Proponents of digital soil mapping sometimes criticize traditional soil mapping for using a discrete data model to describe a continuous surface (field) and lacking a quantified estimation of error. Although most of my research is on digital soil mapping, I like to give due credit to the accomplishments of traditional soil mapping. I understand the need for digital soil mappers to contrast their work to earlier methods, but this has led to some myths that should be debunked (or at least clarified). Simply put, traditional soil mapping methods are not as bad as they are sometimes made out to be, which can cause us to misunderstand the advancements we’ve made with digital soil mapping. In many cases, the most significant difference between traditional and digital soil maps is the quality and amount of data available to make the map. This is where digital soil mapping truly shines, because it would be practically impossible to utilize all of the data we have now for mapping soil using traditional methods.
First, let’s talk about spatial data models. Polygon and raster data models are the most commonly used for mapping applications today. The polygon data model (points, lines, and polygon shapes) is generally the one used on paper maps, largely because those shapes can be efficiently drawn by hand. However, this data model continues to be very useful in the digital age. The raster data model is popular for digital maps because it is a very efficient model for storing and analyzing spatial information in a digital format. It is a regularly spaced grid of cells, so the amount of geometric information needed is minimized.
In digital soil mapping, it is sometimes argued that raster data models are the most appropriate because they can represent continuous fields, and the soil landscape is a continuous field. While I don’t disagree with the conclusion, both of these things are only partially true. We should remain aware that there are many examples of discontinuous breaks in soil landscapes (e.g., changes in parent material). However, assuming the soil landscape being mapped is continuous, raster data models are only better suited to represent continuous fields than polygons. Why? It is because rasters can store data more efficiently that we choose to use rasters to map continuous fields. For example, a circle drawn with 100 vertices (points connected by lines) would look more like a circle than when drawn with five vertices. Nonetheless, the raster data model still assigns a single value to a defined geometric area. Even if a continuous function was used to determine that value, it was converted into a discrete model when it was calculated for the raster. The only way to have a truly continuous map is to have an interactive interface that adapts the values to the pixels on the screen directly from a mathematical function.
So what about the polygon data model, can it represent continuous fields? Yes, the polygon data model has been used to represent continuous fields for centuries by way of isolines or contours. Admittedly there are limitations in how much information can be communicated with isolines, but sometimes that is a good thing. For example, it can be useful to convert raster data into contour lines. The resulting map can often show the spatial patterns in the data more. An interesting question then is, why did traditional soil mappers choose to map units of classified soil instead of drawing contour lines for individual soil properties? One possibility is the high uncertainty associated with spatial predictions, especially with the limited data available.
The obvious and most important differences between traditional soil mapping and digital soil mapping are the better base maps now available combined with the efficiency and consistency that data can now be processed in geographic information systems (GIS). If soil mappers working in the 1950s had only the base maps we have available today, their maps would have been much more accurate. However, a key limiting factor would still have been how much detail could be included in the map when drawing delineations by hand. This is one of the real contributions of digital soil mapping. Using digital methods, we can quickly analyze the base map data and apply quantitative determinations, which make them consistent across the map area.
This limitation in the base maps is an important consideration when evaluating traditional soil maps. In addition to not having the time to delineate small areas, the soil mappers knew that there were limits to how well they could predict variation in the soil landscape with the data they had available. For example, using aerial photography from the 1980s to map forested areas, changes in vegetation were often the most useful spatial information that could be extracted. Stereophotographs provided some information about the relief, but the changes in elevation had to be fairly dramatic to be detectable under the forest cover.
Knowing these limitations, 20th-century soil mappers employed certain strategies to communicate how much they did or did not know. One of these strategies was to recognize that it was easier to hit the broadside of a barn than to hit a single nail. I’m not saying they purposely chose wide targets, but they understood their limitations and were careful not to overstate their confidence. Although delineating smaller areas would decrease the variety or range of soil properties within that area, the soil mapper was limited in how accurate the placement of that delineation could be and in how accurate the prediction of the soil properties could be. To deal with this, the soil mappers would delineate areas compatible with their level of spatial certainty and then classify those areas in a way that encompassed the range of soil properties they thought were most likely to be there.
In modern digital soil mapping, we like to apply Gaussian statistics to estimate the potential error in our predictions (aka uncertainty), which certainly has its benefits. However, it isn’t fair to say that traditional soil mapping did not provide a measure of uncertainty. Compare the soil property ranges included in a traditional soil map’s attribute table with what we would today call prediction intervals. Yes, prediction intervals are based on better statistical methods, but could those methods be applied to the limited data available to soil mappers in the past? In today’s digital soil mapping, we manage a lot more data, and that is a big benefit.
Digital methods have greatly advanced our ability to map soil. Among those advances are greater consistency and efficiency of synthesizing data. The speed in which we can analyze large amounts of data has made it possible for us to leverage the large increase in data available (e.g. remote sensing, etc) and to use the data in new ways. As we move forward with digital soil mapping, we should recognize the reasons behind traditional soil mapping methods and the real reasons behind why we choose certain techniques in digital soil mapping.
Do you see other benefits from digital soil mapping? Tell us about them by leaving a comment.