Spatial variability

Scale (ratio)

Toolbox

Statistics

Cartography

Computer science

Filter (signal processing)

Physical geography

Geology

Regression analysis

Spatial analysis

Regression

Mathematics

Geography

Sampling (signal processing)

Kriging

Environmental science

Soil science

Programming language

Categorical variable

Variogram

Geostatistics

Machine learning

Computer vision

Geoderma

Appropriate scale, justifiably reliable, categorical and continuous spatial soil information is urgently needed to address environmental problems and ensure sustainability of ecosystem services at local, regional and global scales. Regression Kriging (RK) is one of the most popular, practical and robust hybrid spatial interpolation techniques in the digital soil mapper's toolbox that enables the modeling of soil distribution patterns at multiple scales in space and time. Several theoretical and applied aspects of RK have been discussed; however, there are no review studies, which quantify the essential factors affecting the performance of RK. Materials for this review were gathered from high-quality international soil science journals: Catena, Geoderma, and Soil Science Society of America from 2004 to 2014. A total of 142 different models from 40 different articles were examined. The following criteria were considered to evaluate their impacts on the prediction efficiency of RK: i) soil geographic region, ii) area of extent, iii) spatial resolution, iv) target soil properties and/or classes v) sampling design, vi) sampling size and density, vii) sample depth viii) soil-environmental factors as predictors, ix) methods of transformation, x) factor analysis, xi) regression type, xii) model used for variogram, xiii) nugget to total sill ratio, xiv) spatial autocorrelation range, xv) coefficient of variation of observed dataset, xvi) evaluation method (note that in previous publications the term ‘validation’ has been used extensively in publications in pedometrics) and xvii) coefficient of determination. The historical development of RK, limitations and strengths of current RK studies, research gaps, and future trends in RK are discussed. A major finding is the inverse relationship between the accuracy of RK models and the variation of soil properties in the original datasets. Novel modified RK methods are proposed for further investigation to predict soil properties and classes.

Regression kriging as a workhorse in the digital soil mapper's toolbox