The research, conducted by Xinrui Gao and Yangyang Zheng of Wenzhou University and Rumiao Li of Zhejiang Financial College, breaks with a long tradition in agricultural economics. Most previous models of technology adoption treated farmers as independent decision-makers interacting randomly with one another, an assumption known as homogeneous mixing. The authors argue this misses the point. In real rural communities, information flows through dense webs of kinship, geography, and professional ties, and a handful of well-connected, influential growers—large-scale farmers, cooperative leaders—sit at the hubs of these networks while most ordinary farmers have far fewer connections.

To capture this reality, the team built their model on a scale-free complex network, the kind of topology that characterizes everything from the internet to social media, where a few nodes accumulate enormous numbers of links. Onto this network they layered an evolutionary game: each farmer, boundedly rational and working with incomplete information, repeatedly decides whether to adopt or reject a digital technology—in this case, integrated water and fertilizer management, a precision agriculture technique highlighted in China’s 2025 Central Document No. 1. Farmers compare their payoffs with a randomly chosen neighbor and switch strategies with a probability governed by Fermi dynamics, the same rule physicists use to model cooperation in spatial games.

The payoff structure is deliberately rich. Adopters bear acquisition and learning costs but gain yield increases, government subsidies for both equipment purchase and operating expenses, and economies of scale when neighboring adopters coordinate. Non-adopters avoid risk and transition costs but remain locked into lower-output traditional methods. The model also encodes three psychological and informational variables: the intensity of information benefits (the inverse of the information gap), farmers’ risk perception, and the technology’s yield-increase multiplier. Parameter values were triangulated from field surveys of grape vineyards in Pujiang County, Zhejiang Province—where fertigation visibly improved grape quality—alongside prior literature and a structured consultation with eight agricultural economics experts.

Simulating networks of 50, 200, and 500 nodes over 100 game iterations in MATLAB, the researchers found that network scale itself matters profoundly. In the small 50-node network, efficient information transmission produces rapid convergence: under favorable parameters, complete adoption is achieved within roughly ten iterations. But in larger networks, heterogeneity dilutes the influence of central hubs, delaying convergence as path complexity and hub overload slow the cascade of adoption. Small networks, the authors suggest, are ideal testing grounds for pilot programs precisely because policy interventions there are direct—though unstable—while large systems demand more patient, structurally aware strategies.

The most provocative result concerns subsidies. Equipment purchase subsidies do accelerate initial adoption by lowering entry barriers, and higher subsidies raise both the diffusion rate and the peak adoption probability. But the curves tell a cautionary tale: under high subsidies, adoption rates decline sharply after peaking, and in the 50-node network farmers eventually abandon the technology altogether, with steady-state values diverging significantly across subsidy levels. In other words, acquisition subsidies buy a short-term burst of adoption, not sustained diffusion. The authors attribute this to the way subsidies lower the threshold for trying a technology without addressing the deeper market failures—information asymmetry, risk, and cost—that determine whether farmers keep using it.

Operating cost subsidies behave differently. They do not dramatically raise the peak of adoption, but they accelerate stabilization after the initial rise, reducing curve volatility to under five percent. By continuously sharing ongoing costs, they relieve long-term cash-flow pressure and enhance what the authors call technology stickiness. Yet even here there is a threshold effect: very high operational subsidies trigger a “delayed response” across all network sizes, and persistent reliance on subsidies may erode the technology’s endogenous competitiveness. The policy contradiction is clear—subsidies are necessary but structurally insufficient, and one-size-fits-all, long-term high subsidies risk creating dependency rather than self-sustaining diffusion.

What does sustain adoption, according to the simulations, is genuine technological advantage. In medium and large networks, a higher yield-increase multiplier systematically raises the steady-state adoption probability, and the effect exhibits a “rich-get-richer” Matthew dynamic: the better the technology performs, the more it reinforces itself through network effects. The authors frame this as an endogenization process, in which farmers accumulate human capital that converts an exogenous technology into an endogenous growth engine. But the advantage must dynamically match farmers’ actual needs; a mismatched technology, however impressive on paper, is eventually abandoned for poor applicability.

On the barrier side, the information gap emerges as the most insidious force. When the gap is large, adoption curves fluctuate wildly in small networks as boundedly rational farmers oscillate between trial-and-error and blind imitation of neighbors. In the 500-node network, an extreme information gap drives late-stage adoption probability to a negligible fraction of the high-information scenario—a “marginal exclusion effect” in which remote, information-poor farmers are permanently shut out of the technology. Risk perception compounds this: drawing on expected utility theory and prospect theory, the authors show that cognitive discounting of uncertain gains and loss aversion over potential losses suppress expected utility, making high-risk-perception farmers especially resistant in large networks where dependence on agricultural income is high. Technology cost, by contrast, acts directly: low-cost conditions produce fast, stable adoption, while high costs suppress both willingness and speed—consistent with field data showing adoption rates of expensive technologies among smallholders averaging only 15 to 20 percent, versus over 50 percent for cost-reduced alternatives.

The practical implications are pointed. Governments should sequence subsidies across the technology lifecycle: front-load equipment subsidies to activate influential hub nodes such as large growers and cooperatives, then shift toward operational support to prevent attrition, while avoiding indiscriminate long-term high subsidies. Extension agencies should attack the information gap directly with standardized, visually demonstrable performance evaluations and regional demonstration bases built on “seeing is believing” field trials. Farmers themselves, the authors argue, should organize into cooperatives to share upfront costs and trial-and-error risks, and invest in digital literacy to shift from experience-based to data-driven decisions. The study’s authors acknowledge their framework’s limits—it is simulation-based, focuses on farmer-to-farmer interaction rather than the full ecosystem of governments and agribusinesses, and models a single technology type. But its core message stands: digital agriculture will not spread by checkbook alone. It spreads through networks, and policy that ignores the topology of rural social life will keep losing the last mile.

Subject of Research: Evolutionary game modeling of digital agricultural technology adoption on scale-free farmer networks

Article Title: Towards digital agriculture: uncovering the drivers and barriers to the adoption of digital agricultural technologies

Article References: Gao, X., Zheng, Y., & Li, R. (2026). Towards digital agriculture: uncovering the drivers and barriers to the adoption of digital agricultural technologies. Frontiers in Sustainable Food Systems, 10, Article 1894648. https://doi.org/10.3389/fsufs.2026.1894648

Image Credits: AI Generated

DOI: 10.3389/fsufs.2026.1894648

Keywords: digital agriculture, technology adoption, evolutionary game theory, scale-free networks, government subsidies, information gap, risk perception, smallholder farmers, precision agriculture, complex networks, technology diffusion, agricultural economics