The study was conducted in three districts: Babati (Manyara region), Kondoa (Dodoma region), and Singida Rural (Singida region), which are among the focus areas of the Agriculture and Fisheries Development Programme (AFDP) for the 2020-2026 period. This programme, funded by the International Fund for Agricultural Development (IFAD), focuses on developing climate change adaptation technologies in the drier Agro-Ecological Zone (AEZ) of Tanzania’s central mainland corridor as shown in Figure 1 . The selected districts were representative of semi-arid agro-ecological zones of Tanzania. The rainfall and temperature in the study area are presented in Table 1 .

To assess the effectiveness of CSA technologies on maize and common bean intercrops, the study used the TRICOT approach [24] [35] . This is a farmer-centered, on-farm participatory research method ( Figure 2 ), explaining that (A) trials are designed on ClimMob ( https://climmob.net/ ) following a trial protocol derived from the target technologies profile under testing [36] . (B) Technology options (varieties, management practices among others), the selection was based on the aims of the experiment. (C) Sets of three technology options are assigned randomly as incomplete blocks from a broader set of technology packages. Field agents denoted (D) register participants on ClimMob and distribute the trial packages for farmers, the identifiable data (name, age, village, district, GPS) are recorded using Open Data Kit (ODK). (E) Participants guided by researcher and extension agent establish the experiment in their own farm and evaluate list of traits as per trial protocol (e.g., growth parameters, plant height, pest and disease resistance, drought tolerance, yield, etc.) using “tricot rankings” by indicating the option with best performance (1st in the ranking) and the option with worst performance (3rd in the ranking) for the given trait. The 2nd place in the ranking is added to the option not mentioned as best or worst for the given trait. (F) Participants’ assessments are registered using ODK, sent to the ClimMob platform, and aggregated for data analysis and production of automated reports [24] . This participatory approach enhanced scalability and relevance by placing farmers at the center of the research process.

A balanced incomplete randomized block design was used to plant the trials, with 135 farmer sites serving as blocks. Table 2 shows the number of farmers who participated in the Tricot experiment per each village following the use of the formula below as indicated in equation 1.

Equation: $\text{n}=\frac{N}{1+N{\text{e}}^2}$ (1)

where:

n = sample size,

N = population size,

e = sampling error 5 percent [38] .

Therefore, having the number of participants, allowed to assign treatments for evaluation. A total of 15 treatments, thus $\frac{15}{3}=5$ sets of the treatments, that assigned (replicated) to 135 of the project participating farmers (blocks) $\frac{135}{5}=27$ replicates, with three (3) entries as test packages of technologies in each block as it was recommended by [39] .

The CSATs such as farmyard manure (FYM), flat cultivated fields devoid FYM and farmers’ field management techniques (FP) were used as the field management techniques as shown in Table 3 .

The study focused on the intercrop sets made up of drought tolerant maize (DTMV) and early maturing common bean (EMCBV) varieties, as well as a set of known and available local check maize and common beans varieties. The treatments were randomized and assigned into incomplete randomized block designs (RCBD) as shown in Table 4 .

Whereby;

CSA seed varieties for maize include: DTMV₁ for maize variety 1, DTMV₂ for maize variety 2, and LCMV for maize local check; CSA seed varieties for common beans: EMBV₁ for beans variety 1, EMBV₂ for beans variety 2, LCBV for beans local check.

The CSA management practices (P) under maize-common bean intercrops include: P1 with Farmyard Manure (FYM); P2 without farmyard manure (FYM): P3 farmer practice (FP).

Specific growth stages (GS) of maize and common bean were recorded during the growing season. The data collected from sole maize, sole common bean and maize-common bean intercrop plots during main and short rain planting seasons were analyzed using a combination of mixed models and analysis of variance (ANOVA) [40] [41] .

In a mixed model framework, genotypes, cropping systems, spacing and FYM application were treated as fixed effects while farm (location) and season were treated as random effects [42] . This was to account for natural variability across the farmer-managed plots and seasonal weather conditions, enhancing the generalizability of the results [43] . The random structure addressed spatial and temporal heterogeneity that may otherwise bias fixed effect estimates [44] .

The coding of experimental treatments included combinations of sole maize, sole common bean, and maize-common bean intercrops, across different spacing levels and FYM application rates. This structure allowed the testing of the hypothesis that inter-row spacing and FYM rates significantly influence the yield of maize-common bean intercrops in semi-arid regions of Tanzania.

To test the assumptions of normality and homogeneity of variances, the Shapiro-Wilk test and Bartlett’s test were applied to the residuals before conducting ANOVA. The validity of the mixed model approach relied on the fulfillment of these assumptions. Additionally, multiple linear regression was performed with grain yield as the response variable and agronomic indicators such as cob and pod weight, total biomass, and plant population as explanatory variables, to assess their contributions to yield.

All statistical analyses were performed using GenStat Discovery Edition 21. For the maize–common bean intercrop data, a 3-way ANOVA was conducted using the model shown in Equation 2:

Equation: $Y_{ijk}=\mu +{\alpha }_i+{\beta }_j+{\Upsilon }_{\kappa }+{\left(\alpha \beta \right)}_{ij}+{\left(\alpha \Upsilon \right)}_{i\kappa }+{\left(\beta \Upsilon \right)}_{jk}+\left(\alpha \beta \Upsilon \right)i_{j\kappa }+{\epsilon }_{ij\kappa }$ (2)

where:

$Y_{ijk}$ = observed value in the ijk_th treatment combination;

$\mu$ = overall mean;

${\alpha }_i$, ${\beta }_j$, ${\Upsilon }_{\kappa }$ = main effects of variety, spacing, and FYM;

${\left(\alpha \beta \right)}_{ij}$, ${\left(\alpha \Upsilon \right)}_{i\kappa }$, ${\left(\beta \Upsilon \right)}_{jk}$ = two-way interactions;

$\left(\alpha \beta \Upsilon \right)i_{j\kappa }$ = three-way interaction;

${\epsilon }_{ij\kappa }$ = random error.

The 3-way ANOVA was appropriate because it allowed simultaneous evaluation of the main and interaction effects of three agronomical relevant factors—variety, spacing, and FYM application—on intercrop performance. This was critical for uncovering synergies or trade-offs essential for CSA recommendations in resource-constrained, semi-arid environments.

For the sole common bean data collected during the short rainy season, a 2-way ANOVA was conducted as shown in Equation 3:

A 2-WAY ANOVA was used for the data of common bean collected during the short rainy season and the factor effects model (Equation 3) was:

Equation: $Y_{ij}=\mu +{\alpha }_i+{\beta }_j+{\left(\alpha \beta \right)}_{ij}+{\epsilon }_{ij}$ (3)

where:

$Y_{ij}$ = observation in the ij_th treatments;

$\mu$ = overall mean;

${\alpha }_i$, ${\beta }_j$ = main effects of spacing (S) and farmyard manure (FYM);

${\left(\alpha \beta \right)}_{ij}$ = interaction effect between S and FYM;

${\epsilon }_{ij}$ = random error associated with the observation in the ij_th factors.

Similarly, for sole maize, a 2-WAY ANOVA was applied (Equation 4) as follows:

Equation: $Y_{ij}=\mu +{\alpha }_i+{\beta }_j+{\left(\alpha \beta \right)}_{ij}+{\epsilon }_{ij}$ (4)

where:

${\alpha }_i$, and ${\beta }_j$ = main effects of spacing (S) and farmyard manure (FYM) application.

Soil data collected at the end of each harvest (2023 and 2024) were analyzed using one-way ANOVA to compare soil fertility indicators across intercropping systems and sole cropping with different FYM rates, using Equation 5:

Equation: $Y_i=\mu +{\alpha }_i+{\epsilon }_i$ (5)

where:

$Y_i$ = Value of soil nutrient in the ith intercropping system;

$\mu$ = overall mean;

${\alpha }_i$ = effect of the intercropping system;

${\epsilon }_i$ = random error.

All significance tested were evaluated at the 95% confidence level (P = 0.05). Post hoc comparisons were made using Tukey’s test to determine statistically significant differences among the treatment means [41] [45] .

Evaluation of land equivalent ratio (LER)

The Land Equivalent Ratio (LER) was a crucial index used to evaluate the productivity and efficiency of intercropping systems compared to sole-cropping [31] [46] . It indicates how much land would be required under sole cropping to achieve the same yields obtained in an intercrop. An LER greater than 1.0 suggests a yield advantage for intercropping, meaning the system uses land more efficiently. The evaluation of the productivity of intercropping systems relative to monoculture was done using the LER, of each intercropping system was obtained by summing up the relative yields, that is, yields (ton/ha) of maize and beans in the same intercropping system were divided by that of maize and beans in sole cropping in order to give relative yields and then added together to give LER as shown (Equations 6, 7 and 8).

The general formula for calculating LER in a two-crop system was:

Assumption: LEF = PLER

Equation: LER = PLER maize + PLER common bean (6)

Equation: PLER maize = Yield of maize in intercrop (7)

Yield of sole maize

Equation: PLER common bean = Yield of common bean in intercrop (8)

Yield of sole bean

Similarly, the data set was subjected to analysis of variance (ANOVA) using GenStat Version 21 as well the R-software using the following statistical model (Equation 9) was

Equation: $Y_{ij}=\mu +T_i+B_j+e_{ij}$ (9)

where:

$Y_{ij}$ = observation i_th of the treatments and j_th = blocking factor;

$\mu$ = overall mean;

$T_i$ = treatment effects;

$B_j$ = effect in block $e_{ij}$.

The treatment means for the different parameters were separated using Tukey HSD test at 0.05 level of significance.

An ANOVA was performed to test the hypothesis that at least one technology or treatment had a significantly higher performance across key traits compared to the others. The results, illustrated in Figure 3 , showed that for most traits, the p-values were less than 0.05, confirming that the performance differences observed among the items were not due to random chance. This suggests that at least one item in the set performed significantly better in traits such as cob size, plant vigor, and pod number. These findings confirm that the adoption of improved varieties and management practices can lead to significant performance enhancements under semi-arid conditions, a critical consideration for increasing resilience to climate variability.

Traits such as cob size for maize and pod count for common beans had the most significant variance across treatments, which directly correlates with yield potential and marketable produce. To assess which traits most strongly influenced farmers’ overall preference, a Kendall tau correlation analysis was conducted and among all traits evaluated, Cob size [Maize harvest yield data]’ exhibited the strongest correlation with ‘Overall Preference’, with a Kendall tau coefficient of 0.41, indicating a moderate positive association.

The results imply that larger cob size was a strong determinant of technology preference by smallholder farmers, likely due to its direct impact on food security and market value. On the other hand, the trait ‘Germination (Agronomic performance)’ had the weakest correlation with overall preference, with a Kendall tau of 0.01, suggesting that while germination is important for stand establishment, it may not heavily influence farmers’ ultimate selection decisions, especially if later-stage performance (e.g., yield, cob size) is strong. These findings provide empirical support for the effectiveness of CSATs in improving key agronomic traits that influence both biological yield and farmers’ perceptions of performance as reported by [26] . The results also emphasize the importance of targeting traits that are both statistically superior and practically valued by end-users [47] . The strong performance of certain improved maize and bean varieties, especially in cob size and pod number, reinforces their suitability for scaling in semi-arid zones.

The log-worth values by trait for each technology tested in this experiment are illustrated in Figure 4 . This graphical representation provides an intuitive assessment of how different maize and common bean varieties performed relative to the local checks the SITUKA (maize) and JESCA (common bean). The figure displays log-transformed worth values that quantify the likelihood of each technology outperforming the local checks on a trait-by-trait basis. The log-worth parameter is a statistical transformation of the performance probability that aids in distinguishing superior and inferior entries across multiple traits. A log-worth greater than 0 (blue bars in the chart) indicates that a given variety has a statistically higher chance of surpassing the performance of the local checks for that particular trait. Conversely, a log-worth less than 0 (red bars) denotes underperformance relative to the checks. Therefore, the approach was useful for visualizing not only the general performance trend of an item but also the variability in its performance across different agronomic parameters such as cob size, pod number, biomass yield, plant vigor, and pest resistance. For instance, technology with consistent positive log-worth across multiple traits can be considered highly promising candidates for broader dissemination in similar agro-ecological zones.

In this experiment, several improved maize and bean varieties displayed superior log-worth values in traits closely associated with yield, including cob size, grain weight, and pod count. These traits are crucial for determining productivity and farmer satisfaction. Notably, varieties such as T105 (maize) and TARIB6 (common bean) exhibited consistently high log-worth scores across multiple traits, suggesting their robust adaptation to the semi-arid environments under evaluation. The study also applied stratified performance analysis reinforces the value of combining statistical modeling with field-level observations to support decision-making in technologies selection. The log-worth analysis complements the ANOVA and Kendall tau results by adding a probabilistic dimension to performance.

Reliability (probability of outperforming a check) estimates based on worth estimates from PL model were presented in Figure 5 . Reliability measures the precision of estimated worth and the potential response to selection compared to a check. It is a breeding metric proposed by [48] . The reliability estimates were calculated considering the Overall Preference with Msituka + Bjesca (Local checks) as the check item. The blue vertical line set to 0.5 indicates a threshold from where a given item presents a potential probability in outperforming the check (reliability > 0.5).

The estimated log-worth for the ‘Overall Preference’ reference trait as shown in Figure 6 , the purpose was to be able to distinguish the item with the superior performance. Mean separation analysis was also conducted to indicate which item is significantly different (or similar). When items have at least one letter in common, there is not enough evidence from the trial to be confident about their relative order with p-value = 0.05. The Overall Preference’ assessed in this study trials and the intervals were based on quasi-standard errors.

The results showed that DTMV and EMBV intercrops grown and managed using application of FYM resulted into higher yields of both maize and common beans T105 + Tari B6 (PI) in ton ha⁻¹ at (4.809 ± 1.034); followed by T105 + Selian 13 (PI) in ton ha⁻¹ at (4.788 ± 0.991) and T104 + TariB6 (P1) in ton ha⁻¹ at (4.56 ± 0.27). The trials managed with farmer practices (FPs) without applications of FYM yield less of both maize and common beans Msituka + Bjesca (local checks-P3) in ton ha⁻¹ at (2.866 ± 0.726), Msituka + Bjesca (local checks-P2) in ton ha⁻¹ at (2.705 ± 0.687) and T104 + Selian13 (P3) in ton ha⁻¹ at (2.156 ± 0.804). These findings are in line with the study by [49] , the results showed that maize-bean intercropping with application of manure enhanced the resilience of the drought-tolerant bean genotypes giving a 44.4% yield increase as compared with the intercropping of maize and common beans without application of FYM. The study concluded that the improved DTMV and EMBV managed with FYM as significant effects in climate resilience and increase yields of maize and common beans as the results shown in Figure 7 .

The evaluation of gender in maize-bean intercrops reveals distinct priorities between male and female farmers [22] [36] [47] . Understanding these differences was crucial for developing and promoting agricultural practices that were both effective and equitable [47] . The study evaluated the gender preferences on twelve (12) traits were evaluated which was the bean maturity, disease resistance, drought tolerance, germination, maize cob size, maize maturity, pest resistance, plant survival, seed color, seed size, and vigor as shown in Figure 8 . The results shows that women prioritized the traits that enhance household food security and ease of processing. This includes preferences for seed color and size, which can affect culinary qualities and marketability [24] [39] . The results also show that female farmers targeted characteristics based on both agronomic and post-harvest grain attributes to ensure the sustainability of household food intake as shown in Figure 8 which also reported by [50] . The women have shown a preference to drought tolerance, and resistance to pests. To adapt to climate variability, women were preferred the use of improved seeds, aiming to mitigate risks associated with unpredictable weather patterns [39] [47] .

The men typically prioritize traits that maximize economic benefits, such as larger maize cob size and higher overall yields. The men decisions were driven by market-oriented goals thus, were more focusing on technologies that can significantly boost production as also reported by [47] . Generally, both male and female farmers valued traits that confer resistance to diseases and pests, recognizing the importance of plant survival and vigor in ensuring consistent yields [50] [51] . Given the increasing challenges posed by climate change, drought tolerance is a universally desired trait among farmers to safeguard against drought. These gender-specific preferences highlight the need for inclusive agricultural research and extension services that consider the distinct priorities of male and female farmers [47] . By addressing these differences, interventions can be better tailored to meet the diverse needs of farming communities, leading to more sustainable and equitable agricultural development [51] .

Baseline soil fertility analysis conducted before FYM application in 2023 indicated low levels of organic matter and key nutrients. The average organic carbon content was 2.24%, and total nitrogen averaged 0.15%. Phosphorus levels ranged from 6.2 to 8.5 mg/kg, and exchangeable calcium and magnesium were generally below optimal levels, with mean values of 6.1 and 2.2 CmolKg⁻¹, respectively. These values highlight the poor fertility status typical of semi-arid soils and provide context for the substantial improvements observed after FYM application, particularly in the 8 t/ha treatment. The increase in soil organic matter and nutrient availability demonstrates the importance of organic amendments for sustainable soil fertility management [20] [52] .

Based on the experimental results, the best-performing intercropping system was SP50 × 50 spacing with FYM at 8 t/ha. The corresponding yields were: the intercrop maize yield 5.416 ton ha⁻¹, the intercrops bean yield: 0.5356 ton ha⁻¹, and the sole maize yield about 4.978 ton ha⁻¹ and the sole bean yield 0.5485 ton ha⁻¹.

The Partial LERs were calculated as:

PLER for maize: 5.416 / 4.978 = 1.088

PLER for common bean: 0.5356 / 0.548.5 = 0.977

Thus, the total LER: 1.088 + 0.977 = 2.065

This LER value indicates that the intercrop system with SP50 × 50 spacing and 8 t/ha FYM application is 107% more efficient in land use than growing maize and beans separately on the same area. These findings confirm the high potential of CSAT intercropping approaches for improving productivity and land-use efficiency in semi-arid environments as were reported by [31] [53] .

To evaluate the economic feasibility of the CSAT recommendations, a cost-benefit analysis was conducted comparing input costs and the additional value of grain yield under different manure rates and spacing. The highest-performing treatment combination (FYM at 8 t/ha and 50 × 50 cm spacing) resulted in maize and bean yields of 7367 kg/ha and 1040 kg/ha, respectively. Assuming average farm-gate prices of 500 TZS/kg for maize and 1500 TZS/kg for beans, the gross income per hectare amounts to 3,683,500 TZS (maize) and 1,560,000 TZS (beans), totaling approximately 5,243,500 TZS/ha. Input costs for 8 t/ha of FYM application were estimated at 800,000 TZS/ha (including transport and labor), while labor and seed costs amounted to approximately 600,000 TZS/ha. This gives a net return of approximately 3,843,500 TZS/ha, confirming the economic viability of this package under semi-arid conditions also were reported by [26] [54] the benefits of the use of CSAT.

Despite the robust experimental design and analysis, several limitations may have influenced the observed yield differences. Firstly, variability in rainfall across the growing seasons introduced inconsistencies in crop performance, especially in water-limited environments. Secondly, heterogeneity in farmer management practices including planting time, weeding frequency, and fertilizer application may have introduced uncontrolled variability across trial sites. Additionally, other potential confounders such as differences in soil fertility, pest pressure, and access to inputs could not be fully standardized across the decentralized on-farm trials. While efforts were made to account for these factors through the inclusion of farm and season as random effects in the statistical model, their residual influence on yield outcomes cannot be entirely ruled out. These limitations highlight the importance of cautious interpretation and the need for further controlled trials to complement farmer-participatory evaluations.