Goal and scope of the model
This work aims to develop an LCA model of California almond production that captures variability in cropping systems over space and time. Key attributes include accounting for the effects of changing environmental (e.g., water availability for irrigation) and industrial conditions (e.g., the siting or closure of biomass energy facilities), as well as the changing location and extent of cultivation, embedded crop responses (yield and biomass production) to agronomic and management practices, and consideration of the changing fates of orchard co-products (Section S1, Supplementary Information). The reported functional unit is 1 kg of raw brown skin almond kernel, but the model is designed to report results in other plausible functional units including hectares, kilocalories, or specific nutritional values.
Almond production generates by-products and co-products (referred to collectively as co-products) alongside almond kernel. Conforming to the attributional nature of this LCA, economic allocation is used to apportion impacts to almonds as a baseline approach. Other partition-based allocation methods, such as those based on mass or energy content, attribute the majority of impacts to woody biomass generated from almond orchards and do not reflect the drivers for the almond production system. To deepen understanding of the environmental and resource implications of almond co-products and inform decision-making around co-product fates, a substitution (i.e., displacement) approach is also tested. This provides particular insights for orchard biomass which is used in bioenergy generation but has no economic value. Despite the fact that practice and policy have often applied substitution-based allocation in attributional contexts, the debate on using substitution-based allocation in attributional LCA has long held that it constitutes an incontrovertible inconsistency and should not be done (e.g., Majeau-Bettez et al. 2018). However, here substitution-based allocation provides insights unavailable from partition-based allocation methods and can be viewed in a consequential context around decision-making about how to utilize co-products given evolving priorities, markets, and policies that shape industry-wide and grower decisions.
The TRACI 2.1 impact assessment method is used to characterize the following impact categories; global warming for 20- and 100-year timeframes, total primary energy use, smog formation, ozone depletion, human health particulate, acidification, eutrophication, fossil fuel use (USEPA 2017). Two non-TRACI impact categories were also included; total freshwater use and the AWARE water scarcity indicator (WULCA 2018).
Perennial crops have multi-year to multi-decade lifespans, and as such require explicit accounting for temporal variation in production characteristics (illustrated in Fig. S2 of the supplementary material). The orchard life cycle is organized into distinct developmental stages: nursery production and sapling planting, establishment, growth, maturity, decline, and EoL—each of which entails a unique combination of tree physiological characteristics, input demands, and management operations. The scope of this analysis includes orchard establishment (including nursery production) through orchard end-of-life (tree removal and one fallow year), a total time period of 19–24 years, under California cultivation conditions for the three primary growing regions in the state leading up to the year 2018. Temporally variable data such as groundwater level and kernel price make use of 5-year mean values (2013–2018).
Given the intense scrutiny that almond production and the agriculture sector as a whole have received in post-drought California and given increased interest by retailers and consumers in understanding the impacts of food and diet choices, the audience of this study includes producers, retailers, and consumers, as well as relevant research communities. Study results are intended to support industry efforts towards sustainability (ABC 2019), demand for environmental labeling in foreign markets (Del Borghi 2013; EPD International 2020), and policy efforts for environmental regulation (CDFA 2020; California State Assembly 2006).
System definition, boundaries, and model assumptions
Because this LCA attempts to characterize almond production across the state of California, the systems that comprise the background and foreground of almond production are conceived at three scales: global background processes (e.g., nitrogen fertilizer production), which rely on reference LCIs from commercial databases; regional processes (e.g., sapling production or transportation of inputs and outputs from the orchard and post-harvesting sites), which rely largely on modeling or primary data collection; and the orchard agroecosystem (e.g., on-field operations and tree growth), which are also largely modeled processes. Figure 1 illustrates key processes within each of these scales, differentiating between physical flows, processes happening over time (temporal flows), and carbon pools. The system boundary is also included, showing that value-added processing, distribution, and consumption are not included in this assessment. Section S1 in the Supplementary Material provides additional information on the data structure and sources used in the SPARCS-LCA model.
The three major hydrologic/growing regions of California’s CV are included in the model: the Sacramento Valley (SV), San Joaquin Valley (SJV), and Tulare Lake (TL) regions (Fig. 2). Each of the modeled regions has distinct infrastructure, agronomic, and environmental conditions that affect input demands, management operations, and orchard productivity. For example, the SV region is generally characterized by greater water availability throughout the growing season, resulting in less demand on groundwater resources and lower energy consumption for irrigation. The SJV region is characterized by longer orchard lifespans and a greater availability of bioenergy facilities to act as a sink for EoL biomass by-product, while the TL region is characterized by higher water and nutrient demand and higher annual kernel yields. The geographic scale of this analysis is driven by technology and management practice adoption (including irrigation system, water source, nutrient and pest management choices), which is in turn driven by the general environmental and growing conditions that characterize these three regions.
Description of orchard life cycle and business-as-usual conditions
The life cycle of an orchard begins with sapling production at a nursery, and a fallow year between orchard plantings (year 0). Planting of saplings occurs in year 1 of the orchard life cycle, along with irrigation system installation, land preparation, fumigation, and a number of key management decisions like planting density and variety selection. Establishment (years 2–3) involves rapid tree growth and attendant biomass accumulation, a higher rate of tree loss and replacement, and structural pruning. Harvest and related operations (pollination, harvest, and postharvest) begin in year 3. Harvested almonds are comprised of a hull, shell, and kernel. Post-harvest operations thus yield almond kernel, along with hulls and shells.
The growth phase follows (years 4–7), during which biomass accumulation continues to increase logarithmically; agrochemical inputs increase in proportion to tree volume and nutrient demand. Maturity is reached at about year 8, when biomass accumulation and yield pass the inflection point of the logarithmic growth curve, and orchard productivity and inputs reach a relatively steady state until orchard EoL, which is initiated when maintenance costs exceed income. The business-as-usual (BaU) scenarios for SV, SJV, and TL regions consider EoL to occur in years 21, 24, and 19, respectively. EoL operations entail pushing and disposal of trees, either by in-field burning or chipping with subsequent delivery to bioenergy facilities, surface mulching, or soil incorporation of the chips. Assuming the orchard is replanted, the fallow year will begin anew (year 0).
BaU almond production for each region is based on regionally specific economic cost and return studies (Duncan et al. 2016; Pope et al. 2016; Yaghmour et al. 2016). These studies are based on consensus estimates of typical orchard inputs and management practices obtained from interviews and focus groups consisting of farm advisors, growers, and industry representatives. Additionally, the Almond Board of California (ABC) conducts grower self-assessment surveys through their California Almond Sustainability Program (CASP), which collects data on a few orchard inputs and a wide range of practices (ABC 2019).
Data on orchard planting parameters, agrochemical, fuel, water and material inputs, and orchard management operations (Table 1) from these sources were used to define BaU conditions, with the exception of orchard age at removal, which was obtained from analysis of historic aerial imagery (Google Earth Pro V. 7.3 2019). The BaU scenario is intended to represent a “typical” almond orchard in each of the three growing regions, as well as serving as a baseline for comparison with scenarios representing potential changes in management or environmental conditions. The results for each growing region are aggregated as an area-weighted mean to estimate results for the CV as a whole.
Yield and biomass productivity
Yield is modeled on an individual tree basis, using cost and return study estimates of typical regional yield, tree replacement rate, planting density (Figure S3 of the supplementary material), orchard lifespan, and production costs. These data were coupled with county-specific almond prices (USDA NASS 2019) and tree loss rates obtained from analysis of aerial imagery (Google Earth Pro V. 7.3 2019) to estimate individual tree lifetime productivity increase and decline, assuming that under typical circumstances a grower will remove the orchard once economic return no longer exceeds cost of production (Table 1). Under BaU conditions, therefore, the estimated kernel yield of the almond production system matches the regional cost and return study estimates with which the yield model was parameterized.
Regionally specific applied water quantities were taken from recent UC Cost Studies, making no assumption about effective rainfall, instead relying on grower and farm advisor estimates of actual applied water by region. These quantities should thus account for irrigation practice in both drought and normal years. Irrigation in California relies on an extensive surface water conveyance system and local groundwater. The location of use and the water supply type (surface versus groundwater) determine the energy intensity of irrigation water. No data on water supply type used in almonds at state, regional, or orchard scale are available. Instead, results from the Statewide Agricultural Productivity (SWAP) Model (Medellín-Azuara et al. 2011), an economic optimization model that uses water supplies and prices to simulate farmer decision-making, were used. The percentage of surface and groundwater used for irrigation by county calculated in SWAP was assigned to the orchard area in each county.
Once supply type is determined, the energy intensity of the supply in each county must be estimated. The impacts of irrigation water use are a function of total water demand and the energy intensity of water delivery. Groundwater pumping energy is a function of depth and was estimated using data from California Department of Water Resources test wells (California Natural Resources Agency 2019) for years 2012–2017 (including several years of drought conditions). In addition to water depth, groundwater pump technology and design affects energy demand. Technology-specific pump energy and efficiency data (Goulds Water Technology 2019) were used to generate a geospatial model estimating regional groundwater pumping energy demand across the CV (Figs. S4–S5 of the supplementary material).
Surface water energy intensity is a function of location in California; the extensive surface water delivery infrastructure in California includes large areas of gravity-fed water supplies, as well as areas where significant pumping energy is required. For surface water, aerial imagery (Google Earth Pro V. 7.3 2019) and spatial data from select irrigation districts in the CV were used to map surface water delivery infrastructure including the California Aqueduct and the Delta-Mendota Canal to almond orchards served.
In the state of California, there is increasing interest in the potential for groundwater recharge, whereby application of surface water can help maintain or increase groundwater levels, exceeding what would occur just with normal precipitation, a potential co-benefit of irrigation that can help sustain groundwater levels. Groundwater recharge potential was estimated based on spatial data (O'Geen et al. 2015; Kimmelshue et al. 2014), distribution of irrigation system types by county (Deng and Salas 2017), and irrigation system-specific water application efficiency values (UNFAO 1989). Groundwater recharge is assumed to occur incidentally during normal irrigation operations, with quantity determined by the water application efficiency of the local mix of irrigation system types obtained from California Almond Sustainability Program (CASP) data (ABC 2019).
Almond orchards yield significant quantities of co-products at two stages: woody biomass generated at orchard EoL, and during post-harvest operations. While woody biomass and shells might arguably be designated by-products or waste in some locations because they have no value, or may even be a cost to farmers and operators to manage, all are referred to collectively as co-products. Table 2 summarizes the fates of these co-products in the SPARCS-LCA BaU cases.
Previous analyses (Kendall et al. 2015; Marvinney et al. 2015) assumed that a flat 95% of EoL biomass went to energy proportion based on data from an agriservices company active in the SJV region. The current analysis updates these estimates by developing a geospatial model that accounts for biomass energy facility closure and idling from 2014 to 2018, as well as ground-truthed orchard distributions and ages for each growing region (Kimmelshue et al. 2014).
These data were used as primary inputs to an R-based geospatial model that simulates the California orchard landscape over decadal timescales to predict the potential supply of energy feedstock derived from almond EoL biomass co-product (Table 2; Figs. S2, S6 of the supplementary information). This model also used the CropScape Data Layer (USDA NASS 2019) to simulate a competing supply of biomass energy feedstock from other perennial cropping systems in the CV, as well as the logarithmic almond growth model developed previously (Kendall et al. 2015) and spatial and operational data on biomass energy facilities (CARB 2019). The biomass energy feedstock landscape was simulated from years 2014 through 2020, and the estimated values for almond biomass to energy feedstock for year 2018 were used in this analysis.
Data on biomass-to-energy transport costs and payments for feedstock obtained from an agriservices company were used to generate facility-specific economically feasible transport radii (i.e., the distance within which the value of biomass feedstock does not exceed the cost of transport), allowing a regionally specific “feedstock-shed” based analysis of biomass delivery from almond orchard EoL operations to bioenergy facilities. This analysis treats the bioenergy facility as gatekeeper for potential fossil fuel substitution benefits attributed to the orchard production system, and accounts for facility-specific energy production capacity (and thus, feedstock demand) and competition for feedstock delivery from other perennial crops, municipal sources, and forestry by-products.
Biomass fate modeling assumes that the almond grower prioritizes between fates based on expense and feasibility (i.e., available capacity), with energy production being the most cost-effective, followed by in-field burning, surface mulching, and soil amendment or “whole orchard recycling” (WOR) (Holtz et al. 2014; Wolff and Guo 2019). Soil amendment may also be used to dispose of hulls, shells, and prunings in either tractor row soils during the orchard productive lifespan or off-site, whereas WOR, a relatively new possible practice, refers specifically to the incorporation of chipped EoL tree removals back into the floor soil of the orchard from which they were removed. Energy production is limited by the capacity of biomass energy facilities within the economically feasible transport radius of any given orchard block to accept additional biomass feedstock, and in-field burning is limited by air quality regulations. Once these pathways have reached their maximum, 99% of the additional biomass produced in that region is assumed to be directed to surface mulching, and 1% to WOR—a relatively new practice in California.
Post-harvest co-products are generated at hulling and shelling operations, where harvested in-hull almonds are transported. Data on post-harvest operations, inputs, and biomass co-products were obtained from a survey of hulling and shelling operations throughout the CV (Kendall et al. 2015). The primary co-products are hulls and shells, though hash (a mixture of crushed shells and kernel) and twigs are also generated, albeit in relatively small quantities. Almond hulls and hash are assumed to be used as a dairy feed, while shells and twigs tend to find lower-value uses as livestock bedding or energy feedstock. Any of these co-products may also be incorporated into soils or surface mulched if not directed to another use. Table 2 describes the proportion of each co-product directed to each possible fate in each region.
For economic allocation, orchard EoL biomass is assigned a value of zero, because growers manage trees at a cost, rather than generating income. For post-harvest co-products, county-specific 5-year mean almond hull and kernel prices were obtained from the county crop commission reports for California, via the US Department of Agriculture (USDA) National Agricultural Statistics Service (NASS) database (USDA NASS 2019). The price of shell as bedding in 2019 was obtained from the UC Davis Animal Science Department and used in lieu of a mean commodity value. Hash and twig are included respectively within hull and shell production and price estimates, as specific commodity values for these co-products were not available (Table 3).
Using the substitution approach, orchard EoL biomass used as power plant feedstock (Table 2) are assumed to substitute for average electricity in the California grid mix. An average rather than marginal approach is used here because biomass power plants have long served baseload power functions on the California grid. The substitution calculation is conducted by taking the reference LCI for California electricity obtained from Gabi ts v6 (PE International 2019), minus the impacts of emissions from biomass combustion. For substitution calculations of post-harvest co-products, almond hull is treated as a dairy cow feed input, similar in nutritional characteristics to alfalfa (Robinson 2015). Thus, alfalfa is assumed to be substituted by almond hulls. Data on the nutritional value of almond hash as a dairy feed was not available, so this co-product was treated identically to hulls.
Reference LCI datasets
LCI data were obtained from ecoinvent v3.3 and Professional databases accessed via GaBi ts v6 (PE International 2019). We used a cutoff system model, the default provided by GaBi ts. The reference LCI datasets and their sources are described in Tables S1–S6 of the supplementary material. Where an exact match between a flow and reference LCI was not available, the most appropriate available substitute was used or a new LCI was generated. Where regionally specific LCI datasets were not available, LCI data for the same material or process from the most similar region available were used. Wherever possible, transformation process LCI datasets were used. Where the only available option for a particular input LCI was a market process dataset, explicit freight transport modeling was foregone in order to avoid double-counting transport flows.
A process-based LCI for almond pollination services was generated using the SPARCS-LCA model framework and data from an earlier study of GHG and air emissions of US honey production (Kendall et al. 2013), making use of updated LCI datasets (Table S2 of the supplementary information) accessible through GaBi ts. For quantification of avoided impacts due to displacement of dairy feed, a process-based LCI of alfalfa production in California was generated by parameterizing the SPARCS-LCA model framework with input and operation data (Table S3) obtained from UC Davis cost and return studies on alfalfa production (Clark et al. 2016; Long et al. 2015) and USDA Cropscape geospatial data (USDA NASS 2019). The full LCI datasets so produced are available as supplemental spreadsheets.
Emission factors for combustion and soil emission
Emission factors (EFs) for biomass combustion in bioenergy facilities were generated from data obtained from the CARB pollution mapping tool on annual emissions by facility (CARB 2019) and feedstock quantity. EFs for in-field burning of almond biomass (Table S7 of the supplementary information) were obtained from CARB air emissions reports (Jenkins et al. 1996). Soil nitrous oxide (N2O) EFs sensitive to region, irrigation system, and fertilizer nitrogen source were obtained from an analysis that used the denitrification-decomposition (DNDC) model (Deng and Salas 2017), a geochemical process-based model, to assess GHG emissions and carbon storage in California’s agricultural soils (Table 1). Biogenic CO2 emissions, such as those from biomass combustion and decomposition of soil organic matter, were treated as carbon neutral for purposes of Global Warming Potential calculation.
Scenario and sensitivity analysis
The influence of individual parameters on model results was quantified through sensitivity analysis in which model parameters relevant to biomass co-product fate and irrigation were varied systematically, with results expressed as percent change in model output vs percent change in parameter value. These two key processes were identified based on findings from an earlier LCA of California almonds showing that biomass co-product fate and irrigation were the most sensitive and influential processes in determining the impact of almonds (Marvinney et al. 2015). Parameter sensitivity values and response equations are recorded in Tables S8–S10 of the supplementary material associated with this article.
The effects of reduction in applied irrigation water on yield are modeled using data from recent research efforts establishing a water production function in California almond production (Goldhamer and Fereres 2017). No recent California-specific analysis for the effects of water application on biomass accumulation was available, so a biomass productivity function for almond was derived from analysis of irrigated almond production in Spain (Egea et al. 2010).
In addition to sensitivity analysis, potential orchard inputs and management practices are assessed using scenario analysis, to allow comparison of regional BaU practices with results for a hypothetical hectare of almond orchard under different management assumptions for EoL, annual, and postharvest biomass co-product fate (in-field burn, energy feedstock, surface mulch, or soil amendment), as well as irrigation system, pump, and water source. Scenarios affecting tree number or yield (e.g., planting density, replant rate, orchard EoL age, deficit irrigation) are assessed with explicit accounting for yield and biomass changes. The specific assumptions and model parameter changes used in each scenario are shown in Table S12 of the supplementary material.