Runoff generation and soil erosion processes after clear cutting

Authors


Abstract

[1] Timber harvesting by clear cutting is known to impose environmental impacts, including severe disturbance of the soil hydraulic properties which intensify the frequency and magnitude of surface runoff and soil erosion. However, it remains unanswered if harvest areas act as sources or sinks for runoff and soil erosion and whether such behavior operates in a steady state or evolves through time. For this purpose, 92 small-scale rainfall simulations of different intensities were carried out under pine plantation conditions and on two clear-cut harvest areas of different age. Nonparametrical Random Forest statistical models were set up to quantify the impact of environmental variables on the hydrological and erosion response. Regardless of the applied rainfall intensity, runoff always initiated first and yielded most under plantation cover. Counter to expectations, infiltration rates increased after logging activities. Once a threshold rainfall intensity of 20 mm/h was exceeded, the younger harvest area started to act as a source for both runoff and erosion after connectivity was established, whereas it remained a sink under lower applied rainfall intensities. The results suggest that the impact of microtopography on surface runoff connectivity and water-repellent properties of the topsoil act as first-order controls for the hydrological and erosion processes in such environments. Fast rainfall-runoff response, sediment-discharge-hystereses, and enhanced postlogging groundwater recharge at catchment scale support our interpretation. At the end, we show the need to account for nonstationary hydrological and erosional behavior of harvest areas, a fact previously unappreciated in predictive models.

1 Introduction

[2] The practice of clear cutting is known to cause severe environmental impacts [e.g., Ziegler et al., 2006]. In most cases, clear-cut harvesting involves the use of heavy timber machinery which causes soil compaction and reduction of both macroporosity and infiltration capacity [e.g., Malmer and Grip, 1990]. As a consequence, peak flows may increase in both frequency and magnitude, promoting sediment transport [e.g., Birkinshaw et al., 2011; Carr and Loague, 2012; Croke et al., 2001; Iroumé, et al., 2005; Iroumé et al. 2006; Jones and Grant 1996; Malmer and Grip, 1990]. Postlogging acceleration of landsliding rates may additionally intensify sediment transport after logging activities [Montgomery et al., 2000], providing abundant sediment to be purged preferentially during low-frequency but high-magnitude rainfall-runoff events [e.g., Coppus and Imeson, 2002]. These landslides are often associated with the drainage of timber roads [Montgomery et al., 2000] that are regarded as main sources for runoff and sediment in such environments [e.g., Croke et al., 2001; Lane and Sheridan, 2002; Motha et al., 2003]. In contrast, the role of the harvest area itself remains less clear, although its absolute contribution of runoff and sediment fluxes may often exceed that of forest roads as a consequence of its greater area. Owing to relatively low soil compaction compared with timber roads, skidder tracks or log landings, harvest areas are mainly considered as sinks for runoff and sediment transport due to restricted connectivity and limited sediment supply along the slopes [Croke et al., 1999b; Wallbrink and Croke, 2002]. However, Croke et al. [1999a] also demonstrated that runoff generation and sediment transport do occur on harvest areas, owing to uneven degrees of topsoil disturbance which allows the connectivity of erosive surface runoff along tracks of lower hydraulic conductivity. Similarly, Brooks et al. [1994] showed intensified runoff and sediment transport on compacted harvest areas of gentle slopes but increased infiltrability along the steeper slopes associated with enhanced surface retention capacity. Moreover, sediment supply on logged slopes is not necessarily limited. Logging activities are reported to increase soil erodibility by intermixing more dispersive subsoil into topsoil which augments the sediment storage along recently logged slopes with erodible sediment [e.g., Burt et al., 1983; Croke et al., 2001].

[3] Hence, the role of harvest areas is ambiguous. This comprehensive study explores the hydrological and erosional behavior of such harvest areas and focuses in particular on the following two key objectives:

  1. [4] Under which environmental conditions do harvest areas act as sources and under which conditions do they act as a sinks for surface runoff and fine sediment transport?

  2. [5] Is the hydrological and erosional response of such areas stable, or do there exist tipping points where their hydrological and erosional performance dynamically switches along intrinsic threshold values?

[6] To this end, the soil hydrological response to (simulated) rainfall was studied. For this purpose, rainfall simulation experiments were carried out in a mature pine plantation and on two clear-cut harvest areas of different age with various rainfall intensities that conformed with the local rainfall regime. Nonparametric statistical Random Forest models were used to gain insight in the relation between environmental conditions and the soil hydrological and erosive responses. The plot-scale observations are then compared and discussed with observations made at the catchment outlets.

2 Study Area

[7] The study area is located on the eastern slopes of the metamorphic Coastal Range of South Central Chile, about 500 km south of Santiago in the Biobío Region, close to the city of Nacimiento (Figure 1a). All creeks within the study area are part of the Bío-Bío river basin, which drains more than 24,000 ha of southern Central Chile. The dominant land use within the Bío-Bío region is forestry, which has promoted the development of this region into the national center of timber and pulp production [Patterson and Hoalst-Pullen, 2011]. Today, the area shows one of the fastest-growing rates in timber and pulp production worldwide [FAO, 2010]. The clear-cutting technique remains the dominant practice, and single clear-cutting episodes may span areas of several hundreds of hectares. Here, a network of 11 experimental catchments (ranging between 250 and 480 m above sea level (asl)) was established in order to analyze hydrological and erosional processes of different forest management practices [Huber et al., 2010]. The network involves clear-cut catchments which were harvested by the same techniques but differ in age and season of their latest clear cutting. At the same time, the small headwater catchments are homogeneous in terms of size, topography, geology, and soil type and provide together with their simple geometry and vicinity exceptional settings for intercatchment comparison (see Mohr et al., [2012] for morphometric features and details). As a result, the network offers a promising opportunity to study the (apparent) contradictory role of harvest areas as previously outlined. To this end, this study is restricted to three catchments (Table 1), of which one persisted as a mature Pinus radiata D. Don. plantation during the study period (Figure 1b; San Antonio #1). This catchment was last logged and reforested in 1983 following a previous rotation also on P. radiata. Both the other catchments, hereafter named Pichún and San Antonio #3 after their administrative affiliation, were reforested in 1987 and 1983, respectively, following a previous rotation of P. radiata. All other catchments were excluded from this study because they were covered with secondary native forest, Eucalyptus globulus Labill plantations, or showed only little difference to the P. radiata stands of the control catchment. The Radiata pine plantations of Pichún and San Antonio #3 were clear cut, and rubber-tired skidders were used to drag logs uphill to landings, while cable logging was performed only in steep terrain.

Figure 1.

(a) Location of the study area is represented by the black triangle. The inset shows the Bío-Bío drainage system as a dotted line. The representation of the elevation (m asl) is derived from GTOPO30 data (http://demex.cr.usgs.gov/DEMEX/). (b) Sites of the rainfall simulation experiments within the experimental catchments are shown. The positions of the rain gauges and the closest meteorological station are represented. Contour lines correspond to 20 m, derived from a LiDAR digital terrain model (DTM). The numbers correspond to (1) Pinus radiata control and (3) former P. radiata plantation, clear cutting during winter 2009 in San Antonio.

Table 1. Main Features of the Studied CatchmentsThumbnail image of
  • a

    Referring to pre- and post-logging experiments, respectively, conducted across the same catchment.

  • b

    Year of latest preceding clear-cutting corresponds to the year of plantation.

  • c

    Planted with Pinus radiata.

  • d

    Courtesy of Andreas Bauer.

  • [8] Pichún was logged during the dry summer season in 2006, and then the area was artificially laid fallow through the application of herbicides. San Antonio #3 was logged during the rainy season between 14 July and 10 August 2009. Timber harvesting during the rainy season is economically beneficial because the immediate reforestation can be performed under saturated soil conditions, which in turn assures maximum soil water availability for the seedlings when the growing season starts.

    [9] The area is characterized by a subtropical Mediterranean climate showing a pronounced seasonality. Annual average precipitation is 1150 mm, concentrated between April and September, a period which contributes 95% of the total annual yield [Huber et al., 2010]. Annual average temperature is 13°C, and monthly mean temperatures range between 7°C in July and 19°C in January (Figure 2a). The summer temperature can exceed 40°C, enhancing high evapotranspiration rates and promoting very low topsoil water content during summer [Huber et al., 2010]. Rainfall events are cyclonically or orographically driven and may last for several days. However, rainfall intensity is low, and median rainfall intensity equals ~8.8 mm/h (mean 10 mm/h) for the Meñir station during the period from January 2000 to December 2008. Twenty-five percent of the rainfall events fall below 4.8 mm/h, while 75% do not exceed 13.8 mm/h and 95% are less than 23.2 mm/h. The highest intensity recorded reached 27.6 mm/h (Figure 2b). Although the Meñir meteorological station is farther away from the study area than the Pichún station, it provides longer continuous rainfall records which were preferred to assess statistical rainfall properties owing to their similar local conditions in altitude and aspect.

    Figure 2.

    (a) Mean monthly temperature and rainfall for Meñir meteorological station (37°58′, 72°77′, 647 m asl) during the period from January 2000 to December 2008. Bars represent monthly rainfall (mm) and the line monthly average temperature (°C). (b) Cumulative distribution function for maximum 30 min rainfall intensity at Meñir station during the period from January 2000 to December 2008.

    [10] The dominant soil type is a clayey to loamy Luvisol, and its structure is variable on a small scale due to embedded fragments of bedrock, a complex distribution of recent and former root systems, and direct disturbances by timber harvest. The soil is underlain by a deep saprolite layer [Mohr et al., 2012]. Recent channel cuts and truncated soil profiles exposing low-conductive B-horizons show evidence of active soil erosion and landsliding processes triggered by forest clearing [Montgomery et al., 2000].

    3 Methods

    [11] Rainfall simulations have been successfully applied to define major water and sediment fluxes, their feedbacks, as well as their controlling factors triggering source or sink behavior [e.g., Cerdà, 1997; Cerdà, 1998; Cerdà and Doerr, 2005; Croke et al., 1999a; Croke et al., 1999b; Croke et al., 2001; Croke et al., 2006; Imeson et al., 1992; Michaelides et al., 2009; Wainwright et al., 2000]. They permit the simultaneous observations of runoff initiation and erosion processes, complementing the quantitative data with qualitative observations and allowing the comparison of the hydrologic and erosional responses between different land use managements under controlled conditions [e.g., Cerdà, 1997; Michaelides et al., 2009].

    [12] In order to meet the local topographic conditions and to provide low and long-lasting rainfall intensities, we modified the drip-plate type rainfall simulator based on Bowyer-Bower and Burt [1989]: Telescopic extensions (Figure 3a) allow adjustment to slopes up to 45° without inclining the drop-former box to ensure the homogeneous spatial distribution of the simulated rainfall. The rainfall intensity is controlled by interchangeable glass tubes (Figure 3b) of different diameters (here: 2 and 5 mm) based on the Bernoulli principle and confirmed by measuring the water volume leaving calibrated water supply containers (Figure 3c). An acrylic box (Figure 3d) features approximately 600 equally spaced drop-formers that distribute the water over the 0.5 m2 plot. In order to generate more realistic drop size distribution, a 5 mm-spacing-mesh randomly resizes the drops (Figure 3e).

    Figure 3.

    Sketch of rainfall simulator, courtesy of Odette Morales (a: telescopic extensions; b: control panel including interchangeable glass tubes; c: water supply containers; d: drop-former box; e: drop size randomizer).

    [13] The streamflow discharge was monitored at a temporal resolution of 3 min by V-notched Thompson-type weirs equipped with custom-built water stage loggers [Huber et al., 2010; Mohr et al., 2012]. The loggers provide an accuracy of 2 mm. Rainfall was registered at 0.2 mm accuracy by two Hobo tipping bucket rainfall gauges and a meteorological station (Davies Instruments, Advantage Pro Series) located at suitable and accessible sites (Figure 1b).

    3.1 Experimental Strategy and Sampling Design

    [14] During the dry summer season of 2009 (January–February) and 2010 (March 2010), a set of 92 rainfall simulations at 10, 20, and 40 mm/h intensity were performed under a mature 26 year-old P. radiata stand (hereafter S.A. pre), an area which was logged by clear cutting in 2009 (hereafter S.A. post), and an area logged by clear cutting during the dry summer period in 2006 which had been laid fallow (hereafter Pichún). Hence, the time between the last preceding clear cutting and the time when the experiments were conducted was 26, 0.5, and 3 years for S.A. pre, S.A. post, and Pichún, respectively (Table 1). The S.A. pre and S.A. post experiments were all conducted within the San Antonio #3 catchment.

    [15] The applied 10 and 20 mm/h intensities represent the local natural rainfall intensities (Figure 2b). Additionally, a high intensity was applied to accentuate differences in hydrological and erosion responses [e.g., Michaelides et al., 2009].

    [16] The simulations were conducted along upper, middle, and lower slopes, and the plots were selected by the representativeness of the surface cover and accessibility. The timber harvest of San Antonio disturbed the surface to such an extent that this approach was replaced in 2010 by a more suitable two-stage random sampling design [de Gruijter et al., 2006]: In a first step, the study area was divided into equally sized sub-areas (“primary units”), three each for each slope segment. In a next step, one primary unit was randomly selected out of each slope segment, and ten sites were randomly selected, which were assigned in equal parts to the rainfall intensities randomly. This sampling design permits a high sample size, spread across the whole study area efficiently [de Gruijter et al., 2006; Hassler et al., 2011].

    [17] The infiltration rate was calculated by subtracting runoff rate from rainfall intensity after the runoff rate had been stabilized but had lasted for at least 2 h. In order to account for the local slope, runoff volume was normalized to 1 m2 of surface.

    [18] Runoff rates were determined by volume of runoff that was collected in sterile polyethylene bottles per time interval at a runoff gutter downslope. Time to runoff was estimated during the experiments and recalculated to rainfall (mm) necessary to initiate runoff. Runoff yields were determined by accumulating the runoff over the 2 h from initiation of runoff. The collection of the first sample started at the moment of runoff initiation and ceased once sufficient sample was available for sediment analysis [Michaelides et al., 2009]. The remaining samples were taken at intervals of 3 up to 30 min depending on the runoff rate and became longer during the latter part of the simulations when runoff rates became more stable. For the S.A. post series, a tipping bucket rainfall gauge (0.2 mL accuracy) was used to determine runoff rates to improve temporal resolution. A receptacle beneath was used as a sediment collecting device. Sediment concentrations were determined gravimetrically with an accuracy of 0.5 mg after filtering the runoff samples through preweighed glass fiber filters (Advantec Glass Fiber Filter GF 75 47 mm) and drying at 105°C for 48 h.

    [19] Sediment yields were calculated by summing the products of runoff volume and sediment concentration for each sampled time interval during 120 min of surface runoff (equation ((1))).

    display math(1)

    in which SSY corresponds to the total sediment yield (g/m2) after 120 min, R to runoff volume (L/m2) measured during the sampling interval, and SSC to the sediment concentration (g/L).

    [20] Brilliant blue dye tracer at a concentration of approximately 4 g/L was applied to a randomly selected subset of simulations to estimate infiltration patterns and depth prior to and after logging activities [e.g., Blume et al., 2008; Weiler and Naef, 2003]. The depth of the wetting front was measured manually at 5 cm increments along profiles perpendicular to the plot. Standard deviation was used as a proxy for the uniformity of the wetting front.

    3.2 Soil Conditions

    [21] Prior to the experiments, each plot was characterized by slope, position, and aspect according to Jahn et al. [2006]. The percentage of the vegetation, bare soil, stone, and litter cover was estimated by a simple grid method as proposed by Cammeraat [1993]. Organic horizons (lhf) were described after Green et al. [1993]. Initial topsoil moisture (0–5 cm depth) was assessed gravimetrically using soil cores which were extracted adjacent to the upper and lower boundary of the plot or by ThetaProbe soil moisture sensor (ML2x, Delta-T Devices) at an accuracy of ±1%, which had been validated by core samples. Bulk density was determined from the same soil core samples [Cammeraat, 1993]. Soil texture was determined by Rubilar (unpublished soil data, 2008) in 14 pits spread across the study area. Hydraulic conductivity was estimated by double-ring infiltrometer measurements at approximately 30 cm depth following Wu and Pan [1997].

    3.3 Data Analysis

    [22] Due to non-normality of the data, nonparametric tests for statistical interference were applied. We assessed changes in infiltration rate, the depth of rain applied to initiate runoff, runoff yields, peak runoff, sediment yields, erosion rates, and differences in initial conditions due to forest management using a Wilcox rank sum test at a significance level of alpha = 5% [e.g., Hassler et al., 2011].

    [23] We also used a statistical method called Random Forest model, which belongs to a family of methods using decision trees. Decision-tree-based methods are flow chart like structures and allow quantification of relevant predictor variables in high dimensional settings involving interaction [Strobl et al., 2008]. Such decision trees are applied to differentiate data into various groups by separating them along finite predictor variables [e.g., Vorpahl et al., 2012]. Random Forest models [Breiman, 2001], hereafter named RFs, consist of an ensemble of such decision trees. RFs were set up to quantify the impact of environmental variables on runoff generation and sediment transport. RFs are a robust nonparametric statistical method capable of handling large nonlinear, noisy, fragmented, or intercorrelated data for regression [Law and Wiener, 2002; Strobl et al., 2008]. They have been applied for a variety of hydro-geomorphological studies including runoff and sediment transport prediction [e.g., Francke et al., 2008; Zimmermann et al., 2012] or quantifying the impact of driving factors for landsliding [Vorpahl, et al. 2012].

    [24] RFs combine bootstrap aggregating (“bagging”) with random variable selection [Breiman, 2001]. RFs use a randomly selected subset of data (called bootstrapped data) to grow decision trees. The predictions of the tree grown on that data are then tested against data not included in the bootstrapped data (called out-of-bag data). As a result, the out-of-bag data provide an unbiased model performance assessment. In practice, each tree is grown recursively by partitioning the data. At each split, the data are divided into two groups according to a simple rule based on one of a random subset of predictor variables aiming in minimization of overall variance. Thus, the main parameters controlling RF models are the number of trees, the tree complexity in terms of maximum number of terminal nodes or maximum tree depth, the number of randomly selected predictor variables at each split, and the size of the out-of-bag fraction for performance assessment. The overall RFs prediction is then approximated by averaging all single trees' predictions [Breiman, 2001]. Although RFs consist of many noisy but approximately unbiased models and each classification tree itself is relatively inaccurate, they produce the right prediction when averaged [e.g., Liaw and Wiener, 2002; Strobl et al., 2008]. Even though RFs can handle strongly intercorrelated data, the quantification of the real impact of predictor variables is uncertain. In order to overcome this limitation, the variables were conditioned. Conditioning the variables is a way to avoid spurious correlations by revealing the true impact variables while excluding the covariate ones [Strobl et al., 2008].

    3.3.1 Random Forest Model Setup

    [25] Response variables included steady state infiltration rate, mm to runoff, runoff and sediment yield after 120 min, and maximum erosion rates. The set of predictor variables contained forest management practices, surface cover properties (bare soil, vegetation, stones, litter), days after last rainfall, slope, bulk density, depth of organic horizons, and applied rainfall intensities (see section 4.1). In the case of sediment yield, constant infiltration rate, mean suspended sediment concentration (SSC), and maximum erosion rates were added. The maximum erosion rates model additionally contained mean and maximum sediment concentration, mm to runoff, and peak runoff (see section 4.2). An overview of the variables used in this analysis is given in Table 2.

    Table 2. Random Forest Predictor and Response Variables
    Predictor VariablesResponse Variables
    Steady State Infiltration RateMm to Runoff120 Min Sediment YieldMaximum Erosion Rates
    1. X indicates whether included or not in the corresponding RF model.

    2. a

      Predictor variable was used but left out in further analysis due to dominant predictor impact potentially hiding the influence of other predictor variables or after splitting into consistent forest management practices.

    3. b

      Refers to suspended sediment concentration (g/l).

    Forest management(X)a(X)a(X)aX
    Applied rainfall intensities(X)aXXX
    SlopeXXXX
    Days after last rainfallXXXX
    Soil moisture prior to the experimentXXXX
    Bulk densityXXXX
    Surface cover properties• Bare soilXXXX
    • VegetationXXXX
    • StoneXXXX
    • LitterXXXX
    Depth of organic horizonsXXXX
    Steady state infiltration rate  XX
    Mm to runoff   X
    Peak runoff   X
    Maximum erosion rates  X 
    Maximum SSCb   X
    Mean SSCb   X

    [26] Derived from iterative tuning, the RFs were grown for 500 individual trees, and the number of randomly selected variables at each node m was set to 5. Maximal tree complexity was also set to 5 terminal nodes. This setup showed good agreement to recommended model setups [Law and Wiener, 2002]. For each predictor, the variable importance (VI) was quantified as the loss of model performance when that predictor was omitted from the model. Following Strobl et al. [2008], the importance of each predictor (VI) was calculated based on its conditional importance measure. The conditional VI of predictor P is calculated in different steps. First, the mean square error (MSE) of the “out-of-bag” (OOB) predictions for each tree is calculated, and then the values of the predictor variables are randomly permutated before their MSE is estimated again. Finally, the difference d between MSE of the “out-of-bag” predictions for each tree and the MSE of the predictions with randomly permuted (equation ((2))), conditioned values of predictor P* reflect the unscaled permutation importance of the predictor variable averaged over all trees [Strobl et al., 2008] (equation ((2)))

    display math(2)

    with “out-of-bag” abbreviated as OOB and MSE referring to mean square error, which is determined by

    display math(3)

    where RV refers to the observed and modeled value of each response variable, and n is the number of records in out-of-bag data.

    [27] In a next step, the sum over all individual differences d in MSE for each tree t with t  ∈ {1, …,ntree} was averaged over all trees and normalized by the standard error

    display math(4)

    [28] By doing so, unimportant predictors yield low impact on model quality which in turn is reflected in low VI values. Finally, the predictor importance was normalized to 100% for comparison reasons.

    [29] Overall model performance was estimated by the squared Pearson coefficient R2 between modeled and observed responses. All calculations were realized by the statistical environment R [R Development Core Team, 2009]. Random Forest model building was set up using the R packages randomForest [Law and Wiener, 2002] and party [Strobl et al., 2008].

    4 Results

    4.1 Soil and Surface Properties

    [30] The texture of the topsoil is relatively homogeneous across the whole study area (Figure 4) and can be described as loam and its minor variations [Schoeneberger et al., 2002]. Carbon fragments are frequently embedded into the topsoil matrix across the whole area.

    Figure 4.

    Topsoil texture (0–30 cm depth) according to U.S. Department of Agriculture (USDA) soil classification. Samples were taken across the whole study area from 14 pits (unpublished data).

    [31] All plots lie between 140 and 295 m asl along the upper, middle, and lower slope sections. The mean plot slopes reached 20 ± 8.1° for S.A. pre, 15 ± 6.6° for S.A. post, and 21 ± 8.1° for Pichún. Although statistical differences of local gradients of the plot between the distinct series do occur (p-value <0.01), they are insubstantial considering their extensive range.

    [32] Apparent soil density prior to the clear cutting of San Antonio #3 and in Pichún showed no differences and averaged very similar values of 1.06 ± 0.13 g/cm3 and 1.07 ± 0.20 g/cm3, respectively (p = 0.36). After clear cutting San Antonio #3, the topsoil was compacted to a density of 1.42 ± 0.21 g/cm3, a significant increase of more than 30% (p < 0.05).

    [33] Topsoil hydraulic conductivity (ks) was significantly higher only at Pichún (18.6 mm/h) and double that at San Antonio #3, where it yielded similar values of 8.8 and 8.1 mm/h prior to and after the logging activities, respectively.

    [34] In general, topsoil moisture was very low and reached only 3.4 ± 2.77 vol % (S.A. pre), 3.7 ± 1.64 vol % (Pichún), and 5.2 ± 2.96 vol % (S.A. post). However, significant differences are indicated between all series (p < 0.001). The duration of the preceding dry period ranged over 19–32 days for Pichún, 35–56 days for S.A. post, and 4–20 days for S.A. pre.

    [35] Surface cover is relatively homogeneous and dominated by pine needle litter (95 ± 7%) with a thickness of 2.7 ± 1.2 cm for S.A. pre (Table 3). Native species like Greigia landbeckii, Muehlenbeckia hastulata, or Hypochaeris radicata contributed to the remaining 5 ± 7% of the surface cover, while both stones and bare soil were negligible. According to Green et al. [1993], the organic horizons of the forest floor were classified as Mors, which typically shows a tenacious consistency and a compact matted structure in the partly decomposed organic horizon. Although litter cover, which is mainly composed of recent harvest residues like branches and bark, remains high (39 ± 43%) in S.A. post, bare soil cover is dominant (61 ± 43%). The thickness of the organic horizons exhibited the highest variability under S.A. post conditions and yielded 1.7 ± 2.3 cm as a result of denudation removing the litter from steeper slopes and depositing it in local depressions. Similar patchy patterns were observed in Pichún, where the variability in surface cover is highest: bare soil (53 ± 21%) and litter (38 ± 19%) cover most part of the surface, but pioneer species, e.g., Brassica rapa, H. radicata, Cirsium vulgare, or Rubus fruticosus, and stones were also present in some plots. The litter is similar in composition to postlogging San Antonio but had a lower average thickness (0.7 ± 1.3 cm).

    Table 3. Average Surface Cover in % of Total Plot Cover According to Rainfall Simulation Series
     Litter (%)Bare Soil (%)Vegetation (%)Stones (%)
    1. a

      Uncertainty corresponds to standard deviation.

    S.A. pre95 ± 7a0 ± 0a5 ± 7a0 ± 0a
    S.A. post39 ± 43a61 ± 43a0 ± 0a0 ± 0a
    Pichún38 ± 19a53 ± 21a6 ± 6a3 ± 3a

    [36] Topsoil disturbance due to former harvest action accentuated the microtopography which was most pronounced under the recent clear-cutting conditions of S.A. post, with height differences of up to 10 cm within the plots.

    4.2 Results of Drip-Type Rainfall Experiments

    [37] An overview of the hydrological response is presented in Table 4 below.

    Table 4. Hydrological and Erosional Responses According to Simulated Rainfall Intensity and Forest Management Practice
    40 mm/h intensity simulations
                    Runoff Yields (mm) After
     Applied Intensity (mm h−1)Generated Runoff (mm)Runoff coefficientFinal Infiltration Rate (mm min−1)Peak Runoff (mL min−1)30 min60 min90 min120 min
    • a

      Simulation # 5 not considered in the sediment analysis due to very low runoff volume samples affecting the accuracy of sediment transport calculation.

    • b

      Simulations # 9 and 23 excluded in sediment analysis due to missing samples or too small sample volume.

    • c

      Calculated max. SSC 34.63 (simulation # 24) excluded due to very low runoff volume.

    • d

      Simulations # 10 and 24 excluded in sediment analysis due to missing runoff samples or small sample volume.

    • e

      Simulations # 11, 16, and 23 excluded in sediment analysis due to missing runoff samples or small sample volume.

    • f

      Simulation # 27 was excluded in sediment analysis due to small sample volume.

    • g

      Simulations # 17, 30, and 31 were excluded in sediment analysis due to small sample volume.

    • ***

      At 0.001 significance level.

     means.d.nmeans.d.nmeans.d.nmeans.d.nmeans.d.nmeans.d.nmeans.d.nmeans.d.nmeans.d.n
    S.A. pre37.32.1123.35***2.84120.270.241227.28.912229.9147.3125.353.341211.457.701217.2112.131222.5016.5212
    S.A. post39.10.6108.846.79100.220.181030.67.210198.7130.7103.882.46109.016.061013.589.431018.1913.5610
    Pichún37.32.51412.997.85140.180.111430.54.914140.075.9141.461.08123.982.91146.935.01149.886.8914
     
    Erosion response to 40 mm/h intensity simulations
     SSC (g L−1)Erosion Rates (g min−1 m−2)Sediment Yield (g m−2) After   
     means.d.nmeanminmax30 min60 min90 min120 min   
        means.d.nmeans.d.nmeans.d.nmeans.d.nmeans.d.nmeans.d.nmeans.d.n   
    S.A. prea0.090.172670.010.012670.000.00110.020.03110.360.31110.560.42110.710.52110.840.6011   
    S.A. postb1.842.342380.440.742340.130.2380.981.41811.7116.48826.2839.06837.9258.6845.7470.408   
    Pichúnc1.872.373290.200.163290.080.07140.400.28146.344.771412.118.961417.6212.211422.1415.4714   
     
    20 mm/h intensity simulations
     Applied Intensity (mm h−1)Generated Runoff (mm)Runoff CoefficientFinal Infiltration rate (mm min−1)Peak Runoff (mL min−1)Runoff Yields (mm) After
                    30 min60 min90 min120 min
     means.d.nmeans.d.nmeans.d.nmeans.d.nmeans.d.nmeans.d.nmeans.d.nmeans.d.nmeans.d.n
    S.A. pre18.71.1123.31.6120.260.221213.94.412112.476.4121.541.47124.103.44126.985.35129.897.5212
    S.A. post19.20.8105.14.7100.230.211014.84.01095.955.1101.321.22103.192.75105.184.35108.045.929
    Pichún19.01.1913.314.590.140.13916.32.5942.640.790.780.7492.271.9484.902.9967.833.835
     
    Erosion response to 20 mm/h intensity simulations
     SSC (g L−1)Erosion Rates (g min−1 m−2)Sediment Yield (g m−2) After   
     means.d.nmeanminmax30 min60 min90 min120 min   
        means.d.nmeans.d.nmeans.d.nmeans.d.nmeans.d.nmeans.d.nmeans.d.n   
    S.A. pre0.090.032930.010.012930.000.00120.020.03120.180.21120.360.48120.450.52120.540.5612   
    S.A. postd1.110.931480.140.181530.030.0480.230.3083.054.9986.2910.1588.6813.06815.5719.456   
    Pichúne2.802.371190.170.141020.130.2460.430.4763.843.1667.886.59614.669.81523.5710.44   
     
    10 mm/h intensity simulations
     Applied Intensity (mm h−1)Generated Runoff (mm)Runoff CoefficientFinal Infiltration Rate (mm min−1)Peak Runoff (mL min−1)Runoff Yields (mm) After
                    30 min60 min90 min120 min
     means.d.nmeans.d.nmeans.d.nmeans.d.nmeans.d.nmeans.d.nmeans.d.nmeans.d.nmeans.d.n
    S.A. pre9.60.6125.55.4110.180.16127.91.71256.247.7120.390.47121.041.09121.992.00112.832.8911
    S.A. post9.60.31011.59.070.090.14108.71.31017.524.4100.220.42100.581.0290.941.6091.482.358
    Pichún9.30.4314.415.330.180.1337.61.2332.622.830.450.3931.080.7731.881.2632.741.843
     
    Erosion response to 10 mm/h intensity simulations
     SSC (g L−1)Erosion Rates (g min−1 m−2)Sediment Yield (g m−2) after   
     means.d.nmeanminmax30 min60 min90 min120 min   
        means.d.nmeans.d.nmeans.d.nmeans.d.nmeans.d.nmeans.d.nmeans.d.n   
    S.A. pref0.130.16138<0.01<0.01138<0.01<0.01110.01<0.01110.060.05110.110.09110.160.13110.210.1511   
    S.A. postg0.660.92480.020.0248<0.01<0.0170.010.0270.250.2870.490.6570.691.0070.851.426   
    Pichúng1.242.36500.050.0650<0.01<0.0130.180.1732.712.2833.592.9524.453.4835.654.513   

    4.2.1 Infiltration and Runoff Production

    [38] Runoff initiated fastest under plantation cover (S.A. pre) after 5.5 ± 5.4 mm, 3.3 ± 1.6 mm, and 3.4 ± 2.8 mm of applied rainfall under 10, 20, and 40 mm/h intensities, respectively (Table 4). Surface runoff initiated faster on the recent clear cutting (S.A. post) than on the older one (Pichún) under all rainfall intensities although not indicated as substantial (p > 0.05).

    [39] About two thirds of all experiments followed a runoff pattern with a rapidly rising runoff limb towards a constant equilibrium runoff rate. The other experiments (n = 26) showed an enhanced infiltration rate towards the end of the simulations (Figure 5), which was observed under 20 and 40 mm/h rainfall intensities, mostly under the plantation cover of S.A. pre (n = 14).

    Figure 5.

    Example hydrographs showing runoff and erosion response to applied rainfall of different rainfall intensities and forest management practice. Prelogging conditions are represented by the simulations #16 and #28 of S.A. pre, recent logging conditions by #14 of S.A. post and older logging by #7 of Pichún. Note different scales of Pichún #7. Runoff rates (mL/m2/min) correspond to black lines, and red points refer to erosion rates (g/m2/min). Gray bars on the top represent applied rainfall intensity in mm/h during intervals.

    [40] Normalized to 1 m2, peak runoff was highly variable among all forest management practices and rainfall intensities. Regardless of the applied intensities, S.A. pre generated the highest peak runoff rates (Table 4). Under 40 mm/h intensity, the recent clear cutting reached peak runoff rates of 198.7 ± 130.7 mL/min, and this clearly exceeded the old clear cutting (140.0 ± 75.9 mL/min). This order changed under lower rainfall intensities, when Pichún nearly double those in S.A. post (95.9 ± 55.1 versus 42.6 ± 40.7 mL/min and 32.6 ± 22.8 mL/min versus 17.5 ± 24.4 mL/min, respectively). Despite differences among all forest management practices and rainfall intensities, they are significant only between the pre- and postlogging conditions of San Antonio under 10 mm/h intensity (p = 0.01).

    [41] Infiltration rate is a function of rainfall intensity and infiltration capacity because infiltration rates increase as rainfall intensities become greater and finally converge at an intrinsic infiltration capacity [Dunne et al., 1991]. Steady state infiltration rates did not show significant differences between the forest management practices among all applied rainfall intensities (p > 0.05). However, a slight change in infiltration rates of ~10, 6, and 12% at 10, 20, and 40 mm/h intensity, respectively, was observed under the recent clear-cutting conditions of S.A. post and −4, 18, and 12% at 10, 20, and 40 mm/h intensity, respectively, under the older clear cutting of Pichún (Figure 6a). Minor differences between the applied rainfall intensities fail to explain the observed changes (Table 4). The mean infiltration depths reached 12.2 ± 12.1 cm prior to and 8.9 ± 5.8 cm after the logging activities of San Antonio (Figure 6b) (p = 0.51).

    Figure 6.

    (a) Steady state infiltration rates (mm/h) as a function of applied rainfall intensity and forest management practice. Error bars show standard deviation (n = 12, 12, and 12 for S.A. pre under 10, 20, and 40 mm/h; n = 10, 10, and 10 for S.A. post under 10, 20, and 40 mm/h; n = 3, 9, and 14 for Pichún under 10, 20, and 40 mm/h rainfall application). (b) Infiltration depth (cm) prior to (n = 32) and after clear cutting (n = 30) San Antonio. The red dots represent the mean values.

    [42] The unlogged plots of S.A. pre yielded most runoff under all intensities (2.8 ± 2.9 mm, 9.9 ± 7.5 mm, and 22.5 ± 16.5 mm at 10, 20, and 40 mm/h intensity, respectively) (Figure 7a). At 10 mm intensity, Pichún clearly produced more runoff than S.A. post (2.74 ± 1.84 mm versus 1.48 ± 2.35 mm), which flipped vice versa under 40 mm/h rainfall intensity when S.A. post yielded nearly double that for Pichún (18.2 ± 13.6 mm versus 9.9 ± 6.9 mm). At 20 mm/h rainfall, both clear cuttings ended up similar (p = 0.90), yielding 8.04 ± 5.92 mm and 7.83 ± 3.83 mm, respectively. Despite high variability, there is a strong positive linear relationship between runoff ratios between both clear cuttings (runoff yield S.A. post/ runoff yield Pichún) and increasing applied rainfall intensities. This relationship depicts a rainfall intensity threshold of ~19 mm/h above which the recent clear cutting starts exceeding the older one in terms of runoff yield (Figure 7b). Significant differences are only weakly indicated between Pichún and S.A. pre under 40 mm/h intensity ( p = 0.05).

    Figure 7.

    (a) Mean runoff yields as a function of forest management practice and applied rainfall intensity. Error bars represent standard deviation. (b) Ratios between the runoff yields of the young and old clear cuttings after 120 min of runoff as a function of applied rainfall intensity. The ratio is expressed as fraction (S.A. post/Pichún) of the mean values. The linear fit and the model performance (squared Pearson coefficient) are given. y corresponds to the mean runoff yield ratio, and x to the applied rainfall intensity. The horizontal gray dashed line shows 1:1 ratio, and the vertical gray dashed line illustrates the threshold rainfall intensity (n = 12, 12, and 11 for S.A. pre under 10, 20, and 40 mm/h; n = 8, 9, and 10 for S.A. post under 10, 20, and 40 mm/h; n = 3, 5, and 14 for Pichún under 10, 20, and 40 mm/h rainfall application).

    4.2.2 Sediment Transport and Erosion Rates

    [43] Sediment yields were low and did not exceed 1 g/m2 after 120 min of rainfall application regardless of the intensity under the unlogged conditions of S.A. pre (Table 4). Under 10 mm/h rainfall intensity, S.A. post yielded only 0.85 ± 1.42 g/m2 sediment, a value within the same order of magnitude of S.A. pre (Figure 8a), while Pichún yielded 5.65 ± 4.51 g/m2 (p < 0.05).

    Figure 8.

    (a) Mean sediment yields as a function of forest management practice and applied rainfall intensity. Error bars represent standard deviation. (b) Ratios between the sediment yields of the young and old clear cuttings after 120 min of runoff as a function of applied rainfall intensity. The ratio is expressed as fraction (S.A. post/Pichún) of the mean values. The linear fit and the model performance (squared Pearson coefficient) are given. y corresponds to the mean sediment yield ratio, and x to the applied rainfall intensity. The horizontal gray dashed line shows 1:1 ratio, and the vertical gray dashed line illustrates the threshold rainfall intensity (n = 11, 12, and 11 for S.A. pre under 10, 20, and 40 mm/h; n = 6, 6, and 8 for S.A. post under 10, 20, and 40 mm/h; n = 3, 4, and 14 for Pichún under 10, 20, and 40 mm/h rainfall application).

    [44] Under intermediate intensities, both harvest areas had similar sediment yields (p = 0.26) at 15.57 ± 19.45 and 23.57 ± 10.40 g/m2 for the recent and old clear cutting, respectively. Overall maximum sediment yields were measured at 40 mm intensity for S.A. post, being twice as high as for the Pichún sites (45.74 ± 70.40 g/m2 versus 22.14 ± 15.47 g/m2), which reached its maximum under 20 mm/h. However, differences do not appear to be significant (p = 0.71). Like runoff yields, sediment ratios between both clear cuttings showed a similar threshold of 22.6 mm/h above which S.A. post starts to exceed Pichún (Figure 8b).

    [45] Despite high variability, significant differences among erosion rates occurred between all forest management practices. Regardless of the rainfall intensities, erosion rates did not exceed 0.01 g/m2/min under unlogged conditions of S.A. pre. These values are clearly exceeded by rates observed under both clear cuttings even under low rainfall intensity (Table 4). While both clear cuttings showed similar erosion rates under intermediate intensity (0.14 ± 0.18 g/m2/min versus 0.17 ± 0.14 g/m2/min), the younger harvest area doubled the rates of Pichún when 40 mm/h rainfall was applied. Under these conditions, S.A. post reached maximum rates of up to 3.8 g/m2/min. Figure 9 depicts a similar positive linear relationship between the erosion ratios between both clear cuttings, and increasing rainfall intensities points to similar threshold intensity.

    Figure 9.

    Ratios between the erosion rates of the young and old clear cuttings as a function of applied rainfall intensity. The ratio is expressed as fraction (S.A. post/Pichún) of the mean values. The linear fit and the model performance (squared Pearson coefficient) are given. y corresponds to the mean erosion rate yield ratio, and x to the applied rainfall intensity. The horizontal gray dashed line shows 1:1 ratio, and the vertical gray dashed line illustrates the threshold rainfall intensity.

    [46] The temporal evolution of the erosion rates showed two different patterns (Figure 5). Under prelogging conditions of S.A. pre (17 out of 36) and the majority of the Pichún experiments (17 out of 26), erosion rates showed an initial increase up to a peak erosion rate followed by a gradual decline over time. In contrast, under the recent clear-cutting conditions of S.A. post, runoff rates are closely tracked by the erosion rates (Spearman's ρ ~ 0.74).

    4.3 The Hydrological and Erosional Response as a Function of Environmental Variables

    4.3.1 Infiltration and Runoff Response

    [47] Rainfall intensity was the most important predictor variable for infiltration rate since mean square error (MSE) increased by 42.2% when rainfall intensity is omitted (Figure 10c). An infiltration model including all forest management practices explained 71.8% of variance (Figure 10a). The same held true for the runoff yield model where rainfall intensity remained the most important variable (VI = 31.5%), although the model explained only 15.8% of overall variance (Figures 10b and 10d).

    Figure 10.

    Performance of (a) infiltration (n = 92) and (b) runoff yield (n = 84) models including all rainfall simulations. Numbers show model performance given as squared Pearson coefficients between modeled and observed values (in Figures 10c and 10d). Predictor importance is estimated for (c) infiltration model and (d) runoff yield model (predictors: rainfall intensity refers to the normalized applied rainfall intensity (mm/h) considering local slope; antecedent dry period refers to the dry period (days) prior to the experiments; bare soil, litter, vegetation, and stones are expressed as percentage of total surface cover; forest management practice describes the classes of S.A. pre, S.A. post, or Pichún; antecedent soil moisture (vol %) was measured immediately prior to the experiments; bulk density (g/cm3) was measured adjacent to upper and lower boundary of the plot prior to the experiments; organic horizons correspond to the compound thickness of all lhf layers (cm); and slope represents the local slope (°). Note: Even slight negative impact is given for distinct predictors.

    [48] Since a dominant predictor may hide other potentially important predictors [Law and Wiener, 2002], applied rainfall intensity was excluded for further analysis. However, none of the remaining predictors showed substantial impact on infiltration rate, and model performance decreased to only 12.5% of explained variance. In a next step, the distinct forest management practices were analyzed separately. Models based on such homogeneous classes did not perform satisfactorily, and only the prelogging-model (S.A. pre) performed slightly better (Figure 11a). This is surprising since variance is expected to be smaller within homogeneous classes. Here, the antecedent dry period showed some impact on infiltration rates (VI = 15.6%; Figure 11d). However, overall model performances and predictor identification for either infiltration or runoff yield models were weak under all forest management practices (Figures 11a–11f).

    Figure 11.

    (a–c) Infiltration rate model performance and (d–f) predictor importance according to each forest management practice excluding applied rainfall intensity as predictor variable (n = 36 for S.A. pre, n = 30 for S.A. post, and n = 26 for Pichún). For variable explanation, see Figure 10. Numbers show squared Pearson coefficient and indicate model performance (in Figures 11a–11c). Note: even slight negative impact is given for distinct predictors.

    [49] The runoff initiation model performed poorly and explained only 18.8% of variance including all forest management practices. However, runoff initiation was more a function of depth of organic layers (VI = 9.6%) and antecedent dry period (VI = 10.5%) than related to forest management practice (VI = 8.8%) or rainfall intensity (VI = 2.8%). Analyzing the distinct forest management practices separately, the particular models performed even worse.

    4.3.2 Erosion and Sediment Yield

    [50] Only when including antecedent sediment yields (after 30, 60, and 90 min of runoff) did the models perform well. These variables, however, were excluded due to strong intercorrelation effects. Such spurious correlation effects only mimic direct causal connection and may hide relevant but weaker predictor variables [Strobl et al., 2008]. The sediment yield model for all forest management practices explained 51.8% of variance, slightly worse than under prelogging conditions of S.A. pre (62.3%) and comparable to under recent clear-cutting conditions of S.A. post (53.1%). For Pichún, the model performed very poorly (17.7%; Figure 12c). Maximum erosion rates were the most important predictors under all forest management practices (VI ≤ 20.9%), while mean SSC showed some minor impact only under recently logged conditions (Figures 12d–12f). Maximum erosion rates, in turn, are (weakly) related to maximum SSC (VI = 10.9 %) and peak runoff (VI = 10.9%), while the impact of the forest management practice itself is negligible. These models explained at most 51.0% of variance (S.A. post).

    Figure 12.

    (a–c) Sediment yield model performance and (d–f) predictor importance according to each forest management practice (n = 35 S.A. pre, n = 21 S.A. post, n = 22 for Pichún). For variable explanation, see Figure 10. In addition, maximum erosion rates (g/m2/min) registered during the simulations and mean SSC (suspended sediment concentration in g/L) were added as predictors. Numbers show squared Pearson coefficient and indicate model performance (in Figures 12a–12c). Note: Even slight negative impact is given for distinct predictors.

    5 Discussion

    [51] In contrast to an intensified surface runoff which is normally associated with postlogging topsoil compaction [e.g., Bathurst et al., 2011a; Bathurst et al., 2011b; Birkinshaw et al., 2011; Carr and Loague, 2012], slightly higher postlogging infiltration and, thus, lower runoff, were registered. These findings are counterintuitive because clear cutting is normally associated with decreased macroporosity after mechanical soil disturbance decreasing both infiltrability and hydraulic conductivity due to compaction [Huang et al., 1996; Malmer and Grip, 1990]. Interestingly, a similar effect was also registered on a catchment scale. Figure 13 shows an increased probability of approximately 10% of lower streamflow discharge of a recently logged catchment (San Antonio #3) compared with an untreated control catchment (San Antonio #1). Hence, increased infiltration after logging activities is not restricted to plot scale. The apparent inconsistency of lower streamflow even prior to the timber harvest can be explained by the construction of timber road and site preparation which took place several weeks before the clear cutting started.

    Figure 13.

    Blue hydrographs show specific daily discharge (m3) per hectare during the years (a and c) 2008 and (b and d) 2009 for the control and treatment catchments, respectively. Months are indicated at the bottom. Hourly rainfall intensities (mm/h) are given as black bars. Red dashed lines represent the period of logging activities. At the bottom, specific discharge quotients between control and treatment catchments are shown during the years (e) 2008 and (f) 2009. The black dotted line represents 1–1 ratio. Refer to Table 1 for catchment details.

    [52] In consequence, the impact of logging activities lowering infiltration must have been compensated up to such an extent that postlogging enhancement of infiltration rates was established, or the time between two consecutive clear cuttings was too short to allow complete recovery of soil hydraulic properties to initial conditions [e.g., Hofstede et al., 2002; Ziegler et al., 2006]. Enhanced matrix flow promoting infiltration is not feasible as hydraulic conductivity remained unaffected. Moreover, soil compaction after the use of heavy timber machinery even exceeds the range of values reported for harvest areas under comparable environmental conditions [Croke et al., 2001; Gayoso and Iroumé, 1991], which favors decreased porosity and thus lower matric flow conductivities [e.g., Malmer and Grip, 1990]. Thus, preferential flow processes are required to bypass water through the compacted and low-conductive soil. The dye tracer experiments revealed preferential flow paths (Figure 14). Under dry and hydrophobic soil conditions, such processes may be initiated owing to low potential differences between macropores and soil matrix [Weiler and Naef, 2003].

    Figure 14.

    Observed preferential flow patterns in topsoil along recent and former root systems and desiccation cracks of recent clear cutting of San Antonio (simulation #31). White dashed line represents the surface edge.

    [53] Prelogging observations suggest strong hydrophobic behavior of the organic layer. Firstly, antecedent dry periods show a (weak) impact on infiltration rates (Figure 11d), and runoff initiation (weakly) responded to the organic layers only under pine stands. Under comparable environmental conditions (undisturbed pine plantation, dry season, and low soil moisture content), Cerdà [1998] showed reduced infiltration rates and fast runoff response owing to a water-repellent surface. Pine and cypress litter consists of similar amounts of water-repellent resins, waxes, or aromatics [Doerr et al., 2000]. Miyata et al. [2009] confirmed the importance of hydrophobic organic layers for fast runoff initiation and higher peak runoff under cypress plantation cover. Moreover, increasing infiltration rates towards the end of the experiments are primarily observed under plantation cover (39% for S.A. pre, compared to only 27% and 15% for S.A. post and Pichún, respectively) and imply declining water repellency of the organic horizons over time [e.g., Imeson et al., 1992]. Finally, frequently observed charcoal fragments embedded in the topsoil provide independent evidence for former forest fires which are reported to promote water repellency [e.g., Cerdà and Doerr, 2005, Doerr et al., 2000; Imeson et al., 1992]. Considering the relatively high water volume applied under low rainfall intensities before runoff initiation (Table 4), the degree of declining water repellency by advancing topsoil wetting was high. Therefore, it is not surprising that water repellency effects only emerged under higher rainfall intensity experiments [Seifert, 2011] (Figure 5).

    [54] As a result, infiltration increases when water-repellent surfaces are broken up, as is the case during timber harvest. Heterogeneous microtopography shows variable degrees of surface disturbance/impact across the harvest area. Hence, infiltration is expected to enhance non-uniformly but in a patchy spatial distribution. However, variability in either infiltration rates or wetting front depth remained unaffected by the logging activities (Table 4, Figure 6b).

    [55] Preferential flow along macropores initiates when water supply exceeds all losses of matrix flow [Beven and Germann, 1982; Bronstert and Plate, 1997]. Weiler and Naef [2003] reported enhanced infiltration rates towards the end of their experiments when preferential flow had been triggered. The observations obtained here conform to their interpretation when the threshold hydraulic conductivities of 8–9 mm/h for San Antonio and 18–19 mm/h for Pichún are considered. Owing to trapped air, macropore flux may be temporarily restricted [Beven and Germann, 1982], which may be reflected in the observed short peak in runoff preceding an intensified infiltration under 20 mm/h and 40 mm/h rainfall intensities (Figure 5).

    [56] Preferential flow processes do not only occur at the plot scale. Table 5 lists the lag times for the catchments San Antonio #1 and #2. Lag time refers to the time interval between the center of mass of a rainfall event and the observed peak flow at the catchment outlet [Dingman, 2002]. Two comparable rainfall events (June 2009 and August 2010) showed fast response (27–33 and 39–54 min, respectively). However, only when connectivity between fast draining vertical and underlying horizontal flow paths is provided, a rapid response to rainfall events may occur [Blume et al., 2008; Montgomery et al., 1997]. This requirement is met with the fast draining interface between the saprolite and bedrock as recently demonstrated by Mohr et al., [2012].

    Table 5. Estimated Lag Times for Control and Treatment Catchmenta
     June 2009bAugust 2009cAugust 2010d
    1. a

      No discharge data is available for Pichún.

    2. b

      June: 27 June 2009 at 23:00 to 28 June 2009 at 12:00 A.M.; soil moisture: 28.4 and 24.7 (vol %) in catchments 1 and 3 measured on 17 June.

    3. c

      August 2009: 13 August 2009 at 23:00 to 14 August 2009 at 11:00; soil moisture: 38.4 and 36.9 (vol %) for catchments 1 and 3 measured on 13 August.

    4. d

      August 2010: 26 August 2010 at 8:00 to 26 August 2010 at 22:00; no soil moisture data available.

    Total precipitation (mm)41.273.540.6
    Hourly peak intensities (mm/h)6.912.910.2
    Event duration (h)131214
     
     CentroidQpeakLag Time (min)CentroidQpeakLag Time (min)CentroidQpeakLag Time (min)
    San Antonio #14:24 A.M.4:57 A.M.335:10 A.M.7:45 A.M.15512:3613:1539
    San Antonio #34:24 A.M.4:51 A.M.275:10 A.M.7:57 A.M.16712:3613:3054

    [57] Although water repellency and preferential flow paths provide a reasonable explanation for the observations, by themselves they are insufficient to explain the runoff and erosion threshold behavior of the harvest areas. Timber harvest enhanced hydraulic surface roughness by affirming the local relief [Clarke and Walsh, 2006; Malmer and Grip, 1990]. Under the surface conditions of the recent harvest area, runoff connectivity was established, particularly during the 40 mm/h rainfall intensity experiments after the retention capacity of surface storage was exceeded. Surface runoff initiated then as a function of saturated surface and rainfall intensity [Dunne et al., 1991] along cascading runoff paths, e.g., skidder tracks or drag lines. These observations are in line with the threshold rainfall intensity of approximately 20 mm/h, above which the runoff on the younger harvest area started to overcome the retaining effect of local topography and started to outperform the less rugged terrain on the older one (Figure 7b). The runoff threshold is very similar to those observed for sediment yields and erosion rates (Figures 8b and 9). As long as connectivity has not been established, prolonged ponding promotes infiltration and finally percolation recharging the groundwater. The catchment scale response supports that interpretation (Figure 13). Thus, erosion may be surprisingly low even under the bare soil conditions of a young harvest area under low rainfall intensities. Once the threshold rainfall intensity has been exceeded, runoff and erosion severely intensifies.

    [58] Hence, our results suggest that timber harvest areas do not perform statically either as sinks or sources for runoff or erosion but instead switch their behavior along an intrinsic threshold. In consequence, one may ask if hydrological rainfall response models based on static rainfall-intensity-infiltrability thresholds are sufficient under disturbed terrain conditions such as after logging activities [e.g., Brenner, 2011; Carr and Loague, 2012; Ebel et al., 2008]. Under the local meteorological conditions, the young harvest area is expected to perform in 90% of the rainfall events as a sink for runoff and erosion (Figure 2b). Thus, severe erosive runoff will occur only during high intensity rainfall events (in approximately only 10% of cases), pointing to the importance of lower frequency but higher magnitude events following Wolman and Miller [1960].

    [59] The erosion process itself differed considerably across the forest management practices. While the erosion rates are closely tracked by runoff rates under recent clear-cutting conditions, the simulations conducted under the litter cover of S.A. pre reflect maximum transport during the early stages of the experiments (Figure 5; Table 4). Hence, transport limitation provides a first-order control of the erosion process under recent clear-cutting conditions owing to abundant sediment supply or high erodibility [e.g., Burt et al., 1983; Croke et al., 2001]. Here, the soil erosion is intensified by abundant sediment availability after breaking up formerly covered soil than by raising erodibility. Aggregate stability remained high after logging activities and reached 5.4 ± 0.7 at the surface (n = 468) and 5.3 ± 0.7 at 20–25 mm depth (n = 468) on a relative scale from 0 to 6 [Herrick et al., 2001]. Although only sheet erosion processes may be simulated on plot scale, Figure 15 indicates similar processes on catchment scale. A clear figure-of-eight loop supports abundant sediment supply for the recently logged catchment, while the clockwise loop, as observed under undisturbed forest plantation cover, suggests supply limitation [Williams, 1989; Faul, 2011].

    Figure 15.

    Hysteresis loop show temporal suspended sediment concentration (g/L)-discharge (L/s) relationship for (a) untreated control catchment (catchment #1 in Figure 1b) and (b) logged catchment (catchment #3 in Figure 1b) during the same rainfall-runoff event on 11–15 August 2009. This event yielded 104.1 mm distributed over 92 h with a maximum hourly intensity of 10.4 mm/h.

    [60] A surprisingly good agreement between catchment scale response and low intensity rainfall experiments emerged. Normalizing the total transported sediment (273.3 kg/ha and 10.1 kg/ha for S.A. post and S.A. pre [Faul, 2010], respectively) on plot scale and experiment duration, erosion yields 0.59 g/m2/2 h and 0.02 g/m2/2 h under logged and untreated conditions, respectively (Table 4). Although soil erosion is high compared to undisturbed plantation conditions [e.g., Miyata et al., 2009], the results lie more in the lower range of reported soil erosion on different aged harvest areas [e.g., Croke et al., 2001]. Nevertheless, the peak erosion rates registered on the young harvest area are comparatively high [Michaelides et al., 2009].

    [61] Nevertheless, differences in applied rainfall intensities and properties limit the comparison between studies on similar spatial scales [Seeger, 2007]. Transferability from plot to catchment scale is even more challenging owing to nonlinear scaling relationships of runoff and erosion processes [e.g., Benavides-Solorio and MacDonald, 2001; Parsons et al., 2006]. The overall effects are observable only on catchment scale, and studies based on that scale will always incorporate greater level of complexities and interaction at the cost of better initial understanding. Initial but fundamental process understanding, the definition of the major water and matter fluxes and their feedbacks, in turn, are required for larger scale process understanding but only obtainable under the controlled conditions on plot scale [e.g., Wainwright et al., 2000]. This point is basic, because results obtained on small scale may be crucial for elucidating some of the more complex interactions, which may be “averaged” out on larger scales [Wainwright et al., 2000]. In consequence, the scaling issues between both spatial scales are rather apparent when both scales are regarded as completive.

    [62] Finally, considering the patchy distribution of the hydrological and erosional response reversing with rainfall intensity, it is not surprising that the various hydrological and erosional responses did not exhibit a clear mutual relationship with other environmental predictor variables. For example, sites with high final infiltration rates sometimes needed only a few mm of applied rainfall to initiate surface runoff, while other sites showed low erosion rates despite high runoff rates. A similar difficulty in linking environmental variables with soil hydrological response was found by Coppus [2002], which finally limits the unequivocal determination of the underlying processes. In this context, the “memory” of the landscape in terms of former land uses may play an important role, e.g., when assessing soil recovery times to its initial hydraulic properties and the duration of intrarotational cycles [e.g., Croke et al., 1999b; Croke et al., 2001; Hassler et al., 2011; Ziegler et al., 2006]. In fact, a persistent impact of topsoil disturbances by previous logging is consistent with the unexpected higher postlogging infiltration rates and the missing clear relationships with environmental variables. Nevertheless, as data from unlogged and previously undisturbed sites are unavailable, this hypothesis cannot be tested.

    [63] In the end, this study largely represents a “snapshot” of a specific set of environmental conditions at a distinct time [Wainwright et al., 2000]. Repeating the experiments under different seasonal conditions, e.g., changes in soil moisture [e.g., e.g., Cerdà, 1997, Cerdà, 1998], and quantification of surface roughness may overcome some of the limitations towards a more definite determination of the underlying processes.

    6 Conclusions

    [64] We examined the hydrologic response of soils to clear cutting using rainfall simulations. Surprisingly, infiltration increased after clear cutting, a finding which apparently contradicts previous studies. This conflict, however, can be explained by a combination of two processes: (1) breakup of water-repellent surface properties and (2) increased surface storage owing to interrupted surface runoff connectivity which in turn prolongs ponding. However, this is only the case under rainfall intensities of less than 20 mm/h. Under higher rainfall intensities, microtopography-connectivity establishes, and intense surface runoff and subsequently soil erosion may be triggered. In consequence, harvest areas behave nonstationarily and may switch from being a sink of runoff and erosion to being a source, a fact previously disregarded in predictive models.

    [65] Despite restrictions in spatial transferability, both spatial scales showed agreement in hydrological and erosional responses. In particular, very similar sediment yields observed on both scales highlight the reinforcing effect of clear cutting on soil erosion.

    [66] The results suggest that logging activities may have had such an impact that interharvest periods (here 26 years) may be too short for the soil to recover to its initial soil hydraulic properties. Considering the trend towards planting even faster growing species in Chile, e.g., E. globulus, the problem is likely to be aggravated.

    [67] Finally, this study confirms the severe impact of clear cutting on water and soil resources and highlights the dominance of preferential flow processes in high-impact areas, such as intensive forestry. At the end, our findings emphasize the importance of pronounced surface roughness, e.g., by infiltration trenches, and immediate reforestation in order to attenuate erosive surface runoff under such a low rainfall intensity regime.

    Acknowledgments

    [68] The work presented herein is funded by the Chilean Government (Conicyt/BMBF 2009–092, Conicyt/BMBF 243–2010 and Fondecyt 1070218), the International Bureau of the German Ministry of Education and Research (CHL 08/03), and the Graduate School for Natural Disasters (GS NADI) of the University of Potsdam. We thank Forestry SA Mininco for access to our study catchments and financial support for the instrumentation of the catchments and Rafael Rubilar for providing soil data. The authors are particularly grateful to Rodrigo Bravo, Cristian Frêne, Odette Morales, Juan Pablo Navarro, Simon Plate, Franziska Faul, Johannes Brenner, Christian Gläser (University of Braunschweig), and Johanna Lein (University of Greifswald) for helping during the fieldworks. The authors would especially like to acknowledge Andreas Bauer, who assisted also with graphics, and Winnie Seifert for her help in the field and during data analysis. We thank Alexander Densmore, Simon Mudd, Artemi Cerdà, and two anonymous reviewers for helpful critiques of a draft manuscript.