Optimizing site-specific variety, sowing density and nitrogen fertilizer recommendations for maize in the Nigerian Savannas using field experiments and modelling

Maize (Zea mays L.) has over the years become an important crop in the Nigerian Savannas including the semi-arid Sudan Savanna zone where production was initially not feasible. The annual maize output in the country changed from 1.06 million tonnes in 1976 to about 11.6 million tonnes in 2017, but the increase is due to expansion of area and not the much-needed intensification. The average yield per hectare has been below 2 Mg ha-1 since the 1970s, although yields >7 Mg ha-1 have been reported in research stations and best farmer fields. The reasons for the low per hectare yield have been attributed to the inherently poor soils, frequent droughts, pests & diseases and most importantly to lack of adherence to improved agronomic practices and use of improved inputs like fertilizers and seeds. In recent years, new maize varieties that are tolerant to most of the biotic and abiotic constraints have been developed for the Nigerian Savannas by the International Institute for Tropical Agriculture (IITA) and its partners. Several agronomic technologies have also been developed to increase the productivity of these varieties with a view to increasing maize yields. Dissemination of such varieties and technologies and their subsequent adoption requires setting up expensive and time-consuming multi-locational trials for evaluation. Selection of appropriate varieties across agro-ecologies and adoption of appropriate agronomic practices like optimum sowing density and site-specific fertilizer applications will be the key requirements for increase in production per unit area.

Crop simulation modeling offers an opportunity to explore the potential of new varieties and crop management practices in different environments (soil, climate, management) prior to their release. Since most models have been developed elsewhere in Europe and USA, their use outside their domain of development requires a great deal of data for their calibration and evaluation. In addition, the shortage of technical know-how makes the use of those models more difficult especially by policy makers, farmers, technologists and extension agents.

Overall, this research was conducted to evaluate the ability of a dynamic crop simulation model (DSSAT-CSM-CERES-Maize model) in matching maize varieties to the Sudan and Northern Guinea Savannas of Nigeria. The research also aims to use the model in making agronomic recommendations with respect to optimum sowing densities of the different varieties produced in the Nigerian maize belt. To achieve the set aims and objectives, data sets were collected from three different sources. Two of the data sets were collected by setting up field experiments while the third was collected from maize breeders in IITA.

The first set of experiments were conducted in the rainy and dry seasons of 2016 in four research stations in the Nigerian Savanna. In the experiments, 26 maize varieties were planted under near-optimal environments (moisture and nutrient non-limiting). Growth, phenology and yield characteristics of each variety were measured with a view to developing “virtual” genotypic characteristics and incorporating it into the model. In addition to crop data, detailed soil data was collected from two profiles pits dug in each location together with daily weather data (minimum and maximum temperatures, daily rainfall and solar radiation). The purpose of these experiments was to calibrate the existing varieties and agro-ecological conditions of our trial sites into the model.

The second sets of experiments were conducted in the rainy seasons of 2016 and 2017 across farmer fields in the Sudan and Northern Guinea Savannas of Nigeria. The experiments consisted of 10 maize varieties (different varieties were used in the two agro-ecologies) planted under three different sowing densities (2.6, 5.3 and 6.6 plants m-2). In each agro-ecology the experiments were conducted in 30 farmer fields in both years, data was collected on the response of the varieties to the elevated sowing densities as well as soil and weather records from each farmer field and trial location. These experiments were conducted to evaluate the response of diverse maize varieties to elevated sowing density and to evaluate the ability of the model to predict the response of increased sowing density.

The third data-set was collected from long-term varietal evaluation experiments conducted by breeders before varietal release. These data-sets were used to demonstrate how available information from breeder trials can be used to develop genotype specific parameters (GSPs) for use in CERES-Maize model.

Using the data from the detailed calibration experiments and the breeder evaluation experiments, two sets of GSPs for 26 current maize varieties produced in the Nigerian maize belts were developed. Comparison of the two different data sources showed that GSPs generated from the detailed experiments were more accurate, but the breeder evaluation experiments could also be used but implied lower accuracy. The sequential approach method used in the genotype calculator (GENCALC) tool for calculating GSPs in the model was also optimized. Additionally, we used the detailed experimental data to evaluate the ability of the model in predicting observed genotype by environmental interaction (GEI). The model accurately predicted the observed GEI and the predicted grain yields were used to rank the stability of the different varieties across different environments. Long-term weather data (1992-2017) from the dry and wet savannas were then used to conduct seasonal analysis. This revealed that, contrary to current recommendations, intermediate maturing varieties which were suggested only for the wet savannas can also be grown in the dry savannas.

Data from the sowing density experiments were used two-fold. First, a detailed analysis of the response of the maize varieties to elevated sowing density was conducted. A heterogenous covariance structure (Eberhart-Russel factor analytic model (FAM)) was used to model the genotype by environment by density (G×E×D) interaction. From this analysis, it was established that higher yields are expected with increasing sowing density only in optimal environments. The results also show that, under optimal environments maize varieties can be sown above 6 plants m-2 which is beyond the highest density tested and beyond the current recommendation. Second, the observed grain yields from farmer fields were used to evaluate the already calibrated varieties in the CERES-Maize model. The calibrated and evaluated model was then used to provide sowing density recommendations for the different maize varieties under varying nitrogen fertilizer rates. Detailed bio-physical and economic analyses were conducted using the long-term weather records. The model simulations revealed that, early and extra early maize varieties could be planted under sowing densities of up to 8.8 plants m-2 under high Nitrogen (90 kg N ha-1) in the Sudan Savanna providing higher grain yields and money returns per hectare. Sowing density of 6.6 plants m-2 and 90 kg N ha-1 was shown to produce the highest money returns to family labour. For the late and intermediate varieties in the Northern Guinea Savanna, sowing density of 6.6 plants m-2 and N fertilizer application of 120 Kg N ha-1 produced the highest grain yields and money returns per unit land. But highest returns to family labour was simulated for sowing density of 5.3 plants m-2 and N fertilizer rates of 120 kg N ha-1.  These simulated results show that for optimum economic returns, small holder farmers need to increase the planting density of maize in reduced areas of their farms and apply all the N fertilizers they can afford on that area. The remaining area  can then be used for legumes and other low input crops.