TY - JOUR
T1 - Extrapolation of a structural equation model for digital soil mapping
AU - Angelini, M.E.
AU - Kempen, B.
AU - Heuvelink, G.B.M.
AU - Temme, A.J.A.M.
AU - Ransom, M.D.
PY - 2020/5/15
Y1 - 2020/5/15
N2 - In theory, two separate regions with the same soil-forming factors should develop similar soil conditions. This theoretical finding has been used in digital soil mapping (DSM) to extrapolate a model from one area to another, which usually does not work out well. One reason for failure could be that most of these studies used empirical methods. Structural equation modelling (SEM) is a semi-mechanistic technique, which can explicitly include expert knowledge. We therefore hypothesize that SEM models are more suitable for extrapolation than purely empirical models in DSM. The objective of this study was to investigate the extrapolation capability of SEM by comparing different model settings. We applied a SEM model from a previous study in Argentina to a similar soil-landscape in the Great Plains of the United States to predict clay, organic carbon, and cation exchange capacity for three major horizons: A, B, and C. We concluded that system relationships that were well supported by pedological knowledge showed consistent and equal behaviour in both study areas. In addition, a deeper understanding of indicators of soil-forming factors could strengthen conceptual models for extrapolating DSM models. We also found that for model extrapolation, knowledge-based links between system variables are more effective than data-driven links. In particular, model modifications can improve local prediction but harm the predictive power of extrapolation.
AB - In theory, two separate regions with the same soil-forming factors should develop similar soil conditions. This theoretical finding has been used in digital soil mapping (DSM) to extrapolate a model from one area to another, which usually does not work out well. One reason for failure could be that most of these studies used empirical methods. Structural equation modelling (SEM) is a semi-mechanistic technique, which can explicitly include expert knowledge. We therefore hypothesize that SEM models are more suitable for extrapolation than purely empirical models in DSM. The objective of this study was to investigate the extrapolation capability of SEM by comparing different model settings. We applied a SEM model from a previous study in Argentina to a similar soil-landscape in the Great Plains of the United States to predict clay, organic carbon, and cation exchange capacity for three major horizons: A, B, and C. We concluded that system relationships that were well supported by pedological knowledge showed consistent and equal behaviour in both study areas. In addition, a deeper understanding of indicators of soil-forming factors could strengthen conceptual models for extrapolating DSM models. We also found that for model extrapolation, knowledge-based links between system variables are more effective than data-driven links. In particular, model modifications can improve local prediction but harm the predictive power of extrapolation.
KW - Homosoil
KW - Pedometrics
KW - Soil spatial variation
KW - Soil-forming factors
KW - Validation
U2 - 10.1016/j.geoderma.2020.114226
DO - 10.1016/j.geoderma.2020.114226
M3 - Article
AN - SCOPUS:85079859457
SN - 0016-7061
VL - 367
JO - Geoderma
JF - Geoderma
M1 - 114226
ER -