**4. Discussion**

#### *4.1. Measured Changes in Bulk Density*

Ampoorter et al. [39], using a meta-analysis of data from several soil compaction studies including this one, found the number of skidding cycles was the only factor responsible for an increase in bulk density, but the relationship was weak. Soil water content was the only common measure of soil wetness considered and was not significant. However, other research has confirmed wet soils are more susceptible to compaction, and 60% to 90% of the increase in bulk density occurs in the first skidding cycle [27,40]. Soil water content alone is not a reliable variable to evaluate compactability of soil because the relationship between soil water content and maximum bulk density primarily depends on soil texture [17,41]. Higher compacted bulk density can only be achieved at lower soil water contents in course-textured soils [42]. Field measurements of soil water potential is a more reliable predictor of when a soil is most susceptible to compaction [11]. Soil resistance to compression increases as soil water potential becomes more negative [43].

Labelle and Jaeger [27] reported a slight increase in bulk density of soil in wheel tracks of forwarders lasting 12 to 24 months at two sites in New Brunswick, Canada. However, a statistically significant increase in the bulk densities of harvested and/or skidded soils after trafficking is unprecedented. Nevertheless, the statistical power of our data set and the consistency of the increase across all treatments and soil depths for at least 4 years make it difficult to assume that the increase is an artifact of sampling, natural variability, or an anomaly [44,45]. We believe that a combination of factors is contributing to the initial postharvest significant increase in bulk density, as well as a similar decrease in bulk density in the skidding treatments by year 7. In this situation, natural processes of decompaction [22] were ineffective. Instead, we considered postharvest changes in soil ecology [46], and the soil mechanics of deformation [41] as plausible factors responsible for an increase in bulk density after year 1 (Figure 5).

We hypothesize that the dynamics and intensity of the below-ground biological processes in conjunction with the protection of the soil provided by a mature forest canopy and intact forest floor allow mature forests to develop a slightly less dense soil. First, the upper 20 cm of forest soils commonly contain more than 80% of the roots in forested ecosystems [47]. To maintain this high root density, complex and dynamic processes of root mortality and regeneration are necessary, which are site and soil environment specific [48]. The root system also requires an equally active root rhizosphere biota to maintain its efficiency. As a result, a substantial proportion of annual gross primary production of forests is allocated to below ground processes [49]. These processes had not been disturbed on our sites for at least a century.

Clearcut forest harvesting removed the predominantly coniferous tree canopy, causing massive root mortality, stopping new root regeneration, and exposing the forest floor to more cyclic climatic variations. At the same time, the soil environment was also conducive to accelerated decomposition of organic matter [50]. As a result, a combination of root mortality, minimal root regeneration and rhizosphere maintenance, and accelerated rate of decomposition is most likely responsible for the natural consolidation of our soils. This consolidation resulted in the small significant increase in bulk density measured after year 1 (Figure 5). These combinations of processes would have occurred across all treatments but are solely responsible for the increase in bulk density in the harvest-only soil.

The increase in bulk density of trafficked soil after 1 year is attributed to a soil mechanics process. In soil mechanics, a small decrease in bulk density occurs whenever a consolidating force applied to a soil is removed [41]. The decrease in bulk density is referred to as rebound. Rebound is an instantaneous process that occurs once a load is removed from a soil and can only be measured in laboratory consolidation tests. Rebound is attributed to the elastic properties of soil [51]. The amount and kind of clay minerals in a soil are the primary determinants of the amount of rebound [52]. Consolidation tests of sieved agricultural soils have found that rebound results in a 2% to 5% decrease in bulk density after a consolidating load is removed. Rebound with similar decreases in bulk density has been reported for undisturbed forest soils [53]. Root systems of trees also increase soil strength [54] and would increase rebound because the root network would provide additional elastic resistance after a wheel passes.

For bulk density to increase in all trafficked treatments and depths at year 1, the soil apparently returned to its most dense condition, which occurred when the soil was under the wheel/track. The massive disruption of soil ecological processes [50] would have adversely impacted the biological elasticity of roots and other biota. It is this disruption that allowed the trafficked soil to 'reconsolidate' after 1 year. The reconsolidation is assumed to occur from the soil collapsing back into the same structural arrangemen<sup>t</sup> of soil particles and aggregates when under the wheel/track. The envisioned process is likened to expansion and contraction of a surface soil from the formation of ice lenses during a freeze–thaw cycle [55]. The increase was consistent across the three skidding treatments and three soil depths, which persisted for at least 4 years in the skidding treatments (Figure 5).

The values of bulk density at year 7 in all skidding treatments decreased to approximately the values measured immediately after trafficking (Figure 5). The most plausible explanation for the decrease in the values of bulk density for only the trafficked soil is that vegetation regenerated enough root growth and soil biota recovery to exploit the pore spaces created by rebound. A survey of vegetation cover of these sites at year 4 found the 12-cycle treatment had a vegetation cover of 80%, decreasing with less trafficking to the harvest-only area with a cover of 57% (Startsev, unpublished). Unpublished data on water infiltration and pore size distribution measured at year 5 at a few of these sites also inferred a slight improvement in macroporosity may have started. These data are supportive of the small decrease in bulk density measured by year 7 (Figure 5).

Soil on harvest-only areas failed to recover the porosity lost during the first year (Figure 5). More time is needed for successional development of a more mature forest canopy and forest floor. This development will support and protect the associated higher level of gross primary productivity found in soils under mature forests [48]. The process is generally slower in conifer dominated boreal forest ecosystems than elsewhere [49]. Other than the assumed role in reversing the rebound effect of soil compaction, biological processes had no other obvious role in restoring these compacted soils after 7 years.

#### *4.2. Freeze–Thaw (F–T) as a Decompaction Process*

Soil temperature and snow depth were measured on these sites because deep snow insulates the soil, which could limit the effectiveness of the F–T process [35]. The thermal conductivities of air, water, and ice are 0.025, 0.555, and 2.24 W m<sup>−</sup><sup>1</sup> <sup>K</sup>−1, respectively [56]. The thermal conductivities of snow range between 0.10 and 0.5 W m<sup>−</sup><sup>1</sup> <sup>K</sup>−1, which is

primarily a function of density. In seasonal snowpacks, thermal conductivity also increases with the weathering of the snow grains [57]. The thermal conductivity of snow is 5 to 20 lower than that of mineral soils [56]. Hence, snow is a high-quality insulation that slows the loss of heat from the soil, which slows or prevents soil from freezing.

Our data confirmed the insulating potential of snow during the first 3 years of monitoring (Figure 3), and that a snowpack of at least 40 cm kept mineral soil temperatures close to zero despite cycles of air temperatures dropping to −30 to −20 ◦C [56]. Thinner snowpacks were measured in 1997/1998, and later years only resulted in soil temperatures decreasing to −4 to −2 ◦C to a depth of 20 cm. The low soil temperatures were not sustained when air temperatures returned to above −20 ◦C. The warming of the soil is due to the heat released from melting ice in the soil [56].

When subzero soil temperatures are sufficient to contribute to soil decompaction is a complex question. Soil water must first undergo the process of nucleation, which is a structural transformation of water, before ice crystals can begin to form and grow [58]. The process only occurs in soil when water is supercooled. The nucleation temperature can be a few degrees below 0 ◦C in a pure clay–water mixture [59]. In natural fine-textured soils, nucleation generally occurs at soil temperatures < −1 ◦C, and initially occurs on the chemically inert surfaces of grains of sand [60] and high silt soils [61]. The growth of ice crystals next to sand grains pushes the finer soil particles away as the ice crystals grow, which can leave clean grains of sand visible on the surfaces of soil peds. Nucleation generally occurs at the surface of the same sand grains and locations when soil undergoes repeated cycles of subzero temperatures [55].

Without an external supply of water, the water/ice ratio decreases as ice crystals grow. For a high silt soil with 75% silt and 25% clay, Azmatch et al. [61] found about half the soil water had changed to ice at a temperature of −0.65 ◦C. At a temperature of −9.0 ◦C, approximately 25% of the soil water remained unfrozen. Surface chemistry of clay minerals slows the decrease in the water/ice ratio as the clay content and its chemical activity increase [58]. Soil must be nearly saturated if only in situ soil water is available for growing ice crystals that will be large enough to decrease bulk density [62]. In this situation, the maximum volume expansion of a saturated soil remains less than about 3% at a soil temperature of −6.7 ◦C. Such conditions are not common unless the soils are poorly drained. The partial freezing of soil water, including those high in silt, can increase the tensile strength of soil by 100-fold at a temperature of only −0.65 ◦C [61]. Therefore, the measurement of subzero soil temperatures between −2 and 0 ◦C or encountering frozen surface soils in the field cannot automatically be interpreted as an indicator that the F–T process may be decompacting the soil (Figure 3).

The environmental conditions for the F–T process to effectively decompact soil requires a substantial amount of soil water moving upward from deeper in the soil [58]. When the freezing front is stationary, water and vapor moving upward from lower soil layers causes ice crystals to grow into ice lenses and heave the soil upward [60,63,64]. The freezing front will advance deeper into the soil when the heat loss to the atmosphere exceeds the supply of water and heat to the freezing front from deeper in the soil. This advance can be triggered by colder air temperatures, decreasing unsaturated hydraulic conductivity of the nonfrozen soil because the deeper soil no longer contains as much water, or a combination of the two. Freezing of drier soil deeper in the profile will only produce small ice crystals within soil pores that cannot effectively decompact the soil [62]. This is a critical factor limiting the effectiveness of the F–T to decompact subsoils.

A sustained cycle of cold temperatures is more effective at cooling the soil profile. Abrupt decreases in soil temperature do not penetrate soil as deeply as slowly changing soil temperatures of the same amplitude occurring over a long period of time [65]. Snow cover, even a thin one, also moderates the temperature fluctuations (Figure 3) [66]. A sustained balancing of the heat loss to the atmosphere and input of heat from deeper in the soil is called the zero-curtain effect [67]. The depth of the zero curtain in our soils was about 50 cm. The most notable feature of the zero-curtain effect was a synchronized decrease in

soil temperature at all depths during the cold cycle starting on 3 January and persisting to 14 January 1998. The effect caused a slight change in the amplitude of soil temperature with increasing depth, and the absence of the normal time lag associated with the peak decreased in soil temperature with increasing depth [65]. This response is indicative of soil cooling as a result of the simultaneous growth of in situ ice crystals at all depths. Otherwise, water flowing upward in the soil during the cold cycle would have increased the volume of ice forming near the surface, and soil temperatures deeper in the soil would not have decreased as much or as quickly with increasing soil depth. As air temperatures increased at the end of the cold cycle, the recovery of soil temperatures was also a synchronized response. This response occurred because the source of most of the heat warming the soil at these depths originated from the latent heat of fusion released from the in situ melting of ice crystals. Our interpretation is also supported by the low saturated hydraulic conductivity of less than 2 × 10−<sup>6</sup> m sec<sup>−</sup><sup>1</sup> in these fine-textured soils starting at a depth of 10 cm [14]. The actual rate would have been 100 or more times slower than this value because of the lower viscosity of subzero water and the formation of ice crystals blocking the larger pore spaces. The zero-curtain effect persisted until early March. The occurrence of the zero-curtain effect under seasonal snowpacks may be a limited phenomenon in northern temperate forests. The effect would not occur in the absence of a temperature moderating snowpack or a sustained cycle of cold air temperatures occurring before the snowpack was established [56]. It would not have been nearly as obvious without a temporary large decrease in air temperatures previously discussed (Figure 3).

The F–T process is potentially most effective in surface soils where the cycles of freezing soil temperatures are greater and occur more frequently [55,60]. The potential for ice formation to decompact soil is reduced because ice normally forms horizontal lenses in the soil that collapse when the ice melts [55]. In forestry, this phenomenon is responsible for frost heaving small germinates and new planted seedlings out of the ground [60]. In soil engineering, most of the adverse impacts of F–T on roads occur from one to three F-T cycles; additional cycles cause minimal additional adverse impacts [68,69]. The F–T process in subsoils is seldom effective because cycles of F–T are fewer, unsaturated permeability much lower, and external water supply insufficient. While the surface soils of severely compacted temporary forest roads on high clay content soil have been partly loosed to depths of 10–15 cm, the subsoil remains compact and nonforested after 28 years [70]. The subsoil in two early wagon roads crossing the prairies in southeastern Alberta have also remained compacted after 80 and 100 years [71]; these soils are assumed to have frozen every year.

#### *4.3. Shrink–Swell (S-S) as a Decompaction Process*

As a decompaction process, S–S primarily depends on the presence of plastic clay minerals [22]. However, soil must undergo numerous cycles of wetting and drying for evapotranspiration and precipitation cycles to be effective. The clearcut harvesting and mostly intact forest floor on these sites had severely limited evapotranspiration. Five of the sites were imperfectly or poorly drained, and the other four had an argillic subsoil horizon prior to harvesting [14]. As a result, the excess soil moisture reduced soil aeration in the compacted and harvest-only soils on four of these sites, and all sites were moist to wet when sampled in late fall. The future effectiveness of the S–S process in these soils will be poor because other forest soils in the center of the study area have a low plasticity index [70]. A low plasticity index is indicative of soils with lesser amounts of expandable clay minerals [41,59].

## *4.4. Management Implications*

Trafficking and rebound collapse caused a 16.6% increase in bulk density at 5 cm, decreasing to 8.6% at 20 cm for the significantly compacted soil (Table 2). The increase in bulk density only decreased the macropore space [15], but these are the pore sizes primarily responsible for gas diffusion and air permeability [72] and water permeability on these

sites [15]. These boreal forest soils commonly have poor internal drainage because they are finer-textured soils and the sites have low slope classes [73]. The undisturbed soil at six of these nine sites had mottles within the top 20 cm [1]. Harvesting and trafficking caused four of these sites to exhibit morphological changes, and two of these sites changed from imperfectly drained to poorly drained soil after 3–4 years because of prolonged anaerobic soil conditions.

Mechanical site preparation is commonly used to ensure the prompt reforestation of conifers in harvested areas across the boreal forests of Alberta. Disc trenchers and small mounders mounted on wheeled skidders are most common; on wetter soils, larger mounds are built with tracked excavators using mounding buckets. Hence, the sustained poor drainage and loss of air-filled porosity [11,14,15] will likely require a switch to the more aggressive and expensive mechanical site preparation practices on moderately well-drained soils and other more poorly drained areas.

The skidders trafficking soils in this study were in the 16–17 Mg weight class; new skidders are in the 22 + Mg class. New machines have the ability to compact soil to a higher bulk density across a wider range of soil wetness. These changes will drive the need to use aggressive forms of site preparation on more areas and more frequently.

The biological consequences of a postharvest increase in bulk density are unlikely to be a long-term ecological concern on most boreal forest sites in the region. The 25% increase in postharvest bulk density was important on these sites because of the dynamic changes occurring in air-filled porosity and redox potential [11,14]. The postharvest increase in bulk density is also a short-term issue (Figure 5) but has the potential to adversely affect the prompt reforestation of some sites at a critical time. Whether the current change in drainage class is a long-term issue is probably unlikely. The data show that changes in redox potential is dynamic and sensitive to the establishment of a new vegetation cover [14]. Therefore, a long-term change in drainage class in these soils from the machines used is probably unlikely. However, changes in machine systems and climate could result in a different outcome because natural drainage class is the dominant factor affecting site productivity of *Picea glauca* (Moench) Voss [74].

#### *4.5. Measuring Bulk Density*

Bulk density is an increasingly important soil property for measuring the ecological consequences of soil compaction, assessing soil health [75], and quantifying soil carbon storage. Hence, the collection of bulk density samples using the core (cylinder) method requires a higher level of precision and quality control. Page-Dumroese et al. [28] also reported these types of issues affect the monitoring of field trials and operational practices in forestry.

The standard errors decreased with depth, which is attributed to more natural soil perturbations closer to the surface and differences in morphological soil development (Tables 1 and 4). Nevertheless, kurtosis and skewness were consistent for the three depths. The variance in standard errors widened at the base and was shewed to the right. The range of values may have only started to decrease at 20 cm. The consistency of these results suggests that this distribution should be regarded as a normal soil behavior (Figure 6) [59]. The largest standard errors mostly occurred at Site 9 at year 3 and were attributed to site conditions and a probable lapse in diligence in discarding some soil cores (Figure 6). These extreme values were obvious, but other sources of errors may not be during the collection of soil cores.

The widening of the base of the standard error histogram suggests two sources of variation are likely (Figure 6). First is the natural variability of soil bulk density. These data sugges<sup>t</sup> that the minimum is about 0.02 Mg m<sup>−</sup>3. McNabb and Boersma [53] reported standard errors between 0.023 and 0.059 Mg m<sup>−</sup><sup>3</sup> (*n* = 19–23) for four forest soils collected at a higher-quality control standard (7–12 cm depth) because the cores were used in soil mechanics tests. The lowest values were for three Andisols that were quite uniform, and the highest value was for an older Xeric Haplohumult with more spatial and profile variability. Forest soils likely have higher natural variability than soils in other ecosystems because of large root decay, windthrow, and other types of perturbations that occur close to the soil surface.

The second source of variation in standard error is the quality control used during the collection of soil cores. These two errors are not necessarily additive but are responsible for widening of the base of the distribution of the standard errors (Figure 6). Grossman and Reinsch [29] concluded without evidence that mass had less effect on bulk density than sample volume. D2937-17 [76] recommends a core volume of approximately 940 cm3, which is the size of the mold used in the standard compaction test, but 7.5 cm diameters have worked well in most agroforestry tests of cylinder sizes [77,78]. A slide-hammer assembly with a removable sleeve with a diameter of about 7.5 cm has long been regarded as a superior design for agroforestry applications [29,79,80]. However, Grossman and Reinsch [29] provided an equation to calculate whether a hammer-driven core sampler is a good design. The specification is the ratio of the cross-sectional area of steel/material in the sampler (i.e., inside wall thickness of the cylinder) relative to its outside diameter. For double-walled samplers, these measurements include the sleeve and driving cylinder. The ratio of the area in material(s) in the sample cylinder to the outside area of the cylinder should not be greater than 10% to 15%. Hvorslev [81] established these values, and they have been part of the ASTM standard for decades. For our stainless steel rings with a wall thickness of 0.16 cm, the ratio is 8.3%. For a double-walled sampler with a diameter of 7.5 cm and a total wall thickness of 0.6 cm, the area ratio is 27%, and for a smaller sampler, 5 cm double-walled sampler with a wall thickness estimated at 0.5 cm, the ratio is about 32%.

The higher ratio requires more of the energy applied to a driven core sampler to displace soil so that the cutter edge can advance deeper. As a result, thick-walled samplers are far more likely to fracture and loosen dry or compacted soils, compact wetter cohesive soils, or limit the entry of soil into the ring. Woody roots in forest soils compound this problem when a driven core sampler temporarily bounces on an uncut root. In cases where the ring does not fill with soil, Gross and Reinsch [29] and Hao et al. [82] propose measuring the height of the unfilled ring with a ruler or filling the void with beads to correct for soil volume. This is an unacceptable practice and becomes a major source of unknown error in the measurement of bulk density. Soil engineers have long measured the height of soil within a sampling cylinder and compared it with the depth that the cylinder penetrates the soil as a measure of how severely the soil core had been disturbed [81]. As a result, the North American project to evaluate soil health lists Blake and Hartge [80] as their preferred methods for measuring bulk density [75].

Blake [79,80] considered the volume changes in soil from shattering or compaction to be a genuine problem. D2937-17 [76] provides a design specification for a driven sampler with a thin-wall sampler to reduce these errors. The design can be scaled to a 7.5 cm diameter ring. The design includes a space in the driving head for the expansion of a sample if it shatters. If the driving head is removed prior to extracting the cylinder, the elevation of the soil inside the ring can be compared with the surrounding soil. This is a critical practice to ensure that only high-quality soil cores are extracted from the soil profile. A tolerance for error should be established, and samples failing to meet this standard should be abandoned while still in the ground. D1937-17 [76] also provides helpful advice on the soil and conditions when the drive-cylinder method should not be used as well as several recommendations when it may not be applicable.
