We see that inland regions tend to feature lower average TDS and the parcels within the DWZ seem to have lower salinity than coastal parcels without access to recycled water. Data on land use were compiled annually for the 2009 and the 2011-2020 growing seasons by PV Water using visual inspection. The district did not collect land use information in 2010. To collect these data, PV Water staff drove from parcel-to-parcel and surveyed the land for crop type and acreage. The survey work involved several quality checks. Hydrology staff evaluated if any areas of land were missed by the initial inspectors, and then digitized the survey. Then, a randomly selected group of 100 parcels were chosen to do a final quality control check to verify that spatial coverage and crop type were correct. Agricultural land use types include fallow ground, vegetable row crops, strawberries, blackberries and raspberries, blueberries, vine crops, artichokes, orchards, nursery, and unknown agricultural use.Given the limited number of fields that were classified as blueberries, vine crops, and artichokes, we combined vine crops with blueberries, blackberries, and raspberries into one category , and combined artichokes in with the rest of the vegetable row crops. Our sample consists of the 726 parcels that remain in agriculture for the full time period with which we observe. We couple this detailed land survey data with tax assessor ownership and parcel boundary data from the County Assessor offices to form appropriate decision units. These data delineate property boundaries and enable us to assign land use and groundwater quality to each farm at the land parcel level,low round pots designated by the Assessor’s Parcel Number in the tax assessor data.

The average size of a parcel is 31.94 acres. Figure 2.7 shows shares of agricultural acreage in the Pajaro Valley classified as one of the primary six crop categories over time.Trends show decreases in strawberry over time, and increases in vegetable production. Both nursery and orchard crops have held fairly steady during this time period. Finally, we collected data on other relevant observables that may affect both groundwater salinity and land use decisions–namely, environmental conditions and crop prices. Data on weather are sourced from PRISM Climate Data, whose roughly 800 meter gridded data are the official climate data of the U.S. Department of Agriculture, and considers elevation, location, and coastal proximity, among other factors in the development of the climate model. We use the gridded monthly mean temperature and cumulative precipitation data in our analysis. We take the centroid of each land parcel in our dataset, and use the weather data for the grid cell corresponding to that centroid. Crop price and average per acre yield data were collected from the annual County Agricultural Commissioner reports which enabled us to estimate a price per acre across each crop category. Spatially and time-varying groundwater quality and land use data at the parcel level allow us to estimate a discrete choice crop switching model to understand the impact of salinity on farmer welfare. Our research design relies on observable changes in groundwater quality and allows us to control for a suite of observable and unobservable factors that may be correlated with both salinity, crop prices, and crop choice. In what follows, we present our modeling framework and empirical specification, address potential challenges to identification, and discuss the interpretation of the parameters.

One concern with estimating the likelihood that farmers switch crops in response to a change in salinity is the potentially endogenous nature of salinity, si . The diverse array of crops that are profitably grown in this region implies that there are many factors influencing crop choice, spanning agronomic, environmental, policy, and market conditions. While primarily driven by weather and climate, which are out of the control of the individual farmer, changes in groundwater salinity can be influenced by basin-wide groundwater pumping and thus potentially correlated with other unobserved economic factors. To address this concern, we condition on an array of observable and unobservable parcel specific and basin-wide factors that may be systematically correlated with both salinity and crop choice. Temporal variation in groundwater salinity is primarily driven by two factors: groundwater elevation, which is driven by aggregate groundwater pumping, and weather, both of which may correlate with crop choice. To account for these, we control for precipitation, temperature, and the groundwater elevation at the parcel level. We also condition on distance to the coastline since this is an important determining factor of soil texture and the spatial distribution of salinity in the Valley, and because coastal microclimates determine how well crops grow in certain areas. Features of the parcel, such as its size and crop history, are also likely to influence planting decisions and could be correlated with salinity, and motivate our inclusion of lagged crop choices. Additionally, management and policy choices on behalf of PV Water or the county, like the delivery of recycled water supplies to parcels inside the DWZ and rates charged to generate revenues to fund the program, also influence groundwater dynamics and farmer behavior.

Data on these variables allows us to directly account for these factors. Finally, the inclusion of a linear time trend accounts for basin-wide unobservables that may trend linearly with both salinity and crop choice over time. Residual variation captures other drivers of spatial and temporal changes in salinity that are plausibly independent from crop choice. This includes variation in the landscape that drives percolation and local proximity to wetlands and streams which carry water and flush saline soils. Groundwater flow patterns change when pumping occurs at different depths depending on the subsurface geology being encountered at any given layer in the aquifer, driving plausibly exogenous changes in salinity. Since our estimates of the WTP are derived by scaling the effect of changing salinity by the absolute value of the coefficient on per-acre crop revenues, we must also concern ourselves with the identification of price effects.14 The relevant geographic market for the crops grown in this region appears to be worldwide; for example, the U.S. is both an importer and exporter of strawberries, lettuces, fresh vegetables, blueberries, and apples. In fact, the U.S. produces only about 11% of the world production of strawberries in 2017 based on UN Food and Agriculture Organization statistics. Given Pajaro Valley’s small share in the world market, producers as individual sellers in our setting are unlikely to exercise influence on world prices, lending support for the assumption that the assignment of crop prices is as good as random. A related concern stems from our use of crop revenues instead of profits to derive WTP. When choosing a crop, what matters to farmers is the difference in profits across crops, not the difference in prices or revenues. Our approach accounts for this with the inclusion of crop dummies and a time trend which capture consistent cost differences across crop and common changes in costs that trend over time. Changes in cost by crop and time not controlled for by crop indicators and a time trend is unlikely to be correlated with changes in salinity or revenues. However, some downward bias may remain in the estimation of α,plastic pots 30 liters due to the measurement error in using revenue rather than profit. Identification of the WTP hinges on the assumptions that, conditional on this suite of relevant spatial and time-varying parcel-level observables and an annual time trend, unobservable factors are not correlated with both crop choice and salinity levels or crop prices. To provide support for this identifying assumption, we will consider a number of alternative specifications. First, we will show the insensitivity of our results to the inclusion and exclusion of a suite of potentially relevant parcel-specific confounders. Then we will show robustness to the exclusion of parcels in the DWZ, the small subset of farmers that receive limited recycled water deliveries, and to a different measure of salinity. In addition to the identifying assumption, our approach hinges on several modeling assumptions that are critical for the interpretation of our parameter estimates. First, by designating fallow or idle land as our outside option, we limit our sample to parcels that remain in agriculture throughout our study period and interpret our estimates as the WTP conditional on the land being used for farming.

One concern with this is that farmers may be able to exit the agricultural industry entirely in response to changing salinity, which may lead to the conversion of those parcel to industrial uses, growing suburban developments, or natural open spaces and preserves. Heavy agricultural zoning laws in this region prevent this kind of switching from occurring in many places and justifies our assumption to denote fallowed land as the outside option. Additionally, the choice to leave agriculture and convert to a different land use type is an irreversible decision best modeled in a dynamic framework. Second, our approach assumes that farmers can only mitigate the cost of rising groundwater salinity through the channel of crop choice. In other words, our model assumes that no improvements in groundwater salinity can be made in absence of crop switching. In reality, farmers may be able to adjust behavior in other unobservable ways, and these unobservable strategies likely have non-zero costs. For example, a farmer may be able to leach salts through the soil with alternative sources of water or adjust inputs to production to compensate for yield declines due to changing salinity. These mitigation strategies could affect groundwater salinity levels and the agricultural profits that would have been experienced in the absence of crop switching. Because of the potential existence of other unobservable mitigation strategies, the true short-run marginal WTP is likely higher than what we estimate from observable crop switching. We discuss the role of possible alternative sources of water in Section 5.1. Finally, farmers likely have imperfect information about groundwater salinity levels on their parcel at any given time. Our approach assumes that farmers know the TDS measured in the groundwater at their parcel prior to the growing season and use this information to make their crop choice decisions. However, it is possible that farmers have less than complete information and that observable crop switching reflects this. If so, our revealed preference estimates would likely underestimate their full-information preferences . Anecdotal evidence suggests that farmers in the region run groundwater quality tests in their private wells at least once a year, with more sophisticated growers testing multiple times a year. Based on PV Water’s experience, combined with state-level annual reporting requirements, we believe that the impact of imperfect information is minimal in our setting relative to others. Results from our discrete crop choice models are presented in Table 3.3 where each column represents an alternative specification. We report crop-specific estimates of the effect of changing salinity in rows, where salinity is measured as the total dissolved solids from March-May of the growing season. The coefficients on the crop type indicator variables are treated as random variables and are allowed to vary across parcels. Their coefficients and estimated standard deviations are reported in the table. Coefficients on additional independent variables are suppressed. We first report results from a simple specification with only county fixed effects in column . Columns , , and gradually introduce time-invariant and time-varying, parcellevel observables. Column includes parcel-specific factors that directly affect salinity, including the depth to the groundwater table, the cumulative precipitation from March-Mayof the growing year, and the average mean daily temperature from March-May of the growing year. Column further conditions on factors related to the parcel that may affect crop choice, including the parcel size, access to an alternative water source of different quality, distance to the coastline, and last season’s crop choice. Our preferred specification in column , which has the lowest AIC, adds in additional aggregate controls including the agency’s water pumping fee and a linear annual time trend. All reported coefficients are relative to fallow ground, which increases in response to an increase in salinity. Our results demonstrate that, compared to fallow ground, an increase in groundwater salinity decreases the probability that a farmer will grow a cash crop. We see the largest effects among caneberries, one of the most salt-sensitive crops grown in the region. However, vegetables and strawberries are also negatively impacted. Negative but imprecisely estimated effects are observed for orchard crops.