Agricultural input shocks affect crop yields more in the high-yielding areas of the world

Global food production increasingly relies on industrialized systems dependent on off-farm inputs like fertilizers, machinery, energy, pesticides, seeds, and animal feed. Many of these are imported, often from a limited number of countries, making food production vulnerable to disruptions in input availability. Simultaneously, there's growing pressure to reduce synthetic fertilizer and pesticide use for environmental reasons. However, the combined global impact of agricultural input shocks on crop yields hasn't been systematically assessed. This study aims to fill this gap by modeling the impacts of combined agricultural input shocks on crop yields globally, mapping vulnerable areas and crops, and identifying the most impactful input shocks for each crop type. This information is crucial for understanding and enhancing national and global food security in times of geopolitical instability.

While the effects of extreme weather and trade disruptions on food production are relatively well-understood, the impact of reduced agricultural input availability is less clear. Existing studies offer limited geographical scope and consider fewer inputs. For example, Beckman et al. (2020) used economic models to study the effects of EU Green Deal strategies, predicting significant wheat production decreases in the EU. Jansik et al. (2021) used expert interviews to assess the impact on Finnish agriculture, estimating yield reductions of 10–40% from a total input shock. This study aims for a broader, more comprehensive assessment.

This research employed a random forest machine learning model to assess the impact of agricultural input shocks on 12 globally important food crops at a 5 arcmin resolution (approximately 10 km at the equator). The model estimated the impact of various shock levels (25%, 50%, 75%) and combinations (nitrogen, phosphorus, potassium fertilizers individually; machinery, pesticides, all fertilizers together, and all inputs together). The globe was divided into 25 climate bins to control for climate effects, and soil parameters (soil P, soil N, and soil organic carbon) were included. Model performance was evaluated using the Nash-Sutcliffe efficiency (NSE), with most models showing good (NSE 0.65–0.75) or very good (NSE > 0.75) performance. Baseline yields were validated against FAOSTAT data, showing very good agreement (R² > 0.85). A control scenario (all inputs set to zero) further confirmed model robustness. The model predicts the shock scenario yields by selecting observations in the same climate bin where input use is in the baseline similar to scenario use. Decreased scenario shock yields indicate that within the climate bin in question, the baseline yields were only attainable with original input values. Increased yields after scenario shocks mean that in the same climate bin, similar or better yields are possible with less commercial agricultural inputs. The scenario results were then converted from yield to production volumes by multiplying crop-specific yields (tonnes per hectare) with harvested area (ha) for each shock scenario. An R Shiny application, Agri.Input.Shock-explorer, was developed to visualize and explore the extensive results.

The validated model revealed that areas with the highest initial yields suffered the most from input shocks. Across all crops and climate bins, larger shock scenarios led to greater yield decreases, as expected. For wheat, for example, shocks in phosphorus and pesticides showed smaller yield declines compared to nitrogen, potassium, machinery, combined fertilizer, and all-input shocks. A 50% nitrogen shock significantly reduced wheat yields in Central Europe, parts of North America, and areas in Southern Africa, China, and India. A combined 50% shock across all inputs caused more widespread and severe yield reductions. The impacts varied across crops; no areas showed yield declines for all 12 crops. Interestingly, some areas (mostly in sub-Saharan Africa and South Asia) experienced yield increases after shocks, potentially indicating yield gap areas where better yields are achievable with reduced inputs. Examining the impacts by climate bins revealed differences in how shocks affect yields; for maize, certain temperate climate bins showed strong responses to phosphorus shocks, while others were more sensitive to potassium or machinery shocks. Pesticide shocks had minimal effects globally but showed moderate effects in Finland in the 75% shock scenario. A 50% shock in all inputs resulted in global production decreases exceeding 25% for maize and 20% for wheat. The United States experienced the largest absolute production decline (-140 million tonnes), followed by Brazil (-114 million tonnes). Many African countries and others (like Finland and the Baltics) showed relatively smaller decreases. Production decreases in major exporting countries were substantial and would likely disrupt food security in import-dependent regions.

The complex relationship between agricultural inputs and crop yields necessitates a sophisticated modeling approach. The random forest model effectively captured nonlinear relationships and predictor interactions. This study's findings align with some existing research, such as Beckman et al. (2020) concerning EU wheat production. However, this study provides a much broader geographical scope and considers a more comprehensive range of inputs than previous studies. The identification of yield gap areas highlights the potential for improved yields in certain regions with more balanced input use. While the model shows some areas experiencing yield increases after input shocks, this might be due to the algorithm's limitations in extrapolating beyond the training data or the limited detail of data in low-yielding areas. It's important to note that the model doesn't capture long-term impacts from, for instance, soil nutrient depletion. Future research could integrate these factors.

This study demonstrates that agricultural input shocks would severely impact high-yielding regions, threatening global food security. While no single input is universally most influential, synthetic fertilizers play a major role. To enhance food system resilience, regions heavily reliant on synthetic fertilizers, especially imported ones, should transition toward more sustainable, locally sourced alternatives. This research provides a valuable tool for regional food security assessment and risk identification, emphasizing the importance of considering agricultural input availability in building a more resilient and sustainable food system. Future studies should incorporate additional factors such as seed availability and conduct a more detailed economic analysis.

The primary data used were averages around the year 2000, and the model assumes that the relationships between inputs and yields haven't significantly changed since then. Data limitations in some areas may influence model accuracy, particularly regarding the detail of agricultural input data, which is known only at subnational or county level. The model doesn't capture long-term effects, like soil nutrient depletion, or the complex economic and trade dynamics influencing input availability. A full-scale uncertainty analysis wasn't feasible due to the lack of uncertainty measures in many datasets.