The harvest blocks for this study were selected across the mid-western to northwestern sections of New Brunswick (Figure 1). Table1 informs about block-specific attributes pertaining to species composition, harvesting type with and without brushmats laid down along tracks, soil type, elevation, slope and aspect. Information about soil association, landform, and lithology for each block can be found in TableA1 (Appendix).

Northwestern Uplands (NWU)

Blocks 1 - 5 are located within the Southern Uplands Ecoregion of New Brunswick. Mean annual air temperature is 3.6˚C. Mean monthly temperatures vary from −5.3˚C (January) to 12.5˚C (July). The area has a mean precipitation of 1140 mm with 310 mm as snow, inclusively (Department of Environment and Climate Change Canada, 2016a). The forested vegetation mainly consists of sugar maple (Acer Saccharum Marsh.), balsam fir (Abies balsamea L.), yellow birch (Betula alleghaniensis Britt.), and black spruce (Picea mariana Mill.). Blocks 1, 2, 4, 5 involved white and black spruce plantations, whereas Block 3 involved naturally grown birch and maple trees. The bedrock formations of the area consist of late Ordovician deep water marine clastics. The topography comprises gently rolling till-covered plateaus with steeply incised valleys.

Midwestern Uplands (MWU)

Blocks 8, 10 and 11 are located on the Miramichi Caledonia Highlands, which are also part of the Southern Uplands Ecoregion of New Brunswick. Mean annual air temperature is 5.2˚C. Mean monthly temperatures vary from −3.5˚C (January) to 13.9˚C (July). The area has mean precipitation of 1180 mm with 280

¹SWPL: softwood plantation, IntHW: intolerant hardwood, MW: mixedwood, TolHW: tolerant hardwood; ²BS: black spruce; WS: white spruce, BI: birch, YB: yellow birch; RM: red maple; SM: sugar maple; EH: eastern hemlock; ³CT: commercial thinning, CC: clear cut, SHW: Shelterwood, SC: select cut. ⁴Tracks brushmatted.

mm as snow (Department of Environment and Climate Change Canada, 2016a). Block 8 consisted of a naturally regenerated tolerant hardwood stand, with mostly sugar maples and yellow birch trees. The bedrock formations of the area consist of early Devonian felsic plutons, generally overlain by loamy lodgment tills and sandy glaciofluvial outwash sediments.

Lowlands (LL)

Blocks 6, 7 and 9 are located in the midwestern portion of the Continent Lowlands Ecoregion of New Brunswick. Mean annual air temperature is 5.5˚C. Mean monthly temperatures vary from −2.8˚C (January) to 13.8˚C (July). Annual precipitation amounts to 1100 mm, with 250 mm as snow (Department of Environment and Climate Change Canada, 2016a). Blocks 6, 7 and 9 supported natural mixedwood and tolerant hardwood stands, comprised of Eastern hemlock (Tsuga canadensis L. Carrière), yellow birch, sugar maples, beech (Fagus grandifolia Ehrh), and some balsam fir, Eastern white cedar (Thuja occidentalis L.), and black spruce. The bedrock formations of the area consist of early Devonian mafic volcanic and late Ordovician deep-water marine-clastics, generally overlain by ablation and boulder tills, glaciofluvial sediments and eskers.

Data layers needed for the spatial and temporal evaluation and modelling the plot-by-plot data involved:

1) producing region-wide digital elevation models (DEMs, 1 m resolution) from GeoNB’s LiDAR elevation point cloud data using LAS tools;

2) producing the cartographic depth-to-water layer (DTW), for the purpose of emulating plot-by-plot changes on soil moisture content as affected by weather conditions at the time of field sampling (see below);

3) using the province-wide 10 m resolution soil property data layers for soil bulk density (D_b), coarse fragments (CF), organic matter (OM), and texture developed by Furze (2018); this development used topographic, climatic and geological data layers to emulate soil drainage, horizon depth, depth, texture, coarse fragment content, organic matter content, and bulk density data as specified for 12,058 geo-referenced soil pedon locations;

4) obtaining daily weather records for daily rain and snow amounts and snowpack depth from weather station, within or near the NWU, MWU and LL regions (Department of Environment and Climate Change Canada, 2016a) for block-specific soil moisture emulations;

5) obtaining daily stream discharge records from hydrometric monitoring stations representative of stream water flow within or near the NWU, MWU and LL regions for hydrological model calibration (Department of Environment and Climate Change Canada, 2016b) to ensure that the soil moisture emulations were consistent with regional weather and stream discharge events;

6) acquiring forest inventory and road layers, to navigate to and access harvest blocks.

All raster and shapefile data layers were assembled with ArcMap software, using the same projection system (NAD 1983 CSRS New Brunswick Stereographic). The transect plots within blocks were georeferenced. The resulting coordinates were used to extract plot-specific data values from the elevation, DTW, D_b, CF, OM, and soil texture data layers. The DEM layer was also used to determine mean elevation, slope, and aspect for each block (Table 1), needed as block-specific soil-moisture modelling input.

Soil property evaluations were done along 696 geo-referenced transect plots inside and outside wood-forwarding tracks 1) for Blocks 1 to 9 intermittently from May 27, 2014 through November 21, 2014, and 2) for Blocks 10 and 11 in June 2013 (Table 1). The plots within blocks were established pairwise, each pair 100 m apart containing one plot within and one plot adjacent to the track on undisturbed soil, 10 m apart (Figure 2).

Soil samples were retrieved from the top 15 cm of mineral soil. Each sample was placed into labeled freezer bags for storage. Samples were dried in a forced-air oven at 75˚C for 24 hours, then crushed and passed through a 2 mm sieve to separate the fine earth from the CF. The latter was used to determine CF%. The former was used to determine 1) sand, silt and clay for each sample (Sand% + Silt% + Clay% = 100%), using the hydrometer method (Shelrick & Wang, 1993),

and 2) soil carbon% by oven-dry weight using a LECO CNS-2000 analyzer. The resulting soil carbon numbers were converted into soil organic matter via Equation (1).

OM = 1.72 × Carbon(%) (1)

Soil density samples were also collected from each plot by tapping and extracting a metal cup of known volume (85 cm³) into the mineral soil. Following frozen storage, these samples were weighed and dried in a forced-air oven at 75˚C for 24 hours. From this, the non-sieved soil D_b also containing roots and coarse fragments was determined as per Equation (2).

D bcup = Weight of dried soil ( g ) + Weight of Coarse Fragments ( g ) Volume ( cm 3 ) (2)

The fine-earth bulk density of the soil between rocks and roots was estimated using the oven-dry weight fractions of sand (Sand_W), organic matter (OM_W) and mineral soil depth (cm) as predictor variables as per Equation (3) (Balland et al., 2008).

D b = 1.23 + [ D p − 1.23 − ( 0.75 × Sand W ) ] [ 1 − exp ( − 0.0106 × Depth ) ] ( 1 + 6.83 × OM W ) (3)

A Humboldt digital cone penetrometer (cone base = 1.5 cm²; cone angle 60˚) was used to determine soil penetrability through recording CI in MPa at 15, 30, 45, and 60 cm depths within each sampling plot. Similarly, a Delta T HH2 moisture meter (TDR: a time-domain reflectometer) was used to determine the volumetric soil moisture content of the fine-earth fraction between coarse fragments for each plot at 15 cm mineral soil depth. Five CI and MC_V readings were obtained for each plot: one at the center point (location of GPS coordinates), and four arranged at each cardinal direction (north, south, east, and west), 1 m apart from center. For blocks 10 and 11, the CI measurements were taken once per subplot to the deepest depth reached with the penetrometer. Measurements were not recorded if obstructed by surface rocks, roots or logs. The subplots were then averaged to provide single numbers per plot for further analysis.

Since CI tends to be related to pore-filled soil moisture content (MC_PS, see Vega-Nieva et al., 2009), it was important to infer soil pore space (PS) and MC_PS from the soil particle and bulk densities as follows (Balland et al., 2008):

1 D p = OM D OM − 1 − OM D OM (4)

PS = 1 − D b D p (5)

MC PS = MC V PS (6)

with organic and mineral particle densities set to D_OM = 1.3 and D_Min = 2.6 g∙cm⁻³, respectively.

Soil moisture was modelled block-by-block over time using the temporal aspatial Forest Hydrology Model (ForHyM) (Arp & Yin, 1992; Yin & Arp, 1994; Jutras, 2012). This model uses 1) daily weather records for temperature, and precipitation for input (Table 2), 2) block-specific specifications for topography (elevation, slope, and aspect), 3) soil-surveyed properties (horizon depth, texture, depth, OM, CF) as per mapped soil type (Table 3), and 4) vegetation type and % canopy closure. These specifications were needed to emulate daily soil moisture, temperature, snowpack conditions and stream discharge. In this model, soil permeability at saturation was empirically related to D_p, D_b, and Sand_w by setting

log 10 K sat = − 0.98 + 7.94 × log 10 ( D p − D b ) + 1.96 × Sand W ; R² = 0.80 (7)

In ForHyM, K_sat and snowpack are further calibration-adjusted to quantify the extent of inflow (lateral flow) along layers versus downward percolation (vertical flow) into layers (Table 2, FigureA1). In soils where soil density increases with depth, downward Ksat is general less than the Equation (7) specifications. Also of note is the ForHyM formulation for water retention at field saturation (FC_W), given by

¹ Department of Environment and Climate Change Canada (2016a); ²Black Brook Watershed Research site (2014).

S = sand, SL = sandy loam, SCL = sandy clay loam, L = loam.

FC W = SP W ( 1 − exp { [ − 0.588 ( 1 − Sand W ) − 1.73 × OM ] SP W } ) ; R² = 0.96 (8)

where FC_W, SP_W, and Sand_W all refer to total weight fractions per dried and gently crushed fine-earth soil that passes through a 2 mm sieve. In combination, any ForHyM generated block-by-block soil moisture output by soil layer is, apart from daily weather records for rain, snow and air temperature and topography specifications is also a function of and soil bulk density and sand, organic matter and coarse fragment content. Specifically, Equation (8) implies that soil moisture retention at field capacity increases with increasing organic matter content and decreasing bulk density, sand and coarse fragment content (Balland et al., 2008).

In general, soil moisture increases from ridges and steep slopes to water-saturated wetlands, depressions and hyporheic zones adjacent to streams, rivers, lakes and shores. To some extent, this tendency is quantitatively related to the cartographic depth-to-water index (DTW). This index represents the least elevation rise to any point on the land away from the nearest open water locations, where DTW = 0. Conceptually, soil moisture levels should therefore decrease as DTW increases. For the changes in volumetric and pore-filled soil moisture content over time, this can be expressed as follows (Vega-Nieva et al., 2009):

MC PS,DTW = MC V,DTW PS = 100 − [ 100 − MC PS0 ] × [ 1 − exp ( − k mc × DTW ) 1 − exp ( − k mc × DTW max ) ] p mc (9)

with MC_{PS, DWT} and MC_{V, DWT} as pore-filled and volumetric soil moisture contents at any point in the landscape in relation to DTW, with MC_PS0 as pore-filled moisture content on top of the ridge, where DTW = DTW_max, and with DTW = 0 at open-water locations such as stream and lakes. The local flow-channel network was generated by applying the D8 flow-accumulation algorithm to the 1 m resolution LiDAR-generated digital elevation model for New Brunswick, as described in Figure 3. The seasonality extent of the resulting flow network was emulated by varying the minimum upslope flow-accumulation area for flow initiation (termed flow initiation area, or FIA for short) by season and weather, as illustrated in Figure 4.

In Equation (9), parameters k_mc and p_mc (0.5 and 1.5, respectively) quantify how MC_PS,DTW varies with DTW across areas of interest as MC_PS0 varies from, for example, 20% on ridges to 100% at and near water-filled flow channels. When soils become uniformly saturated, MC_PS,DTW = 100% everywhere. At other times MC_PS0 ≤ MC_PS,DTW < 1.

Since soil moisture at any point in the landscape are also affected by block-by-block and plot-by-plot variations in soil properties and wood-forwarding tracks, MC_PS also needed to be evaluated as function forest cover and soil properties inside and outside wood-forwarding tracks. This was done by setting:

MC ps = f ( DTW,elevation,forestcover,soil properties,block,track,brushmat ) (10)

with Forest Cover (SW, HW, and MW), Block, Track and Brushmat coded 1 when applicable and 0 when not. The elevation variable accounted for additional topographic variations as they existed block-to-block across the NWU, MWU and LL regions. The soil properties refer to plot-by plot variations in soil texture (sand, silt, clay content), D_b, CF, OM, and layer depth. The block-by-block and plot-by-plot CI variations were evaluated similarly by setting:

CI = f ( DTW,elevation,forestcover,soil properties,block,track,brushmat ) (11)

The block- and plot-descriptive, field, laboratory, ForHyM-generated and raster-extracted data were compiled into a single datasheet, were summarized, and were subjected to correlation, factor, multivariate regression analyses in R (R Core Team, 2015), using MC_V, MC_PS and CI as dependent variables. The resulting regression models for Equations (9)-(11) were assessed in terms of scatterplots for the actual versus modelled values, the best-fitted regression coefficients for each independent variable of non-zero significance values (p-values < 0.01), and R² and RMSE goodness-of-fit indicators. In addition, the conformance levels between actual and model projected 0 - 10, 10 - 20, 20 - 30, … % MC_PS classes and 0 - 0.5, 0.5 - 1, 1 - 1.5,… MPa CI classes were evaluated in terms of confusion matrices, non-randomness, and cumulative conformance probabilities to determine the uncertainty range of best-fitted model projections. Subsequently, the MC_V, MC_PS and CI regression evaluations were repeated using variables amenable for area-wide projection purposes. These variables referred to 1) DTW, elevation, digitally-generated province-wide soil property rasters for Sand, CF, OM and D_b (Figure 5; Furze, 2018), 2) block-, and season- or weather-specific assignments for FIA, 3) soil depth, and 4) forest cover type for non-tracked soil conditions. The process of doing so by soil depth and season (month) is illustrated in Figure 6.

The plot-based determinations for sand, silt, and clay, OM, CF, MC_PS, and CI are summarized per block in Figure 7 by box plots and in terms of number of plots, and minimum, maximum, average and standard deviation values in TableA2 (Appendix). These plots reveal that the MC_PS, CI and D_b were not always higher inside than outside the wood-forwarding tracks, as one would expect (Allen, 1997; Han et al., 2009; Labelle & Jaeger, 2011; Cambi et al., 2015; Solgi et al., 2018). Possible reasons for the lack of systematic MC_PS, D_b, and CI increases refer to 1) the wide range of operational multitrack observations across ground conditions comprised of rocks, stumps, and variable forest litter depths including brushmats, and 2) differences in MC_PS, D_b, and CI surveying methodologies, e.g., using D_b and MC sensors that required no soil displacement (Labelle & Jaeger, 2011) and hydraulic cone penetrators (Cambi et al., 2015; Solgi et al., 2018). The results reported below were obtained through manual soil extraction and probe insertions.

In terms of tracks with and without brushmats, measured values for D_b, CI, and MC_PS for the top mineral soil 15 cm were somewhat lower and less variable

under matted than non-matted tracks (Figure 8). Systematic increases for D_b and CI from the matted and non-matted tracks as reported by Han et al. (2009) and Labelle et al. (2015) were therefore not found, likely due to strongly varying soil, root and rock conditions along the tracks.

In terms of the higher trending MC levels on matted versus non-matted plots, Roberts et al. (2005) and Moroni et al. (2009) reported lower cumulative soil moisture losses under matted than non-matted tracks likely through mulch-related shading, insulation, and lowered evaporation benefits.

The overall trends between MC_PS and CI versus CF, sand, OM and D_b are presented in Figure 9. As shown, increasing CF generally lowered MC_PS presumably due to 1) increasing porosities between the fragments, 2) subsequent increases in soil moisture loss due to increased soil permeability, and 3) enhanced evaporation due to fact that CF conduct heat faster than soil (Poesen & Lavee, 1994; Chow et al., 2007).

The increasing MC_PS trend with increasing D_b relates to decreasing pore space. Similarly, decreasing MC_PS with increasing OM% relates to increased pore space due to OM-facilitated soil granulation (Han et al., 2006). There is also the possibility of decreasing MC_PS due to increasing hydrophobicity as OM-containing soils dry out (Vogelmann et al., 2013). In contrast, the increasing MC_PS trend with increasing Sand% may be due to preferential pore space saturation in sandier topsoil portions.

In terms of CI, increasing levels of CF contribute to increasing soil resistance to penetration because of skeletal soil stabilization (Manuwa, 2012), and the need to push coarse fragments to the side to gain greater penetration depths. Increasing Sand% generally decreases soil strength,, due to lower particle-to-particle cohesion. The opposite occurs with increasing Clay% (Raven et al., 1999). The trend between increasing CI and increasing soil depth (Figure 10) is due to depth-related increases in D_b (Equation (3); Carter et al., 2007).

Analyzing the trends among the MC_PS and CI affecting variables more closely produced the correlation matrix and its factor analysis results in Table 4. The bolded Factor 1 - 3 loadings indicate that:

1) Factor 1 is positively associated with increasing CI, CF, D_b, elevation and tracks, but negatively associated with increasing soil organic matter content.

2) Factor 2 is positively associated with higher elevations where tolerant

hardwood forests dominate, where brushmats were rarely used, and where soils are somewhat sandier.

3) Factor 3 is positively associated with increasing MC_PS but negatively associated with increasing CF, Sand, and decreasing DTW, as to be expected. Also, soil pores tend to be drier at higher elevations but fill up more easily and remain wet or moist longer inside than outside the tracks.

The ForHyM-generated results for daily soil moisture, stream discharge, snowpack depth and depth of soil frost are illustrated in Figure11, which also displays daily weather records (rain, snow snowpack depth, air temperature) for Blocks 1 and 9 from 2011 to end of 2014. The daily MC_PS0 output so produced block-by-block (Table 5) was obtained through calibrating the ForHyM output with reported snowpack depth and cumulative stream discharge values for each of the three regions (FigureA1). These calibrations involved adjusting the ForHyM parameters for snowpack depth and lateral versus vertical layer-by-layer soil permeability, as detailed in Table 5.

Fitting the actual plot-specific MC_PS determinations for the top 15 cm of mineral soil versus the corresponding log₁₀DTW values with FIA = 1 ha across all the plots by way of Equation (9) produced the following equation once FIA and MC_PS0 was specified for each block:

MC PS,DTW = 100 − [ 100 − MC PS0 ] × [ 1 − exp ( − 0.5 DTW ) 1 − exp ( − 0.5 DTW max ) ] 1.5 ; R 2 = 0.29 (12)

This result improved considerably by using log₁₀DTW_1ha in combination with MC_PS,DTW:

MC PS = 48.16 − 5.80 log 10 DTW + 0.19 MC PS,DTW ;   R 2 = 0.38 (13)

In detail, the inclusion of log₁₀DTW spread the MC_PS prediction range from about 30% to 75% [Equation (12)] to about 20% to 90% [Equation (13)]. Even further improvements were obtained by including the plot determinations for D_b, Elevation, HW and Track as independent variables. This led to the best-fitted MC_PS regression results listed in Table 6, and to Equation (14):

MC PS = 32.18 − 1.49 log 10 DTW + 0.27 MC PS,DTW + 42.56 D b                       − 0.10 Elevation + 13.09 HW + 3.93 Track ;   R 2 = 0.61 (14)

The corresponding best-fitted MC_V results, also listed in Table 6, led to:

MC V = 35.25 − 1.33 log 10 DTW + 0.19 MC PS,DTW + 13.37 D b                     − 0.07 Elevation + 9.38 HW + 2.43 Track ;   R 2 = 0.51 (15)

Units: MCPS, DTW and OMDSM in %; DTW and Elevation in m; Db in g/cm³; HW Blocks and Tracks: 1 when present, and 0 otherwise.

Based on the t-value entries in Table 6, one determines that the significance contributions of all the MC_PS and MC_V predictor variables follow these sequences:

for MC_PS: D_b ≈ MC_PS,DWT ≈ Elevation ≈ HW > log₁₀DTW > Tracks > Sand > CF;

for MC_V: MC_PS,DWT ≈ Elevation ≈ HW > log₁₀DTW > Sand > Tracks > D_b ≈ CF.

As to be noted, there is a high to low D_b regression coefficient and significance change from MC_PS [Equation (14)] to MC_V [Equation (15)] due to the greater dependence of MC_PS on D_b via Equation (5) and Equation (6) while the significance levels for the predictor variables remain about the same. Also note that the best-fitted R² and RMSE values drop from MC_PS to M_V due to the narrower MC_V data range.

The + and − signs for the regression coefficients in Equation (14) and Equation (15) follow the expected trends for MC, namely:

1) MC decreases with increasing DTW but increases as MC_PS0 (and therefore MC_PS,DWT) increase, as to be expected.

2) MC decreases towards higher and generally steeper elevations.

3) MC is higher inside than outside soil tracks.

4) Soils under hardwood cover tend have higher MC values than elsewhere, in part due to deeper rooting and higher top-soil OM accumulation.

5) The contributions of Sand and CF (both negative) to MC were also found to be significant, but only marginally so. Excluding these from the analysis changed R² for MC_V from 0.51 to 0.50, and for MC_PS from 0.63 to 0.60 (details not shown).

The best-fitted MC_PS and MC_V equations―obtained from substituting the plot-generated values for MC_PS,DWT with the block-assessed FIA values and with plot-level Equation (3) estimates for OM generated from the digital soil layers for OM% (labelled OM_DSM)―are as follows (Table 6):

MC PS = 70.88 + 0.26 MC PS,DTW − 1.84 log 10 DTW + 0.98 OM DSM                         − 0.10 Elevation + 9.96 HW + 4.32 Track ;   R 2 = 0.47 (16)

MC V = 45.22 + 0.19 MC PS,DTW − 1.75 log 10 DTW + 0.13 OM DSM                   − 0.07 Elevation + 8.48 HW + 2.53 Track ;   R 2 = 0.48 (17)

The corresponding actual versus best-fitted scatterplots associated with MC_PS via Equations (12)-(14) and Equation (16) are shown in Figure 12.

Using Equation (14) produced the block-by- block presentation in Figure 13, which also shows the field-determined MC_PS plot values for the top 15 cm of soil outside and inside the tracks. Visually, there is a general agreement between the off-track field determinations and the corresponding projections, with off-track MC_PS generally higher than inside track.

Grouping the off- and in-track projections and corresponding field determinations into 10% MC_PS classes produced the MC_PS confusion matrix and associated MC_PS cumulative conformance plots in Figure 14. In summary, the ±5% MC_PS conformance level so found amounts to 34% across the 20 < MC_PS < 100% range, but increases to 81% and 96% as the conformance uncertainty increases to ΔMC_PS ≤ ±15 and ±25%, respectively. The possibility of randomly achieving

these conformance levels have kappa values of 0.26, 0.73 and 0.92, respectively. The kappa value for random chance agreements equals zero, by definition.

The best-fitted CI results using plot-determined and plot-projected data are listed in Table 7 and led to the following equations:

CI ( plot-determined ) = 1.60 − 0.01 MC PS + 0.03 Depth                                                               + 0.76 Track + 0.40 SW ;   R 2 = 0.41 (18)

CI ( plot-projected ) = 1.60 + 0.03 log 10 DTW − 0.02 MC PS0 + 0.02 Depth                                                         + 0.64 Track + 0.32 SW ;   R 2 = 0.39 (19)

where 0.03 log₁₀DTW − 0.02 MC_PS0 in Equation (19) replaced −0.01 MC_PS in Equation (18). These results affirm that the plot-determined CI values, as to be expected, increase with decreasing soil depth, are higher inside that than outside

the tracks, are higher under softwood forest cover, decrease with increasing soil moisture content, and increase with increasing soil density. The associated significance levels vary as follows:

for Equation (18): Track > Depth > MC_PS > SW > D_b

for Equation (19): Track > Depth > SW ≈ DTW_1ha > MC_PS0

Other variables such as Sand, Clay, CF and OM would also affect CI, but these variables were not found to make additionally significant CI contributions to the plot-determined or projected values. Quantitatively, Equation (18) indicates that a change in MC_PS from 0% to 100% would lead to a change in CI by 3.1 MPa (all other conditions remaining the same). In comparison, CI would on average increase by 2.6 MPa from 0 to 1m soil depth. In comparison, CI values are about 0.7 MPa higher inside than outside tracks.

The scatterplots associated with Equation (18) and Equation (19) are presented in Figure 15, with the former more heteroscedastic than the latter. Log-transforming CI the best-fitted results as listed in Table 7 produced an even scatterplot in Figure 15, based on the following equation:

log 10 CI = 0.152 − 0.002 MC PS + 0.006 Depth                             + 0.189 Track + 0.091 SW ;   R 2 = 0.41 (20)

The overlays of the field-determined on the Equation (18) projected CI values for the top 15 cm of soil are shown by Block in Figure 16, which also shows the corresponding values at 30, 45, and 60 cm depth for Block 3.

Grouping the field-determined and projected CI values into 0.25 MPa classes produced the CI confusion matrix and associated cumulative conformance plots in Figure 17. As shown, the CI confusion matrix has an actual with projected

ΔCI ≤ 0.25 MPa class conformance of 32% across the 0.5 < CI < 4.5 MPa range. This conformance increases to 78% and 95% as the uncertainty range for CI increases from ±0.5 and ±0.75 MPa, respectively. Achieving these agreements has kappa values equal to 0.23, 0.67 and 0.88, respectively. In comparison with the MC_PS kappa values, this suggests that the CI projections are slightly more random than the MC_PS projections.

The above results show that the combination of emulating layered soil properties and daily soil moisture simulations produces non-random soil moisture and cone penetrability projections across harvest blocks of varying topography, forest cover, substrates and weather conditions. To this effect, the extent of pore-filled moisture content can be predicted with a tolerance of ±15%, 8 times out of 10 within the 20% to 100% soil moisture range. Similarly, the cone penetrability index can be predicted with a tolerance of ±0.5 MPa, 8 times out of 10 within the 0.2 to 4.5 MPa range.

In summary, generalizing similar results for the purpose of soil-trafficability forecasting block-by-block requires:

1) a 10 m resolution DEM layer with a tolerance of at least ≤ 2 m, 8 times of out 10 (Furze et al., 2017);

2) a digital soil mapping process that emulates the required soil property layers at a general conformance level of 80% (Furze, 2018);

3) an assessment of the season and weather dependent minimum upslope open-channel flow initiation area (FIA) as this could vary from ≤0.25 to ≥8 ha (Figure 4);

4) a forest hydrology model that can be used to estimate weather- and season-affected soil moisture content (MC_PS0) for ridge-top soils, to initiate, e.g., the Equation (15) and Equation (18) calculations (Jones & Arp, 2017), based on block-by-block elevation, slope, aspect, texture, D_b, OM, CF, vegetation type and % canopy closure.

The MC and CI results so produce are projected in Figure 18 for hardwood

and softwood areas somewhat larger than Block 3 and Block 4. This was done for weather conditions that varied from essentially frozen in February to wet in May, dry in July, and moist in October. The blue shading in these plots suggests where rutting would occur on account of elevated MC and lowered CI values once wood-forwarding operations cross the areas so marked. For the most part, areas prone delineations as these vary by weather and season. Least rutting would have occurred in February (ground mostly frozen) and in July and August (ground mostly dry), but would have been most extensive in May and late fall (ground mostly moist to wet).

While the above study provides a useful framework for delineating and evaluating soil trafficability restrictions due to changes in soil moisture and soil penetrability, there is a need to validate the generality of this framework by:

1) testing across all dominant and minor soil types within the same region;

2) repeating the same across other regions;

3) conducting sequential MC and CI determinations as ground conditions transit from wet to dry and from frozen to non-frozen;

4) advancing the digital soil property mapping process from coarse textured DEMs (Furze, 2018) to 1 m LiDAR-generated terrain elevation resolution. Doing so will likely increase the significance by which projected sand, clay, organic matter, coarse fragments and bulk density gain significance as MC and CI predictors.

Repeated topsoil moisture and density testing should ideally be done without probe insertions, i.e. via nuclear moisture-density gauge (Corns, 1988; Brown et al., 1998; Labelle & Jaeger, 2011). CI testing would further benefit from hydraulic rather than manual testing (Hooks & Jansen, 1984; Herrick & Jones, 2002). Doing so would assist in reducing manually generated errors, inconsistencies, and biases, but would increase sampling expense.

There are other DEM-utilizing soil MC projection techniques. Most notably among these is the topographic wetness index (TWI; Beven & Kirkby, 1979; Sørensen et al., 2006). This index relates soil wetness to local upslope flow-accumulation areas and slopes, i.e., TWI = ln (flow accumulation area/slope). Comparative studies revealed that TWI-inferred MC depends strongly on TWI-cell averaging (Murphy et al., 2009a, 2011; Ågren et al., 2014), while DTW-inferred MC can directly be related to field-based DTW measurements (Dobbie & Smith, 2006; Murphy et al., 2009b).

Optimal TWI-based MC interpretations are generated by filling noise-generated DEM pits, followed by changing the focal search radius for cell-centered mean elevation differences, slope and TWI (Southee et al., 2012). The resulting MC correlations with TWI had R² values ranging from 0.06 to 0.36, with best-results obtained for the 5 to 10 m cell-size range. Attempts focused on analyzing soil moisture conditions from spectral surface images tend to produce mixed results, especially for soils covered by vegetation and vegetation litter (Njoku & Entekhabi, 1996; Das & Paul, 2015). Best results are generally observed for bare to open mineral soil surfaces (Jackson, 1993; Engman & Chauhan, 1995; Wang & Qu, 2009; Das & Paul, 2015).

Further digital soil property mapping improvements can likely be obtained using LiDAR-generated DEMs at 1 m resolution. Doing so would likely increase the significance by which projected sand, clay, organic matter, coarse fragments and bulk density gain significance as MC and CI predictors.

With reasonable soil- and weather-informed MC and CI projections, it is feasible to generate machine- and load-specific rut depth maps as affected by machine type, loads, tire dimensions, and number of wood-forwarding passes, as demonstrated by Vega-Nieva et al. (2009) and Jones et al. (2018). In turn, such projections can be used to evaluate existing traffic-induced soil disturbance impacts as demonstrated by Campbell et al. (2013).