction of expanse of the areas with lower EC diminished over the time which can show that the salinity is transferring from other regions to the south of this region. A closer look at the region shows that in places where there are narrow valleys, the water of the ephemeral rivers is trapped and the EC values are increased consequently. Fig. 8EC map of Kashmar (µmho/cm) (a) Calculated by OK in 2008, (b) calculated by IDW in 2013, (c) calculated by LPI in 2018. EC map of Kashmar (µmho/cm) (a) Calculated by OK in 2008, (b) calculated by IDW in 2013, (c) calculated by LPI in 2018. Relationship between meteorological parameters and EC After finding the best interpolation method for the study areas, finding the best interpolation method for a new neighboring region is considered which can assess to predict the best interpolation by using its meteorological parameters. An algorithm has been designed to generalize the best interpolation method of the study area to other regions in Fig. 9. For solving this flowchart an Excel sheet has been designed. The details of the analyses of the meteorology, morphology, hydrology and hydraulic data of the study areas are presented as the supplementary sections. In flowchart shown in Figs. 9 and 14 parameters which are effective on EC in any region were considered as the input data. Classification of the parameters for giving points is depicted in Fig. 10. First of all, in Khorasan Razavi, the best interpolation method was determined for several sub basins using the “cross validation” technique in GIS. In the next step, the best interpolation method was generalized to the neighboring sub basins using the proposed algorithm. The algorithm is based on classification of all 14 parameters and putting the interpolation methods into any of them. For example, if the precipitation of the new sub basin is 130 mm, the priority is given to the method which located in the range of 100 to 140 mm (see Fig. 10). For a new sub-basin, the values of 14 parameters are studied to see which class they fall in and based on that, the new sub basin gains some points for each interpolation method. The interpolating method that has got the most points is set to be the best interpolation method for the new sub basin. If the precipitation in a basin is 150 mm, it is located within the range of 140 to 180 mm as it can be seen in this class, RBI has the priority. This process is continued for all 14 parameters and eventually one method gets the most points. An Excel sheet has been designed that shows the procedure. Also, it should be mentioned that the used hydraulic and methodology parameters were presented into the supplementary section. Fig. 9Proposed algorithm for giving points and choosing the best interpolation method. Proposed algorithm for giving points and choosing the best interpolation method. Fig. 10Classification of parameters for the algorithm. Classification of parameters for the algorithm. In the following sections the hydrology, meteorology and hydraulic data of the study area are presented. These parameters are classified and the new region falls into one of these classes and if it gains the necessary points, the interpolation method of that region is assigned to that. Conclusion By plotting the semi-variogram in GS + program based on different method as Global Polynomial Interpolation (GPI), Inverse distance weighing (IDW), Radial basis function (RBF), Kriging method, Global Polynomial Interpolation (GPI), Local Polynomial Interpolation (LPI), the best variogram model fitted to spatial structure of electrical conductivity. Finally, by considering the acceptable range for different parameters which impact on EC and evaluating their impacts by scaling, the best interpolation method has been selected for that area for employing their neighborhood basin. Result indicated that the precipitation located within the range of 140 to 180 mm, RBI has the priority. This process is continued for all 14 parameters and eventually one method gets the most points. Finally, by using the provided algorithm, the best interpolation methods were proposed based on the input parameters. It worth to be mentioned that using the interpolation methods based on the environmental parameters can provide a numerical process to predict the EC. In future studies, the spatial distribution of groundwater EC can be studied by the use of equations governing the flow of water and solutes in the soil and the results compare with the present approach. Also, an optimization model can be added in the presented methodology. Moreover, other interpolation methods can be used to increase the applicability of the present methodology. Electronic supplementary material Below is the link to the electronic supplementary material. Supplementary Material 1 Supplementary Material 1