The monitoring of cultivated soils appears difficult to carry out due to high spatial variability, combination of short and long term temporal variations and large areal extent of the support. Besides, increasing regulation constraints require the implementation of survey programs to assess environmental policies efficiency. To assist the effective design of future monitoring networks, we developed a method based on realistic virtual landscapes to test the accuracy of several sampling strategies differing by their sampling designs and intensities. Considering soil and water quality issues, we studied soil phosphorus content. A realistic distribution of soil phosphorus was generated on a 15km by 15km area, using known spatial structures and crop field patterns. The evolution of this virtual landscape was modeled over 50 years with a shift of practices in manure spreading fertilizations after the 20th year. This spatio-temporal evolution of soil phosphorus content was the virtual landscape on which we tested the accuracy and the ability to detect evolution trends of the following sampling designs: regular grid, stratified sampling based either on initial or on current land use, and transect sampling. Five sampling densities (0.1, 0.6, 1.1, 2.6, and 4.6 site per km2) were systematically tested and two time intervals were used to assess trend detection (2 and 8 years). Increasing sampling density systematically improved the accuracy and trend detection of all the strategies, particularly between 0.1 and 1.1 sites per km2. Concerning the four sampling designs tested here, regular grid designs proved to be the most accurate whereas designs stratified on the initial land use introduced the largest bias and error. Designs stratified on current land use succeeded poorly in detecting trend compared to designs with fixed monitoring sites. Indeed, the change of sites at each date and the global evolutions of land use proportions produced a scattering of statistical parameters over time. This was explained by differences between sampled and real land use proportions. Both methodological results and design comparisons are transferable to environmental stakeholders willing to efficiently design survey programs.