The Big Bang, the point in space and time from which all matter and energy in the universe supposedly emanated, is thought to have occurred sometime around 13.7 billion yr ago (Bloom, 2010). During the past 4.5 billion yr since the Earth coalesced and cooled, the planet has undergone alterations and transformations via (for example) plate tectonics, volcanism, and orogenies, which spawned severe changes in the composition and structure of the atmosphere, the oceans, and the land surface—including the biosphere and pedosphere. Other major natural forcing factors have included solar processes and orbital and galactic variations, which changed the amount of solar energy the Earth received (Intergovernmental Panel on Climate Change, 2007). Solar insolation and the structure and composition of the Earth's surface drive many ecosystem processes that have formed the soils we observe on the landscapes of today.
Value of soils in the anthropocene
During the last three centuries, human actions have produced profound shifts in the Earth system, becoming the main driver of global environmental change. Crutzen (2002), Steffen et al. (2005), Rockström et al. (2009), and others have described this new epoch, named the Anthropocene, and argued that human-driven changes are pushing the Earth system well outside of its normal operating range. About 30 to 50% of the planet's land surface is exploited by humans (Crutzen, 2002). Global population, which was just under one billion in 1800, is currently about 6.9 billion (U.S. Census Bureau, 2010) and is projected to be approximately 9 billion by 2050 (United Nations, 2004), indicating exponential growth rates. Crossing beyond critical values, the so-called “planetary boundaries” that provide safe operating space for humanity with respect to the Earth system, could cause severe, abrupt, and untenable environmental change (Steffen et al., 2005). Out of nine planetary boundaries, three of the Earth-system processes—global climate change, the rate of biodiversity loss, and interference with the N cycle (i.e., the amount of reactive N)—have already transgressed their boundaries, thus jeopardizing the resilience of major components of Earth-system functioning (Rockström et al., 2009). In the Anthropocene, soil change and soil formation and degradation have also accelerated, jeopardizing soil quality and health. As such, the need for up-to-date, high-quality, high-resolution, spatiotemporal, and continuous soil and environmental data that characterize the physicochemical, biological, and hydrologic conditions of ecosystems across continents has intensified.
The Need to Sustain Soil Resources
According to the National Academy of Sciences (2001), changes in climate, land use dynamics, biological diversity, ecosystem functioning, biogeochemical cycles, and the quality of soil and water resources have accelerated during the past decades at a rate proportional to human-induced activities and population growth. To address such global challenges, the National Academy of Sciences (2010) has identified four high-priority, interdisciplinary research initiatives in Earth surface processes: (i) interacting landscapes and climate; (ii) quantitative reconstruction of landscape dynamics across time scales; (iii) coevolution of ecosystems and landscapes; and (iv) future of landscapes in the Anthropocene. The latter two priority initiatives target the development of integrated human–landscape systems that are undergoing rapid change under varying climate and land use conditions and projections. Clearly there are profound needs to better quantify and reconstruct spatial and temporal Earth surface (soil) patterns and landscape dynamics.
Humans have fundamentally altered global patterns of ecosystem processes, biodiversity, pedodiversity, and landscape dynamics. Ellis and Ramankutty (2008) linked global populations to land use and land cover, showing that >75% of Earth's ice-free land has been altered as a result of human residence and land use, culminating in the delineation of anthropogenic biomes (anthromes). Between 1700 and 2000, the terrestrial biosphere made the critical transition from mostly wild to mostly anthropogenic biomes, passing the 50% mark early in the 20th century (Ellis et al., 2010). According to these researchers, anthropogenic transformation of the biosphere during the Industrial Revolution resulted about equally from the expansion of urban and agricultural land uses into wildlands and intensification of land use within seminatural anthromes.
According to Scherr (1999), the land surface of the Earth totals 13.0 billion ha, of which about 8.7 billion ha are under human use, mostly suitable only for forest, woodland, grassland, or permanent vegetation. Only 3.2 billion ha are potentially arable. About half of this potentially arable land is currently cropped, but overuse threatens to severely degrade the soil quality. The International Soil Reference and Information Center (1990) concluded that 1.97 billion ha (23% of globally used land) was degraded between 1949 and 1990, primarily by water erosion, followed by wind erosion, soil nutrient depletion, and salinization. Overgrazing was the leading proximate cause, followed by deforestation, and agricultural activity. Of all degraded soils, 58% were in drylands and 42% in humid areas. Soil degradation was assessed as being highest in Central America (31%), followed by Europe (20%), Africa (19%), Asia (16%), South America (9%), and North America (7%) (Scherr, 1999). According to the World Resources Institute (1990) about 25% of the globally used land is at risk for future degradation, including irreversible desertification. Soil degradation translates into reduced crop productivity, diminished livelihood, famine, undernourishment, and other societal disasters. There is a tremendous need for monitoring of soil degradation, which includes tracking of spatially explicit soil change.
Globally, humanity uses the equivalent of 1.4 planets to provide the resources we use and to absorb our waste (Global Footprint Network, 2010). Ecological footprints in high-income countries (6.1 ha per capita) differ tremendously from those of low-income countries (1.9 ha per capita), and of the world (2.6 ha per capita), indicating that humanity is consuming resources well beyond sustainability. Soils provide supporting, regulating, provisioning, and cultural ecosystem services (Millennium Ecosystem Assessment, 2005) and play a major role in the global system regulating major biogeochemical cycles and energy and water fluxes. To sustain soil resources and address the challenges of the Anthropocene at global and local scales, soil resource data are critical, requiring concerted efforts that transcend social, economic, and political boundaries (Grunwald, 2006b). For example, the agrocentric approach to soil mapping in countries with food deficiencies, which mainly targets soil fertility and soil degradation (Blum, 2006), must be integrated with the envirocentric approach to soil mapping that has gained momentum in developed countries concerned with environmental quality (Grunwald, 2009). Efforts to map soil properties, quantify changes in the soil ecosystem, and assess soil–atmosphere, soil–hydrosphere, and soil–biosphere fluxes at spatial and temporal scales matching other ecosystem components and processes have been hampered, however, by various factors as outlined here.
Soils—Part of the Global Biogeochemical Cycles
At local, regional, and global scales, the biogeochemical reactor of the Earth's surface responds to natural and human-induced forcings through chemical weathering and erosion of bedrock or surface deposits, the availability of nutrients in soils, the fate of anthropogenic contaminants, and the properties of ecosystems (National Academy of Sciences, 2010), which inherently provide positive or negative feedbacks to climate, water, and major biogeochemical cycles. Currently, major concern is focused on the influence of terrestrial C fluxes on global climate change, particularly the role of humans in modifying storage and fluxes in the global C cycle. In Norvig et al. (2010), David Montgomery pointed out that preserving the thin layer of minerals, living microorganisms, and dead plants blanketing the planet is critical to soil C sequestration as well as sustaining terrestrial life. Because the soil C is magnitudes of order larger than C in the atmosphere, even small increases in the rates of soil organic C loss could greatly enhance CO2 concentrations in the atmosphere, potentially creating a positive feedback on climate (Cox et al., 2000). On the other hand, the soil's ability to sequester large amounts of C is high, thus providing ample opportunity to counteract global climate change through mitigation and adaptation strategies.
Global assessments of the soil organic C pool have demonstrated major uncertainties. The estimates for the global soil C pool in the upper 1-m profile are vastly different depending on the data and method used to upscale soil C observations, including 1395 Pg (Post et al., 1982), 1462 to 1548 Pg (Batjes, 1996), 1600 Pg plus 360 Pg in peat (Jacobson et al., 2004), 2011 Pg (Intergovernmental Panel on Climate Change, 2000), 2500 Pg (Lal, 2004), and 3250 Pg (Field et al., 2007). Trumbore (1997) pointed out that soil C estimates in such global studies are based on relatively few soil C inventories (pedon data) from important regions and the uncertainties involved in estimating total soil C stocks from coarse-scale soil map units are extremely high. Furthermore, these global studies do not specify which fraction of the total soil C pool is in active, intermediate, or passive pools (Trumbore, 1997), thus they lack the link to processes involved in modulating soil C change. As Vasques et al. (2010a) pointed out, predictions of C pools and other dynamic soil properties across large regions are still rare. Given the large uncertainty associated with current assessments of soil C stocks, which often rely on historic soil data, no reliable hindcast and forecast estimates for soil C and other properties are available. This deficiency is in contrast to estimates of climate change, biodiversity, or population dynamics compiled by other disciplines, which provide predictions into the past and future
(Fig. 1). There are profound needs for better assessment of the global distributions of soil properties, such as soil C, N, P, moisture, and others, but also their linkages to ecosystem processes, biogeochemical cycles, and environmental and human-induced forcings. Accurate, high-resolution soil-environmental data spanning the globe are essential to assess ecosystem services, soil degradation, decoupling of major cycles (C, N, P, and S), soil–plant–water relationships, and soil contamination across spatial and temporal scales.
Significance of Digital Soil Mapping and Modeling
Digital soil mapping (DSM) and modeling techniques have proliferated during the past decades to address these soil data and information needs (Grunwald, 2006a; Lagacherie et al., 2007; Hartemink et al., 2008; Boettinger et al., 2010) and have the potential to overcome some of the limitations imposed by labor-intensive and costly traditional soil surveys. But there is still terra incognita ahead of soil science to provide a universally accepted digital soil model responding to global societal needs and pressures and guiding society toward environmentally sustainable management. The notion that soil surveys should be more encompassing than static soil maps was evoked earlier by Young (1973), who argued that soil surveys should include predictions of soil response to changes in land use, soil-specific crop-yield forecasts, and more soil interpretations than provided in traditional soil maps. Albeit a major evolution in soil survey methods is taking place, most operational soil surveys are still focused on the generation of soil maps that delineate areas (polygons) of one or more taxonomic classes. Furthermore, the making of these maps usually does not use predictive quantitative models derived from mathematical algorithms, geostatistics, or statistics with associated uncertainty and accuracy assessment for the estimated soil properties or processes (Grunwald, 2009). Grunwald (2009) also noted that current DSM research studies emphasize modeling soils across space and rarely across both space and time. Thus the DSM challenge of describing the spatial and temporal variability of physicochemical soil conditions in diverse ecosystems across continental and global scales is great.
Factors Limiting Soil Mapping and Modeling
There are several factors that have constrained soil mapping and modeling at continental and global scales, including the costs and labor involved in developing inventories across large regions. The costs for soil mapping are dependent on the map scale (or spatial resolution), field sampling, and the adopted processing methods to derive soil properties or soil classes. Current efforts to maintain soil survey programs are expensive. For example, in the United States, the NRCS is responsible for soil surveys covering about 9.7 million km2 of the United States and its territories, disseminating vector-based soil map products at map scales of 1:12,000 to 1:24,000 (the soil survey geographic database, SSURGO), 1:250,000 (U.S. general soil map, STATSGO2; formerly the state soil geographic database, STATSGO), or coarser. Most of the work is focused on maintenance, validation, and updates of surveys by about 450 staff members. Soil survey appropriations in the United States were US$43.46 million in 1980 and have risen to US$93.939 million in 2010 (Levin and Benedict, NRCS, personal communication, 2010). The current cost for soil surveying in the United States amounts to roughly US$10.30 ha−1.
This is in contrast to DSM projects, which have utilized environmental covariates (e.g., derived from remote sensing, digital elevation models [DEM]) and predictive modeling techniques to map soils). For instance, MacMillan et al. (2010b) conducted predictive ecosystem mapping using a knowledge-based fuzzy semantic import model to assess soils across 8.2 million ha in Canada, costing about Canadian $0.34 ha−1, with an average accuracy of 69%. Even lower costs could be achieved by automatic predictive mapping across a 3 million ha forested area in British Columbia, Canada, at an effective map scale of 1:20,000 (grid resolution of 25 m) at US$0.20 ha−1 (MacMillan et al., 2007). The GlobalSoilMap.net project (www.globalsoilmap.net/; verified 9 May 2011), which aims to predict 10 soil properties at six specific depth intervals at 90-m grid resolution across the globe (Sanchez et al., 2009; MacMillan et al., 2010a) (a land area of ?150 million km2), has estimated costs at US$0.20 ha−1. These estimates suggest that spatially explicit DSM that utilizes layers of globally available environmental covariates and modeling techniques may lower the cost for soil predictions at continental and global scales with spatial resolutions matching other environmental variables such as DEM and remote sensing images (≤30–90-m spatial resolution). These finer spatial resolutions resemble more closely the inherent spatial variability of soil and environmental properties. In a meta analysis, McBratney and Pringle (1999) found that spatial autocorrelations were ≤300 m for soil C, NO3–N, and K, and ≤100 m for soil P, sand, clay, and pH, which suggests that soil map products should resemble these spatial resolutions (or pixel sizes). In addition to high-resolution soil models that represent the spatial variability and distribution of soil properties, monitoring of soil change will require a major investment to establish a soil monitoring network across the globe. Recent studies have demonstrated the capability to assess soil C change across large areas in Java (Minasny et al., 2010), China (Yan et al., 2010), and Belgium (Meersmans et al., 2011), which required space–time soil data sets. Such soil change analysis from regional to continental and global scales is needed to address emerging questions of the Anthropocene.
Currently Available Digital Soil Data at Global Scales
Most of the Earth's land area is covered by existing soil maps at various scales, from low-resolution (e.g., the 1:5,000,000 FAO-UNESCO Soil Map of the World), to moderate resolution (e.g., 1:24,000 NRCS soil survey maps), to high resolution (e.g., the 1:5000 pedologic map of Belgium). Furthermore, some of these maps and their associated databases have been digitized and are available in digital form. However, digital soil mapping is more
than digitizing existing soil maps. Finke (2007) described several means of assessing the accuracy of digital soil maps, in terms of both producer accuracy and user accuracy. When evaluating currently available digital soil maps, or when considering the potential for new digital soil map products, data quality can be viewed as a function of positional quality, attribute quality, completeness, semantic quality, currency, logical consistency, and lineage (Finke, 2007).
The FAO-UNESCO Soil Map of the World (Nachtergaele, 1999), originally published as paper maps between 1971 and 1981, has been digitized, generalized, modified, and updated to produce several global digital soil databases (Table 1). While there is no current alternative to these digital versions of the FAO-UNESCO Soil Map of the World at the global scale, definite shortcomings to this map are recognized (Sanchez et al., 2009). In particular, the map does not adequately represent the current condition of soils, nor the current state of knowledge about soils and soil classification. All of the digital versions of the Soil Map of the World are at coarse scales (1:25,000,000 or 1:5,000,000) and represent information from soil classes. The earliest versions—the World Soil Resources Map (produced in 1990) and the World Reference Base (WRB) World Soil Resources Map (produced in 2003)—were digitized as generalized versions of the paper map and presented at a scale of 1:25,000,000. The primary difference between these two databases is that the legend for the WRB World Soil Resources Map was updated to conform to the WRB classification system. The more recent Digital Soil Map of the World (produced in 2007) provides a digital rendering of the FAO-UNESCO map at the original resolution of 1:5,000,000 and, like its predecessors, represents the dominant soil types within each soil map unit polygon. It may be noted that soil types (classes) only indirectly relate to soil-environmental change induced by the Anthropocene. The other permutations of the Soil Map of the World, the Global Data Set of Derived Soil Properties (from 2005) and the Derived Soil Properties Database (from 2006), were created by the International Soil Reference and Information Center (ISRIC), and combine spatial data from the Digital Soil Map of the World with measured soil property data from the World Inventory of Soil Emission potentials global soil profile data set (Table 1). Both databases are of coarse spatial resolution (e.g., at the equator, the 0.5° resolution raster of the Global Data Set of Derived Soil Properties is approximately equivalent to a horizontal resolution of 55 km, whereas the 5 arc-min resolution of the Derived Soil Properties Database is approximately 9 km). Both data sets report approximately 20 soil properties for two to five depth intervals (Table 1), including available water capacity, base saturation, bulk density, cation exchange capacity, coarse fragment content, drainage class, electrical conductivity, organic C, particle size distribution, pH, and total N. The digital Harmonized World Soil Database (FAO/IIASA/ISRIC/ISS-CAS/JRC, 2009) is the most recent spatial data set derived from the Soil Map of the World. It is a raster data set with a 30 arc-sec resolution (?1 km) and it provides data on soil classes and 13 selected soil properties. While the Harmonized World Soil Database is also derived from the FAO-UNESCO map, it incorporates regional and national soil information from around the globe to update both the spatial and tabular data in the database to produce a more seamless and consistent representation of world soil resources (FAO/IIASA/ISRIC/ISS-CAS/JRC, 2009).
|Product||Date of release||Currency of data||Scale||Data model||Variables|
|Map of World Soil Resources||May 1990(version 1)||1981 + updates||1:25,000,000||vector||Soil classes (using 1988revised legend)|
|World Reference Base (WRB) Map of World Soil Resources||Jan. 2003||1:25,000,000||vector||Soil classes (using WRB)|
|Digital Soil Map of the World||Feb. 2007(version 3.6)||1:5,000,000||vector||Soil classes (using WRB)|
|Global Data Set of Derived Soil Properties||December 2005(version 3.0)||1:5,000,000||raster(0.5° resolution)||22 soil properties at 0–30 and 30–100 cm|
|Derived Soil Properties Database||June 2006(version 1.1)||1:5,000,000||raster(5 arc-min resolution)||19 soil properties at 20-cm depth intervals to 100 cm|
|Harmonized World Soil Database||Mar. 2009(version 1.1)||1981 + updates||1:5,000,000||raster(30 arc-s resolution)||harmonized soil class and13 soil properties|
Table 2 lists several digital soil map products available at regional to continental scales but does not include non-digital products (e.g., atlas publications such as the Soil Atlas of the Northern Circumpolar Region, eusoils.jrc.ec.europa.eu/library/maps/Circumpolar/; verified 9 May 2011), nor does it include non-map products (e.g., soil profile databases such as the NRCS National Soil Characterization Database). It is apparent that most of these digital map products were produced by digitization of older paper maps without sampling of current soil resources, and in most cases, the digital database represents a compilation of multiple paper maps created at different times, by different individuals, at different scales, and for different purposes. Such compilations have varied currency and lineage and often contain logical inconsistencies (most evident where two or more separate maps have been joined, e.g., at political boundaries), all of which diminish the data quality (Finke, 2007). Efforts to harmonize the multiple data sets are required to improve both the attribute quality and the semantic quality of the resulting digital soil map. With the exception of digital soil survey databases from Canada and the United States, all of the regional and continental data bases are at coarse scales (1:1,000,000 or coarser). The most detailed data—the Soil Landscapes of Canada (version 3.1.1) and the SSURGO database—have incomplete spatial coverage.
|Product||Date of release||Scale||Data model||Variables|
|European soil database (ESDB), including soil geographical database of Eurasia (SGDBE)||Mar.–Nov. 2006 (version 2)||1:1,000,000||raster, vector||associations of soil typological units with associated soil properties|
|Soil map of the European Communities||June 1986||1:1,000,000||vector||taxonomic classes (associations)|
|Soil and terrain database (SOTER), land degradation status and soil vulnerability assessment for central and eastern Europe (SOVEUR)||2000 (version 1.1)||1:25,000,000||vector||soil classes + terrain elements, with measured soil property data|
|Soil and terrain database for Latin America and the Caribbean (SOTERLAC)||December 1998||1:5,000,000||vector||soil classes + terrain elements, with measured soil property data for 1800 soil profiles|
|Soil and physiographic database for north and central Eurasia||December 1999||1:5,000,000||vector||soil classes + terrain elements|
|Digital atlas of Australian soils||1:2,000,000||vector||soil–landscape units (consisting of multiple soil types)|
|SOTER for northeastern Africa||1998||1:1,000,000–1:2,000,000||vector||1200 homogeneous agro-ecological mapping units|
|SOTER for central Africa (SOTER-CAF)||September 2006||1:2,000,000 (Congo), 1:1,000,000 (others)||vector||dominant soil type, number of soil components|
|SOTER for southern Africa (SOTERSAF)||2003||1:2,000,000||vector|
|U.S. general soil map (STATSGO2)||2006||1:250,000||vector||taxonomic classes with estimated soil properties|
|Soil survey geographic database (SSURGO)||multiple||1:12,000–1:24,000||vector||taxonomic classes with estimated soil properties|
|CONUS-Soil||1998||1:250,000||raster (1 km)||estimated soil properties at six depth increments|
|Soil landscapes of Canada (SLC)||December 1996 (version 2.2)||1:1,000,000||vector||taxonomic classes with estimated soil properties|
|Soils of Russia||1:25,000,000||vector||soil classes with associated properties|
The Soil and Terrain Digital Database (SOTER) project has been a source for multiple national and regional digital soil maps and databases (Table 2). The goal of the SOTER project was to develop a global soil database coverage at 1:1,000,000 scale (Batjes, 1990). Through cooperation among the United Nations Environmental Program, FAO, and ISRIC, soil class maps that represent standardized soil and terrain attributes have been developed for many regions of the globe, including South America, southern and central Africa, and eastern and central Europe at scales of 1:2,000,000 to 1:5,000,000 (Table 2). The map units of the SOTER databases delineate land areas with distinct patterns of soils and associated landforms and parent materials. These SOTER products are expected to have greater positional accuracy, improved attribute accuracy, greater semantic accuracy, and improved logical consistency than the Digital Soil Map of the World. However, they are still at relatively coarse scales.
In summary, it is evident that the available digital global soil data do not meet the profound needs of the Anthropocene. If such coarse-scale soil data are included in global ecological biodiversity models, global climate change simulation models, or ecosystem service assessments, major uncertainties arise with unknown outcomes. Global soil monitoring networks describing soil change have not been implemented at this point in time. Assessing the impact of anthropogenic forcings on soil health, quality, services, degradation, and change requires higher spatial- and temporal-resolution soil data, which are currently not available at continental and global scales.
Are People Ready for Raster Digital Soil Data?
As illustrated above, most spatial soil data are available as generalized vector polygon maps—either as paper or downloadable digital products. Given the relative dearth of publically available pixel-based soil data, we asked the question, “Are people ready for raster digital soil maps?” To address this, we informally assessed people's awareness of existing soil survey information and their preferences for a future soil map format. We interviewed 65 participants with a wide variety of ages (18–60+) and backgrounds (12% agriculture, 23% natural resources, 65% other), with about half of the participants indicating that they were familiar with soil as a natural resource. To establish the same minimum level of basic understanding for all participants, each interview began with a brief review of a customized soil survey and interpretations for building site development for a site near the Utah State University campus in Logan, UT, generated from Soil Survey Staff (2011a). We asked the participants to compare vector polygon and digital pixel products for a 140-km2 area of northern Utah (centered at 41°32′37″ N, 112°19′31″ W): polygon lines over a black-and-white air photo base, a colored
pixel-based map over an air photo base, and a colored pixel-based map only. Participants were asked to rate the maps as high, medium, or low in terms of (i) visual appeal, (ii) conveying information about the spatial distribution of soils, and (iii) determining the soil type mapped for a specific location.
The visual appeal of the colored pixel maps was rated much higher than the polygon map; however, all the maps were rated similarly for concisely conveying soil spatial distribution and determining the soil type at a specific location (Fig. 2). Interestingly, the participants who had previously used soil surveys (32%) rated the polygon map much higher in all aspects than those who had no prior experience with soil maps. The most common response by participants choosing to offer comments was that the pixel maps would be useful in an interactive environment or web application where spatially explicit soil information can be accessed. We concluded that soil data and information can be effectively represented by a variety of map types and that pixel map products would be favorably received by the public, regardless of age and background.
Advances in Soil Mapping
Soil Mapping Paradigms
Pivotal events and paradigm shifts in soil mapping were described by Grunwald (2006a) and are summarized in Fig. 1. Technologies introduced to soil science in the early 1980s, such as global positioning systems (GPS), soil and remote sensing, and geographic information systems, have facilitated the upscaling of site-specific soil observations to larger landscape scales. In particular, soil sensors, such as visible near-infrared spectroscopy (VNIR) (Shepherd and Walsh, 2002; Vasques et al., 2009, 2010b), mid-infrared spectroscopy, and laser-induced breakdown spectroscopy (LIBS) (Harmon et al., 2005; Martin et al., 2010), offer new opportunities for rapid, accurate, and dense collection of soil properties. National soil survey programs have focused on the mapping of soil classes based on pedon descriptions in the field and laboratory characterization of soil morphological and physicochemical parameters of selected pedons at map scales of 1:24,000 and 1:100,000 to very coarse map scales of 1:1,000,000 and coarser (Fig. 3). These soil maps are “double crisp” because they use crisp map unit boundaries and crisp soil classes that ignore the internal heterogeneity of properties within them. Considering that fine-scale variability of many soil properties and processes has created intricate spatial patterns across the soil-landscape continuum, such double-crisp polygon-based soil maps impose major constraints. As Hartemink et al. (2010) pointed out, polygon-based soil maps have been useful for generalized land use planning and management, but their drawbacks are numerous, including that they are (i) static without representing the dynamics of soil conditions (e.g., nutrient depletion), (ii) inflexible for quantitative studies (e.g., C balance) because they require functional soil properties, (iii) losing information because soil classes provide a summarized account of the soils of a region, (iv) providing soil data often represented at a scale that is seldom useful for particular questions, and (v) difficult to integrate with other grid-based resource data (e.g., satellite images and DEM). In addition, crisp soil maps do not provide uncertainty and error assessments. This is in contrast to pixel-based soil prediction models that provide error metrics from cross-validation or validation procedures. With the advent of advanced computational capabilities (e.g., cloud computing), large soil grids can be processed, overcoming the limitations of the pre-digital era, which constrained soil maps to vector-based formats.
Implications of Soil Mapping and Modeling across Space and Time
The phenomena of space and time, the distributions of soil properties and processes at escalating spatial scales, and the predominant current DSM approaches are summarized in Fig. 3. Some inherent soil and environmental phenomena on Earth cannot be changed, such as geographic domain space and entropy of a soil ecosystem, whereas mapping of soil properties and processes and soil representation models can be adapted. As the spatial scale increases from fine (field) to coarser scales (landscapes, continents, and global), the increasing extent and geographic domain space translates into increased variance of soil attributes (McBratney, 1992, 1998) and increased Shannon's information entropy, which measures the diversity or disorder (unpredictability) of a system (Vieux, 1993; Culling, 1988; Martin and Rey, 2000; Seuront, 2010). If a landscape exhibits constant soil property values, the probability to predict them is high, at 1.0, resulting in zero entropy and zero uncertainty; however, many research studies have documented the high variability of soil properties (Cambardella et al., 1994; McBratney and Pringle, 1999; Lin et al., 2005). Increasing variances also mean that spatial and temporal autocorrelations of soil and environmental landscape properties increase at escalating spatial scales (Vasques et al., 2011).
Upscaling of Site-Specific Soil Observations to Global Scales
Soils have been mapped based on genetic horizons or fixed depth intervals within soil profiles because the representation of soil properties and processes requires time and space to be turned into discrete units. As the spatial scale increases, larger pixel (grain) sizes or map unit polygons have been used to represent soil properties. This increase in pixel or grain size inherently influences the upscaling behavior of soil models, which are dependent on the grain, extent, and variance of soil-environmental observations (Vasques et al., 2011). The three sampling scales of spacing, extent, and sample support, termed the scale triplet by Blöschl and Sivapalan (1995), impact up- and downscaling of soil models. Scale-independent behavior (self-similar or fractal behavior) assumes that the coarser scale system behaves like the average finer scale system, which implies that processes are linear. Such linear behavior of soil processes may only occur across a specific range of scales, reaching a threshold (the “tipping point”) at which processes become nonlinear, resulting in multifractal behavior. Nonlinear dynamics with thresholds, hysteresis, and alternate states are well known in ecological systems (deYoung et al., 2008; Contamin and Ellison, 2009) but poorly investigated in the soil science discipline. Such inherent scaling behavior suggests that it is incorrect to assume that simple aggregation of soil property or process data at fine grain (or map units) to coarser grain will represent the soil system at global or continental scales. In essence, summing soil map units or pixels may not correctly represent the whole soil system because of the nonsimilarity of soil property or process behavior across spatial scales. In the past, however, simple aggregation methods have commonly been adopted to generalize regional soil maps (e.g., a map scale of 1:250,000 or coarser) to global maps (e.g., 1:1,000,000) to provide global soil estimates lacking uncertainty assessment and an understanding of the scaling behavior of soils.
To upscale our understanding of pedogenic processes from the pedon to regional and global scales, it is critical to quantify the scaling behavior between soil properties–processes and natural and anthropogenic drivers, which may become nonlinear and lack stationarity at escalating spatial and temporal scales. Our understanding of the scaling of soil properties and processes is still limited, partially due to the soil models and representations as described above. At fine spatial scale, pedogenic, hydrologic, and many ecosystem processes have been represented at high temporal resolution for biogeochemical speciation and in-depth soil process studies, whereas time steps for monitoring usually increase at escalating spatial scales. These spatial and temporal discretization phenomena illustrate how the relationships between soil and environmental properties, which are used in many factorial-based DSM projects, tend to decrease in strength as the spatial scale increases from the field to large landscape scales. Worldwide, operational digital soil map products have been focused on soil classes, whereas research-oriented digital soil map products have emphasized quantitative factorial modeling using CLORPT (CLimate, Organisms, Relief, Parent material, and Time) (Jenny, 1941) or SCORPAN (S, soils; C, climate; O, organisms, biotic factor; R, relief; P, parent material; A, age; and N, space) (McBratney et al., 2003) and statistical and geostatistical methods to estimate soil properties and classes (Grunwald, 2009).
Digital Soil Mapping Projected into the Future
Besides the provision of soil taxonomic and property data at coarse scales, there is need for interpretations of soil–environmental relationships and formation of higher level soil biogeochemical and soil process models formalizing knowledge of pedogenic and ecosystem processes. The latter have been developed at local (site-specific) scales (Minasny et al., 2008; Rasmussen et al., 2010); however, upscaling these models to larger spatial and temporal scales is hampered by the large amounts of soil and environmental input data required to run them. This drawback has created opportunities for global climate change simulation models (Intergovernmental Panel on Climate Change, 2007) to address critical questions pertaining to the Anthropocene, which are used at coarse spatial resolutions to model ecosystem changes and forcings but often rely on generalized (historic) soil data inputs with high uncertainties. Other studies have used meta analysis, which synthesizes disparate soil data to derive new knowledge addressing critical questions of the Anthropocene but often relies on scarce data sets. For example, global meta analysis was used by Post and Kwon (2000) to assess soil C sequestration and land use change across various ecosystem types, by Guo and Gifford (2002) to estimate soil C stocks and land use change, and by Bond-Lamberty and Thomson (2010) to assess global soil respiration.
Digital soil mapping can advance in the future if the window of perception widens, as the density of observations in space and time increases, and mapping techniques advance (Fig. 4). Improved soil sensors and in situ soil data collection methods would allow the collection of soil data at higher spatial and temporal resolution and at denser grids at escalating spatial scales. We envision modeling continuous soil depth functions in soil profiles, rather than crisp soil horizons, and developing three-dimensional soil-landscape models (Grunwald, 2006c; Malone et al., 2009). Denser soil data sets would allow us to more accurately quantify the relationships between soil and environmental attributes (SCORPAN approach), where the density and scale of soil observations resemble more closely the spatial resolution of the SCORPAN factors, and to elucidate the scaling behavior of soil properties and processes across spatial and temporal scales. Because the assessment of spatial and temporal autocorrelations are dependent on the density, amount, and distribution of soil observations within a landscape, advanced soil collection methods would also advance predictions of digital soil models in space and time. Thus, major advancements to quantify soil properties and processes are highly dependent on improvements in inferential delineation of soil observations and less so on improved methodologies or models. It has been demonstrated in various studies that soil predictions can be made at ≤30-m pixel resolutions (Thompson and Kolka, 2005; Grunwald et al., 2007;
Vision for Global Digital Soil Mapping and Modeling: The Future Soil Pixel
Given the constraints and issues related to currently available digital soil data at continental and global scales, the urge to identify an ideal soil pixel is high. Such a soil pixel (i) is knowledge rich (i.e., provides detailed pedogenic information), (ii) provides an ideal dimension (width by length) to represent the spatial variability of multiple soil properties and outcomes of soil ecosystem processes, and (iii) is contiguous in space and time across continents and the globe. This future soil pixel is probably not uniform and universal across the globe, depending on the geographic location, inherent soil variability, relationships among soil-SCORPAN factors, and disparately acting anthropogenic and natural forcings, which transform and reshape the soil pixel as time evolves. Although this soil pixel of the Anthropocene is unknown, we propose the following conceptual modeling framework to explicitly account for anthropogenic and natural forcings that determine and modulate soils:where SA is the target soil property (e.g., soil organic C), S represents ancillary soil properties (e.g., soil texture, soil spectral data), T represents topographic properties (e.g., elevation, slope gradient, slope curvature, compound topographic index), E represents ecological and geographic properties (e.g., physiographic region, ecoregion), P is the parent material and geologic properties (e.g., geologic formation), A represents atmospheric properties (e.g., precipitation, temperature, solar radiation), W represents water properties (e.g., surface runoff, infiltration rate), B represents biotic properties (e.g., vegetation or land cover, land use, land use change, spectral indices derived from remote sensing, organisms), H is human-induced forcings (e.g., contamination, greenhouse gas emissions), j is the number of properties from j = 1, 2, …, n, px is a pixel with size x (width = length = x) at a specific location on Earth, tc is the current time, ti is the time to tc with time steps i = 0, 1, 2, …, m, and z is depth. (Note: the H factor includes human activities that force the change, such as greenhouse gas emissions, which may alter other factors. For example, global climate change is the result of feedbacks of anthropogenic and natural forcings and processes. The actual change in climate is represented by A.)
The envisioned STEP-AWBH (phonetically, “step-up”) model is spatially and temporally explicit and enhances previous factorial modeling frameworks. The soil property of interest, SA, is estimated from various spatially explicit environmental variables (STEP) that tend to be static within a human time frame and thus is represented in the model at one time (tc or, if available, ti). Soil properties (S) may be derived from existing soil maps or databases, field or laboratory investigations, and include soil taxonomic classes, physicochemical and biological properties, and soil spectral or other soil properties derived from sensors (e.g., VNIR, LIBS, electromagnetic induction, or γ radiometrics). Geologic (P) and ecological (E) properties usually show orders-of-magnitude higher spatial variation than soil properties and stratify a given landscape into subsets. The T factor represents a variety of primary and secondary topographic properties derived from DEM as described by Wilson and Gallant (2000). The pixel size (px) may vary among STEP factors because of disparate spatial autocorrelations and the variability of STEP attributes across a given landscape.
The AWBH factors explicitly account for space (i.e., pixel location) and time, whereby the time component may be aggregated to represent different time vectors. Atmospheric (A) properties (e.g., temperature and precipitation) include the current climate at time tc at a location (px) derived from regional and global climate monitoring networks; the seasonal impacts of climate on SA(z,px,tc) represented by the aggregation of climatic properties across weeks or months preceding tc; and the longer term trends in warming or cooling, droughts, or other climate oscillations, such as the El Niño Southern Oscillation, on SA, which can be represented by annual or decadal climatic aggregates. Similarly, the W factor represents time-dependent knowledge of water quantity and quality related features (e.g., soil moisture, water table amplitude, or mean suspended sediment values within a drainage basin). Biotic (B) properties, such as natural or human-induced vegetation or land use change can be assessed through remote sensing imagery or aerial photographs across a specific period of time (ti). Remote sensing imagery is globally available, such as the European Space Agency's Earth Observation data on soil moisture (soil moisture and ocean salinity mission), the Moderate Resolution Imaging Spectroradiometer (MODIS, 250–1000 m), Landsat (30–60 m), Advanced Spaceborne Thermal Emission and Reflection Radiometer (ASTER, 15–90 m), and regionally available at fine spatial resolutions, such as Quickbird (0.61-m panchromatic, 2.44-m multispectral), GeoEye1 (0.41-m panchromatic, 1.65-m multispectral), and light detection and ranging (LIDAR) at submeter resolutions. Remote sensing images have been used widely to infer the biophysical state and composition of vegetation, land use, and aboveground features (e.g., biomass and C content). The H factor represents different anthropogenic forcings that can act across shorter or longer periods of time on SA(z,px,tc) to shift SA into a different state, such as greenhouse gas emissions, contamination (e.g., an oil spill), disturbances, overgrazing, and others. The modeling framework represented by Eq.  can be adapted to model soil degradation, losses in fertility, functions and values of soils, and more. Depending on the geographic setting and history of a soil landscape, one (or more) of the factors in Eq.  impart(s) major control on SA, which may shift into a different state due to scaling up of models to coarser landscape scales, crossing geographic or attribute domain boundaries, or disproportional impact of anthropogenic forcings. The STEP-AWBH model can be adapted to forecast and hindcast soil properties; however, answering critical questions of the Anthropocene will require investment in the spatially explicit collection and monitoring of soil data to populate such models and test and validate results.
Carré et al. (2007) proposed to go beyond DSM and outlined the concepts of digital soil assessment (DSA) and digital soil risk assessment (DSRA). Digital soil assessment is the quantitative modeling of difficult-to-measure soil attributes, necessary for assessing threats to the soil (e.g., the decline of soil organic matter or biodiversity and erosion) and soil functions (e.g., biomass production) using DSM outputs. Digital soil risk assessment is the quantitative evaluation of soil-related scenarios for providing policy guidance using the outputs from DSA fused with socioeconomic data and more general information on the environment (Carré et al., 2007). The main advantages of a DSM–DSA–DSRA chain are reduced costs; formalized, consistent, and transparent methods; and models that are easily updated and allow assessment of error propagation for soil risk assessment. Fused soil systems facilitate higher order modeling of complex soil-environmental systems that are exposed to natural and human-induced stressors. Predictions of SA pixels are viewed not as the endpoint but as a stepping stone toward the assessment of soil functions, soil quality, the risk of soil degradation, and ecosystem services. Ecosystem services assess “the benefits human populations derive, directly or indirectly from ecosystem functions or ecosystems” (Costanza et al., 1997; Millennium Ecosystem Assessment, 2005); thus, they add the socioeconomic dimension to products derived from DSM. Bouma (1997, 2001) has advocated the role of soil science (and soil scientists) in environmental, social, and economic policy, particularly the application of soil maps and quantitative methods to land use management (e.g., Wösten et al., 1985; Droogers and Bouma, 1997; Sonneveld et al., 2002). In the future, closer linkages between soil prediction maps or models and environmental and socioeconomic applications are desirable to assess the value of soils as “soil natural capital” (e.g., Hewitt et al., 2010)—and of equal importance, as “monetary capital” or “social capital” in a global, interconnected world.
Vision for Global Digital Soil Mapping and Modeling: The Modern Soil Scientist
The future of DSM and modeling depends not only on the scientific expertise of researchers but also on stakeholder needs, people's perception of soil information, and the training and education of the modern soil scientist. For instance, current qualifications for a soil scientist in the NRCS, the agency responsible for leading and coordinating activities of the National Cooperative Soil Survey and for advancing soil survey technology for global applications (Soil Survey Staff, 2011b), are “a bachelor's degree or higher in soil science or a closely related discipline that includes 30 semester hours or equivalent in biological, physical, or earth science with a minimum of 15 semester hours in such subjects as soil genesis, pedology, soil chemistry, soil physics, and soil fertility” (NRCS, 2011a). While these are sound fundamental qualifications, better education and training are required for DSM and modeling at multiple spatial scales. In addition to the theory and field application of pedology, the modern soil scientist should (i) be able to access and manipulate geospatial, topographic, remotely sensed spectral, and proximally sensed spectral data to represent soil and other environmental covariates, and (ii) have strong quantitative skills, particularly in statistics and spatial analysis. Along with an increasing public awareness and use of geospatial information (e.g., online mapping and Earth viewing tools, portable GPS units), there is an increasing number of college-level courses available in geographic information systems and analysis. It is often difficult, however, to find advanced courses in nonparametric statistics, spatial analysis, and sampling design useful in DSM and modeling. We propose that the education and training of the modern soil scientist can be achieved via the development and delivery of short courses that focus on specific knowledge and skills. The short-course model for training and education of the modern soil scientist can help (i) innovate curriculum development, particularly for the education of graduate students and professionals, (ii) accelerate curriculum revision, and (iii) provide opportunities for professional societies (e.g., the SSSA and the International Union of Soil Sciences) to be actively involved in advancing DSM and modeling by facilitating short-course offerings in conjunction with international, national, and regional meetings.
The challenges facing human societies in the Anthropocene will require spatial information about soil resources that can be used by planners, modelers, scientists, and policymakers. This information must be digital, be compatible with other geospatial resource and environmental data, and convey knowledge of specific soil properties and processes across three-dimensional space and time with estimates of uncertainty. Existing soil maps are generally inadequate because of limitations that include scale, currency, completeness, logical consistency, and lineage. Future digital soil maps must be pixel-based, multiresolution representations of spatial and temporal patterns of soil properties. Our vision for global DSM provides a modeling framework that will support the development of this future soil pixel based on STEP-AWBH, as well as a context for the training and education of the modern soil scientist to use DSM and advance toward DSA and DSRA.