Can Soil Survey become the basis for land management?

One of the most challenging aspects of land management and restoration is to balance the generation of new information and the application of existing accepted knowledge.  At one end of the spectrum is the argument that we can always learn more but we should not wait to get on with the job.  At the opposite end is the rationale for a learning-based approach (adaptive management)-as we do more, we should learn more and we should constantly be revising the information base.  These questions are complicated by the lack of a centralized decision-making structure in most landscapes. There are also competing uses and groups, many of whom don’t even know they are in the mix. But long-term adaptive management is critical for land management into a future with changing climate and land use pressures.
We are at a turning point with regard to the development of a national-level adaptive management approach in the United States.  The National Cooperative Soil Survey (NCSS) has been a model for effective use of taxpayer funds to collect, store and disseminate information relative to land use and management (Miller 2012).  It has been so good in fact, that it has just about completed the initial soil survey (Figure 1).  While it is certainly not perfect, the information associated with the NCSS is accurate, highly accessible and exceptionally useful.  There are some gaps, but on the whole, this effort represents a standard that most of the rest of the world can only dream about.  The measured and interpreted information in this survey is far from an esoteric or academic exercise; basic information about soil performance provides a basis for decision-making for farmers, agribusiness and local, state and federal government.  On the global level, readily available resource information provides a competitive advantage that helps insure efficiency in production systems and government programs, even when the analysis is confined to on-site benefits (Giasson et al 2000).  Generally, the return on investment is far greater than most other potential agriculture investments (Pradhan 2009).
While we can be confident that the existing information in the NCSS has been well worth the investment, what is next?  One school of (non) thought is “Well, it’s done.  Now we can close the program and save the money”.  Another, almost as ill-conceived idea is “Good, now we can use the money to do more conservation”.  Both of these illogical options are based on the assumption that we have garnered sufficient knowledge from the existing work to solve conservation challenges and that the problems of the 20th Century are also the problems of the 21st Century.  Neither of these premises requires much effort to invalidate, simply because our existing interpretations are based on some pretty shaky assumptions about climate stability.  There is way more than enough evidence to dictate that we reexamine the way we interpret our existing data.
Added to climate change implications are a variety of changing assumptions about what we expect from land.  Most of our existing interpretations are limited to ideas about how to optimize and maintain production of a fairly narrow range of commodities.  As the pressure on land, and ecosystems, increases to produce not only more, but a wider variety of goods and services, our interpretations have to consider a broader array of inputs, processes and outputs.  For example, we struggle mightily with trying to predict the effects of changes in land use and management on hydrologic processes when we only consider landscapes as collections of fields.  Land ecology is, without overstating it, complicated.
Initiatives to reinterpret existing soil and vegetation information that we have collected as part of the existing NCSS effort, as well as emerging ideas about how to extend the concepts are loaded with promise.   Far from winding down, the early efforts at rethinking our basic assumptions and developing new ideas are really starting to pay off.  Not the least of which are the soil systems initiative and the ecological site effort that the NCSS has elevated in importance.  The questions that appear on the horizon demand that we put some serious effort into developing completely new, and sometimes seemingly contradictory, approaches.  
So the idea that we can get finished with the soil survey--call it a day and head to the house--may be appealing to bean counters and soon-to-be-retirees who thought they would complete their careers and the job on the same day, it’s a siren song that we have to actively resist.  But we have to realize that the contribution of soil science to land ecology does not have a finish line.
Miller, B.A. 2012. The Need to Continue Improving Soil Survey Maps. Soil Horizons. 53.
Giasson, E., C. van Es, A. van Wambeke and R.B. Bryant. 2000.  Assessing the economic value of soil information using decision analysis techniques.  Soil Science 165: 971-978. 
Pradhan, A. 2009.  Economic benefits of the National Cooperative Soil Survey Program.  Ph.D. Thesis Publication No. 3381204.  West Virginia University, Morgantown WV.


OK, we sent a bunch of soil scientists out there. I assume they know soil very well. I also assume that they are not botanists. So... to what extent are the soil parameters measured by soil scientists relevant to plants? They have a classification system that (I assume) makes sense for a soil-centric worldview; does it also make sense for a plant-centric worldview? My best guess is: partially.
In an ideal world, if we want a soil map that can be used for predictions about plant communities, the soil map would be based directly, explicitly, and traceably on the particular aspects of soil variation that are the most important drivers of plant communities. This requires that two difficult and time-consuming areas of expertise be joined--soils and botany. I don't know anyone who combines these skill sets. Perhaps there are people out there fitting that description, but the botanists I know have (as they must, to be good botanists) a basic familiarity with soils... and likewise the soil scientists I know have (as, likewise, they must) a basic familiarity with plants. You don't ask the soil scientists to tell you which species of Amaranthus is on a plot and you don't ask the botanists to tell you what kind of sandy soil the amaranth is growing in, because in neither case are you likely to get a useful answer. But you need both--you need to know which amaranth it is and to have a good description of the soil in which it is growing. This becomes difficult. Multiply across all the thousands of plants in my area and the however-many (not being a soil sicentist, I don't even have a good guess) kinds of soils in my area and you have a nearly insoluble problem. So, I tend to think that is the next step--for any given area we need both excellent botanists and excellent soil scientists working in collaboration to determine how plant diversity and soil diversity are related. At present, we are very lucky if we have one or the other, never mind collaboration.

"I don't know of anyhonw who combines these skill sets."
Then it seems like you have finally fallen in with the right crowd.  Stick around, you'll get the picture soon enough... 

kurby22's picture

There was a special edition of Rangelands (Dec 2010) that explains in great detail a lot of the things you mention--and how they are being applied in the development of ecological sites, which are directly correlated to soil components in soil surveys.  
You are absolutely right that plant people won't know everything about soils and soils people won't know everything about plants, but they should know enough to do their jobs properly.  In a perfect world there should always be a plant person and a soils person working together--although that's not always possible, the soil science division is attempting to build a program where there is at least one plant ecologist in each major soil survey office working with the soil scientists.  However, we also expect all of our plant ecologists to know their soils well enough to dig a hole in the ground and know what they're looking at in the soils and how those soil and site properties relate to the plant communities correlated to them, as it is the crucial medium for all plant growth, one cannot understand plants if they do not understand the soil and site conditions the plants require for growth. 
You might also take a look at Web Soil Survey (, which allows you to look at the area you work in and gather a list of all the soils found there.  It's an invaluable tool for anyone working in plant science.

I'm finally getting around to responding. :-)
Since my earlier comment, I've encountered several cases that might indicate problems with using SSURGO data in a botanical context. Suppose you walk an area and find that the northern half is very depauperate mesquite / fourwing saltbush shrubland with very low forb diversity & abundance, few grasses of any kind, and little surface sand while the southern half of the area is diverse mesquite / sand sage / fourwing saltbush shrubland with high forb diversity & abundance, relatively high grass diversity & abundance, and deep, loose surface sand. Then you check the SSURGO data and find that it's all one map unit. Clearly there is a botanically important change in soil characteristics that is not represented in the SSURGO data.
Alternatively, suppose you encounter a similar vegetation pattern but check the SSURGO data and find that there are three mapped soil units but that their distribution is independent of the variation in plant communities (each mapped soil unit includes parts of the depauperate area and parts of the diverse area, none of them follow the observed plant community composition / diversity boundary). To me, this indicates that SSURGO scientists identified changes in soil that were important in their classification scheme but which do not match changes in plant-relevant soil characters.
Or, looking at it from another angle, suppose you're trying to understand the distribution of a particular plant species rather than looking at patterns in plant communities as a whole. The plant you're interested in is rare and apparently occupies a fairly narrow niche. You map all the observed individuals of that species relative to SSURGO data and find that it occupies a wide range of mapped soils and ecological sites; a very small portion of those soil units / ecological sites is occupied by the species, however, so if you tried to estimate the plant's distribution based on soil data you would get a truly massive false-positive rate. However, subjectively (to a non-soil-scientist in the field), there is a distinctive common pattern of surface soil appearance and associated plant communities (e.g., relatively large surface rocks over finer soils in high-diversity creosote shrubland). To me, this indicates that the plant is responding to a set of common soil properties that correspond poorly with how soil variation is mapped.
I've encountered all three scenarios. I'm not sure to what extent these are indications that we simply need more and finer-scale soil mapping efforts (although surely we need those, either way!) or are indications that the way soils are mapped by soil scientists has partial overlap with the soil characteristics that are particularly relevant to plants.
I've encountered more easily-diagnosed cases of the latter in geological mapping. Geological mapping of sedimentary rocks is focused primarily on age. However, plant communities seem to respond primarily to variation in chemical composition, texture, and decomposition patterns of rocks. For instance, you will often get very different plant communities on limestone, sandstone, and gypsum. So if you're trying to use geological data in a botanical context, you might want a map of all limestone in an area. Instead, you will have available to you a map of different age-defined geologic formations, many of which will contain some limestone but few of which will be composed mostly or entirely of limestone. The mapped variable is not the variable that you're interested in, and has a complicated intersectional relationship to the variation you're interested in. This does not mean, of course, that classifying rocks by age is not the right approach for sedimentary geologists; it just means that the classification approach best suited to study of sedimentary geology is not the classification approach that is best suited to the study of botany. If we're going to use mapping of sedimentary geology -for botanical purposes-, though, we need the classification that is suited to this purpose and not the classification that is best suited to study of sedimentary geology for its own sake.
Similarly, if you're interested primarily in botany from the viewpoint of grazing ungulates, what you'll want is a classification of plants based on palatability, nutrition value, and toxicity. Botanical classification, however, is based primarily on phylogenetic inference and morphology, which does not correlate very well with the aspects of plant variation that are important for ungulate nutrition; some plant families are usually toxic, some are rarely, if ever, toxic, and many fall somewhere in between. The classification of plants that is best suited to those interested in ungulates is not the classification of plants that is best suited to understanding plants.
Going back to soils, I don't know if we're in an analogous situation. However, it would be kind of surprising to me if different classification goals of soil scientists and botanists were not an issue.

I don't think any of the scenarios that you described are at all uncommon.  Without know the area intimately, it would be difficult to repond directly, but I think your comment clearly defines the conundrum of mapping and describing spatially and temporally variable phenomena at a scale relevant to management.  First of all, a soil map unit (especially in more arid and semi-arid areas) can contain a lot of variability in soil properties.  From a design standpoint, a map unit frequently contains multiple soil components.  Those components can be pretty drastically different in the physical and chemical make-up, but occur at a spatial scale that does not warrant mapping.  An Ecological Site can (and frequently does) represent a range of soil properties contained in several components. Thus, an Ecological Site can contain mulitple components, but a component can only be assigned to one unique Ecological Site.  Rarely, a map unit and an Ecological Site have a 1:1 relationship.  So, in your comparision, you could easily be looking at different soil components (and different Ecological Site) within a single map unit-a common occurrence.
Another possibility is that you are actually on the same ecological site, and the vegetation communities you see are real and  reflect different disturbance regimes (ecological states).  This is also pretty common and sometimes difficult to detect without some relatively intensive sampling and/or extensive experience.  Again, sorting out whether these are design inaccuracies, mapping errors, sub-map unit scale spatial variability or temporal variability requires some systematic decision-making at the specific point of interest and for the specific property (soil or vegetation) of interest.
This is where I disagree with your suggestion that the solution to these issues is more fine-scale mapping.  Given the problems I described above, as well as financial realities, I am not sure that more fine scale mapping would give us any better information.  What would be more helpful, I think, is much more attention to the hierarchal nature for the distribution of soil properties, microclimate and vegetation patterns.  A hierarcical key in the form of a decision tree would allow the user to not only to eliminate bad assumptions, but to increase the probability of finding what you are looking for, in your case rare plants. 
In short, I think the maps that we create for soil property distribution or vegetation dynamics are assumption laden.  We would be much more effective developing a set of diagnostic keys to expose and test those assumptions at increasingly fine scales than we would  by trying to create increasingly inaccurate maps of spatial and temporal gradients.

To be clear, I don't know if more fine-scale mapping would address the problems I've encountered, but think it is good of itself--and might address many of the problems one encounteres as a botanist looking at soil-based classification of the landscape. For instance, when a soil map unit includes several components, mapping those components separately might make the soil map correspond with vegetation better. But even if it doesn't, I assume that better resolution of soil variation will be inherently good (in the same way that I assume soil mapping to be inherently good overall).
As I continue to think through these issues, I am slowly arriving at the conclusion that categorization has inherent problems. So far as I can tell, "All variation is continuous," is probably a generally applicable dictum that is accurate most of the time in ecology as a whole. From a big-picture perspective, the exceptions (which certainly do exist) are outliers. One can sometimes point to nice clear boundaries in soil properties, plant communities, and the like. Most of the time, though, you're looking at massive variation without clear lines dividing that variation. Or you might have clear lines in one component of the landscape without clear lines in others! (E.g., imagine you have a very clear line in soil charactistics, without a corresponding line in plant community characteristics; a geological unconformity might be a good model here, where a clear break in a chronological classification may or may not correspond with a clear break in those geological characteristics that are relevant to plant communities--if it's limestone on either side of the unconformity, the plants probably don't care about the natural break in a chronological classification.) This leads me to the conclusion that an ideal mapping would be one that is based on presenting continuous variation rather than one based on categorization. I assume that all soil or ecological site mapping is based on underlying set of variables that are mostly continuously distributed. Those underlying variables may or may not be explicitly mapped, but form the basis of the categorization. I.e., one might explicitly map all of the underlying variables and then segment that variation into categories, or one might start with categories (delineations based on aerial imagery, for instance) and then try to associates the variables with the categories post-hoc. Either way, one has essentially a pattern of mostly continuous variation in abiotic attributes across the landscape, which one attempts to segment into categories. Because the underlying variation is mostly continuous, those categories cannot help but have fairly arbitrary boundaries. The categories are a proxy for the variation in those underlying variables. Categorization is often necessary either in terms of simplifying communication (it is easier to talk about the "loamy ecological site" than to talk about the particular values of and variation in a set of ecologically-relevant variables) or to make up for gaps in our knowledge (if you categorize first and measure second, the task requires dramatically less data than if you map each potentially relevant variable first and then use that mapped variation to create categories), but categories are ultimately not what we want if we are to understand the landscape. Any categorization of continuous variation is inherently lossy and will create edge problems that will be problematic for land management. Returning explicitly to the problem of mapping rare plants, suppose the plant in question is present where there is loose surface sand and a very deep carbonate horizon. If you want to predict this plant's distribution, you need a map of surface soil texture and depth to carbonate. Instead, what you have available is a set of mapped units, each of which both has substantial variation in those attributes and overlaps with other mapped units in those attributes. The result is that the mapping of categorized soil variation, although it might be based on variation in the attributes of interest, does not preserve that variation. For a simpler example, you only need to look at edges of soil map units. Just imagine that you are trying to map a rare plant that occurs at that bit of continuous variation precisely where someone decided to draw the line. So half of your plants occur on one map unit, half occur on the other. In such a situation, predicting the plant's distribution based on categorized map units is very unhelpful (either you include both map units and dramatically overestimate its distribution, or include neither and dramatically underestimate its dristrubtion)
Long story short: one should expect that:
1) ecologically-relevant variables are continuously distributed;
2) plant communities and particular plants will occur in some segment of that continuous variation;
3) categorization of continuous variation must lose relevant information;
4) therefore, mapping of continuous variation of each ecologically-relevant attribute is the best-case scenario, with categorization serving as "short-hand" but neither reducing nor replacing our understanding of that variation.

I know I'm late to this conversation, I've been thinking about this all year. Actually I just read it during my lunch break. Here's my two cents:
Having used ecological site concepts and soil map units to help botanists map the location of rare cacti, I can say with some surety that both maps and ecological keys have a role in the process. Given current technological and financial contstraints, soil mapping simply cannot capture, organize, and display all soil properties relevant to the thousands of plant species that may or may not occupy any given point on the landscape. That is why botanists have jobs mapping rare plants in the first place. Also, keep in mind that soils are not static entities, and that species composition (or presence/absent) may reflect changes in dynamic soil properties (e.g. nutrient pools, pH, etc.) and/or disturbance. Perhaps as raster-based soil mapping improves, continuous maps of soil properties will become amenable to multivariate analyses that can produce a suite of soil properties which combine on the landscape to map POTENTIAL locations for rare plants with increased precision. Still, these technologies are a long ways off (the computing power and data input requirements are massive), and do not guarantee improved accuracy (despite improved precision) OR the ability to map variables relevant to predicting where rare plants occur. Besides, we can do a pretty good job with the resources we have now.
Some people have successfully used known locations of rare plants to identify the ecological sites AND the ecological states in which a rare plant is known to occur. This is a simple extraction process in GIS that intersects soil map unit lines with plant occurrance shapefiles. In my experience, the rare plants exhibited patterns in terms of the ecological sites and ecological states to which they are best suited. This makes sense because generalists are less likely to be rare, and despite the fact that ecological sites are typically defined by the common and not the rare species, they account for soil-plant relationships better than any other resource currently available. They are not perfect, but they are adequate for the time being. So, the botanists would target field sampling in soil map units containing a soil component correlated to the ecological sites of interest and the expected disturbance regime (e.g. time since fire, grazing intensity, etc) of the ecological states of interest. This reduced the sampling frame by better than 90% of the 120,000+ acres for the project. While in the field, the botanists used ecological site keys, as Joel described, to ensure that they were on the ecological site and state that they expected. The use of these keys requires basic understanding of soils, as most botanists have, and has the benefits of eliminating bad assumptions and increasing the probability of finding locations that are truly similar to the areas where a rare species is known to exist.
I understand the desire for more and better data, trust me, but I also know that clarity comes through the process of using existing information to the fullest extent possible.

I certainly agree that there's a lot more we can do with existing data. I'm just not convinced that we can't keep improving that data, nor that a categorized system based on soil map units is the right conceptual approach. I think raster mapping of continuous variables is what we're going to need if we really want to realize the full potential for soil science in land management. Which doesn't mean that this is realistically achievable in the near future, of course. It's just going to be sitting in my mind as the ideal goal...
Leaving rare plants for the moment, I have a land management example in southern New Mexico: we are applying herbicides to creosote shrubland in an attempt to restore grassland that was (or is believed to have been) historically present. Ecological sites give us a neat and tidy conceptual framework. We look at the reference state for an ecological site; we look at what's there now. If they don't match, we attribute this to a change from the reference state to the current, degraded state. Now we have an identified restoration need, so we do an herbicide treatment. Obviously it's not that simple in practice, but that's the basic conceptual framework. Let's say we're looking at the gravelly ecological site, R042XB010NM. I've spent a lot of time in that ecological site. From what I can tell, existing grass cover within it shows continuous variation, for which elevation appears to be the biggest explanatory variable: very low at the bottom end of the elevational range (here's an example, at 4200 ft), relatively high, sometimes up to reference state levels, at the top end of the elevational range (here's an example, at 5200 ft). Both of those linked pictures are on White Sands Missile Range; although they may well not be in their historical state, they're at least ungrazed & generally un-messed-with for the last 70 years. Outcomes from previous herbicide treatments seem to follow the same basic pattern: within the same ecological site, poor results (i.e., low perennial grass cover after treatment) at the bottom end of the elevational range, relatively good results at the top end of the elevational range. That leads me to believe that this particular category is not very useful in designing herbicide treatments; it doesn't capture the variables we're interested in very well. My hunch is that creosote shrubland with very low perennial grass cover is in fact the historical climax state for much (perhaps most) of the gravelly ecological site, or at minimum, that it is the closest thing to the historical climax state that is realistically achievable at present.
In any case, my point here isn't to pick on this ecological site. I don't think there's going to be a "right" categorization that will work well across a wide range of research and land management questions. I think this kind of failure is inherent in a categorical approach. That doesn't mean there isn't plenty of cool stuff we can do with existing soil mapping. We just need to keep in mind that the real world doesn't exist in neat categories; if we pretend it does, that alone is going to cause us to make some mistakes.

Excellent comment on a challenge that we all acknowledge, but struggle with.  We know that the world out there (especially the ecological part of it) is really organized continuously; but we have to deal with it discretely in order to actually get anything done.  If you take any given soil series, and the map unit component manifestations, there is a defined (but often ignored) amount of variability within that category-which we can quantify, given enough time and money.  For some, the question is "how do we reduce those conceptual and applied units down to a finer level so that we know better what we are looking at?"; for others the question is "how do we describe the range of attributes to guide management decisions?".  Your example of the elevation gradient really shows that conundrum.  Assuming that the only think that differentiates those sites is elevation (and not some overwhelming management actions), a logical question would be "do we divide that set of soil:vegetation properties into multiple sites" or "do we leave it a one site and spend more time providing evidence and synthesis of the within site dynamics that managers can use to make, implement and monitor decisions"?  Obviously, we have opted for the latter, not because the differences in the two locations is not real, but because, in a management context, we are more likely to get better decisions with more focus on important processes, the variability associated with them, the uncertainty of the information and finally the relevance to management decisions.  This is not unique to rangelands.  Nobody in their right mind would use a soil survey map to make a decision about whether to build a skyscraper on a particular piece of land.  That decision might start with a soil survey, but before too long, a soil mechanical engineer and a structural engineer (at a minimum) would be brought in to investigate.  What we are doing is much the same, just bigger and less intense.  One reason we have opted for accuracy over precision is that most of the values and processes in areas where we use ecological site information are better reflected at larger scales.  Thus, a STM becomes a collection of (hopefully) well-organized decision points for both classification and management response.  I think we are always going to be drawn to false precision.  If you can give an answer to six decimal points, it sounds so much more impressive than "well, it is probably between 1 and 100".  Our challenge is how do we retain and communicate the importance of that temporal spatial variability to decision makers, but still provide a relatively clear path for decision-making that can be recreated and tested?