Eco-friendly farming methods were introduced to 100 farmers through a Training of Trainers (ToT) program in July 2023. From this group, 34 farmers were purposively selected based on (i) adoption of eco-friendly practices before April 2024; and (ii) availability of complete, internally consistent field and yield records for the 2024/2025 cycle. Therefore, the sample does not represent a statistically random draw from the cooperative population.

Yield measurements were collected from the first crop cycle (October 2024–February 2025). To control for environmental heterogeneity, each eco-friendly farmer was paired with a neighboring conventional farmer located within 100–200 m and matched for altitude, slope, shade structure, and soil type. To ensure independence and reduce spatial autocorrelation, farm pairs were separated by at least 200 m and treated as independent observational units.

Training for eco-friendly practices was delivered initially by an international GNGIF trainer and subsequently by trained local ToTs with staff facilitation. Conventional farming training was provided separately by Afridve, an agricultural consulting firm, during the same period.

In this study, eco-friendly farming is defined as an integrated package combining (i) no-tillage; (ii) continuous organic mulching; (iii) compost application; and (iv) field-made microbial inputs, including Indigenous Microorganisms (IMO), Fermented Plant Juice (FPJ), and Fermented Herbal Juice (FHJ). Additional inputs such as calcium phosphate from animal bones and eggshell-based calcium were also incorporated. No synthetic fungicides or herbicides were routinely applied. This package differs from general organic or agroecological labels because it incorporates IMOs and specific preparation and application frequencies unique to this project.

Conventional farming is defined as a system relying on tillage, synthetic fertilizers (CAN, NPK, foliar N–K sprays), copper-based fungicides, and limited use of organic matter.

To account for inherent differences between plots, baseline characteristics—including tree age and prior fertilizer history—were documented. The demonstration farm consisted of uniform 7-year-old Ruiru-grafted trees, whereas farmers' fields contained predominantly 10–15-year-old SL34 and SL28 trees. Conventional plots followed standard fertilizer programs prior to the study, providing the baseline nutrient and disease management conditions.

Table 1 summarizes the contrasting management practices between eco-friendly and conventional farming systems across soil management, nutrient inputs, pest control, and input sourcing.

A controlled demonstration farm was established in January 2024 to complement the farmer-level observational analysis. The farm consisted of two adjacent plots managed under contrasting systems: an eco-friendly plot implementing a standardized set of eco-friendly farming practices and a conventional plot managed according to prevailing local practices. Each plot contained 20 coffee trees of the same age and variety to ensure comparability.

Yield and disease data were collected according to a predefined schedule summarized in Table 2. Yield measurements were recorded during three periods: the early crop (30 April−11 June 2024), the main crop (4–25 November 2024), and the subsequent early crop (15 May−7 July 2025). Coffee disease incidence—Coffee Berry Disease (CBD) and Coffee Leaf Rust (CLR)—was assessed once in each cropping season on 13 May 2024 (early crop), 15 October 2024 (main crop), and 13 May 2025 (early crop).

For each assessment, CBD and CLR were evaluated on all 20 trees per plot. The number of infected berries (CBD) and infected leaves (CLR) were counted for every tree, providing precise and fully comparable tree-level disease metrics. All disease evaluations were conducted by trained research staff to ensure procedural consistency and minimize observer bias. Data collection employed standardized survey tools and was verified through on-farm supervision.

Monthly treatment schedules followed the demonstration farm's management logbooks (summarized in Tables 3, 4).

Eco-friendly treatments included the application of Indigenous Microorganisms (IMO), Fermented Plant Juice (FPJ), Fermented Herbal Juice (FHJ made from garlic and ginger), calcium phosphate (animal bones fermented in vinegar), and eggshell-derived calcium, all applied using standardized dilution protocols alongside compost-based nutrient amendments. Conventional treatment relied on Borozinc, NPK, and Mancozeb with more frequent use of synthetic foliar sprays and pesticides.

Table 5 summarizes the pesticides and fungicides applied in the conventional farming plot during the 2024–2025 period. For disease management, Green Copper produced by Twiga Chemical Industries Ltd. was used at a rate of 140 grams per 20 L of water. Its active ingredients, copper oxychloride and copper hydroxide, target both Coffee Leaf Rust (CLR) and Coffee Berry Disease (CBD). Over the monitoring period, a total of 10 applications were conducted between January 2024 and June 2025, amounting to 1,400 g in total. For weed control, Weedal, manufactured by Hangzhou Agrochemical Industries E.A Ltd., was applied at a rate of 120 ml in 20 L of water. Containing 2,4-D Amine and Glyphosate as active ingredients, Weedal was applied twice during the same period, with a total application volume of 240 ml. This pesticide and fungicide regime reflects the reliance on chemical inputs for disease and weed management in the conventional farming system.

Yield data from famers were collected during three harvest periods from 2023 to September 2025. Farmers harvested cherries from the designated trees along the transect, and the quantities were documented through structured interviews and field surveys administered by the research team. All farmer-reported harvest amounts were cross-checked with on-site observations and farm sales records to ensure consistency and reliability of the yield data.

Yield data were collected jointly by trained researchers and farmers. To ensure data reliability, all field observations were independently re-measured by a second assessor, achieving 100% consistency. Field sheets were reviewed and all data were digitized using a double-entry verification system.

All statistical analyses were performed in R (version 4.3). Data preparation and visualization were conducted using the tidyverse package suite.

For the farmer comparison dataset, per-tree yield differences between eco-friendly and conventional plots were evaluated using paired t-tests based on the matched-pair design. The Shapiro–Wilk test was used to verify the normality of paired differences prior to applying parametric tests. Effect sizes (Cohen's d) and 95% confidence intervals (CIs) were calculated to quantify the magnitude of treatment effects.

For demonstration-farm analysis (yield and disease incidence), CBD and CLR incidence were measured exclusively at the demonstration farm. To account for repeated measurements across seasons and clustering at the tree level, we fitted linear mixed-effects models (LMMs) using the lme4 package. Models specified tree ID as a random intercept and included management system (eco-friendly vs. conventional), season, and their interaction as fixed effects.

Model performance was summarized using marginal and conditional R². Paired t-tests were additionally used as descriptive comparisons for demonstration-farm yield and disease incidence.

Statistical significance was defined as α = 0.05 for all analyses. Results are reported with test statistics, model estimates, standard errors, and confidence intervals in accordance with Frontiers reporting guidelines.

The study spanned approximately 20 months (January 2024–September 2025), covering two major harvests and one minor harvest typical of Kenya's bimodal coffee production system. While this period represents one extended production cycle, it does not capture inter-annual climatic variability or long-term soil ecological change. Results should therefore be interpreted as short-term outcomes rather than long-term sustainability trends.

This analysis compares the coffee yield per tree between eco-friendly farming and conventional farming methods based on data collected from 34 farmers during the 2024/2025 first crop season.

In Table 6, the mean yield per tree under the eco-friendly farming practice was 2.91 kg, whereas the mean yield under conventional farming practice was 1.79 kg. A paired t-test was conducted to compare the two farming systems. The results indicated a statistically significant difference in yield between eco-friendly farming and conventional farming practices (t = 3.837, p < 0.001). These findings demonstrate that eco-friendly farming practices were associated with significantly higher yields per tree compared to conventional practices.

Table 7 is shown the comparison of the yield per tree in both farming methods and assess statistical significance within each factory. Kianjuki showed the largest difference with +1.48 kg/tree. Kevote and Ngaindethia followed with +0.85 and +0.79 kg/tree, respectively. For the 2024/2025 first crop, eco-friendly management increased per-tree yield across all factories, with statistically significant gains in Kianjuki, marginally significant gains in Kevote, and a positive but non-significant trend in Ngaindethia due to small sample size. However, these results highlight that local contextual factors and factory-level dynamics influence outcomes, and suggest promising trends for eco-farming that require further validation through larger samples and repeated studies.

Per-tree coffee yield differed between eco-friendly and conventional management across the two cropping seasons and adoption-year groups (Table 8). During the 2023/2024 second crop, eco-friendly plots showed slightly higher yields than conventional plots for both adoption cohorts.

However, these differences were not statistically significant. Among farmers who adopted eco-friendly practices in 2023, eco-managed trees produced 0.54 kg more coffee per tree than conventionally managed trees (95% CI: −0.24 to 1.32, p = 0.159). A similar non-significant pattern was observed among 2024 adopters, with a mean difference of 0.74 kg/tree (95% CI: −0.58 to 2.07, p = 0.250).

In contrast, significant differences emerged during the 2024/2025 first crop. For 2023 adopters, eco-friendly plots yielded 1.48 kg/tree more than conventional plots (95% CI: 0.32–2.64, p = 0.016). Among 2024 adopters, the yield advantage remained positive (+0.85 kg/tree) and reached the threshold of statistical significance (95% CI: 0.00–1.70, p = 0.050). These results indicate that although eco-friendly farming showed no measurable advantage during the early adoption period (2023/2024 second crop), productivity benefits became pronounced as farmers advanced into their first full production season under eco management.

Table 9 summarizes the mixed-effects model evaluating the effects of farming method, adoption year, and season on per-tree coffee yield. Consistent with the descriptive patterns, eco-friendly management showed a strong and statistically significant positive effect on yield. After adjusting for season and adoption year, eco plots produced approximately 1.30 kg/tree more than conventional plots (95% CI: 0.63–1.98, p < 0.001). Season also had a clear influence, with the 2023/2024 second crop yielding significantly less than the 2024/2025 first crop (Estimate = −0.91, p < 0.001). Adoption year alone did not significantly affect yield (Estimate = 0.40, p = 0.443), indicating no systematic baseline difference between 2023 and 2024 adopters.

The interaction between farming method and adoption year approached statistical significance (Estimate = −0.85, p = 0.069). As shown in Table 9, this reflects a tendency for the yield advantage of eco-friendly management to be larger among 2023 adopters (≈ +1.30 kg/tree) compared with 2024 adopters (≈ +0.45 kg/tree), although the difference was not statistically conclusive.

Random-effects estimates further indicate substantial heterogeneity in yield. The farmer-level random intercept (SD = 1.15) demonstrates marked differences in baseline productivity across farms, whereas the residual SD (1.34) reveals considerable within-farm variability attributable to tree-level and micro-environmental factors. These variance components highlight the importance of accounting for both farm-level and within-farm noise when evaluating management effects in smallholder conditions.

The following Figure 1, Tables 10, 11 illustrates the comparative outcomes of Yield, Coffee Leaf Rust and Coffee Berry Disease for both conventional and eco-friendly farming methods over the trial period (2024–2025):

In Figure 1, paired-sample Cohen's d effect sizes comparing eco-friendly and conventional yields across seasons. The 2024 first season shows a large negative effect (d = −1.15), indicating substantially lower yields under eco-friendly management; however, this early deficit should be interpreted cautiously because the demonstration farm was established only in January 2024, and eco-friendly practices had been applied for barely 2 months before harvest—an interval too short for agronomic effects to fully manifest. In contrast, the 2024 second season shows a large positive effect (d = +1.12), and the 2025 first season exhibits a medium-to-large positive effect (d = +0.92), suggesting stronger eco-friendly yield gains once sufficient time for soil and plant responses had elapsed.

In Table 10, CBD incidence—measured as the number of infected berries per tree—showed clear differences between management systems and seasons. In 2024, eco-friendly management produced a marked reduction in CBD, with the mixed-effects model estimating a mean decrease of 38.15 infected berries per tree compared with conventional plots (p < 0.001). This effect corresponded to a very large paired effect size (Cohen's d = −2.24), representing roughly a 90% reduction in disease burden.

Seasonal patterns were also evident. Overall CBD incidence declined sharply in 2025 (−35.70 berries per tree; p < 0.001), reflecting substantially lower disease pressure during the second year of monitoring. The significant interaction between management system and season (β = 34.05, p < 0.001) indicates that although eco-friendly plots continued to exhibit lower CBD incidence than conventional plots in 2025, the magnitude of this difference was reduced as overall disease levels dropped. This finding aligns with the medium-sized paired effect observed in 2025 (Cohen's d = −0.60).

Model explanatory power was high. Fixed effects—management system, season, and their interaction—accounted for 78% of the variance in CBD incidence (marginal R² = 0.78), underscoring their dominant influence on disease dynamics. Incorporating random tree-level variability increased the total explained variance to 82% (conditional R² = 0.82), indicating modest but meaningful heterogeneity among individual trees that was appropriately captured by the mixed-effects framework.

In Table 11, CLR incidence—measured as the number of infected leaves per tree—also exhibited clear differences across management systems and seasons. In 2024, eco-friendly management resulted in a substantial reduction in CLR, with the mixed-effects model estimating a mean decrease of 27.05 infected leaves per tree compared with conventional plots (p < 0.001). This reduction corresponded to a very large paired effect size (Cohen's d = −2.10), indicating that eco-friendly practices lowered CLR severity by more than 90% during the high-disease season.

Seasonal dynamics showed a similarly pronounced pattern. Overall CLR incidence declined sharply in 2025 (−24.70 leaves per tree; p < 0.001), reflecting substantially lower pathogen pressure in the second year of observation. The significant interaction between management system and season (β = 27.00, p < 0.001) indicates that although eco-friendly plots continued to display lower CLR incidence than conventional plots in 2025, the magnitude of the difference diminished as overall disease levels fell. This pattern is consistent with the near-zero paired effect size observed in 2025 (Cohen's d = −0.05).

Model explanatory power was likewise high. Fixed effects—management system, season, and their interaction—accounted for 91% of the variance in CLR incidence (marginal R² = 0.91), demonstrating their strong influence on disease outcomes. Incorporating random tree-level variability increased the total explained variance to 92% (conditional R² = 0.92), indicating limited but non-negligible heterogeneity among individual trees that was effectively partitioned within the mixed-effects structure.

Table 12 summarizes the cost–benefit comparison between eco-friendly and conventional management. Total input costs were nearly identical, with eco-friendly practices costing KES 10,030 vs. KES 10,590 for conventional inputs, indicating that ecological management does not require higher financial investment. Despite similar costs, eco-friendly plots showed substantially lower CBD and CLR incidence—up to 90% reductions in 2024—and maintained comparable or slightly higher yield trends during the early adoption period. Overall, the combination of equal or lower costs and markedly reduced disease pressure demonstrates a clear agronomic and economic advantage of eco-friendly management at the demonstration farm.