2. Study site and Methods
2.1. Study site, experimental treatments
The study was conducted in the Valdivian Coastal Reserve on the coast of south central Chile (39°58ʼS, 73°33ʼW) (Figure 1). Mean annual precipitation is 2500 mm and mean annual temperature is 10°C; 95% of precipitation occurs in fall, winter, and spring. The geology consists of Paleozoic metamorphic rocks, partially overlaid by Tertiary marine sediments with a slope of 30° (60%). Soils have a volcanic origin and are Typic Haplohumults (Ultisols, Hueycolla series) with a low pH (4.2–4.8) (CIREN, 2001).
Prior to the conversion to Eucalypt plantations in the 1990s, the vegetation of the study site was Valdivian temperate evergreen forest in areas of abundant annual precipitation (2000 to 5000 mm) from near sea level to nearly 1000 m in the Andes and the Coastal Cordillera from 38°30’ to 47°S (Veblen, Donoso, Kitzberger, & Rebertus, 1996). These forests are dominated by approximately fifteen species, most of which are endemic to south central Chile and adjacent areas of Argentina (Donoso, 2006). Historically, native forests of the study site were selectively logged by local people for wood and wood fuels. Between 1993 and 1999, 3000 ha of native forests in this area were clear-cut, burned, and converted to exotic Eucalyptus plantations (Little, Lara, & González, 2013; Lara et al., 2014). In 2003, The Nature Conservancy purchased an area of 500 km2 from the timber company that established the Eucalyptus plantations and created a reserve to protect rainforests and coastal marine ecosystems. At the start of the experiment in 2006, vegetation in the study catchments consisted of 7-yr-old commercial plantations of Eucalyptus globulus and native forest along streamside buffers (Table 1).
The paired-catchment study is a collaborative agreement among Universidad Austral de Chile (UACh), The Nature Conservancy (TNC) and Masisa S.A., a private forestry company. Masisa S.A. harvestedEucalyptus plantations and planted native tree species in a 45-ha pilot area within the Valdivian Coastal Reserve. In conjunction with TNC and Masisa S.A., UACh managed the design and planning of the restoration experiment including monitoring of streamflow, precipitation and permanent vegetation plots. TNC owns, protects and manages the site.
Precipitation and stream gaging began in 2006 in four small catchments (Figure 1, Figure 2). Catchment size ranges from 3.7 to 5.3 ha, elevation ranges from 6 to 195 m, and mean slope gradient is 41 to 47% (Table 1). In 2010, Eucalyptus plantations occupied 64 to 76% of catchment area, and native forest riparian buffers occupied 24 to 36% of catchment area. Eucalyptus plantations in catchments RC5, RC10 and RC11 were clear-cut in February to April of 2011, when the plantations were 12 years of age (Table 1). Clear-cut catchments were replanted in 2011 with a native tree species, Nothofagus dombeyiat a density of 1,500 seedlings/ha, with supplemental planting in 2012 (Little et al., 2013, Lara et al., 2014). Nothofagus dombeyi is a dominant species in the forests of south central Chile; it may reach 40 m in height (Donoso, 2006). It was chosen because it is a pioneer fast-growing species that is present in the Reserve, and therefore it was expected to rapidly form closed-canopy forest stands (Lara et al., 2013). Areas under restoration were fenced to exclude cattle, and cut stumps of Eucalyptus were treated with herbicide in catchments RC10 and RC11 but not in catchment RC5. Native trees (fifteen species) have seeded in from nearby forest stands, and together with regenerating shrubs, ferns and epiphytes have produced a diverse young forest in the areas that were planted with Nothofagus spp. (Lara et al., 2013, Lara et al., 2014). This study covers five years prior to the clear-cutting of Eucalyptus (2006 to 2010, the pre-treatment period) and nine years after clear-cutting the Eucalyptus and planting of native trees in RC5, RC10, and RC11, as well as continued growth of the Eucalyptus plantation in RC6 (2011 to 2019, the post-treatment period or the period under restoration).
2.2. Field data collection
Precipitation has been measured in canopy openings at five sites over varied periods since February 2006 (Little, Cuevas, Lara, Pino, & Schoenholtz, 2014) (Figure 1) using tipping-bucket gages (Model DAVIS 7852) with a resolution of 0.2 mm, equipped with HOBO event loggers. Precipitation data were compiled on a daily basis. A complete record of daily precipitation for the period April 2006 through March 2020 was created for the rain gage nearest to the four study catchments based on linear relationships with other stations (R2 ranged from 0.94 to 0.98).
Streamflow is measured using 90° V-notch weirs constructed in 2006 (RC5, RC6) and 2008 (RC10, RC11) (Little et al., 2014). The water year starts in April (fall) of the named year and ends in March (summer) of the following year. Stage height was measured manually for water years 2006 to 2008 in RC5 and RC6 and using automated measurements at all catchments for water years 2009 to present. Atmospheric pressure and water pressure at the weir were measured at all stream gaging locations using pressure transducers (HOBO U20-001, with a resolution of 40 Pa). Data were downloaded, compiled at 15-minute resolution, quality checked, converted to discharge and summarized at the daily scale. Mean daily streamflow for each day of the record was expressed on a unit-area basis (mm). Missing values of daily streamflow were filled based on adjacent values (for gaps of one to three days) and relationships with precipitation (see below) for longer gaps. Analyses reported here used observed streamflow for RC5 and RC6 (2006 through 2019) and RC10 and RC11 (2009 through 2019) and filled values for RC10 and RC11 (2006 through 2008). The resulting dataset will be available after publication fromhttp://www.cr2.cl/datos-cuencas-restauracion-reservavaldiviana/ .
Eucalyptus plantations were surveyed in 2010, prior to clear-cutting, by Masisa S.A. (Figure 2). Permanent vegetation plots were established in the understory of the Eucalyptus plantations before they were clear-cut in 2011 (Figure 2). Each circular plot is 12.6 m radius and 500 m2 and contains 20, 1-m radius subplots distributed at 3 m intervals along the cardinal axes. Data on survival, health, height, and diameter at breast height (dbh) were obtained for N. dombeyi and Eucalyptus individuals in two quadrants of the 500 m2 plot. The number of seedlings (< 2 m height), saplings (> 2 m, <5 cm dbh), and trees (dbh > 5cm) and the cover of all non-tree species was recorded in the subplots. Plots were measured in 2010, 2012, 2014, and 2016. Of the 45 plots established in 2010 in the total restored area, four were in RC11, two were in RC5, and one was in RC10. In 2020, plots were resampled and additional vegetation plots of the same design were installed to provide equal sampling density (four plots in each study catchment) (Figure 2). In 2020, one vegetation plot was also established in the native vegetation buffer in RC5, RC10 and RC11, and three plots were sampled in RC6 to determine basal area ofEucalyptus .
2.3. Data analyses
Three methods were applied to estimate the effects of the treatment on streamflow: (1) double-mass plots and runoff ratios; (2) a before-after analysis contrasting post-treatment streamflow to pre-treatment streamflow, 2006 to 2010; and (3) a before-after, control-impact analysis using precipitation and streamflow data for the pre-treatment water years 2009 and 2010. In addition, a base flow separation analysis was performed on the daily data, and seasonal base flow values were correlated to prior precipitation. Analyses were conducted at the multi-year, annual (April to March water year), and seasonal time scales. Seasons were defined as austral fall (April to June), austral winter (July to September), austral spring (October to December), and austral summer (January to March). These methods and their advantages and limitations are described below.
2.3.1. Runoff ratios and double-mass curves
Runoff ratios (Q/P, where Q = streamflow and P = precipitation) were calculated for each year and season. Double-mass curves of cumulative streamflow vs. cumulative precipitation were constructed for all study catchments for the period of record. Runoff ratios and double mass curves include both the effects of changing vegetation and changing climate.
2.3.2. Before-after analysis
A before-after analysis of streamflow was conducted following the method of Swank and Douglass (1974). The average and standard deviation of streamflow during the pre-treatment period was calculated for each catchment. The treatment effect Δ was the difference between observed streamflow in each year of the post-treatment (under restoration) period and the average pre-treatment streamflow,
Δt = Qt – Q [1]
where Qt = streamflow in period t andQ = average streamflow for the pre-treatment period. The pre-treatment period was water years 2006 to 2010. For the 2006 to 2008 water years, the analysis used measured streamflow for RC5 and RC6 and daily streamflow modeled using precipitation for RC10 and RC11 (see below). The before-after comparison presumes that long-term climate (precipitation, potential evapotranspiration) is stationary.
2.3.3. Observed vs. predicted analysis
A before-after, control-impact analysis accounts for non-stationarity in climate. The treatment effect is defined as the change over time in the relationship of streamflow between a treated and a control catchment, which is assumed to have no vegetation change (Eberhardt & Thomas, 1991; Perry & Jones 2017). However, changes in vegetation over time in the control catchment significantly affect the estimated treatment effect (Jones & Post, 2004). To separate the effects of changes in precipitation from changes in vegetation, the relationship of streamflow to precipitation was estimated in each catchment for water years 2009 and 2010, the pre-treatment period when all catchments were instrumented with continuous water level sensors, and used to predict streamflow in the remaining years. The treatment effect was the difference between observed and predicted streamflow, as described below.
Daily antecedent precipitation was calculated from the complete daily precipitation record:
APt = Pt + Pt-1k [2]
where APt = antecedent precipitation on dayt , Pt = precipitation on day t , andk = exponent indicating the “memory” of past precipitation events. Two values of k (0.7 and 0.9) were selected to represent relatively short (k = 0.7) and long (k = 0.9) memory.
Linear models were fitted to predict daily streamflow (Qt ) as function of daily antecedent precipitation (using two values of k ) for each month during the two-year pre-treatment period:
Qt = α + β APt [3]
This produced four models of daily precipitation (2009, 2010, each withk = 0.7 and k = 0.9) for each month of the year. Daily values of streamflow (Qt ) were estimated for all days in the period of record using each of these four models, and the average of the predicted values from the four models and its standard error was determined for each day in the record. The treatment effect,Δ , was then determined as the difference:
Δ = Q’t - Qt [4]
where Q’t = observed streamflow (mm) andQt = predicted streamflow (mm) for each day in the record. The values of Δ were summed by year and by season.
2.3.4. Base flow separation and memory
Total daily streamflow was separated into quick flow and base flow following the method of Chapman and Maxwell (1996). Base flow was calculated as:
k 1  k
Qb(i) =  Qb (i 1) +  Q(i) [5]
2  k 2  k
where Qb = base flow (mm),Qt = total streamflow (mm) and k is a parameter ranging between 0 and 1. Higher values of k increase the fraction of total streamflow represented by Qb. After testing different k values ranging from 0.4 to 0.97, we chose k = 0.95 for spring and summer and k = 0.90 for fall and winter, which produced stable base flow estimates.
The influence of past precipitation on streamflow in each catchment (“memory”) was estimated by correlating seasonal streamflow to precipitation in the current and past seasons. The water balance for time period t is:
Qt = Pt – ETt – ΔSt [6]
where Qt = streamflow, Pt= precipitation, ETt = evapotranspiration, and ΔSt = change in deep soil moisture in periodt . The lagged effect of prior precipitation on current streamflow is:
Qt = Pt-n – ETt-n – ΔSt-n [7]
where Qt is a function of P , ET , and ΔS in time period t-n , a prior season.