Year: 2024
Type: dissertation
Abstract: Since the Green Revolution in the 1960s mineral fertilisers have contributed to a great extent to increase food production, which was required to meet the world's growing food demand. Today, about 174 Tg of nitrogen fertiliser is used globally each year to produce our food. However, more than half of it is lost to the environment, causing severe air and water pollution, biodiversity loss, and climate change. This is also a major issue in areas with intensive vegetable production where high fertiliser doses are applied to obtain high yields. Improving the nutrient use efficiency without compromising yield is very challenging as a complex interaction between the weather, and biological and geophysical soil processes influences the nutrient uptake. Numerous tools exist to assist farmers in determining the correct dose for their vegetables. However, these tools are often not capable of taking variable soil conditions or management into account. On the other hand, governments are imposing more strict policies on farm management to reduce agricultural nitrogen pollution in order to meet European environmental regulations. To reduce the nitrogen pollution of our surface and ground water by vegetable cultivation this PhD research aimed to improve the current advisory systems for fertilisation using process-based simulation models, local soil information and weather data. Firstly, the effectiveness of the commonly used 'Kulturbegleitende-Nmin-Sollwerte' (KNS) advisory system and the Flemish residual soil nitrate (RSN) policy were assessed by analysing simulation data for 42 growing seasons of a cauliflower-leek rotation. To this end, the ECOFERT simulation model was calibrated for a Flemish field and simulations were performed for different fertilisation application rates and methods. The results indicated that row application of the nitrogen dose recommended by the KNS system based on soil analyses could reduce nitrate leaching by 18 % compared to a broadcasted fixed application rate. This was mainly attributed to increased cauliflower dry biomass (+18 %) and N uptake (+13 %), while the effect on the leek production was negligible. Despite the reduced nitrate leaching, the simulated average RSN was still 35 kg NO3‑N ha‑1 above the legal limit of 85 kg NO3‑N ha‑1. The ineffectiveness of the system was attributed to the fixed estimates for the daily mineralisation rate and N uptake, which could vary considerably depending on the weather conditions. The main goal of this PhD was to reduce the environmental impact of vegetable cultivation by elaborating improved decision support tools for fertilisation. To this end, a decision support system (DSS) based on a process-based simulation model was elaborated. First, the ECOFERT simulation model was recalibrated and validated on an extensive dataset from fertilisation experiments conducted across different research centres in Flanders in 2019 and 2020. After calibration, the ECOFERT model could accurately simulate leek dry biomass (DBM) and N uptake on most studied fields. Despite the substantial variation in the accuracy of the soil mineral nitrogen simulation between fields the results were in line with other multi-site studies on process-based crop models for arable cropsand considered sufficient to include the model in a decision support system. Subsequently, the ECOFERT model was used to determine the fertilisation application rate that maximized yield while limiting the risk to violate the legal RSN thresholds stated by the Flemish manure action plan (MAP). The proposed ECOFERT DSS reduced the mineral N-input between 16.3 % and 21.1 % (34 - 44 kg N ha‑1) compared to the KNS advisory system without significant yield loss in a field with low soil organic carbon (SOC) and low initial soil mineral nitrogen content. In a field with high SOC, the ECOFERT DSS recommended no application of mineral fertilisers, while the KNS recommendation varied between 117 and 162 kg N ha-1. This resulted in RSN values ranging from 35 to 54 kg NO3‑N ha-1 and 65 to 158 kg NO3-N ha-1 for the ECOFERT DSS and KNS, respectively. Despite its success in maintaining the RSN below the legal limit, the ECOFERT DSS resulted in a decrease of 10.2 % gross yield and 13.3 % gross income. However, these differences were not significant (p>0.3) due to the large variability between the repetitions of each strategy. The ECOFERT DSS provides a fertilization advice before any leeks are planted. However, leeks are mostly fertilised at planting and six to ten weeks after planting and some farmers are even interested in dividing the fertiliser application across even more moments. Therefore, it could be beneficial to update the fertilisation advice using information acquired between the application moments. To this end, the potential of unmanned aireal vehicle (UAV) based leek DBM and N‑uptake monitoring was evaluated. During the calibration and validation experiments of the ECOFERT model, multi-spectral images were acquired at the plant sampling moments with a UAV. Thes data were utilised to determine the most robust statistical model to estimate leek DBM and N uptake throughout the leek development. To this end, the performance of multiple linear and non-linear models was compared. Each model was repeatedly trained on the data from all but one flights, and evaluated on the data from the remaining flight. This approach was preferred as the research indicated that randomly splitting the dataset for training and validation resulted in overoptimistic results that were biased towards the most complex non-linear models during model selection, due to the leakage of flight-specific information between calibration and validation data. The models were calibrated on the data from the 2019 experiments and then tested on the data from 2020. Eventually, leek DBM could be estimated using multi-spectral UAV data with an RMSE of 6.60 g plant‑1, and a bias of 0.57 g plant‑1. Leek N uptake could be estimated with a RMSE of 0.22 gN plant‑1, and a bias of 0.00 gN plant‑1. While these results look promising, several challenges were identified that impede the use of multispectral UAV data for reliable leek monitoring throughout the growing season across multiple fields. The models were less accurate when evaluated on data from the next growing season. Both models were also not able to capture the observed variation at the beginning of the growing season and underestimated DBM or N uptake at the end of the growing season. In summary, this research has demonstrated the potential of process-based simulation models for optimizing leek fertilisation to comply with regional environmental regulations. However, future research is required to improve the balance between yield and RSN by adapting the objective function in the DSS. Additionally, inclusion of other management decisions such as planting date, plant density, and the mixture of fertilisers could further improve the advice. It is also recommended to integrate the dimension of long-term soil health in the optimization.
Source: Lirias
Author: Haumont, Jérémie
Language: English
Subfield: Environmental Chemistry
Field: Environmental Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7005920954