Drivers of domestic wells vulnerability during droughts in California’s Central Valley

Over the past decade, California has experienced two multiyear droughts, resulting in water insecurity for communities and significant economic losses for the agricultural sector. Despite the recognition of water as a human right in the state since 2012, droughts consistently lead to the failure of thousands of domestic wells due to intensified groundwater pumping for irrigation purposes. In the Central Valley alone, groundwater sustains the livelihoods of thousands of individuals (and millions across the state) serving as their sole water source, rendering them vulnerable due to inadequate groundwater management. In this study, we present a spatial statistical model to identify critical localized factors within the food-water-human system that contribute to the vulnerability of domestic wells during droughts. Our results indicate that the depth of domestic wells, density of domestic and agricultural wells, socioeconomic conditions, and the extent of perennial crops play significant roles in predicting well failures during droughts. We show the implications of addressing these factors within the context of ongoing groundwater sustainability initiatives, and we propose strategies to safeguard the water source for thousands of individuals necessary to protect domestic wells.


Introduction
Groundwater in California is an important source of water for human use and agriculture.However, community water supply systems, small water supply systems, and households with private domestic wells are vulnerable to water shortages and water quality issues due to extended agricultural groundwater pumping (Smith et al 2018, Pauloo et al 2020, Levy et al 2021).In the state, approximately 1.3 million people are served by domestic wells, mostly concentrated in the San Joaquin Valley with over 450 thousand people (Pace et al 2022).In 2012, California established a landmark law (Assembly Bill 685) recognizing the human right to water, ensuring that all individuals, including low-income and minority communities, are entitled to access clean, safe, and affordable water to drink, cook, and bathe.The recent occurrence of two consecutive drought periods (2012-2016 and 2020-2022) has exacerbated the depletion of groundwater levels, primarily due to intensified agricultural pumping practices, particularly in the San Joaquin Valley (Medellín-Azuara et al 2016, 2022).As a consequence, thousands of domestic wells were reported dry, especially from rural disadvantaged communities, who have been severely impacted by unsustainable groundwater use (Feinstein et al 2017, Klasic et al 2022).The increasing risk of water shortages from groundwater sources for human consumption is not confined to California and the western United States, as evidenced by research (Perrone et al 2017, Scanlon et al 2021).Similar challenges are prevalent in other semiarid regions around the world, characterized by the depletion of groundwater resources .Consequently, there is a growing need for sustainable groundwater management practices (Elshall et al 2020).
The vulnerability of California to drought stems from its Mediterranean climate, and high dependence on snowpack and atmospheric rivers for water supply (Diffenbaugh et  During the 2012 to 2016 drought, the California Department of Water Resources (DWR) established the Dry Well Reporting System (DWR 2023).Such platform allows for voluntary reporting of domestic well failures and aims to provide water shortages mitigation assistance.Since its establishment in 2014 until the end of 2022, a total of 5259 domestic wells have been reported as dry or failed in the state.The majority of these reported failures occurred during the past two drought periods, with 2425 wells experiencing failure from 2014 to 2016 and 2588 wells from 2020 to 2022.The greatest concentration of reported dry wells was observed in the Central Valley, where a total of 3929 domestic wells were affected between 2014 and 2022.Significantly, the subbasins in the San Joaquin Valley (southern region of the Central Valley), namely Kaweah, Kings, Madera, and Tule, accounted for 2878 of these reported dry wells (figure 1).Given the substantial magnitude of domestic well failure and groundwater depletion (figure S1), as well as the region's significant agricultural importance, these four subbasins serve as a representative case study area within the broader context of the Central Valley.Insights derived from studying this region can offer valuable perspectives for addressing domestic well vulnerability.Figure 1(B) shows the number of agricultural and domestic wells by section (1mi 2 ) of the Public Land Survey System (PLSS) and figure 1(C) the count by section of reported dry wells.
Since the California Sustainable Groundwater Management Act (SGMA) was passed in 2014, Groundwater Sustainability Agencies (GSAs) have been established to develop and implement groundwater sustainability plans (GSPs).These plans aim to conceptualize water budgets, groundwater dynamics, and frame strategies for achieving sustainable groundwater management in each groundwater subbasin.However, since their submission in 2020 there have been inadequate engagement and representation of marginalized communities (Leach et al 2021, Dobbin et al 2023, Perrone et al 2023) and assessment of domestic well vulnerabilities within these plans.Thus, is crucial to gain a comprehensive understanding of the factors that contribute to domestic well failure, as this knowledge is vital for sustainable groundwater management and the protection of domestic wells.Notably, many of these factors can be addressed at the local level, particularly those related to agricultural practices and their associated impacts.For instance, agricultural wells, which are typically deeper and have higher yields than domestic wells (Perrone and Jasechko 2019), contribute to localized drawdown and groundwater quality degradation (Perrone et al 2017, Pauloo et al 2020, Levy et al 2021, Gailey 2023).Previous studies have focused on forecasting the occurrence of dry wells in the Tule subbasin (Gailey et al 2019) and at a large scale in the Central Valley (Pauloo et al 2020) or other basins in western USA (Perrone et al 2017).These studies utilized interpolated groundwater levels and drilled depths of domestic wells to identify hotspots of dry wells or vulnerable wells in anticipation of future drought events.While these studies demonstrated high accuracy in predicting dry wells following groundwater declines, they did not comprehensively characterize key localized factors that contribute to the vulnerability of domestic wells.
In this study, we implement a spatial statistical model that utilizes publicly available spatial data sets, including reported dry wells, well completion records, groundwater levels and land use information.Our objective is to identify specific factors that contribute to the vulnerability of domestic wells during droughts.The used spatial model incorporates a Bayesian approach to account for spatial autocorrelation among observations.By employing this approach, we are able to predict the probability of failure of a domestic well within the four subbasins under investigation.Our model results provides valuable insights into the most significant factors within the food-water-human system which merit higher consideration for the development of effective groundwater management strategies.The findings from this study have broader applicability, extending beyond our study area to other groundwater basins in California, Western US and other regions in the world where domestic wells are situated in semi-arid regions where agriculture heavily relies on groundwater resources.Additionally, with the exacerbating global groundwater depletion and impacts to domestic users there is a need to develop methods, such as the presented data driven study, that can be replicated in other places to inform groundwater sustainability policy.

Data and exploratory analysis
The spatial data set used in this study uses records on dry well reports, well completion reports, interpolated groundwater levels from monitoring wells, crop land use, surface water supply and poverty index.These data sets have different spatial resolutions and processing, explained in detail in the following sections.Each well in the data set is located in a section of 1 mi 2 (2.6 km 2 ) from the PLSS.When specific coordinates of the reported dry well or completed well are unknown, the centroid of the section is registered as coordinates of the well by the DWR reporting system.For this reason, we used the PLSS section as reference location.We utilized two spatial resolutions to summarize the data sets for each well in the analysis (they are described in the following sections).First, the section where the well is located and adjacent sections referred as 9 mi 2 (23.3 km 2 ), depicted in yellow in figure 2 and a second resolution using a 25 mi 2 polygon (64.7 km 2 ), depicted in green in figure 2, that includes second-level section neighbors.

Dry wells and well completion reports
Domestic dry wells are voluntarily reported to the publicly available reporting system from DWR (2023), which since 2014 collects domestic groundwater supply shortages.These reports are collected from individuals, local agencies, and organizations.Given the volunteer nature of the reports there are uncertainties on the actual number, location, and well depths reported.A potentially omission bias from the non-reported dry wells is inherent to the data set and the analysis.However, is the only publicly available data set that collects systematically water shortages from domestic wells.To overcome some of the limitations of this data set, we used the upper boundary of the reported well depth if its a range and removed dry wells that did not report a well depth or is unclear.Figure 3 shows the count of reported dry wells in the study area before removing incomplete reports.
Our data set contains reported dry wells and randomly sampled domestic wells from PLSS sections where no dry wells were reported in a way that we build a balanced data set with equal number of reported dry and not reported dry wells by year and groundwater basin.The objective of this approach is to have a balanced data set that achieves the best prediction performance of domestic well failure.We calculate the number of agricultural and domestic wells by section using the well completion reporting system from DWR (2023).For each well and year of analysis we calculated number of wells at the two spatial resolutions (figure 2).We assumed a lifespan of 30 years for domestic wells, as suggested by Pauloo et al (2020) and Gailey et al (2019), and 50 years for agricultural wells, from conversations with well drilling technicians.Given the geologic characteristics of California's Central Valley, wells can extract water from the unconfined, semiconfined, or confined aquifer.Given that most reported dry domestic wells (figure S2) and domestic wells from the well completion reporting system within our study area are situated within the unconfined to semiconfined aquifer, our study specifically focused on agricultural and domestic wells located within these two aquifers (refer to figures S3 and S4).It is essential to note that  wells within the same aquifer have the potential to mutually influence one another, forming the basis for our selection criteria.Additionally, we assumed that agricultural wells and domestic wells from the well completion reporting system may have been inactive for each year of analysis (2014-2022), using interpolated groundwater levels for the unconfined and semiconfined aquifers (section 2.1.2).For more details about this process refer to supplementary material (SM).

Groundwater levels
Groundwater levels are reported in the DWR's periodic groundwater level measurements from monitoring wells.We followed the same classification process to the corresponding aquifer location to use only those from the unconfined to semi-confined aquifers.We validate our classification for the monitoring wells reported in the GSP's groundwater monitoring network of each GSA, available in the SGMA portal.In this study we want to capture the conditions before each year's irrigation season starts, thus we used reported levels from January to April for each year.Finally, we employed an ordinary kriging interpolation method following Pauloo et al (2020), using the log-transformed groundwater levels, and the correction of Laurent (1963) to obtain unbiased groundwater level estimates.The spatial interpolation via kriging was performed using the gstat R package (Gräler et al 2016).The final groundwater level raster was used in two ways, first to filter wells with top screens (or approximated top screen) shallower than the interpolated groundwater level, and second to calculate the ratio of the total drilled depth to the interpolated groundwater level for each well in the analysis, used as covariate in the statistical model.There are temporal and spatial variations in the monitoring well data, in addition to spatially sparse reports, thus this interpolation is a representation of the regionwide groundwater levels but not localized drawdown.We used the groundwater level before the irrigation season (January -April).
2 Surface water deliveries were used at the GSA scale, thus these values represent statistics across GSAs that can be inconsequential.

Cropland
Cropland was obtained from the cropland data layer (CDL), referred as CropScape, produced by USDA National Agricultural Statistics Service (NASS) (Boryan et al 2011).We downloaded the spatial layers from 2014 to 2022 using the CropScapeR package for R (Chen et al 2023).CropScape classifies each pixel, with 30 m resolution, into 134 different crops and other uses.We categorized crops into three categories: annual crops, perennial crops and forage crops.Table S2 in SM relates CDL crops to each category.Finally, using the cropland rasters (figure S5) we calculated zonal statistics for each PLSS, generated using the R package exactextractr (Baston 2022).

Surface water supply
Surface water supplies in the San Joaquin Valley vary by water agency and irrigation district.Sources may have contracts with state (State Water Project) or federal (Central Valley Project) projects, water rights to divert water from rivers or streams, as well as transfers with other districts.Due to the lack of a spatial data set that relates each section of the PLSS to a surface water source, we used the historical allocations to each GSA.GSAs reported historical surface water allocation in their GSPs and in their yearly updates, both available in the SGMA portal.

Sociodemographic
We are interested to know if the socio-economic status affects the risk of domestic wells to cause residential water shortages.We hypothesize that lower income communities are often times isolated rural communities surrounded by agriculture, thus are likely more vulnerable to agricultural pumping.To characterize socioeconomic status, we employed the poverty index from CalEnviroScreen 4.0 (OEHHA 2021).This index represents the percent of population living below twice the federal poverty level at the census tract.

Data sets limitations
Our study possesses limitations associated with the nature of the data sets employed, and we acknowledge the inherent uncertainties and potential errors that may arise from their use.Many of these pertain to publicly available data on groundwater levels, completed wells, and dry wells, which have been thoroughly discussed by prior research (Gailey et al 2019, Jasechko et al 2020, Pauloo et al 2020).We outline the most significant limitations of the study.(1) DWR's voluntary dry wells reports and well completion reports may under represent the real number of dry wells and number of wells that exist.The location of the wells is uncertain and often times the centroid of the nearest PLSS section is used.Additionally, the characteristics of the wells (e.g.well depth and screen location) are approximations and may lack of accuracy.
(2) There are temporal and spatial variations in the groundwater levels monitoring data, which can lead to inconsistencies and inaccuracies in reported groundwater levels.Thus, these uncertainties cascade into our groundwater level interpolations, in addition to potential errors inherent to the selected geostatistical interpolation method.
(3) Land use information obtained from USDA's CropScape, relies on remote sensing which have classification errors as reported by Espinoza et al (2023).However, in our study we are only interested in proportions by large crops categories which may reduce the error.

Exploratory data analysis
The curated data set contains 2866 wells distributed in the four groundwater basins of the study area.
The predictor variables or covariates of the model are summarized in table 1, that also includes well depth and groundwater level used to calculate the ratio between well depth and groundwater level.These variables are selected to represent the localized characteristics of the system that can affect the domestic well vulnerability during dry years.We conducted data exploration to identify relationships between covariates.We estimated Pearson correlation coefficients, shown in figure 4. We measured the variance inflation factors (VIF) which resulted in values lower than 1.5 suggesting no collinearity among covariates.Given that wells may be located in the same section or adjacent sections, spatial autocorrelation was detected for the covariates using the Moran's I test (figure S6).Finally, before running the spatial model (section 2.2), we run linear models to identify associations between domestic well failure and each covariate.

Spatial model
To model the probability of domestic well failure we used the R-Integrated Nested Laplace Approximation (INLA) package that relies on the INLA using a spatially correlated random effect (Blangiardo andCameletti 2015, Lindgren andRue 2015 Fichera et al 2023).We used a constrained refined Delaunay triangulation to define the spatial random effect, where triangles define the basis of SPDE spacial process (Zuur et al 2017).To tune the triangulation, we use the distances between wells (figure S8), employing a distance of 16 km for the maximum edge length of the finer inner mesh and 80 km for the edge of triangles outside of it to avoid boundary effects.A cutoff value was defined to constrain the assignment of wells with a distance of less than 3 km to the a single vertex.The final triangulation is shown in figure S9.
We predict domestic well failure (n = 2866 observations) using the model defined by equation ( 1).Where W i is the binary condition failure (W = 1) or no failure (W = 0) of a domestic well i.Equation ( 2) is the logit-link function for the Bernoulli probability function where p i is the expected value of the probability of a well i to fail, α is the intercept, X j is the j covariate of well i and β j is the fixed effect coefficient.β c are the interaction effects between crop categories (c = {r_area_annual,r_area_perennial,r_area_forage}) and agricultural wells density.We include a groundwater subbasin (Basin g,i ) random intercept effects (iid) (β g ) to identify if well failure vulnerability changes across groundwater basins.S i accounts for the SPDE spatial random effects.Additionally, we used 20% of the wells with equal number of positive or negative failure for validation, sampled using block cross-validation (Valavi et al 2019).Variables were transformed using z-score standardization to facilitate the interpretation of the results.Finally, we use the R-INLA default uninformative priors.
We run the model described by equations ( 1) and (2) comparing different model configurations and two spatial resolutions (9mi 2 and 25mi 2 ), defined in SM 6.1.Model configurations were used to assess how information affect the prediction of domestic well failure.We use different cross-validation and information criteria (Gelman et al 2014) to compare the prediction capacity and performance among model configurations.These metrics include Watanabe-Akaike information criterion (WAIC) (Watanabe 2010) and the Area Under the receiver operating characteristic Curve (AUC) (Fawcett 2006).Lower WAIC values suggest superior model performance and higher AUC suggest superior classification performance.Readers can refer to the following GitHub repository to access model script used for modeling, model diagnostics and visualization of results: https://github.com/josemrodriguezf/Domestic_Dry_Wells.

Results
Figure 5 shows the results of the linear fits, where the orange line is the mean coefficient and in grey the 95% credible interval.Figure 5(F) depicts the association between ratio between well depth and groundwater level showing a strong negative association with well failure.Additionally, we would expect a higher probability of domestic well failure in places with high domestic well density (figure 5(D)) and agricultural wells density (figure 5(E)).Cropland distribution (perennial, annual, and forage) shows similar positive association with domestic failure (figures 5(A)-(C)).Finally, in places with higher poverty index a higher probability of domestic well failure is expected (figure 5(G)).Since there is not data available on groundwater pumping at a fine resolution we asses if cropland ratio has an interaction effect with the agricultural wells density that can potentially affect domestic well failure (figures S7).
Our spatial model effectively predicted the probability of a well going dry.The model described by equations ( 1) and (2), using the 9 mi 2 resolution, resulted in the lowest information criterion metrics (table S3) and an AUC of 0.88 for the fitting data set (figure S10) and 0.92 for the validation data set (figure S11).This model also demonstrates good spatial performance on domestic well failure prediction across space for the validation data set (figure S12) predicting correctly 85% of the wells.Additionally, the model shows no spatial autocorrelation of the residuals (figure S13).
The most important single effect predictor (figure 6(A)) is the ratio between well depth and groundwater level that has a negative effect (mean of −2.76 log-odds).This result indicates the impact that an increase in the ratio between well depth and groundwater level by one standard deviation (1.5) has on domestic well failure, while the other variables are at their reference levels.However, other factors attribute significantly to domestic well failure, such as domestic wells density that has the largest positive coefficient (0.75 mean), followed by agricultural wells density (0.50 mean) and poverty index (0.24 mean).
Surface water deliveries distribution overlaps with zero, the rationale behind this value is that we focused on dry years and there are no meaningful differences in surface water supplies across these drier years.Each coefficient from covariates without interaction effects (surface_water, poverty, r_well_depth_gw_level, and dom_wells_density) are interpreted as the change in log-odds corresponding to a one standard deviation change in the covariate.
An increase in perennial crops increases the effect that agricultural wells can have on domestic well failure, shown by the interaction between perennial crops ratio and agricultural wells density (0.26 mean).This coefficient can also be interpreted in the other direction, given an increase in agricultural wells density the expected impact of perennial crops to domestic well failure also increases.Although the coefficients for forage crops, annual crops and their interaction effects with agricultural wells density are not far from zero, this does not imply that pumping to irrigate forage and annuals crops do not affect domestic wells, but rather these coefficients are less important predictors of domestic well failure considering all the other covariates in the model.
The model includes groundwater subbasin random effects which allow us to compare expected domestic well failure given the variability that exists across basins (e.g.geology, groundwater depletion, groundwater management, and land use).Though the 75% credible intervals of these coefficients overlap with zero, we can observe that there is a higher expectation of domestic well failure in Tule and Kaweah subbasins than Madera and Kings subbasins.Finally, it is important to note that the distribution of the intercept is very wide with a 75% credible interval overlapping zero.This coefficient indicates the log-odds of domestic well failure when all the other covariates are at their reference level (mean).Given that there is large uncertainty for many of the covariates, such as the interpolated groundwater levels and derived uncertainties from the spatial resolution (9mi 2 ), the intercept shows a wide posterior distribution, including larger values compared to the other factors.

Discussion
As groundwater planning aims to address the impacts of groundwater depletion and achieve long-term sustainability, it becomes crucial to identify priority areas for effective groundwater sustainability implementation.Our model offers predictive capabilities for domestic well failure, serving as a valuable tool to inform land and water management strategies focused on their protection.While the results reveal significant factors influencing domestic well failure prediction, it is essential to acknowledge that a comprehensive understanding of groundwater flow dynamics is necessary for accurate interpretation.0).This analysis ignores spatial random effects and fits were generated using the model: W i ∼ Bernoulli(p i ) where logit(p i ) = β j X j and W i = 1 if domestic well failure was reported and Additionally, it is important to note that during droughts, dry wells occur in hotspots with a higher concentration of domestic wells.This does not imply a causal relationship between domestic wells density and domestic well failure, but rather the nature of these hotspots.It is also important to acknowledge the heterogeneity of the communities in the region, and that there is not a silver bullet to solve water insecurity problems in the San Joaquin Valley.Each location must be understood individually as much as holistically.In the following sections, we explore in-depth the key insights derived from the results and their implications within the social, agricultural, and policy contexts.In SM sections 8 and 9 we explain more in detail characteristics of agricultural wells, domestic wells, reported dry wells and groundwater level trends in the study area.

Domestic wells, agriculture, and droughts
Though there is unavailability of fine scale groundwater level monitoring and complete domestic well coverage in the groundwater level monitoring network that can inform about localized draw-down conditions, the ratio between well depth and groundwater level (at the regional scale) emerges as the most crucial factor.For example, the median of this ratio for reported dry wells is 1.1 (well depth 10% deeper than the groundwater level before the irrigation season starts) and a 75th percentile of 1.4, whereas the mean of this ratio for wells that did not fail (of wells used in the model) is 1.8 and a 25th percentile of 1.4.Is important to highlight that groundwater level decline from agricultural pumping is a basin wide phenomena, particularly in the San Joaquin Valley where the aggregate of industrial agricultural pumping has a significant impact on the regional groundwater levels.
In dry years, a rise in agricultural pumping, resulting from both longer operational hours for existing wells and the activation of previously idle or newly installed wells, can amplify drawdown effects.This intensified drawdown has the potential to lead to domestic well failure even before the basin's groundwater table descends to depths beyond the pump intake (Gailey et al 2022).Thus, shallower domestic wells are more vulnerable than deeper agricultural wells, which is a common characteristic of the Central GSAs have defined minimum groundwater level thresholds to avoid SGMA undesirable results (e.g.chronic lowering of groundwater levels).This threshold facilitates operational flexibility, allowing groundwater levels to be deeper than the sustainability objective ('measurable objective') during drought periods while recovering the groundwater level during wet years, as shown by Rodríguez-Flores et al (2023).Additionally, this threshold provides certainty to well owners that their wells will be functional even after multi-year dry periods.However, as discussed by Bostic et al (2023) and Perrone et al (2023), minimum thresholds did not consider domestic wells depths in many of the GSPs, leaving domestic wells unprotected.Our result indicates that the most important factor to protect domestic wells is to maintain the groundwater level to a safe minimum threshold that can protect the domestic wells for localized drawdown.
Our analyses suggest that domestic wells in areas with more perennial tree crops are more vulnerable.The study area has shown an expansion of perennial crops, mainly nut trees such as almonds and pistachios.Though these crops are highly profitable, they have a high water demand, for example almonds and pistachios use in average 12 000 m 3 /ha per year (DWR 2020), and require irrigation every year despite surface water shortages, relying on groundwater pumping.These factors serve as primary drivers of ongoing groundwater depletion, amplifying the complexity and reducing the adaptive capacity of groundwater management policies (Mall and Herman 2019, Qin et al 2019, Rodríguez-Flores et al 2023).

Social impacts
Our results indicate that domestic wells in places with higher poverty index are more vulnerable.As reviewed by Johnson and Belitz (2015) and London et al (2021), rural areas or outside of incorporated city boundaries are less likely to be served by municipal or other larger public water supply systems.Consequently, rural areas heavily depend on private domestic wells for their water supply, which significantly increases their risks of water shortages and compromise their water quality (Balazs and  In the study area, there is a higher concentration of domestic wells in census tracts with a poverty index between 30th and 60th percentile.Areas with a poverty index between 0 and 30 percentile have the second most number of domestic wells, followed by areas with high levels of poverty (between 60th and 90th percentile) (figure S26).However, this trend changes for the reported dry wells.After areas with moderate poverty (30th to 60th percentile), high poverty areas reported the second most number of domestic dry wells (figure S27).Though well depths are similar for all areas despite of the poverty level (figure S28), domestic well vulnerability of higher poverty areas arises from being more rural areas impacted by agriculture.These low-income communities are inhabited mostly by people of color (London et al 2021, Méndez-Barrientos et al 2022, Pace et al 2022), who have limited infrastructure and are exposed to other environmental risks (Anderson et al 2018, Fernandez-Bou et al 2021, Flores-Landeros et al 2022).Though the poverty index is reported at the census tract level, which may misrepresent conditions of small rural communities designated to larger census tracts, the results are consistent with prior studies (Johnson and Belitz 2015, Perrone et al 2017, Klasic et al 2022).
Within the dry well reporting system, numerous records reported interim solutions to address the water shortage or that the well failure was resolved.Common adaptations include lowering pumps and replacing dry wells with deeper wells (Gailey 2023).However, drilling deeper wells entail higher upfront costs (Feinstein et al 2017, Gailey et al 2019, Perrone and Jasechko 2019), larger operating cost (given the increased lift and energy use), and in some cases, deeper groundwater may contain elevated levels of salts or other contaminants making it unsuitable for domestic use (Kang and Jackson 2016).For instance, the costs related to drilling domestic wells, typically fall within the range between $118/m and $200/m (Gailey et al 2022, SWRCB 2022).This poses a burden for low-income communities to be self-sufficient considering that wells in the study area would need to be drilled deeper than 80 m and in some places more than 100 m (in the Tule subbasin).Thus, communities are compelled to depend on state's support or alternative resources such as tanker trucks and bottled water (London et al 2021, Méndez-Barrientos et al 2022).

Implications for water and land management
Though SGMA is in early implementation stages, the impacts of groundwater pumping during last drought showed again the lack of management that in the future will need to be implemented by substantial pumping constraints (Escriva-Bou et al 2020) and cropland retirement (Escriva-Bou et al 2023), regardless of the upcoming dry periods.Our findings suggest that effective groundwater management should prioritize addressing localized factors with the greatest impact on the risk of well failure.These factors can be effectively mitigated through various land and water management strategies, including strategic cropland repurposing (Bourque et al 2019, Bryant et al 2020, Fernandez-Bou et al 2023)  Based on our findings, higher density of wells increases the vulnerability of domestic wells.Thus, GSAs should exercise control over the approval of new agricultural well drilling, taking into account the density of agricultural wells and agricultural pumping in the vecinity of domestic wells.Furthermore, comprehensive monitoring of pumping activities and groundwater levels should be conducted in regions characterized by a significant density of domestic wells.The state passed a bill in 2022 (Assembly Bill 2201), which mandates that the approval of new agricultural well permits be subject to the oversight of GSAs.
Finally, to successfully implement land and water management strategies, it is essential to take into account their overall cost-effectiveness.In general, the costs associated with crop production losses from pumping constraints exceed significantly the costs associated with dry well mitigation costs (Gailey et al 2022).This underscores the significance of investigating alternative avenues to internalize externalities from agricultural pumping, such as cost-sharing mechanisms (Stone et al 2022), or economic instruments like pumping fees designed to cover domestic dry wells mitigation costs.Additionally, maintaining state programs to mitigate well failure, such as delivering water tanker trucks is economically costly and unsustainable in the long-term (Feinstein et al 2017, Méndez-Barrientos et al 2022).For example the DWR's small community drought relief program had a budget of $190 million available in 2021, and $95 million to continue the program in 2022 in addition to $20 million allocated to the water storage tank and hauling program.

Conclusion
In this study, we conducted a comprehensive spatial analysis of domestic well failure in the southern part of the Central Valley, California, focusing on the subbasins of Tule, Kaweah, Kings, and Madera.This drought-susceptible region, known for its vast agriculture and remarkably unsustainable groundwater overdraft, has experienced thousands of domestic wells failures in recent years.We employ a reproducible modeling framework that integrates key components of the food-water-human system to understand domestic wells' vulnerability.We identified the proximity between domestic well depths and groundwater levels, the density of domestic and agricultural wells, and the extent of perennial crop areas as the most important factors in predicting the probability of a domestic well failing.With the impending risk of droughts and the imperative to establish measures for groundwater sustainability and domestic wells protection, local GSAs need to identify effective strategies.This analysis offers insights to help water agencies prioritize their land and water management strategies.Though the spatial analysis in this study provides a data-driven analysis that can be used as reference for understanding the association among key variables, it is essential to acknowledge the complexity of groundwater flow dynamics.Achieving a more precise aquifer dynamics is crucial for comprehending the effectiveness of potential land and water management strategies.Finally, the findings and discussion in this study are relevant not only to other basins within California but also to regions across the western United States and semiarid areas worldwide facing similar challenges, where unsustainable agricultural pumping practices render domestic wells highly vulnerable during droughts, posing a barrier to rural communities' access to reliable water.

Figure 1 .
Figure 1. Figure (A) shows the location of reported domestic dry wells between 2014 and 2022 in the Central Valley California, highlighted in red color is the study area.Figure (B) shows the number of domestic and agricultural wells per PLSS section (1mi 2 ) of the build since 1970.Figure C shows the number of dry wells reported in the 2014-2022 period by PLSS section.
al 2015, Gershunov et al 2019, Payne et al 2020), which result in multiyear drought periods and significant surface water shortages.The agricultural sector, which holds substantial economic importance in the region and is recognized as one of the most economically valuable in the United States (CDFA 2022), heavily relies on groundwater extraction for irrigation purposes.During dry periods, there is a notable surge in the construction of agricultural wells and subsequent pumping (Lund et al 2018, Jasechko et al 2020, Medellín-Azuara et al 2022).This trend has triggered a range of adverse effects, including increased depth to groundwater (Liu et al 2022, Vasco et al 2022), reduced groundwater storage (Alam et al 2021), land subsidence (Ojha et al 2019), impacting groundwater dependent ecosystems (Howard et al 2023), and deterioration of groundwater quality (Levy et al 2021).

Figure 2 .
Figure 2. Example of a reported domestic dry well location, and surrounding agricultural and domestic wells.The section of the PLSS where a dry well is located is illustrated in red.The 9 mi 2 (23.3 km 2 ) resolution is illustrated in yellow and the 25 mi 2 (64.7 km 2 ) in green.

Figure 3 .
Figure 3. Number of reported dry wells in the four groundwater subbasins of the study area.Dry periods are highlighted in red.

Figure 5 .
Figure5.Results from performing linear fits for the covariates used in the study.The mean of the linear fit is depicted in orange and the estimated 95% credible interval in gray.Dots represent observations of reported domestic well failure (1) or not (0).This analysis ignores spatial random effects and fits were generated using the model: W i ∼ Bernoulli(p i ) where logit(p i ) = β j X j and W i = 1 if domestic well failure was reported and W i = 0 if not.
Valley and other places in western USA (Perrone and Jasechko 2019, Jasechko et al 2020, Pauloo et al 2020).In addition, drawdown can create localized water quality degradation (Smith et al 2018, Levy et al 2021).

Figure 6 .
Figure 6.Subfigure (A) shows the mean and 95% credible intervals of the marginal posterior distribution of fixed effects and interaction effects.Subfigure (B) shows the posterior distribution for groundwater subbasin specific random effects.Estimates are in the log-odds scale, and are also summarized in tables S4 and S5.Surface water supply (surface_water), ratio between well depth and groundwater level (r_well_depth_gw_level), proportion perennial crops (r_area_perennial), proportion forage crops (r_area_forage), proportion annual crops (r_area_annual), poverty index (poverty), density agricultural wells (ag_wells_density) and domestic wells density (dom_wells_density).

Table 1 .
Summary of the variables before standardization used in the statistical model for the 9mi 2 resolution.Wells density and crop rate areas were calculated using ≈5760 acres for each well's 9mi 2 resolution polygon.
). R-INLA accounts for spatial autocorrelation estimating a Gaussian Markov random field via stochastic partial differential equation (SPDE) with a Matérn covariance function (Krainski et al 2018).This method has proven to efficiently model spatially dependence of residuals and has been widely applied in socio-environmental and environmental systems (Nelson et al 2017, Expósito-Granados et al 2019, Burchfield and Nelson 2021, Gong et al 2021, Jaffé et al 2021, Bosmans et al 2022, Ndolo et al 2022, and Managed Aquifer Recharge (MAR)(Marwaha et al 2021, Ulibarri et al 2021, Wendt et al 2021, Levintal et al 2023).For example, the California Department of Conservation is implementing the Multibenefit Land Repurposing Program (MLRP) which objective is to retire high irrigation demand cropland to reduce groundwater use, and allocate it to other less water intensive uses, while providing other benefits for communities and ecosystems (Espinoza et al 2023).Another example is the LandFlex program funded by DWR and implemented by GSAs, that provides financial incentives to retire cropland in close proximity to domestic wells, providing economic incentives for each acre-foot (1233 m 3 ) of permanent overdraft eliminated.
Effects of the structure of water rights on agricultural production during drought: a spatiotemporal analysis of California's central valley Water Resour.Res.53 8293-309 OEHHA 2021 CalEnviroScreen 4.0 OEHHA (available at: https:// oehha.ca.gov/calenviroscreen/report/calenviroscreen-40)Ojha C, Werth S and Shirzaei M 2019 Groundwater loss and aquifer system compaction in san joaquin valley during 2012-2015 drought J. Geophys.Res.: Solid Earth 124 3127-43 Pace C, Balazs C, Bangia K, Depsky N, Renteria A, Morello-Frosch R and Cushing L J 2022 Inequities in drinking water quality among domestic well communities and community water systems, California, 2011-2019 Am.J. Public Health 112 88-97 Pauloo R A, Escriva-Bou A, Dahlke H, Fencl A, Guillon H and Fogg G E 2020 Domestic well vulnerability to drought duration and unsustainable groundwater management in California's central valley Environ.Res.Lett.15 044010 Payne A E, Demory M-E, Leung L R, Ramos A M, Shields C A, Rutz J J, Siler N, Villarini G, Hall A and Ralph F M 2020 Responses and impacts of atmospheric rivers to climate change Nat.Rev. Earth Environ. 1 143-57 Perrone D et al 2023 Stakeholder integration predicts better outcomes from groundwater sustainability policy Nat.Commun.14 3793 Perrone D and Jasechko S 2017 Dry groundwater wells in the western United States Environ.Res.Lett.12 104002 Perrone D and Jasechko S 2019 Deeper well drilling an unsustainable stopgap to groundwater depletion Nat.Sustain. 2 773-82 Qin Y, Mueller N D, Siebert S, Jackson R B, AghaKouchak A, Zimmerman J B, Tong D, Hong C and Davis S J 2019 Flexibility and intensity of global water use Nat.Sustain. 2 515-23 Rodríguez-Flores J M, Gupta R S, Zeff H B, Reed P M and Medellín-Azuara J 2023 Identifying robust adaptive irrigation operating policies to balance deeply uncertain economic food production and groundwater sustainability trade-offs J. Environ.Manage.345 118901 Scanlon B R, Rateb A, Pool D R, Sanford W, Save H, Sun A, Long D and Fuchs B 2021 Effects of climate and irrigation on GRACE-based estimates of water storage changes in major US aquifers Environ.Res.Lett.16 094009 Smith R, Knight R and Fendorf S 2018 Overpumping leads to California groundwater arsenic threat Nat.Commun.9 2089 Stone K M, Gailey R M and Lund J R 2022 Economic tradeoff between domestic well impact and reduced agricultural production with groundwater drought management: Tulare county, California (USA), case study Hydrogeol.J. 30 3-19 SWRCB 2022 Supplemental Appendix e: Future Socioeconomic Indicator Considerations for SSWSs and DWs (California State Water Resources Control Board) Tariqi A Q and Naughton C C 2021 Water, health and environmental justice in California: geospatial analysis of nitrate contamination and thyroid cancer Environ.Eng.Sci.38 377-88 Ulibarri N, Escobedo Garcia N, Nelson R L, Cravens A E and McCarty R J 2021 Assessing the feasibility of managed aquifer recharge in California Water Resour.Res.57 e2020WR02929 Valavi R, Elith J, Lahoz-Monfort J J and Guillera-Arroita G 2019 blockCV: An r package for generating spatially or environmentally separated folds for k-fold cross-validation of species distribution models Methods Ecol.Evol. 10 225-32 Vasco D W, Kim K H, Farr T G, Reager J T, Bekaert D, Sangha S S, Rutqvist J and Beaudoing H K 2022 Using sentinel-1 and GRACE satellite data to monitor the hydrological variations within the tulare basin, California Sci.Rep. 12 3867 Watanabe S 2010 Asymptotic equivalence of bayes cross validation and widely applicable information criterion in singular learning theory J. Mach.Learn.Res.11 3571-94 Wendt D E, Loon A F V, Scanlon B R and Hannah D M 2021 Managed aquifer recharge as a drought mitigation strategy in heavily-stressed aquifers Environ.Res.Lett.16 014046 Zuur A F, Ieno E N and Saveliev A A 2017 Spatial, Temporal and Spatial-Temporal Ecological Data Analysis With r-Inla (Highland Statistics Ltd) p 1