From Farmer to Farmer: How Climate Knowledge Really Travels

Climate adaptation is not only about giving farmers better information. The people carrying that information, the relationships connecting them, and the boundaries separating farming communities can determine whether useful advice travels a few kilometres or remains trapped within one locality.

A new study, "Social network structure and climate change adaptation information diffusion among dairy farmers: evidence from northern Tunisia," published in Frontiers in Climate, examines this problem among dairy farmers in Tunisia's Bizerte Governorate. The researchers studied a network of 239 people reconstructed from interviews with 60 dairy farmers, asking where farmers obtain information and advice about adapting dairy production to climate change and weather variability.

The findings point to a problem that agricultural extension programmes can easily overlook: communities may already possess useful climate knowledge, yet the social connections needed to move that knowledge between communities can be surprisingly weak.

Climate information is available, but it does not travel evenly

Northern Tunisia's dairy farmers are working in increasingly difficult conditions; rising temperatures, declining rainfall, heat stress, water shortages and changing disease risks make timely local knowledge particularly valuable. Formal extension services cannot always provide that information frequently or widely enough, leaving farmer-to-farmer relationships as an important route for sharing practical adaptation knowledge.

The researchers found a remarkably thin information network: The 239 actors were connected by only 274 observed directed ties, producing a network density of 0.00482. The largest connected group contained 199 people, or 83.3% of the network, while another 40 were spread across eight smaller components.

Access to information was even more uneven: The in-degree Gini coefficient reached 0.797, signalling extreme inequality, while 74.9% of actors received information from no identified source in the observed network. This does not mean one exceptionally powerful farmer controls everything; overall network centralisation was actually low. Instead, many people have very few connections while a relatively small group occupies much stronger information positions.

The distribution was clearly heavy-tailed, meaning a minority of actors had disproportionate informational influence, but the researchers warn against describing it as a proven "scale-free" network. Only nine observations were available in the fitted tail, far too few for a reliable power-law conclusion, and statistical comparisons could not distinguish the power-law model from log-normal or exponential alternatives. That distinction matters because extension programmes should not assume that finding a handful of popular farmers will automatically unlock the whole network.

Geography creates walls around farming knowledge

The strongest barrier uncovered by the study was geography. Farmers were drawn from four areas of Bizerte, Utique, Sejnane, Joumine and Mateur, and their information relationships overwhelmingly remained inside those administrative zones. About 81.4% of observed ties connected people within the same zone, compared with only 26.3% expected under random mixing. Cramér's V of 0.729 showed a very strong relationship between geographic location and network community membership. Mateur had the strongest within-zone concentration at 87.2%, while even the lowest, Utique at 73.5%, remained far above the random expectation.

The network map presented on page 7 makes this fragmentation visible: clusters of farmers form relatively dense local pockets linked by only a limited number of bridging actors. Those bridges matter because information circulating successfully inside one cluster has few natural routes into another. The researchers call this "geographic lock-in." A useful adaptation technique introduced in one part of Bizerte may become well known locally while remaining largely unknown to farmers in a neighbouring area facing similar climate pressures. The study detected 19 communities using the Louvain method, although the authors are careful about interpreting that number. Because relationships between 179 non-interviewed actors were not observed, the network appears sparser and more divided than it might really be. A sensitivity test showed that adding only 10% of possible missing alter-to-alter connections could reduce modularity from about 0.81 to around 0.20. The exact strength of community separation is therefore uncertain, but several independent geographic tests still provide strong evidence that spatial segregation itself is real.

The right farmers can dramatically expand the reach of advice

The study's most useful insights come from separating two ideas that can look similar on paper: being recognised as an information source and being socially positioned to connect otherwise separated groups. A survey asking a farmer, "Who gives you information?" creates a directional relationship. Yet real conversations between farmers are often mutual. Someone who does not appear important when researchers follow only the direction of nominations may still be the person socially connecting two communities.

The researchers therefore analysed a nomination layer, using directed network measures, alongside a social-proximity layer, where relationships were treated as bidirectional. The difference was substantial. Only three actors appeared in both top-10 brokerage rankings, giving an overlap of just 30%. Some important social bridges had zero directed betweenness, meaning a conventional directed analysis could miss them entirely.

That finding became even more important when the team simulated information diffusion. When 28 structural brokers identified through undirected betweenness were selected as starting points, they could reach 128 people, or 53.6% of the entire network, in a single step. Randomly selecting the same number of people reached an average of only 74.5 actors, or 31.2%. After two steps, broker targeting reached 76.6% of the network compared with 65.9% under random selection.

Independent Cascade simulations produced the same basic result. Broker-based seeding outperformed random seeding at transmission probabilities of 0.1, 0.2 and 0.3, although the advantage became smaller as information became easier to transmit.

There is another complication: you cannot reliably identify these brokers from their demographic profiles. Age, education, farming experience, herd size and membership in cooperatives or associations showed no statistically significant relationship with brokerage in this sample. The authors caution that the sample of 60 interviewed farmers limits the ability to detect weaker associations, yet the practical message remains clear — demographic shortcuts are not a dependable substitute for mapping actual relationships.

The brokers themselves also face what the researchers describe as a "structural double bind." They showed somewhat greater exposure to specific information about genetic improvement and herd-management technologies, yet only 43% considered phones useful for sharing adaptation information, compared with 72% of non-brokers. A similar, though statistically weaker, pattern appeared for internet use. Total adaptation-information exposure was not significantly different between the groups, suggesting brokers are not simply better-informed farmers; their importance comes from where they sit in the social network.

For agricultural extension services, that changes the question from "Who should receive information?" to "Who can carry information across the boundaries where it currently stops?"

The authors recommend mapping networks to identify genuine structural brokers, giving those people better communication and institutional support, and combining recognised information sources with cross-community bridges. They also argue for developing additional bridges so the system does not depend on a tiny number of people connecting otherwise separated communities.

The study captures the network at one point in time, relies on self-reported relationships, cannot observe all connections involving non-interviewed farmers, and does not measure whether receiving information actually leads farmers to adopt an adaptation practice. Its findings are also specific to Bizerte and should not automatically be assumed to describe farming networks elsewhere.

Creating more climate information may achieve little when the social infrastructure needed to move it is missing. In geographically fragmented farming communities, investing in the people who connect groups — and helping new connections form — could be just as important as producing the next piece of climate advice.