Get reliable science

The Global Assessment of Soil Degradation (GLASOD) has been the most influential global appraisal of land quality in terms of environmental policy. However, its expert judgments were never tested for their consistency and could not be reproduced at unvisited sites, while the relationship between the GLASOD assessments of land degradation and the social and economic impact of that degradation remains unclear. Yet, other methodologies that could respond to urgent calls for an updated assessment of the global environmental quality are not operational or, at best, in progress. Therefore, we evaluate the reliability and social relevance of the GLASOD approach and assess its candidacy for new global environmental assessments. The study concentrates on the African continent, capitalizing on new GIS data to delineate and define the characteristics of GLASOD map units. Consistency is tested by comparing expert judgments on soil degradation hazard for similar combinations of biophysical conditions and land use. Reproducibility is evaluated by estimating an ordered logit model that relates the qualitative land degradation classes to easily available information on explanatory variables, the results of which can be used to assess the land degradation at unvisited sites. Finally, a cross-sectional analysis investigates the relation between GLASOD assessments and crop production data at sub-national scale and its association with the prevalence of malnutrition. The GLASOD assessments prove to be only moderately consistent and hardly reproducible, while the counter-intuitive trend with crop production reveals the complexity of the production-degradation relationship. It appears that increasing prevalence of malnutrition coincides with poor agro-productive conditions and highly degraded land. The GLASOD approach can be improved by resolving the differences in conceptualization among experts and by defining the boundaries of the ordered classes in the same units as independent, quantitative land degradation data.