30 October 2026 | 16:00 - 19:00
Nitrogen fertilization in temperate arable systems must reconcile two competing requirements: providing crops with enough N to express their yield potential, and avoiding the environmental losses caused by excess mineral N in the soil. In Wallonia (Belgium), this trade-off is handled through plot-scale N balance approaches developed within the REQUASUD advisory network.
One of the most uncertain terms of this balance is the contribution of soil organic matter through in situ net N mineralization, which still relies on bulk descriptors (total organic carbon and total nitrogen) and on empirical coefficients. The aim of this thesis is to improve the estimation of this term, and through it the reliability of N fertilization advice in Wallonia.
The work is built on three complementary case studies.
The first uses a two-year monitoring of three plots after grassland ploughing to evaluate anaerobic incubation as a proxy for in situ mineralization. Potentially mineralizable nitrogen (PMN) was strongly correlated with in situ mineralization estimated with the LIXIM model (R² ≈ 0.91, n = 6) and differed from it by less than 10% in five of the six plot-years; the exception, a weed-invaded plot, was matched by a simple mineral N balance.
The second case study uses a seven-year, 121-plot regional dataset to develop and compare predictive models of in situ mineralization. Multiple linear regression (MLR) and a Bootstrap Forest model (BFM), both integrating PMN, permanganate oxidizable carbon (POxC), pH and the TOC/N and TOC/clay ratios, reduced the RMSE of in situ mineralization estimates by 41% and 51%, respectively, and the interquartile range of relative prediction errors by 46% and 60%, compared with the current REQUASUD method, without substantially changing average N rates at the regional scale.
The third case study investigates the use of visible and near-infrared spectroscopy (Vis–NIRS) on 1,171 Walloon soils to provide rapid estimates of PMN and POxC. POxC was predicted with an accuracy already consistent with quantitative use on the full validation sets (CCC = 0.90–0.94; SEP = 101.6–120.8 mg kg⁻¹; RPIQ = 2.8–3.3), rising to CCC = 0.96, SEP = 80.9 mg kg⁻¹ and RPIQ = 4.3 under routine quality-control screening, which supports routine deployment. PMN was predicted with moderate accuracy (CCC ≤ 0.84; SEP ≥ 58.7 kg N ha⁻¹; RPIQ ≤ 2.3), and about half of its prediction error variance arises from the uncertainty of the incubation assay used as reference.
Taken together, these results support a shift away from indirect estimates based on bulk pools and fixed coefficients, towards a tiered advisory framework combining biologically meaningful indicators (PMN, POxC), multivariate models, and high-throughput spectroscopic measurements. The main expected operational benefit is not a change in regional fertilizer use, but a reduction of the large plot-level errors that drive both yield losses and nitrate leaching.
Prof. Jacques Mahillon (UCLouvain) (Supervisor)
Prof. Richard Lambert (UCLouvain) (Co-Supervisor)
Prof. Stephan Declerck (UCLouvain) (Chairperson)
Prof. Guillaume Lobet (UCLouvain) (Secretary)
Dr. Christophe Vandenberghe (ULiège, Gembloux Agro-Bio Tech)
Dr. Jean-Pierre Destain (CRA-W and ULiège, Gembloux Agro-Bio Tech)