We employed an integrated assessment methodology based on crop, economic, and LCA modeling to investigate climate change adaptation and mitigation scenarios for processing potato and processing tomato supply chains, starting with current conditions through the year 2050.
Selection of representative counties for modeling
Crop Reporting Districts (CRDs) were selected by first sorting them in a descending manner by total crop area for eight fruit and vegetable crops (including potatoes and tomatoes) that are targeted in the broader project comprising this study. For more detail see the Supplementary Information (SI). We then included the CRDs necessary to capture 80% of the total production area for the crops, resulting in a list of 31 CRDs. The counties having the highest target crop production area within each of these CRDs were then selected for the crop modeling, with one additional county (St Johns, Florida), added in order to better represent potatoes in that state).
Selection of processing varieties for modeling
The crop modeling was based only on processing varieties and excluded fresh market varieties. Crop varieties for processing have more homogeneous growth patterns and harvest periods while varieties for fresh markets are extremely variable in terms of growing season, shape, color, and yield to adapt to specific markets, which makes them more difficult to parameterize for modeling purposes.
Estimation of yield impacts from climate change
We utilized a multi-model approach based on AgMIP protocols32 to estimate changes in yield, irrigation water requirements, and crop nutrient requirements (N, P, K) in all cropping areas of interest through 2050. Climate change impacts by 2050 on potatoes in 32 main potato growing districts in the US were estimated with an ensemble of five process-based models (SIMPLE33, CropSyst34,35, LINTUL-POTATO-DSS36, EPIC37,38, and DSSAT-Substor-Potato39,40,41), and one statistical model42. Three crop models (SIMPLE, CropSyst, and DSSAT CSM-CROPGRO-tomato43) and a statistical model were used to estimate the impact of climate change on tomatoes in eight main tomato districts for processing tomatoes across the US. National or regional level impacts were derived from district averages by weighting the corresponding crop areas. The crop models were calibrated to field-experimental-based-corrected district yields44 for potatoes45. Due to the lack of tomato data, the tomato models used previous cultivar calibrations46. The statistical model was trained on data from the USDA NASS dataset47. Crop and statistical model estimates used gridded downscaled48 daily weather data (4 km x 4 km) for a baseline (1981-2010)49 and two future time slices (2021-2050 and 2041-2070) from five general circulation models, GCMs48,49 for a Representative Concentration Pathway 8.513. No water or nitrogen limitations were assumed in the potatoes and tomato cropping systems. As a possible adaptation to a warmer climate, an earlier planting date was considered. Nitrogen, phosphorus and potassium fertilizer demand was calculated after the crop simulations based on simulated yield and nutrient concentrations50,51 and their changes with elevated atmospheric CO2 concentrations17. The simulated baseline yields were bias-corrected to the regression yield for 2017 based on CRD yields for each CRD (see crop modeling protocol45 for more details).
Future climate scenarios
Higher future atmospheric CO2 concentrations will stimulate growth if other nutrients are not limiting52. The yearly changing atmospheric CO2 concentrations for the baseline (1981-2010) and future periods (2030s and 2050s) under the RCP 8.5 scenario were applied25. Five GCM’s were used (GFDL-ESM2M, HadGEM2-ES365, IPSL-CM5A-LR, MIROC-ESM-CHEM, NorESM1-M), consistent with ISIMIP – the Inter-Sectoral Impact Model Intercomparison Project53.
Yield projection method
The projected future yields included climate change impacts (temperature and CO2 change), the effect of earlier planting as an adaptation and a projected technology trend on yield improvement. The technology trend is a combination of improved seeds, more effective use of fertilizer, water, and various inputs, as well as better equipment and other improvements. A step-wise process was utilized. First, a regression line was fitted to the observed yield trends for each CRD (based on USDA NASS), with the slope of this line assumed to have two linear components: technology and climate: The technology component was determined as the difference in slopes between the overall observed trend and the simulated baseline trend due to climate change during that same time period. The technology component observed in the past was then attenuated to 90% by 2030 and 70% by the year 2050, causing the partial flattening of the yield curves over time15. The climate component was determined based on the percentage linear increase in simulated crop yield from the baseline period through the 2030s and then removed from the observed historical yield trend (to create a climate-corrected technology trend). The overall future yield trend was constructed from the simulated climate change effects with the attenuated technology trend added. In order to characterize overall modeling uncertainty, the same yield projection methodology was applied to the 25th- and 75th-percentile ensemble results, in addition to the ensemble median, which was treated as the best single estimate of future yield.
Economic modeling overview
Structural partial equilibrium models for the US fresh and processed potato and tomato were developed to simulate the impact of climate variation and mitigation practices on crop net returns and land use change. In order to capture the geographically detailed output from the crop yield simulation models, area, yield, and production equations were developed for the 31 US CRD’s with one additional region to capture the remainder of the US. Each area equation is driven by the ratio of gross market returns with the cost of production for the crop they are producing and the previous year’s planted area (otherwise known as the “lagged planted area”). Based on input from agricultural extension personnel and growers regarding production practices, processors and growers select specific varieties of potatoes and tomatoes depending on the end use of the product. Therefore, substitution between the processed and fresh sectors is very limited or nonexistent. The implication for economic models is that crops produced for the fresh sector are generally considered a different commodity than the same crop produced for the processing sector. The area of specialty crops like potato and tomato is often linked closely to the number of contracts offered by a processor rather than being driven by competing crop returns. The inclusion of lagged planted area in the economic model reflects a short-term constraint against significantly changing processing capacity with processors preferring to operate their facilities at optimal capacity54.
Demand equations were developed at the national level based on the fresh and processed demand as detailed in USDA’s Economic Research Service datasets. Whether processed or fresh, future consumer demand for potatoes and tomatoes was driven by inflation-adjusted income, population, overall consumption trends (reflecting tastes and preferences), and inflation-adjusted price.
International partial equilibrium economic potato and tomato models focused on the primary US trading partner countries were also developed and used as part of the modeling system employed in this study (see SI for details). Because climate impacts on crops and yield vary by region, it was important to determine if trade would be affected by the climate impacts on other trading partners55. The international models and US models were combined to form a set of global models. The global partial equilibrium models solve simultaneously for the set of crop prices that balance the global supply and demand of each commodity.
Land use change
Although they are the two vegetable crops with the largest total cropped area in the US, the area devoted to potatoes and tomatoes is still very small in comparison with major row crops (e.g., maize, soybeans, etc.). Typically, returns per hectare for vegetables are significantly higher than row crop returns, and combined with their relatively small area footprint and rotational requirements they do not significantly compete with row crops for area. However, in some cases they do compete for irrigation water. Although the specialty crop returns usually exceed row crop returns substantially, row crop producers with water rights may choose to continue their operations rather than leasing those rights to specialty crop producers. Therefore, the modeling system also incorporated WAEES existing models of global row crops (see SI). While the economic models constrain total irrigated area to the existing CRD irrigated area, the constraint was not found to be significantly binding in this analysis.
One important consideration with respect to land use for both potatoes and tomatoes is the presence of shared soil borne diseases (nematodes) for these Solanaceae crops. Neither crop can typically be cultivated on the same land within a period of 4 years. The need for such a lengthy crop rotation may put a burden on land cultivated by vegetable growers when demand increases, where, just as with the irrigation, farmers growing row crops may not be that forthcoming to share land with vegetable farmers.
Data limitations
The crop modeling teams noted that there was significant variation in climate impacts on yield within individual states necessitating the use of sub-state production regions. Both county and CRD’s were evaluated as possible geographies, but ultimately CRD’s were chosen due to more complete data sets. Data on the production of US fruits and vegetables by CRD primarily relies on the five-year agricultural censuses with many missing data points because of USDA disclosure rules. To the extent that NASS reported historical annual CRD data, this was used in the analysis. When needed, interpolation between the census years was done by aligning the sum of the CRD data with the annual NASS data reported at the state level. In order to estimate the supply elasticities, historical time series of area, yield, and production for each CRD over the 2000 to 2017 period were assembled.
Interdisciplinary data exchange among the modeling teams
The economic modeling drew on information from across the interdisciplinary teams. The crop model teams provided yield impacts with and without adaptation for US potatoes and tomatoes for the 31 CRD’s under the RCP 8.5 GHG emissions scenario. The extension teams provided insight into crop production practices, input use and costs of production. The yield impacts under the RCP 8.5 scenario on regions outside the US and/or for crops other than potatoes and tomatoes were provided by the IFPRI IMPACT model56,57. Finally, technical parameters such as fruit and vegetable water content were provided by the LCA team. Outputs on the current and projected levels of input use, realized yield, processing use, technology trend and land use change were reported to the other modeling teams, as needed.
Life Cycle Assessment (LCA)
The life cycle assessment methodology included the development of life cycle inventories (LCIs) of the supply chains for two processed products made from potatoes and tomatoes: frozen French fries and pasta sauce. This work was governed by a protocol that fully describes the “cradle to grave” approach (see SI for details). An integrated supply chain model was constructed to account for all major raw materials needed at each stage of the supply chain. Data on yield, fertilizer inputs and irrigation were derived using the results provided by the crop and economic modeling teams. The on-farm LCI represents the average farm management and production of each crop reporting district (CRDs). Post-harvest stages include processing plants with some LCI data based on engineering estimates. The protocol also specifies assumptions used for evaluating future crop production scenarios. Mitigation analyses were performed for a full cradle-to-grave system, including farm-to-processer transport, processing and packaging, distribution through retail, consumption and final disposal. The LCA assessment methodology is compliant with ISO standards.58
Life-cycle Inventory modeling
The life-cycle inventory model couples the output from process models of potato and tomato production using a semi-automated workflow to map data into the LCA software. The data were supplemented with and verified against available information from USDA statistical websites, including National Agricultural Statistical Service (NASS), Agricultural Resource Management Survey (ARMS), and Economic Research Service (ERS). Some data were difficult to obtain, and for the processing stages, data were partly based on the processing plant (e.g. for tomato) and on engineering estimates and available literature.
LCA modeling
The model is constructed of three elements: (i) Production; (ii) Postharvest; and (iii) Biowaste-handling (see SI for detail). In brief, the first element characterizes crop production in each CRD45, and the subsequent stages of the supply chain include processing (with warehouse storage of potatoes), retail/supermarket and consumer activities which are modeled to account for material and energy consumption and related emissions. The third element models three alternate methods for biowaste (scraps and food waste) handling.
System boundaries and Functional Unit
A full “cradle-to-grave” perspective (farm to consumer, including waste management) was adopted to define the system boundary. It is common in agricultural LCA, to define the functional unit (FU) as product mass (fresh or dry) or as land occupied (hectare). Although mass is widely used as FU, its appropriateness is debated59, particularly considering the large variation among foods’ characteristics: water and nutritional content, for example. Alternate functional units have been suggested; however, these have not been adopted for this study60–62. Reference flows are the quantitative outputs from processes contained in a product system that are required to deliver the functional unit. (see SI). The defined FU’s for potato and tomato are 1 kg French fries consumed or 1 kg pasta sauce consumed, respectively. Because the FU includes consumption at the consumer stage, the reference flows of the raw crops fully account for the loss fractions at each stage of the supply chain. As an example, to consume 1 kg of frozen fries, 1.22 kg must be purchased based on consumer stage waste of 18%. Ultimately, to deliver the 1 kg of frozen fries, 2.16 kg of raw potato must be produced (see SI for waste fractions along the supply chain).
LCA impact categories and impact assessment methods
ISO 14044:2006 recommends that the choice of impact categories and impact assessment methods be based on the specific requirements of the LCA practitioner to meet the objective of a study63. This study protocol considered three impact categories: global warming potential (GWP100) (in kg CO2 eq), water consumption (m3 eq), and land use (m2-a). These were considered most relevant in the context of resiliency of specialty crop supply chains under climate change scenarios.
Handling of products and co-products
Most production systems generate multiple products with various functions and services. The handling of multifunctionality in LCA requires a choice among different approaches, such as subdividing the multi-functional processes, system expansion or allocation64. This often occurs in the food processing industry, where processing plants are built with multiple processing lines, which generate arrays of products (e.g. raw potato processed to frozen fries, chips, dehydrated products; and tomato to paste, diced, sauces etc.). In such cases, as suggested by others65, physical causal relationships can be applied to distribute the burdens among the multiple products. In this study, it is assumed that the production lines are independent, that is the quantity of frozen fries produced does not affect the quantity produced for potato-chips, when both are manufactured in a single facility. Hence, from the total annual raw materials consumed in an ideal processing plant, the sub-division of raw material inputs to each processing line was estimated based on typical product yields from the facility. For calculating the energy inputs at retail/supermarkets, we relied on data available for shelf space occupied by product category66.
Biowaste treatment scenarios
Estimates of the quantity of waste generated across the supply chain were based on Buzby67, and other sources68–70. Biowaste includes peels and scraps as well as damaged products removed at sorting. In the basic scenario we have considered biowaste management as: composting (on-farm waste), livestock feed (processor and retail waste) and for consumer waste following composting, incineration and land-fill71. The features and assumptions for the alternative biowaste management scenarios are fully described in the SI. Transportation of biowaste to conversion facilities is excluded, considering the high uncertainty of the distances in different CRDs.
Uncertainty Assessment
The mitigation scenarios were compared using a Monte Carlo bootstrap statistics approach72,73. Briefly, 1000 simulations were conducted with a fixed seed for random number generation to provide paired samples for each of the mitigation and baseline models. Subsequently 300 replications of 100 samples with replacement were performed and a distribution of Student-t and associated p-values produced for each pair. If the upper 95th confidence interval p-values was less than 0.01, the null hypothesis that the mean values of the two distributions are equal is rejected.