The fact that conventional agricultural practices have many detrimental effects is widely acknowledged (Rabbinge, 1997). To mitigate these effects, Dutch policy makers have implemented environmental laws that are essentially based on characteristic indicators for groundwater quality. This has resulted in progressively tighter restrictions on the input of N fertilisers and a consistent reduction of the number of registered pesticides. Enforcement of these laws still creates considerable problems, which is partly caused by their generic character: no provision is made for the significant variation among soil types. As this variation is well known to farmers, their affinity with the imposed rules and regulations is limited. This thesis follows a different approach, in which soil variability is placed at the starting point of research, using the techniques of precision agriculture. The objectives are:Develop a methodology that efficiently describes soil variability at the within-field level. Soil variability should be described in terms of functional properties that are directly relevant to farm management operations. In other words: describe soils in terms of their water regimes, nutrient cycling and sorption characteristics rather than using traditional taxonomic properties such as texture, soil organic matter (SOM) content and colour.Based on the above, develop methods to: (i) optimise the application of N fertiliser and (ii) evaluate and control the environmental risks associated with pesticide use. These methods should be developed through prototyping (Vereijcken, 1997) with ample attention for operational aspects.In line with the desired setting, research was conducted on a commercial arable farm in the central-western part of the Netherlands (51 o17'N, 4 o32'E). The farm covers an area of approximately 100 ha and applies a crop rotation of winter wheat, consumption potatoes and sugar beet. Soils originate from marine deposits, are generally calcareous and have textures ranging from sandy loam to clay. They are characterised as fine, mixed, mesic Typic Fluvaquents (Soil Survey Staff, 1998) or Mn25A-Mn45A on the Dutch 1:50,000 soil map (Vos, 1984). Soil variability is large and mainly expressed through differences in texture, soil organic matter (SOM) content and subsoil composition (peat or mineral matter). Drainage conditions are excellent and in general terms the area is considered prime agricultural land.In order to fully exploit the potential of precision agriculture, an understanding is required of the biophysical processes that govern the growth conditions of a crop. Simulation models provide a powerful tool in this respect and play a crucial role throughout this thesis. To maximize the accuracy of modelling results, much effort was invested in producing high-quality input data. Four methods were compared to derive soil hydraulic parameters from the basic soil properties collected in a 1,5000 soil survey of the study area. Applied methods included: (A) laboratory measurements, (B) class pedotransfer functions, (C) continuous pedotransfer functions and (D) continuous pedotransfer functions combined with simple laboratory measurements. Modelling performance was evaluated by comparing simulated and measured soil moisture contents for three sites and two depths. The combination of continuous pedotransfer functions and simple laboratory measurements (method D) clearly produced best results. Modelling performance was highest overall and results were consistent for individual profiles and depths. Modelling uncertainty was lowest, far lower than the uncertainty resulting from the measured data set. (Chapter 2)With the soil hydraulic parameters available, attention focused on describing soil variability. Soils were characterised in terms of simulated functional properties relating to water regimes and nutrient dynamics. Four properties were considered: water stress, N-stress, N-leaching and residual N-content at harvest. Sensitivity to water stress was evaluated for a dry year (1989); other properties were quantified for a wet year (1987). Based on functional similarity, individual soil profiles were grouped into functional classes. Standard interpolation techniques and a boundary detection algorithm subsequently identified soil functional units in each field. Analysis of variance revealed that over 65% of the spatial variation could thus be accounted for. This confirmed that soil characterisation had been efficient and that the resulting units were suitable entities to be used as management units for precision agriculture. (Chapter 3)Once the management units had been established, two field experiments were conducted to compare precision and conventional N management. The experiments were conducted in consecutive years (1998 and 1999) and on different winter wheat fields ( Triticum aestivum L.). Precision management used real-time simulations to monitor soil mineral N levels in each management unit. Early warning was provided when mineral N concentrations dropped below a critical threshold. Used as a trigger , this information served to optimise the timing of four consecutive N fertilisations. Fertiliser rates were determined through exploratory simulations, which calculated the amount of mineral N required under normal conditions. Compared to conventional management, fertiliser input was reduced by 15-27% without affecting grain yield. Grain quality was either not affected (1999) or significantly increased (1998; P