Identifying agrisolar PV arrays across the CCV

We used remotely sensed imagery of existing solar PV arrays and geographic information system (GIS) datasets to develop a comprehensive and publicly available dataset of ground-mounted arrays co-located with agriculture in the CCV through 2018. We extracted all existing non-residential arrays from two geodatabases (Kruitwagen et al.4,57 and Stid et al.5,58) within the bounds of the CCV alluvial boundary59. We removed duplicate arrays and applied temporal segmentation methods described in Stid et al.5 to assign an installation year for Kruitwagen et al.4 arrays. We acquired Kruitwagen et al.4 panel area within array bounds by National Agriculture Imagery Program imagery pixel area with solar PV spectral index ranges suggested in Stid et al.5 and removed commissions (reported array shapes with no panels). We then removed arrays with >70% overlap with building footprints60 to retain only ground-mounted installations. Finally, overlaying historical CDL crop maps with new array shapes, we removed arrays in areas with majority non-agricultural land cover the year before installation (Supplementary Fig. 4 and Supplementary Discussion).

The resulting dataset (925 agrisolar co-located arrays) included 686 ground-mounted arrays from Stid et al.5 plus 239 from Kruitwagen et al.4. For these sites, we calculated array peak capacity (kWp) by61:

$$\mathrm{Capacity}={\mathrm{Area}}_{\mathrm{panel}}\times\eta\times{G}_{\mathrm{STC}}$$

(1)

where \({\mathrm{Area}}_{\mathrm{panel}}\) is the total direct area of PV panels in m2, \(\eta\) is the average efficiency of installed PV modules during the array installation year62 (Supplementary Fig. 5) and \({G}_{\mathrm{STC}}\) is the irradiance at standard test conditions (kW m–2). Arrays were split into ‘Commercial-’ (p) and ‘Utility-’ (≥1 MWp) scale arrays following the California Public Utility Commission NEM capacity guidelines63.

Scenario summary and assumptions

We computed annual FEW resource and economic values for each ground-mounted agrisolar PV array identified across the CCV for four scenarios: (1) reference, business as usual with no solar PV installation and continued agricultural production on the same plot of land, (2) baseline, agrisolar PV installation with moderate assumptions related to each component of the analysis, (3) worst case, PV installation with high negative and low positive effects for each component, (4) best case, similar but opposite of the worst-case scenario. We compare baseline to the reference scenario to estimate the most likely FEW and economic effects and use the differences between best- and worst-case scenarios to estimate uncertainty. Supplementary Tables 2 and 3 provide an overview of scenarios for each resource and Supplementary Tables 4 and 5 for baseline agrisolar lifespan FEW resource and economic value outcomes, respectively.

Identified arrays were installed between 2008 and 2018 and were assumed to have a 25-year lifespan for arrays due to performance, warranties, module degradation and standards for electrical equipment64,65. We assumed that land-use change effects ceased following 25 years of operation to simplify assumptions about module replacement, resale or continued use. We then summarized the FEW and economic effects of all arrays across the CCV and divided our temporal analysis into three phases: (1) addition (2008–2018) where arrays were arrays were being installed across the CCV, (2) constant (2019–2032) with no array additions but all arrays installed by 2018 are operating and maintained and (3) removal (2032–2042), where arrays are removed after 25 years of operation.

We performed several sensitivity analyses to address limitations in the available data and methods and to show how changes in future policy (NEM) could affect incentives displayed here. Sensitivity analysis included the capacity cut-off between commercial- and utility-scale (5 MW), solar PV lifespan (15 and 50 years), nominal discount rate (3%, 7% and 10%), solar PV direct area bias (proportional direct to total infrastructure area and a uniform perimeter buffer) and irrigation redistribution (assuming 50% of irrigation water-use offset is redistributed rather than conserved), all else equal (Supplementary Discussion and Supplementary Tables 6–20). We discuss additional assumptions and limitations in Supplementary Discussion.

Displaced crop and food production

Replacing fields (or portions thereof) with solar PV arrays affects crop production by (1) lost production of food, fibre and fuels and (2) reduced revenue from crop sales. We simplify the complex effects of lost production and include solely the foregone calories through both direct and indirect human consumption, which is justified because CCV crop production is largely oriented towards food crops. Future analyses could evaluate the lost fibre (primarily via cotton) or fuel (via biofuel refining) production.

We evaluated the economic and food production effects of displaced crops through a crop-specific opportunity cost assessment of land-use change, incorporating actual reported; yields, revenue, caloric density and regionally constrained caloric conversion efficiencies for feed/silage and seed oil crops. All crop type information was derived from the USDA National Agricultural Statistics Service (NASS) CDL22 for the array area in both prior- and post-installation years (Supplementary Fig. 4 and Supplementary Methods provide the adjacent fallowed land analysis). Each array was assigned a majority previous crop from the spatially weighted means of crop types within the array area for the five years before the installation.

We converted all eligible crop types to kcal (also called calorie) for human consumption after Heller et al.25. Foregone food production (\({\mathrm{Food}}_{\mathrm{Foregone}}\) in kcal) following PV installation was then defined for each array as:

$${\mathrm{Food}}_{\mathrm{Foregone}}={\mathrm{kcal}}_{\mathrm{density}}\times\mathrm{Yield}\times\mathrm{Area}$$

(2)

where \({\mathrm{kcal}}_{\mathrm{density}}\) is in kcal kg–1, \(\mathrm{Yield}\) is in kg m–2 and \(\mathrm{Area}\) of each array in m2. Crop-specific caloric density data (kcal kg–1) were derived from the USDA FoodData Central April 2022 release66. FoodData food descriptions and nutrient data were joined and CDL specific crop groupings were made through a workflow described in Supplementary Fig. 6. Crop-specific yield data (kg m–2) were derived from the USDA NASS Agricultural Yield Surveys67. State-level (California) yield data were processed similarly, with missing crop data filled based on national average yields. We used caloric conversion efficiencies for feed, silage or oil crop to account for crop production that humans do not directly consume.

For each array, we calculated annual revenue of forgone crop production in real (inflation adjusted) dollars, calculated by:

$${\mathrm{Crop}}_{\mathrm{Foregone}}={\mathrm{Price}}_{\mathrm{crop}}\times\mathrm{Yield}\times\mathrm{Area}$$

(3)

where \({\mathrm{Price}}_{\mathrm{crop}}\) is in US$ kg–1, \(\mathrm{Yield}\) is in kg m–2 and \(\mathrm{Area}\) of each array in m2. We used the annual ‘price received’ for all crops in the USDA NASS Monthly Agricultural Prices Report for 2008 through 201868. For the baseline case, we assumed that food prices will scale directly with electricity prices through 2042 given that they respond to similar inflationary forces69. Supplementary Fig. 6 and Supplementary Methods provide a more complete workflow including best- and worst-case scenario assumptions.

Change in irrigation water use and cost savings

Irrigation water use can only be offset by agrisolar co-location if the prior land use was irrigated. The presence of irrigation was inferred from the Landsat-based Irrigation Dataset (LanID) map for the year before installation70,71 (Supplementary Fig. 4). If the array area contained irrigated pixels, then we assumed the cropland area and all respective crops within the rotation were irrigated.

We calculated the total forgone irrigation water use (\({\mathrm{IrrigWater}}_{\mathrm{Foregone}}\) in m3) by:

$${\mathrm{IrrigWater}}_{\mathrm{Foregone}}=\frac{{\mathrm{IrrigWater}}_{\mathrm{year}}}{{\mathrm{IrrigWater}}_{\mathrm{survey}\; \mathrm{year}}}\times{\mathrm{IrrigDepth}}_{\mathrm{crop}}\times\mathrm{Area}$$

(4)

where \({\mathrm{IrrigDepth}}_{\mathrm{crop}}\) in m is the crop-specific irrigation depth, \({\mathrm{IrrigWater}}_{\mathrm{year}}\) in m3 is the annual county-level irrigation water-use estimate and \({\mathrm{IrrigWater}}_{\mathrm{survey}\; \mathrm{year}}\) in m3 is the county-level irrigation water-use estimate for the respective survey year irrigation depths.

We estimated annual crop-specific county-level irrigated depths from survey and climate datasets for each array. Crop-specific irrigation depths (\({\mathrm{IrrigDepth}}_{\mathrm{crop}}\)) were taken from the 2013 USDA Farm and Ranch Survey72 and 2018 Irrigation and Water Management Survey73, and logical crop groupings were applied (for example, almonds, pistachios, pecans, oranges and peaches were considered orchard crops). Because irrigation depths depend on the total precipitation in each survey year, we used multilinear regression to build annual county-level irrigation water-use estimates (\({\mathrm{IrrigWater}}_{\mathrm{year}}\)) from five-year US Geological Survey (USGS) water use74, gridMET growing season average precipitation75, with year as a dummy variable to incorporate temporal changes in irrigation technologies and practices. For the installation phase (2008 to 2018), these depths varied based on historical climate and survey data, whereas the projection phases (constant and removal) used a scenario-dependent moderate, wet (worst-case, least water savings) or dry (best case, most water savings) year estimate from the historical record (discussed in Supplementary Methods).

Assigning an economic value to water use is difficult and varies based on the temporally changing supply and demand76. We calculated the economic value of the change in water use (Water in real US$) to the farmer by:

$$\mathrm{Water}=(\Delta\mathrm{Water}_{\mathrm{use}}\times\,{\mathrm{Irrig}}_{\mathrm{Energy}}\times\,{\mathrm{Price}}_{\mathrm{Elec}})+{\mathrm{Water}}_{\mathrm{right}}$$

(5)

where \({\Delta\mathrm{Water}}_{\mathrm{use}}\) (m3) is the offset irrigation water use for the co-located crop minus O&M projected water use, \({\mathrm{Irrig}}_{\mathrm{Energy}}\) (MWh m–3) is the irrigation electricity required to irrigate the co-located crop given local depth to water and drawdown estimates from McCarthy et al.77, \({\mathrm{Price}}_{\mathrm{Elec}}\) (US$ MWh–1) is the utility-specific (commercial-scale) or regional average (utility-scale) annual price of electricity based on the electricity returns and modelled electricity generation described in Supplementary Methods and \({\mathrm{Water}}_{\mathrm{right}}\) is a CCV-wide average water right contract rate of ~ US$0.03 m–3 (ref. 78). Here we assume that water (and thus energy) otherwise used for irrigation was truly foregone and not redistributed elsewhere within or outside the farm. Change in O&M water use was based on Klise et al.79 reported values, described in Supplementary Methods.

Electricity production, offset and revenue

Installing solar PV in fields has three benefits: (1) production of electricity by the newly installed solar PV array, (2) reduction in energy demand due to reduced water use and field activities and (3) revenue generation via net energy metering (NEM) or land lease. Here we assume that on-farm electricity demand is dominated by electricity used for irrigation and ignore offset energy (embodied) used for fuel.

We modelled electricity generation for each array using the pvlib python module developed by SANDIA National Laboratory80. Weather file inputs for pvlib were downloaded from the National Renewable Energy Laboratory (NREL) National Solar Radiation Database81. We also estimated annual on-farm load to differentiate offset electricity use and surplus generation. Not only is electricity generated by the arrays, but electricity consumption is foregone for each array due to not irrigating the array area. The annual change in electricity consumption due to water use (\({\mathrm{Electricity}}_{\mathrm{water}\; \mathrm{use}}\) in GWh) is calculated by:

$${\mathrm{Electricity}}_{\mathrm{water}\; \mathrm{use}}={\mathrm{IrrigElec}}_{\mathrm{demand}}\times{\Delta\mathrm{Water}}_{\mathrm{use}}$$

(6)

where \({\mathrm{IrrigElec}}_{\mathrm{demand}}\) is the county-level rates for irrigation electricity demand in GWh m–3 and \({\varDelta\mathrm{Water}}_{\mathrm{use}}\) is the change in water use in m3 from equation (5). County-level electricity requirements to irrigate were calculated using irrigation electricity demand methods described in McCarthy et al.77 modified with a CCV-specific depth to water (piezometric surface) product for the spring (pre-growing season) of 201882.

Revenue from electricity generation was calculated separately for each array depending on array size and the installation year. Commercial-scale arrays (Supplementary Methods and Supplementary Table 21). Thus, for commercial-scale arrays, annual cash flow from solar PV (NEM in US$) is calculated as:

$$\mathrm{NEM}={\mathrm{Saved}}_{\mathrm{offset}\; \mathrm{load}}+{\mathrm{Earned}}_{\mathrm{surplus}}$$

(7)

where \({\mathrm{Saved}}_{\mathrm{offset}\; \mathrm{load}}\) is real US$ saved by offsetting annual on-farm electric load and \({\mathrm{Earned}}_{\mathrm{surplus}}\) is real US$ earned by surplus PV electricity generation sold to the utility under NEM guidelines. Both \({\mathrm{Saved}}_{\mathrm{offset}\; \mathrm{load}}\) and \({\mathrm{Earned}}_{\mathrm{surplus}}\) are estimated based on pvlib modelled electricity generation and valued at the historical utility-specific energy charge retail rates. Historical energy charges are available either through utility reports83,84,85 or the US Utility Rate Database via OpenEI86. We made several assumptions that resulted in omission of fixed charges including transmission and interconnection costs from the analysis. Details about electricity rates and omitted charges are summarized in Supplementary Methods.

For utility-scale arrays (≥1 MW), annual revenue from agrisolar co-location (Lease in US$) was assumed to be given by:

$$\mathrm{Lease}={\mathrm{Land}}_{\mathrm{lease}}\times\mathrm{Area}$$

(8)

where Lease is the economic value estimated to be paid to the farmer by the utility for leasing their land in US$ m–2 and Area of each array in m2.

We assumed commercial-scale arrays installed before 2017 were grandfathered into NEM 1.0 guidelines for the duration of their lifespan. However, arrays installed in 2017 and 2018 fall under NEM 2.0 guidelines which include a US$0.03 kWh–1 non-bypassable charge removed from \({\mathrm{Earned}}_{\mathrm{surplus}}\)21,87,88. Annual on-farm operational load was estimated and distributed across the year based on reported California agricultural contingency profiles89 and Census of Agriculture county-level average farm sizes90,91,92 (Supplementary Figs. 7 and 8 and Supplementary Methods). With distributed hourly load estimations and modelled solar PV electricity generation, we delineated electricity and revenue contributing to annual load (\({\mathrm{Saved}}_{\mathrm{offset}\; \mathrm{load}}\)) from surplus electricity and revenue that would have been sold back to the grid and credited via NEM (\({\mathrm{Earned}}_{\mathrm{surplus}}\)).

Future electricity revenue was projected using 2018 conditions (contribution to annual load, to surplus) and energy charge rates, modelled electricity production described above (includes degradation, pre-inverter, inverter efficiency and soiling losses) and projected changes in the price of electricity. The Annual Energy Outlook report by the US Energy Information Administration (EIA) provides real electricity price projections annually between 2018 and 2050 for ‘Commercial End-Use Price’93. This annual rate of change was used to estimate projected deviations from 2018 energy charges (2018 US$ kWh–1) during the constant and removal phases (2019–2042), with projected solar PV generation including discussed losses.

We used solar land consultant and industry reports for solar land-lease (\({\mathrm{Land}}_{\mathrm{lease}}\)) rates that ranged from US$750 ha–1 yr–1 to US$4,950 ha–1 yr–1, with high-value land averaging IS$2,450 ha–1 yr–1 in the CCV94,95. Comparable lease rates (~US$2,500 to US$5,000 ha–1 yr–1) were reported by developers in the CCV region17 and used in a solar PV and biomass trade-off study in Germany18 (~US$1,000 to US$2,950 ha–1 yr–1).

Array installation and O&M costs

Historical installation costs (Installation) were taken from the commercial-scale PV installation prices reported in the Annual Tracking the Sun report where reported prices are those paid by the PV system owner before incentives62. The baseline scenario is the median installation price, whereas the best- and worst-case scenarios are the 20th and 80th percentile installation costs, respectively. These reported values are calculated using NREL’s bottom-up cost model and are national averages using average values across all states. Installation cost was not discounted, as it represents the initial investment for commercial-scale installations at day zero. All future cash flows, profits and costs are compared to this initial investment. We also included the 30% Solar Investment Tax Credit in the Installation for commercial-scale arrays96. The system bounds of this impact analysis were installation through the operational or product-use phase. We, therefore, did not assume removal expenses or altered property value (terminal value) to remove uncertainty in decision making at the end of the 25-year array lifespan.

Historically reported and modelled O&M values (pre-2020) range from US$0 kWp–1 yr–1 (best case) to US$40 kWp–1 yr–1 (worst case) with an average (baseline) of US$18 kWp–1 yr–1 (refs. 97,98). Projected O&M costs were based on modelled commercial-scale PV lifetime O&M cost to capital expenditure cost ratios from historical and industry data that provided scenarios varying on research and development differences (conservative, moderate, advanced). The annual reported values are provided from 2020 to 2050 for fixed O&M costs including: asset management, insurance products, site security, cleaning, vegetation removal and component failure and are detailed in the Annual Technology Baseline report by NREL97, which are largely derived from the annual NREL Solar PV Cost Benchmark reports.

Farm operation costs

Business-as-usual farm operation costs (Operation) were derived from the ‘Total Operating Costs Per Acre to Produce’ reported in UC Davis Agricultural and Resource Economics Cost and Return Studies99. We removed operational costs to ‘Irrigate’ from the total because we estimate that as a function of electricity requirements and water rights (described in ‘Change in irrigation water use and cost savings’) while retaining ‘Irrigation Labour’ as this was not included in our irrigation cost assessment. Best- and worst-case scenarios for farm operation costs coincided with yield scenarios described in ‘Displaced crop and food production’.

Discounted cash flow for agrisolar co-location

For each commercial-scale array in the CCV, we computed the annual real cash flow as:

$$\mathrm{Commercial}=\mathrm{NEM}+\mathrm{Water}+\mathrm{Operation}-\mathrm{Food}-\mathrm{O}\& \mathrm{M}-\mathrm{Installation}$$

(9)

and for each utility-scale array as:

$$\mathrm{Utility}=\mathrm{Lease}+\mathrm{Water}+\mathrm{Operation}-\mathrm{Food}$$

(10)

where Commercial is the real return in 2018 US$ for commercial-arrays (p) and Utility is the real return in 2018 US$ for utility-scale arrays (≥1 MWp). Each of the terms on the right-hand side of these equations are defined in the sections above.

We then computed real annual discounted cash flow (\({\mathrm{DCF}}_{\mathrm{real}}\)) for each array to estimate the total lifetime value of each array. The \({\mathrm{DCF}}_{\mathrm{real}}\) at any given year n is calculated for each array by:

$${\mathrm{DCF}}_{\mathrm{real}}=\mathop{\sum }\limits_{n=1}^{25}\frac{{\mathrm{CF}}_{n}^{\mathrm{real}}}{{(1+{r}_{\mathrm{real}})}^{n}}$$

(11)

where \({{\mathrm{CF}}_{n}}^{\mathrm{real}}\) is the real annual cash flow at year n (either Commercial or Utility as relevant for each array) and \({r}_{\mathrm{real}}\) is the real discount rate without an expected rate of inflation (i) from the nominal discount rate (\({r}_{\mathrm{nom}}\)) calculated using the Fisher equation100:

$${r}_{\mathrm{real}}=\frac{\left(1+{r}_{\mathrm{nom}}\right)}{\left(1+i\right)}-1$$

(12)

Vartiainen et al.101 clearly communicates this method in solar PV economic studies and discusses the importance of discount rate (in their case, weighted average cost of capital) selection. For i, we use 3%, which is roughly the average producer price index (PPI) and consumer price index (CPI) (3.4% and 2.4%, respectively) between 2000 and 2022 and comparable to other solar PV economic studies101,102. We use a 5% \({r}_{\mathrm{nom}}\)103 and perform a sensitivity analysis using 3%, 7% and 10% \({r}_{\mathrm{nom}}\) and discuss discount rates used in literature in Supplementary Discussion. Separately from the sensitivity analysis for \({r}_{\mathrm{nom}}\), we also calculated our best-case and worst-case scenarios for each array.

All prices were adjusted to 2018 US dollars for calculation of real cash flow terms in equations (11) and (9). We adjusted consumer electricity prices and installation costs for inflation to real 2018 US$ using the US Bureau of Labor Statistics Consumer Price Index for All Urban Customers104. We adjusted all production-based profits and costs (all other resources) using US Bureau of Labor Statistics Producer Price Index for All Commodities105.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.