Groundwater overexploitation along with decreasing precipitation exacerbates groundwater level decline and causes groundwater drought. Efficient assessment of the drought is critical to water management, especially in drought-prone agriculture regions. It remains challenging to characterize groundwater drought quantitatively due to the difficulty in obtaining groundwater observation data and the complexity of groundwater flow systems. To this end, agglomerative hierarchical cluster analysis was performed on long-term groundwater levels to classify wells in the San Joaquin River Basin, California. A Modified Mann–Kendall (MMK) test was undertaken to detect seasonal groundwater level trends from 1980 to 2019, and the magnitude was calculated using Sen's slope estimator. A nonparametric Standardized Groundwater level Index (SGI) was used to quantify the characteristics of groundwater drought. Results show that long-term (40-year) temporal patterns in groundwater levels varied significantly over the San Joaquin River Basin. Significantly decreasing trends were observed among more than 34.6% of wells, with an average decline of 0.69 m/year. Wells suffered frequent and severe groundwater droughts in the last decade, which were mainly driven by heavy groundwater exploitation. Findings provide useful information about the long-term behavior of regional groundwater levels, which in turn help stakeholders monitor droughts and adapt groundwater management strategies.

  • The long-term temporal patterns in groundwater levels were efficiently measured.

  • Characteristics of groundwater drought in the San Joaquin River Basin were quantified through the SGI.

  • Groundwater droughts varied significantly in space and time.

  • Heavy groundwater abstraction mainly drives groundwater drought.

C2VSim-FG

California Central Valley Groundwater-Surface Water Simulation Model-Fine Grid

ETDI

Evapotranspiration Deficit Index

GRACE

Gravity Recovery and Climate Experiment

GRI

Groundwater Resource Index

MMK

Modified Mann–Kendall

SGI

Standardized Groundwater level Index

SGMA

Sustainable Groundwater Management Act

SMDI

Soil Moisture Deficit Index

SPEI

Standardized Precipitation Evapotranspiration Index

SPI

Standardized Precipitation Index

Groundwater resources serve as a primary buffer against drought for agriculture irrigation worldwide, particularly in places like California that have both highly developed agricultural water supply infrastructure (Ho et al. 2016; Felfelani et al. 2017). However, groundwater resources are increasingly pressured due to intensive anthropogenic activity, threatening water availability to cope with future droughts (Scanlon et al. 2012; Goodarzi et al. 2016; Guo et al. 2021; Wang et al. 2022). The occurrence of frequent droughts can hinder groundwater recharge, which in turn, leads to heavily stressed aquifers and severe reduction in groundwater levels. Groundwater drought is caused by decreasing recharge and/or excessive pumping, and thereby results in various detrimental effects including a reduction in groundwater storage, land subsidence, and seawater intrusion (Ho et al. 2016; Mahlknecht et al. 2017; Pauloo et al. 2020). In California's Central Valley, home to one of the world's most productive agricultural regions, groundwater system dynamics is vulnerable to both natural and anthropogenic interventions. As a major source of California's water supply, groundwater provides close to 40% of the supply during normal years and up to 60% in drought years (Medellín-Azuara et al. 2015). Extensive reliance on groundwater leads to basins suffering from chronic groundwater overexploitation beyond natural recharge (Ho et al. 2016; Massoud et al. 2018). A large proportion of the San Joaquin Valley aquifer has suffered from chronic overdraft during the past few decades owing to high water demand for irrigation (Faunt 2009). According to the 2014 Sustainable Groundwater Management Act (SGMA), recent legislation requires critically overdrafted basins to reach balance between recharge and extraction within 20 years (Owen et al. 2019; Liu et al. 2021). Compared with surface water resources, groundwater storage may take significantly longer to be replenished. In this regard, it is imperative to assess and characterize groundwater drought for efficient management.

Groundwater drought can be quantified on the basis of groundwater levels, volume, recharge, and discharge. Many drought indices have been exploited to enable aspects of drought severity, duration, and/or the spatial extent to be characterized (Mishra & Singh 2010; Shahid & Hazarika 2010; Hao et al. 2014; Kim & Rhee 2016). Tirivarombo et al. (2018) applied the Standardized Precipitation Index (SPI) and the Standardized Precipitation Evapotranspiration Index (SPEI) to capture the variability of droughts at different time scales. Mendicino et al. (2008) presented a new Groundwater Resource Index (GRI) as a tool in a multi-aquifer system for drought monitoring and forecasting. Narasimhan & Srinivasan (2005) derived the Evapotranspiration Deficit Index (ETDI) and the Soil Moisture Deficit Index (SMDI) for agricultural drought monitoring based on weekly soil moisture and evapotranspiration deficit, respectively. Matera et al. (2017) derived a novel agricultural drought index, called DTx to describe transpiration deficit calculated by a water balance model. Zhang et al. (2005) exploited the capabilities of the Moderate resolution Imaging Spectroradiometer to forecast crop production using a satellite-based Climate-Variability Impact Index. Due to the difficulty in obtaining observational groundwater data and the complexity of groundwater flow systems, it remains challenging to characterize groundwater drought quantitatively. In this regard, this study adopts a Standardized Groundwater level Index (SGI) for standardizing groundwater level time series and quantifying groundwater drought with long-term groundwater level data. The SGI was modified from the Standardized Precipitation Index (SPI) and was initially developed by Bloomfield & Marchant (2013).

Investigating long-term groundwater trends is essential to groundwater drought monitoring and assessment. Previous studies have been carried out to investigate groundwater fluctuations (Panda et al. 2012; Mishra & Nagarajan 2013; Halder et al. 2020). Tabari et al. (2012) investigated annual groundwater trends and witnessed dominant positive trends in spring and summer seasons. Tiwari et al. (2009) explored groundwater storage change in Northern India and concluded severe groundwater loss caused by heavy irrigation. Rodell et al. (2009) used Gravity Recovery and Climate Experiment (GRACE) data to evaluate groundwater depletion rate in the Indian states of Rajasthan, Punjab, and Haryana (including Delhi). Trends in groundwater levels provide a thorough understanding of long-term temporal behavior exhibited by groundwater level time series and could be either negative (decreasing), positive (increasing), or zero (stable) (Le Brocque et al. 2018). Significantly decreasing groundwater trends lead to severe and frequent groundwater drought (Castle et al. 2014). Trend analysis characterized by groundwater-level in long-term time series (Famiglietti & Rodell 2013) covers both dry and wet periods, which in turn conveys implicit drought information. Despite previous studies of groundwater drought through analysis of groundwater hydrographs, more research should be specific to groundwater drought assessment by quantifying groundwater level fluctuation patterns in an agriculture-dominated area. Groundwater droughts are not a simple function of meteorological drivers. Groundwater droughts can vary in space and time due to the response of groundwater systems to changes in groundwater levels and spatial variations in aquifer characteristics. To date, there is a lack of systematic investigation of heterogeneities in groundwater droughts at the regional scale. Understanding how long-term groundwater level trends affect groundwater drought is necessary for sustainable water supply. To fill in the gap, the present study focuses on (1) cluster analysis in groundwater level time series; (2) long-term pattern detection in seasonal groundwater level time series; and (3) groundwater drought characteristics quantification in a heavily drafted agricultural region (the San Joaquin River Basin), where groundwater pumping is a main driving factor.

This work is structured as follows. In section 2, we first start with a description of the study area. Then, we present data materials and the associated methods. Section 3 presents results and discussion. Section 4 concludes this study with a brief outlook of future study.

Study area

The study area (∼8,429.05 km2) shown in Figure 1 covers most of the San Joaquin River Basin in California, which is bounded by the Sierra Nevada mountains on the northeast and the Coast Ranges on the southwest and located in California's Central Valley Aquifer. The Central Valley is a large structural trough filled with sediments of Jurassic to Holocene age, where sediments comprise unconfined, semi-confined and confined gravel, sand, silt, and clay. Most of freshwater is contained in the upper part of the sediments consisting of post-Eocene continental rocks and deposits, with thicknesses ranging from 1,000 to 3,000 feet (Williamson et al. 1989). The San Joaquin Valley covers southern two-thirds of the Central Valley. The upper semi-confined aquifer consists of three hydrogeologic units (Laudon & Belitz 1991) that grade into each other: Coast Ranges alluvium, Sierran alluvial deposits, and flood-basin deposits. The Coast Ranges alluvium varies from sands and gravels in creek-channel deposits (Laudon & Belitz 1991) near the heads of fans to silts and clays in interfan and distal fan areas, and is around 800 feet thick along the Coast Ranges (Miller et al. 1971). The Sierran alluvium comprises well-sorted medium-to-coarse-grained fluvial deposits (Miller et al. 1971), thinning eastward and westward and varying from 400 to 500 feet thick in the valley. The Coast Ranges alluvium and Sierran alluvium interfinger near the surface, and the contact extends westward with increasing depth (Laudon & Belitz 1991).

Figure 1

Map of the study area. (a) and (b) show groundwater levels at two wells. Please refer to the online version of this paper to see this figure in colour: http://dx.doi.org/10.2166/nh.2022.048.

Figure 1

Map of the study area. (a) and (b) show groundwater levels at two wells. Please refer to the online version of this paper to see this figure in colour: http://dx.doi.org/10.2166/nh.2022.048.

Close modal

wThe San Joaquin River Basin has experienced large changes in groundwater storage. The drought-prone area includes several major groundwater subbasins such as Delta-Mendota Subbasin, Turlock Subbasin, Merced, Chowchilla, and Madera Subbasins, which are key subareas comprising Depletion Study Areas (DSAs) (Williamson et al. 1989). Approximately two-thirds of the area was used for agriculture. Crops are dominated by cotton, almond orchards, deciduous fruit, nut orchards, and pasture. Groundwater is consumed intensively for irrigation in conjunction with surface water, causing overdrafts in groundwater basins and subbasins. Yin et al. (2021) presented more details about overdraft-induced groundwater depletion in the San Joaquin River Basin. It was estimated that 30% of annual water demand for agricultural lands and urban areas is provided by groundwater pumping during wet years and up to 70% during extremely dry years (Faunt 2009). Figure 2 estimates groundwater extraction percentage as well as agricultural water supply requirement in the study area (Figure 1) during 1980–2015, where data are available from California Central Valley Groundwater-Surface Water Simulation Model-Fine Grid (C2VSim-FG) (Dogrul & Kadir 2020). An estimated 53% of annual average groundwater was pumped for agricultural water supply, with a rising trend during a period of 36 years (1980–2015). Such an increase exacerbated the lowering of groundwater levels in the area, which also intensified the vulnerability of groundwater supply sources. In addition, Figure 3 estimated the total precipitation amount of the study area. An average decreasing precipitation of 0.015 mcf/year (∼50,690 m3/day) was detected during the past 36 years.

Figure 2

Estimated groundwater pumping percentage (left axis) as well as agricultural water supply requirement (right axis) in the study area during 1980–2015 (Data source: Water budget of C2VSim-FG from Dogrul & Kadir 2020). Please refer to the online version of this paper to see this figure in colour: http://dx.doi.org/10.2166/nh.2022.048.

Figure 2

Estimated groundwater pumping percentage (left axis) as well as agricultural water supply requirement (right axis) in the study area during 1980–2015 (Data source: Water budget of C2VSim-FG from Dogrul & Kadir 2020). Please refer to the online version of this paper to see this figure in colour: http://dx.doi.org/10.2166/nh.2022.048.

Close modal
Figure 3

Estimated precipitation in the study area (Data source: PRISM from Daly et al. 1994). Please refer to the online version of this paper to see this figure in colour: http://dx.doi.org/10.2166/nh.2022.048.

Figure 3

Estimated precipitation in the study area (Data source: PRISM from Daly et al. 1994). Please refer to the online version of this paper to see this figure in colour: http://dx.doi.org/10.2166/nh.2022.048.

Close modal

Data collection and processing

The periodic groundwater level dataset applied in the study contains long-term seasonal groundwater level measurements collected by the Department of Water Resources (DWR) and cooperating agencies in groundwater basins statewide (data can be accessed at https://data.ca.gov/dataset/periodic-groundwater-level-measurements). This dataset is maintained in the DWR Enterprise Water Management database, and contains information specific to monitoring well locations and groundwater level measurements collected at these wells. Typically, the groundwater level measurements (above mean sea level (msl)) are taken manually twice per year. The raw data were checked for data availability with continuous observations. Finally, available seasonal groundwater levels were collected along with locations of 104 observation wells that are widely distributed in the domain. The wells with long-term groundwater level measurements spanning from March 1980 to October 2019 were analyzed in the study.

Agglomerative hierarchical clustering

The groundwater level considered for drought assessment varies significantly from well to well. As a result, it is impractical to propose a groundwater drought for each well. Therefore, cluster analysis was carried out to classify well groups according to the similarity of water-level fluctuations. Cluster analysis is an efficient statistical method, which classifies elements based on similarity or dissimilarity associated with variables. In the study, elements exhibiting similar groundwater fluctuations were clustered into different groups by accounting for long-term seasonal groundwater levels. Agglomerative hierarchical clustering with a complete linkage method was performed to identify homogeneous wells based on groundwater level hydrographs. Compared with partitional clustering algorithms such as K-means, agglomerative hierarchical clustering does not require feature extraction or dimensionality reduction and hence is suitable for handling real-world data (Bouguettaya et al. 2015). Each data point is initially considered as an individual cluster in the agglomerative hierarchical clustering procedure. At each iteration, similar clusters merge with others until one cluster or K clusters are formed, which can be visualized using a dendrogram. The Euclidean Distance is used to measure the similarities between the wells (Shiau & Lin 2016). A distance matrix is calculated first. Each well is considered as a separate group. Then, the two groups that are most similar are combined into a new bigger group on the basis of the complete linkage method. In this study, the agglomerative hierarchical cluster analysis was performed with ‘hclust’ in R statistical environment (R core Team 2017).

Nonparametric trend test for groundwater levels

The nonparametric Mann–Kendall test is highly suitable for trend detection in hydrological variables since it does not require data to be normally distributed. The test was originally developed by Mann and later further developed by Kendall (Mann 1945; Kendall 1948). The Mann–Kendall statistic (S) is expressed as:
(1)
(2)
where and are the sequential data values at the time of i and j, respectively. N represents the length of the time-series data. If N >8, statistics S approximates to normal distribution. The mean of S is 0 and the variance of S is calculated as follows:
(3)
where var(S) is the variance of the Mann–Kendall Statistic.
Then, the test statistic Z is denoted as follows:
(4)
A Z value greater than zero indicates an increasing trend, and vice versa. Given a significance level , the sequential data would experience a statistically significant trend if where is the corresponding value of following a standard normal distribution. The study adopted for the test.
The original Mann–Kendall test is not appropriate for time-series data where observations are highly seasonal and serially correlated. In order to incorporate autocorrelation effects on trends of groundwater level fluctuations, a Modified Mann–Kendall (MMK) test (Kendall 1948; Hamed & Rao 1998) was conducted to identify seasonal trends of groundwater levels for all wells in the study area. The trend test was conducted in the R programming language with a confidence level of 95%. The variance in Equation (3) is corrected as follows:
(5)
where n is the number of tied group and is the number of data in the ith group (i = 1∼n).

Sen's slope estimator

The magnitude of the trend is analyzed using Sen's slope estimator as developed by Sen (Yu et al. 1993). The trend is calculated as follows:
where is Sen's slope estimate. An upward trend in a time series is indicated by a positive . Otherwise, the data imply a downward trend during the time period of interest. Computer programs were coded using R to calculate Z (Equation (4)) and (Equation (6)) in the procedures of the Mann–Kendall trend test and the Sen's slope estimator.

Calculation procedure of the SGI for groundwater drought quantification

Groundwater drought was evaluated in the study by adopting the SGI, which was proposed by Bloomfield & Marchant (2013). The SGI has been identified as related to the application of SPI-like normalization methods in the context of groundwater level data where groundwater levels at observation boreholes are a useful measure of the quantitative status of groundwater resources. Specifically, the SGI is a nonparametric method that transforms normal scores of the time series of groundwater level data with an inverse normal cumulative distribution function. Then, the transformed values are arranged consecutively to obtain the SGI time series. The flexibility of the calculation procedure over different time series allows for monitoring of both short-term and long-term droughts (Mishra & Singh 2010). Groundwater drought characteristics such as duration, severity, and intensity are finally calculated to quantify groundwater drought. Drought duration refers to a period in which the magnitude of the SGI is below the threshold. Severity is the sum of negative SGI values during drought, and intensity is defined as the ratio of the drought severity to the drought duration. The severity of a drought is determined by the departure of a negative SGI value from zero (McKee et al. 1993). A threshold level approach for SGI is used (Peters et al. 2003), signifying the influence of drought at a specific site. According to the calculation, this study categorized SGI ≤ –1.0 as severe groundwater drought, −1.0 < SGI ≤ 0 as moderate drought, and SGI ≥ 0 as no groundwater drought. The nonparametric SGI is calculated in MATLAB R2016a based on the groundwater level hydrographs of representative well sites from each cluster (Farahmand & AghaKouchak 2015).

Classification of wells based on cluster analysis

Cluster analysis was performed to classify wells based on their similarities in groundwater-level fluctuations. Agglomerative hierarchical clustering with a complete linkage method was carried out using the library ‘cluster’ and ‘factoextra’ in the R programming language, and the resulting dendrogram was shown in Figure 4. In order to identify distinct clusters, the dendrogram has to be cut at a specific linkage distance level called the threshold. The distance shows the similarity between observations. Generally, threshold selection is subjective. For example, a threshold at 800 would produce two clusters, whereas a threshold at 600 would yield three clusters. A gap statistic method was employed in the study to determine the optimal number of clusters during the agglomerative hierarchical clustering process (Tibshirani et al. 2001). The gap statistic compared the total intracluster variation for different number of clusters with their expected values under the null reference distribution of data. The optimal number of clusters was obtained from the corresponding gap statistic in such a way that the increment rate from those clusters was not significant. The resulting number of clusters in the study case was five, as shown in Figure 4. The spatial variation of the wells within the resulting five different clusters was displayed on the map of the study area (Figure 5). It can be noticed from Figure 5 that the group including 39 wells in the northern portion showed similar water-level patterns and were hence termed as cluster 5 (red dots), whereas other wells were dispersed over the area and entitled as different clusters. The box plot in Figure 6 indicated the groundwater level median of each cluster with their respective variations. The groundwater level median of the wells in cluster 1 was 31.25 m above msl while wells of cluster 3 had a relatively higher groundwater level (93.75 m) among all clusters. According to the variations, Figure 6 manifested that groundwater levels of the wells from cluster 1 fluctuate more significantly than others.

Figure 4

Cluster dendrogram visualization. Please refer to the online version of this paper to see this figure in colour: http://dx.doi.org/10.2166/nh.2022.048.

Figure 4

Cluster dendrogram visualization. Please refer to the online version of this paper to see this figure in colour: http://dx.doi.org/10.2166/nh.2022.048.

Close modal
Figure 5

Map of classified wells in the five clusters. Please refer to the online version of this paper to see this figure in colour: http://dx.doi.org/10.2166/nh.2022.048.

Figure 5

Map of classified wells in the five clusters. Please refer to the online version of this paper to see this figure in colour: http://dx.doi.org/10.2166/nh.2022.048.

Close modal
Figure 6

Groundwater level variations in the wells of each cluster. Please refer to the online version of this paper to see this figure in colour: http://dx.doi.org/10.2166/nh.2022.048.

Figure 6

Groundwater level variations in the wells of each cluster. Please refer to the online version of this paper to see this figure in colour: http://dx.doi.org/10.2166/nh.2022.048.

Close modal

Trend detection of seasonal groundwater level

The nonparametric Mann–Kendall trend test was applied at a significance level of 0.05 to assess seasonal groundwater levels for all observation wells in the study domain. Prior to the Mann–Kendall test, the serial correlation of groundwater levels for all the wells was conducted in R programming, and a significant lag 1 autocorrelation was observed for seasonal groundwater levels for a few wells. This study employed the modified Mann–Kendall test to incorporate the autocorrelation effects on trends of groundwater level fluctuations and identify seasonal trends of all the wells. A significance level of 0.05 was used to test if the trend is significant or not. The results are presented in Table 1 along with their respective Sen's slopes. Positive values of Z in Equation (4) indicate upward trends while negative values represent downward trends. The response of groundwater levels is influenced by the balance between recharge and discharge. Significant trends were observed in the seasonal groundwater levels, as shown in Figure 7. Table 2 describes details of the groundwater trend detection statistics. 77 wells showed a downward trend, and 48% of these wells decreased significantly (a significance level of 0.05). 27 wells behaved an upward trend and nearly 30% increased significantly. Specifically, the 36 significantly decreasing wells include four from cluster 1, one from cluster 2, eight from cluster 3, nine from cluster 4, and 14 from cluster 5. The average decrease in groundwater levels is 0.69 m/year. The widespread groundwater head declines were attributable to heavy pumping in the system. Especially, lowered groundwater levels in cluster 5 responded to intensive groundwater withdrawals for irrigation. Similarly, a total number of eight wells belonging to clusters 2–5 presented significantly increasing trends with an average rise in groundwater levels by 0.81 m/year. Besides, a few wells that are close spatially were found to show distinct groundwater level patterns. This is due to the wells that are located in distinct depths of unconfined, semi-confined, and confined hydrogeologic units, as described in section 2.

Table 1

Mann–Kendall statistics and Sen's slope values for seasonal groundwater levels

ClusterWell IDZSen's slopeClusterWell IDZSen's slope
Cluster 1 27 −4.4335 −1.7735  37 2.5044 2.6750 
22 −4.3372 −1.6250 38 2.4247 2.5650 
25 −5.1860 −1.4231 −2.9948 −2.1434 
91 −0.6622 −0.6993 56 −2.6357 −0.3176 
92 −0.8226 −2.0227 13 −1.1256 −0.4265 
26 −2.3550 −0.9025 57 −2.5839 −0.2408 
88 −0.7905 −1.0575 103 −4.0183 −0.9527 
104 −0.8195 −0.7978 67 −2.0013 −0.1522 
Cluster 2 96 0.2661 0.1500 60 −1.8064 −0.0412 
85 −0.6818 −0.2914 78 −2.4918 −0.0682 
43 1.0171 0.4274 50 −2.5433 −0.2600 
44 −1.4583 −0.0551 49 −2.9391 −0.4145 
76 −0.5330 −0.0143 58 −3.6468 −0.7986 
95 2.1995 2.5607 Cluster 5 −1.5767 −1.4222 
42 −0.2310 −0.3250 47 −1.3098 −0.0889 
97 −0.7489 −0.1667 48 −1.8770 0.1402 
23 −0.9432 −0.5529 51 −1.4718 −0.1133 
28 −3.7320 −1.3636 71 −0.8191 −0.0583 
Cluster 3 100 1.0234 0.5071 55 1.0816 0.3086 
41 0.2452 0.0762 45 −0.0835 −0.0001 
101 −1.3621 −0.4250 46 −1.3761 −0.0540 
102 −2.0858 −0.7125 61 −1.9868 −0.0368 
90 0.3788 0.4200 86 −2.2727 −0.8050 
83 −1.5881 −0.1468 81 −0.8257 −0.0615 
82 −0.5081 −0.0027 52 −3.0721 −0.2400 
84 −0.9214 −0.0188 53 −0.8361 −0.2000 
34 −1.5540 −2.4825 21 1.2280 0.2857 
93 0.6818 0.3457 72 −0.4613 −0.0143 
24 0.8040 0.1306 94 0.0379 0.0100 
89 1.0444 1.0657 19 0.3411 0.0800 
98 1.8399 2.3644 74 −0.8468 −0.0343 
11 −1.7559 −1.1392 54 −0.6313 −0.0056 
12 −1.7559 −0.7750  17 −0.2029 −0.0575 
10 −3.5171 −0.6813 18 −0.3791 −0.0500 
−1.6614 −0.2507 66 0.0112 0.0050 
3.3544 0.3269 14 2.1697 0.1323 
0.0021 0.0214 15 0.4959 0.0254 
−3.9644 −1.3477 16 −0.1519 −0.0050 
−3.6312 −0.9050 73 −2.0789 −0.0263 
−2.4337 −1.2266 80 −1.1779 −0.0373 
59 1.4431 0.0433 87 −4.0150 −0.9025 
65 −2.1110 −0.0414 64 −5.3625 −0.1656 
68 0.4810 0.0091 77 −3.6042 −0.2175 
69 −0.0873 −0.0050 30 −2.5287 −0.4290 
−2.5663 −0.8750 32 −2.6925 −0.3357 
79 1.9767 0.0756 62 −3.6661 −0.4321 
99 −1.3951 −0.0997 20 −2.5156 −0.6667 
29 −2.5888 −0.5500 63 −3.4270 −0.2600 
33 −1.2902 −1.2174 70 −1.1723 −0.5120 
Cluster 4 40 2.7839 3.3417 75 −0.4557 −0.0864 
39 1.7249 2.9650 31 −3.1665 −0.6767 
36 2.8622 2.8383 35 −1.4772 −1.0758 
ClusterWell IDZSen's slopeClusterWell IDZSen's slope
Cluster 1 27 −4.4335 −1.7735  37 2.5044 2.6750 
22 −4.3372 −1.6250 38 2.4247 2.5650 
25 −5.1860 −1.4231 −2.9948 −2.1434 
91 −0.6622 −0.6993 56 −2.6357 −0.3176 
92 −0.8226 −2.0227 13 −1.1256 −0.4265 
26 −2.3550 −0.9025 57 −2.5839 −0.2408 
88 −0.7905 −1.0575 103 −4.0183 −0.9527 
104 −0.8195 −0.7978 67 −2.0013 −0.1522 
Cluster 2 96 0.2661 0.1500 60 −1.8064 −0.0412 
85 −0.6818 −0.2914 78 −2.4918 −0.0682 
43 1.0171 0.4274 50 −2.5433 −0.2600 
44 −1.4583 −0.0551 49 −2.9391 −0.4145 
76 −0.5330 −0.0143 58 −3.6468 −0.7986 
95 2.1995 2.5607 Cluster 5 −1.5767 −1.4222 
42 −0.2310 −0.3250 47 −1.3098 −0.0889 
97 −0.7489 −0.1667 48 −1.8770 0.1402 
23 −0.9432 −0.5529 51 −1.4718 −0.1133 
28 −3.7320 −1.3636 71 −0.8191 −0.0583 
Cluster 3 100 1.0234 0.5071 55 1.0816 0.3086 
41 0.2452 0.0762 45 −0.0835 −0.0001 
101 −1.3621 −0.4250 46 −1.3761 −0.0540 
102 −2.0858 −0.7125 61 −1.9868 −0.0368 
90 0.3788 0.4200 86 −2.2727 −0.8050 
83 −1.5881 −0.1468 81 −0.8257 −0.0615 
82 −0.5081 −0.0027 52 −3.0721 −0.2400 
84 −0.9214 −0.0188 53 −0.8361 −0.2000 
34 −1.5540 −2.4825 21 1.2280 0.2857 
93 0.6818 0.3457 72 −0.4613 −0.0143 
24 0.8040 0.1306 94 0.0379 0.0100 
89 1.0444 1.0657 19 0.3411 0.0800 
98 1.8399 2.3644 74 −0.8468 −0.0343 
11 −1.7559 −1.1392 54 −0.6313 −0.0056 
12 −1.7559 −0.7750  17 −0.2029 −0.0575 
10 −3.5171 −0.6813 18 −0.3791 −0.0500 
−1.6614 −0.2507 66 0.0112 0.0050 
3.3544 0.3269 14 2.1697 0.1323 
0.0021 0.0214 15 0.4959 0.0254 
−3.9644 −1.3477 16 −0.1519 −0.0050 
−3.6312 −0.9050 73 −2.0789 −0.0263 
−2.4337 −1.2266 80 −1.1779 −0.0373 
59 1.4431 0.0433 87 −4.0150 −0.9025 
65 −2.1110 −0.0414 64 −5.3625 −0.1656 
68 0.4810 0.0091 77 −3.6042 −0.2175 
69 −0.0873 −0.0050 30 −2.5287 −0.4290 
−2.5663 −0.8750 32 −2.6925 −0.3357 
79 1.9767 0.0756 62 −3.6661 −0.4321 
99 −1.3951 −0.0997 20 −2.5156 −0.6667 
29 −2.5888 −0.5500 63 −3.4270 −0.2600 
33 −1.2902 −1.2174 70 −1.1723 −0.5120 
Cluster 4 40 2.7839 3.3417 75 −0.4557 −0.0864 
39 1.7249 2.9650 31 −3.1665 −0.6767 
36 2.8622 2.8383 35 −1.4772 −1.0758 
Table 2

Statistics of the Mann–Kendall test

Increasing positive trends (27) Cluster 1 Significantly increasing (8) Cluster 1 
Cluster 2 Cluster 2 
Cluster 3 12 Cluster 3 
Cluster 4 Cluster 4 
Cluster 5 Cluster 5 
Decreasing negative trends (77) Cluster 1 Significantly decreasing (36) Cluster 1 
Cluster 2 Cluster 2 
Cluster 3 19 Cluster 3 
Cluster 4 11 Cluster 4 
Cluster 5 32 Cluster 5 14 
Increasing positive trends (27) Cluster 1 Significantly increasing (8) Cluster 1 
Cluster 2 Cluster 2 
Cluster 3 12 Cluster 3 
Cluster 4 Cluster 4 
Cluster 5 Cluster 5 
Decreasing negative trends (77) Cluster 1 Significantly decreasing (36) Cluster 1 
Cluster 2 Cluster 2 
Cluster 3 19 Cluster 3 
Cluster 4 11 Cluster 4 
Cluster 5 32 Cluster 5 14 
Figure 7

Groundwater level trends in the study area. Please refer to the online version of this paper to see this figure in colour: http://dx.doi.org/10.2166/nh.2022.048.

Figure 7

Groundwater level trends in the study area. Please refer to the online version of this paper to see this figure in colour: http://dx.doi.org/10.2166/nh.2022.048.

Close modal

Quantification of groundwater drought by SGI hydrographs

Groundwater drought was evaluated using the SGI for all the collected 104 wells over the San Joaquin River Basin. Since the wells were classified into five clusters based on similarities in the water-level patterns, a selection of wells from each respective cluster was presented within the manuscript to summarize the SGI assessment. Considering groundwater level time series and spatial distribution, the wells with ID 91, 44, 82, 50, and 73 were selected from cluster 1 to cluster 5, respectively. Results of the SGI time series of the five wells are presented in Figure 8, which depicted SGI temporal variations of the representative wells. Negative values of SGI indicate different groundwater drought conditions while positive values of SGI (SGI > 0) represent normal conditions (no groundwater drought). The drought characteristics including severity, duration, and intensity disclosed that wells from cluster 3 and cluster 4 suffered maximum severe drought with a longer duration of 12 seasons (SGI ≤ –1.0), as shown in Table 3. 11 seasons of maximum drought were observed in the selected well from cluster 2. The well from cluster 2 experienced severe groundwater drought during 2014–2019 with an average intensity of SGI equal to −1.41, while suffering from moderate groundwater drought (−1.0 < SGI ≤ 0) during 1988–1996. The selected well from cluster 5 suffered a severe groundwater drought in 1996. However, a longer duration of drought occurred thereafter from 2009 to 2019. The drought in cluster 5 can relate to long-term significantly decreasing wells according to trend analysis. This was mainly driven by increased groundwater consumption of crop irrigation during the last decade as well as rather limited surface water availability. Besides, the decreasing precipitation (see Figure 3) may also intensify the likelihood of groundwater droughts in some basins, especially during the past few years. However, for deep confined aquifers, precipitation impact can be neglected. Groundwater withdrawal has also increased to sustain rapidly expanding perennial crops during drought. The temporal variations of SGI revealed that the selected well from cluster 5 experienced more frequent droughts in the last decade. In contrast, groundwater drought alleviated (−1.0 < SGI ≤ 0) in the well from cluster 3 in.the past decade while the situation was severe during 1992–1997. Groundwater drought situations varied in different areas and time periods due to the complexity of geographical location, agricultural irrigation and other natural factors. Groundwater drought observed in the representative wells of each cluster provides insight into future risks for agricultural groundwater supply in the basin. Recurrent groundwater droughts affecting the wells from clusters 1, 2, 4, and 5 during the last decade strongly conveyed that the underlying aquifer was getting deteriorated from year to year. In addition to the prolonged droughts, significant declines in groundwater levels have exacerbated drought intensity, indicating that some existing shallow agricultural wells may be vulnerable to run dry.

Table 3

Drought characteristics based on SGI values for representative wells

Well clustersDrought intensityMaximum durationMaximum severity
Cluster 1 −1.37 −9.57 
Cluster 2 −1.41 11 −15.55 
Cluster 3 −1.35 12 −16.23 
Cluster 4 −1.33 12 −16.01 
Cluster 5 −1.38 −6.91 
Well clustersDrought intensityMaximum durationMaximum severity
Cluster 1 −1.37 −9.57 
Cluster 2 −1.41 11 −15.55 
Cluster 3 −1.35 12 −16.23 
Cluster 4 −1.33 12 −16.01 
Cluster 5 −1.38 −6.91 
Figure 8

SGI time series of each representative well from five clusters of (a)–(e). Please refer to the online version of this paper to see this figure in colour: http://dx.doi.org/10.2166/nh.2022.048.

Figure 8

SGI time series of each representative well from five clusters of (a)–(e). Please refer to the online version of this paper to see this figure in colour: http://dx.doi.org/10.2166/nh.2022.048.

Close modal

Continuous groundwater overuse exacerbates pervasive overdraft which results in groundwater level declines, and groundwater drought issues in agriculture-dominated areas. It remains critical but challenging to characterize groundwater drought quantitatively due to the difficulty in obtaining observational groundwater data and the complexity of groundwater flow systems. An extensive study is needed to focus on long-term groundwater level trend detection and regional groundwater drought quantification. In this study, seasonal patterns in groundwater level time series were identified through investigating 104 real-world monitoring wells widely distributed in the San Joaquin River Basin, California, from 1980 to 2019. Before evaluating groundwater droughts, an efficient agglomerative hierarchical cluster analysis was first carried out to classify wells according to the similarities in groundwater level fluctuations. Results of the gap statistics implied that five clusters were sufficient to efficiently manifest groundwater level patterns in the basin. The modified Mann–Kendall test and Sen's slope estimator were performed to detect seasonal groundwater level trends of all the wells. The results revealed both increasing and decreasing trends in the groundwater level time series of the study area. However, 77 out of the 104 wells were found decreasing. 36 from the decreasing wells showed significant downward trends with an annual average decline of 0.69 m. Almost half of the descending wells from cluster 5 (mainly concentrated in the northern counties) showed significant depletion in groundwater levels, which can be attributed to continuous groundwater overexploitation together with declining rainfall. Besides, long-term overdraft increases the likelihood of lengthened groundwater droughts.

Along with identifying long-term trends, groundwater drought characteristics were effectively quantified using the nonparametric SGI. Analysis of SGI confirmed a large number of groundwater drought events occurred in this area. Also, groundwater drought varied significantly in different clusters due to the complexity of geographical location and agricultural irrigation. In particular, the wells in clusters 2, 4, and 5 have frequently suffered from severe groundwater droughts over the past decade, and the area would be more vulnerable to upcoming groundwater drought. Special and timely attention would be required in the protection, use, and management of groundwater resources in these affected areas. The findings of this study will assist decision-makers and planners in adapting or optimizing the existing land use and groundwater use policies, especially in an overdrafted groundwater basin. High-resolution spatio-temporal groundwater data may be preferable in the future to facilitate assessment of potential groundwater drought risk.

The work was sponsored by the National Key R&D program of China (2021YFC3200500), the National Natural Science Foundation of China (52109080), Fundamental Research Funds for the Central Universities (B220201013). The study also acknowledges support from School of Engineering through the Water Systems Management Lab of UC Merced, California. The authors would like to thank the California Department of Water Resources for providing the periodic groundwater levels dataset in groundwater basins statewide. The CASGEM (California Statewide Groundwater Elevation Monitoring) Program is acknowledged for granting access to groundwater well location information.

J.Y. conceptualized the study; performed data curation; designed the methodology; did formal analysis; acquisition of funds; validated, visualized, and wrote the original draft. J.M.-A. collected funds; conceptualized the study; designed the methodology; supervised the study, collected the resources; validated the study; wrote, reviewed, and edited the article. A.E.-B. performed data curation; designed the methodology; validated, wrote, reviewed, and edited the article.

Data cannot be made publicly available; readers should contact the corresponding author for details.

The authors declare there is no conflict.

Bloomfield
J. P.
&
Marchant
B. P.
2013
Analysis of groundwater drought building on the standardised precipitation index approach
.
Hydrology and Earth System Sciences
17
,
4769
4787
.
https://doi.org/10.5194/hess-17-4769-2013.
Bouguettaya
A.
,
Yu
Q.
,
Liu
X.
,
Zhou
X.
&
Song
A.
2015
Efficient agglomerative hierarchical clustering
.
Expert Systems with Applications
42
(
5
),
2785
2797
.
https://doi.org/10.1016/j.eswa.2014.09.054.
Castle
S. L.
,
Thomas
B. F.
,
Reager
J. T.
,
Rodell
M.
,
Swenson
S. C.
&
Famiglietti
J. S.
2014
Groundwater depletion during drought threatens future water security of the Colorado River Basin
.
Geophysical Research Letters
41
(
16
),
5904
5911
.
https://doi.org/10.1002/2014GL061055.
Daly
C.
,
Neilson
R. P.
&
Phillips
D. L.
1994
A statistical-topographic model for mapping climatological precipitation over mountainous terrain
.
Journal of Applied Meteorology and Climatology
33
,
140
158
.
Dogrul
E. C.
&
Kadir
T. N.
2020
Theoretical documentation for the integrated water flow model (IWFM-2015), Revision 961
.
California Department of Water Resources
:
Sacramento, CA, USA
.
Famiglietti
J. S.
&
Rodell
M.
2013
Water in the balance
.
Science
340
(
6138
),
1300
1301
.
doi:10.1126/science.1236460
.
Farahmand
A.
&
AghaKouchak
A.
2015
A generalized framework for deriving nonparametric standardized drought indicators
.
Advances in Water Resources
76
,
140
145
.
doi:10.1016/j.advwatres.2014.11.012
.
Faunt
C. C.
2009
Groundwater Availability of the Central Valley Aquifer, California: U.S. Geological Survey Professional Paper 1766 225 p
.
Felfelani
F.
,
Wada
Y.
,
Longuevergne
L.
&
Pokhrel
Y. N.
2017
Natural and human-induced terrestrial water storage change: a global analysis using hydrological models and GRACE
.
Journal of Hydrology
553
(
10
),
105
118
.
https://doi.org/10.1016/j.jhydrol.2017.07.048.
Goodarzi
M.
,
Abedi-Koupai
J.
,
Heidarpour
M.
&
Safavi
H. R.
2016
Development of a new drought index for groundwater and its application in sustainable groundwater extraction
.
Journal of Water Resources Planning and Management
142
(
9
),
04016032
.
https://doi.org/10.1061/(ASCE)WR.1943-5452.0000673.
Guo
M.
,
Yue
W.
,
Wang
T.
,
Zheng
N.
&
Wu
L.
2021
Assessing the use of standardized groundwater index for quantifying groundwater drought over the conterminous US
.
Journal of Hydrology
598
,
126227
.
https://doi.org/10.1016/j.jhydrol.2021.126227.
Hamed
K. H.
&
Rao
A. R.
1998
A modified Mann-Kendall trend test for autocorrelated data
.
Journal of Hydrology
204
(
14
),
182
196
.
https://doi.org/10.1016/S0022-1694(97)00125-X.
Hao
Z.
,
AghaKouchak
A.
,
Nakhjiri
N.
&
Farahmand
A.
2014
Global integrated drought monitoring and prediction system
.
Scientific Data
1
(
140001
),
1
10
.
doi:10.1038/sdata.2014.1
.
Ho
M.
,
Parthasarathy
V.
,
Etienne
E.
,
Russo
T. A.
,
Devineni
N.
&
Lall
U.
2016
America's water: agricultural water demands and the response of groundwater
.
Geophysical Research Letters
43
(
14
),
7546
7555
.
https://doi.org/10.1002/2016GL069797.
Kendall
M. G.
1948
The Advanced Theory of Statistics
.
Griffin
,
London
.
Kim
D.
&
Rhee
J.
2016
A drought index based on actual evapotranspiration from the Bouchet hypothesis
.
Geophysical Research Letters
43
(
19
),
10
277
.
https://doi.org/10.1002/2016GL070302.
Laudon
J.
&
Belitz
K.
1991
Texture and depositional history of Late Pleistocene-Holocene alluvium in the central part of the western San Joaquin Valley, California
.
Bulletin of the Association of Engineering Geologists
28
(
1
),
73
88
.
Le Brocque
A. F.
,
Kath
J.
&
Reardon-Smith
K.
2018
Chronic groundwater decline: a multi-decadal analysis of groundwater trends under extreme climate cycles
.
Journal of Hydrology
561
(
6
),
976
986
.
https://doi.org/10.1016/j.jhydrol.2018.04.059.
Liu
Z.
,
Herman
J. D.
,
Huang
G.
,
Kadir
T.
&
Dahlke
H. E.
2021
Identifying climate change impacts on surface water supply in the southern Central Valley, California
.
Science of The Total Environment
759
,
143429
.
Mahlknecht
J.
,
Merchán
D.
,
Rosner
M.
,
Meixner
A.
&
Ledesma-Ruiz
R.
2017
Assessing seawater intrusion in an arid coastal aquifer under high anthropogenic influence using major constituents, Sr and B isotopes in groundwater
.
Science of the Total Environment
587
,
282
295
.
https://doi.org/10.1016/j.scitotenv.2017.02.137.
Mann
H. B.
1945
Nonparametric tests against trend. Econometrica
.
Journal of the Econometric Society
13
,
245
259
.
https://doi.org/10.2307/1907187.
Massoud
E. C.
,
Purdy
A. J.
,
Miro
M. E.
&
Famiglietti
J. S.
2018
Projecting groundwater storage changes in California's central Valley
.
Scientific Reports
8
(
1
),
1
9
.
https://doi.org/10.1038/s41598-018-31210-1.
Matera
A.
,
Fontana
G.
,
Marletto
V.
,
Zinoni
F.
,
Botarelli
L.
&
Tomei
F.
2017
Use of a new agricultural drought index within a regional drought observatory
. In:
Methods and Tools for Drought Analysis and Management
.
Springer
,
Dordrecht
, pp.
103
124
.
McKee
T. B.
,
Doesken
N. J.
&
Leist
J
, .
1993
The relationship of drought frequency and duration time scales
.
Proceedings of the 8th Conference on Applied Climatology
17
(
22
):
179
183
,
Anaheim, California
.
Medellín-Azuara
J.
,
MacEwan
D.
,
Howitt
R. E.
,
Koruakos
G.
,
Dogrul
E. C.
,
Brush
C. F.
,
Kadir
T. N.
,
Harter
T.
,
Melton
F.
&
Lund
J. R.
2015
Hydro-economic analysis of groundwater pumping for irrigated agriculture in California's central Valley, USA
.
Hydrogeology Journal
23
(
6
),
1205
1216
.
https://doi.org/10.1007/s10040-015-1283-9.
Mendicino
G.
,
Senatore
A.
&
Versace
P.
2008
A Groundwater Resource Index (GRI) for drought monitoring and forecasting in a Mediterranean climate
.
Journal of Hydrology
357
(
3–4
),
282
302
.
https://doi.org/10.1016/j.jhydrol.2008.05.005.
Miller
R. E.
,
Green
J. H.
&
Davis
G. H.
1971
Geology of the compacting deposits in the Los Banos-Kettleman City subsidence area, California. U.S. Geological Survey Professional Paper 497-E, 45 p
.
Mishra
S. S.
&
Nagarajan
R.
2013
Hydrological drought assessment in tel river basin using standardized water level index (SWI) and GIS based interpolation techniques
.
International Journal of Conceptions on Mechanical and Civil Engineering
1
(
1
),
1
4
.
Mishra
A. K.
&
Singh
V. P.
2010
A review of drought concepts
.
Journal of Hydrology
391
(
1–2
),
202
216
.
https://doi.org/10.1016/j.jhydrol.2010.07.012.
Narasimhan
B.
&
Srinivasan
R.
2005
Development and evaluation of Soil Moisture Deficit Index (SMDI) and Evapotranspiration Deficit Index (ETDI) for agricultural drought monitoring
.
Agricultural and Forest Meteorology
133
(
1–4
),
69
88
.
https://doi.org/10.1016/j.agrformet.2005.07.012.
Owen
D.
,
Cantor
A.
,
Nylen
N. G.
,
Harter
T.
&
Kiparsky
M.
2019
California groundwater management, science-policy interfaces, and the legacies of artificial legal distinctions
.
Environmental Research Letters
14
(
4
),
045016
.
Panda
D. K.
,
Mishra
A.
&
Kumar
A.
2012
Quantification of trends in groundwater levels of Gujarat in western India
.
Hydrological Sciences Journal
57
(
7
),
1325
1336
.
https://doi.org/10.1080/02626667.2012.705845.
Pauloo
R. A.
,
Escriva-Bou
A.
,
Dahlke
H.
,
Fencl
A.
,
Guillon
H.
&
Fogg
G. E.
2020
Domestic well vulnerability to drought duration and unsustainable groundwater management in California's central Valley
.
Environmental Research Letters
15
(
4
),
044010
.
https://doi.org/10.1088/1748-9326/ab6f10.
Peters
E.
,
Torfs
P. J. J. F.
,
Van Lanen
H. A.
&
Bier
G.
2003
Propagation of drought through groundwater-a new approach using linear reservoir theory
.
Hydrological Processes
17
(
15
),
3023
3040
.
https://doi.org/10.1002/hyp.1274.
R Core Team
2017
R: A Language and Environment for Statistical Computing
.
R foundation for statistical computing
,
Vienna
,
Austria
.
Rodell
M.
,
Velicogna
I.
&
Famiglietti
J. S.
2009
Satellite-based estimates of groundwater depletion in India
.
Nature
460
(
7258
),
999
1002
.
Scanlon
B. R.
,
Faunt
C. C.
,
Longuevergne
L.
,
Reedy
R. C.
,
Alley
W. M.
,
McGuire
V. L.
&
McMahon
P. B.
2012
Groundwater depletion and sustainability of irrigation in the US High Plains and Central Valley
.
Proceedings of the National Academy of Sciences
109
(
24
),
9320
9325
.
Shahid
S.
&
Hazarika
M. K.
2010
Groundwater drought in the northwestern districts of Bangladesh
.
Water Resources Management
24
(
10
),
1989
2006
.
https://doi.org/10.1007/s11269-009-9534-y.
Shiau
J. T.
&
Lin
J. W.
2016
Clustering quantile regression-based drought trends in Taiwan
.
Water Resources Management
30
(
3
),
1053
1069
.
https://doi.org/10.1007/s11269-015-1210-9.
Tabari
H.
,
Abghari
H.
&
Hosseinzadeh Talaee
P.
2012
Temporal trends and spatial characteristics of drought and rainfall in arid and semiarid regions of Iran
.
Hydrological Processes
26
(
22
),
3351
3361
.
https://doi.org/10.1002/hyp.8460.
Tibshirani
R.
,
Walther
G.
&
Hastie
T.
2001
Estimating the number of clusters in a data set via the gap statistic
.
Journal of the Royal Statistical Society: Series B (Statistical Methodology)
63
(
2
),
411
423
.
https://doi.org/10.1111/1467-9868.00293.
Tirivarombo
S.
,
Osupile
D.
&
Eliasson
P.
2018
Drought monitoring and analysis: standardised precipitation evapotranspiration index (SPEI) and standardised precipitation index (SPI)
.
Physics and Chemistry of the Earth, Parts A/B/C
106
,
1
10
.
https://doi.org/10.1016/j.pce.2018.07.001.
Tiwari
V. M.
,
Wahr
J.
&
Swenson
S.
2009
Dwindling groundwater resources in northern India, from satellite gravity observations
.
Geophysical Research Letters
36
,
18
.
https://doi.org/10.1029/2009GL039401.
Wang
W.
,
Chen
Y.
,
Chen
Y.
,
Wang
W.
,
Zhang
T.
&
Qin
J.
2022
Groundwater dynamic influenced by intense anthropogenic activities in a dried-up river oasis of Central Asia
.
Hydrology Research
.
https://doi.org/10.2166/nh.2022.049.
Williamson
A. K.
,
Prudic
D. E.
&
Swain
L. A.
1989
Ground-water flow in the Central Valley, California, 1401
.
US Government Printing Office
,
Washington, DC
.
Yin
J.
,
Medellín-Azuara
J.
,
Escriva-Bou
A.
&
Liu
Z.
2021
Bayesian machine learning ensemble approach to quantify model uncertainty in predicting groundwater storage change
.
Science of The Total Environment
769
,
144715
.
https://doi.org/10.1016/j.scitotenv.2020.144715.
Yu
Y. S.
,
Zou
S.
&
Whittemore
D.
1993
Non-parametric trend analysis of water quality data of rivers in Kansas
.
Journal of Hydrology
150
(
1
),
61
80
.
https://doi.org/10.1016/0022-1694(93)90156-4.
Zhang
P.
,
Anderson
B.
,
Tan
B.
,
Huang
D.
&
Myneni
R.
2005
Potential monitoring of crop production using a satellite-based Climate-Variability Impact Index
.
Agricultural and Forest Meteorology
132
(
3–4
),
344
358
.
https://doi.org/10.1016/j.agrformet.2005.09.004.
This is an Open Access article distributed under the terms of the Creative Commons Attribution Licence (CC BY-NC-ND 4.0), which permits copying and redistribution for non-commercial purposes with no derivatives, provided the original work is properly cited (http://creativecommons.org/licenses/by-nc-nd/4.0/).