Statistical Spatial Models of Soil Parameters. An Approach Using Different Methods at Different Scales

Abstract

The Romanian Soil Survey System does not imply, up to the present date, the use of digital methods in representing field campaign results or for mapping soil parameters. The presented study tests several geostatistical methods to model some soil parameters (soil pH and topsoil humus content), mainly in order to observe the differences induced by the scale of the approach and to test existing data. In this respect, three differently dimensioned analysis scales were chosen, all parts of the same larger region, located in Iaşi County. On the chosen areas the main three categories of methods used in pedometrics were tested: methods of the kriging family (ordinary kriging, cokriging), regression methods applied both globally and locally (Geographically Weighted Regression) and the combined approach of regression-kriging respectively. In order to test the results were used crossvalidation and independent sample validation. The root mean square error (RMSE) was used as selection criteria for the choice of the optimum method. The study proves that among the various interpolation methods tested, the regression-kriging approach gives better results and that the local approach, using GWR, is superior to the global regression approach. Moreover, the pH proved to be more spatially predictable compared to the topsoil humus content.

Comparative Multi-Criteria Assessment of Hydrological Vulnerability—Case Study: Drainage Basins in the Northeast Region of Romania

“…The East-Carpathian Crystalline-Mesozoic Zone [38], with hard rocks of low permeability, contributed to the formation of elongated catchments and developed along fault lines. In the platform area, towards the northeast part of Romania, the Sarmatian and Quaternary geological deposits, made of clay, marls, and sands with significant thickness, generated extensive drainage basins [39,40].…”

Spatial modelling of topsoil properties in Romania using geostatistical methods and machine learning

“…Other studies focused on comparative applications of DSM techniques in different regions of the country. [ 24 ] used kriging and regression-kriging to derive quantitative soil parameters and logistic regression–kriging to derive occurrence spatial probabilities for soil horizons in a part of Iaşi County; [ 11 ] tested the used of logistic regression and fuzzy techniques for the prediction of qualitative soil variables; [ 25 ] applied comparatively several geostatistical methods including ordinary kriging, geographically weighted regression, regression-kriging to map soil quantitative properties; [ 26 ] investigated the effect of scale and resolution on DSM; [ 27 ] used regression-kriging to map soil heavy metals in Cluj County; [ 28 ] explored the potential of land-surface segmentation technique to create predictors for DSM of qualitative soil parameters (eg. soil taxonomic units, textural classes); [ 29 ] used RK and OK to interpolate pXRF-derived soil chemical data in an archaeological site.Among DSM methods, kriging was one of the first to be used for the interpolation of soil properties in the 80’s [ 30 , 31 ].…”

Assessment of Water Quality State Dynamics Using Adaptive Filtering Methods and Neural Networks Approaching. Case study - Danube River in Galati area

“…Many authors have indicated that due to the seasonality of river water flows, the assessment of temporal variations in water quality could be made [1,[12][13][14]. These seasonal variations become an important aspect of the physical and chemical characterization of aquatic environments, and each investigation should include this research [1,[15][16][17].…”