Insights into the Interconnected Dynamics of Groundwater Drought and InSAR-Derived Subsidence in the Marand Plain, Northwestern Iran

Abstract

Groundwater drought, a significant natural disaster in arid and semi-arid regions, contributes to numerous consecutive issues. Due to the inherent complexity of groundwater flow systems, accurately quantifying and describing this phenomenon remains a challenging task. As a result of excessive agricultural development, the Marand Plain in northwestern Iran is experiencing both groundwater drought and land subsidence. The present study provides the first in-depth investigation into the intricate link between groundwater drought and subsidence. For this purpose, the open-source package LiCSBAS, integrated with the automated Sentinel-1 InSAR processor (COMET-LiCSAR), was utilized to assess land subsidence. The Standard Groundwater Index (SGI) was computed to quantify groundwater drought, aquifer characteristics, and human-induced disturbances in the hydrological system, using data collected from piezometric wells in a confined aquifer. The results revealed a negative deformation of 65 cm over a 75-month period, affecting an area of 57,412 hectares within the study area. The analysis showed that drought duration and severity significantly influence land subsidence, with longer and more severe droughts leading to greater subsidence, while more frequent drought periods are primarily associated with subsidence magnitude. Multi-resolution Wavelet Transform Coherence (WTC) analysis revealed significant correlations between groundwater drought and InSAR-derived land deformation in the 8–16-month period.

1. Introduction

In recent years, drought has emerged as a significant challenge in arid and semi-arid regions [1]. Drought is a complicated natural disaster characterized by its extensive spatial impact and prolonged duration [2]. Droughts are generally classified into four types, including meteorological drought, agricultural drought, hydrological drought, and socio-economic drought [3]. Groundwater drought represents another distinct type of drought, typically emerging later in the drought propagation process and often influenced by other drought types. The weak and delayed response of groundwater systems to climatic factors can sometimes prevent the onset of groundwater droughts [4]. This type of drought typically develops over a period of months to years and is characterized by an initial decline in groundwater recharge, followed by reductions in groundwater levels and discharge [5]. Global climate change, coupled with human activities such as the expansion and change in agricultural practices, has contributed to the occurrence and exacerbation of groundwater drought [6]. Groundwater drought can result in significant socio-economic impacts, particularly in arid and semi-arid regions. These include decreased water availability for agriculture and drinking purposes, as well as land subsidence [7]. Therefore, the quantification and monitoring of groundwater drought in arid and semi-arid regions are essential for effective water resource management and the mitigation of groundwater drought’s impacts [8]. In order to comprehensively investigate the role of groundwater drought within the scope of studies conducted in the field of drought, bibliometric analysis was conducted. To achieve this, VOSviewer software (version 1.6.20) was utilized to generate a network map illustrating the co-occurrence of keywords extracted from 12,335 scientific documents. A total of 46,078 keywords related to drought and groundwater were identified, with only those occurring at least 50 times being considered for analysis. Figure 1 provides the various research areas related to drought as illustrated as a keyword co-occurrence network.

VOSviewer classifies keywords into four distinct clusters. The first cluster (in red) illustrates analyses and studies related to drought phenomena and their connections with groundwater resources, extreme events, and hydrological events. The second cluster (in green) shows the plant responses to the drought phenomenon. The third cluster (in blue) represents research aimed at unveiling the intricate relationship between climate change and drought, and the fourth cluster (in yellow color) includes keywords with frequent occurrences in the field of drought and water quality. The size of the nodes reflects the weight and significance of the words; larger nodes denote more important words [9]. Although groundwater drought has emerged as a significant challenge in recent years, limited research has been conducted on the occurrence and propagation of droughts within groundwater systems. The analysis of the keyword “groundwater drought” in cluster 1 revealed that it appears only 66 times across various publications. Furthermore, the keyword co-occurrence analysis indicated that “subsidence” and its related terms, as a key consequence of groundwater drought, were not present among the investigated keywords. The limited research on groundwater drought, coupled with the lack of investigation into its effects on subsidence, highlights a significant research gap in this field.

The accelerating rate of land subsidence worldwide has emphasized the necessity for comprehensive monitoring in recent decades. Among conventional monitoring techniques, the station-based Global Positioning System (GPS) is widely utilized due to its ability to provide accurate three-dimensional displacement measurements [10]. However, despite its high accuracy, the GPS method faces significant challenges, including limited spatial coverage, the scarcity of measurement stations, and the substantial costs associated with station establishment [11]. Remote sensing approaches have proven to be an effective tool in monitoring natural disasters, providing timely and accurate data for early warning systems, damage assessment, and disaster response planning [12,13]. Among them, Synthetic Aperture Radar (SAR), has emerged as a reliable alternative for measuring land deformation over the past two decades [14]. This method offers significant advantages over traditional approaches, such as leveling surveys, by eliminating the need for time-intensive field data collection. Additionally, SAR provides higher spatial coverage and denser measurements compared to station-based GPS data, making it a more efficient and comprehensive solution for subsidence monitoring.

Extended and frequent droughts reduce groundwater levels, increasing pressure on the soil and contributing to land subsidence [15]. This factor has prompted researchers to investigate the relationship between these two destructive phenomena in recent years. Among the limited studies conducted, Smith and Knight (2019) [16] estimated that 98% of the total subsidence observed in California, USA, during the 2007–2010 drought was permanent, resulting in a loss of groundwater storage equivalent to 9% of the groundwater used during that period. Miller et al. (2020) [17] demonstrated that a prolonged drought that began in 2012 in California intensified in 2014 and persisted until 2016, significantly accelerating subsidence in the region. Barthelemy et al. (2024) [18] utilized a daily soil wetness index (SWI) to introduce a novel annual drought index that can be associated with subsidence damage. A comparison with claims data revealed that the magnitude of the developed drought index was effective in identifying subsidence events while maintaining spatial consistency. In their comprehensive study of the Ardabil Plain in northwestern Iran, Ghorbani et al. (2022) [19] employed interferometric synthetic aperture radar (InSAR) to monitor long-term subsidence and assess its relationship with groundwater extraction and climate change. Monitoring rainfall and the Standardized Palmer Drought Index (SPDI) for the region indicated an increase in both precipitation rates and the magnitude of drought, as reflected in the SPDI, during the years 2019 and 2020. This trend led to a reduction in land subsidence in the region. Welch et al. (2024) [20] investigated inelastic subsidence in coastal areas through the analysis of decade-long GPS datasets. Results revealed that drought-induced subsidence is most pronounced in open-field areas with expansive soils, while it remains minimal in urbanized regions or areas with non-expansive soils.

In recent years, numerous studies have been conducted to assess the relationship between groundwater levels and land subsidence, contributing to a growing body of literature on the subject [21,22,23,24,25,26]. An analysis of 290 case studies across 41 countries globally reveals a strong direct correlation, following an exponential trend, between the average rate of land subsidence and both groundwater withdrawal, with a determination coefficient of 0.950, and groundwater level decline, with a determination coefficient of 0.888 [27]. Although several studies have demonstrated a significant correlation between groundwater levels and land subsidence, the connection between land subsidence and groundwater drought remains undiscovered. Investigating the relationship between land subsidence and the characteristics of groundwater drought can provide deeper insights, as this connection is not apparent from raw groundwater level data. Focusing on the dynamics of groundwater drought, including temporal variations, magnitude, and intensity, enables a more comprehensive understanding of subsidence behavior.

Deployable output from water supply boreholes is a function of groundwater level. Therefore, standardized indices serve as valuable measures for quantitatively assessing the status of groundwater resources during regional drought conditions compared to raw groundwater level data [28]. This standardized index can serve as an early warning indicator for groundwater drought and potentially provide the ability to assess subsidence risks [29]. Moreover, in comparison to raw groundwater level data, standardized indices exhibit a reduced seasonal component, making it easier to distinguish between actual drought conditions and normal seasonal variations.

A review of previous studies reveals a lack of research addressing the relationship between groundwater drought and subsidence. To fill this gap, the present study leverages the benefits of the groundwater drought index to perform a comprehensive analysis, investigating its connection with subsidence for the first time in one of the most critical plains in northwestern Iran. The comprehensive land cover deformation map of the plain was generated using InSAR. Following an analysis of subsidence rates and magnitudes in the study area, the Standardized Groundwater Index (SGI), as the most commonly used groundwater drought index, was calculated for drought monitoring based on data from six piezometric wells located in the confined aquifer of the plain. The final stage involved assessing the relationship between groundwater drought and subsidence by using drought characteristics and correlation methods across various scales (frequencies) and time periods.

2. Materials and Methods

2.1. Study Area

The Marand Plain, located in East Azarbaijan Province in northwestern Iran, is recognized as one of the most critical plains in Iran. Covering an area of approximately 826 km², the plain has a semi-arid climate characterized by an average annual rainfall of 247 mm and an average annual temperature of 11.4 °C [30]. The Marand Plain is primarily drained by two rivers, Zilbir-Chay and Zonouz-Chay. Zilbir-Chay, the main river of the plain, flows from east to west across the area. Zonouz-Chay enters the plain from the north, merging with Zilbir-Chay in the western part of the plain before exiting and joining the Aras River [31]. The Marand aquifer comprises 62 monitoring wells, including both observation and piezometric wells, with water levels recorded on a monthly basis. Of these, 12 piezometric wells are situated within the semi-confined and confined aquifers, while the remaining 50 observation wells are located in the unconfined aquifer. Subsidence predominantly occurs in confined aquifers as a result of excessive groundwater extraction, leading to a decline in pore pressure and the compaction of compressible sediments [32]. Therefore, monthly groundwater level records from six piezometric wells in the confined aquifer were acquired from the East Azerbaijan Regional Water Authority (EARW) over a 13-year period spanning 2008 to 2021. The selection of these wells was based on their appropriate distribution throughout the confined aquifer and the completeness of their recorded measurements. The groundwater level data for the six piezometric wells are provided in Supplementary Table S1. Figure 2 illustrates the location of the Marand aquifer, along with the piezometric wells included in the study. Furthermore,

Table 1. Details of piezometric wells (confined aquifers).

Station ID	UTM:X	UTM:Y	Minimum GWL (m)	Maximum GWL (m)	Mean GWL (m)	Standard Deviation	Depth of Well (m)
P1	534,039	4,259,630	1036.73	1048.5	1041.39	2.55	104
P2	551,155	4,258,163	1087.83	1099.45	1092.19	1.83	180
P3	534,462	4,263,199	1031.36	1036.79	1033.94	1.20	210
P4	543,192	4,261,306	1044.77	1057.34	1051.44	3.13	225
P5	554,478	4,260,382	1102.76	1114.03	1106.51	3.10	200
P6	562,537	4,259,338	1138.5	1152.9	1143.02	3.54	135

presents the statistical characteristics of the piezometric wells.

An analysis of the geophysical properties of boreholes in the Marand Plain reveals the presence of three distinct types of aquifers: unconfined, semi-confined, and confined [33]. The unconfined aquifer, with a depth ranging from 220 to 270 m, is primarily located in the eastern and southeastern regions. This aquifer consists of coarse alluvial deposits, including sand, gravel, silt, and clay in the eastern area. In the southern part, it transitions into Plio-Pleistocene deposits characterized by low-permeability and semi-permeable materials containing layers of marl and clay. The confined aquifer, ranging in thickness from 10 to 30 m, is located in the western and central parts of the Marand Plain. It is composed of alluvial deposits that include distinct layers of clay and marl. The semi-confined aquifer, with a thickness varying between 10 and 150 m, is located in the western region of the Marand Plain [34]. Figure 3 displays six distinct geological cross-sections of the Marand Plain aquifer, derived from geophysical investigations, illustrating the various layers of alluvial deposits. Moreover, Figure 4 illustrates the locations of the geoelectrical sondes within the study area.

2.2. InSAR Data

The Centre for the Observation and Modelling of Earthquakes, Volcanoes, and Tectonics (COMET) is a world-leading research center dedicated to studying the Earth’s dynamic processes. The Looking Inside the Continents from Space (LiCS) initiative, led by COMET, uses Sentinel-1 satellite data to monitor large-scale land deformations. This effort is supported by the LiCSAR system, an automated interferometric processing platform for Sentinel-1 data, which provides free access to its products through an online portal [35]. LiCSBAS is an open-source package developed for InSAR time series analysis, capable of automatically processing time series data using the freely accessible products provided by LiCSAR [36]. This freely accessible package allows users to perform InSAR time series analysis efficiently, minimizing both processing time and disk space requirements [37]. The present study utilized Sentinel-1 satellite images spanning from January 2015 to February 2024. The acquisition interval, which was 24 days until the end of 2016, was reduced to 12 days after 2017 following the launch of Sentinel-1B operations. However, following the cessation of Sentinel-1B operations at the end of December 2021, the acquisition interval reverted to 24 days. The LiCSAR-generated interferograms over LiCSAR frames covering the Marand plain in descending orbits (LiCSAR frame ID 079D_05210_131313) were processed through the LiCSBAS package. Furthermore, to perform cross-validation between ascending and descending deformations, the ascending track with LiCSAR frame ID 174A_05216_131313 was utilized. Detailed information of the imagery is listed in

Table 2. Detailed information from LiCSAR data for the study area.

Frame ID	Date		Period (Year)	Epochs Processed (%)	GACOS (%)	Products
	Start	End
079D_05210_131313	15 January 2015	9 February 2024	10.1	98	100	2706
174A_05018_131313	10 November 2014	15 February 2024	9.3	87	100	1234

.

The LiCSAR processor produces each interferogram using three sub-swaths derived from Sentinel-1 data in the Interferometric Wide Swath (IWS) mode. These sub-swaths comprise 13 bursts, covering an area of 250 km × 250 km and providing a spatial resolution of 0.001°. The total number of interferograms was 2264, out of which 1355 were removed based on internal criteria to filter out low-quality interferograms. This corresponds to retaining 227 images out of a total of 333 (Figure 5). In this figure, Bperp, or perpendicular baseline, refers to the component of the baseline (the distance between two satellite positions) that is perpendicular to the line of sight from the satellite to the ground. The perpendicular baseline is crucial because it affects the sensitivity of the interferometric measurements to ground deformation and topography.

A variety of noise indices are employed to mask the results while ensuring the generation of accurate Line-of-Sight (LOS) time series deformation accumulation maps and deformation velocity. Moreover, atmospheric delay noise can be effectively removed by utilizing the General Atmospheric Correction Online Service (GACOS) for InSAR [38]. Figure 6 illustrates the workflow for obtaining land deformation data through LiCSAR and LiCSBAS time-series analysis. This figure illustrates the workflow of LiCSBAS, which includes the preparation of unwrapped (unw) interferometric phases and coherence (coh) data. Under the assumption of negligible horizontal displacement, the vertical displacement is determined using the following equation:

$$D_{\upsilon }={\displaystyle \frac{D_{LOS}}{\cos \theta }}$$

(1)

where $D_{\upsilon }$ represents the vertical displacement, $D_{LOS}$ denotes the LOS displacement, and $\theta$ refers to the incidence angle of the SAR sensor.

2.3. Standardized Groundwater Index

The present study employed the Standard Groundwater Index (SGI) as an innovative metric for monitoring groundwater drought and assessing its association with subsidence [28]. This index serves as a robust tool for assessing the severity and characteristics of hydrological drought [39]. In contrast to many conventional drought indices, the SGI eliminates the need for an accumulation period, owing to the continuous nature of groundwater levels [40]. Furthermore, the SGI uses a non-parametric normal scores transformation on raw data, rather than relying on a fixed transformation based on a specific distribution. This approach accounts for the varying distributions that can characterize groundwater time series [41]. The SGI drought index is determined through the following equation:

$${SGI}_{ij}={\displaystyle \frac{P_{\left(D\right)ij}-{\mu }_{\left(D\right)i}}{{\sigma }_{\left(D\right)i}}}$$

(2)

In the equation provided, ${SGI}_{ij}$ denotes the Standardized Groundwater Index for well i during time period j,$P_{\left(D\right)ij}$ represents the cumulative probability corresponding to the groundwater level in well i for period j, and ${\mu }_{\left(D\right)i}$ and ${\sigma }_{\left(D\right)i}$ represent the mean and standard deviation, respectively, of the groundwater level time series in well i.

2.4. Probability Density Functions

Fitting the appropriate Probability Distribution Function (PDF) in the calculation process for drought indices is crucial, as it significantly influences the assessment and interpretation of drought conditions. The monthly variations in groundwater levels across different monitoring stations exhibit distinct ranges and distributions, presenting a key challenge in fitting parametric distributions to groundwater level time series [42]. Furthermore, groundwater level time series demonstrate irregularities due to a complex interplay of factors including long-term trends, autocorrelation effects, and anthropogenic activities, making groundwater level data incompatible with standard PDFs [42]. The mentioned limitations have made non-parametric distributions an invaluable tool for calculating the SGI in the assessment of groundwater drought [43,44]. This study employed the flexible, non-parametric method of Kernel Density Estimation (KDE) to develop groundwater drought time series for the selected wells [45]. The KDE approach can be used to determine the PDF for the random variable x through the following equation [46]:

$$P\left(x\right)={\displaystyle \frac{1}{Nh}}{\sum }_{k=1}^{N}K\left({\displaystyle \frac{x-x_k}{h}}\right)$$

(3)

where N represents the total number of data points, K denotes the kernel function,$x={\left(x_1, x_2,\dots , x_d\right)}^T$ are the scale parameters, and $x_k$ is a d-dimensional data vector ${\left(x_{1,i},x_{2,i},\dots ,x_{d,i}\right)}^T$ given i = 1, 2, …, n, and h stands for the bandwidth parameter, which is a smoothing parameter. In the present study, the Gaussian Kernel was used as the core tool of the KDE method.

2.5. Drought Characteristics

Drought characteristics are essential for understanding the behavior and impact of droughts in a given region. A drought event can be evaluated through various components, as depicted in Figure 7. Based on this figure, a +ve Run signifies a period where conditions exceed a specific threshold, suggesting wet conditions, whereas a −ve Run indicates a period where conditions fall below the threshold, reflecting drought conditions. A drought event is characterized by some primary components: (a) Drought initiation time (t_i), which is the starting of the water shortage period and signifies the beginning of a drought; (b) Drought termination time (t_e), the point at which water scarcity diminishes sufficiently, ending the drought conditions; and (c) Drought duration (D_d), the continuous time span, measured in years, months, or weeks, during which a drought parameter remains below a critical threshold. Essentially, drought duration represents the period between the initiation and termination of the drought. This study identifies groundwater drought events using the SGI index, where negative SGI values indicate drought events. Therefore, the total number of distinct drought occurrences during the study period is considered the count of drought events, while the longest continuous drought period observed is referred to as the maximum duration. Drought periods, or drought duration, are a critical characteristic of drought events, representing the number of distinct drought phases occurring throughout the study period. The average length of drought periods, usually expressed in months or years, is calculated as

$$Mean Drought Period={\displaystyle \frac{Number of Drought Events}{Number of Drought Periods}}$$

(4)

Run sum (Deficit) is drought severity, and it refers to the cumulative sum of drought severity (negative SGI values) over the entire duration of a drought event. Maximum severity reflects the most extreme drought index value recorded within the study period:

$$Drought Severity=\sum _{j=1}^x{SGI}_{ij}$$

(5)

Here, j represents the first month of a drought and increments sequentially until the drought concludes (x) within any given i time interval. Drought intensity represents the average drought index value for a given drought event, calculated as the ratio of drought severity to its duration:

$$Drought Intensity={\displaystyle \frac{\sum Drought Severity}{Number of Drought Periods}}$$

(6)

2.6. Nonparametric Mann–Kendall Test

In this study, the widely used Mann–Kendall (MK) test was employed to capture the trend of groundwater drought [47,48]. The MK test, as a nonparametric method, provides the advantage of analyzing data without assumptions of normality, linearity, or the absence of missing values. The MK Z_MK statistic is applied to capture both positive and negative trends. A negative Z_MK value indicates a decreasing trend, while a positive value reflects an increasing one. The calculation process for the standard Z_MK value is formulated as follows:

$$Z_{MK}=\left\{\begin{array}{c}\begin{array}{cc}{\displaystyle \frac{S-1}{V\left(S\right)}}& for S>0\end{array}\\ \begin{array}{cc}0& for S=0\end{array}\\ \begin{array}{cc}{\displaystyle \frac{S+1}{V\left(S\right)}}& for S<0\end{array}\end{array}\right.$$

(7)

$$V\left(S\right)={\displaystyle \frac{n\left(n-1\right)\left(2n+5\right)}{18}}$$

(8)

$$S=\sum _{k=1}^{n-1}\sum _{j=k+1}^{n}sgn\left(x_j-x_k\right)$$

(9)

$$sgn\left(x_j-x_k\right)=\left\{\begin{array}{cc}+1& if \left(x_j-x_k\right)>0\\ 0& if \left(x_j-x_k\right)=0\\ -1& if \left(x_j-x_k\right)<0\end{array}\right.$$

(10)

where V(S) denotes the variance, and S represents the Kendall sum statistic. The signs of the differences between successive values are calculated as positive (+1), negative (−1), or neutral (0). These signs are expressed using the sgn(…) function, as illustrated in Equation (10). Here $x_j$ and $x_k$ correspond to the time series values at time steps j and k, respectively, derived from a dataset $\left\{x_1, x_2,\dots , x_k,\dots ,{ x }_j,\dots , x_n\right\}$ consisting of n observations. For this study, trend analysis was performed at a 1% significance level. To assess the presence of a monotonic trend, at a significance level of $\alpha$, the null hypothesis (H₀) is rejected if $\left|Z_{MK}\right|>Z_{1-\alpha /2}$. The critical value, $Z_{1-\alpha /2}$, is determined from the standard normal cumulative distribution table. For example, at a 5% significance level, H₀ is rejected if $\left|Z_{MK}\right|$ > 1.96. A larger absolute $Z_{MK}$ value indicates a stronger statistical significance of the trend. In this study, the MK trend test was conducted at a 0.05 significance level, a widely adopted threshold in drought monitoring research [49,50].

2.7. Trend-Free Pre-Whitening (TFPW) Mann–Kendall Test

Although the MK test is widely applied in hydrological trend analysis, it lacks the capability to effectively characterize trends in serially correlated time series. This limitation increases the likelihood of identifying false trends or change points in the time series [51]. Considering the potential for autocorrelation in some of the wells [52], the Trend-Free Pre-Whitening (TFPW) method [53], a modified version of the Mann–Kendall (MK) test, was employed to analyze the groundwater drought process. The detailed procedure for the TFPW method is outlined as follows:

(1)

Calculate the linear trend (β) for each index series, X_t (where t = 1, 2,…, N):

$$\beta =Median{\displaystyle \frac{X_j-X_i}{j-i}}{\forall }_i<j$$

(11)

(2)

Generate a new series, Y_t, by eliminating the trend component from the original series:

$$Y_t=X_t-\beta ·t$$

(12)

(3)

Calculate the first-order autocorrelation coefficient, r, for the new series Y_t, and conduct a significance test on r with a significance level of 0.1. If the test is significant, proceed by applying the MK method directly to test the original series, X_t. If the test is not significant, proceed to step (4) for further preprocessing.

(4)

Create a new series, $Y_t^{\prime }$, by eliminating the autocorrelation terms from the original series. Then, reintroduce the trend component to form a new series, $Y_t^{″}$, which is free from the influence of autocorrelation interference:

$$Y_t^{\prime }=Y_t-r·Y-1$$

(13)

$$Y_t^{″}=Y_t^{\prime }+\beta ·t$$

(14)

(5)

Substitute the new series, $Y_t^{″}$, into the MK test.

The TFPW method effectively eliminates serial dependence, a key issue in testing and interpreting time series data [54,55]. This process enhances the accuracy of trend analysis by minimizing biases in trend detection.

2.8. Continuous Wavelet Transform (CWT)

For the given time series $X=\left\{x_i, i=1, 2, 3, \dots , N\right\}$ with a uniform sample step ${\delta }_t$, the CWT $W_i\left(s\right)$ at scale s and time $t_{n^{\prime }}=n^{\prime }\delta t (n^{\prime }=0, 1, \dots , n-1)$ can be represented as

$$W^X\left(s, t_i\right)=\sqrt{{\displaystyle \frac{{\delta }_t}{s}}{\sum }_{m=1}^{n-1}{x}_m\psi \left({\displaystyle \frac{t_m-t_{n^{\prime }}}{s}}\right)}$$

(15)

where $\psi$, the mother wavelet, represents a zero-average wavelet basis function. In the proposed scheme of this study, the Morlet wavelet (represented as ${\psi }_0$) serves as the mother wavelet to climate time series analyses since it grants the best balance between time and frequency resolution [56]. The Morlet wavelet is formulated as follows:

$${\psi }_0\left(\eta \right)={\pi }^{-{\displaystyle \frac{1}{4}}}{e}^{i{\omega }_0\eta }{e}^{-{\displaystyle \frac{1}{2}}{\eta }^2}$$

(16)

where $\eta$ and ${\omega }_0$ refer to the non-dimensional time and frequency, respectively.

2.9. Wavelet Transform Coherence (WTC)

Wavelet Transform Coherence (WTC), an advanced extension of the Continuous Wavelet Transform (CWT), serves as a powerful analytical method for simultaneously examining the relationship between two time series across both time and frequency domains [57,58]. Wavelet coherence of two time series is defined as

$$R_{Y, X}^2\left(s, t_{n^{\prime }}\right)={\displaystyle \frac{{\stackrel{⃡}{W}}^{Y, X}(s, t_{n^{\prime }})\stackrel{-}{{\stackrel{⃡}{W}}^{Y, X}(s, t_{n^{\prime }})}}{{\stackrel{⃡}{W}}^{Y, Y}(s, t_{n^{\prime }}){\stackrel{⃡}{W}}^{X, X}(s, t_{n^{\prime }})}}$$

(17)

where $\stackrel{⃡}{\left(\right)}$ stands as the smoothing operator and $\overline{\left(\right)}$ stands as the complex conjugate operator.

$$W^{Y, X}\left(s, t_{n^{\prime }}\right)=W^Y(s, t_{n^{\prime }})·W^{X_{*}}(s, t_{n^{\prime }})$$

(18)

where $W^X(s, t_{n^{\prime }})$ and $W^Y(s, t_{n^{\prime }})$ show the wavelet transform of X and Y series, respectively. The symbol “⁎” represents the complex conjugate. $W^{Y, X}\left(s, t_{n^{\prime }}\right)$ ranges from 0 to 1 and similarly to Pearson’s r, a value close to 0 means no association, whereas 1 means a complete or perfect correlation.

3. Results

3.1. Long-Term Deformation Rates

Figure 8 illustrates the long-term LOS displacement rate in the northern region of Urmia Lake, located in northwestern Iran, over the period from 2015 to 2024. Long-term deformation rates are computed by fitting a linear trend to cumulative displacement data obtained from multiple interferograms. The northern region of Urmia Lake includes four prominent subsidence zones. During this period, the Marand Plain, located northeast of Urmia Lake, experienced a maximum land subsidence rate of −69.36 mm per year based on the spatial reference point. Moreover, significant subsidence has occurred across extensive areas of the Shabestar Plain, located in the eastern part of Urmia Lake. In the northwestern and western regions of the lake, the Khoy and Salmas plains have experienced subsidence at a slower rate.

3.2. InSAR Validation

The vertical deformation time series was generated using the InSAR technique to enable a comparison of its accuracy with the time series from the permanent GNSS station [59]. To evaluate the accuracy of the InSAR-derived subsidence measurements, data from a permanent GNSS station with daily observations were utilized. The Iranian Permanent GPS Network (IPGN) consists of 107 continuously operating GNSS stations equipped with high-precision dual-frequency receivers. This network was established in 2005 and is strategically distributed across seismically active regions of Iran. The National Cartographic Center of Iran (NCC) is responsible for maintaining and processing the data from these stations. The primary objectives of the network include monitoring crustal deformation and providing a highly accurate geodetic reference frame for various geoscientific applications. The GNSS data processing strategy employed by the NCC is based on network solutions (relative positioning), utilizing the GAMIT software (version 10.2) [60]. In the initial phase, 107 permanent stations were deployed, with an average spacing of 25 to 30 km in denser regions. Tasuj is the only city within the study area with a GNSS station (TASJ station), which was used to validate the subsidence maps derived from LiCSBAS InSAR time-series analysis. The location of this station is indicated by a black circle in Figure 8. To validate the findings, the Root Mean Square Error (RMSE) criterion was used, considered as the most widely used statistical indicator for InSAR result validation [61]. To achieve this, RMSE values were calculated by taking into account the vertical displacement data recorded at the GNSS station and averaging the 10 nearest InSAR pixels over the overlapping time period [62]. The InSAR data, initially recorded in the LOS direction, were projected onto the vertical axis for direct comparison. The analysis covered the period from January 2015 to December 2021, with GNSS measurements processed relative to a fixed reference point. The date 15 January 2015 was chosen as the temporal baseline (zero displacement reference) for consistency in displacement calculations. The comparison between the InSAR results and the data from the TASJ GNSS station is demonstrated in Figure 9. Between 2015 and 2017, the InSAR-derived vertical displacement time series closely followed the seasonal subsidence trends observed in the GNSS data. However, from 2018 onward, a significant divergence emerged, with GNSS measurements consistently indicating greater vertical displacement compared to InSAR results. In some periods, the discrepancy reached up to 28 mm. The results indicate that at the TASJ station, an RMSE value of 11.51 mm was calculated over a period of 6.8 years.

To evaluate the accuracy of the LiCSBAS processing, a single static GNSS station with daily measurements was available. Consequently, in addition to GNSS data, a cross-validation method among the ascending and descending deformations was employed to assess the consistency of the results. Comparing displacement values obtained from ascending and descending orbits is a widely used and reliable approach for assessing the compatibility between measurements [63,64]. Therefore, to verify the self-consistency of measurements obtained from two different geometries, the mean displacement rates within the overlapping region of both directional datasets were calculated. A quantitative comparison of the estimated displacement rates from the ascending and descending orbits is presented in Figure 10. The displacement rate maps along the LOS were compared for all common pixels in both acquisition paths. Subsequently, the distribution of subsidence rates from both paths was analyzed, revealing a correlation coefficient (R²) exceeding 0.95 between the estimated displacements. Figure 10b presents a density scatter plot illustrating the relationship between subsidence rates from both paths. Each point in the plot is colored according to the local density of the data points, with the color bar indicating the density values on a logarithmic scale. Higher density values correspond to regions with a greater concentration of data points, while lower values indicate sparser regions. The distribution of differences, expressed in millimeters per year, indicated that the difference between the two series was nearly zero. These values suggest a strong correlation between the subsidence patterns (average rates) derived from different datasets, indicating that both are predominantly influenced by vertical motion [65].

3.3. Subsidence Map of the Marand Aquifer

Figure 11 depicts the long-term subsidence map of the Marand aquifer from January 2014 to February 2024, derived from LOS displacement data, along with the locations of the selected piezometric wells in the confined aquifer. The land deformation map reveals the presence of both uplift and subsidence phenomena within the aquifer. As depicted in Figure 11, a negative deformation of 65 cm was observed over a period of 75 months, impacting an area of 57,412 ha across confined, semi-confined, and unconfined aquifers. Severe subsidence has been recorded over approximately 34,539 ha in the central regions, predominantly within the confined aquifer.

The excessive extraction of groundwater for agricultural development has resulted in aquifer compression and significant subsidence in the study area. This conclusion is further supported by Figure 12, which illustrates the spatial NDVI trend over the study period (2015–2024) using the Kendall tau approach. The findings indicate a predominantly positive NDVI trend across the studied aquifer areas, reflecting extensive vegetation expansion in the region. The significant downward trend in the average groundwater level within the confined aquifer, reflected by Z-index values of −7.32 (MK test) and −11.02 (TFPW test), combined with the significant upward trend in the average NDVI across the study area, with Z-index values of 1.88 (MK test) and 2.85 (TFPW test), confirmed the extensive extraction of groundwater resources from the confined aquifer to support agricultural development.

3.4. Identification of Critical Impact Zone

Aquifer behavior was analyzed through assessing the spatial influence of groundwater extraction from individual wells on land deformation in their surrounding areas. To achieve this, the average InSAR displacement was quantified within a radial distance ranging from 250 to 4000 m around the piezometric wells [66]. Subsequently, the statistical characteristics of land deformation derived from the generated maps were visualized using box plots (Figure 13). Subsidence values tend to be more variable and extreme within smaller buffer zones (e.g., 250 m and 500 m). However, as the radial distance increases, the subsidence values become more stable and demonstrate reduced variation, with the exception of well P5, which showed the highest fluctuations within the 1500 m and 2000 m radii. At a distance of 4000 m, the results were found to be not significant or effective.

The western part of the aquifer shows stability in terms of land subsidence. Wells P1 and P3, located in the western part of the aquifer, demonstrated a relatively small range of subsidence values with fewer fluctuations across various distances. Wells P2, P5, and P6 showed a higher range of subsidence, especially in smaller buffer zones (250 m and 500 m), indicating more intense ground deformation near the wells. In the present study, the greatest area of aquifer exploitation was limited to a radius of 500 m around the wells [67], and subsequent analysis was carried out accordingly.

3.5. Trend Analysis of Groundwater Drought

This section presents the trend analysis of groundwater drought for the period emphasizing its role in identifying long-term patterns and changes in groundwater conditions. To maintain consistency and enhance the reliability of the trend analysis, the groundwater drought trend was assessed using the same time period as that for the InSAR data (2015–2024). The results, shown in

Table 3. Trend analysis results of SGI drought index.

Well	MK Test		TFPW Test
	Z	Sen’s Slope	Z	Sen’s Slope
P1	6.39 *	0.03069	8.77 *	0.03141
P2	−8.51	−0.02639	−9.88 *	−0.02641
P3	6.93 *	0.03351	9.35 *	0.03325
P4	−5.56 *	−0.01756	−7.80 *	−0.01793
P5	−4.84 *	−0.01034	−7.61 *	−0.01087
P6	−5.66 *	−0.01389	−7.46 *	−0.01372

Numbers with power ‘*’ indicate significant trend at a 95% confidence level.

, reveal that wells P1 and P3 exhibit a significant positive trend at the 95% confidence level, indicating they have entered a wet phase. These wells are located in the western part of the Marand Plain aquifer, an area where no significant subsidence is observed. The absence of subsidence in this region may contribute to the stability of ground conditions, with the positive effect of groundwater recharge helping to alleviate pressures on the aquifer. Consequently, these areas are likely to maintain more stable water reserves and demonstrate reduced sensitivity to excessive water withdrawal. In contrast, the analysis highlights critical conditions in wells P2, P4, P5, and P6, which show a strong negative trend at the 95% confidence level. The significant negative trends in wells P5 and P6, located in the central part of the Marand aquifer, indicate a significant decline in groundwater levels and a deterioration of drought conditions in these regions. Notably, the most significant subsidence occurred in the vicinity of well P2, where the updated trend analysis for 2015–2024 using the MK and TFPW tests revealed the most pronounced negative trend in terms of Z values at the 95% confidence level.

3.6. Groundwater Drought Monitoring

To investigate the relationship between subsidence and groundwater drought, the first step involved evaluating groundwater drought characteristics through the SGI index for the selected piezometric wells. Figure 14 illustrates groundwater drought fluctuations, represented by the SGI index, for these wells. In the figure, months experiencing drought conditions (indicated by negative SGI values) are marked in red, while months experiencing wet conditions (indicated by positive SGI values) are marked in green. The results indicated that wells P1, P2, and P3 experienced alternating consecutive drought periods, with the highest occurrences recorded at 11, 10, and 9 periods, respectively. In contrast, wells P4, P5, and P6 exhibited a significant negative trend in drought severity. Initially, these wells were in wet conditions, but they gradually transitioned into periodic drought conditions starting in September 2013. Specifically, well P4 entered a sustained drought phase in June 2018, while wells P5 and P6 transitioned into full drought phases in March 2017. Overall, wells P4, P5, and P6 recorded the highest number of drought events, with 80, 88, and 81 events, respectively. Notably, well P4 experienced the highest number of extreme droughts (SGI ≤ −2), with four events occurring between September and December 2020. Well P2 recorded the highest number of severe drought events (SGI ≤ −1.5), with a total of seven occurrences, four of which occurred consecutively between October 2012 and January 2013. Furthermore, wells P5 and P6 have not experienced any severe or extreme drought events. Details of the drought characteristics for the studied wells are summarized in

Table 4. Groundwater drought characteristics for representative wells.

Well	No. of Drought Events	No. of Drought Periods	Mean Drought Period	Intensity	Severity	Maximum Intensity	Maximum Duration (Months)
P1	75	11	6.81	0.74	56.10	6.22	10
P2	78	10	7.8	0.73	57.70	12.44	10
P3	75	9	8.33	0.76	57.25	19.09	19
P4	80	5	16	0.71	57.01	36.50	34
P5	88	6	14.66	0.58	51.60	35.97	47
P6	81	7	11.57	0.64	52.47	37.87	47

.

3.7. Relationship Between Groundwater Drought and Land Subsidence

To investigate the relationship between drought conditions in the selected wells and land subsidence, the average subsidence within a 500-m radius of each well was calculated. The results showed that the highest subsidence occurred in the vicinity of well P2, reaching 22.4 cm in November 2020. Similarly, the maximum recorded subsidence near wells P4, P5, and P6 measured 16.3 cm, 17.2 cm, and 21.2 cm, respectively. The western part of the aquifer remained stable in terms of land subsidence, as evidenced by the minimal subsidence rates observed for wells P1 and P3, both located in this region, with a maximum recorded subsidence of approximately 2 cm. In contrast, well P6 exhibited the highest average subsidence, estimated at around 11.8 cm, suggesting prolonged exposure to subsidence in this area. The average subsidence recorded for the remaining wells, including P1, P2, P3, P4, and P5, was 0.1 cm, 11.7 cm, 0.2 cm, 8.3 cm, 10.2 cm, and 11.8 cm, respectively. Figure 15 illustrates the land subsidence rates at the selected wells in the confined aquifer. Upon investigating the subsidence time series, it is evident that the fluctuations in subsidence values occur nearly simultaneously and with comparable amplitudes. This pattern suggests a unified response of the aquifer to common environmental or anthropogenic factors, such as seasonal agricultural harvesting, variations in rainfall, or changes in aquifer recharge.

The analysis of the results indicates that among the drought characteristics, drought duration and maximum intensity have a direct impact on land subsidence. Specifically, in the areas around wells P4, P5, and P6, which have experienced longer and more severe droughts, the greatest subsidence effects are observed. It can be concluded that in areas where wells have entered a phase of continuous drought, characterized by a high number of drought events and a low number of drought periods, leading to a high mean drought duration, a higher rate of subsidence is expected. On the other hand, in wells P1, P2, and P3, which experience consecutive drought events and have a higher number of drought periods (11, 10, and 9 respectively), the severity of the drought appears to be a determining factor for subsidence. Specifically, the area around well P2, with the highest severity (57.70), exhibits the most significant subsidence.

3.8. Correlation Analysis

To evaluate the impact of groundwater drought variations on land deformation, a correlation analysis was conducted using lag times ranging from zero to six months (Figure 16). The findings indicate a strong positive correlation between the SGI groundwater drought index and vertical land deformation in the central and critically affected regions of the aquifer. The correlation coefficients for P4, P5, and P6 are 0.911, 0.721, 0.695, and 0.817, respectively. In regions experiencing high subsidence rates, the correlation between vertical land deformation and the groundwater drought index decreases across all wells (P2, P4, P5, and P6) as the time lag increases. The zero-lag positive correlation observed in critical areas is attributed to the presence of fine-grained sediments, which experience rapid consolidation as groundwater drought conditions intensify. In contrast, in the vicinity of wells P1 and P3, where land subsidence remains stable, a weak negative correlation is observed between the groundwater drought index and the subsidence time series. The correlation coefficients for wells P1 and P3 are −0.157 and −0.202, respectively. In wells P1 and P3, located in regions with stable subsidence conditions, the correlation between the groundwater drought index and vertical land deformation demonstrates an increasing trend as the time lag increases. Analysis indicates that in well P1, the relationship between vertical land deformation and groundwater drought is most significant at a 3-month lag, with a correlation coefficient of −0.731. Similarly, in well P3, the strongest correlation is observed at 3- and 4-month lags, with correlation coefficients of −0.654 and −0.656, respectively. A negative correlation could occur if the depletion of the surficial aquifer caused geostatic unloading, leading to the expansion of the underlying confined aquifer. In this case, the reduction in groundwater levels in the surficial aquifer would result in the uplift of the ground surface [68].

3.9. Wavelet Coherence Analysis

Subsequently, multi-resolution analysis of Wavelet Transform Coherence (WTC) was employed for the detection of correlations at various temporal scales. The WTC results between groundwater drought and InSAR-derived vertical land deformation are depicted in Figure 17. In this figure, the thick black line represents the 5% significance threshold when compared to red noise. The horizontal axis represents time, while the vertical axis denotes frequency, with lower values indicating higher scales. The WTC analysis identifies regions in the time-frequency domain where the two time series exhibit covariance. Significantly correlated regions are indicated by warmer colors, typically depicted in yellow, while colder colors, such as blue, signify weaker dependencies. Regions in colder colors beyond the significant areas correspond to time-frequency domains where the two time series exhibit no apparent dependence. Moreover, the WTC uncovers phase patterns through the direction of arrows in the wavelet coherence plots, enhancing our understanding of lead–lag associations between parameters. The orientation of the arrows has the following meaning: arrows pointing to the right indicate an in-phase (or favorably connected) relationship between x(t) (SGI) and y(t) (second variable, vertical land deformation) and vice versa. Arrows pointing in different directions indicate lags or leads between variables. In this context, an upward-pointing arrow denotes a leading relationship from the first variable to the second, while a downward-pointing arrow signifies the reverse relationship. Regions outside the Cone of Influence (COI) must not be considered. Based on the obtained results, it was noted that the groundwater drought–vertical land deformation relationship exhibited continuous significant coherence areas at the 95% confidence level for the 8–16-month period, which is probably caused by short-term or seasonal phenomena such as precipitation patterns, groundwater recharge, or drought cycles. In the vicinity of wells P1 and P3, located in relatively stable regions, the upward direction of the arrows suggests that groundwater drought plays an influential role in leading vertical land deformation. On the other hand, in other wells located in critical areas, the arrows primarily point to the right, indicating an in-phase relationship between groundwater drought and subsidence.

4. Discussion

The validation results demonstrated an acceptable agreement between InSAR measurements and GNSS data, with an RMSE of 11.51 mm, supporting the reliability of the calculated subsidence. The observed divergence between GNSS and InSAR measurements from 2018 onward may be attributed to temporal decorrelation, where variations in surface scattering properties between SAR acquisitions degrade measurement accuracy [69]. Factors such as vegetation growth, fluctuations in surface moisture, and human-induced land use changes contribute to reduced coherence in SAR signals, increasing noise in the interferogram and affecting displacement estimates [70]. The accelerated rate of vegetation expansion and extensive land use modifications observed since 2018 are likely key contributors to this discrepancy. It is important to note that in validating InSAR results, a difference of approximately 7–8 mm is generally considered acceptable. However, variations of up to ±20 mm may occur due to various noise sources [71]. Therefore, considering the GNSS-derived deformations as true measurements and the observed consistency between the InSAR results and the GNSS data, it can be concluded that the deformation measurements derived from InSAR in this study demonstrate acceptable accuracy.

The land deformation map indicates the occurrence of both uplift and subsidence phenomena within the Marand aquifer. Land uplift can be caused by several factors, including atmospheric noise effects, issues during the recovery phase [72], and aquifer recharge [73,74] driven by both natural and artificial influences. Considering that the eastern regions of the Marand aquifer experience aquifer recharge, it appears that land uplift in other areas, such as the western regions, is primarily attributed to atmospheric noise effects or the fault recovery phase [30].

NDVI serves as a reliable indicator of agricultural activities, effectively reflecting productivity dynamics [75]. Increased irrigation for agriculture leads to over-extraction of groundwater, particularly from confined aquifers, where recharge is much slower than extraction. The geological structure of a confined aquifer is characterized by compressible clay and silt layers, which experience permanent compaction as water pressure declines. This process serves as a key limiting factor, significantly increasing the risk of subsidence [76]. As a result, the central part of the study area, located within the confined aquifer, experiences significant subsidence despite the widespread expansion of vegetation cover across most parts of the aquifer, as indicated by the positive NDVI trend.

The variance in deformation values observed for P1–P6 in Figure 13 is attributed to multiple factors, including the spatial heterogeneity of subsurface geology, variations in groundwater extraction rates, and differential compaction of aquifer materials. Subsidence values demonstrate higher variability at smaller buffer zones (e.g., 250 m and 500 m), mainly due to localized compaction effects caused by concentrated pumping near wells. As the radial distance increases, the deformation patterns tend to stabilize, indicating a more distributed impact of groundwater withdrawal. However, differences between wells remain significant. Wells P2, P5, and P6 exhibit significant fluctuations, particularly in smaller buffer zones, which suggests that these locations experience more intense compaction or that the subsurface materials are more compressible. In contrast, wells P1 and P3, located in the western part of the aquifer, show less variation, possibly due to a more rigid geological structure or lower extraction rates. Notably, well P5 demonstrates high variance even at larger distances (1500–2000 m), indicating either a stronger cumulative extraction impact or local geological anomalies that enhance subsidence.

Groundwater drought monitoring revealed a critical condition in the central regions of the Marand aquifer, resulting in widespread land subsidence. Trend analysis of the average NDVI showed a notable positive trend in both the central areas and the confined aquifer over the 13-year period from June 2008 to September 2020. The uncontrolled expansion of agricultural activities in this region has contributed to a decline of approximately 5 m in the average groundwater level within the confined aquifer. A substantial decline in the groundwater level within a confined aquifer has resulted in the compression of inter-aquifer beds composed of fine-grained, highly compressible sediments, such as clay and silt, thereby causing land subsidence. A review of studies indicates that irrigated agriculture and orchards in the Zilbirchay basin, including the Marand plain, expanded by approximately 50% and 45%, respectively, between 1987 and 2007 [30]. Additionally, as shown in Figure 18, alfalfa cultivation, a high-water-use crop, increased by around 43% between 2007 and 2015, while 20% of other land use was converted to orchard development. This expansion has significantly contributed to the depletion of groundwater resources in the region, leading to the drying up of numerous wells. It is estimated that there were more than 60 piezometric wells in the Marand plain, of which over 10 have dried up in the past decade [77]. Based on current trends, alfalfa cultivation is projected to increase by 16%, and orchard coverage by 31% from 2015 to 2030. Moreover, pumping practices have remained unchanged, and following current policies is expected to exacerbate pressure on groundwater resources, potentially leading to the drying up of other wells in the near future. The groundwater drought in the study area is further intensified by the implementation of Inappropriate cropping patterns. Since 2007, there has been a steady decrease in land used for rainfed agriculture, a trend that is expected to persist until 2030. This shift is accompanied by a significant 39% reduction in the cultivation of water-efficient crops, including irrigated wheat.

The findings of this study indicate that monitoring groundwater drought, rather than relying on raw groundwater data, offers a more comprehensive perspective in understanding the relationship between groundwater resource conditions and subsidence. Analyzing the characteristics of groundwater drought, such as its duration and magnitude, can act as an early warning system for potential subsidence in areas surrounding the study wells. The absence of field data on groundwater is considered a significant limitation of this study. However, the use of GRACE satellite data for monitoring groundwater droughts presents a reliable solution to address this gap and can be investigated in future work.

5. Conclusions

This study provides a comprehensive assessment of the interrelationship between groundwater drought and InSAR-derived land deformation in the Marand aquifer, a critical region in northwestern Iran. InSAR-derived measurements were validated using GNSS data with an RMSE of 11.51 mm. Analysis of long-term deformation rates (2015–2024) revealed severe subsidence, with the Marand Plain experiencing the highest rate of −69.36 mm/year. Spatial analysis indicated that subsidence exceeded 30 cm in central regions, particularly within the confined aquifer, coinciding with high vegetation density, as indicated by the NDVI distribution.

Groundwater drought trends showed critical conditions in central wells with significant groundwater decline at the 99% confidence level, while western wells exhibited positive trends aligning with stable subsidence. Buffer analysis around piezometric wells confirmed the localized influence of groundwater extraction, with subsidence effects decreasing at greater distances. Longer and more severe droughts were linked to greater subsidence, while frequent droughts influenced subsidence magnitude. Correlation analysis found a strong positive relationship between SGI and subsidence in critical zones, especially with zero lag, while stable regions exhibited a delayed negative correlation due to geostatic unloading. Multi-resolution WTC analysis highlighted seasonal-scale correlations with dominant periodicities of 8–16 months.

Overall, this study provides valuable insights into the interconnected dynamics of groundwater drought and land subsidence, emphasizing the critical role of sustainable groundwater management in mitigating subsidence risks. To further enhance the precision of subsidence monitoring, future studies should aim to report deformation trends with an accuracy of 0.1 mm/year, which would significantly improve our understanding of subtle changes in land deformation. Furthermore, future research should incorporate a more comprehensive approach to quantifying the relationship between vegetation expansion dynamics and land subsidence by integrating additional influencing factors.