Informations et ressources scientifiques
sur le développement des zones arides et semi-arides

Accueil du site → Doctorat → Belgique → Digital soil mapping, downscaling and updating conventional soil maps using GIS, RS, statistics and auxiliary data (in an arid and semi-arid region)

Ghent University (2017)

Digital soil mapping, downscaling and updating conventional soil maps using GIS, RS, statistics and auxiliary data (in an arid and semi-arid region)

Zeraatpisheh Mojtaba

Titre : Digital soil mapping, downscaling and updating conventional soil maps using GIS, RS, statistics and auxiliary data (in an arid and semi-arid region)

Digitale bodemkartering, neerschalen en actualiseren van conventionele bodemkaarten gebruikmakend van GIS, remote sensing, statistiek en hulpkaarten

Auteur : Zeraatpisheh Mojtaba

Université de soutenance : Ghent University

Grade : Doctorat 2017

Spatial distribution of soil types and soil properties in the landscape are important in many environmental researches. Conventional soil surveys are not designed to provide the high-resolution soil information required in environmental modelling and site-specific farm management. The objectives of this study were to investigate the relationship between soil development, soil evolution in the landscape, updating legacy soil maps and pedodiversity in an arid and semi-arid region. The application of Digital Soil Mapping (DSM) techniques was investigated with a particular focus to predict soil taxonomic classes and spatial distribution of soil types by soil observations and covariate sets representative of s,c,o,r,p,a,n factors. In the first study, focus is on establishing relationships between pedodiversity and landform evolution in a 86,000 ha region in Borujen, Chaharmahal-Va-Bakhtiari Province, Central Iran. From an overview study, we could conclude that landform evolution was mainly affected by topography and its components. A second study compares various DSM-methods and a conventional soil mapping approach for soil class maps in terms of accuracy, information value and cost in central Iran. Also, the effects of different sample sizes were investigated. Our results demonstrated that in most predicted maps, in DSM approaches, the best results were obtained using the combination of terrain attributes and the geomorphology map. Furthermore, results showed that the conventional soil mapping approach was not as effective as DSM approach. In the third study, different models of the DSM approach were compared to predict the spatial distribution of some important soil properties such as clay content, soil organic carbon and calcium carbonate content. Among all studied models, the terrain attribute “elevation” is the most important variable to predict soil properties. Random forest had promising performance to predict soil organic carbon. But results revealed that all models could not predict the spatial distributions of clay content properly. The minimum area of land that can be legibly delineated in a traditional (printed) map is highly dependent upon mapping scale. For example, this area at a mapping scale of 1:24,000 is about 2.3 ha but at a mapping scale of 1:1,000,000 it is about 1000 ha. A mapping scale of 1:1,000,000 is just too coarse to show a fine-scale pattern or soil type with any degree of legibility, but finer-scale soil maps are more expensive and time-consuming to produce. Thus, spatial variation is often unavoidably obscured. The fourth study of this dissertation focuses on downscaling and updating soil map methods. Thus, the objectives were to apply supervised and unsupervised disaggregation approaches to disaggregate soil polygons of conventional soil map at a scale of 1 : 1,000,000 in the selected area. Therefore, soil subgroups and great groups were selected because it is a basic taxonomic level in regional and national soil maps in Iran. In general, we conclude that DSM approach and also disaggregation approach are capable to predict soil types and properties, produce and update legacy soil maps. However, still a number of challenges need to be evaluated e.g. influence of expert knowledge on CSM approach, resolution of ancillary data, georeferenced legacy soil samples data to validate disaggregated soil maps.

Mots clés : Data mining, Soil-landscape modeling, Cubist, Random Forest, Disaggregation, Conventional soil mapping, Cost efficiency, Soil Properties


Page publiée le 18 septembre 2017