ICT4Agroecology part II: outcomes for maize production systems

ABSTRACT Agroecology is gaining increased attention by experts and global organizations; however, it’s practical application in the field remains understudied, particularly in developing countries. Here, we aimed to evaluate common agroecological practices at three sites in Tanzania, including organic soil amendments, intercropping, and biological pest control – individually, in pairs, or in three-way combinations at each site. Information Communication Technology tools – the AgroEco Research and AgroEco Analysis application – were used for data gathering & storage and data visualization & statistical analysis, respectively. We found that high maize kernel weights (as proxy for yield) can be obtained from nutrient-poor soils specifically when grown with a combination of organic soil amendments and intercropped with legumes – ranging from 3 to 4.5 t/ha (with a maximum of 7 t/ha on one occasion) and exceeding national maize yields of 1–2 t/ha in Tanzania. However, intercropping and biological pest control individually and in combination did not affect maize yield. Under flood or drought conditions, only plots with soil fertility amendments produced some harvestable maize kernels. Our results provide a substantial reference for recommending and advocating agroecological methods to smallholder farmers, farmer training organizations, and policy makers. GRAPHICAL ABSTRACT

Maize (Zea mays) is a staple cereal crop for people living in Sub-Saharan Africa (Government of Tanzania 2021).However, current maize yield is generally low.In the 2019/2020 season, the average productivity of maize was 1.5 tons/ha in mainland Tanzania and 1.9 tons/ha in Zanzibar (Government of Tanzania 2021).In mainland Tanzania, the highest yield (2.0 tons/ha) was reported in the Ruvuma and Njombe regions, followed by the Mbeya and Songwe regions, each with a yield of 1.9 tons/ha (Government of Tanzania 2021).The lowest maize productivity was observed in the dry Dodoma region (1.1 t/ha, Government of Tanzania ( 2021)).The country's population is increasing, but the arable land area is decreasing as a result of land degradation and the allocation of agricultural land for other development uses.A further increase in land used for maize production is, therefore, currently infeasible and unsustainable.Therefore, research must identify sustainable agroecological practices to increase maize yield (Constantine et al. 2021;Shiferaw et al. 2011).
This study is the outcome of a 5-year Agroecology Research and Advocacy project across three agroecological zones of Tanzania.This project aims to address this knowledge gap and provide local scientific evidence related to maize cultivation (Hilbeck et al. 2024a).We conducted experiments on maize and a second staple crop, cassava (Manihot esculenta).Here, we report the findings for maize, while the outcomes for cassava were reported in a sister publication (Hilbeck et al. 2024b).The primary goal of this project was to validate the effectiveness of selected agroecological practices (applied singly or in combination) that can be recommended to smallholder farmers, farmer training organizations, policy makers, and the general public.
We tested agroecological practices from three categories of common agroecological methods (e.g., Bezner Kerr et al. (2021); Constantine et al. (2021); Himmelstein et al. (2017); Paracchini et al. (2020)): (1) soil fertility and conservation measures (compost and mulching); (2) increased biodiversity through intercropping (maize -cowpea legumes); and (3) biological pest control measures (based on ash or botanicals such as neem, chili, garlic, and aloe).The research design allowed testing of each practice individually, in pairs, or in three-way combinations at each location.Our research was coordinated with a broad range of experts and practitioners from academic institutions, civil societies, government organizations, and smallholder farmers (Hilbeck et al. 2024a).
For many years, Tanzania has experienced a rapid increase in mobile phone usage and a corresponding increase in investment in digital services, including within the agricultural sector.Hilbeck et al. (2022) found that although there is a wide range of agricultural assets (e.g., crop-or value chain-specific) and technologies (e.g., drones, Internet, SMS) within the agricultural sector, the majority -if not all -align partly or fully with conventional farming approaches and none of the technologies specifically align with agroecological practices.Thus, another goal of this project was to use readily available technologies on mobile phones already owned and operated by most farmers, in order to use Information Communication Technologies (ICTs) to facilitate context-based, farmer-centered transitions to agroecology.Our project was supported by two specifically developed ICT tools: AgroEco Research application (AER) was used for data gathering and storage, and the data recorded using the AER application were automatically read by the AgroEco Analysis application (AEA).The AEA tool is unique, in that it reads data directly from the AER website, allowing immediate (real-time) data visualization, statistical analyses, and data download for further use.
To meet our research goals, we considered four research questions: (1) Do common soil fertility management measures (i.e., with local ingredients and mulching) affect maize yield, plant size, and plant survival?; (2) How does intercropping legumes with cowpeas affect maize yield, plant size, and plant survival?; (3) Do common ecological pest controls comprising local ingredients affect maize yield, plant size, and plant survival?; and (4) What is the interplay between soil fertility management, intercropping, and pest control?
The research goals regarding the development and testing of the specifically developed ICT applications and their usage by farmers have been published in detail elsewhere (see Hilbeck et al. 2024a, Hilbeck et al. 2023)

Materials and methods
We carried out multi-year field plot studies at three field sites, each in a different agroecological zone in Tanzania: Chambezi field station (northeastern coastal region of Bagamoyo District Council); Mvomero field site (Morogoro District, 200 km inland from Dar es Salaam); and Mumbaka field site (Masasi District, southern region bordering Mozambique).We conducted field trials over four field seasons (2018)(2019)(2020)(2021) to better consider the effects of local and unpredictable events, such as droughts or flooding.Additional details on the materials and methods for this integrated project are explained in a separate paper (Hilbeck et al. 2024a).

Field plots
We created 48 plots of 18 m 2 each (6 × 3 m) at each of the three sites.Treatments were replicated three times at each field site, each replication consisting of 16 experimental plots.Maize was grown in eight plots per site (24 plots in total), while the remaining eight plots were used for cassava cultivation, as reported in the Hilbeck et al. (2024a) methodology companion paper.We tested three agroecological practices: (1) single crop vs. maizecowpea intercropping; (2) ecological pest control vs. no pest control; and (3) soil fertility amendment vs. no soil fertility amendment.The treatments and corresponding sets of agroecological practices are abbreviated as follows: • P -ecological pest control (application of common natural pest control remedies) • S -soil fertility amendments (addition of local organic material to the soil, i.e., compost and mulching) • L: legume intercropping with cowpea (Vigna unguiculata) with a reduced number of maize plants to allow space for the intercrop.• C -control (no treatments, standard farmer practice without fertilizers or pesticides) Season and replicate area (Rep) were included as block factors.We used a full factorial design, in which every treatment was applied individually or in combination with other treatments, and compared against the negative control plots.This allowed for estimation of the effect of a single treatment factor on the response variable, as well as potential interaction effects.Given that pest control and soil fertility amendment were more difficult to vary across plots than was legume intercropping, we applied a split-plot design, where pest control and soil amendment varied across larger areas than intercropping.Specifically, each replicate area was split into four and individual quarters treated with P, S, P + S, and one with no treatment.Within each quarter, legume intercropping was applied to half of the plots.Combinations of these treatments are indicated by stacking the abbreviations together, for example the abbreviation "PS" indicates both pest control and soil amendment.
The same open-pollinated maize variety (TMV1) was used at all field sites owing to its drought tolerance and high yield capacity. Similarly, the same non-crawling cowpea variety (Tumaini) was used for maize -legume intercropping at all field sites.For both crops, untreated seeds were obtained from TARI, Ilonga and Morogoro.Although the amount and timing of soil amendment was standardized, as well as the preparation method and the ratio of compost to mulch, the precise composition varied between (but not within) field sites owing to the availability of different local organic materials to farmers.Similarly, biological pest control was applied using comparable rigor across field sites, but the ingredients (i.e., ash, chili, aloe vera, garlic, and/or neem) differed depending on local practices.Therefore, we emphasize that our conclusions about the utility of these treatments need to be interpreted with respect to their local application.Further details on the applied treatments, including the ingredients and preparation methods of inputs (compost, mulching, and pest control), are provided by Hilbeck et al. (2024a).

Local drought and flooding events
Unpredictable periods of drought affected all field sites.The Mumbaka and Mvomero field sites were affected by low amounts of precipitation, leading to droughts in 2019 and 2021.At these field sites, maize seeds did not germinate after the first sowing and, thus, were re-sown and irrigated at a low level.To avoid moisture variation, all maize plots were irrigated with the same amount of water at the same frequency (two times a day for two days in three weeks).In addition to drought spells, floods significantly affected the Chambezi field site during the long rainy seasons in 2018 and 2020.No replacement planting or harvesting could be performed in either year.Instead, we allowed the surviving plants to continue to grow until harvest, and analyzed these 2 years separately to highlight how the tested agroecological practices impacted the measured parameters under extreme environmental conditions.

Response variables
We measured four different variables: plant size, weight of maize kernels per plot (one proxy for yield), weight of maize kernels per number of planted seeds (another proxy for yield), and number of surviving plants.The size of the maize plants was measured when they reached their final height at the tasseling and flowering stages.Crop height was measured from the soil surface to the top of the tassel.The maize yield was recorded as the total weight of kernels (in kg) per plot (18 m 2 ), converted to tons per ha (t/ha), and the total weight of maize kernels per plot (yield) was divided by the number of maize seeds planted per plot.The numbers of seeds planted was 200 (mono-cropped) or 130 (intercropped) per plot.The number of surviving maize plants per plot was recorded on the day of the harvest.This allowed us to evaluate possible linkages between treatment type and plant survival.

Data collection
Data was collected using the AER application (Hilbeck et al. 2024a) which is a mobile application 1 and web-based data-gathering tool that allows researchers to record the selected response variables.The application ensured data standardization across all geographical locations and seasons.The collected data were stored in a centralized online database, thus facilitating real-time querying and analysis, as well as ensuring safe storage through daily backups.Data collection using the AER application was directly supervised by scientists and/or by field staff and farmers trained by scientists.At these field sites, scientists or staff used the AER application to record agroecological inputs and practices, such as seeds or manure, field work activities (e.g., planting, harvesting), and response variables used to track the development, health, and performance of crops.This data collection could be performed using a regular smartphone or computer.

Data analyses
The data recorded using the AER application were automatically read by the AEA application (Hilbeck et al. 2024a).In short, the AEA application 2 reads the response variables, field sites, replication areas, experiment dates, and treatment factors from the AER application website 3 and automatically plots the response variables as a function of treatment and time.The user can select a subset of the data for analysis, for example by excluding seasons with crop failure due to flooding.Once the data subset was chosen, the effects of the treatments and their interactions on the response variables were automatically tested using Analysis of Variance (ANOVA) and considering the split-plot experimental design.The replicate area is always included as a block factor, while the season is included as a block factor once the season end dates are input by the user.Residual analysis plots indicate data skew by single measurements and/or potential violations of model assumptions.The user can then elect to apply first-aid transformations to the response variables and interaction terms (all main effects and two-way interaction effects by default) can be included to improve the model fit.The AEA application also facilitates the download of raw and aggregated data (means per plot).Data were analyzed for each field site (Chambezi, Mumbaka, and Mvomero).We refrained from a single common analysis with the field site as an additional block factor, as there were too many uncontrolled differences between field sites, owing to different climatic conditions, soil types, and timing of sowing and harvesting.

Soil analyses outcomes
In this paper, we report the data for the soil analyses carried out a total of three times for each field station, first sampling took place just before the first season planting and, subsequently, every other year in 2019 and 2021.Measured soil parameters included soil organic carbon (SOC), total Nitrogen (N), and available phosphorous (P).Methods of analysis were described in Hilbeck et al. (2024a andin Hilbeck et al. 2023).The soil data equally apply also to the cassava field trials and will be cited there accordingly.

Results
Here, we present the results for each measured parameter at each station and conclude with a comparison between field sites.

Plant size
Highly significant differences in maize plant sizes between seasons were observed, regardless of treatment (season p < 0.0001).Average maize plant size was the greatest during 2018 (Figure 1b).Years 2019-2021 were characterized by drier weather conditions than 2018, with similar average plant sizes for all treatments.
Soil amendment (S) had the most significant effect across all seasons (S p < 0.0001) (Figure 1a), although no differences were observed during 2018 between S-treated plots and the control (Figure 1b).In 2018, most plant sizes were 1.5-2.5 m, regardless of treatment (Figure 1b).In the following seasons (2019-2021), however, S-treated maize plants exhibited greater heights (up to 2-3 m) than those in the control plots and all other plots without soil amendments (heights of 1-2 m) (Figure 1b).This increase in height was the most pronounced in 2020, but consistently and, therefore, statistically significant in 2019-2021 for PS-treated plots (p = 0.0404) and PSLtreated plots (p = 0.0157) (Figure 1b).In three of the four seasons, mean maize plant sizes in L-treated plots were often smaller than those in the untreated control, which led to a near-significant legume intercropping effect (L p = 0.0782).Although S-treated maize plants tended to grow larger than those without soil amendments, plant sizes showed a wide variation; thus, they did not exhibit a statistically significant S effect across all seasons.Neither single treatments of P nor L, or their combinations (PL, SL), had a measurable effect on plant size, except when all treatments were combined (PSL).

Plant survival
The number of surviving maize plants remained constant across all treatments in comparison to the control.A significantly lower number of surviving maize plants were recorded in L-treated plots (p < 0.0001) owing to the lower number of maize seeds planted in these plots.Otherwise, no specific treatment or treatment combination showed a significant effect on plant survival in the block analysis or the mean comparison analysis (Figure 2b).
On average, 30-60 maize plants survived in L-treated plots, which was 25-50% of the planted (120-130) maize seeds.For plots without legume intercropping, an average of 70-100 maize plants survived, which constituted 50% or less of the 190-200 planted seeds.However, the number of surviving maize plants varied both within each season and significantly between seasons (seasonal effect p < 0.0001) (Figure 2).

Weight of harvested kernels per planted seeds
At the Mumbaka field site, the weights of harvested maize kernels per planted seed differed significantly between the seasons (p < 0.0001).In particular, low precipitation in 2018 and 2021 led to lower weights of the harvested kernels.
Soil treatments and maize -cowpea (legume) intercropping significantly increased the average weight of harvested maize kernels per planted seed (with S p < 0.0001 and L p = 0.0027) (Figure 3a).However, as illustrated in Figure 3b, the effects of these two treatments increased when used in combination.Across all seasons, SL-and PSL-treated plots resulted in average kernel weights 200-600% greater than did the control or individual treatments (S, L, or P).Maize kernel weights treated with a combination of soil fertility and legume intercropping ranged from 0.02 to >0.06 kg/planted seeds, which led to highly significant effects for SL (p = 0.001) and PSL (p = 0.001) compared to the control with kernels weights of approximately 0.01 kg/planted seeds.While kernel weights for S-, SL-, or PSL-treated plots varied by a factor of 2 or more between seasons, those in the untreated control plots varied very little between seasons and remained constant at ~0.01 kg/planted seeds or slightly below (2018) (Figure 3b).Soil amendments alone (S), or in combination with pest control (PS), had inconsistent effects on kernel weights.In 2018, kernel weights for S-treated plots were the same as those in the control; kernel weights were, however, higher under a combination of pest control measures (PS) (Figure 3b).In 2019, we observed the inverse result, where kernel weights for S-treated plots were higher than those in the control and in PS-treated plots.However, maize kernel weights were not significantly affected by single treatments of L (p = 0.9999) or P (p = 0.7246), or their combinations (PL p = 0.9999; PS p = 0.3191) compared to those in the untreated control plots.In fact, the weight of harvested kernels per planted seeds for L-treated plots occasionally measured below those in the untreated control plots.

Weight of harvested kernels per area
The data reported in this section do not account for the lower maize plant densities in the intercropped plots.However, despite lower plant densities, weights of maize kernels harvested from SL-and PSL-treated plots were either similar to weights in plots that received soil amendments, both alone (S) and combined with pest control measures (PS), or exceeded the weights for plots with soil amendments (around 3 t/ha to < 5 t/ha in 2019 and 2020).As the weights of maize kernels were confounded by different plant densities, in the means comparison analyses, only in plots with soil amendments only (S p = 0.036) or when all treatments were combined (PSL p = 0.061) significant differences to the untreated control were observed.
In contrast, maize kernel weights in L-treated plots y (<0.5 t/ha in most seasons) were typically similar to the untreated control plots (>0.5 to < 1 t/ ha); this can be explained by the lower seed and subsequent plant numbers.We therefore observed that legume intercropping, applied in conjunction with soil amendments, positively influenced the weights of harvested kernels, in a manner where it compensated for the lower plant densities.We found that legume intercropping as a (negative) main factor became less significant (L main factor p = 0.02393; Figure 4a), whereas soil amendments remained the most important main factor (S main factor p < 0.001, Figure 4).
Furthermore, no significant effects were observed on the weights of harvested kernels from P or PL treatments.Overall, the levels of harvested maize kernels differed significantly between the seasons (p < 0.0001).Moreover, the weights in the control plots (i.e., without treatment) remained stable but low.

Chambezi field site
In two of the four seasons (2018 and 2020), maize plants were severely affected by floods.Unlike cassava, it was not advisable to re-seed maize, as it was in the mid-rainy season.The rainy season is not sufficient to sustain a new maize crop until harvest without irrigation.Hence, surviving maize plants were allowed to continue to grow, in order to assess treatment effectivity in years with severe weather events.Therefore, we analyzed the data for certain parameters separately for flood seasons and normal seasons.

Plant size
Soil amendment (S) was a significant main factor across all years, regardless of weather conditions (p < 0.0119; Figure 5a).Maize plants in S-treated plots were consistently and significantly taller than those in the untreated control plots and all other plots without soil amendments (Figure 5b).This led to statistically significant effects for all treatments in the means comparison analysis of the individual treatments and their combinations (S p = 0.0001; PS p = 0.0001; SL p = 0.0002; PSL p = 0.0001).However, the legume intercropping main factor was also significant (L p = 0.0029).This is possibly due to the negative impact of legume intercropping on maize plant size in 2018, which may have been confounded by the severe floods of that year (e.g., overgrowth by cowpea legumes) (Figure 5a and b).
As expected, highly significant seasonal differences in plant size were observed (season, p < 0.0001).Maize plants were smaller in the two flood- affected seasons (2018 and 2020) than in the seasons unaffected by flooding (2019 and 2021).In 2018 and 2020, mean plant sizes in S-treated plots still grew to sizes of 1.3-1.5 m, while in the controls and all other treated plots, plant sizes were typically <1.0 m with average sizes of 0.7-1.1 m (significant S effect, p = 0.037, Figure 5b).In the two seasons unaffected by flooding (2019 and 2021), average plant sizes in soil-treated plots (S, PS, SL, PSL) ranged from 1.8-2.0m, while plant sizes were ~0.5 m smaller in all other treated plots and the control plots (1.3-1.5 m) (significant S effect, p = 0.0024) (Figure 5b).
No measurable effect of any of the other treatment factors was observed on plant size.In addition, there were no significant effects of replication (p = 0.61).

Plant survival
There were no differences in plant survival that could be attributed to the applied treatments in comparison with the controls in any season.The consistently and, thus, significantly lower numbers of surviving maize plants in L-treated plots resulted from the lower number of maize seeds planted.This led to a highly significant legume intercropping main factor (L main factor p < 0.0001; Figure 6a) and highly significant differences when comparing the number of surviving plants in L-treated plots to the untreated controls (L p < 0.0001; PL p < 0.0001; SL p < 0.0001; PSL p < 0.0001; Figure 6b).
However, highly significant differences were observed in the survival of maize plants between the seasons (p < 0.0001) due to the flooding impact in the two years.Between the two flood-affected years, no seasonal differences were observed (2018 vs. 2020, p = 0.5113).In 2018 and 2021, an average of 60-85 maize plants (30-40%) survived regardless of the treatments, for all plots that were not intercropped with legume cowpeas.In all intercropped plots, on average, 48-55 maize plants survived, i.e., ~ 35-40% of the maize seeds that were sowed.
We observed a significant seasonal effect between the two years unaffected by flooding (2019 vs. 2021, p = 0.00034).Although the overall average plant survival remained relatively constant between these two years, survival numbers were much more variable in 2021 than in 2019, leading to an overall lower number of surviving maize plants in 2019.In 2019, in all single-crop plots, approximately 90 maize plants (±5, ~45%) survived; slightly lower and more variable plant survival was observed in 2021 (75-95, ~37-47%).This pattern was also true for the surviving maize plants in L-treated plots, but to a lesser extent (63, ~48%, plants survived in 2019 vs. 55-62, ~40-47% survived in 2021).
Overall, less than half of the seeded maize plants survived until harvest across all treatments and years.

Weight of harvested kernels per planted seeds
In the Chambezi field experiments, all treatments significantly increased mean kernel weights per planted seeds, reflected in significant main factors for soil amendments (S p = 0.0169), cowpea legume intercropping (L p = 0.0025), and pest control (P p = 0.0109) (Figure 7a).However, as illustrated in Figure 7b, the effects of these treatments increased when they were combined.Treatments led to increased kernel weights in the following order of significance: soil amendments alone (S p = 0.0733), soil amendments with pest control (PS p = 0.0287), soil amendments with cowpea legume intercropping (SL p = 0.0012), and the highest average yields were recorded in plots that received a combination of all treatments (PSL) (p < 0.0001).Single treatments of P or L did not lead to significant increases in kernel weights per planted seeds compared to those in untreated control plots, nor did their combination (PL).Pest control measures and cowpea legume intercropping only resulted in increased kernel weights when used in conjunction with soil amendments.
However, clear differences in kernel weights between seasons were observed (p < 0.0001), due to the impact of flooding on maize yields in 2018 and 2020.
Hence, when separately analyzing seasons with and without flooding, there was no significant difference between overall kernel weights between the two flood-affected seasons (p = 0.18619).In both flood-affected seasons, the only main factor was soil amendment (S) at a statistical significance level of p = 0.0817.Both pest control and cowpea legume intercropping, either applied alone or in combination, did not exert a measurable impact on maize kernel weight per planted seeds.The significance of soil amendment was due to the highly significant effect on kernel weights in plots that received a combination of soil treatments and legume intercropping (SL p = 0.0075; PSL p = 0.0045) (Figure 7a).That is, in flood-affected years, the only plots that yielded any harvestable maize kernels -albeit at a very low level (0.003-0.01 kg/number planted seed) -were those treated with both soil amendments and legume intercropping.Both treatments likely helped to stabilize the soil, thus preventing the floodwater from uprooting the maize plants and providing some nutrients once the floods receded.All untreated control plots, and most plots receiving only pest control or legume intercropping, or both, did not produce a maize yield in either flood season.
When analyzing the data for the two non-flood seasons (2019 and 2021) only, the seasonal effect was still significant (p < 0.0001) with overall kernel weights per planted seeds in 2021 (between 0.01 and <0.04 kg/planted seeds) dropping below those in 2019 (between 0.02 and >0.04 kg/planted seed; Figure 8b).2021 was characterized by drier conditions than 2019.In both 2019 and 2021, all main treatment factors had a significant overall effect on maize kernel weights in the factor analyses (S main factor p = 0.0091; P main factor p = 0.0007; L main factor p = 0.0001).However, these main factors revealed their impact only when they were combined.Significantly higher maize kernel weights were harvested (S p = 0.0443) in all plots treated with soil amendments, with weights increasing as other treatments were added (PS p = 0.0124; SL p < 0.001).The highest kernel weights were achieved when all three treatments were combined (p < 0.001).Despite drier conditions in 2021, the largest relative differences in kernel weights per planted seeds were observed in PSL-treated plots, where the weight of harvested maize kernels was 10 times greater than that in plots without soil amendments (p < 0.002 to < 0.01, Figure 8b).Maize kernel weights in untreated control plots were extremely low and remained approximately constant across all years, except in 2019, where control plots exhibited higher kernel weights (>0.01 to 0.02 kg/planted seeds) than those in all other seasons.As for other field sites, it was observed that despite lower plant densities, the weight of maize kernels harvested in SL-and PSL-treated plots were either similar to, or greater than, those in S-or PS-treated plots.Thus, the beneficial effects of legume intercropping, in addition to soil amendments, on the weights of harvested kernels compensated for the lower plant densities in intercropped plots.Therefore, legume intercropping became insignificant as the main factor (L main factor p = 0.301; Figure 9a).However, as already observed for the above analyses of maize kernel weights per planted seeds, the addition of pest control treatment had a significant effect on harvested kernel weights per area (P main factor p = 0.0058), in addition to the main factor soil amendments (S main factor p = 0.0165).
Consequently, in the means comparison analyses, all treatments that included soil amendments were significant: alone (S p = 0.008), in combination with legume intercropping (SL p = 0.016) and pest control (PS p = 0.001), and when all three treatment types were combined (PSL p = 0.001).
Maize kernel weights obtained from L-or P-plots were typically lower than or equal to those for the control plots, which can be attributed to the lower initial seed number (Figure 9b).Without soil amendments, legume intercropping, or pest control alone did not increase the weight of harvested kernels.
The weight of the harvested maize kernels was much lower in the two floodaffected years (2018 and 2020).Only S-treated plots yielded harvestable maize kernels, but the kernel weights were < 1 t/ha (Figure 9b).In 2019 (not affected by floods), the weights of harvested maize kernels in S-treated plots were 2-3.5 t/ha, which was approximately double the harvested weights for control plots and plots without soil amendments.2021 was drier than 2019, resulting in lower harvested kernel weights (<0.5 t/ha) in plots without soil amendments.In contrast, the weights of harvested kernels in plots with soil amendments were three to five times higher than those in plots without soil amendments (Figure 9b).The soil moisture protection function of the mulch layer was likely also a contributing factor.In both seasons unaffected by flooding, the weight of harvested kernels was the highest in plots where soil amendments were combined with either pest control (PS, kernel weights of 3.5 t/ha in 2019 and 1.5 to 2.5 t/ha in 2020) or with a combination of pest control and legume intercropping (PSL, kernel weights of 2 to 3.5 t/ha in 2019 and roughly 1.5 to a little over 3 t/ha in 2021).In these cases, the combined treatments almost entirely compensated for the lower plant densities in the intercropped plots (Figure 9b).In S-treated plots, harvested kernel weights were lower than, or equal to, those in intercropped plots (2019: S vs. SL and PSL; 2021: S vs. PS, SL, and PSL) (Figure 9b).

Plant size
Soil amendment (S) was the only main factor that significantly affected plant growth (p < 0.0001) (Figure 10a).Most maize plant plots that received compost and mulching were taller than those in untreated control plots, which led to statistically significant effects for the S-, PS-, and PSL-treated plots for all years (S p = 0.00972; PS p = 0.01366; PSL p = 0.00936) (Figure 10b).The lack of statistical significance for maize plants in SL-treated plots is likely due to the large plant size variability, although the plants in these plots tended to be taller than those in the control.The effect of soil amendments was the most pronounced in 2018, when maize in S-treated plots grew twice as tall as in all other plots, including the untreated control (Figure 10b, left panel).In the control plots and most non-S plots, average plant sizes varied between 1 and 2 m, but could reach 2-2.5 m in S-treated plots.This pattern was, however, not true for 2019, when all plant sizes were similar to those in the untreated control plots.Consequently, the overall final plant size across all treatments and the control differed significantly between seasons (p < 0.0001) (Figure 10b).

Plant survival
The significantly lower numbers of surviving maize plants in L-treated plots resulted from the lower number of planted maize seeds.This result was therefore independent of treatment method.No data for maize plant survival were collected in 2020.
There was also a significant pest control effect on surviving plants, due to (unexpected) reduced survival of maize plants in some PS-and PSL-treated plots, mainly in 2021.This led to a significant (negative) effect of P in the split plot analysis (P p = 0.0227; Figure 11b).In 2019, plant survival for all treatments was lower than that in the untreated control plots.This led to statistically significant seasonal differences in the split-plot analysis (season p = 0.002974).
In the L-treated plots, 40-60 plants (30-50% of maize seeds sowed) survived, except in 2021 when only 30 plants survived until harvest in PSL-treated plots.In non-intercropped plots, the average number of surviving maize plants was 40-90 plants (25-50% of the maize seeds sowed).Hence, survival rates were quite variable but roughly similar, regardless of the number of seeds sowed.

Weight of harvested kernels per planted seeds
Kernel weights per planted seeds varied significantly between seasons (season p < 0.0001), with the highest weights recorded in 2018.In all other seasons, kernel weights per planted seeds from S-treated plots were consistently higher than those in the control and plots without soil amendments.This led to a significant effect of soil amendments on kernel weight (S p = 0.0298) (Figure 12a), and a significant S effect in the means analysis (S p = 0.0251) (Figure 12b).Unlike for the other two field sites, we did not observe the same additive effects as treatments were combined at Mvomero, specifically in the combination of soil amendments and legume intercropping.Although kernel weights per planted seeds in SL plots were still much higher than those in L-only and control plots, this kernel weight increase can be attributed to the soil amendments as the kernel weights varied within the ranges observed for soil amendment-related treatments (i.e., S,PS, SL, or PSL).In non-S plots, maize kernel weights were largely unaffected by both pest control and cowpea legume intercropping alone, and by their combination (PL).
In 2018, mean kernel weight levels exceeded 0.06 kg per planted seed in S-treated plots, at 2-4 times the weights of those that did not receive soil amendments (Figure 12b).In 2019 and 2020, kernel weights harvested from S-treated plots were 0.02-0.05kg/per plant.For the first three seasons (2018-2020), mean kernel weights per planted seeds in the control plots remained almost constant at ~0.02 kg/plant.Similar to those of the other stations, the harvested kernel weights in the untreated control plots showed relatively little variation compared to those in the treated plots, specifically in S-treated plots, where kernel weights were both higher and more variable (Figure 12b).The low harvested kernel weights recorded in 2021 can be attributed to severe drought in the area.As for other field sites, the highest maize yields were obtained from plots that had received compost and mulch, except for the PSLtreated plot.In these PSL-treated plots, the exceptionally low levels of harvested kernel weights were due to the low survival of maize plants in 2021, as a result of the adverse effects of the pest control measure.

Weight of harvested kernels per area
The data do not account for the lower maize plant densities in the intercropped plots.As a consequence of the lack of additive effect of soil amendments plus legume intercropping, the overall weight of harvested kernels per L-treated plot was significantly lower because of the lower number of maize plants, as reflected by a significant legume intercropping effect (L p < 0.0001, Figure 13a).
Therefore, unlike for other field sites, it was observed that the weights of maize kernels harvested from SL-and PSL-treated plots were lower than those from plots receiving only soil amendments (S) or combined with pest control measures (PS).Hence, we did not observe a benefit to kernel weights when combining legume intercropping with soil amendments (Figure 13).Single treatments of neither biological pest control (P), legume intercropping (L), nor their combination (PL) had a measurable effect on kernel weight.Consequently, in the mean comparison analyses, only a single treatment of soil amendments (S) was statistically significant (S p = 0.006; Figure 13b).
From 2018-2020, kernel weights from plots treated with soil amendments, either with or without pest control, ranged from ~ 7 t/ha in 2018 to ~ 2.5 t/ha in 2020 (Figure 13b).Kernel weights from plots without soil amendments but with pest control or legume intercropping ranged at or dropped below those for the control plots at 1-3 t/ha.Harvested kernel weights for untreated control plots rarely dropped below 1 t/ha, but reached up to 3 t/ha (2020).The exception to this lower limit was kernel weights recorded in the drought year of 2021, when kernel weights were below those of all other years, and measured ~ 0.5 t/ha.Kernel weights were higher only in plots treated with soil amendments (S, PS, SL, 1-2.5 t/ha).Therefore, kernel weights remained up to 5-fold higher in S-treated plots than in plots without soil amendments (Figure 13b).

Trends across all field sites
All measured parameters varied significantly between seasons owing to unpredictable adverse weather events (i.e., droughts or floods) that occurred at all field sites at different times.Thus, we found unpredictable weather events to be the most important external factor affecting plant size and harvested kernel weights, a proxy for yield.However, our data also showed that under environmental stress conditions, such as floods and droughts, the highest yield per plot and per surviving maize plant was achieved in S-treated plots, albeit at much lower overall levels than those in seasons unaffected by extreme weather events.Soil amendments helped mitigate both flood and drought conditions.The addition of compost and mulch provided essential plant nutrients and organic matter to the soil, while mulch cover also preserved soil moisture under drought conditions and reduced soil erosion in flood situations.
Plant size.Soil amendment was the only treatment that consistently affected plant size at all field sites, although its impact was varied.Maize plants that received compost and mulch as soil treatments grew larger than plants with other treatments.Plant sizes were the smallest at the Chambezi field site and largest at the Mumbaka field site and, also at the Mvomero field site during 2018.
Plant survival.Plant survival was generally low at all stations and across all seasons, with only typically < 50% of the sowed maize seeds surviving until harvest.However, survival was largely independent of treatment.Soil amendments appeared to positively affect plant survival at the Mumbaka field site, but remained < 50%.At the Mvomero field site in 2021, pest control measures even led to reduced survival (as low as 25%) in some plots.
Weight of harvested kernels.Finally, soil amendments significantly increased the harvested kernel weights (by two to six times) at all stations and for almost all seasons.This effect was enhanced by cowpea legume intercropping at the Mumbaka and Chambezi field sites, but not at the Mvomero field site.In contrast, legume intercropping without soil amendments did not affect the harvested maize kernel weights at any station.
Peak harvested kernel weights of 5-7 t/ha in S-treated plots were achieved at the Mvomero and Mumbaka field sites in some years that were unaffected by extreme weather.At the Chambezi field site, however and in contrast to the findings for cassava (Hilbeck et al. 2024b), harvested kernel weights were typically much lower than those at the other field sites.
We did not observe a measurable impact of pest control on kernel weights, when applied as a single treatment.If a positive pest control effect was observed, it was always in conjunction with other treatments (e.g., Chambezi in 2021), but we also observed negative effects (e.g., Mvomero in 2021).
In contrast to S-treated plots, harvested maize kernel weights in untreated control plots often show very little variation, both within and between seasons.Maize yield in the untreated plots was usually low, but fairly stable.Harvested kernel weights in untreated control plots in seasons unaffected by extreme weather events measured ~ 1 t/ha in Mumbaka, 0.5-1.5 t/ha in Chambezi, and 2 t/ha in Mvomero.
We did not observe any significant effects of replication, confirming that the treatments and other environmental (seasonal) conditions were constant across all plots and did not confound any results.

Soil data
When pairwise comparing all treatments individually and in combinations with the respective controls, no significant differences were observed.Data variability, effect and sample size did not reach statistical significance.However, when aggregating the data for the three main factors soil amendments (S), legume intercropping (L) or pest control treatments (P), some changes in soil properties became apparent that we report below and show in Figures 14-16.

Soil organic carbon (SOC).
Lowest, but similar, mean SOC levels were recorded for Chambezi (<0.5%) and Mumbaka field stations (0.5-0.6%), both of which are characterized by sandy soils.Highest mean SOC levels were measured at Mvomero field station (>0.8-1.25%)(Figure 14).When comparing all plots that received soil amendments to untreated control plots, SOC content tended to increase quantitatively over time but likewise increased also the variability of the data.As a result, only in the fourth year in Figure 14.Soil organic carbon content (%) at all three field stations (2017,2019,2021) and three main factor treatments (pest=biological pest control; legume=cowpea legume intercropping; compost=soil amendments).
Chambezi and the second year Mvomero, the increase was statistically significant (p < 0.05).
When comparing all plots that were intercropped with cowpea legumes (L) or received pest control treatments (P), organic carbon increase was only significant in either the fourth year in Chambezi (p < 0.05) or the second year in Mvomero (p < 0.05) -both of which likely were confounded with some plots also having received soil amendments.

Phosphorus (P).
Unlike SOC, P levels of soils at all field stations were very low, ranging from near 0 to 6 mg/kg.Also at all stations, significant increases in p values were recorded in the fourth year on plots that had received soil amendments (mean P content: Chambezi 12.5 mg/kg, Mvomero 25 mg/kg, Mumbaka >30 mg/kg) (p < 0.05) (Figure 15).However, this mean increase of P levels came along with substantially increased data variability unlike on the untreated control plots.A slightly similar trend could be observed on plots that were intercropped with cowpea legumes or treated with organic pesticides although the increase in P content did not reach statistical significance and was likely confounded to some degree with the effects of soil amendments.

Nitrogen (NTot).
Similar to SOC, lowest, but rather similar, N content levels were measured for the sandy soils at Chambezi (0.03-0.06%) and Mumbaka (0.06-0.07%) field stations (Figure 16).At Mvomero field station, slightly higher N levels were measured ranging from around 0.08 to nearly 0.1%.N levels tended to slightly increase on plots treated with soil amendments but did not reach statistical significance and data variability increased visibly.For all other main factor treatments, occasionally in one field season (2 or 3) slightly increased mean values were observed associated with increased data variability but no difference was significant and most mean values were quite similar to the respective control values.

Discussion
Of the three common agroecological practices tested here (organic soil amendment, legume intercropping, and pest control), we found that soil amendments in the form of compost application and mulch cover was the single most effective treatment, leading to an increase in yield in all three agroecological zones.This finding is not necessarily unexpected, because it is widely known that the addition of organic matter (e.g., compost, manure) to nutrient-poor soils exerts an immediately measurable effect (e.g.Jjagwe et al. (2020); Manlay, Feller, and Swift (2007);Thierfelder, Mwila, and Rusinamhodzi (2013)).Such nutrient-poor soils are common in Africa, but agroecological practices vary across African smallholder farming (Constantine et al. 2021;Muluneh et al. 2022).We found only a few published studies investigating the impact of organic soil amendments in the African context.These studies primarily considered manure, but often in conjunction with inorganic synthetic fertilizers (e.g., Thierfelder, Mwila, and Rusinamhodzi (2013)).In contrast, compost is uncommon in African smallholder farming systems.We found that the effect of soil amendments differed between field sites.Of the three field sites, initial harvested kernel weights were lowest in Chambezi; thus, addition of soil amendments led to the highest relative yield increase (maximum 10-fold, up to 3.5 t/ha).However, this yield increase only occurred in conjunction with legume intercropping.The next highest maximum relative yield gains were observed for the Mumbaka field site, with up to a 5-fold increase in kernel weights (maximum of 5 t/ha).The yield increase at Mumbaka was also only achieved in conjunction with legume intercropping.At the Mvomero field site, however, harvested kernel weights were typically higher than those at the other field sites, even in untreated control plots.Therefore, although the highest absolute kernel weight levels were recorded for this site in 2018 (>7t/ha), the relative increase compared to control plots or plots without soil amendments did not exceed three to four-fold.The complementary effects of legume intercropping and soil amendments were not as clearly observed at Mvomero as for the other field sites.
When soil amendments were applied in combination with cowpea legume intercropping, kernel weight levels tended to increase, at least in some years (e.g., 2020) and surpass those of legume intercropping alone.There were no cases where cowpea legume intercropping alone had a significant effect on maize yield, which is in contrast to the marked benefits from soil amendments alone.Our finding that intercropping alone does not increase maize yield proxies agrees with results of other studies, where maize -legume intercropping has even been shown to even reduce yields on occasion, specifically the legume cowpeas (e.g.Kutamahufa et al. (2022); Namatsheve et al. (2020)).Researchers have attributed this to competition of both crops for nutrients, light (shading or overgrowth), and water (e.g., Himmelstein et al. (2017); Kutamahufa et al. (2022); Namatsheve et al. (2020)).Such competition usually arises from poor spatio-temporal arrangements of the intercrops.In our study, we did not observe an increase or decrease in either of the measured yield proxies (i.e., weight of kernels per planted seeds or per plot) as a result of intercropping alone.
Yet, we found the complementary nature of maize and cowpea legume intercropping in conjunction with organic soil amendments a particularly notable outcome of this study.This interaction has rarely been reported in the scientific literature, and if so, it was often, again, in conjunction with inorganic synthetic fertilizers, particularly in Africa (e.g.Kermah et al. (2017); Kiwia et al. (2019); Kutamahufa et al. (2022); Masvaya et al. (2017)).While inorganic nitrogen fertilizers had a significant effect on maize yields, intercropping either had no effect or small effects on maize yields, as observed in our study.However, the authors did not observe an additive effect of inorganic N fertilizer and cowpea intercropping.In this case, the addition of mineral nitrogen fertilizer may have inhibited the nitrogen-fixing capacity of cowpeas, as reported by Malunga, Lelei, and Makumba (2018).
Numerous studies have shown that high levels of nitrogen can reduce root nodule number and inhibit nodule growth and nitrogen-fixing capacity in legumes, although this is dependent on the bacterial strain (Magadlela et al. (2016), and references therein; Ardley, Sprent, and van der Heijden (2021)).However, Kiwia et al. (2019) studied the role of inorganic fertilizer use in cereal-pigeon pea (Cajanus cajan) intercropping and found that adding small amounts of synthetic fertilizers combined with intercropping increased yield, when compared with unfertilized crops.However, studies investigating the combination of legume intercropping with organic fertilizers are limited, although our results suggest that yield increase can be achieved with the addition of organic fertilizers.Himmelstein et al. (2017) conducted a metaanalysis of intercropping in Africa that considered 58 studies; the authors also noted a lack of studies on the impact of intercropping when coupled with other conservation agriculture practices (such as integrated pest management or fertilizer use).
We therefore believe that the knowledge gap of the combination of intercropping and other methods is worthy of further attention, which could offer untapped potential for ecological intensification of crop production.Combining organic soil amendments (such as adding compost, manure, or other forms of organic fertilizers and mulching) with intercropping of legumes also contributes to stabilization or improvement of soil fertility, even under increased yield gains.This leads to higher nutrient extraction due to the addition of nitrogen from legumes.In our study, we showed that the combined effect of soil amendments and intercropping can fully compensate for the lower plant densities in intercropped plots than in mono-cropped plots.
We hypothesize that one reason for the additive effects of soil amendments and cowpea legume intercropping may be the supply of phosphorus and other essential nutrients for nitrogen-fixing microbes and symbionts.This can also be attributed to phosphorus mobilization and the nutrient solubilization of legume roots and root exudates.The phosphorus content in soils is key for the nitrogen-fixing efficiency of symbiotic bacteria associated with legumes and, consequently, the capacity of these legumes to, subsequently, enrich the nitrogen content of nutrient-poor soils (Magadlela et al. (2016), and references therein).Similarly, Li et al. (2007) showed that phosphorus mobilized by a legume crop species (faba beans, Vicia faba) led to increased growth of a second crop species, maize, grown in alternate rows, resulting in large yield increases in phosphorus-deficient soils.Phosphorus deficiency, which was confirmed in all soils at our field sites, can impair nodulation and symbiotic nitrogen fixation (see Figure 14).Hence, the applied soil amendments may have not only added the necessary nutrients for maize plant growth, but also enabled the symbiotic bacteria to carry out nitrogen fixation at higher capacities than in plots without soil amendments.The observed differences may also be attributed to the positive effect of organic amendments on soil physical properties, such as water infiltration and retention, soil aeration, soil color, and soil temperature.Improved physical properties of the soil affect the populations and activities of soil organisms and the associated transformation of plant nutrients in the soil.
These changes in soil physical and biological properties may also explain the observed interaction of legume intercropping with organic soil amendments at the Chambezi and Mumbaka field sites.Both sites share similar sandy soils, which exhibited poorer fertility conditions than soil at in Mvomero (Figure 14).While all field sites exhibited severe phosphorus deficiencies, soils at the Mvomero field site had somewhat higher SOC and N contents than both other field sites.
Furthermore, rainfall patterns and the subsequent water content of soils have been shown to interfere with nitrogen availability for crops under different fertilizer treatments.Falconnier et al. (2020) showed that while maize benefited less from an increase in atmospheric CO 2 without nitrogen fertilizers for nutrient-poor soils, maize was less affected by higher temperatures and decreasing rainfall (i.e., drought) but was more affected by increased rainfall (i.e., floods) because nitrogen leaching was more critical (Falconnier et al. 2020).The authors found that climate change and nitrogen input interactions have strong implications for adaptation approaches across Sub-Saharan Africa, and that climate change impacts could be reduced through an increase in maize production with balanced nutrient management (Falconnier et al. 2020).This may also hold true for organic soil fertility measures releasing nitrogen either from compost, manure, or biologically fixed nitrogen, and may explain the differences in kernel weights observed between locations.This may also support our observations that kernel weights in untreated plots were low but showed little variation as a result of environmental conditions.
We also found that pest control did not affect maize growth or kernel weights when applied as a single treatment.Pest control occasionally had a positive effect on maize yield when combined with other treatments, for example at the Chambezi field site in 2021.However, we also observed significant adverse effects on plant growth and kernel weights at the Mvomero field site in 2021.We, therefore, cannot identify a systematic effect of pest control measures, although this was not necessarily due to the lack of efficacy of the ingredients used.All botanical ingredients used (i.e., chili, aloe vera, garlic, and neem) in the applied biocontrol solutions are well known to adversely affect pests (e.  2021) and references therein).We hypothesize that the reason for the inconsistency in responses lies in the timing of the pest control application and the lack of knowledge regarding the targeted pest species.
It is fairly common in farming communities that only a few pest species (usually the most damaging) are known, with their life cycles and population dynamics being poorly understood.As a result, knowledge on the most effective pest control compounds and the best timing and pest life stage for their application is lacking.For example, while the fall armyworm is often identified as a pest, these organisms are typically only identified in their late larval stage (i.e., when their feeding damage on maize leaves becomes visible), such that it is too late for treatment.By this time, damage has already been caused by younger larval stages inside the maize whorl.Treatment of maize plants when larvae are large and feeding on leaves is no longer effective, because the sensitive tissue inside the whorl has been damaged by pest feeding activity of younger larvae already, which can result in plant death and/or yield reduction.
Hence, the effective use of pesticides prepared from locally available resources (e.g., ash) also requires knowledge about the timing of application and targeted pests.Without this knowledge, the success of biopesticides is at best coincidental; however, it can also often be ineffective and even adversely affect plant growth or promote insect infestation if applied at too high rates or frequencies (as we observed).There is still a lack of knowledge regarding the suitability of botanical compositions to target specific insect pest species.
Therefore, common recipes for the preparation of biopesticides typically include ingredients with a broad spectrum of activity.In a complementary sister project, neem, an insecticidal compound used in the botanical pest control mix applied in this study, has been tested in more detail.However, as the preparation and application of these biopesticides is time consuming and expensive, our results suggest that it may be more effective to focus on other proven means of pest control: crop diversity (e.g., through legume intercropping), aiming for crop varieties that are less susceptible to pests and diseases, and boosting soil fertility to ensure healthy plants that can better defend themselves against herbivore pests (Bautze, Dietemann, and Singh 2022).Overall, pest damage was rarely a yield-limiting factor in this study.

Agroecological practices
The most reliable return on the efforts of applying the tested agroecological practices regarding yields came from soil amendment practices.However, soil amendments offer many more benefits in addition to increases in yields, e.g.soil water conservation, protection from flood and drought, soil health and supporting a diverse microbiome necessary for organic matter recycling.But the best results in terms of yields were achieved when maize was intercropped with cow pea legumes.Often, maize yields on intercropped plots with soil amendments exceeded those on plots where only soil amendments were applied and always those on intercropped plots without soil amendments.This combined effect of cow pea intercropping with soil amendments should be recommended and become routine practice in agroecological production systems.Legume intercropping also provides additional benefits, like nutritious food crops and added biodiversity for other beneficial organisms that control pests and diseases.In contrast, labor intensive and costly pest biocontrol practices delivered variable outcomes -they require knowledge on type and ecology of pests to be effective which is lacking.A lot more must be invested in terms of research and extension consulting on crop entomology to benefit fully from pest biocontrol practices.

Application of ICT support tools
The AEA application was developed in conjunction with the data gathering and storage application (AER).The AEA application was coupled to the AER to allow for automated and rapid analyses (graphical and statistical) of the field datasets.The automatization of data gathering, visualization, and statistical analyses enabled us to draw conclusions and practical recommendations from the collected data without the need for specific programming skills.However, basic statistical knowledge is required to conduct meaningful analyses and correctly interpret results.We suggest that coupling the two ICT tools could be useful also in other study areas, because this approach facilitated the complete workflow from conceptualization and conduction of experiments and data acquisition by local farmers and field staff to graphical and statistical analyses and practical recommendations by researchers.The AEA application was based on open-source software and built to allow adaptation of key aspects to other settings (e.g., link to input data, choice of experimental design, selection of response and treatment variables).This would require adjustments to the underlying R code, depending on the conceptual differences between our experiment and the different target setting.We avoided hard coding (i.e., the embedding of data or code that is not generally valid, but experimentspecific) where possible during the development of the AEA application.Therefore, most parts of the core application (app.R) may be re-used for analysis of other similar experiments, provided that the global parameters (e.g., the link to the database or the names of treatment and response variables) are properly defined in the provided parameter file (GetParameters.R).For more details regarding codes and design, we refer readers to the methodology publication of this series (Hilbeck et al. 2024a).

Figure 8 .
Figure 8. Chambezi field site: mean maize kernel weights per planted seeds recorded in different seasons: a) flood-affected seasons in 2018 and 2020; and b) seasons not affected by flooding in 2019 and 2020.