Feed the crop, not the soil! : Explaining variability in maize yield responses to nutrient applications in smallholder farms of western Kenya

Crop productivity intensification in smallholder farming systems of sub-Saharan Africa (SSA) is urgently required to improve food self-sufficiency. Increased fertilizer use can address nutrient deficiencies that limit crop productivity in SSA. There is however large uncertainty in crop yield responses to fertilizer applications on farmer fields. This uncertainty has been linked to strong heterogeneity in soil fertility between and within farms. Fertilizer recommendations that account for this spatial heterogeneity are therefore required to better advise farmers, reduce investment and environmental risks for sustainable crop productivity intensification. The main objective of this study was to better understand and explain patterns of maize yield and yield responses to fertilizer applications under heterogenous fertility conditions in smallholder farming systems. This would allow for improved targeting of fertilizer applications, and enable better prediction of expected crop yield response to fertilizer use. A series of on-farm experiments assessing maize yield response to fertilizer application under variable soil fertility conditions were established on 23 farmers’ fields in Siaya, western Kenya across multiple seasons. Prior to experiment establishment, farmers were extensively interviewed to obtain information on past crop and nutrient management practices in selected fields. The experiment used was comprised of nutrient omission trials (NOTs) on farmer fields in Siaya to assess patterns of maize yield response to fertilizer applications of 150 kg ha-1 nitrogen (N), 40 kg ha-1 phosphorus (P) and 60 kg ha-1 potassium (K). In Phase 1, plots with treatments including control, PK, NK, NP and NPK were repeated for 7 consecutive seasons in the same plots. In Phase 2 of the experiment, a second set of NOTs including PK, NK, NP and NPK were established in every plot on 6 fields that were previously part of Phase 1. On 13 other fields from Phase 1, all plots received NPK in Phase 2. Yields and above ground biomass were measured every year, soil samples were taken in 2013, 2014, 2016 and 2018 and plant samples were taken in 2016 and 2018. Spatial-temporal patterns in yield and yield responses were studied and compared with soil and farmer characteristics. Yield response to soil nutrient supply was studied with the QUEFTS model, and the RC-P model was used to study fate of fertilizer P. Nutrient balances were calculated. The frequency and magnitude of maize yield response to fertilizer N, P and K varied strongly over space and time, yet observed patterns were not adequately explained by soil chemical parameters or texture. Fertilizing with N, P, and K substantially reduced observed spatial-temporal variability in maize yield response, and resulted in consistently enhanced maize yields. All fields were responsive to N, most fields to P and only 7 to K. On average, NPK yields were about 5 to 5.5 tons ha-1 in the short- and long rainy seasons respectively. Application of only NP or NK resulted in strongly declining yields within a few seasons, with large differences between farms in resilience of soil P and K stocks. Based on observed spatial-temporal patterns, we concluded that blanket fertilizer recommendations in such farming systems result in low fertilizer use efficiencies. We further concluded that current methods for soil analysis do not adequately explain the observed variation in maize yield response to application of N, P and K fertilizers under the highly variable soil fertility conditions encountered in smallholder farming systems. Accounting for past manure application in Phase 1 of the experiment improved our ability to explain the variation in maize yield response to fertilizer application. Mean maize yield response to N, P and K application was 2.8, 1.1 and 0.6 t ha-1 in fields with animal manure previously applied, and 2.3, 3.0 and 1.6 t ha-1 in farms without past manure applications over 7 cropping seasons. Differences in maize yield response in fields with and without past manure applications were mainly related to enhanced soil phosphorus (P) and potassium (K) supply, and larger recovery of applied nitrogen (N) in fields with manure previously applied. Based on these findings, we concluded that the strong influence of past animal manure application on yield response to fertilizer applications merits the inclusion of past manure application as a co-variate in analysis of yield response data from smallholder cropping systems of SSA. The Quantitative Evaluation of the Fertility of Tropical Soils (QUEFTS) model did not adequately estimate crop yield responses to fertilizer applications under variable soil fertility conditions. This was linked to poor estimation of soil N, P and K supply based on current relationships for potential soil nutrient supply in the QUEFTS model. In particular, soil organic carbon (SOC) and P-Olsen were poor indicators of crop N and P uptake from the soil. Maize grain yield in unfertilized control treatment plots provided better estimates of potential soil N, P and K supply, resulting in improved predictions of maize yield response to fertilizer applications. These findings suggest that the standard soil parameters analysed do not accurately inform on the soil fertility status of the field and are of little use for smallholder farmers. Improved relations for estimation of potential soil nutrient in QUEFTS are required for better prediction of expected maize yield response to fertilizer application under variable soil fertility conditions. Maize crops in strongly nutrient-depleted soils responded strongly to balanced NPK fertilization, with yields comparative to long-term means within three seasons. Placement of P fertilizer strongly improved recovery, reducing the need for larger soil P stocks on soils that will typically develop a large insoluble P pool under P fertilization. The RC-P model provided insights in long-term recovery of P and could describe the observed P uptake patterns reasonably well. We concluded that strongly nutrient depleted tropical soils such as those in Siaya with high clay contents that are typical for western Kenya, do not require prior investments to rebuild nutrient stocks and soil organic matter to substantially increase crop yields to 5-5.5 t ha-1. This has important implications for crop productivity intensification in SSA as a large proportion of soils under cultivation are strongly nutrient depleted, and earlier approaches have suggested the need for costly and capital intensive soil fertility replenishment. Results in this thesis clearly demonstrate that sustainable intensification of crop productivity on smallholder farms of SSA is very well possible on all fields under good management, even when soils are strongly nutrient depleted. However, the need for P and K fertilizers and amounts applied should be tailored to specific field conditions to reduce farmer costs in the short term. Accounting for past farm management and assessment of current yields under minimal or no fertilizer applications provides a means for improved targeting of fertilizer applications at the farm level. In the long term, farmers should aim for balanced fertilization to prevent mining of soil stocks. Simplified decision support tools that use field level information to develop improved estimates of fertilizer N, P and K requirements based on refined relationships between soil nutrient supply, nutrient uptake and yield, are required to derive fertilizer recommendations in future.