PhD Defence Ruamporn Moonjun


Ruamporn Moonjun is a PhD student in the department of Earth Systems Analysis. Her supervisor is prof.dr. V.G. Jetten from the faculty of Geo-information Science and Earth Observation.

Thailand has a long tradition of using detailed soil information in agriculture and watershed management. Agricultural extension workers (locally named “soil doctors”), give advice on farm level on increasing agricultural production and combat land degradation. Unfortunately, detailed soil information is only available for the flat areas where most of the agriculture used to be concentrated, with conventional soil mapping methods based on the US Soil Taxonomy classification system. As more and more of the hilly and mountainous areas are being used there is now a distinct lack of detailed soil data for these services. Also the complexity of the services has increased, dealing with integrated watershed management, addressing sustainability and more complex forms of land degradation, soil conservation and infrastructure development.

The Thai government cannot afford to do a detailed soil survey for the entire country as it is prohibitively expensive. Large scale soil survey products are not adequate, either categorically or cartographically, and cannot be easily downscaled for detailed applications. This study investigates an alternative soil survey method, developing a framework as a standard or guidelines to implement soil survey projects efficiently based on digital soil mapping (DSM) techniques. The framework should cope with the generation, maintenance and use of digital soil map products to meet the increasing demand of soil data for multi-purpose use and also offering possibilities for the update of soil information. The purpose of this research, therefore, is to investigate DSM methods for fine-scale soil mapping. The specific objectives include: 1) an investigation of high-resolution DEM and digital terrain modelling techniques, 2) application of airborne gamma-ray imagery and 3) use of fuzzy logic for fine-scale soil mapping. The study was conducted in Lomsak, Phetchabun province in Thailand, an area characterized by a variable terrain (flat to mountainous) and a large variety in soil types.

Eight terrain attributes were computed from two DEM resolutions (original 5-m. and degraded to 10-m.) using three neighborhood sizes (5x5, 10x10 and 20x20 cells). These attributes and their standardized principal components were then used as predictive variables for soil series and properties using logistic and linear regression, respectively. The results show that DEM derivatives based on grid resolution alone are not sufficient in analyzing their applicability in soil mapping. The neighborhood size also becomes important. The application of high resolution DEM at 5 m resolution with neighborhood size of 10x10 pixels gave good result to help map soil series. Single terrain variables could only model about 20% of the variability in subsoil bulk density and pH, with no clear advantage to either resolution or window size. Principal components derived from 5-m DEM with 20x20 neighborhood size and 10-m DEM with 10x10 neighborhood size were moderately successful (20-25% of variance explained) for these two properties. Probabilities of occurrence for two of three representative soil series were successfully modeled (area under ROC curve about 0.9) from the 5-m DEM with 10x10 neighborhood and 10-m DEM with 5x5 pixel of neighborhood. Predictive maps generally conformed to expert knowledge of experienced mappers, but showed large differences in detail among window sizes.  No general conclusion can be drawn about appropriate resolution and window size; these must be investigated per-property or series. Further, relief alone is a poor predictor of soil properties in this landscape.

The relationship between gamma-ray data and geological units was examined with box-and-whisker plots, using rock and soil samples. Rock and soil sample classifications were compared with the gamma-ray image and to typical radioelement responses found in the literature. To interpret AGRI data in terms of regolith and soil genesis, we compared AGRI to two existing soil maps: geopedologic and soil series maps. First, the geopedologic map was split into four maps according to the geopedologic hierarchy:  landscape, lithology, relief, and landform; at the latter (lowest) level, soil units are also associated. Secondly, soil series and geopedologic soil units were used to examine the distribution of radioelement response to selected soil characteristics: parent material, texture, mineralogy, and thickness. The correlation in both soil maps was interpreted in terms of the radioelement changes during pedogenic and geomorphic process, based on a review of literature and supported by soil samples.

AGRI provided useful information in three forms (single signal, ratio, and so called ternary images  enhanced with a hill shaded DEM) by relating these to lithology, material transport, and internal pedogenic processes. AGRI correlated well with the classes of the geopedologic map (1:50,000) at the two higher levels (landscape and lithology) but to a lesser degree at the two lower levels (relief and landform in geopedologic approach). In the mapping stage, AGRI showed deficiencies in the soil series map (1:50,000) made by conventional aerial photo analysis and limited field surveys, especially in inaccessible areas but also in low-relief terraces and flood plains, which provided a basis for future field sampling to correct these deficiencies. AGRI suggested new boundaries, differentiating topsoil properties and the presence of plinthite, despite its coarse resolution. Clustering of gamma ray and elevation data (DEM) was carried out using fuzzy logic to generate various classification layers. Class labels were assigned to the one with the largest total inverse distance over the entire set of fuzzy classification bands. The result shows relatively higher classification accuracy for soil parent material differentiation (overall accuracy of 72%) as compared to the classification for soil types (67%). Therefore, the result also shows that gamma-ray helps in determining lithology/parent material, weathering index and topsoil texture.

Soil series and topsoil texture mapping in a complex landscape has been carried out using fuzzy logic (SoLIM). An expert system is used whereby rule-based reasoning is applied for mapping soils in which the soil-landscape relationship is taken into account. The accuracies of the fuzzy logic derived soil map and that of conventional soil map are tested using a set of validation data. The results show that a soil series map generated by fuzzy logic has an overall accuracy of 67%, the highest accuracy is found in the Ct series (88 %) and the lowest in So series (57%). The results depend on the degree in which a series are related to a landscape position, and the broadness of the definition of the series. Regarding the topsoil texture, three texture classes give the highest accuracy (greater than 80%) are Silty Clay, Slightly Gravelly Clay Loam and Rock outcrops, while the lowest accuracy was found in Clay Loam (53%). The overall accuracy is about 65%. The accuracy of the soil map prepared by the conventional method shows an overall accuracy of 13%. The results confirm that Fuzzy Logic is advantageous in providing detailed information about spatial variations and representing realistic spatial patterns in soil series and topsoil texture maps. It has also the potential for reducing inconsistency associated with the traditional soil mapping processes, and as mapping can be carried out with a relatively low density of soil samples it may also reduce costs.

Soil survey works in Thailand are the responsibility of the Land Development Department (LDD), Ministry of Agriculture & Cooperatives. Currently, there is an increased demand of soil information not only for farm level planning but also for addressing complex sustainability issues. LDD has adopted geo-information system (GIS) and remote sensing techniques (RS) for digitizing existing soil maps, map visualization and data retrieval, but the LDD has not yet implemented digital soil mapping (DSM) techniques. The research results can be used to support soil survey works in Thailand in developing guidelines and framework for digital soil mapping, also for soil mapping in complex sloping landscapes, based upon specific Thai needs and conditions.