Groundwater Level Prediction through GMS Software – Case Study of Karvan Area, Iran

This area with the code 4207 is a section of the Gavkhnoi Basin, Isfahan Province, Iran (Fig. 1). The geographic status of the area is of 32°10′ to 33°40′ N and 40°20′ E. The subject area is 225 km², of which 100 km² is mountainous and 125 km² is plain, with an average altitude of about 2,400 m a.s.l. Morghab River is a seasonal river. According to the Koppen classification, the climate of the area is mild semi-arid. Most of the vegetation coverage consist of orchards and cereals such as wheat and barley. The annual mean total long-term precipitation in the altitudes and plains is 328 and 260 mm, respectively, and the annual mean evaporation at altitudes and plains is 2,059 and 2,313 mm, respectively.

Fig. 1

The subject aquifer contains 496 groundwater discharge sources, including 401 operating wells with annual discharge of about 31.2 Mm³, 61 subterranean with annual discharge of about 10.7 Mm³ and 34 springs with discharge of about 21.6 Mm³ (Table 1). The geographical distribution of these three categories are shown in Figure 2.

Fig. 2

The selected mathematical code in GMS software is of MODFLOW, which is a 3D finite difference model and the nodes are located at the centre of the cell, to which the hydraulic load is assigned. To simulate the subject aquifer, the size of the model grid cells is calculated by sensitivity analysis for all the observed wells with the mean cell size of 400 × 400 m². The count of observed wells and exploration wells are 9 and 2, respectively. The water level data of the observation wells measured on monthly basis is applied for model calibration and validation and the exploration well information is applied to estimate the aquifer hydrodynamic coefficients.

The most important aquifer recharge factors consist of the direct infiltration of the Kordolia dam reservoir, precipitation, infiltration of surface floodwater, groundwater entering aquifer and the consumed water return. The aquifer discharge factors consist of water consumed in agricultural, drinking and industry, rainfall runoff, aquifer evaporation and water discharged from wells, springs and aqueducts, the volumes of which are tabulated in Tables 2 and 3.

The details of groundwater balance of the alluvial aquifer are presented in Table 2 revealing that about 15% of the infiltration of perception is added to the aquifer volume.

With respect to the content of this table, precipitation of 2.6% of its annual volume is added to the aquifer water. For calculating the aquifer balance (groundwater balance) Eq. 1 is applied (Roscoe Moss Company 1990): (1) ΔV=∑inlet−∑outlet=−6.1 Mm3 \Delta V = {\sum\nolimits_{{\rm{inlet}}}} - {\sum\nolimits_{{\rm{outlet}}}} = - 6.1\;{\rm{M}}{{\rm{m}}^3} where: –

ΔV is the water volume change (p m⁻³),

To calculate the changes in water volume in the aquifer Eq. 2 is applied (Roscoe Moss Company 1990): (2) ΔV=Sy×A×h \Delta V = {{\rm{S}}_{\rm{y}}} \times {\rm{A}} \times {\rm{h}} where: –

ΔV is the water volume changes (p m⁻³),

The final volume of the aquifer capacity is 683.8 Mm³, with an average annual reduction of 6.1 Mm³, that is, 1%. The 3D modelling for Karvan aquifer is performed in steady state for one month which is then calibrated. The unsteady model is constructed and calibrated for 68 months, which is verified for the following 18 months, after which the whole modelling period is verified. The error rate is defined for the calibrated and verified model by observing well water level difference at 1 m for both states. The period of this modelling is between January 2013 and February 2020, with about seven-year forecast period (2020–2027). Time steps were considered monthly. The number of layers with boundary conditions are shown in Figure 3.

Fig. 3

To calibrate this model in steady and unsteady states, uncertain parameters should be calibrated through pilot or polygon methods to prevent severe statistical bias on a single parameter. The transferability coefficient at the inlet and outlet boundaries of the aquifer and the transferability of the river waterways network are optimised through polygon method by assigning a negative key number to the conceptual model envelopes and with the initial guess at a reasonable minimum and maximum limit. The optimum values of horizontal hydraulic conductivity parameter with 15 pilots, surface recharge of aquifer with 12 pilots, horizontal anisotropy of hydraulic conductivity with 11 pilots and specific yield coefficient with 15 pilots through pilot point method as samples horizontal hydraulic conductivity are shown in Figure 4.

Fig. 4

Here, the sensitivity analysis method combined with the calibration step is adopted. The results of the sensitivity analysis of the parameters affecting the calibration of the Karvan aquifer indicate the maximum effect of surface recharge at pilot 7 and the high impact of horizontal hydraulic conductivity at pilot 4 (Fig. 5).

Fig. 5

The relative error value of RMS or RMSE is 1.68, indicating high simulation accuracy. To better understand the value of this relative error it should be normalised. To convert RMSE into normalised MSE (NRMSE), Eq. 3 is applied (Mentaschi et al. 2013): (3) NRMSE=RMSE(γmax−γmin) {\rm{NRMSE}} = {{{\rm{RMSE}}} \over {\left( {{\gamma _{\max }} - {\gamma _{\min }}} \right)}} where: –

the numerator is the RMSE value and the denominator is the fractional difference between the highest observed well data in the observation wells and the lowest of the same among all observation wells at 172.06 m yielding the value 0.0097; because it is less than 0.1, the modelling is ideal.

The model is validated for the initial 18 months, and for the total modelling period (86 months) for observation wells is performed as shown, for example, in well #2 in Figure 6.

Fig. 6