Climate indices strongly influence old-growth forest carbon exchange

We present a decade and a half (1998–2013) of carbon dioxide fluxes from an old-growth stand in the American Pacific Northwest to identify ecosystem-level responses to Pacific teleconnection patterns, including the El Niño/Southern Oscillation (ENSO). This study provides the longest, continuous record of old-growth eddy flux data to date from one of the longest running Fluxnet stations in the world. From 1998 to 2013, average annual net ecosystem exchange (FNEE) at Wind River AmeriFlux was −32 ± 84 g C m−2 yr−1 indicating that the late seral forest is on average a small net sink of atmospheric carbon. However, interannual variability is high (>300 g C m−2 yr−1) and shows that the stand switches from net carbon sink to source in response to climate drivers associated with ENSO. The old-growth forest is a much stronger sink during La Niña years (mean FNEE = −90 g C m−2 yr−1) than during El Niño when the stand turns carbon neutral or into a small net carbon source (mean FNEE = +17 g C m−2 yr−1). Forest inventory data dating back to the 1930s show a similar correlation with the lower frequency Pacific North American (PNA) and Pacific Decadal Oscillation (PDO) whereby higher aboveground net primary productivity (FANPP) is associated with cool phases of both the PNA and PDO. These measurements add evidence that carbon exchange in old-growth stands may be more sensitive to climate variability across shorter time scales than once thought.


Introduction
Old-growth forests represent a small fraction of western North America; in fact, only about 1.1 million hectares or 10% of their historic range exists today (Franklin and Spies 1991). While these late seral stands cover a relatively small fraction of land, they represent a significant pool of carbon that has been stored for centuries in the soil, aboveground biomass, and woody debris. This large carbon pool has the potential to be released suddenly into the atmosphere after a disturbance event, such as a forest fire, or more slowly under rising global atmospheric temperatures via increased decomposition. Without external change, however, it had been assumed that mortality and respiration balance growth in late seral forests making them carbon neutral (Odum 1965, 1969, Franklin et al 1981. A number of recent ecosystem carbon studies (Hollinger et al 1994, Anthoni et al 2002, Knohl et al 2003, Loescher et al 2003, Desai et al 2005, Falk et al 2008, Luyssaert et al 2008, Hudiburg et al 2009, Lichstein et al 2009, Tan et al 2011, Wharton et al 2012 have put this ecological paradigm into question. For example, old-growth forests have the potential to sequester carbon at high rates (>300 g C m −2 yr −1 ) similar to younger forests if climatic conditions are favorable. This behavior has been observed in old-growth stands both in the American tropics (Loescher et al 2003) and along the temperate American West Coast (Falk et al 2008, Wharton et al 2012. Forests along the western coast of North America are subject to variations in climate via the eastward movement of weather patterns caused by equatorial and extratropical ocean-atmospheric oscillations over the Pacific Ocean (Mote et al 2003). These correlations between distant oscillation events and local climate are called teleconnections. In the western Americas, major teleconnection events occur during strong warm and cool phases of the Pacific Decadal Oscillation (PDO), Pacific/North American Oscillation (PNA), and the El Niño-Southern Oscillation (ENSO). Cool phases of each usually bring cooler, wetter winter weather to the Pacific Northwest for time periods lasting from years (ENSO) to a decade (PNA) or multiple decades (PDO).
The interannual variability associated with ENSO has made it possible to robustly study its impacts on net ecosystem carbon exchange ( F NEE ) in US West Coast stands (Goldstein et al 2000, Morgenstern et al 2004, Schwalm et al 2007 and elsewhere (e.g., Loescher et al 2003, Schwalm et al 2011, Parazoo et al 2015 as flux tower records are now long enough to extend one or two decades so that they contain multiple El Niño and La Niña events. Lower frequency oscillations such as the PDO and PNA have also been linked to F NEE variability (e.g. Wharton et al 2009a, Thorne and Arain 2015; however, the robustness of these findings is limited by the lack of multiple events at even the longest running flux tower sites. Sites with biometric measurements, however, can have much longer measurement records (e.g., >50 years) spanning multiple major PNA or PDO cycles.
Here, we present an unparalleled record of ecosystem carbon measurements at the oldest forest in the Fluxnet network, including fifteen years (1998)(1999)(2000)(2001)(2002)(2003)(2004)(2005)(2006)(2007)(2008)(2009)(2010)(2011)(2012)(2013) of eddy covariance (EC) measurements and nearly sixty years  of biometric measurements at the Wind River old-growth conifer forest in southern Washington, USA. Our EC record was long enough to capture six major ENSO phase events including a strong El Niño in 1997-1998, moderate La Niña from 1998, moderate El Niño in 2002-2003, moderate La Niña in 2007, strong El Niño in 2009-2010, and strong La Niña in 2010 To the authors knowledge this paper presents the longest ENSO-F NEE analysis published to date. The biometric data captured PNA phase changes arounds 1958, 1970, 1977, 1988, and 1998, a major shift in the PDO in 1976-1977, and a second less certain PDO shift in the mid 1990s. The extended years of EC and biometric data build and advance earlier teleconnection findings at this site (Wharton et al 2009a(Wharton et al , 2012 and show the utility of long term carbon measurement studies in assessing the response of long-living ecosystems, such as natural old forests, to climate fluctuations which occur across a wide range of time scales.
The major goals of this paper were to: (1) Identify any correlations between the Pacific climate indices and variability observed in climatic records of air temperature and precipitation at Wind River.
(2) Assess whether the Wind River old-growth forest has been a stable net carbon sink, source, or carbon neutral based on EC measurements over the last fifteen years. If interannual variability is found to be significant, identify any correlations between ENSO-related climate variability (e.g., precipitation, air temperature) and annual or seasonal ecosystem flux variability (e.g., F NEE , gross primary productivity, F GPP , and ecosystem respiration, F Reco ).
(3) Assess whether biometric measurements of aboveground net primary productivity (F ANPP ) at Wind River correlate with observed variability in the lower frequency Pacific oscillations, PNA and PDO.

Site description
The Wind River Field Station (formerly the Wind River Canopy Crane Research Facility) is located in an unmanaged, 500 hectare, old-growth evergreen needleleaf forest in southern Washington State, USA (45°4 9′ 13.76″ N; 121°57′ 06.88″ W, 371 m above sea level). Stand characteristics are briefly given here while Shaw et al (2004) provide a detailed ecological description.
Stand density is approximately 427 trees per hectare, tree ages range from 0 to ∼500 years , and leaf area index measurements range from 8.2 to 9.2 m 2 m -2 with little seasonality (Thomas and Winner 2000, Parker et al 2002, Roberts et al 2004. The stand is dominated by evergreen conifer species including Douglas-fir (Pseudotsuga menziesii (Mirbel) Franco), the oldest and largest trees in diameter and height, and western hemlock (Tsuga heterophylla (Raf.) Sarg). Average tree heights are 52 m for Douglas-fir and 19 m for western hemlock (Ishii et al 2000). Total estimated biomass is 619 Mg C ha −1 , of which 398 Mg C ha −1 is stored in live biomass and 221 Mg C ha −1 in soil and coarse and fine woody debris (Harmon et al 2004).

Climate data
Continuous meteorological records are available for the last century from the nearby USFS Wind River Ranger Station (1919-1977) (45°28′ 47″ N, 121°33′ 36″ W, 351.1 m a.s.l) and Carson Fish Hatchery NOAA weather station (1977-2015) (45°31′ 12″ N, 121°34′ 48″ W, 345.6 m a.s.l). Historical mean annual air temperature is 8.8°C and mean total water-year (October-September) precipitation is 2325 mm. Precipitation is highly seasonal and is best described by water-year instead of calendar year. The forest receives more than 2 m of precipitation on average during the water-year but less than 15% usually falls during the warmest summer months when water demand is the highest.
Periods of anomalous precipitation were identified using the standardized precipitation index (SPI) (McKee et al 1993). Positive SPI values indicate above normal precipitation while negative SPI values indicate below normal. By definition, the SPI ranges from −3 (extremely dry) to +3 (extremely wet). We calculated a 6 month SPI using the 1919-2013 precipitation record to identify seasonal precipitation anomalies that may reflect anomalies in soil reservoir storage.
Interannual and decadal climate variability in the region is driven in part by the presence and magnitude of interrelated equatorial and extratropical oceanatmospheric oscillations, including the PDO (20-30 year cycle) (Mantua et al 1997), PNA (∼10 year cycle) (Wallace and Gutzler 1981), and ENSO (2-7 year cycle) (Wolter and Timlin 1998). Positive phases of the PDO, PNA, and multivariate ENSO index (MEI) bring significantly warmer and drier winters to the region while negative phases bring cooler and wetter conditions (Mote et al 2003). While all three oscillations affect the local climate, this paper focuses first on the role of ENSO. ENSO had multiple phase shifts within the fifteen year flux record which allows us to assess its influence on old-growth land surface-atmospheric carbon exchange. The influence of the PDO and PNA are considered at the end of the paper when the longterm biometric record of above-ground net primary productivity is shown for last six decades.

Ecosystem carbon data 2.3.1. Biometric
The old-growth forest has 40 ha of permanent measurement plots in nine parallel belt transects. These plots were established in 1947 and are re-measured approximately every five years to gather data on tree recruitment, growth, and mortality. Average aboveground net primary productivity ( F ANPP ) (g C m −2 yr −1 ) is calculated as the mean change in live woody tree carbon storage plus tree mortality and recruitment over the re-measurement interval following methods in Harmon et al (2004). Here, F ANPP does not include biomass changes due to foliar production nor does it include heterotrophic respiration fluxes, belowground autotrophic respiration, or belowground carbon storage (e.g., changes in root biomass).

Micrometeorological
A full suite of half-hourly micrometeorological measurements have been taken at Wind River since 1998, including multiple heights of air temperature, relative humidity, and radiation, and multiple depths of soil temperature and soil moisture. Ecosystem flux measurements are taken above the canopy using the EC method. Net ecosystem fluxes of CO 2 ( F NEE ) are integrated across half-hours to obtain daily, monthly and annual sums, and include the canopy storage flux term ( F s ) as well as the direct EC measurement ( F c ). The flux measurements are quality controlled to identify and flag periods of heavy precipitation, fog, tower shadowing, insufficient fetch, and insufficient turbulence. All missing and excluded carbon fluxes were gap-filled using algorithms devised by Reichstein et al (2005). The ustar correction method was used to identify and 'correct' nighttime carbon fluxes taken during low turbulence conditions. Ecosystem respiration fluxes ( F Reco ) were calculated based on an empirical relationship between nighttime F NEE , taken during sufficient turbulence conditions, and soil temperature and moisture. Photosynthesis (or gross primary production, F GPP ) was calculated as the difference between F NEE and F Reco . F Reco and F GPP are always assigned positive values such that if F NEE is negative, carbon uptake by photosynthesis is greater than carbon loss by ecosystem respiration. For complete details on these measurements and processing techniques see (Paw et al 2004, Falk et al 2005 and Wharton et al 2012).

Results and discussion
Compared to the historical climate record, the oldgrowth forest over the last fifteen years has, on average, experienced drier and near normal air temperature conditions. Within this long-term trend, however, is a large amount of seasonal and interannual variability spanning the full range of climate variability observed over the last century. Important seasonal anomalies include a warm winter in 2003, cool spring in 2008, 2011 and 2012, warm summer in 1998, 2003, and 2009, dry summer in 2002, 2006, and 2012, wet summer 2004, dry spring in 2007, dry winter in 2001, 2005, and 2013, and wet winter in 1999 (figure 1). 2000-2001 had the lowest measurable precipitation since records began in 1919, while 1998-1999 was in the upper 20th percentile of historically wettest years. Some years experienced precipitation that fell more or less evenly throughout the year (2001,2005,2013), while in other years, the vast majority of precipitation (>70%) came during the winter months (1999,2000,2002,2006,2007). Other years experienced relatively wet springs (2003,2011,2012). Most years experienced a very pronounced dry season during the summer which began, on average, in late June to early July and ended in late September to mid-October.
A fast Fourier transform (FFT) was performed on the historical climate records using the statistical software package Origin 8.6 (OriginLab Corp., Northampton, MA). Power peaked at frequencies corresponding to 2.6, 11.7, and 23.5 water-years in the 1919-2013 precipitation record and 4.8 and 27.6 years in the 1930-2013 air temperature record (figure 2). The peaks at 2.6 and 4.8 years overlap with the average phase length for ENSO, the 11.7 years peak corresponds to the PNA, and the 23.5 and 27.6 years peaks correspond with the PDO.
Year-to-year variability in net ecosystem carbon exchange has been high since EC measurements began in 1998 (figure 3). Years with significant F NEE anomalies ( F NEE >mean F NEE ±one standard deviation or On average, annual gross primary productivity (15 yr mean F GPP =1384 g C m −2 yr −1 ) at Wind River is slightly larger than annual ecosystem respiration (15 yr mean F Reco =1352 g C m −2 yr −1 ). The strongest net carbon sink years ( F NEE <−100 g C m −2 yr −1 ) included 1999, 2006, and 2008 and were driven by greater than average F GPP (mean F GPP =1440 g C m −2 yr −1 ) and less than average respiration (mean F Reco =1292 g C m −2 yr −1 ). Strong source years were largely driven by higher rates of respiration than average (mean F Reco =1596 g C m −2 yr −1 ).
The power spectra in figure 2 confirmed that air temperature and precipitation variability is associated with time periods corresponding to phase lengths of ENSO, PNA and PDO. To examine this relationship more closely, mean annual air temperature and winter season SPI were compared to the winter season MEI. Figure 4 shows that both air temperature and the SPI are moderately correlated with the winter MEI. A stronger correlation (r=0.62) between air temperature and MEI was found, whereby warm phases of ENSO are correlated with warmer than average annual temperatures (e.g., 1998, 2003) ( figure 4(a)). While the correlation (r=−0.33) between precipitation and MEI was smaller, a relationship between warm-phase ENSO and drier than normal conditions is also apparent (e.g., 2003) ( figure 4(b)).
Annual F NEE was next plotted against the MEI to assess if a correlation between ENSO and forest carbon sink/source strength has also existed over the last fifteen years. First, it is apparent that the source or sink strength of the old-growth forest is clearly correlated (r=−0.73) with the winter SPI. Drier than normal winters are moderately correlated with small sink years (e.g., 2001), carbon-neutral years (e.g., 2004, 2005, and 2009), or source years (e.g., 2003, 2013) ( figure 5(a)), while wetter winters coincide with larger sink years (e.g., 1999, 2006). A smaller, but still moderate correlation (r=0.55) is also found when annual F NEE is plotted against the winter MEI. Cool phases of ENSO are correlated with larger than normal carbon sink years (e.g., 1999, 2008, and 2011). Likewise warm ENSO phases are correlated with carbon-neutral (e.g., 2005, 2010) or source years (e.g., 1998, 2003) ( figure 5(b)). Environmental drivers linked to precipitation and air temperature include soil water availability, atmospheric water vapor pressure deficit (δe), photosynthetically active radiation (Q p ), and amount and type of clouds. δe is an important driver of canopy gasatmosphere exchange because it influences stomatal conductance and can limit photosynthetic CO 2 uptake at high levels, particularly when soil moisture is limited. The F NEE record at Wind River indicates that maximum half-hourly F NEE uptake occurs, on average, around 12°C and quickly declines once mean air temperatures and δe reach 20°C and 2.0 kPa, respectively (Wagle et al 2016). Net C uptake can occur even at very low temperatures (0°C-5°C) given that there is sufficient light for photosynthesis (Q p >50 MJ m −2 mo −1 ) (Falk et al 2005). Figure 6 highlights the 15 yr averages for some of these environmental drivers and ecosystem fluxes. In figure 6(b), Point 1 shows the average ecological transition from net carbon source to net carbon sink in the early spring as light levels increase, atmospheric water demand is relatively low and soil moisture is plentiful.
Point 2 shows the transition from net carbon sink to source in the middle of summer when water demand peaks and soil water availability becomes limiting. Note the asymmetric timing of the peaks in F GPP and F Reco which results in maximum cumulative net carbon uptake occurring usually in early June. Point 3 shows the region where light levels have high control over F NEE, microclimate conditions favor carbon uptake, and F GPP reaches its annual peak. Point 4 shows the region where atmospheric water demand (and limited available soil moisture) largely limits F GPP and subsequently the old-growth forest turns into a net source of carbon to the atmosphere for the rest of the year until Point 1 is reached the following winter.
Interannual F NEE differences begin to occur, on average, between day 100 and 150 (mid-April-end of May) and the deviation from average continues to increase throughout the rest of the year for warm and cool ENSO phase years (figure 7). Most years begin net carbon uptake around the same time (day 70-79, mid-March) except for cool phase years which on average start net uptake about 2-3 weeks earlier (day 56, end of February). Warm phase years end net carbon uptake on average earlier (day 148, end of May) than neutral or cool phase years (day 169-175, mid to late June).
Continuous net carbon uptake lasts on average 105 days at Wind River but this length depends on ENSO phase strength. Warm phase years see the shortest period of net uptake (70 days) while cool phase years have the longest period of continuous daily −F NEE (115 days). Wind River's net carbon uptake period is relatively short compared to other US evergreen forests (mean=266 days, n=9 forests, stand ages range from 22 to 110 years, mean age=70 years) (Wagle et al 2016). The shorter net uptake period is due to a combination of geography and climate, and forest age and height. Unique site factors that limit F GPP include dry, warm summers with high atmospheric water demand, tall tree heights which induce stomatal closure even at moderate δe levels, and relatively low northern latitude winter/early spring light levels. The presence of large soil and woody biomass/ debris carbon pools favor high R eco .
Our carbon flux measurements show that the strongest interannual carbon flux anomalies largely occur during spring and summer months and appear to be related to ENSO climate variability. Important interannual anomalies in F GPP and F Reco are shown by the numbered points in figures 7(a)-(c). Point 1 shows that during neutral phase years, F Reco is higher than normal in the spring and is greater than F GPP anomalies. Point 2 shows that neutral year summer months have larger F GPP anomalies than F Reco and F GPP is higher than the 15 year mean. These deviations cause average annual F NEE during neutral years to be very close to the 15 year mean.
During warm phase years, Point 3 shows that F Reco is much higher than normal in the spring/early summer due to warmer temperatures while F GPP anomalies are near normal. Although note that by midsummer (Point 4), F GPP significantly decreases compared to the 15 year mean due to higher water stress. Past work has shown that the upper canopy Douglasfir foliage is largely driving observed variability in ecosystem carbon exchange at the old-growth forest (Wharton et al 2009b). When δe is moderate or high and soil moisture is limiting stomatal conductance in these tall trees sharply declines before midday which greatly limits F GPP Yoder 1997, Wharton et al 2009b). By late summer F Reco has also significantly decreased compared to normal due to low soil moisture availability and is perhaps driven by a reduction in root respiration rates. Root respiration at Wind River has been shown to decrease during drier periods (Taylor et al 2015). Overall mean annual F NEE during warm phase years shows that the forest acts as a small carbon source.
During cool phase years both F GPP and F Reco are attenuated in the late spring due to lower temperatures although reductions in F Reco are slightly larger (Point 5) leading to greater net carbon uptake than normal. By early June, Point 6, F Reco stays strongly attenuated due to lower temperatures in the canopy while F GPP moves closer to normal conditions. Both points 5 and 6 favor net carbon uptake and the forest acts as a moderate net carbon sink during cool-phase ENSO years.
Attributing anomalies in F NEE to specific environmental drivers is confounded by the fact that the different ecosystem processes which determine respiration and photosynthesis are correlated with climate variables across a wide range of time scales.
Wavelet cospectra analysis provides some insights into these complex relationships. A recent study by Wagle et al (2016) has shown that Wind River F NEE and Q p are most highly correlated on daily time scales, F NEE and δe on weekly and monthly scales, F NEE and Ta on seasonal time scales, and F NEE and soil moisture on interannual time scales. This analysis also showed that drivers that vary seasonally or annually were more important for F NEE in semi-arid and Mediterranean sites, including Wind River, than in locations with evenly distributed precipitation. Larger variability in F NEE in semi-arid and Mediterranean sites at longer timescales might be related to larger year-to-year variability in soil water availability and drought duration, and the length of cumulative net carbon uptake in these sites (Wagle et al 2016).
While the EC flux record at Wind River encompasses multiple phase changes in ENSO and reveals anomalies on the daily, seasonal and interannual time    drier conditions (mean SPI=−0.2) after the PDO switch. A less certain PDO switch to a cool phase may have occurred in the mid-1990s although the SPI data at Wind River do not show a subsequent increase in precipitation (mean SPI=−0.3). Figure 8 shows that the PDO and PNA have been largely in-phase with the exception of the most recent five years as the PDO has been in a cool phase but the PNA has stayed weakly positive. The biometric F ANPP measurements (five year averages ending in the year shown) indicate an overall decrease in F ANPP since the 1950-1960s. Exceptions to this occur in 1958and 1978 (shown as open circles in figure 8(c)); both measurement years include increased mortality following significant Douglas-fir beetle tree kills in 1951 and 1973, respectively. The short period of increasing F ANPP from 1960 to 1972 is likely due to stand recovery following the first disturbance event which created gaps in the canopy leading to increased recruitment and growth in the lower canopy. Correlations between F ANPP and the PDO and PNA are shown in figure 9. The post-disturbance F ANPP measurements taken in 1958 and 1978 were removed from these plots. Higher F ANPP is correlated with cooler phases of both the PNA and PDO. This finding agrees with tree ring studies in the lower Cascade forests which indicate that radial growth is limited by low summer precipitation and high summer temperatures and increases with higher winter precipitation (Case and Peterson 2005).   Luyssaert et al (2008) suggest that many old-growth stands are, on average, moderate to large net sinks of atmospheric carbon (∼ −200 g C m −2 yr −1 ) due to the imbalance between slow decomposition of dead wood and fast recovery of living vegetation from natural gap events. At Wind River we have not observed this over the last fifteen years; instead the 500 year old forest has been a small average net carbon sink (−32 g C m −2 yr −1 ). Our EC measurements are in better agreement with modeled (−49 g C m −2 yr −1 ) (Turner et al 2015) and biometric estimates (−15 g C m −2 yr −1 ) (Gray et al 2016) of other oldgrowth stands in the region. However, interannual variability in F NEE is high at Wind River and favorable meteorological conditions during La Niña years have the potential to turn the old-growth forest into a much stronger sink as we observed significantly lower F Reco during these events. The magnitude of F NEE interannual variability adds evidence that old-growth forests may be more pliable to climate forcings on shorter time scales than the traditional ecological theory predicts. The sixty year record of F ANPP also shows a correlation with the two major lower frequency oscillations, the PNA and PDO, such that higher F ANPP is also associated with cooler phases of both. This longer time record contained at least two major disturbance events. Confounding influences of climate-induced disturbances (e.g., insect outbreaks, fungal disease, fire) on mortality and forest growth highlight the need for both long-term (e.g., biometric forest inventory data) and short-term records (e.g., EC data) to fully assess climate phase-ecosystem interactions, as natural regeneration and biomass decomposition occur across different time scales (Luyssaert et al 2008).