Increases in the Amounts of Agricultural Surfaces and Their Impact on the Sustainability of Groundwater Resources in North-Central Chile

Abstract

Water is a fundamental resource for Chile’s productive structure, which is more important in arid areas, and especially with agricultural uses. This study was based on two basins (Cogotí and Illapel) located in the Coquimbo Region of north-central Chile. In this region, surface water rights were closed in 2002 and the only current option is the use of groundwater. These basins have high water demands due to the use of surface and groundwater for agricultural purposes, a fact that should influence the sustainability of groundwater reserves over time. The objective of this study was to determine how much agricultural use has affected the availability of groundwater in two basins. Under the previous context, the evolution of agricultural irrigation surfaces was evaluated using Landsat images and forest classifications. Similarly, groundwater reserves were evaluated using the recessive curves of hydrographs associated with the beginning of each hydrological year. The results show an increase in the agricultural area between 1996 and 2016, with a subsequent decrease, while groundwater reserves denoted significant decreases over time. In conclusion, a significant decrease in the volumes of groundwater reserves in both basins was observed, a decrease that is consistent with the increase in irrigated areas.

1. Introduction

Water resources are a fundamental component for life, which are involved in a host of biological and physical processes [1]. The presence of groundwater constitutes a key factor for most physiological and biochemical activities [2,3,4], as well as the functioning of various ecosystems [5].

In Chile, a generalized decrease is projected for water resources due to the increase in demands from water users, changes in land use, and climatic variability [6,7]. Likewise, the availability of water is manifested with limitations in the national territory due to climate change [8], a phenomenon that is manifested as a lower supply of water resources in Chile.

Water demands in the north and central zones of Chile permanently exceed the existing availability, with agriculture, mining, and drinking needs being the main demanders of the resource [9,10]. There is clear empirical evidence in Chile that indicates that where mining and large-scale agriculture coexist in the same region, these areas are affected by water scarcity [11].

In this scenario, the evaluation of hydrological variables that account for the general water balance of basins is important. Thus, the flow variable, as an expression of the water production capacity of a hydrographic system, is a determining factor in the objective of knowing the state of a given watershed, in the face of scenarios of climate change and the use of water resources [12].

In the above context, changes in land use for intensive agricultural purposes is a variable that would have a very strong impact on the water production of various basins, as well as on groundwater reserves [13]. In this sense, various authors (e.g., [14,15,16,17,18,19]) have pointed out that current agricultural practices are not sustainable over time, given that this activity represents 85% of global water consumption and it is likely that this will double by 2050 [17]. Another aspect to consider, especially in areas with low amounts of rainfall, is the use of groundwater to supply the differential demand required for crop irrigation [13,20,21,22]. However, this practice reduces the volume of water stored in aquifers and, if maintained, it is possible to alter groundwater availability or even dry terminate it [22,23,24,25]. This imbalance between water supply and demand is also a factor that will impact local ecosystems and populations, and its effects will be accentuated by climate variability and change [26].

Though the decrease in surface water has been widely studied in Chile (e.g., [13,27,28,29,30,31]), the evolution of groundwater availability is not well understood because its analysis is more complex [32]. However, it is possible to analyze the variation in reserves using baseflows as a proxy unit [32,33], thus facilitating the estimation of recharge. Thus, baseflow and its impact based on overuse is an interesting factor to analyze, which can shed light on the impact on aquifer reserves. One way to estimate baseflows is through exponential distributions [34], distributions that have been used to model surface and groundwater flows at a basin level [35,36].

Based on the above, it is proposed to carry out an analysis of the effects of land use changes for large-scale agricultural purposes in two basins in north-central Chile, establishing whether the trends in the values of water reserves have remained stable over time or have varied significantly to determine whether or not those land use changes have affected the availability of groundwater in the last two decades.

2. Materials and Methods

2.1. Study Area

The Cogotí and Illapel basins are located in the Coquimbo administrative Region, between latitudes 29° S and 32° S (Figure 1). The local climate is arid to semi-arid with a precipitation pattern defined by 2-to-3 wet winter months and a territorial gradient where precipitation increases from north to south and from the coast (west) to the Andes (east) [13,37].

Qgis 3.28 [38] and Rstudio 2023 [39] software were used to carry out this study. The research aims to determine how the surface area of intensively irrigated agricultural areas has varied in four time periods (1996, 2006, 2016, and 2022). In addition, it aims to establish the behavior of trends in groundwater storage volumes in each basin studied from 1996 to 2022.

2.2. Quantification of Land Use Changes

2.2.1. Selection and Delimitation of Basins

The selection of the basins was carried out between the regions of Coquimbo and Valparaíso, in the north-central zone of Chile, which were subjected to a sorting process based on the hydrological information available at fluviometric stations. The procedure began with the selection of fifteen basins, which were identified using the “CR2 Search” (CR2, 2023), to end with a selection of two basins (Cogotí and Illapel) with fluviometric information between 1996 and 2022, as previously pointed out. Both basins were delimited using the QGIS program, based on the methodology proposed by Dixon and Uddameri [40] (Figure 1 and

Table 1. Summary of the fluviometric stations used in this study.

Name of Station	ID	Drainage Area (km2)	Recording Period
Río Cogotí Entrada Embalse Cogotí	4531002	741	1996–2022
Rio Illapel en el Peral	4726001	2000	1996–2022

).

2.2.2. Collection and Processing of Landsat Images across Different Time Frames

The effective irrigation areas were determined for four years (1996, 2006, 2016, and 2022) using QGIS and Landsat images. Landsat images (

Table 2. List of images used for classification.

Year	Sensor	Cloud Cover (%)	Satellite Area *	Image Date
1996	Landsat 5-TM	<10	001 081	15 January 1996
			233 082	28 March 1996
2006	Landsat 7-ETM+	<10	001 081	31 March 2006
			233 082	16 March 2006
2016	Landsat 8-OLI	<10	001 081	23 February 2016
			233 082	4 April 2016
2022	Landsat 9	<10	001 081	4 April 2022
			233 082	28 March 2022

* Images sourced from United States Geological Survey (USGS).

) from the United States Geological Survey (USGS) were used for the established periods, thereby allowing the analysis of the variation of the irrigation surfaces, especially for intensive uses within each basin.

The satellite codes used were “001 081” and “233 082”. After downloading these USGS images, the bands from each year were merged using the “Build Virtual Raster” tool. Subsequently, both areas were combined into a single “Tiff” layer using the “Mosaic” tool of the Orfeo Toolbox extension (OTB, [41]). The result was clipped to the area of the two selected basins using the “Clip Raster by Mask Layer” tool.

2.2.3. Training Polygons

Once the Landsat images were obtained for the areas of interest (basins) and the four periods, the bands were configured for agricultural areas. Training polygons were identified, defining two categories: “Agriculture” and “Others”. Intensively irrigated agriculture was defined as that visible in satellite images, including grazing areas. “Other” encompasses non-agricultural areas such as urban areas, bodies of water, native vegetation, and bare soils, among others. Photointerpretation of satellite images (1996–2022) was carried out due to the lack of records documenting agricultural growth within the study area.

2.2.4. Land Use Classification

Land cover monitoring using remote sensing data requires robust and accurate classification methods that enable the mapping of complex land cover and land use categories [42]. The estimation of the agricultural area was carried out using “Random Forest (RF)”, a machine learning classifier applied to the Landsat images described in Table 2. RF is a technique that collects several base models and combines them to obtain improved results [43]. This technique consists of the combination of the concept of “Decision trees” [44] and Breiman’s “Bagging” [45]. Thus, the concept of “Decision trees” represents an algorithm to classify information from a set of training data (previously selected) with the aim of automating and simplifying this process [46]. On the other hand, “Bagging” or “bootstrap aggregation” would be a type of joint learning, which improves the precision of a weak classifier by creating a set of classifiers [47]. That is, “Bagging” helps with the generation of multiple decision trees through the combination of these methods, improving the precision and generalization of the model [43]. A simplified version of its operation is represented in Figure 2.

This technique has been previously used in other Chilean studies, providing good results [48,49]. Furthermore, training polygons were provided to RF, allowing unsupervised classification data analysis by estimating missing values through the algorithm. The use of the Google Earth Engine cloud geoprocessing platform allowed the implementation of RF-supervised classification, pixel by pixel, for the four proposed years (1996, 2006, 2016, and 2022) of the Landsat images.

2.2.5. Estimation of Irrigation Surface

Once classifications were obtained, the results of the RF algorithm were transformed into vector algorithms using QGIS to calculate the amount of intensive irrigation surfaces in hectares for the four periods. Its result gave rise to the generation of intensively irrigated agricultural areas, for each basin and for the four defined years.

2.3. Groundwater Volume Estimation Methodology

Unfortunately, information regarding the size and shape of aquifers in Chile is limited (non-existent for the studied basins), making it difficult to determine whether surface basins coincide with them. However, the methodology used in this study allows for estimating the volume of recharge that the surface basin contributes to the aquifer, as detailed further down in this paper.

2.3.1. Preprocessing of Data for Calculating Baseflows

The calculation of baseflows was carried out in the two selected basins. This calculation was based on the information provided by the CR2 flow database [50], which includes daily flows recorded at its gauging stations (see Figure 1).

The flow analysis period considered data from 1996 to 2022, with particular interest in those data that were recorded between the beginning of the hydrological year and the end of the previous one, that is, between 30 March and 2 April for each year of analysis. This was carried out to determine the recessive behavior of floods when there were lower water reserves in the basin.

2.3.2. Hydrograph Separation and Model Fitting for the 1996–2022 Period

The focus of this research is on baseflow recession, observable in the typical recession hydrograph illustrated by Jain [51]. Similarly, Balocchi et al. [52] shows that the recession curve used includes the second break point of the descending curve of the hydrograph. According to Pizarro et al. [53], the exclusive inflow of groundwater begins from the third breakpoint, especially in arid and semi-arid areas. Keeping this in mind, for this research, the mathematical model of Remenieras, Potential, and Exponential Model 2, here and after the study by Balocchi et al. [52] (

Table 3. Equations for baseflow estimation.

Model	Equation	Source
Remenieras	$\mathrm{Q}\left(\mathrm{t}\right)={\mathrm{Q}}_0{e}^{-\mathsf{\alpha }\mathrm{t}}$	Balocchi et al. [52]
Potential	$\mathrm{Q}\left(\mathrm{t}\right)={\mathrm{Q}}_0{(1+\mathsf{\alpha }\mathrm{t})}^{-2}$	Cirugeda [54]
Balocchi et al.	$Q\left(\mathrm{t}\right)={\mathrm{Q}}_0{e}^{(-2\mathsf{\alpha }\sqrt{\mathrm{t}})}$	Balocchi et al. [52]

Where Q(t): Recessive flow at time t in m³ s⁻¹, Q₀: Initial recessive flow at time t₀. When the flow starts to be fed exclusively by groundwater reserves in m³ s⁻¹, e: Neper’s constant, t: time, α: Recession constant.

), were used.

To estimate the groundwater volumes stored in each basin, it was necessary to adjust each predefined model to the data available for each year and each basin at the beginning of the hydrological period (end of March or beginning of April). The first step was to define the value of Q₀, that is, the maximum flow rate of the recessive or exhaustion curve and the moment at which this occurs. To do this, on 31 March of each year and in each basin, the last major flood produced in the area was selected, establishing the last maximum flow or peak that defined the flood. This could be close to the date under study or could be delayed by a few months, especially if these are basins located in arid and semi-arid areas, as was the case. Thus, once the maximum flow is established, the hydrograph separation methodology [52] establishes that the logarithms of the flows must be graphed as a function of time, establishing a semi-logarithmic relationship, lnQ versus t to graphically establish the second break point of the descending curve, which defines the beginning of the recessive curve at the coordinate (t₀, Q₀). To clarify how to obtain the initial flow rate Q₀, Figure 3 shows an example of a last flood for the Illapel basin, before the start of the hydrological year, which occurred in January 1998. An average daily maximum flow can be seen to be slightly higher than 47 m³s⁻¹. The previous result (Figure 3) is transformed into a semi-logarithmic graph to estimate the break points in the curve (Figure 3). Thus, for the second break point, a flow value is obtained on the “Y” axis (applying an antilogarithm), which is called the baseflow, hereinafter “Q₀”, and from the “X” axis a time, hereinafter “t₀”. Once the initial baseflow Q₀ is defined, another value is taken corresponding to subsequent weeks or days, that is, a Q(t) at time t₀ + Δt. With this new value, the value of the parameter α for each model considered is obtained (Table 3).

2.3.3. Goodness of Fit (GOF)

In the field of hydrology, error metrics play a fundamental role in evaluating the quality of fit or in quantifying disparities between predicted and observed values in a time series [55]. They provide a measure for evaluating and interpreting hydrological models in terms of their predictive effectiveness based on real data.

The most popular system-scale performance metrics in hydrological modeling are Nash–Sutcliffe Efficiency (NSE) [56] and Kling–Gupta Efficiency (KGE) [57]. To validate the authenticity of the data, the use of NSE is ruled out as its value may not be as significant for highly variable data sets [58]. Instead, KGE is used, which is a more robust metric for outliers or that considers the synchronization of peaks and valleys [59]. The KGE equation is defined by Gupta et al. [57] as expressed in Equation (1). For this purpose, Rstudio was used to establish the KGE values corresponding to each basin and year, using the “Metric” package [60].

$$KGE=1-\sqrt{{\left(r-1\right)}^2+{\left(\alpha -1\right)}^2+{(\beta -1)}^2}$$

(1)

where r is the correlation between the observed and simulated series, α is the relative standard deviation of the simulated series compared to the observed ones, and β is the bias (the difference between the averages of the simulated and observed series).

2.3.4. Estimation of Base Volume

With the fitted models, a Q(t) or baseflow (Equation (2)) function was obtained. Subsequently, to calculate the annual volume of stored groundwater (m³), the expression of each model adjusted with its corresponding parameters was used. In this context, the objective was to estimate the volume stored as of 31 March of each year under analysis, or dates close to them. To this end, it was assumed that there will be no new contributions of water to the channel, whether from precipitation, snowmelt, or other sources. Therefore, the area under the curve of the equation that represents each model defined the total volume stored in the basin after each flood. Since it is very likely that the coordinate (t₀, Q₀) is on a date before 31 March, the volume flowing between t₀ and 31 March must be subtracted to find the volume stored as of 31 March, which is represented in the expressions shown in Equations (2) and (3).

$${\int }_{t_0}^{\mathrm{\infty }}Q\left(t\right)dt-{\int }_{t_0}^{31\mathrm{M}\mathrm{a}\mathrm{r}\mathrm{c}\mathrm{h}}Q\left(t\right)dt={\int }_{31\mathrm{M}\mathrm{a}\mathrm{r}\mathrm{c}\mathrm{h}}^{\mathrm{\infty }}Q\left(t\right)dt$$

(2)

$$\mathrm{Volume} \left({\mathrm{Hm}}^3\right)={\displaystyle \frac{24 \left(\mathrm{h}\right) \ast 3600 \left(\mathrm{s}\right)}{\mathrm{1,000,000}}}{\int }_{31\mathrm{M}\mathrm{a}\mathrm{r}\mathrm{c}\mathrm{h}}^{\mathrm{\infty }}Q\left(t\right)dt$$

(3)

2.4. Estimation of Precipitation

The precipitation calculation was carried out at two rainfall stations within the selected basins using daily data from CR2 [61], from 1996 to 2022 (Figure 1). The selected stations were Río Cogotí en Fraguita (04531003) and Río Illapel en Huintil (04726003).

The analysis was carried out at a monthly level (i.e., the monthly variation of precipitation per year was evaluated) to determine the behavior of precipitation and its possible impact on the water reserves of the basin.

2.5. Trends

Trend analysis focuses on the long-term trend component. However, it is unfeasible to achieve infinitely high sampling rates or perfect measurements [62], hence the need for statistical trend analysis methods. Therefore, in this investigation, the non-parametric Mann–Kendall (MK) test was applied, which has been frequently used to calculate the significance of trends in hydrometeorological time series [63].

The MK test examines each possible pair of data points, (x_i, y_i) and (x_j, y_j), to determine whether the pairs have the same relationship with each other, that is, whether there is a monotonic relationship or not [64]. This makes it possible to determine whether trends in water reserves show a significant decline over time. According to Pizarro et al. [13], the MK statistic for a time series of data {Z_k, k = 1, 2, …, n} is developed through the following process: (a) The values of the variables are listed in order, that is, from x₁ to x_n (x₁, x₂, …, x_n). (b) The sign of the difference in each pair of values is sought by comparing their magnitudes (x_j − x_k), with (j > k), following Equation (4). (c) The Mann–Kendall statistical S is obtained, where if the value of S is positive, it is inferred that the trend is increasing, while if it is negative, it is inferred that there is a decreasing trend, in which j and k represent two consecutive years (Equation (5)). (d) Then, based on the indicators, a variance is estimated for the MK S statistic, which considers the case of ties (Sign x_j − x_k = 0) obtained in step “b”, using Equation (6). And, (e) the statistical calculation of Z is carried out using Equation (7).

$$Sgn \left(X_j-X_k\right)=\left\{\begin{array}{c}1 if \left(X_j-X_k\right)>0\\ 0 if \left(X_j-X_k\right)=0\\ -1 if \left(X_j-X_k\right)<0\end{array}\right.$$

(4)

$$S=\sum _{k=1}^{n-1} \sum _{j=k+1}^{n}Sgn \left(X_j-X_k\right)$$

(5)

$$VAR={\displaystyle \frac{1}{18}}\ast \left[n (n-1) (2 n+5)-\sum _{q=1}^g{t}_q (t_{q }-1) (2 t_q+5)\right]$$

(6)

where n is the sample size, t_q is the frequency of the ties in group q, and q is the number of groups with ties.

$$Z=\left\{\begin{array}{c}{\displaystyle \frac{S-1}{{\left[VAR\left(S\right)\right]}^{0.5}}} if S>0\\ 0 if S=0\\ {\displaystyle \frac{S+1}{{\left[VAR\left(S\right)\right]}^{0.5}}} if S<0\end{array}\right.$$

(7)

Once the Z statistic is calculated, it is compared with the p value (0.05), allowing the null hypothesis to be accepted or rejected. Thus, H₀ is rejected if the p value returned by the MK test is less than 0.05 [65].

The MK calculation was performed using the Rstudio program with the “rkt” package [66] for the volume series of both basins. Additionally, the rate of change in the time series was estimated using the Theil–Sen formula (TS), a derivative that is robust to outliers compared to traditional linear regression [67]. From Equation (8), it is considered that i < j; i = 1, 2, …, (n − 1); j = 2, 3, …, n. Therefore, by combining them, it is possible to analyze trends in time series, and their use together offers a more comprehensive assessment of the presence and significance of these trends [67].

$$Theil-Sen slope=median {\displaystyle \frac{(y_j-y_i)}{(x_j-x_i)}}$$

(8)

The application of the non-parametric MK test to the data series of groundwater volumes stored in each basin and each year was carried out in two periods. First, the 1998–2010 period was analyzed, excluding the megadrought that has affected the country since that year, and the period between 1998 and 2020. The objective was to visualize the trend that stored volumes had before the incidence of a prolonged drought. The simplified process is illustrated in Figure 4.

3. Results

3.1. Land Use Changes

The results of the classification of agricultural land use are presented in Figure 5 for the four years under evaluation. This surface was estimated by Random Forest and Landsat images, using the basins delimited in previous steps (“Cogotí” and “Illapel”). Moreover, each year presented an accuracy of 0.9; 0.8; 0.8, and 0.8, respectively, through Random Forest classification and using the training polygons (i.e., the estimate is appropriate). In general, an incremental trend in the agricultural irrigation area amount is observed over time for each basin until 2016, with a strong decrease in 2022 (Figure 6), most likely due to the lack of water availability.

3.2. Estimation of Underground Volumes

During the investigation, it was confirmed that the daily flow data obtained from CR2 showed a lack of information for different years in the two basins. As a result, the absence of certain years in the analysis was observed. However, this lack did not prevent an adequate analysis of the data.

In terms of the determination of the start times of the selected recessive flows as of 31 March, the first aspect that needed to be addressed was the determination of the pairs of values (t₀, Q₀) for each year and for each basin, in order to model the baseflow in a given period. This determined the need to estimate the second break point of the descending curve of the flood hydrograph closest to and prior to 31 March. Examples of results associated with the breakpoints identified in the data for the 1996–2020 period for the two basins are detailed in Figure A1 and Figure A2, which provide a visual representation of the results segment, allowing for a more detailed understanding of the trends and patterns identified. An important detail to keep in mind is that the flows, having small values, which is typical of arid and semi-arid areas, had to be converted to liters per second (l s⁻¹), so that the logarithm could have value.

Once the second break points were determined, it was necessary to adjust the mathematical models of Remenieras, Potential, and Balocchi et al. [52] to the flow curve. Goodness of fit was assessed with the KGE test and the results of which are detailed in

Table A1. Values obtained for the Kling–Gupta Efficiency (KGE) Test for the Río Cogotí Station, the Cogotí Reservoir Entrance (04531002), for the estimation of recessive flows.

Year	Remenieras	Potential	Balocchi et al. [52]
1998	0.92	0.87	0.79
2000	0.80	0.81	0.76
2001	0.56	0.58	0.60
2002	0.70	0.71	0.65
2003	0.73	0.88	0.73
2004	0.68	0.85	0.77
2005	0.58	0.58	0.53
2006	0.89	0.89	0.81
2007	0.74	0.74	0.62
2009	0.87	0.87	0.95
2010	0.69	0.69	0.57
2011	0.59	0.46	0.35
2012	0.67	0.33	0.34
2013	0.50	0.83	0.69
2014	0.69	0.60	0.45
2017	0.69	0.70	0.58
2018	0.94	−0.04	0.13
2019	0.91	0.78	0.62
2020	0.74	0.70	0.53
$\overline{x}$	0.73	0.67	0.61

and

Table A2. Values obtained for the Kling–Gupta Efficiency (KGE) Test for the Rio Illapel Station, in El Peral (04726001), for the estimation of recessive flows.

Year	Remenieras	Potential	Balocchi et al. [52]
1998	0.85	0.95	0.84
1999	0.76	0.81	0.67
2000	0.83	0.84	0.69
2001	0.66	0.68	0.62
2002	0.63	0.64	0.59
2003	0.83	0.92	0.73
2004	0.79	0.79	0.68
2005	0.63	0.64	0.55
2006	0.98	0.62	0.53
2007	0.86	0.49	0.43
2008	0.55	0.56	0.50
2009	0.86	0.64	0.54
2013	0.53	0.55	0.54
2016	0.62	0.77	0.64
2017	0.51	0.51	0.41
2020	0.57	0.59	0.64
$\overline{x}$	0.72	0.69	0.60

. In the evaluation of the three models, the average for each basin was considered. When observing these averages, it stands out that both basins show better performance in terms of KGE with the Remenieras model, which is the reason it was finally chosen for further analyses (Figure 7).

An analysis of the results of the estimation of underground volumes as of 31 March, for each year and each basin, based on the Remenieras Model, was performed. This volume estimate is a default because part of the underground volumes in a basin may be evacuated to another basin, may be evacuated to the sea, or may evaporate from areas upstream of the control station. However, it is a first approximation to estimating water reserves at a given time. With this, it was possible to obtain an estimate of the volumes stored in various temporal scenarios. Based on the integral of the Remenieras’s expression (Equation (9)), storage volumes were obtained for each basin and times considered. Results are illustrated in Figure 8. The figure visually denotes a decrease in the volumes stored underground for the two basins. In fact, the values for some years are so low that they are not seen in the graph, as exemplified in Illapel for 2013 or in Cogotí for 2020. Cogotí exhibits the lowest average volume (

Table A3. Estimated underground volumes for each basin (Hm³).

Year	Cogotí	Illapel
1996	No data	No data
1997	No data	No data
1998	0.4918	15.211
1999	No data	0.254
2000	0.0168	0.314
2001	0.0340	2.953
2002	0.0767	1.181
2003	0.6887	8.704
2004	0.0289	0.420
2005	0.0311	0.504
2006	0.0124	0.145
2007	0.0075	0.116
2008	No data	0.338
2009	0.0085	0.110
2010	0.0226	No data
2011	0.0007	No data
2012	0.0406	No data
2013	0.0054	0.02655
2014	0.0011	No data
2015	No data	No data
2016	No data	0.10089
2017	0.00972	0.73790
2018	0.15065	No data
2019	0.00175	No data
2020	0.00297	1.3 × 10−7
2021	No data	No data
2022	No data	No data

), a situation that can be attributed to its smaller area compared to the other basin under study.

$${\int }_0^{\infty }{Q}_0 e^{-\mathsf{\alpha }\mathrm{t}}dt={\displaystyle \frac{Q_0}{\mathsf{\alpha }}}$$

(9)

At the macro level, the trends found are negative and significant in both periods. The results obtained are presented in

Table 4. Mann–Kendall trend results for two periods in the two basins.

Basin	Period	Theil-Sen Slope (Hm3)	p-Value	Z
Cogotí	1998–2009	−0.0056	0.049	−1.97
	1998–2020	−0.00178	0.014	−2.45
Illapel	1998–2009	−0.1389	0.0467	−1.99
	1998–2020	−0.0296	0.006	−2.75

, where it is possible to observe a significant negative trend of 0.0056 and 0.00178 Hm³ for both the first and second periods of analysis, respectively, at the Cogotí basin. Therefore, it is inferred that an important part of the underground volumes has decreased considerably towards the end of March or the beginning of April, which is the beginning of the hydrological year.

For the Illapel basin, on the other hand, a significant negative trend is observed for both periods (Table 4), with 0.1389 and 0.0296 Hm³, respectively. In the 1998–2009 period, the volume is almost five times greater than in the 1998–2020 period, indicating an accelerated decline in Illapel even before 2010, when the megadrought began, attributing such a depletion to the overconsumption of the resource. This is reinforced when considering that the trends in precipitation, despite being negative, are not significant (

Table 5. Mann–Kendall trend results for rainfall in two periods in the two stations.

	Cogotí			Illapel
Month	Theil-Sen Slope (mm)	p-Value	Z	Theil-Sen Slope (mm)	p-Value	Z
January *
February	0	1	−1.247	0	0.66	−0.633
March	0	0.21	−1.247	0	0.53	−0.633
April	0	0.52	−0.641	0	0.09	−1.72
May	0	0.88	0.2	−0.003	0.58	−0.547
June	−0.0012	0.98	−0.026	0.008	0.84	0.248
July	−0.0204	0.51	−0.653	−0.018	0.47	−0.72
August	−0.0054	0.8	−0.255	−0.018	0.26	−1.118
September	0	0.5	−0.681	0	0.53	−0.622
October	0	0.6	−0.531	0	0.73	−0.352
November	0	0.41	0.909	0	0.44	−0.765
December	0	0.16	1.576	0	0.16	1.576

* The test was not carried out for January due to the fact that there was no variation in the data.

). In other words, the results of the MK test and the TS slope for the Cogotí and Illapel basins do not verify trends that are significantly different from zero in both basins.

3.3. Analysis of the Relationship between Intensive Agricultural Use of Irrigated Land and the Trend in Stored Underground Volumes over Time

Figure 9 shows a comparison between the variables of intensive agricultural use under irrigation and the trend in groundwater reserves for the Cogotí and Illapel basins.

4. Discussion

The results achieved show an increase in the amounts of agricultural areas between 1996 and 2016 at the two study areas; such an increase was 64% and 45% for the Cogotí and Illapel basins, respectively. However, in 2022, there was a notable drop in the amount of irrigation surfaces, returning to the 1996 amount. These results coincide with the agricultural area and production reported by the inventories of the Chilean Natural Resources Research Center (CIREN) for the Coquimbo Region, where there was a higher value in 2018, but a significant decrease in 2021 [68]. The above could be explained by the presence of a megadrought that affected the central area of Chile between 2010 and 2020 [69,70,71].

On the other hand, the drip irrigation method has exceeded 89% in both basins, highlighting the production of high-value crops such as table grapes, avocados, and citrus. In this scenario, it is relevant to note that the evapotranspiration of grapes can vary between 30% and 70% [72], while citrus trees exhibit transpiration rates that fluctuate between 0.15 and 2.30 mm/day [73] and agro-industrial avocado production consumes up to 120% of the surface and groundwater volumes allocated to agricultural use during dry years [74]. Based on the above, the intense use of agricultural land, with the incorporation of highly water-demanding crops, is an evident example of the overuse of resources in both basins. Thus, the most evident proof is that from 2016 onwards, the ecosystem was not able to withstand the pressure and in a scenario of restricted supplies, it was necessary to abandon a significant proportion of the land because the amounts of water resources were not sufficient.

The estimation of underground stored water volumes was made through the integration of baseflows and the application of various mathematical models, where the Remenieras model presented the best fit to the existing data. As noted, this analysis was carried out at the beginning of the hydrological year in the southern hemisphere, that is, 31 March to 1 April of each year, with results showing relevant situations in both basins under study. Thus, it is important to highlight that the estimated volumes present low values in general, which is to be expected in basins located in arid areas with average annual rainfall that is less than 200 mm. In this way, the maximum reserve estimated for Cogotí in the period studied was 0.69 Hm³, while for Illapel, it was 15.2 Hm³. Likewise, the lowest value for the same period for Cogotí was 0.3 Hm³ and for Illapel, it was only a few cubic meters, values that are generally very low and reflect the situation of arid areas.

A similar study was conducted by Balocchi et al. [75], where the influence of native and exotic vegetation (Pine and Eucalyptus) on baseflows in eight basins in central Chile was evaluated. The results of this study show no significant differences between native and exotic vegetation, nor an explicit relationship between cover and baseflow. That is, no significant effect of cover on baseflow was detected. It should be noted that the exotic cover used in by Balocchi et al. [75] does not require irrigation and, therefore, the effect on flows is lower. In this study, an inverse relationship was found between the cover of fruit plantations and water reserves, mainly explained by a higher water consumption given that this type of monoculture needs constant irrigation. Another factor to consider is the prolonged drought affecting the country [69,70], derived from the fact that authors such as Lee and Ajami [76], when analyzing 358 basins in the United States, found that prolonged droughts (9–104 months) can impact baseflow amounts, and that these effects can last up to 41 months after the drought ends. This is important, since the megadrought in Chile has lasted more than 180 months and, therefore, can have a significant impact on baseflows. Parra et al. [77] studied water reserves in central-southern Chile, concluding that the depletion coefficient of the baseflow model is sensitive to dry periods, decreasing the value of the parameter during these periods.

When analyzing the trends obtained through the MK test, these were separated into two periods (1998–2009 and 1998–2020). The objective was to be able to see if the situation prior to the megadrought (i.e., the first period) was similar to the entire period integrating the megadrought. The results achieved indicate that for both the first and second periods in the Cogotí basin, a significant negative trend of 0.0056 and 0.00178, respectively, was observed. The existence of a negative and significant trend, even before the megadrought, suggests that this situation could be attributed to a variable other than this event. In this context and as previously mentioned, the excessive use of water resources emerges as a much more influential factor.

In the case of the Illapel basin, a significant negative trend (TS test) is also seen for both periods (0.1389 and 0.0296, respectively). The 1998–2009 period presents a volume almost five times greater than the 1998–2020 period, that is, an accelerated decrease is observed in Illapel even before 2010, which again reaffirms the overuse of the resource before the megadrought occurred.

The above is relevant because it indicates that before the existence of the mega-drought, stored volumes already showed a significant decrease. Likewise, carrying out flow analysis at the beginning of the hydrological year, that is, when water reserves reach their lowest point and do not receive new water inflows, provides crucial information on the retention, storage, and depletion of groundwater [51]. In other words, performing the analysis in the month of March reduces the relevance of precipitation, since more than 85% of it occurs between the months of June and September [78].

Having carried out the analysis on the recession flow is relevant because although it represents the gradual depletion of water storage during periods of little or no precipitation [79], it could still be slightly dependent on the distribution of precipitation [80].

Previous studies [13,81,82] analyzed precipitation trends in the Coquimbo Region, finding no significant variations. The results from precipitation analysis for the Illapel and Cogotí basins show a similar behavior, where precipitation trends are negative, but not significant. Regarding surface flows, negative and significant trends are observed in the average and maximum flows, which has determined an increase in the use of groundwater [13]. Similarly, Pizarro et al. [13] investigated the decline of groundwater in the region and found that the main driver of this decline is overuse (i.e., an anthropogenic factor). Therefore, this study focuses on crop production and its impact on baseflows. Moreover, future investigations should re-evaluate the behavior of baseflows and how climate change has influenced them.

Furthermore, Pizarro et al. [13] point out that in 2004, the basins of the region were closed for the granting of surface water rights, increasing since that date due to the use of groundwater. Thus, the apparent growth trend in this type of crop then focused on a very drastic use of groundwater, which only decreased in the second half of the 2010s, when extensive surfaces had to be left unused due to the non-availability of water in a period affected by the megadrought that hit the country and from which the Coquimbo Region has not yet recovered. Therefore, it follows that water resources were overexploited beyond their real capacities. This overexploitation had a significant impact on groundwater availability, especially considering that this water source supported the increase in agricultural activity from 2004 onwards.

It is important to mention that rainfall in the central-northern area of the country is greatly affected by climatic phenomena such as the El Niño Southern Oscillation (ENSO) or the Pacific Decadal Oscillation (PDO) [83], which determine that the recharge of aquifers occurs sporadically [84]. In this framework, the fraction of precipitation that eventually converts to recharge is therefore comparable in magnitude to the uncertainty range in recharge quantification methods [85]. However, despite these episodic events, there are studies that have not considered this variable for the estimation of recharges [86,87].

An important aspect of water scarcity is reflected in the quality of life of nearby populations. Reduced water availability limits crop and livestock yields, thereby decreasing family income, a situation that has deteriorated the mental health of these communities [85]. This deterioration has led to migration from rural to urban areas in search of better job opportunities and quality of life [85,86]. Furthermore, the overuse detected by Pizarro et al. [13] has the potential to accelerate water scarcity in rural areas, which are usually used (through deep wells) to address water scarcity in the middle term [87].

On the other hand, it is important to note that in Chile, there is a clear lack of water resources management at the basin level [9,88]. This deficiency makes it difficult to implement sustainable water resource use policies that integrate productive, social, and environmental aspects in their management and thus mitigate the impact of water scarcity on the population.

Finally, though cadastral the amount of information on agricultural land in Chile is limited, extending the timeframe or examining different time scales could provide a more comprehensive understanding of dynamic changes in groundwater reserves. Additionally, the studied basins were selected because they are in an area with high amounts of agricultural production and low amounts of annual rainfall. However, further studies should expand the geographic area under analysis to identify patterns in the behavior of baseflows. Similarly, factors such as urbanization and mining, as well as climate variation models, should be integrated in further analyses.

5. Conclusions

Based on the results from this investigation, it is possible to conclude that there was an increase in the areas allocated to crops with high water demand, in a proportion that was above the ecosystem’s capabilities, which determined that, in the midst of a megadrought, large areas of crops were abandoned.

It is also possible to conclude that the use of groundwater has increased and has determined a significant reduction in existing reserves, as demonstrated by the negative and significant trends obtained in this study.

Finally, it is concluded that overuse is a problem in these basins, above and beyond climatic factors, since the negative trends in groundwater availability manifest prior to the existence of the 2010–2021 megadrought and that determines the need to rethink the use of water resources under sustainability schemes.