When two of the world’s most widely trusted crop simulation engines were asked the same question about feeding sub-Saharan Africa, they agreed on the direction but disagreed on the numbers — and that disagreement may be the most instructive finding to emerge from a new Nigerian study. In research published in Discover Agriculture, Blessing Funmbi Sasanya and Ayebaemi Akono of the University of Port Harcourt deployed APSIM, Australia’s Agricultural Production Systems Simulator, and AquaCrop, the Food and Agriculture Organization of the United Nations’ water-productivity model, to grow maize and potato virtually under 27 different combinations of irrigation, nitrogen and weed management. No seeds were sown, no rain was waited for and no weeds were hoed. Instead, more than four decades of climate records and laboratory-characterised soil profiles were pushed through both models to test how much yield is won or lost through the timing of water, the dose of fertiliser and the diligence of weeding — and whether two independently built models would even tell the same story.
The stakes could hardly be higher. Most farmers across sub-Saharan Africa depend on rain-fed agriculture, leaving crops hostage to rainfall that arrives erratically or not at all, and water deficits during critical growth stages translate directly into yield losses. Nitrogen, the nutrient that drives productivity more than any other, is chronically mishandled — lost to leaching, volatilisation and denitrification when applied carelessly, or omitted entirely when fertiliser prices climb beyond the reach of smallholders. Weeds, flourishing in tropical conditions, compete fiercely for light, water and nutrients, while herbicide overuse carries human-health and environmental costs. Crucially, these three levers are interdependent: inefficiency in one compounds losses in the others. Field experiments designed to disentangle their combined effects are expensive, slow and difficult to scale, which is precisely why the researchers turned to simulation. Crop models, they argue, allow entire agronomic experiments to be completed in hours of computation rather than seasons of labour, offering a decision-support framework for regions where data are scarce and growing conditions unforgiving.
The two engines, though built for the same purpose, work in fundamentally different ways. APSIM, developed by Australia’s Agricultural Production Systems Research Unit, is a modular, process-based framework that steps through crop growth day by day, simulating phenological progression, photosynthesis, biomass accumulation and yield formation alongside soil water, nitrogen and carbon cycling and management interventions such as fertilisation, tillage and irrigation. Crucially, it represents weed growth mechanistically, so crop–weed competition can be quantified directly within the simulation. AquaCrop, engineered by the FAO for streamlined, water-focused analysis, derives canopy cover dynamically from leaf expansion, senescence and maturity timing, partitions evapotranspiration into soil evaporation and crop transpiration, computes above-ground biomass from water productivity and converts it into yield through a harvest index. Weeds enter AquaCrop only qualitatively, through control categories such as perfect, moderate and poor. That architectural difference — one model mechanistic and nutrient-centric, the other water-centric and deliberately simple — set up a natural experiment in how model structure shapes predictions when both receive identical inputs.
The virtual experiment was anchored in real conditions. The study site, the University of Port Harcourt Research and Teaching Farm in Rivers State, Nigeria, lies in a humid zone receiving about 2,293 millimetres of rainfall a year at an average temperature of 28 degrees Celsius and 75 percent relative humidity. Daily meteorological records from NASA’s POWER database, spanning 1981 to 2023, supplied rainfall, temperature, humidity, wind speed and sunshine data, while undisturbed soil cores collected to a depth of 1.2 metres were analysed for hydraulic conductivity, particle size distribution, pH, bulk density, total nitrogen, available phosphorus, exchangeable potassium and organic carbon. Both crops were sown on 15 March in the models, with maize maturing 130 days after planting and potato 120. The researchers then built a 3 × 3 × 3 factorial arrangement within a randomised complete block design: three nitrogen rates — 100, 50 and 0 kilograms per hectare for maize, and 150, 75 and 0 for potato — three weed regimes comprising fortnightly removal, monthly removal and a single mid-season weeding at day 65 for maize and day 60 for potato, and three irrigation thresholds triggered when 40, 70 or 100 percent of available soil water had been depleted. In AquaCrop, fertility and weed control were encoded through the model’s soil-fertility and weed-management parameters, while APSIM applied urea at sowing and implemented scheduled weed-biomass removal events, and analysis of variance at the five percent level, with Duncan’s Multiple Range Test for mean separation, quantified how the factors shaped simulated growth.
The maize simulations delivered results with immediate economic implications. Peak above-ground biomass reached 21,435 to 21,467 kilograms per hectare by the 130th day after planting, and maximum grain yield climbed to 10,755 kilograms per hectare — achieved not only under intensive management but also in scenarios that let the soil dry severely between irrigations. With 100 kilograms of nitrogen per hectare and fortnightly weeding, triggering irrigation at 70 percent depletion or waiting until 100 percent depletion produced yields statistically indistinguishable from the other top-performing regimes, suggesting farmers could cut pumping costs and labour without measurable penalty at harvest. Canopy cover, by contrast, barely moved across treatments, ranging only from 0.892 to 0.900, while simulated biomass at the 99-day mark spanned roughly 8,200 to more than 16,500 kilograms per hectare. The penalties were equally stark: scenarios in which weeds were cleared only once per season sank to the bottom of the yield table, and the fully managed control significantly outperformed every combination pairing nitrogen scarcity with relaxed weed control on water productivity.
Potato told a subtly different story. Above-ground biomass peaked between 9,395.85 and 9,400.35 kilograms per hectare at 120 days after planting, and maximum dry root yield reached 9,020.95 kilograms per hectare under 150 kilograms of nitrogen per hectare with fortnightly weeding and irrigation at the 70 or 100 percent depletion thresholds. Nitrogen emerged as the decisive lever: treatments pairing delayed irrigation with adequate fertiliser out-yielded those with timely watering but suboptimal nitrogen, while every zero-nitrogen scenario clustered near the bottom of the rankings. By the 90-day mark, biomass ranged from roughly 2,783 to more than 7,257 kilograms per hectare, and some moderately stressed treatments matched or slightly exceeded the fully managed control by harvest — evidence that potato growers in water-limited settings can defer irrigation provided nutrition is secured. The simulated yields also sit comfortably within ranges reported in the published literature, from fertilised and irrigated Mediterranean potato systems to pre-release varieties trialled in Kenya, lending external credibility to numbers generated entirely in silico.
The heart of the study, however, lies in the head-to-head comparison of the two models. Pearson correlation coefficients between their outputs ranged from −0.36 to 0.78, with relatively strong positive correlations of 0.63 to 0.78 for biomass, dry yield and water productivity at most growth milestones — evidence that both engines responded to management in the same direction. But correlation, the authors stress, measures co-variation rather than numerical agreement. Maize grain yield at maturity showed only a weak correlation of 0.41 even though an independent t-test found no significant difference between the models’ predictions; the root-mean-square error stood at 3,847.41 kilograms per hectare and the mean absolute error at 2,870.80 kilograms per hectare. Potato root yield correlated more strongly at 0.65, yet the models differed significantly. Error metrics ballooned elsewhere: maize biomass RMSE climbed from 427.77 kilograms per hectare early in the season to 8,866.11 at harvest, and potato values spanned 70.55 to 8,874.98 kilograms per hectare. Willmott’s index of agreement was generally low, indicating that the models partially co-varied rather than genuinely concurred — the same compass heading, but markedly different readings of the distance travelled.
The divergence traces directly to model architecture. AquaCrop, designed for water-limited systems with minimal input requirements, proved the more reliable estimator of canopy cover and water productivity, consistent with earlier calibration studies in Egypt, Ethiopia and Australia, while APSIM, with its explicit nitrogen modules and mechanistic crop–weed competition, remains the stronger instrument for nutrient dynamics and cereal productivity. The weed gap matters too: because AquaCrop translates weed pressure into qualitative adjustments to crop growth parameters while APSIM grows and competes actual weed biomass, identical weeding schedules do not translate identically into crop stress. The findings align with previous work arguing that model ensembles — running several models and pooling their predictions — reduce uncertainty in tropical agriculture more effectively than any single model, and this study offers a concrete demonstration of why: two models given the same weather, soils and management can agree on trends while disagreeing substantially on magnitude.
For farmers, the actionable message is that maximum input does not guarantee maximum return. The reduced-input scenarios — Treatment 8 for maize and Treatment 16 for potato — delivered simulated yields statistically comparable to the most intensive regimes, hinting that carefully timed deficit irrigation and moderate nitrogen rates could shave input costs without sacrificing productivity. In a region where fertiliser can consume a large share of a smallholder’s budget and irrigation water is rationed by rainfall rather than choice, such savings are far from trivial. The authors are equally candid about the limitations: the models were calibrated but not validated against fresh field observations, so the outputs describe model behaviour rather than verified real-world yields, and predictive accuracy across other environments remains untested. Future work, they write, should anchor the framework to comprehensive field measurements at representative agricultural sites before it graduates from research tool to farm-level recommendation.
Even with those caveats, the study sketches a template for how data-scarce agricultural systems can modernise. A 27-scenario experiment that would demand years and considerable expense in the field was executed computationally, ranking management priorities, generating hypotheses for future trials and exposing model blind spots in a single pass. Root crops, the authors note, have been sparsely represented in recent agronomic modelling, and pairing maize with potato within one unified factorial framework is itself a methodological advance. As climate variability tightens its grip on rain-fed agriculture across Africa, the ability to rehearse seasons digitally — testing whether to irrigate at 40 or 70 percent depletion, whether 50 kilograms of nitrogen suffices, whether monthly weeding is one interval too lax — could become as fundamental to farming as the rain gauge. The two models, in the end, emerge less as rivals than as complementary lenses: one tuned to water, the other to nitrogen, and together mapping the path toward sustainable intensification.
Subject of Research: Comparative simulation of maize and potato yield responses to irrigation scheduling, nitrogen application and weed management using the APSIM and AquaCrop crop models under resource-constrained conditions in sub-Saharan Africa
Subject of Research: Agriculture
Article Title: Model-based comparison of crop yield responses to water, nitrogen, and weed management using APSIM and AquaCrop
Article References: Sasanya, B. F., & Akono, A. (2026). Model-based comparison of crop yield responses to water, nitrogen, and weed management using APSIM and AquaCrop. Discover Agriculture, 4(1), Article 226. https://doi.org/10.1007/s44279-026-00701-5
Image Credits: AI Generated
DOI: 10.1007/s44279-026-00701-5
Keywords: Modelling and simulation, Maize, Potato, Agricultural productivity, Resource conservation, APSIM, AquaCrop, Crop simulation models, Weed management, Nitrogen management, Irrigation scheduling, Sub-Saharan Africa
Cite Scienmag News
APA MLA Chicago
Alan Morgan. (August 31, 2026). APSIM and AquaCrop simulations weigh water, nitrogen, and weed effects on yields. Scienmag. https://scienmag.com/apsim-and-aquacrop-simulations-weigh-water-nitrogen-and-weed-effects-on-yields/
Copy citation Download RIS
Tags: agricultural modeling for sub-Saharan AfricaAPSIM and AquaCrop water and nitrogen managementclimate data in crop modelingclimate data integration in crop simulationsclimate resilience in rain-fed farmingcomparative analysis of crop simulation enginescomparison of agricultural simulation enginescrop simulation modelsdrought resilience in crop modelingeffects of water timing on crop productivityimpact of irrigation and fertilization on crop yieldsmaize and potato yield predictionmodeling weed effects in agriculturenitrogen fertilization and crop growthrole of nitrogen and water in sustainable agriculturesoil profile analysis for crop simulationsoil profile analysis in crop modelingsub-Saharan African rain-fed agriculturevirtual crop growth and yield predictionwater productivity in agricultureweed management effects on maize and potato yields
