Field Perception of the Boundary Between Soil and Saprolite by Pedologists and its Differentiation Using Mathematical Models

Saprolite plays a central role into hydrologic and nutrient cycles. Despite that, saprolite research is scattered and uses heterogeneous, sometimes conflicting, methods and concepts. During field work, it is difficult to assign the boundary between soil and saprolite. This paper aimed to identify the subjacent logic that pedologists use to assign to a regolith volume its soil or saprolite nature. To achieve this goal, a tree algorithm was used to build a hierarchy of physical and chemical properties of a set of regolith profiles. Such hierarchization expose the inner, subjective criteria used by researchers during the assignment of a certain profile zone as saprolite or soil. The following variables were measured: total porosity (TP); bulk density (Bd); particle density (Pd); total Fe2O3, Al2O3, CaO, MgO, K2O, Na2O, P2O5, and TiO2; selective extraction of iron by ditionite-citrate-bicarbonate (FeDCB) and ammonium oxalate (FeOA); and the FeDCB/FeOA ratio. These measurements were done in a set of 25 regolith profiles (137 horizons and layers), located in the Southeast region and Northeast region of Brazil. The decision tree algorithm was applied using the recursive partition method to identify which of the measured property was most strongly associated with the field assignment of the pedologists to a certain profile zone as saprolite or soil. The Bd, FeDCB/FeOA, MgO, CaO, TP, and P2O5 explained 93 % of the pedologists choice, being Bd responsible for 81 %.


INTRODUCTION
Recent approaches such as Critical Zone and Planetary Boundaries (Rockström et al., 2009) shed light on the importance of the whole regolith to sustain ecosystems and the human societies (Brantley et al., 2007).
The regolith is the section of the lithosphere column changed by weathering, being further divided into soil and saprolite (O'Brien and Buol, 1984). In shallow soils, saprolite is close to the surface and may become a nutrient source and water reservoir to plant development (Melo et al., 1995;Pedron et al., 2009;Santos et al., 2017), but also as a shortcut for surface pollutants to reach underground water.
As a natural resource, characterization and mapping of saprolites are needed for their better use and management. Despite its importance, the concepts and definitions of saprolite are quite diverse, even controversial.
Conceptualization, definition, and characterization are standard operations to allow the registering, organization, classification, and mapping of saprolites. The establishment of a common procedure worldwide would provide the basis to share knowledge and collaborate towards a global understanding of this natural body.
Establishment of a sharp limit between soil and saprolite is debatable. However, classification systems require a definition of the object being classified. In this regard, the operational definition of saprolites should avoid overlapping the soil, that is, the same material should not be classified simultaneously into two classification systems.
At present, two saprolite classification systems were proposed in the soil science community. The Saprolite-Regolith Taxonomy -SRT (Buol, 1994) defines the saprolite as "regolith material that have unconfined compressive strength less than 100 MPa, and are either not penetrated by plants roots, except at intervals greater than 0.10 m, or occur more than 2.00 m below the soil surface, whichever is shallower". The Subsoil Reference Groups -SRG defines "saprolithic material is little affected by pedogenetic process and represents in situ weathering product of the original rock" (Juilleret et al., 2016), and classifies the materials below the lower soil limit of the World Reference Base -WRB (IUSS Working Group WRB, 2015). This paper is based on the perception of two pedologists in describing regolith profiles and assigning the soil-saprolite boundary disregarding any classification system. The study aimed to identify the criteria pedologists use to assign to a certain regolith volume its nature as soil or saprolite, by comparing the saprolite and soil sets made by the algorithm to those made by the pedologists. By doing so, we could identify the laboratory measurements that correlate with other field perception.

Obtaining the data
The data were collected from 25 regolith profiles (P1 to P25) described by Guerra (2015) and Santos (2015) in their thesis, summing up to 137 horizons and layers, developed from: granite, syenite, gneiss, schist, sandstone, and siltstone (Table 1).

Analyzed variables
The variables measured into the lab and considered in the decision tree algorithm were chosen among the most affected by weathering and pedogenesis. Only samples analyzed by the same or similar procedures were considered.
Rev Bras Cienc Solo 2019;43:e0180104 The bulk density (Bd) was determined by the volumetric ring method. After measuring the dry mass (dm) and volume (dv) of the material, the bulk density (Mg m -3 ) was calculated by equation 1: It is worth noting that for profile 21 (P21) there is no result of bulk density for saprolite.
(4) Profile classification was corrected in relation to the thesis (Guerra, 2015).

Data analysis
The decision tree algorithm used the recursive partition method and the Deviance function from the R software library (Ripley, 2016). The algorithm provided a cutting value to split the initial set of samples into two subsets, and a looping procedure further splits each sub-set into two sub-sub-sets and so on, until a limit value is reached or a single object remains in the set. Metaphorically, the method splits the data as a tree splits from the trunk towards the branches and leaves. At each step, the procedures identify a variable and a cutting value that maximize an impurity measurement (Rodrigues, 2005).

RESULTS AND DISCUSSION
The variables that contribute most to group the samples into subsets were Bd, Fe DCB /Fe OA , MgO, CaO, TP, and P 2 O 5 ( Figure 3).

Main variable
The most important variable was bulk density (Bd) which alone explained 81 % of the sample clustering, that is, bulk density was the variable that best fit the pedologists criteria to decide the place of the soil-saprolite boundary (Tables 2 and 3).
Saprolites usually have greater bulk density than soil (Oliveira, 2012) (Figures 3 and 4a), because saprolites tend to be less porous, have less organic carbon, and are also compressed by the weight of the overlaying soil. However, in the present paper, we found profiles in which the saprolite density was smaller than the soil density. In all cases, these profiles had a textural horizon (Table 2 and Figure 4b). The probable origin of the term "saprolite" dates back to the 19st century when Becker (1895) defined it as "the non-transported weathering product which has very little or none loss of volume as related to the original rock". By this concept, the solid phase saprolite is both the residual and neoformed material, and the associated porous system (Calvert et al., 1980;Kretzschmar et al., 1997), resulted from rock weathering. Since the volume is maintained (isovolume) the loss of mass during the alteration of minerals imply in a decrease in bulk density and increase in the porous system (Costa and Cleaves, 1984). The further loss of isovolume in saprolites may occur both by collapse of the saprolite volume due to the overgrowth of the porous system beyond its capacity to sustain the weight of its own weight and of the soil column above it; or by expansion due to the formation of peds and increase in organic carbon (Stolt et al., 1991).  For the sake of simplicity, soil materials were named "horizons" and saprolite materials, "layers". The 81 % agreement obtained using only Bd (first node of the decision tree) means that it missed 15 horizons and 11 layers, from a total of 88 horizons and 49 layers ( Figure 5).
The use of the variables of the three first nodes (Bd, Fe DCB /Fe OA , and MgO) increased the agreement only by 4 %, that is, up to 85 %, missing 7 horizons and 5 layers ( Figure 5). Using all the nodes/variables (Figure 3), the final percentage of agreement between the tree and the pedologists were 93 %.
Most of the samples in disagreement were from metamorphic rocks, particularly schists (Table 1). This suggests that it was more difficult for the pedologists to maintain their criteria when judging saprolite materials inherited from rocks with heterogeneous structure. As Price and Velbel (2003) pointed out, saprolitic materials evolved from heterogeneous rocks are also heterogeneous, entangling the judgement.
Despite these difficulties, the Fe DCB /Fe OA ratio and Bd, taken together, resulted in an error in only three samples, when considering the gneisses. These profiles have in common thinner soil-saprolite transitions, all at depths smaller than 1.00 m. This observation suggests that the contribution of the Fe DCB /Fe OA ratio depends on the degree of weathering/ pedogenesis and the abundance of Fe in the parent material. Bulk density determined by the method of the volumetric ring and paraffin-shaped fragment (Teixeira et al., 2017). Bd = bulk density.

Secondary variables
The variables other than Bd were considered secondary due to the much smaller contribution they did to the overall agreement between the pedologists and the decision tree ( Figure 3).
The total magnesium content (MgO) and the Fe DCB /Fe OA ratio increased only 4 % the agreement between pedologists and the decision tree (from 81 to 85 %), figure 4. The use of variables, such as MgO, is very dependent on the parent material composition. On the other hand, because the ammonium oxalate (Fe OA ) solubilize preferentially the less crystalline oxides (Schwertmann, 1973), and the DCB (Fe DCB ) the pedogenic ones (Mehra and Jackson, 1960), the ratio between the two is less dependent of the total amount of iron.
The fast precipitation of iron during the weathering of iron bearing minerals at the weathering front tends to produce less crystalline oxides, which further, during the pedogenesis, tend to reorganize themselves in more crystalline forms. Therefore, the Fe DCB /Fe OA ratio tends to increase as the profile evolves (Stolt et al., 1991;Pedron et al., 2015).

CONCLUSIONS
The decision tree methodology allowed to estimate the best variable to separate soil from saprolite under the conditions of the present study was the bulk density of materials. This variable alone explained 81 % of the grouping of materials (soil/saprolite) performed by pedologists. The improvement brought by all the variables studied in this mathematical model resulted in 93 % agreement with the logic adopted by pedologists.