The research stands out because it moves beyond case studies and anecdotal accounts of smart farming, which dominate much of the popular discussion about AI in agriculture. Instead, it treats artificial intelligence as a measurable economic input, proxied by AI patent activity, and asks whether innovation in this domain is statistically linked to the output of the food production sector over time. By framing the question this way, the study connects two literatures that rarely speak to each other: the economics of technological innovation and the empirical analysis of food systems. The result is one of the first time-series investigations to quantify the AI-food production relationship for an African economy using a nationally representative, multi-decade dataset.

Methodologically, the study employs the autoregressive distributed lag bounds-testing approach, a widely used econometric technique for detecting long-run relationships between variables that may behave quite differently in the short term. The bounds test confirmed that a long-run equilibrium relationship exists among food production, AI patents, inflation, population growth, and household income. This is a critical first step in such analyses: without evidence of a stable long-run relationship, any estimated coefficients could be statistical artifacts rather than meaningful economic signals. The confirmation of cointegration among these variables means that, despite year-to-year fluctuations driven by droughts, policy shifts, and global market shocks, food production in South Africa moves together with these determinants over the long haul.

Once the long-run relationship was established, the study estimated how each factor contributes to food production. The headline finding is that artificial intelligence carries a positive and statistically significant coefficient in both the short-run and long-run models. In practical terms, periods of heightened AI innovation in the dataset are associated with subsequent increases in food production, and this association persists when the analysis is extended over longer horizons. Yet the author is careful to note that the magnitude of the effect is modest. AI is not a substitute for the fundamental drivers of agricultural output; rather, it appears to act as a supporting factor whose benefits depend on the surrounding economic and institutional environment.

The inclusion of control variables sharpens the picture. Inflation and population growth, both of which shape the demand and cost conditions facing the agricultural sector, were incorporated into the models alongside household income, which captures the purchasing power of consumers. Their presence in the confirmed long-run relationship underscores that food production is embedded in a broader macroeconomic system. Rising incomes can shift demand toward more resource-intensive foods, population growth expands the consumer base, and inflation affects input costs such as fertilizer, fuel, and machinery. Against this backdrop, the persistent positive AI signal suggests that technological innovation exerts an influence that is distinguishable from these conventional economic forces.

One of the most timely aspects of the study is its treatment of the COVID-19 pandemic. The author re-estimated the models with a dummy variable capturing the pandemic period, a standard robustness technique for checking whether an extraordinary shock is distorting the results. The findings held: the AI coefficient remained positive even after accounting for the pandemic, while the dummy itself revealed that COVID-19 had a negative effect on food production. This dual result is reassuring for the credibility of the AI-food production link and simultaneously documents the disruptive footprint of the pandemic on South Africa’s food system, likely reflecting lockdown-related labor disruptions, supply chain bottlenecks, and reduced economic activity.

The modest size of the AI effect is arguably the study’s most policy-relevant insight. The author interprets it as evidence that AI’s contribution to food production is conditional: it depends on complementary infrastructure and adoption capacity. In other words, patents and algorithms do not raise yields by themselves. Their value is realized only when farms have access to reliable digital connectivity and electricity, when agricultural data systems exist to feed models with usable information, and when farmers possess the technical skills to deploy AI-enabled tools effectively. Where these complements are missing, as is often the case for smaller producers, the gap between innovation on paper and impact in the field can be wide.

This conditionality points directly to a policy agenda. The study recommends that policymakers prioritize reliable digital and energy infrastructure, the development of agricultural data systems, investment in technical skills, and affordable access to AI-enabled technologies, with particular attention to smaller producers who are least likely to capture new technologies on their own. South Africa’s agricultural sector is characterized by a dual structure in which large commercial operations coexist with smallholder farmers, and the benefits of any general-purpose technology tend to accrue first to those already well resourced. Deliberate policies to lower the cost and skill barriers facing smaller producers could therefore determine whether AI narrows or widens existing productivity gaps in the food system.

The research also carries implications for how innovation is measured and encouraged. By using AI patents as the indicator of artificial intelligence activity, the study links food production outcomes to formal inventive output, suggesting that intellectual property generation in AI is a meaningful signal of technological capacity relevant to agriculture. For a middle-income economy such as South Africa, this raises questions about how to stimulate domestic AI innovation and ensure it is directed toward sectoral priorities like food security. It also highlights the value of long-run empirical evidence in an area where enthusiasm often outpaces rigorous measurement; the study’s finding that AI matters, but modestly, offers a calibrated baseline against which future adoption trends can be judged.

As climate variability, population pressure, and economic shocks continue to test food systems across the African continent, evidence of this kind is likely to become increasingly valuable. The study’s central message is one of grounded optimism: artificial intelligence is not a silver bullet for food production challenges, but it is a genuine and statistically detectable contributor that can support domestic output when the enabling conditions are in place. For South Africa, the task ahead is less about proving that AI can help agriculture and more about building the infrastructure, data ecosystems, and human capital that allow a modest statistical association to grow into a transformative one. The research, published open access in Discover Sustainability, was conducted without external financial support, and the author declares no competing interests.

Subject of Research: The econometric relationship between artificial intelligence and food production in South Africa

Article Title: The relationship between artificial intelligence and food production in South Africa

Article References: Giwa, F. (2026). The relationship between artificial intelligence and food production in South Africa. Discover Sustainability. https://doi.org/10.1007/s43621-026-04990-0

Image Credits: AI Generated

DOI: 10.1007/s43621-026-04990-0

Keywords: artificial intelligence, food production, South Africa, agriculture, food security, autoregressive distributed lag, AI patents, agricultural economics, Discover Sustainability, COVID-19, digital infrastructure, smallholder farmers

The research, conducted by Malihe Erfani of the University of Zabol and Fatemeh Jahanishakib of the University of Birjand in Iran, introduces what the authors call a cyclical-incremental approach, abbreviated CIA. Rather than treating environmental assessment as a one-off exercise completed before construction begins, the framework builds strong feedback loops connecting the early steps of EIA and ecosystem service valuation. The core idea is that assessment should not be a single snapshot but a cycle: an initial round of valuation is followed by monitoring, and the resulting data feed back into refined estimates, with the process repeating incrementally over the life of a project. According to the authors, this design is intended to improve the realism and objectivity of EIA through at least two assessment rounds, allowing predictions made before a project is built to be checked, corrected, and sharpened against what actually happens on the ground.

The conceptual move at the heart of the framework is deceptively simple but consequential. Ecosystem service valuation has long been promoted as a complement to environmental decision-making, capable of expressing ecological change in units that economists and finance ministries understand. But in practice, the two streams of analysis—EIA on one side, monetary valuation on the other—have often run in parallel without truly informing each other. Earlier scholarship has noted that EIA documents frequently fail to infer meaningful conclusions about ecosystem services, and that scientific shortcomings in environmental impact statements are common internationally. By embedding valuation directly within the assessment cycle, the new approach forces the environmental dimension of cost-benefit analysis to carry quantified ecological costs and benefits, making trade-offs explicit rather than leaving them buried in technical appendices.

To test the framework, the researchers applied its first round to a real and environmentally sensitive case: a sulfuric acid plant at the Sarcheshmeh Copper Complex in Kerman province, Iran. Sarcheshmeh is one of the country’s major copper operations, and sulfuric acid production is closely tied to smelting activities, with associated concerns about air quality in the surrounding industrial region. The team selected five ecosystem services for monetary assessment: habitat quality, carbon sequestration, oxygen production, food production, and forage production. These span the major categories ecologists use—provisioning services such as food and forage that people directly harvest, and regulating and supporting services such as carbon storage, oxygen generation, and the maintenance of habitat that underpin biodiversity.

The results of the first-round cost-benefit analysis were stark. Across the five assessed ecosystem services, no positive benefit was identified. The largest annual financial loss was attached to carbon sequestration, estimated at 72,877 dollars per year, followed by food production at 55,277 dollars per year. Taken together, the total annual financial loss across the five services came to 155,109 dollars, with no corresponding benefit found among the services evaluated in this initial application. The pattern is notable because it shows that both provisioning services, which supply tangible products, and supporting or regulating services, which maintain the broader ecological machinery, suffered the greatest damage. In other words, the plant’s footprint was not confined to a single category of nature’s output but eroded the landscape’s productivity and its climate-regulating capacity simultaneously.

For decision-makers, numbers of this kind change the character of the conversation. A traditional EIA might report that emissions are expected to affect vegetation or that land clearing will reduce habitat, leaving officials to weigh those statements against projections of jobs, output, and revenue. The framework’s output instead places a dollar figure on the environmental side of the ledger, directly comparable with the economic side of a cost-benefit analysis. If a project’s expected financial gains are smaller than the quantified annual losses to ecosystem services, the imbalance becomes visible in the very analysis that approval decisions rely on. The authors argue that these outputs can assist decision-makers by making the environmental costs and benefits of development projects explicit and quantifiable, a step toward what the study’s keywords describe as win-win environmental policies and honest accounting of trade-offs.

The methodological lineage of the approach is worth appreciating. Cost-benefit analysis and environmental impact assessment have been discussed together since at least 1990, when researchers examined how the two tools might be reconciled, and subsequent work has explored monetary valuation of ecosystem services under different future management scenarios, including studies of forest ecosystems facing environmental change. Other efforts have integrated ecosystem services into impact assessment through analytic and deliberative methods, or applied valuation at local scales in data-scarce regions. What distinguishes the cyclical-incremental approach is its insistence on iteration grounded in monitoring data. Ex-ante assessment—the prediction made before a project exists—is inherently uncertain, and the framework treats that uncertainty not as a flaw to be hidden but as a starting point to be progressively corrected through ex-post rounds that validate and refine the initial estimates.

That iterative structure also addresses a long-standing weakness in the EIA literature: follow-up. Scholars have documented that assessments often end once permits are granted, with little systematic checking of whether predicted impacts materialized, a gap that adaptive management frameworks have tried to fill. By requiring at least two rounds and building feedback loops back to the earliest steps of the process, the cyclical-incremental approach institutionalizes follow-up as part of the valuation itself. Monitoring data collected after a project begins operating can reveal whether the predicted losses to carbon storage or food production were overestimated or underestimated, and the valuation can be recalculated accordingly. Over successive cycles, the estimates converge on reality, and the cost-benefit analysis that informs ongoing management decisions becomes progressively more trustworthy.

The Iranian case study also illustrates the practical challenges of applying such methods in data-scarce settings. Valuing services like carbon sequestration and oxygen production requires biophysical models of vegetation and land cover, while food and forage production depend on agricultural and rangeland data, and habitat quality demands ecological assessment of the affected area. The authors draw on prior economic valuation work in Kerman province and on spatial analysis techniques, reflecting a growing body of research that uses remote sensing and geographic information systems to characterize atmospheric and land-use conditions around industrial sites in the region. The study was carried out partly under a research project supervised by Iran’s Department of Environment, with support from the University of Zabol, indicating institutional engagement at the level of the national environmental authority.

The authors are careful to frame the published results as the first round of a longer process, noting that subsequent rounds using monitoring data are required to validate and progressively refine the application. Even so, the demonstration carries weight beyond a single acid plant in Kerman. Mining and heavy industry worldwide face intensifying scrutiny over their effects on ecosystem services, with studies from open-pit coal mines in semi-arid regions to mining landscapes in Ghana documenting substantial losses in the economic value of nature’s outputs. As governments and lenders increasingly demand that environmental costs be internalized in project economics, frameworks that can produce defensible, repeatable, and monetized estimates of ecological change are likely to attract attention. The cyclical-incremental approach offers a template: quantify the services at stake, price the changes, feed the results into cost-benefit analysis, and then return after the machinery starts running to check the math against the living world. If the approach performs as designed across future rounds, the environmental ledger may finally begin to balance against the financial one.

Subject of Research: Integrating ecosystem service valuation into environmental impact assessment and cost-benefit analysis for development projects

Article Title: Integrating Environmental Impact Assessment and Ecosystem Service Evaluation in Cost-Benefit Analysis: A Cyclical-Incremental Approach

Article References: Erfani, M., & Jahanishakib, F. (2026). Integrating Environmental Impact Assessment and Ecosystem Service Evaluation in Cost-Benefit Analysis: A Cyclical-Incremental Approach. Environmental Management, 76(9), Article 316. https://doi.org/10.1007/s00267-026-02628-5

Image Credits: AI Generated

DOI: 10.1007/s00267-026-02628-5

Keywords: environmental impact assessment, ecosystem services, cost-benefit analysis, monetary valuation, carbon sequestration, habitat quality, food production, mining, Iran, environmental management, EIA follow-up, sustainability