A tool for cost-effective soil information
How is it done?
There are three approaches to DSM in South Africa, each yielding different level of outputs, at corresponding costs. With Land Type disaggregation, the information contained in the Land Type inventory is combined with terrain analysis to create a 3 to 4 unit soil association map. Due to a large amount of desktop work required, this approach is the cheapest and has been used for EIA’s. The most common commercially used DSM approach is the expert knowledge approach. In this approach, sufficient field observations are required for the soil surveyor to build a mental model of the soil distribution pattern within the landscape. This model is then transferred to a computer inference system, which generates the map. As there is quite a bit of field work involved, the cost is higher than that of Land Type disaggregation, but it also produced more detail maps. The expert knowledge approach has been used to create soil maps for forestry production potential, EIA’s and hydropedological studies. The Rolls-Royce of DSM is machine learning. It is the most expensive approach, as it requires a large soil observation databases, but it can utilize these databases to create soil property maps. It has been used for hydropedological and land degradation studies.
DSM vs conventional mapping
The conventional method of soil mapping in South Africa is a grid survey. Generally, this entails making a soil observation every 150 m (one observation per 2 ha), and then grouping similar soil observations together when drawing the map. The largest drawback of this method is, of course, the large number of soil observations necessary, which drives up the cost. Spreading the observations further from each other cuts the cost, but also loses some detail of the map. Conventional soil maps are drawn at a known scale, which gives an idea of the detail of the soil map. For agricultural potential, soil maps are generally drawn at a scale of 1: 10 000, which has a 2 ha minimum mapping unit of production units. With DSM the same information will be given through the resolution and number of soil mapping units of the map. Resolution relates to the size of the pixels of the soil map, which generally would be 30 m, while the amount of detail depicted on the soil map is shown in the number of mapping units present.
Expert Knowledge derived soil maps it could be up to six. DSM produced soil maps have the added advantage that they come standard with an accuracy determination, which is a further metric of the accuracy of the soil map. As no accuracy assessment is done on conventional soil maps, they are regarded as 100% accurate, which is not the case. Errors on these maps are due to difficulties drawing soil boundary lines between observations and soils occurring in areas smaller than the minimum mapping area.
A case study from Mozambique
DSM was applied to map around 21 000 ha for forestry production in Mozambique. The area has a high annual rainfall (± 1 800 mm) but also goes through a distinct dry period between May and October. Therefore, it is of utmost importance for the foresters to plant specific cultivars to the known water holding capacity of the soil. A DSM soil map was created in around 12 man weeks, of which 6 were field work and another 6 used to create the maps and write the report. In contrast to this, conventional soil mapping methods would take more than 90 man-weeks for the field work only. Thus, great savings were made in time and money. The overall soil map achieved a validation accuracy of 80%. On a scale of 0 – 5, 37% of the area had a production potential above 4, while 15% was unsuitable for production. Furthermore, of the total area suitable for production, 94% had a low erosion risk, while 74% had a high risk of soil compaction. These maps allowed for specific management plans to be implemented in addition to the specific cultivar choices.