Abstract
Soil monitoring programmes face significant challenges as there is an important trade‐off between detecting significant changes in soil properties on the one hand (which can be achieved by minimizing variability by higher sampling density or stratification approaches), and identifying the driving forces responsible for these changes on the other hand (which requires enough variability). This study aims to reconcile these two objectives by identifying the driving forces of soil organic carbon (SOC) evolution over a long period, based on an extensive but stratified soil monitoring programme. Data at both the finest level (questionnaires to the farmers) and the large scale (agricultural census, climate and soil databases for southern Belgium) were used in a cluster analysis, multiple linear regressions and mixed odels in order to discriminate between the driving forces involved. Results indicated that the negative ‘baseline effect’ (i.e. the inversely proportional effect of the initial SOC content on the SOC evolution) was responsible for an important part of the SOC variability. Consequently, the systems are not at steady state when starting the observations, although this assumption is used by most SOC dynamic models. Moreover, the baseline effect resulted in a trend of the soils to converge towards a regional SOC stock which significantly differed according to land use (36.4 t C ha −1 for the plough depth of cropland and 92.2 t C ha −1 for the 0–30 cm layer of grassland). Despite this strong effect, the main driving forces of the SOC decrease of cropland (−0.2 t C ha −1 yr −1 ) and SOC increase of grassland (+0.2 t C ha −1 yr −1 ) over a period of 50 years were discriminated. The agricultural management (cropland) and the clay content (grassland), together with the change in precipitation (to a lesser degree for cropland) were highlighted as the predominant factors involved in SOC evolution, when land use change is excluded. The use of questionnaires allowed to better understanding the impact of an intensive agricultural management on the SOC content, as the lowest SOC stocks were associated to the most intensively managed fields. The mixed models partly succeeded in predicting SOC evolution as they presented still large uncertainties after validation (mean error from 3% to 25%, root mean square error of prediction from 21% to 242%). While SOC monitoring schemes are increasingly being implemented, our results will likely apply to those using a similar design. It was shown that this strategy succeeded to reconcile both the SOC change detection and the distinction of the driving forces involved at the regional scale.