Groundwater depletion in California’s Central Valley accelerates during megadrought

Groundwater storage variations by integrating GRACE/FO-derived TWS with other terrestrial water storage components for the past two decades

GRACE/FO TWS anomalies for the combined Sacramento, San Joaquin and Tulare basins (Fig. 1, Fig. 2a) were used to calculate groundwater storage anomalies in California’s Central Valley. The GRACE/FO time series (Fig. 2a) for the combined basins is indicative of a region that has experienced successive droughts, punctuated by brief wet periods, resulting in significant cumulative water loss during the study period.

Before estimating groundwater storage changes, GRACE/FO TWS were first evaluated by comparing its monthly changes to those from an observed water balance calculation (see Eq.(1) in Methods). Figure 2b shows the observed water flux components including precipitation (P), evapotranspiration (ET) and streamflow discharge (Q) for the combined river basins, while Fig. 2c shows a close correspondence between dS/dt derived from GRACE/FO, and that computed using P–ET–Q in Eq.(1). The Root Mean Squared Difference between the two is 26.4 mm/month, and is within the range of the mean uncertainty using GRACE/FO measurements (43.6 mm/month). Such a good agreement between GRACE/FO-derived and observed dS/dt demonstrates that GRACE/FO is capable of accurately monitoring basin-wide water balance changes, and provides further confidence in the groundwater storage change estimates described below¹².

Groundwater storage anomalies were estimated by subtracting the anomalies of soil moisture, surface water, and SWE (Fig. 2d) from GRACE/FO TWS anomalies (Fig. 2a) following Eq.(2) as detailed in Methods. The SWE, soil moisture and surface water datasets were obtained from operational, publicly available sources, including the National Oceanic Atmosphere Administration’s Snow Data Assimilation System (SNODAS)⁴³, NASA’s North American Land Data Assimilation System (NLDAS)⁴⁴, and the California Data Exchange Center⁴⁵, respectively, ensuring data accessibility for potential routine monitoring following this approach.

Figure 3a shows the monthly groundwater storage anomalies derived from GRACE/FO and the datasets shown in Figs. 2a, d in the Central Valley between September 2003 and December 2021. Three notable periods of groundwater recharge and loss were identified in the past 18 years. Groundwater recharge occurred during wet periods from October 2003 to July 2006, March 2011 to July 2011, and October 2018 to August 2019, shown as blue arrows in Fig. 3a. Groundwater loss phases correspond to the well-known droughts that occurred during that time period, namely August 2006–February 2011, August 2011–March 2017, and since September 2019, shown as red arrows in Fig. 3a. A pattern of short phases of recharge followed by longer phases of groundwater loss emerges, resulting in longer-term groundwater depletion over the last two decades. Estimated rates and the total volumes of groundwater gains and losses are summarized in Table 1.

A groundwater recharge phase (22.7 ± 16.0 mm/yr; 3.49 ±  2.5 km³/yr) in the Central Valley was observed at the beginning of the GRACE mission during 2003–2006 (1st recharge in Fig. 3a and Table 1), when the precipitation amounts were close to or slightly higher than the 20-year average. The NOAA National Weather Service report⁴⁶ reveals that weak to moderate levels of El Niño events during 2004–2006 resulted in nearly normal amounts of precipitation and snow in the study region. A volume of 9.9 ± 4.2 km³of groundwater was replenished during this phase of the analysis.

This period of groundwater increase was followed by the 4.5-year drought that began in August 2006. During the 2006–2011 drought (1st drought in Fig. 3a and Table 1), a groundwater loss rate of 42.9 ± 7.8 mm/yr (6.59 ± 1.20 km³/yr) was estimated, resulting in 30.2 ± 2.6 km³ of groundwater loss during that period. Compared with the earlier analysis in Ref. 12, an additional year of data was included here, and represented the complete drought phase through 2011, rather than through 2010, as in Ref. 12. Although the groundwater loss rate is slightly higher than the 38.9 ± 9.5 mm/yr reported in Ref. 12, the difference falls within the 95% confidence interval, confirming the consistency between the two analyses.

Prior to the second drought, a short, rapid recharge phase (March–July 2011, 2nd recharge in Fig. 3a and Table 1) replenished 29.6 ± 15.7 km³ of groundwater (462.5 ± 157.8 mm/yr; 71.07 ± 24.25 km³/yr), as a result of the strong El Niño in 2010 that brought abundant precipitation in early 2011⁴⁷.

The groundwater loss rate for the second phase of drought in the GRACE/FO record (2011–2017, 2nd drought in Fig. 3a and Table 1) was 42.7 ± 5.8 mm/yr (6.56 ± 0.89 km³). Although a similar groundwater loss rate was estimated for the drought of 2006–2011, the second drought lasted a year longer, resulting in roughly 7 km³ more groundwater loss (37.1 ± 2.1 km³ total), equivalent to about 23% of surface water storage in the Central Valley, and greater than the volume of Lake Mead (32.2 km³) at full capacity. The GRACE/FO-based groundwater estimated in this study reached an 18-year low by late 2016. This phase of drought was notable for widespread water conservation efforts across California, and for the passage of SGMA in 2014. This second phase of drought ended with atmospheric river events that brought heavy precipitation to California in early 2017⁴⁸.

The original GRACE mission was decommissioned in late 2017 and transitioned to GRACE-FO after its launch in May 2018. Hence there is year-long data gap in the combined GRACE/FO record from August 2017–September. 2018. Studies of that time period^23,34 suggest that groundwater recharge occurred during this data gap. We estimate that during the lifetime of original GRACE mission (2003–2017), 41.8 ± 1.2 km³ of groundwater were lost (Table 1).

We assume that the groundwater depletion followed the 18-year historical trend (2003–2021), but made no assumption about its seasonal dynamics during the data gap between the GRACE and GRACE-FO missions. From October 2018 to August 2019 (3^rd recharge in Fig. 3a) we estimated that groundwater storage increased by 26.6 ± 16.0 km³ (188.8 ± 108.9 mm/yr; 29.02 ± 16.73 km³/yr).

The third phase of drought in the GRACE/FO record began in September 2019. After the recharge event in the winter of 2018, major water inputs in the region, including precipitation and SWE, significantly decreased in the winters of 2019 and 2020 (Figs. 2b and d). These two winters rank the years 2019 and 2020 as fourth driest consecutive 2-year period on record⁴⁹. In particular, precipitation reached an 18 year low in the winter of 2020–2021 (Fig. 2b), and TWS (Fig. 2a) shows this same time period as the driest wet season in the GRACE/FO record. Between September 2019 and December 2021 (Present drought in Fig. 3a), total groundwater losses in the Central Valley were 20.0 ± 5.1 km³ (55.8 ± 21.8 mm/yr; 8.58 ± 3.35 km³/yr), which is roughly 31% faster than the previous two droughts.

During the present megadrought in southwestern North America (2003–2021), groundwater anomalies observed from GRACE/FO in the Central Valley show a trend of groundwater depletion of 15.7 ± 1.4 mm/yr (2.41 ± 0.22 km³/yr), resulting in a total groundwater loss of 44.3 ± 0.9 km³, an amount that is nearly than 1.4 times the full capacity of Lake Mead.

Longer-term trends and comparison to observations

The GRACE/FO groundwater estimates were compared with water table depth anomalies observed from groundwater wells, as shown in Fig. 3b. A valley-wide water table depth was obtained by averaging measurements from available wells located within Central Valley, managed by California’s DWR and USGS²³ (see Methods). Seasonal variations of GRACE/FO derived groundwater storage changes and the observed water table depth were removed by subtracting their climatologies, i.e. deseasonalized groundwater storage and water table anomalies, to avoid seasonal inconsistencies between the two measurements, and to only examine their long term trends. Overall, the two measurements demonstrate similar trends from 2003 to 2021. While there is a greater difference between the well and GRACE/FO estimates following 2017, Fig. 3b shows that the groundwater estimates using GRACE/FO are capable of capturing the periods of loss and recovery observed on the ground, and in particular, the greater rate of groundwater loss since 2019, which appears even stronger in the well observations than in the GRACE/FO estimates. Discrepancies may be attributed to the irregular availability of groundwater well data, and to a major decline in available well observations since late 2018 (see Methods, Supporting Information, and Fig. S3). Both of these factors underscore the challenges of estimating large-area groundwater dynamics from well data alone, and of validating groundwater models and satellite observations.

Figure 4 shows cumulative groundwater losses from 1962–2021 using the CVHM¹³ and GRACE/FO. From 2003 to 2014 when both CVHM and GRACE data were available, the groundwater depletion rate for the CVHM was 16.3 ± 6.3 mm/yr (2.51 ± 0.97 km³), matching that from GRACE, 14.7 ± 6.0 mm/yr (2.25 ± 0.92 km³), indicating that the two methods are compatible and may be combined for the further analysis. The combined CVHM-GRACE/FO groundwater depletion rate was calculated by using both CVHM estimations from 1962–2014 and GRACE-derived groundwater storage changes from 2003–2021 through linear regression analysis. The result shows that the groundwater depletion rate from 1962 to 2021 was 12.1 ± 0.8 mm/yr (1.86 ± 0.12 km³/yr), shown as the black line in Fig. 4, resulting in a total groundwater loss of 111.5 ± 0.9 km³. In addition, Fig. 4 shows that the periods for groundwater recovery were shorter, and mostly driven by extreme weather events^46,47,48,50 in the nearly two decades of the GRACE/FO record. Although groundwater was recharged, these extreme wet events typically generated flooding, and had significant negative social, environmental and economic consequences^46,47,48,50. This sequence of extreme hydrological events—long-term extremely dry conditions with considerable groundwater losses, punctuated by short-term extremely wet conditions with short bursts of groundwater recharge—underscores the challenge of sustainable groundwater management under changing climate.

Figures 3a and 4, along with Table 1, show that the rate of groundwater loss is accelerating in the Central Valley. Groundwater loss rates observed from GRACE/FO (15.7 ± 1.4 mm/yr; 2.41 ± 0.22 km³/yr) between 2003 and 2021 are 28% faster than the longer-term (1962–2021) depletion rate of the combined CVHM-GRACE/FO record (12.1 ± 0.8 mm/yr; 1.86 ± 0.12 km³/yr). The most recent phase of groundwater loss, between September 2019 and August 2021 (55.8 ± 21.8 mm/yr; 8.58 ± 3.35 km³/yr), is nearly 31% faster than GRACE/FO estimated losses the previous two drought phases during the GRACE/FO record, and nearly five times faster than the long-term depletion rate.

Relationship between surface water allocations and estimated groundwater storage changes

Figure 5a compares GRACE/FO estimated monthly groundwater storage variations to annual surface water allocations (in % of annual maximum) via the two primary aqueducts in the Central Valley, the California State Water Project (SWP)⁵¹ and the federal Central Valley Water Project (CVP)⁵². The two aqueducts transport surface water from northern California to the south. Figure 5b compares the annual groundwater storage changes (net fluxes) to the total surface water deliveries from both the CVP and SWP (in km³). The annual groundwater change was calculated as the difference of the mean annual groundwater anomalies between two consecutive years. Figure 5a, b show that when surface water is abundant, greater allocations are made to farmers, relieving stress on groundwater and allowing for recovery, and vice versa.

Between 2003 and 2007, surface water storage was increasing (Fig. 2d), allowing for larger allocations (>60%) from both aqueducts, less reliance on groundwater, and hence increasing groundwater storage. Surface water deliveries in Central Valley reached a high for the study period in 2016, resulting in about 5 km³ recharge (Fig. 5b). Surface water storage, and hence allocations, decreased between 2007 and 2009, resulting in significant groundwater storage decline. Surface water deliveries decreased to 2.30 km³ in 2009, corresponding to the highest annual groundwater storage loss by 7.86 km³ during the 1st drought period.

The second drought in the GRACE/FO record began in August 2011, triggering decreasing surface water allocations that resulted in heavy groundwater demand. During this period, CVP cut its allocation to 0% in 2014 and 2015, and 5% in 2016, while the SWP reached its lowest allocation for the study period, 5% in 2014. The low surface water delivery volumes in 2014 and 2015 drove corresponding annual groundwater losses of 9.66 and 7.64 km³, respectively, and led to intensified groundwater pumping through 2016 (Fig. 5b).

Groundwater storage variations continued to reflect surface water allocations, increasing in 2017 and 2019 with above-average surface water storage, followed by major losses in both surface water allocations, and groundwater storage, through the end of 2021. For example, in 2020, aqueduct allocations decreased to 20% for both projects, and to 0% and 5% in 2021 for the CVP and SWP, explaining in part the increased rate of groundwater loss during this time period. In 2021, the annual groundwater loss was 9.22 km³, matching the greatest annual loss during the study period, which occurred in 2014.

Demonstration of GRACE/FO-derived groundwater storage changes to support regional groundwater management

GRACE/FO-derived groundwater storage changes were also estimated in the Sacramento, San Joaquin, and Tulare basins, as shown in Fig. 6 and Table 2. The same periods of groundwater recharge and loss in the Central Valley are used to calculate the gains and losses for the three basins, including longer-term depletion rates. Overall, the individual basin follows similar trends, i.e. three short recharge phases, followed by three longer droughts, as was presented for the entire Central Valley. During the 1^st recharge phase, similar rates of groundwater recharge can be observed in the Sacramento and Tulare basins, with increasing rates of 39.0 ± 20.0 and 27.5 ± 15.8 mm/yr (2.81 ± 1.44 and 1.17 ± 0.67 km³/yr (Fig. 6a, c and Table 2)), resulting in groundwater increases of 8.0 ± 2.4 km³ and 3.3 ± 1.1 km³ in the two basins, respectively. Although a slight groundwater loss of 0.7 ± 2.0 km³ (6.4 ± 29.6 mm/yr; 0.26 ± 1.21 km³/yr) in the San Joaquin basin is observed for this period (Fig. 6b and Table 2), the loss rate is not statistically significant (within an uncertainty of 95% confidence interval), indicating that groundwater supply and consumption were nearly balanced in the basin.

When entering to the 1st drought phase, results show that the Sacramento, San Joaquin, and Tulare basins all experienced similar groundwater loss rates of ~42 mm/yr (40–44 mm/yr) (Fig.6a–c and Table 2). The drought ended with the strong El Niño in 2010⁴⁷.

During the 2^nd drought, all three basins experienced significant losing trends. Figure 6a–c, and Table 2 show that the Tulare basin suffered more severe groundwater losses than the other basins, with a loss rate of 62.9 ± 4.4 mm/yr (−2.67 ± 0.19 km³/yr). The total groundwater loss in the Tulare basin was 15.1 ± 0.4 km³, which was nearly 40% of the total loss in Central Valley, yet the area of the Tulare basin only occupies about one quarter of the study region. The groundwater storage changes during the 18 year study period show that the depletion rates in the Sacramento, San Joaquin, and Tulare basins, were 12.9 ± 1.8, 16.2 ± 1.9, and 20.6 ± 1.5 mm/yr (0.93 ± 0.13, 0.67 ± 0.08, and 0.88 ± 0.06 km³/yr) (Fig. 6a–c and Table 2), respectively, indicating that the southern Central Valley (combined San Joaquin and Tulare) lost more groundwater than the north, similar to the findings of earlier studies^23,30. However, the situation was reversed in the drought that began in September 2019 (present drought in Fig. 6a–c), during which we found higher groundwater loss rates of 76.1 ± 28.1 mm/yr (5.48 ± 2.02 km³/yr) in the Sacramento basin compared to those of 38.1 ± 25.2 and 60.1 ± 14.0 mm/yr (1.56 ± 1.03 and 2.55 ± 0.60 km³/yr) for the San Joaquin and Tulare basins, respectively.

The deseasonalized GRACE/FO-derived groundwater storage and observed water table anomalies are compared for each of the three basins. Similar to the approach for the whole Central Valley, wells with available measurements within a particular basin boundary were averaged to represent the water table depth variation for the basin (see Methods and Supplementary Information). The two measurements show similar trends and variations for the Sacramento and Tulare basins, except for a strong water table rise in the winter of 2019 for the Tulare basin. As discussed earlier for the entire Central Valley, a dramatic decrease in the number of available well observations after late 2018 may have resulted in an inconsistent record of water table depth.

While the Sacramento and Tulare basins showed generally good agreement between GRACE/FO-derived groundwater storage changes and observed well measurements, less correspondence was observed in the San Joaquin basin, particularly during the 1st drought period. However, the two drought phases from 2011–2017 and after 2019 are clearly recognizable, with water table observations falling in response to increased groundwater pumping.

Figure 6 highlights both strengths and weaknesses of using the GRACE/FO approach at the sub-basin scale of the individual Sacramento, San Joaquin, and Tulare basins. On the one hand, sub-basin analyses provide important insights into groundwater storage variations across the Valley, in particular, sub-basin trends, which could ultimately inform SGMA performance and provide early warning (in the case of the Sacramento basin) for those regions where groundwater losses are unexpected. On the other, the sub-basins are considerably smaller than the ~154,000 km² area of the Central Valley, which corresponds the lower area limit for an acceptable level of error for monthly TWSA detection^36,53,54,55. (Note that the longer time period associated with the trend calculations mitigates this issue somewhat, resulting in greater confidence in the sub-basin trends than the monthly variations). Hence the GRACE/FO-derived groundwater storage variations at these sub-basin scales should be used judiciously.

As with the whole-valley comparisons to observations, the sub-basin analyses are faced with the same challenges as described above, i.e. the difficulties in assembling larger-area water table depth averages from unevenly distributed well observations collected at disparate times and for varying periods of time. In spite of these challenges, the regional groundwater analyses for the sub-basins demonstrates the potential utility of GRACE/FO-derived groundwater storage changes for supporting regional groundwater management efforts.

Source: Resources - nature.com