|Main authors:||Else K. Bünemann, Giulia Bongiorno, Zhanguo Bai, Rachel E. Creamer, Gerlinde De Deyn, Ron de Goede, Luuk Fleskens, Violette Geissen, Thom W. Kuyper, Paul Mäder, Mirjam Pulleman, Wijnand Sukkel, Jan Willem van Groenigen and Lijbert Brussaard|
|Source document:||Bünemann, E. K. et al. (2018) Soil quality - A critical review. Soil Biology and Biochemistry, Volume 120, May 2018, pp 105-125
Increasing the number of indicators can increase collinearity as well as the complexity of the relationships between indicators and management options. Moreover, costs of measurements easily become prohibitive, especially if detailed soil biological parameters are included (O'Sullivan et al., 2017). For these reasons, the number of soil quality indicators that is actually analyzed on a given set of samples needs to be reduced to a minimum dataset.
In the first proposed minimum datasets, this selection was based on expert judgement (e.g. Doran and Parkin, 1994). Subsequently, statistical data reduction by multivariate techniques such as principal component analysis (PCA), redundancy analysis (RDA) and discriminant analysis (e.g. Andrews and Carroll, 2001; Lima et al., 2013; Schipper and Sparling, 2000; Shukla et al., 2006), and multiple regression (Kosmas et al., 2014) became more common. After this initial data reduction, simple or multiple correlation analysis can further decrease the number of indicators (Andrews and Carroll, 2001; Kosmas et al., 2014), sometimes followed by the use of expert judgement for choosing only one out of two or more highly correlated soil properties (Sparling and Schipper, 2002). With these techniques, the number of indicators finally selected typically ranges between 6 and 8. Because soil properties that are relevant for soil functioning but do not show much variation in a given study will not be included in the minimum dataset, validation of the minimum dataset is important, for example by testing its relation to predefined and independently measured management goals (Andrews and Carroll, 2001).
A participatory approach of selecting soil biological indicators from a long list of potential indicators was presented by Ritz et al. (2009). Potential indicators were scored by scientists and end-users in a “logical-sieve” approach, which allowed several iterations. The different requirements for an indicator (»Requirements for soil quality indicators - Table 3) were weighted: reproducibility was considered absolutely essential, whereas the existence of a standard protocol had the lowest weight. A modified version of this method was applied by Stone et al. (2016a) to establish the top 10 biodiversity indicators of soil quality (defined as the ability to perform key soil processes) across the agricultural area of European member states for use in future monitoring.
Finally, the most important soil quality indicators can also be inferred from participatory conceptualization of how complex systems function. For example, Troldborg et al. (2013) and Aalders et al. (2011) established a Bayesian Belief network defining which factors are most influential in determining the risk of compaction and erosion, respectively.
Hence, the selection of a minimum dataset derived from a larger set of soil quality indicators is a necessary step in soil quality assessments because of financial and time limitations and to avoid collinearity. Methodological transparency is imperative to allow wide application of minimum dataset selection.
Note: For full references to papers quoted in this article see