Small scale digital soil mapping in Southeastern Kenya

Abstract

Digital soil mapping techniques appear to be an interesting alternative for traditional soil survey techniques. However, most applications deal with (semi-)detailed soil surveys where soil variability is determined by a limited number of soil forming factors. The question that remains is whether digital soil mapping techniques are equally suitable for exploratory or reconnaissance soil surveys in more extensive areas with limited data availability. We applied digital soil mapping in a 13,500 km² study area in Kenya with the main aim to create a reconnaissance soil map to assess clay and soil organic carbon contents in terraced maize fields. Soil spatial variability prediction was based on environmental correlation using the concepts of the soil forming factors equation. During field work, 95 composite soil samples were collected. Auxiliary spatially exhaustive data provided insight on the spatial variation of climate, land cover, topography and parent material. The final digital soil maps were elaborated using regression kriging. The variance explained by the regression kriging models was estimated as 13% and 37% for soil organic carbon and clay respectively. These results were confirmed by cross-validation and provide a significant improvement compared to the existing soil survey.

Introduction

Increasing environmental concern has augmented the demand for regional land use analysis. While in the past regional land use analysis was often based on qualitative procedures (FAO, 1976), currently more quantitative methods are required and become available (Stoorvogel and Antle, 2001, Bouma et al., 2007). Soil information is important for many regional land use analysis models. This is especially true in models that deal with processes of land productivity and degradation. However, traditional soil surveys do not provide quantitative data at the detailed scale level that is required (McBratney et al., 2000, Ziadat, 2005, Kravchenko et al., 2006a) and new methods of soil mapping are needed.

Standard soil surveying techniques (USDA, 1984, Soil Survey Staff, 1993, USDA, 2007) have had great importance in pedology. However, conventional soil surveys provide qualitative data in the form of chloropleth maps which are a simplification of the existing soil resources (Zhu et al., 2001). Moreover, the traditional methods are expensive and time consuming due to the large number of observations and the limited use of auxiliary information. Recently, with the rapid development of computers and information technology, together with the availability of new types of remote sensors, a more quantitative approach has been developed that may replace the traditional inventory techniques. These new techniques include the modeling of continuous surfaces based on the factors of soil formation, as well as the assessment of accuracy and uncertainty of the predictions (McBratney et al., 2000). This approach is commonly referred to as digital soil mapping (McBratney et al., 2003). In digital soil mapping a limited number of soil observations can be used. These observations are then related to auxiliary information representing important soil forming factors: digital elevation models representing topography, satellite images representing land cover and climate, and geological maps representing parent material and possibly age. These relationships can now be used to predict soil properties for the entire area for which auxiliary information is available. In early applications, soil observations were related only to terrain attribute maps using simple regression models, but later the predictors were broadened to an array of environmental variables giving origin to the terms “environmental correlation” (McKenzie and Ryan, 1999) or the “CLORPT techniques” (McBratney et al., 2000). Alternatively, hybrid methods have been developed from the combination of geostatistics and environmental correlation, where the observations or the residuals of the regression are interpolated using co-kriging or regression kriging (Hengl et al., 2004).

Literature provides a large number of examples where digital soil mapping is presented as an efficient surveying technique. However, in many of these cases the techniques are applied in small areas (less than 100 ha) with at least 200 observations per square kilometer (Bhatti et al., 1991, McBratney et al., 2000, Florinsky et al., 2002, Kravchenko et al., 2006b), or for (semi-) detailed soil surveys in areas of less than 150 km², in which the number of observations per square kilometer ranges from one to 20 (Gessler et al., 2000, Ryan et al., 2000). In addition we see that in many of these successful stories soil variation is induced by a limited number of soil forming factors. For example, by correlating soil reflectance with Landsat Thematic Mapper images, Bhatti et al. (1991) effectively estimated soil properties; Gessler et al. (2000) built a model for soil organic carbon (SOC) that accounted for 78% of the variation using topography and terrain attributes only; and McKenzie and Austin (1993) attained a good prediction of soil clay content with parent material and relief as explanatory variables, using just about 200 soil samples for an area of 500 km². Furthermore, small scale applications of digital soil mapping (Frazier and Cheng, 1989, McBratney et al., 2000, Hengl et al., 2004) indicate that hybrid methods represent a powerful spatial prediction tool, especially up to catchment or regional extent. Many of the examples of digital soil mapping applications come from Western Europe, the United States and Canada where good explorative soil surveys are already available. However, there is a call for explorative soil surveys in many tropical countries where the national surveys have not progressed as much as in many developed countries. In these cases, it is urgent to find methodologies that enable to rapidly and effectively capture information about the spatial variability of the soils and reduce the need for intensive and expensive sampling. Hence, the question that remains is whether the digital soil mapping techniques are suitable for explorative or reconnaissance surveys, where we have to look at larger areas, with limited data availability and considerable inherent soil variation caused by the interaction of different soil forming factors.

In this research we tested the digital soil mapping techniques for a reconnaissance survey in Kenya. The final soil map of this study was intended for the analysis of agricultural productivity focusing on terraced maize fields. We, therefore, focused on SOC and clay content because these properties are important driving factors behind crop production and can be used in crop growth simulation models as indicators of soil fertility and water holding capacity. SOC is expected to be highly variable as it is influenced by land use. In contrary, we expect the clay content to be less variable and more dependent on parent material and soil development. In previous studies (Gessler et al., 2000, Kravchenko et al., 2006b) both properties have shown strong spatial structure, suggesting the potential of using terrain attributes and other auxiliary information in order to model their variability. We will examine if this assumption is still valid when samples are taken one to several kilometers apart in areas that are so large that the spatial prediction is performed with much less than one observation per square kilometer.