Digital soil mapping approaches to address real world problems in southern Africa

Abstract

Listen

Translate article

Soil spatial information is increasingly sought after for various agricultural, environmental and developmental uses, but is often unavailable, also in southern Africa. Digital soil mapping (DSM) can provide the tools to fill this gap, but the uptake in developing countries have been slow. Local research is required to adapt internationally developed methodologies to unique local situations. In southern Africa DSM research have reached the level where DSM tools can now be used in commercial soil related projects. Several DSM case studies, conducted across Southern Africa, have provided the platform from which this work is presented. These case studies were done for a range of situations, including environmental settings with variations, size, data availability and aim of the soil map. Three different approaches have been identified as useful DSM tools, with varying costs and level of information it provides. Land type disaggregation is the cheapest, as it is largely desktop based, but can only produce small scale soil association maps. The expert knowledge approach is the most widely used commercially. Large scale soil associations can be mapped, and 30 soil observations per homogeneous soil distribution area are required. Machine learning methods can map soil properties, but rely on large data sources, consequently it is the costliest. Machine learning is therefore used to produce large scale maps large areas, where cost can be diluted. This paper gives an outline of DSM research in southern Africa and presents a case study of each of the DSM approaches, showing the methodology, potential and limitations of the approach within a commercial context. Specific case studies presented in this paper include the agricultural assessment of 166 km of pipeline for a water distribution project in Limpopo (land type disaggregation), a land capability assessment of a 15,000 ha open coal mining area in Mozambique (expert knowledge) and hydropedological mapping in Johannesburg (machine learning).