Multivariate Sampling Design Optimization for Digital Soil Mapping
In this study, we have extended the spatial simulated annealing (SSA) methodology to be able to simultaneously optimize a completely new sampling design for more than one pedological variable using regression kriging prediction-error variance (RKV) as opt
- PDF / 262,813 Bytes
- 11 Pages / 439.37 x 666.142 pts Page_size
- 66 Downloads / 193 Views
Multivariate Sampling Design Optimization for Digital Soil Mapping Gábor Szatmári, Károly Barta and László Pásztor
Abstract In this study, we have extended the spatial simulated annealing (SSA) methodology to be able to simultaneously optimize a completely new sampling design for more than one pedological variable using regression kriging prediction-error variance (RKV) as optimization criterion. For this purpose, the following soil properties were chosen: soil organic matter content, rooting depth, calcium carbonate content, and plasticity index according to Arany. The number of new observations was set to 100. The methodology is illustrated with a legacy soil dataset and auxiliary information from a study site in Central Hungary. The combined structure of the regression models and the variogram of the dominant soil parameter were applied in the optimization process provided by SSA to calculate the quality measure (i.e., spatially averaged RKV). The resulted sampling design was evaluated by various statistical and point pattern analysis tools. The Kolmogorov–Smirnov test’s results and the observed empty space function showed that the optimized sampling configuration represents properly both the feature and geographic space. Furthermore, the empty space function pointed out that there is an inhibition between the sampling points, which caused a “quasi”-regular point pattern. The extended SSA methodology is suitable to optimize the sampling design for more than one soil variable.
Keywords Model-based sampling Spatial simulated annealing kriging prediction-error variance Variography
Regression
G. Szatmári (&) L. Pásztor Institute for Soil Science and Agricultural Chemistry, Centre for Agricultural Research, Budapest, Hungary e-mail: [email protected] K. Barta Department of Physical Geography and Geoinformatics, University of Szeged, Szeged, Hungary © Springer Science+Business Media Singapore 2016 G.-L. Zhang et al. (eds.), Digital Soil Mapping Across Paradigms, Scales and Boundaries, Springer Environmental Science and Engineering, DOI 10.1007/978-981-10-0415-5_7
77
78
7.1
G. Szatmári et al.
Introduction
Digital soil mapping (DSM) aims at spatial prediction of soil types and properties by combining soil observation at points with auxiliary information, such as contained in digital elevation models, remote sensing images, and climatological records (McBratney et al. 2003; Heuvelink et al. 2007). Hence, the direct observations of the soil are important for two main reasons: (1) they are used to characterize the relationship between soil property and auxiliary information and (2) they are used to improve the predictions based on the auxiliary information, by spatial interpolation of the differences between the observations and predictions (Heuvelink et al. 2007). Regression kriging (RK), also termed universal kriging or kriging with external drift (Hengl et al. 2007), illustrates pretty well that twofold application of the observations, since it combines a regression of the target pedological variab
Data Loading...
