Water shortage risks from perennial crop expansion in California’s Central Valley

California’s Central Valley is one of the world’s most productive agricultural regions. Its high-value fruit, vegetable, and nut crops rely on surface water imports from a vast network of reservoirs and canals as well as groundwater, which has been substantially overdrafted to support irrigation. The region has undergone a shift to perennial (tree and vine) crops in recent decades, which has increased water demand amid a series of severe droughts and emerging regulations on groundwater pumping. This study quantifies the expansion of perennial crops in the Tulare Lake Basin, the southern region of the Central Valley with limited natural water availability. A gridded crop type dataset is compiled on a 1 mi2 spatial resolution from a historical database of pesticide permits over the period 1974–2016 and validated against aggregated county-level data. This spatial dataset is then analyzed by irrigation district, the primary spatial scale at which surface water supplies are determined, to identify trends in planting decisions and agricultural water demand over time. Perennial crop acreage has nearly tripled over this period, and currently accounts for roughly 60% of planted area and 80% of annual revenue. These trends show little relationship with water availability and have been driven primarily by market demand. From this data, we focus on the increasing minimum irrigation needs each year to sustain perennial crops. Results indicate that under a range of plausible future regulations on groundwater pumping ranging from 10% to 50%, water supplies may fail to consistently meet demands, increasing losses by up to 30% of annual revenues. More broadly, the datasets developed in this work will support the development of dynamic models of the integrated water-agriculture system under uncertain climate and regulatory changes to understand the combined impacts of water supply shortages and intensifying irrigation demand.


Introduction
Agriculture in dry and semi-arid regions strongly depends on water availability, which often constrains crop planting decisions and agricultural expansion (Rockström et al 2007). Driven by rising global demand for food, many such regions have developed extensive water delivery infrastructure and/or relied on unsustainable rates of groundwater extraction. While crop choice and water use decisions are often made by individual landowners, the surrounding region also experiences the economic and environmental consequences (Pfeiffer and Lin 2012). These include hydrologic alteration caused by surface water storage and conveyance systems (Döll et al 2009), as well as groundwater depletion, a global challenge occurring primarily in areas of high agricultural development (Siebert et al 2010, Scanlon et al 2012, Famiglietti 2014. Groundwater overdraft leads to higher pumping costs, poorer water quality (Kang et al 2019), decreased well yields (Konikow and Kendy 2005), and land subsidence (Smith et  Any further distribution of this work must maintain attribution to the author(s) and the title of the work, journal citation and DOI. extraction leads to collective depletion of a common pool resource (Hardin 1968).
Agriculture stresses regions where water is the limiting factor of production. Globally, 25% of agricultural regions face strong competition for water resources (Luck et al 2015). Several of these regions have built extensive water storage and conveyance infrastructure to enable the expansion of agriculture (Rufin et al 2018), though over-reliance on this infrastructure can increase the economic risks of water shortage (Srinivasan et al 2012). Viewed as a dynamic system, infrastructure development creates a feedback loop where agricultural water demand (and thus economic productivity) continue to grow, potentially requiring additional infrastructure to mitigate risk (Di Baldassarre et al 2018). Perennial crops are one example of this dynamic, because they generally grow in semi-arid regions and incur substantial costs during water shortages. In many parts of the world, the severity of water shortages due to agricultural development rivals that associated with climate change (Kummu et al 2010, Rodell et al 2018. Estimating historical changes in land use is a key step towards quantifying these consequences and understanding dynamic crop choice decisions in regions of water scarcity, a key component of the foodenergy-water nexus (Cai et al 2018). While multiple datasets exist to support this effort, they face notable limitations, mainly in the scale mismatch between the agriculture and water sectors. For example, planting and irrigation decisions are made at the scale of individual landowners (Foster et al 2019), and water allocations might be determined at the river basin scale, while the water resources impacts of agricultural trade are felt globally (D'Odorico et al 2019). Food-energywater systems are perhaps best analyzed at the mesoscale (Lant et al 2019), i.e. within political boundaries defining counties, districts, and cities. However, finerscale crop data is often needed, ideally over a long historical record if dynamic decisions are to be analyzed.
California's warm Mediterranean climate and expansive water infrastructure support a thriving agricultural industry, with annual revenue exceeding $30 billion (Cooley et al 2015). The state is also prone to multi-year droughts, including one of the most severe on record from 2012 to 2016, representing the increased variability in precipitation expected in a changing climate (Diffenbaugh et al 2015, Swain et al 2018. During this recent drought, surface water deliveries to agriculture ranged from 0% to 30% of contracted amounts (Lund et al 2018). Groundwater pumping increased as a result, amplifying the trend of long-term overdraft (Faunt 2009, Scanlon et al 2012. While this overdraft has primarily supported agriculture (Marston and Konar 2017), the burden of well water outages falls mainly to smaller rural communities lacking access to other supplies (Swain 2015, Perrone andJasechko 2017). Concern over groundwater depletion in California led to the 2014 Sustainable Groundwater Management Act (SGMA), which requires basins to balance groundwater use by 2040 (Conrad et al 2018). Regulations will be implemented by local agencies, with details still being finalized (Kiparsky 2016), and will likely curtail agricultural groundwater use in the future, resulting in economic losses.
Gradual changes in crop planting over time can also contribute to water stress, particularly in the Tulare Lake Basin in the southern Central Valley of California. The basin is one of the state's largest agricultural regions with roughly three million irrigated acres (CA DWR 2013), and has seen a transition to perennial crops in the region from 1980 to 2016, a trend which continued through two large droughts (figure 1). Figure 1 includes the five most common perennial crops (almonds, pistachios, oranges, table grapes, and wine grapes) and annual crops (alfalfa, silage, cotton, wheat, and tomatoes); additional crops are listed in supplemental table S4 is available online at stacks.iop.org/ERL/14/104014/mmedia (USDA National Agricultural Statistics Service California Field Office 2019). The trend follows a widening difference in inflation-adjusted revenue per acre between field crops and perennials. Because perennial crops require substantial capital investment and typically survive for decades, they cannot be fallowed during drought years without significant economic losses, resulting in demand hardening (Johnson and Cody 2015). Despite this limitation, water-intensive perennial crops also generate more revenue per unit of applied water (Medellín-Azuara et al 2015). The absence of crop decline during drought years in figure 1 suggests that revenue is the driving factor for crop choices in the region, and that additional costs of groundwater pumping during drought have not historically influenced crop choices.

Irrigation districts
The Tulare Lake Basin region includes 30 major irrigation districts within four counties (Kern, Tulare, Kings, and Fresno), shown in the supplementary material figure S2. Irrigation districts comprise many landowners who make crop choice decisions. Districts were chosen as the spatial unit of analysis because of their key role in regional water allocations and their mandated reporting of water supplies and demands via Agricultural Water Management Plans (AWMPs). The AWMPs estimate annual surface and groundwater water consumption in each district; a subset of districts include recent surface water delivery estimates for both wet and dry years, and will be selected for more detailed analysis. Each district has identified only a single wet and dry year over the period 2008-2015.

Crop data from pesticide use reports (PUR)
PUR from the California Department of Pesticide Regulation (DPR) provide crop acreage data at a spatial resolution of 1 mi 2 . This dataset includes the county, meridian, township, range, and section of almost all pesticide permits submitted since 1974, covering nearly every agricultural property during this time frame (DPR 2017). This extensive coverage allows for crop acreage analysis at a higher spatial resolution than would otherwise be available over this time period. The PUR archive is available on the DPR website for 1974-2016 and has recently been used by the Department of Water Resources for acreage estimates in the California Central Coast region (DPR 2016, Shipman et al 2018).
Acreage estimates for each crop type listed in the dataset were calculated by filtering the reports in each 1 mi 2 section for unique parcel codes, batch numbers, and crop codes, each of which are described by the metadata (DPR 2016). The supplementary material (section 2) explains this process in more detail. The resolution of this dataset allows for analysis of crop planting patterns and water needs at the district scale, which is not possible with county-level data, though it will not capture crops grown without pesticide application. The county-level data therefore is used to validate the aggregated PUR estimates as shown in figure S5 in the supplementary material. Importantly, while many remote sensing datasets such as CropScape (Han et al 2012) and LandIQ provide a spatial resolution of crop type to the level of individual orchards, they include only recent years and do not contain the longer record of the PUR data needed to analyze trends over the past several decades.

Water demand estimates
The spatially distributed crop data is compiled for the purpose of estimating trends in water demand over time, and the potential for demand hardening due to the expansion of perennial crops. Estimates of applied water per acre are compiled by the California Department of Water Resources (DWR 2018). This database estimates annual depth of applied water for each of 20 crop types for each hydrologic region in California, calculated from evapotranspiration, crop coefficients, soil characteristics, rooting depths, and precipitation, adjusted for irrigation efficiency. These applied water estimates are averages for orchards and fields within the 17 000-square mile hydrologic region, assuming that full watering requirements are fulfilled, and averaging over variability within each hydrologic region.
Water demand for each district is then calculated using the equation below, where D is the district water demand (acre-feet), N is the number of crops grown in the region, AW i is the applied water depth (acre-foot per acre) for crop i, and A i is the planted area (acres).
For perennial crops we also estimate the minimum depth of applied water needed to sustain trees, even in the case of little or no yield. Specifically, we assume a minimum 50% of applied water depth under regulated deficit irrigation (RDI) Soriano 2006, Chai et al 2016). As a sensitivity test, we repeat the analysis with RDI values of 40% and 60% as reported in the supplementary material (figure S9). In practice, RDI decisions depend on individual access to water rights, trading rights, and water conveyance infrastructure. The practice may become more common in the future as an option to for landowners to continue growing high-value perennial crops while reducing the economic risks during drought.

Economic analysis
Estimated water demands for each irrigation district are then compared with the dry-year surface water supplies reported in the AWMPs, reflecting a focus on drought and potential impacts of climate change. Most districts have selected 2014 as their representative dry year, which was the third driest out of a 110-year record at the statewide level. Scenarios are generated by reducing available groundwater by 10%-50%, in increments of 10%, of the estimated dry-year pumping levels. These groundwater reduction levels are subjective and meant to illustrate a range of possible revenue losses should groundwater reductions occur under new SGMA regulations, which are still being determined.
Revenue losses associated with water shortages are then calculated by distributing water shortages among crops, beginning with the crops with the lowest marginal value. Annual crops are fallowed first; followed by the use of RDI for perennial crops, which we assume eliminates yield for the given year; followed by pulling and replanting perennial crops, a costly measure of last resort. In practice, the effect of RDI on crop yield varies by crop type and regional climate (Kirda 2002), and in general may carry over to subsequent years (Goldhamer and Fereres 2017). Fallowed acreage is then converted to revenue loss based on nominal crop yield in the absence of water shortage. The baseline total agricultural revenue is calculated by multiplying crop acreages with their respective yield-per-acre averages and price-per-unit crop drawn from the County Commissioner dataset (USDA National Agricultural Statistics Service California Field Office 2019).
While water supplies are estimated based on data from single dry years in each irrigation district, shortages may lead to long-term effects on perennial crop revenue. Perennial crop pulling and replanting costs are calculated from the tree and vine loss tool created by the UC Davis Agricultural Issues Center (UC Agricultural Issues Center 2017), which assumes that replanting occurs five years into the productive lifespan of the orchard, and incorporates discounted revenue losses over time. Therefore, estimates of revenue losses include this temporal effect. The analysis further assumes that landowners can apply water to the highest-value crops anywhere in the district, which likely underestimates economic losses from groundwater pumping regulations in practice.
Additionally, the analysis does not account for groundwater pumping costs, which are assumed to have negligible effect on crop choice compared to revenue losses (figure 1). However, in practice, curtailment of pumping during drought could result in a rise in the water table and reduction of pumping costs. We also do not consider the costs of drilling deeper wells, which could be obtained by combining this analysis with a spatially distributed well depth dataset (e.g. Perrone and Jasechko 2017).

Crop acreage estimates
Spatially distributed crop type data is compiled from the PUR from 1974 through 2016. Figure 2 shows the expansion of perennial crops in the basin over the past two decades to highlight the recent acceleration of this trend. Perennial cultivation in the region has increased from an average of 567 thousand acres in the early 1990s to over one million acres in 2016. This increase is not uniform across the basin, as shown in figure 2. The central area of the valley remains primarily annual crops, while districts to the east and south have undergone more dramatic shifts towards permanent crops. While the extent of this trend within the basin varies between irrigation districts, it occurs continuously throughout the analyzed time frame.
Natural water availability has historically had little influence over yearly crop planting decisions, due to extensive surface water infrastructure and dependable groundwater reserves to buffer the effects of drought. These resources have enabled the shift to high-value perennial crops, driven primarily by market prices and changes in consumer income and preferences. Other regional factors such as soil quality clearly play a role in crop choice decisions, as figure 2 shows spatial variation within the region. The increase in perennial crops has led to both an increase in water demand and a decrease in the flexibility of this water demand. In other words, many perennial crops require more applied water each year, and consistent watering, without an option to cut back during drought years. At current rates of surface water imports and groundwater extraction, this practice has benefitted landowners as perennial crop yields have become increasingly profitable. However, future changes to water availability from climate trends, SGMA, and other regulations may change the economic outlook of the region's agriculture.

Agricultural water demands
For each district, we estimate the total water demand for annual and perennial crops, as well as the demand for RDI of perennial crops, assumed to be a theoretical minimum of 50% ( figure 3). This lower-bound demand calculation is useful for estimating a minimum water requirement if a drought or regulation significantly constrains the water supply during the year. The upward trend in the minimum irrigation required each year to sustain existing perennial crops suggests a demand hardening of agricultural water supplies, where water demand for agriculture becomes less flexible. The potentially high cost of lost perennials makes landowners less likely to fallow during water shortages and more willing to pay for alternative water sources if necessary.
Many of the irrigation districts depend on groundwater for much of their total water supply, especially during drought. The fraction of groundwater use varies greatly among irrigation districts. As a portion of total supply, surface water varies from roughly 70% to less than 10% for some districts during dry years, according to the AWMP reports. The spatially distributed crop data shows that perennial acreage also varies significantly across districts, ranging from less than 5% to nearly 100%. This suggests that their economic vulnerabilities to water shortage may be quite different.

Revenue loss from groundwater regulation
The growing inflexibility in water demand due to perennial crop expansion has broad implications depending on the local water rights and water demands within each district. While most irrigation districts can meet water demands at current rates of water imports and pumping volumes, a mandated reduction in groundwater pumping could incur extensive revenue losses. Figure 4 shows the potential revenue losses that occur under groundwater pumping restrictions ranging from 10% to 50%, assuming surface water deliveries from a historical dry year. Two crop distributions are considered: historical (1996) and current (2016). The variability in costs between districts shows how the combination of water portfolios and crop choices interact to create water shortage Figure 3. Estimated water demands, in thousand acre-feet (TAF), for selected irrigation districts. (a) District with primarily annual crops, (b) district with a shift from majority annual crops to majority perennial crops, and (c) district with consistently majority perennial crops. The green and yellow lines represent available surface water during a recent wet year and dry year, respectively. The difference between surface water supply and demand has historically been met by groundwater, which is now subject to regulation.
risks. As shown in figure 4, these interactions are not linear (e.g. revenue losses under 50% reductions are greater than twice those under 25% reductions) due to the assumption of fallowing the lowest-value field crops first.
In general, figure 4 shows that groundwater pumping reductions may cause significant losses relative to total revenue, particularly for districts which have seen large expansions of perennial crops relative to the historical scenario. At a 50% level of groundwater reduction, three districts would see revenue losses exceeding 50% of their usual yearly revenue generated from crop yields. The districts with the steepest revenue losses are those with fewer acres of field crops to fallow, instead forcing the expensive task of pulling and replanting perennial crops. Depending on the crop, this would cost an estimated 2-6 times annual revenue in the absence of groundwater constraints. This vulnerability associated with increasing perennial crop acreage from 1996 to 2016 occurs in almost every district, ranging up to a projected 30% increase in revenue losses during dry years relative to the historical crop distribution.
While the findings show potentially steep economic losses for irrigation districts with a strong reliance on groundwater and a high percentage of perennial crops, surface water and groundwater volumes typically vary each year for each irrigation district. Surface water deliveries are especially volatile due to uncertain precipitation, snowpack, and regulations feeding into California streams and reservoirs. Annual deliveries from the Central Valley Project (federal) or State Water Project (state) can vary from complete curtailment to almost double the average volume. While groundwater pumping has historically buffered these fluctuations, this response may be limited with implementation of the SGMA. Although many factors may influence a district's water imports and pumping decisions, the crop choice decisions in the region have been supported by landowners' ability to alleviate surface water shortages with groundwater. Changes to this past norm may result in economic losses as the region transitions from reliance on groundwater pumping to regulated long-term groundwater sustainability, underscoring the value of a diverse portfolio of water sources and crop types to manage uncertain future water availability. This analysis is repeated to test the sensitivity of the RDI parameter with values ranging from 40% to 60% as reported in supplemental figure S9. These results indicate that RDI affects the magnitude of revenue losses, as expected, but does not change the key outcome of increased vulnerability to water shortage with more perennial crops.

Discussion and limitations
These estimates of revenue loss reflect only a few out of many possible scenarios. In particular, the parameter values for groundwater curtailments and RDI are illustrative, and likely too simplistic to represent the eventual implementation of new SGMA regulations in California. In particular, this might involve more flexible opportunities for groundwater banking to augment storage during wet years, making droughts less costly Dahlke 2017, Ghasemizade et al 2019). Further, we do not account for the long-term retirement of productive land, which is likely to occur in this region under sustained curtailments on groundwater use (Hanak et al 2017). While this study focuses on the impacts of perennial crop expansion on water supply risks, perennial crops also provide ecosystem services beyond those of annual crops, including nutrient cycling, soil retention, carbon sequestration, and biodiversity conservation (Dale and Polasky 2007, Jose 2009, Asbjornsen et al 2014. If such benefits are to be managed directly, the composition of crop decisions may need to be modified between wet and dry years (Rapidel et al 2015), and the structure of planting can be chosen to balance the tradeoff between revenue and ecosystem services (Rey Benayas and Bullock 2012). Yet, particularly in California, it is not clear whether the ecosystem services provided by perennial crops are offset by the environmental issues caused by the large-scale surface water storage and conveyance systems that have significantly altered aquatic habitat, primarily to support this agricultural development (Grantham et al 2014).
This study does not account for access to other sources of surface water, for example via water markets (Maples et al 2018). Water sales and leases are an important tool for California farms and have become increasingly significant since the 1980s (Hanak and Stryjewski 2012). Surface water markets may consequently see steep price increases per acre-foot as during the recent drought, though this increase might more accurately reflect the cost of water use within the region. This study only considers costs during a single year; under SGMA legislation, regions are obliged to return groundwater levels to their previous levels after overdraft produced during a drought, meaning that costs might persist for several years beyond the end of a drought. Finally, this study does not include improvements to irrigation technology or institutional governance, which may improve agricultural water management in the coming decades (Zhu et al 2019).

Conclusion
Globally, some of the most productive agricultural regions face chronic water stress. Dry and semi-arid regions are especially vulnerable to the consequences of drought and groundwater overdraft, even when agriculture is supported by extensive water infrastructure. California's Central Valley offers one example where high-value perennial crops have increased consistently over the past four decades, creating water supply challenges for many irrigation districts. This study contributes a spatially distributed crop choice dataset for the Tulare Lake Basin compiled from PUR, which is used to analyze water shortage risks at the irrigation district scale. The looming threat of droughts paired with recent groundwater regulations will likely require many irrigation districts to navigate difficult tradeoffs between reducing dependence on groundwater while maintaining revenues from perennial crops. This analysis of water shortage risks aims to establish a baseline for potential future costs while highlighting the value of diversifying water supplies and crop choices to maintain flexibility in an uncertain future.
Zhu T, Ringler C and Rosegrant M W 2019 Viewing agricultural water management through a systems analysis lens Water Resour. Res. 55