Context. - Many ferralsols and podzols from central Amazon were developed by a long-time pedogenesis over a sandy-clay Cretaceous sediment which mainly consists of kaolinite and quartz [Lucas et al., 1996]. The soils progressively range from clayey microaggregate ferralsols on the plateaux associated with tropical forests, to podzols at the bottom of thalwegs and in some small declivities of the largest plateaux associated with forests of smaller trees or open savannahs [Chauvel et al., 1987; Bravard, 1988; Lucas, 1989; Cornu, 1995]. Their moisture properties are different from those of temperate soils [Arruda et al., 1987; van den Berg et al., 1997; Tomasella et al., 2000].
Methods. - To quantify the moisture properties, mercury injection porosimetry (MIP) seems to be an adequate method [Vachier et al., 1979]. It is easy to use and enables accurate investigation of the major part of the porosity spectrum. We used a Carlo-Erba 2000 which enabled 12 to 20 measurements for pore entry radii from 4 nm to 0.1 mm. It requires dried (here, air-then oven-drying), centimetre-sized samples. The drying of the sample, the large pressures of Hg used, the different surface properties of soil components in presence of water or Hg, may induce discrepancies between MIP data and water desorption measurements, particularly if the organic matter (OM) is abundant. We compared MIP and water desorption data, and established pedotransfer functions to estimate the water desorption curve from MIP data and OM content.
Water desorption was performed with a pressure membrane equipment, which allows us to investigate pores with entry radii from 0.1 μm to 0.5 mm [AFNOR, 1996a], on decimetric samples. We also used water desorption data obtained in situ in other studies from similar sites [Tomasella and Hodnett, 1996]. For some depths, drying under controlled atmosphere [AFNOR, 1996a] enabled us to investigate pores with entry radii down to 1.5 nm.
The total pore volume was determined on the same dry or humid sample (volume 100 cm3) by cylinder measurement or using the paraffin method [AFNOR, 1996b], in order to determine the volume of large pores. Organic carbon mass ratio (C) data are from Bravard  and Cornu ; a C value of 1 % corresponds to a volume ratio of organic matter to solid soil of 4.8 to 6 %.
Notations. - We classified each pore class from its pore size (r), deduced from pressure data with the Laplace law (r is the entry radius for cylindrical pore model or the entry width for slit-shaped pore model):
- nanopores (r < 4 nm) and cryptopores (4 nm < r < 0.1 μm) are residual pores, containing residual water that roots cannot extract. It moves by evaporation in dry conditions;
- micropores (0.1 μm < r < 10 μm) contain bioavailable water;
- mesopores (10 μm < r < 0.1 mm) and macropores (0.1 mm < r) are determinant for hydraulic conductivity, from which water drain in some hours (mesopores) or seconds (macropores) after the rain.
We described their abundance in terms of volume, in order to better quantify their geometrical structure, and took as a reference the volume of dry solid matter in the soil, because the total soil volume is not constant (it depends on moisture content). Then we used the partial void ratio (u) for MIP, the water ratio (n) or the air ratio (a) for water desorption, and the fluid ratio (e) for total porosity:
MIP and water desorption give directly the values of ucrypto, umicro, umeso, nres, nmicro, nmeso. Total pore volume measurements give ehumid and edry.
We must be aware that the maximum water content remains below the fluid content, because of residual air: air bubbles trapped during wetting (we tried to minimize this effect by a progressive, two-week-long wetting), air linked to hydrophobic soil organic surfaces, and air in non-connected pores (not here). Residual air can thus be neglected for MIP, and for the water desorption of samples without organic matter. By subtraction, we obtain:
We can extrapolate the ratio of dry nanopores (unano) with reasonable accuracy because its value is anyway very low. For the ferralsols between 1.1 and 1.7 m depth, (umacro + unano) ≤0.03. We added a small triangle to the left of the porosity spectrum, in order to have unano = 0.01 at 1.4 m depth for the ferralsol. Then unano ranges from 0.002 for the podzol to 0.02 for the ferralsol above 0.7 m depth.
All the porosity spectra we obtained had very few small micropores. Two porosity domains can thus be defined: textural porosity below 0.25 μm pore size, and structural porosity above. We described these two domains using their logarithmic average pore entry radii (rtext or rstruct) and their lobe width λ,. These parameters were defined by fitting an Ahuja and Schwartzendruber curve  (often named a symmetric, m = 1, van Genuchten curve [van Genuchten, 1980]):
[Log(rtext) − λtext;Log(rtext)+ λtext] is the interval gathering half of textural porosity, with λtext = Log3/νtext and it is the same for structural porosity. We set ustruct from e measurements, to give it a physical meaning, and introduced it as a value of associated with the 1mm pore size during the fitting, so that all curves approach their asymptotic value for about the same pore size.
Results on porosity spectra. - The carbon content declines with depth, with a local maximum in the Bh horizon for the podzols (fig. 4). Ferralsols have a high residual porosity and average meso- and macroporosities (fig. 2 and 5), podzols have basically high meso- and macroporosities (figs 3 and 6), and transition soils are intermediate. For ferralsols and transition soils, named clayey soils here, all porosity classes have typical variations with depth: maximum at the surface; local minimum at a compacted horizon at 0.13 to 0.3 m (ferralsol) or 0.4 to 0.6 m (transition soil); local maximum for meso- and macroporosity, at the « termites’ horizon » at 0.4–0.6 m (ferralsol) or 0.8–1.2 m (transition soil); constant values for residual porosity below a depth of 1.2 m, and decreasing values for the other porosity classes. The average textural pore size increases vertically from the surface to the depth, with a local minimum at the compacted level, and laterally from ferralsols to podzols (fig. 7 and 10).
Interpretation. - Residual porosity is correlated to clay content and textural pore size is correlated to the kaolinite particle size (unpublished work). The compacted level, named biological plough pan [Chauvel et al., 1987], may be due to a gravitational accumulation of the finest kaolinites from the above horizons, at the foot of the highly pedoturbated level. Low microporosity is linked to a very low silt and fine sand content, typical of tropical soils. Meso-and macroporosities are due to a clay microaggregation for clayey soils, and to the coarse sand fabric for the podzol. Microaggregation of clay is enhanced near the surface by root or ant pedoturbation, and at depth by termit pedoturbation, and maybe lateral water flows on the slope.
Dispersion. - The dispersion of the values is low for a depth between 0.7 and 1.5 m, which is due to a homogeneous kaolinitic plasma. The dispersion increases with pore size, surface proximity for depth 0.7 m, or the increasing depth below 1.5 m. Near the topsoil, this is due to the heterogeneous biological activity, giving an heterogeneous soil structure. In horizons below 1.5 m, the causes are micronodules, localised water flows, and insufficient root pedoturbation. Therefore, we took more samples near the surface, we quantified soil shrinkage from the same sample before and after drying, we gave a higher statistical weight to data from 0.7 to 1.5 m depth in regressions, and smoothed values with depth (table IV).
Comparison of results on dry or humid samples
Total poral volume. - The shrinkage of the ferralsols samples decreases with depth. On the contrary, the shrinkage of podzol samples increases with depth (fig 5, 6 and table I).
Residual porosity and microporosity. - OM presence is correlated to a decrease in the residual and micro- porosity given by MIP, compared to water desorption data. For soil samples with a low organic matter content, MIP gives a lower residual porosity especially for fine clays and a higher microporosity (table I). We could establish one pedotransfer function for all the soils studied here by taking into account the average textural pore size (table II and fig- 8).
Meso- and macroporosity. — Through drying, mesoporosity increases for clayey soils and shrinks for podzol. These changes are more important for deep than for shallow horizons. Through drying, macroporosity increases, except for deep horizons of clayey soils (fig. 5,6; tables I, III).
Textural porosity of ferralsols and transition soils. - The progressive vertical and lateral variation of MIP data gives reliability to a comparison with the two points where the textural porosity of humid samples was measured. MIP gives a higher pore size (roughly doubled, see table II) and a lower textural lobe width (roughly halved), compared to water desorption data.
Structural porosity. - Our measurements on dry samples, compared to water desorption data, give simultaneous variations of three parameters, average structural pore size, structural lobe width, and macropores/mesopores volume ratio: an increase for the upper horizons of clayey soils or the deep horizons of podzol, a low change for the upper podzol horizons, and a decrease for the deeper horizons of clayey soils (fig. 9 and 10). However, the dispersion on these parameters is too high to predict them for humid samples from our data on dry samples. In order to parameterise the water desorption curve, we propose to apply equations of table I to subdivisions of micro- and mesoporosity, sharing out the organic matter influence between these subdivisions to get a precise water desorption curve, that can then be parameterised [Bastet, 1999].
Interpretation. - Hysteresis and pore geometry effects could not explain differences between water desorption and MIP, as in both cases the non-wetting fluid (air or Hg) enters pore space, and the same pore geometrical model was applied. The contact angle used in the Laplace law may change with the soil composition, especially with OM content for water desorption (hydrophobicity). Then we may have overestimated humid pore sizes at shallow horizons. Correcting this would enhance the discrepancies observed here. Some phenomena are not taken into account in the Laplace law: compression of porosity during MIP [Penumadu and Dean, 2000], or water physiosorption on mineral surfaces during water desorption [Tuller et al, 1999]. The first phenomenon causes the Laplace law to overestimate dry pore sizes (for r < 2 μm), the second causes it to overestimate humid nanoporosity. Both explain that MIP gives a higher microporosity, a higher textural pore size, a lower residual porosity.
The drying of samples causes porosity shrinkage, diminishing pore volumes and pore sizes. Then some macropores of clayey soils could turn into mesopores. However, the shrinkage at a given scale may cause the opening of the porosity at a higher scale, with cracking. Thus, the shrinkage of residual porosity can explain the increase of microporosity for all horizons, the increase of mesoporosity for clayey horizons, and the increase of macroporosity (eye-visible cracks), for shallow clayey horizons. The shrinkage of podzol mesoporosity can explain the increase of its macroporosity as well.
The amount of shrinkage due to drying, or compression during MIP depends on soil OM content and type, soil structure and soil texture. Soil texture could explain a smaller shrinkage of total porosity for podzol than ferralsol, for shallow horizons: the lentil-shaped kaolinites, depending on their relative positions, are more likely to occupy a variable volume than the quartz grains, round-shaped with cavities. Soil structure is determined by its pedogenesis: eluviation/illuviation, pedoturbation. The unexpected large shrinkage for deep podzol seems to be due to a loose structure, inherited from eluviation of clay. Pedoturbation, together with OM presence, may either create porosity, or seal or compact porosity [Sokolowska and Sokolowski, 1998 ; Chauvel et al., 1999]. They may create a new porosity likely to shrink in clayey soils and on the contrary they may consolidate the loose structure of podzols.
Conclusion. - We established pedotransfer functions to define the water desorption curve from MIP, organic carbon content, and bulk density (table I and II). These functions can be used for soils with the following characteristics, thus for most tropical soils:
- negligible fine clay (< 0.2 μm) content, in order to enable the extrapolation of nanopore volume from MIP, without needing nitrogen desorption [Bruand and Prost, 1987];
- soils mainly composed of kaolinite and quartz. The organic carbon content has little influence on the humic podzol studied here, so the equations established here are pertinent for other podzols.
These pedotransfer functions give information on the structure of these soils :
- residual and micro- porosities modify between MIP and water desorption, according to the organic carbon content and the texture of clay (given by the average textural pore size), with the same law for all soils;
- mesoporosity changes according to two different types of structure : (1) clay microaggregates for clayey soils, (2) sandy fabric for podzol, loose at depth due to eluviation of parent clay, consolidated near the surface by pedoturbation;
- total porosity follows, with an attenuation, the changes in the MO-linked porosity and in the porosity linked to soil structuring particles, that is the residual porosity for clayey soils, and the mesoporosity for podzols;
- macroporosity, in reality as in the equations, compensates for the difference.
This method could be tested on soils with other granulometry, mineralogy, or higher OM content.
This work was supported by IRD and the PEGI and PROSE programs, we thank Max Sarrazin for his technical collaboration and the reviewers for their critical reading.